question
stringlengths 12
1.77k
| context
stringlengths 79
71.7k
| answer
stringlengths 1
1.63k
|
|---|---|---|
Which best describes the relationship between the protagonists?
A. They're friendly but their friendship detracts from their ability to problem-solve and be productive.
B. They're both in a tough situation but their hatred for one another pushes them to work independently.
C. They work together and are able to coordinate with each other pretty well.
D. They don't like each other too much; they put up with each other at best.
|
The Monster Maker By RAY BRADBURY "Get Gunther," the official orders read. It was to laugh! For Click and Irish were marooned on the pirate's asteroid—their only weapons a single gun and a news-reel camera. [Transcriber's Note: This etext was produced from Planet Stories Spring 1944. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] Suddenly, it was there. There wasn't time to blink or speak or get scared. Click Hathaway's camera was loaded and he stood there listening to it rack-spin film between his fingers, and he knew he was getting a damned sweet picture of everything that was happening. The picture of Marnagan hunched huge over the control-console, wrenching levers, jamming studs with freckled fists. And out in the dark of the fore-part there was space and a star-sprinkling and this meteor coming like blazing fury. Click Hathaway felt the ship move under him like a sensitive animal's skin. And then the meteor hit. It made a spiked fist and knocked the rear-jets flat, and the ship spun like a cosmic merry-go-round. There was plenty of noise. Too damned much. Hathaway only knew he was picked up and hurled against a lever-bank, and that Marnagan wasn't long in following, swearing loud words. Click remembered hanging on to his camera and gritting to keep holding it. What a sweet shot that had been of the meteor! A sweeter one still of Marnagan beating hell out of the controls and keeping his words to himself until just now. It got quiet. It got so quiet you could almost hear the asteroids rushing up, cold, blue and hard. You could hear your heart kicking a tom-tom between your sick stomach and your empty lungs. Stars, asteroids revolved. Click grabbed Marnagan because he was the nearest thing, and held on. You came hunting for a space-raider and you ended up cradled in a slab-sized Irishman's arms, diving at a hunk of metal death. What a fade-out! "Irish!" he heard himself say. "Is this IT?" "Is this what ?" yelled Marnagan inside his helmet. "Is this where the Big Producer yells CUT!?" Marnagan fumed. "I'll die when I'm damned good and ready. And when I'm ready I'll inform you and you can picture me profile for Cosmic Films!" They both waited, thrust against the shipside and held by a hand of gravity; listening to each other's breathing hard in the earphones. The ship struck, once. Bouncing, it struck again. It turned end over and stopped. Hathaway felt himself grabbed; he and Marnagan rattled around—human dice in a croupier's cup. The shell of the ship burst, air and energy flung out. Hathaway screamed the air out of his lungs, but his brain was thinking quick crazy, unimportant things. The best scenes in life never reach film, or an audience. Like this one, dammit! Like this one! His brain spun, racketing like the instantaneous, flicking motions of his camera. Silence came and engulfed all the noise, ate it up and swallowed it. Hathaway shook his head, instinctively grabbed at the camera locked to his mid-belt. There was nothing but stars, twisted wreckage, cold that pierced through his vac-suit, and silence. He wriggled out of the wreckage into that silence. He didn't know what he was doing until he found the camera in his fingers as if it had grown there when he was born. He stood there, thinking "Well, I'll at least have a few good scenes on film. I'll—" A hunk of metal teetered, fell with a crash. Marnagan elevated seven feet of bellowing manhood from the wreck. "Hold it!" cracked Hathaway's high voice. Marnagan froze. The camera whirred. "Low angle shot; Interplanetary Patrolman emerges unscathed from asteroid crackup. Swell stuff. I'll get a raise for this!" "From the toe of me boot!" snarled Marnagan brusquely. Oxen shoulders flexed inside his vac-suit. "I might've died in there, and you nursin' that film-contraption!" Hathaway felt funny inside, suddenly. "I never thought of that. Marnagan die? I just took it for granted you'd come through. You always have. Funny, but you don't think about dying. You try not to." Hathaway stared at his gloved hand, but the gloving was so thick and heavy he couldn't tell if it was shaking. Muscles in his bony face went down, pale. "Where are we?" "A million miles from nobody." They stood in the middle of a pocked, time-eroded meteor plain that stretched off, dipping down into silent indigo and a rash of stars. Overhead, the sun poised; black and stars all around it, making it look sick. "If we walk in opposite directions, Click Hathaway, we'd be shaking hands the other side of this rock in two hours." Marnagan shook his mop of dusty red hair. "And I promised the boys at Luna Base this time I'd capture that Gunther lad!" His voice stopped and the silence spoke. Hathaway felt his heart pumping slow, hot pumps of blood. "I checked my oxygen, Irish. Sixty minutes of breathing left." The silence punctuated that sentence, too. Upon the sharp meteoric rocks Hathaway saw the tangled insides of the radio, the food supply mashed and scattered. They were lucky to have escaped. Or was suffocation a better death...? Sixty minutes. They stood and looked at one another. "Damn that meteor!" said Marnagan, hotly. Hathaway got hold of an idea; remembering something. He said it out: "Somebody tossed that meteor, Irish. I took a picture of it, looked it right in the eye when it rolled at us, and it was poker-hot. Space-meteors are never hot and glowing. If it's proof you want, I've got it here, on film." Marnagan winced his freckled square of face. "It's not proof we need now, Click. Oxygen. And then food . And then some way back to Earth." Hathaway went on saying his thoughts: "This is Gunther's work. He's here somewhere, probably laughing his guts out at the job he did us. Oh, God, this would make great news-release stuff if we ever get back to Earth. I.P.'s Irish Marnagan, temporarily indisposed by a pirate whose dirty face has never been seen, Gunther by name, finally wins through to a triumphant finish. Photographed on the spot, in color, by yours truly, Click Hathaway. Cosmic Films, please notice." They started walking, fast, over the pocked, rubbled plain toward a bony ridge of metal. They kept their eyes wide and awake. There wasn't much to see, but it was better than standing still, waiting. Marnagan said, "We're working on margin, and we got nothin' to sweat with except your suspicions about this not being an accident. We got fifty minutes to prove you're right. After that—right or wrong—you'll be Cosmic Films prettiest unmoving, unbreathin' genius. But talk all you like, Click. It's times like this when we all need words, any words, on our tongues. You got your camera and your scoop. Talk about it. As for me—" he twisted his glossy red face. "Keeping alive is me hobby. And this sort of two-bit death I did not order." Click nodded. "Gunther knows how you'd hate dying this way, Irish. It's irony clean through. That's probably why he planned the meteor and the crash this way." Marnagan said nothing, but his thick lips went down at the corners, far down, and the green eyes blazed. They stopped, together. "Oops!" Click said. "Hey!" Marnagan blinked. "Did you feel that ?" Hathaway's body felt feathery, light as a whisper, boneless and limbless, suddenly. "Irish! We lost weight, coming over that ridge!" They ran back. "Let's try it again." They tried it. They scowled at each other. The same thing happened. "Gravity should not act this way, Click." "Are you telling me? It's man-made. Better than that—it's Gunther! No wonder we fell so fast—we were dragged down by a super-gravity set-up! Gunther'd do anything to—did I say anything ?" Hathaway leaped backward in reaction. His eyes widened and his hand came up, jabbing. Over a hill-ridge swarmed a brew of unbelievable horrors. Progeny from Frankenstein's ARK. Immense crimson beasts with numerous legs and gnashing mandibles, brown-black creatures, some tubular and fat, others like thin white poisonous whips slashing along in the air. Fangs caught starlight white on them. Hathaway yelled and ran, Marnagan at his heels, lumbering. Sweat broke cold on his body. The immense things rolled, slithered and squirmed after him. A blast of light. Marnagan, firing his proton-gun. Then, in Click's ears, the Irishman's incredulous bellow. The gun didn't hurt the creatures at all. "Irish!" Hathaway flung himself over the ridge, slid down an incline toward the mouth a small cave. "This way, fella!" Hathaway made it first, Marnagan bellowing just behind him. "They're too big; they can't get us in here!" Click's voice gasped it out, as Marnagan squeezed his two-hundred-fifty pounds beside him. Instinctively, Hathaway added, "Asteroid monsters! My camera! What a scene!" "Damn your damn camera!" yelled Marnagan. "They might come in!" "Use your gun." "They got impervious hides. No use. Gahh! And that was a pretty chase, eh, Click?" "Yeah. Sure. You enjoyed it, every moment of it." "I did that." Irish grinned, showing white uneven teeth. "Now, what will we be doing with these uninvited guests at our door?" "Let me think—" "Lots of time, little man. Forty more minutes of air, to be exact." They sat, staring at the monsters for about a minute. Hathaway felt funny about something; didn't know what. Something about these monsters and Gunther and— "Which one will you be having?" asked Irish, casually. "A red one or a blue one?" Hathaway laughed nervously. "A pink one with yellow ruffles—Good God, now you've got me doing it. Joking in the face of death." "Me father taught me; keep laughing and you'll have Irish luck." That didn't please the photographer. "I'm an Anglo-Swede," he pointed out. Marnagan shifted uneasily. "Here, now. You're doing nothing but sitting, looking like a little boy locked in a bedroom closet, so take me a profile shot of the beasties and myself." Hathaway petted his camera reluctantly. "What in hell's the use? All this swell film shot. Nobody'll ever see it." "Then," retorted Marnagan, "we'll develop it for our own benefit; while waitin' for the U.S. Cavalry to come riding over the hill to our rescue!" Hathaway snorted. "U.S. Cavalry." Marnagan raised his proton-gun dramatically. "Snap me this pose," he said. "I paid your salary to trot along, photographing, we hoped, my capture of Gunther, now the least you can do is record peace negotiations betwixt me and these pixies." Marnagan wasn't fooling anybody. Hathaway knew the superficial palaver for nothing but a covering over the fast, furious thinking running around in that red-cropped skull. Hathaway played the palaver, too, but his mind was whirring faster than his camera as he spun a picture of Marnagan standing there with a useless gun pointed at the animals. Montage. Marnagan sitting, chatting at the monsters. Marnagan smiling for the camera. Marnagan in profile. Marnagan looking grim, without much effort, for the camera. And then, a closeup of the thrashing death wall that holed them in. Click took them all, those shots, not saying anything. Nobody fooled nobody with this act. Death was near and they had sweaty faces, dry mouths and frozen guts. When Click finished filming, Irish sat down to save oxygen, and used it up arguing about Gunther. Click came back at him: "Gunther drew us down here, sure as Ceres! That gravity change we felt back on that ridge, Irish; that proves it. Gunther's short on men. So, what's he do; he builds an asteroid-base, and drags ships down. Space war isn't perfect yet, guns don't prime true in space, trajectory is lousy over long distances. So what's the best weapon, which dispenses with losing valuable, rare ships and a small bunch of men? Super-gravity and a couple of well-tossed meteors. Saves all around. It's a good front, this damned iron pebble. From it, Gunther strikes unseen; ships simply crash, that's all. A subtle hand, with all aces." Marnagan rumbled. "Where is the dirty son, then!" "He didn't have to appear, Irish. He sent—them." Hathaway nodded at the beasts. "People crashing here die from air-lack, no food, or from wounds caused at the crackup. If they survive all that—the animals tend to them. It all looks like Nature was responsible. See how subtle his attack is? Looks like accidental death instead of murder, if the Patrol happens to land and finds us. No reason for undue investigation, then." "I don't see no Base around." Click shrugged. "Still doubt it? Okay. Look." He tapped his camera and a spool popped out onto his gloved palm. Holding it up, he stripped it out to its full twenty inch length, held it to the light while it developed, smiling. It was one of his best inventions. Self-developing film. The first light struck film-surface, destroyed one chemical, leaving imprints; the second exposure simply hardened, secured the impressions. Quick stuff. Inserting the film-tongue into a micro-viewer in the camera's base, Click handed the whole thing over. "Look." Marnagan put the viewer up against the helmet glass, squinted. "Ah, Click. Now, now. This is one lousy film you invented." "Huh?" "It's a strange process'll develop my picture and ignore the asteroid monsters complete." "What!" Hathaway grabbed the camera, gasped, squinted, and gasped again: Pictures in montage; Marnagan sitting down, chatting conversationally with nothing ; Marnagan shooting his gun at nothing ; Marnagan pretending to be happy in front of nothing . Then, closeup—of—NOTHING! The monsters had failed to image the film. Marnagan was there, his hair like a red banner, his freckled face with the blue eyes bright in it. Maybe— Hathaway said it, loud: "Irish! Irish! I think I see a way out of this mess! Here—" He elucidated it over and over again to the Patrolman. About the film, the beasts, and how the film couldn't be wrong. If the film said the monsters weren't there, they weren't there. "Yeah," said Marnagan. "But step outside this cave—" "If my theory is correct I'll do it, unafraid," said Click. Marnagan scowled. "You sure them beasts don't radiate ultra-violet or infra-red or something that won't come out on film?" "Nuts! Any color we see, the camera sees. We've been fooled." "Hey, where you going?" Marnagan blocked Hathaway as the smaller man tried pushing past him. "Get out of the way," said Hathaway. Marnagan put his big fists on his hips. "If anyone is going anywhere, it'll be me does the going." "I can't let you do that, Irish." "Why not?" "You'd be going on my say-so." "Ain't your say-so good enough for me?" "Yes. Sure. Of course. I guess—" "If you say them animals ain't there, that's all I need. Now, stand aside, you film-developing flea, and let an Irishman settle their bones." He took an unnecessary hitch in trousers that didn't exist except under an inch of porous metal plate. "Your express purpose on this voyage, Hathaway, is taking films to be used by the Patrol later for teaching Junior Patrolmen how to act in tough spots. First-hand education. Poke another spool of film in that contraption and give me profile a scan. This is lesson number seven: Daniel Walks Into The Lion's Den." "Irish, I—" "Shut up and load up." Hathaway nervously loaded the film-slot, raised it. "Ready, Click?" "I—I guess so," said Hathaway. "And remember, think it hard, Irish. Think it hard. There aren't any animals—" "Keep me in focus, lad." "All the way, Irish." "What do they say...? Oh, yeah. Action. Lights. Camera!" Marnagan held his gun out in front of him and still smiling took one, two, three, four steps out into the outside world. The monsters were waiting for him at the fifth step. Marnagan kept walking. Right out into the middle of them.... That was the sweetest shot Hathaway ever took. Marnagan and the monsters! Only now it was only Marnagan. No more monsters. Marnagan smiled a smile broader than his shoulders. "Hey, Click, look at me! I'm in one piece. Why, hell, the damned things turned tail and ran away!" "Ran, hell!" cried Hathaway, rushing out, his face flushed and animated. "They just plain vanished. They were only imaginative figments!" "And to think we let them hole us in that way, Click Hathaway, you coward!" "Smile when you say that, Irish." "Sure, and ain't I always smilin'? Ah, Click boy, are them tears in your sweet grey eyes?" "Damn," swore the photographer, embarrassedly. "Why don't they put window-wipers in these helmets?" "I'll take it up with the Board, lad." "Forget it. I was so blamed glad to see your homely carcass in one hunk, I couldn't help—Look, now, about Gunther. Those animals are part of his set-up. Explorers who land here inadvertently, are chased back into their ships, forced to take off. Tourists and the like. Nothing suspicious about animals. And if the tourists don't leave, the animals kill them." "Shaw, now. Those animals can't kill." "Think not, Mr. Marnagan? As long as we believed in them they could have frightened us to death, forced us, maybe, to commit suicide. If that isn't being dangerous—" The Irishman whistled. "But, we've got to move , Irish. We've got twenty minutes of oxygen. In that time we've got to trace those monsters to their source, Gunther's Base, fight our way in, and get fresh oxy-cannisters." Click attached his camera to his mid-belt. "Gunther probably thinks we're dead by now. Everyone else's been fooled by his playmates; they never had a chance to disbelieve them." "If it hadn't been for you taking them pictures, Click—" "Coupled with your damned stubborn attitude about the accident—" Click stopped and felt his insides turning to water. He shook his head and felt a film slip down over his eyes. He spread his legs out to steady himself, and swayed. "I—I don't think my oxygen is as full as yours. This excitement had me double-breathing and I feel sick." Marnagan's homely face grimaced in sympathy. "Hold tight, Click. The guy that invented these fish-bowls didn't provide for a sick stomach." "Hold tight, hell, let's move. We've got to find where those animals came from! And the only way to do that is to get the animals to come back!" "Come back? How?" "They're waiting, just outside the aura of our thoughts, and if we believe in them again, they'll return." Marnagan didn't like it. "Won't—won't they kill us—if they come—if we believe in 'em?" Hathaway shook a head that was tons heavy and weary. "Not if we believe in them to a certain point . Psychologically they can both be seen and felt. We only want to see them coming at us again." " Do we, now?" "With twenty minutes left, maybe less—" "All right, Click, let's bring 'em back. How do we do it?" Hathaway fought against the mist in his eyes. "Just think—I will see the monsters again. I will see them again and I will not feel them. Think it over and over." Marnagan's hulk stirred uneasily. "And—what if I forget to remember all that? What if I get excited...?" Hathaway didn't answer. But his eyes told the story by just looking at Irish. Marnagan cursed. "All right, lad. Let's have at it!" The monsters returned. A soundless deluge of them, pouring over the rubbled horizon, swarming in malevolent anticipation about the two men. "This way, Irish. They come from this way! There's a focal point, a sending station for these telepathic brutes. Come on!" Hathaway sludged into the pressing tide of color, mouths, contorted faces, silvery fat bodies misting as he plowed through them. Marnagan was making good progress ahead of Hathaway. But he stopped and raised his gun and made quick moves with it. "Click! This one here! It's real!" He fell back and something struck him down. His immense frame slammed against rock, noiselessly. Hathaway darted forward, flung his body over Marnagan's, covered the helmet glass with his hands, shouting: "Marnagan! Get a grip, dammit! It's not real—don't let it force into your mind! It's not real, I tell you!" "Click—" Marnagan's face was a bitter, tortured movement behind glass. "Click—" He was fighting hard. "I—I—sure now. Sure—" He smiled. "It—it's only a shanty fake!" "Keep saying it, Irish. Keep it up." Marnagan's thick lips opened. "It's only a fake," he said. And then, irritated, "Get the hell off me, Hathaway. Let me up to my feet!" Hathaway got up, shakily. The air in his helmet smelled stale, and little bubbles danced in his eyes. "Irish, you forget the monsters. Let me handle them, I know how. They might fool you again, you might forget." Marnagan showed his teeth. "Gah! Let a flea have all the fun? And besides, Click, I like to look at them. They're pretty." The outpour of animals came from a low lying mound a mile farther on. Evidently the telepathic source lay there. They approached it warily. "We'll be taking our chances on guard," hissed Irish. "I'll go ahead, draw their attention, maybe get captured. Then, you show up with your gun...." "I haven't got one." "We'll chance it, then. You stick here until I see what's ahead. They probably got scanners out. Let them see me—" And before Hathaway could object, Marnagan walked off. He walked about five hundred yards, bent down, applied his fingers to something, heaved up, and there was a door opening in the rock. His voice came back across the distance, into Click's earphones. "A door, an air-lock, Click. A tunnel leading down inside!" Then, Marnagan dropped into the tunnel, disappearing. Click heard the thud of his feet hitting the metal flooring. Click sucked in his breath, hard and fast. "All right, put 'em up!" a new harsh voice cried over a different radio. One of Gunther's guards. Three shots sizzled out, and Marnagan bellowed. The strange harsh voice said, "That's better. Don't try and pick that gun up now. Oh, so it's you. I thought Gunther had finished you off. How'd you get past the animals?" Click started running. He switched off his sending audio, kept his receiving on. Marnagan, weaponless. One guard. Click gasped. Things were getting dark. Had to have air. Air. Air. He ran and kept running and listening to Marnagan's lying voice: "I tied them pink elephants of Gunther's in neat alphabetical bundles and stacked them up to dry, ya louse!" Marnagan said. "But, damn you, they killed my partner before he had a chance!" The guard laughed. The air-lock door was still wide open when Click reached it, his head swimming darkly, his lungs crammed with pain-fire and hell-rockets. He let himself down in, quiet and soft. He didn't have a weapon. He didn't have a weapon. Oh, damn, damn! A tunnel curved, ending in light, and two men silhouetted in that yellow glare. Marnagan, backed against a wall, his helmet cracked, air hissing slowly out of it, his face turning blue. And the guard, a proton gun extended stiffly before him, also in a vac-suit. The guard had his profile toward Hathaway, his lips twisting: "I think I'll let you stand right there and die," he said quietly. "That what Gunther wanted, anway. A nice sordid death." Hathaway took three strides, his hands out in front of him. "Don't move!" he snapped. "I've got a weapon stronger than yours. One twitch and I'll blast you and the whole damned wall out from behind you! Freeze!" The guard whirled. He widened his sharp eyes, and reluctantly, dropped his gun to the floor. "Get his gun, Irish." Marnagan made as if to move, crumpled clumsily forward. Hathaway ran in, snatched up the gun, smirked at the guard. "Thanks for posing," he said. "That shot will go down in film history for candid acting." "What!" "Ah: ah! Keep your place. I've got a real gun now. Where's the door leading into the Base?" The guard moved his head sullenly over his left shoulder. Click was afraid he would show his weak dizziness. He needed air. "Okay. Drag Marnagan with you, open the door and we'll have air. Double time! Double!" Ten minutes later, Marnagan and Hathaway, fresh tanks of oxygen on their backs, Marnagan in a fresh bulger and helmet, trussed the guard, hid him in a huge trash receptacle. "Where he belongs," observed Irish tersely. They found themselves in a complete inner world; an asteroid nothing more than a honey-comb fortress sliding through the void unchallenged. Perfect front for a raider who had little equipment and was short-handed of men. Gunther simply waited for specific cargo ships to rocket by, pulled them or knocked them down and swarmed over them for cargo. The animals served simply to insure against suspicion and the swarms of tourists that filled the void these days. Small fry weren't wanted. They were scared off. The telepathic sending station for the animals was a great bank of intricate, glittering machine, through which strips of colored film with images slid into slots and machine mouths that translated them into thought-emanations. A damned neat piece of genius. "So here we are, still not much better off than we were," growled Irish. "We haven't a ship or a space-radio, and more guards'll turn up any moment. You think we could refocus this doohingey, project the monsters inside the asteroid to fool the pirates themselves?" "What good would that do?" Hathaway gnawed his lip. "They wouldn't fool the engineers who created them, you nut." Marnagan exhaled disgustedly. "Ah, if only the U.S. Cavalry would come riding over the hill—" "Irish!" Hathaway snapped that, his face lighting up. "Irish. The U.S. Cavalry it is!" His eyes darted over the machines. "Here. Help me. We'll stage everything on the most colossal raid of the century." Marnagan winced. "You breathing oxygen or whiskey?" "There's only one stipulation I make, Irish. I want a complete picture of Marnagan capturing Raider's Base. I want a picture of Gunther's face when you do it. Snap it, now, we've got rush work to do. How good an actor are you?" "That's a silly question." "You only have to do three things. Walk with your gun out in front of you, firing. That's number one. Number two is to clutch at your heart and fall down dead. Number three is to clutch at your side, fall down and twitch on the ground. Is that clear?" "Clear as the Coal Sack Nebula...." An hour later Hathaway trudged down a passageway that led out into a sort of city street inside the asteroid. There were about six streets, lined with cube houses in yellow metal, ending near Hathaway in a wide, green-lawned Plaza. Hathaway, weaponless, idly carrying his camera in one hand, walked across the Plaza as if he owned it. He was heading for a building that was pretentious enough to be Gunther's quarters. He got halfway there when he felt a gun in his back. He didn't resist. They took him straight ahead to his destination and pushed him into a room where Gunther sat. Hathaway looked at him. "So you're Gunther?" he said, calmly. The pirate was incredibly old, his bulging forehead stood out over sunken, questioningly dark eyes, and his scrawny body was lost in folds of metal-link cloth. He glanced up from a paper-file, surprised. Before he could speak, Hathaway said: "Everything's over with, Mr. Gunther. The Patrol is in the city now and we're capturing your Base. Don't try to fight. We've a thousand men against your eighty-five." Gunther sat there, blinking at Hathaway, not moving. His thin hands twitched in his lap. "You are bluffing," he said, finally, with a firm directness. "A ship hasn't landed here for an hour. Your ship was the last. Two people were on it. The last I saw of them they were being pursued to the death by the Beasts. One of you escaped, it seemed." "Both. The other guy went after the Patrol." "Impossible!" "I can't respect your opinion, Mr. Gunther." A shouting rose from the Plaza. About fifty of Gunther's men, lounging on carved benches during their time-off, stirred to their feet and started yelling. Gunther turned slowly to the huge window in one side of his office. He stared, hard. The Patrol was coming! Across the Plaza, marching quietly and decisively, came the Patrol. Five hundred Patrolmen in one long, incredible line, carrying paralysis guns with them in their tight hands. Gunther babbled like a child, his voice a shrill dagger in the air. "Get out there, you men! Throw them back! We're outnumbered!" Guns flared. But the Patrol came on. Gunther's men didn't run, Hathaway had to credit them on that. They took it, standing. Hathaway chuckled inside, deep. What a sweet, sweet shot this was. His camera whirred, clicked and whirred again. Nobody stopped him from filming it. Everything was too wild, hot and angry. Gunther was throwing a fit, still seated at his desk, unable to move because of his fragile, bony legs and their atrophied state. Some of the Patrol were killed. Hathaway chuckled again as he saw three of the Patrolmen clutch at their hearts, crumple, lie on the ground and twitch. God, what photography! Gunther raged, and swept a small pistol from his linked corselet. He fired wildly until Hathaway hit him over the head with a paper-weight. Then Hathaway took a picture of Gunther slumped at his desk, the chaos taking place immediately outside his window. The pirates broke and fled, those that were left. A mere handful. And out of the chaos came Marnagan's voice, "Here!"
|
C. They work together and are able to coordinate with each other pretty well.
|
Who seems to have the least to hide in the text?
A. Don Cort
B. Jen Jervis
C. Ed Clark
D. Mayor Civek
|
And Then the Town Took Off by RICHARD WILSON ACE BOOKS, INC. 23 West 47th Street, New York 36, N.Y. AND THEN THE TOWN TOOK OFF Copyright ©, 1960, by Ace Books, Inc. All Rights Reserved For Felicitas K. Wilson THE SIOUX SPACEMAN Copyright ©, 1960, by Ace Books, Inc. Printed in U.S.A. THE CITY THAT RAN OFF THE MAP The town of Superior, Ohio, certainly was living up to its name! In what was undoubtedly the most spectacular feat of the century, it simply picked itself up one night and rose two full miles above Earth! Radio messages stated simply that Superior had seceded from Earth. But Don Cort, stranded on that rising town, was beginning to suspect that nothing was simple about Superior except its citizens. Calmly they accepted their rise in the world as being due to one of their local townspeople, a crackpot professor. But after a couple of weeks of floating around, it began to be obvious that the professor had no idea how to get them down. So then it was up to Cort: either find a way to anchor Superior, or spend the rest of his days on the smallest—and the nuttiest—planet in the galaxy! I The town of Superior, Ohio, disappeared on the night of October 31. A truck driver named Pierce Knaubloch was the first to report it. He had been highballing west along Route 202, making up for the time he'd spent over a second cup of coffee in a diner, when he screeched to a stop. If he'd gone another twenty-five feet he'd have gone into the pit where Superior had been. Knaubloch couldn't see the extent of the pit because it was too dark, but it looked big. Bigger than if a nitro truck had blown up, which was his first thought. He backed up two hundred feet, set out flares, then sped off to a telephone. The state police converged on the former site of Superior from several directions. Communicating by radiophone across the vast pit, they confirmed that the town undoubtedly was missing. They put in a call to the National Guard. The guard surrounded the area with troops—more than a thousand were needed—to keep people from falling into the pit. A pilot who flew over it reported that it looked as if a great ice-cream scoop had bitten into the Ohio countryside. The Pennsylvania Railroad complained that one of its passenger trains was missing. The train's schedule called for it to pass through but not stop at Superior at 11:58. That seemed to fix the time of the disappearance at midnight. The truck driver had made his discovery shortly after midnight. Someone pointed out that October 31 was Halloween and that midnight was the witching hour. Somebody else said nonsense, they'd better check for radiation. A civil defense official brought up a Geiger counter, but no matter how he shook it and rapped on it, it refused to click. A National Guard officer volunteered to take a jeep down into the pit, having found a spot that seemed navigable. He was gone a long time but when he came out the other side he reported that the pit was concave, relatively smooth, and did not smell of high explosives. He'd found no people, no houses—no sign of anything except the pit itself. The Governor of Ohio asked Washington whether any unidentified planes had been over the state. Washington said no. The Pentagon and the Atomic Energy Commission denied that they had been conducting secret experiments. Nor had there been any defense plants in Superior that might have blown up. The town's biggest factory made kitchen sinks and the next biggest made bubble gum. A United Airlines pilot found Superior early on the morning of November 1. The pilot, Captain Eric Studley, who had never seen a flying saucer and hoped never to see one, was afraid now that he had. The object loomed out of a cloudbank at twelve thousand feet and Studley changed course to avoid it. He noted with only minimum satisfaction that his co-pilot also saw the thing and wondered why it wasn't moving at the terrific speed flying saucers were allegedly capable of. Then he saw the church steeple on it. A few minutes later he had relayed a message from Superior, formerly of Ohio, addressed to whom it might concern: It said that Superior had seceded from Earth. One other radio message came from Superior, now airborne, on that first day. A ham radio operator reported an unidentified voice as saying plaintively: " Cold up here!" Don Cort had been dozing in what passed for the club car on the Buckeye Cannonball when the train braked to a stop. He looked out the window, hoping this was Columbus, where he planned to catch a plane east. But it wasn't Columbus. All he could see were some lanterns jogging as trainmen hurried along the tracks. The conductor looked into the car. The redhead across the aisle in whom Don had taken a passing interest earlier in the evening asked, "Why did we stop?" "Somebody flagged us down," the conductor said. "We don't make a station stop at Superior on this run." The girl's hair was a subtle red, but false. When Don had entered the club car he'd seen her hatless head from above and noticed that the hair along the part was dark. Her eyes had been on a book and Don had the opportunity for a brief study of her face. The cheeks were full and untouched by make-up. There were lines at the corners of her mouth which indicated a tendency to arrange her expression into one of disapproval. The lips were full, like the cheeks, but it was obvious that the scarlet lipstick had contrived a mouth a trifle bigger than the one nature had given her. Her glance upward at that moment interrupted his examination, which had been about to go on to her figure. Later, though, he was able to observe that it was more than adequate. If the girl had given Don Cort more than that one glance, or if it had been a trained, all-encompassing glance, she would have seen a man in his mid-twenties—about her age—lean, tall and straight-shouldered, with once-blond hair now verging on dark brown, a face neither handsome nor ugly, and a habit of drawing the inside of his left cheek between his teeth and nibbling at it thoughtfully. But it was likely that all she noticed then was the brief case he carried, attached by a chain to a handcuff on his left wrist. "Will we be here long?" Don asked the conductor. He didn't want to miss his plane at Columbus. The sooner he got to Washington, the sooner he'd get rid of the brief case. The handcuff it was attached to was one reason why his interest in the redhead had been only passing. "Can't say," the conductor told him. He let the door close again and went down to the tracks. Don hesitated, shrugged at the redhead, said, "Excuse me," and followed the conductor. About a dozen people were milling around the train as it sat in the dark, hissing steam. Don made his way up to the locomotive and found a bigger knot of people gathered in front of the cowcatcher. Some sort of barricade had been put up across the tracks and it was covered with every imaginable kind of warning device. There were red lanterns, both battery and electric; flashlights; road flares; and even an old red shirt. Don saw two men who must have been the engineer and the fireman talking to an old bearded gentleman wearing a civil defense helmet, a topcoat and riding boots. "You'd go over the edge, I tell you," the old gentleman was saying. "If you don't get this junk off the line," the engineer said, "I'll plow right through it. Off the edge! you crazy or something?" "Look for yourself," the old man in the white helmet said. "Go ahead. Look." The engineer was exasperated. He turned to the fireman. "You look. Humor the old man. Then let's go." The bearded man—he called himself Professor Garet—went off with the fireman. Don followed them. They had tramped a quarter of a mile along the gravel when the fireman stopped. "Okay," he said "where's the edge? I don't see nothing." The tracks seemed to stretch forever into the darkness. "It's another half mile or so," the professor said. "Well, let's hurry up. We haven't got all night." The old man chuckled. "I'm afraid you have." They came to it at last, stopping well back from it. Professor Garet swelled with pride, it seemed, as he made a theatrical gesture. "Behold," he said. "Something even Columbus couldn't find. The edge of the world." True, everything seemed to stop, and they could see stars shining low on the horizon where stars could not properly be expected to be seen. Don Cort and the fireman walked cautiously toward the edge while the professor ambled ahead with the familiarity of one who had been there before. But there was a wind and they did not venture too close. Nevertheless, Don could see that it apparently was a neat, sharp edge, not one of your old ragged, random edges such as might have been caused by an explosion. This one had the feeling of design behind it. Standing on tiptoe and repressing a touch of giddiness, Don looked over the edge. He didn't have to stand on tiptoe any more than he had to sit on the edge of his seat during the exciting part of a movie, but the situation seemed to call for it. Over the edge could be seen a big section of Ohio. At least he supposed it was Ohio. Don looked at the fireman, who had an unbelieving expression on his face, then at the bearded old man, who was smiling and nodding. "You see what I mean," he said. "You would have gone right over. I believe you would have had a two-mile fall." "Of course you could have stayed aboard the train," the man driving the old Pontiac said, "but I really think you'll be more comfortable at Cavalier." Don Cort, sitting in the back seat of the car with the redhead from the club car, asked, "Cavalier?" "The college. The institute, really; it's not accredited. What did you say your name was, miss?" "Jen Jervis," she said. "Geneva Jervis, formally." "Miss Jervis. I'm Civek. You know Mr. Cort, I suppose." The girl smiled sideways. "We have a nodding acquaintance." Don nodded and grinned. "There's plenty of room in the dormitories," Civek said. "People don't exactly pound on the gates and scream to be admitted to Cavalier." "Are you connected with the college?" Don asked. "Me? No. I'm the mayor of Superior. The old town's really come up in the world, hasn't it?" "Overnight," Geneva Jervis said. "If what Mr. Cort and the fireman say is true. I haven't seen the edge myself." "You'll have a better chance to look at it in the morning," the mayor said, "if we don't settle back in the meantime." "Was there any sort of explosion?" Don asked. "No. There wasn't any sensation at all, as far as I noticed. I was watching the late show—or trying to. My house is down in a hollow and reception isn't very good, especially with old English movies. Well, all of a sudden the picture sharpened up and I could see just as plain. Then the phone rang and it was Professor Garet." "The old fellow with the whiskers and the riding boots?" Jen Jervis asked. "Yes. Osbert Garet, Professor of Magnology at the Cavalier Institute of Applied Sciences." "Professor of what?" "Magnology. As I say, the school isn't accredited. Well, Professor Garet telephoned and said, 'Hector'—that's my name, Hector Civek—'everything's up in the air.' He was having his little joke, of course. I said, 'What?' and then he told me." "Told you what?" Jen Jervis asked. "I mean, does he have any theory about it?" "He has a theory about everything. I think what he was trying to convey was that this—this levitation confirmed his magnology principle." "What's that?" Don asked. "I haven't the faintest idea. I'm a politician, not a scientist. Professor Garet went on about it for a while, on the telephone, about magnetism and gravity, but I think he was only calling as a courtesy, so the mayor wouldn't look foolish the next morning, not knowing his town had flown the coop." "What's the population of Superior?" "Three thousand, including the students at the institute. Three thousand and forty, counting you people from the train. I guess you'll be with us for a while." "What do you mean by that?" Jen Jervis asked. "Well, I don't see how you can get down. Do you?" "Does Superior have an airport?" Don asked. "I've got to get back to—to Earth." It sounded odd to put it that way. "Nope," Civek said. "No airport. No place for a plane to land, either." "Maybe not a plane," Don said, "but a helicopter could land just about anywhere." "No helicopters here, either." "Maybe not. But I'll bet they're swarming all over you by morning." "Hm," said Hector Civek. Don couldn't quite catch his expression in the rearview mirror. "I suppose they could, at that. Well, here's Cavalier. You go right in that door, where the others are going. There's Professor Garet. I've got to see him—excuse me." The mayor was off across the campus. Don looked at Geneva Jervis, who was frowning. "Are you thinking," he asked, "that Mayor Civek was perhaps just a little less than completely honest with us?" "I'm thinking," she said, "that I should have stayed with Aunt Hattie another night, then taken a plane to Washington." "Washington?" Don said. "That's where I'm going. I mean where I was going before Superior became airborne. What do you do in Washington, Miss Jervis?" "I work for the Government. Doesn't everybody?" "Not everybody. Me, for instance." "No?" she said. "Judging by that satchel you're handcuffed to, I'd have thought you were a courier for the Pentagon. Or maybe State." He laughed quickly and loudly because she was getting uncomfortably close. "Oh, no. Nothing so glamorous. I'm a messenger for the Riggs National Bank, that's all. Where do you work?" "I'm with Senator Bobby Thebold, S.O.B." Don laughed again. "He sure is." " Mister Cort!" she said, annoyed. "You know as well as I do that S.O.B. stands for Senate Office Building. I'm his secretary." "I'm sorry. We'd better get out and find a place to sleep. It's getting late." " Places to sleep," she corrected. She looked angry. "Of course," Don said, puzzled by her emphasis. "Come on. Where they put you, you'll probably be surrounded by co-eds, even if I could get out of this cuff." He took her bag in his free hand and they were met by a gray-haired woman who introduced herself as Mrs. Garet. "We'll try to make you comfortable," she said. "What a night, eh? The professor is simply beside himself. We haven't had so much excitement since the cosmolineator blew up." They had a glimpse of the professor, still in his CD helmet, going around a corner, gesticulating wildly to someone wearing a white laboratory smock. II Don Cort had slept, but not well. He had tried to fold the brief case to pull it through his sleeve so he could take his coat off, but whatever was inside the brief case was too big. Cavalier had given him a room to himself at one end of a dormitory and he'd taken his pants off but had had to sleep with his coat and shirt on. He got up, feeling gritty, and did what little dressing was necessary. It was eight o'clock, according to the watch on the unhandcuffed wrist, and things were going on. He had a view of the campus from his window. A bright sun shone on young people moving generally toward a squat building, and other people going in random directions. The first were students going to breakfast, he supposed, and the others were faculty members. The air was very clear and the long morning shadows distinct. Only then did he remember completely that he and the whole town of Superior were up in the air. He went through the dormitory. A few students were still sleeping. The others had gone from their unmade beds. He shivered as he stepped outdoors. It was crisp, if not freezing, and his breath came out visibly. First he'd eat, he decided, so he'd be strong enough to go take a good look over the edge, in broad daylight, to the Earth below. The mess hall, or whatever they called it, was cafeteria style and he got in line with a tray for juice, eggs and coffee. He saw no one he knew, but as he was looking for a table a willowy blonde girl smiled and gestured to the empty place opposite her. "You're Mr. Cort," she said. "Won't you join me?" "Thanks," he said, unloading his tray. "How did you know?" "The mystery man with the handcuff. You'd be hard to miss. I'm Alis—that's A-l-i-s, not A-l-i-c-e—Garet. Are you with the FBI? Or did you escape from jail?" "How do you do. No, just a bank messenger. What an unusual name. Professor Garet's daughter?" "The same," she said. "Also the only. A pity, because if there'd been two of us I'd have had a fifty-fifty chance of going to OSU. As it is, I'm duty-bound to represent the second generation at the nut factory." "Nut factory? You mean Cavalier?" Don struggled to manipulate knife and fork without knocking things off the table with his clinging brief case. "Here, let me cut your eggs for you," Alis said. "You'd better order them scrambled tomorrow. Yes, Cavalier. Home of the crackpot theory and the latter-day alchemist." "I'm sure it's not that bad. Thanks. As for tomorrow, I hope to be out of here by then." "How do you get down from an elephant? Old riddle. You don't; you get down from ducks. How do you plan to get down from Superior?" "I'll find a way. I'm more interested at the moment in how I got up here." "You were levitated, like everybody else." "You make it sound deliberate, Miss Garet, as if somebody hoisted a whole patch of real estate for some fell purpose." "Scarcely fell , Mr. Cort. As for it being deliberate, that seems to be a matter of opinion. Apparently you haven't seen the papers." "I didn't know there were any." "Actually there's only one, the Superior Sentry , a weekly. This is an extra. Ed Clark must have been up all night getting it out." She opened her purse and unfolded a four-page tabloid. Don blinked at the headline: Town Gets High "Ed Clark's something of an eccentric, like everybody else in Superior," Alis said. Don read the story, which seemed to him a capricious treatment of an apparently grave situation. Residents having business beyond the outskirts of town today are advised not to. It's a long way down. Where Superior was surrounded by Ohio, as usual, today Superior ends literally at the town line. A Citizens' Emergency Fence-Building Committee is being formed, but in the meantime all are warned to stay well away from the edge. The law of gravity seems to have been repealed for the town but it is doubtful if the same exemption would apply to a dubious individual bent on investigating.... Don skimmed the rest. "I don't see anything about it being deliberate." Alis had been creaming and sugaring Don's coffee. She pushed it across to him and said, "It's not on page one. Ed Clark and Mayor Civek don't get along, so you'll find the mayor's statement in a box on page three, bottom." Don creased the paper the other way, took a sip of coffee, nodded his thanks, and read: Mayor Claims Secession From Earth Mayor Hector Civek, in a proclamation issued locally by hand and dropped to the rest of the world in a plastic shatter-proof bottle, said today that Superior has seceded from Earth. His reasons were as vague as his explanation. The "reasons" include these: (1) Superior has been discriminated against by county, state and federal agencies; (2) Cavalier Institute has been held up to global derision by orthodox (presumably meaning accredited) colleges and universities; and (3) chicle exporters have conspired against the Superior Bubble Gum Company by unreasonably raising prices. The "explanation" consists of a 63-page treatise on applied magnology by Professor Osbert Garet of Cavalier which the editor (a) does not understand; (b) lacks space to publish; and which (it being atrociously handwritten) he (c) has not the temerity to ask his linotype operator to set. Don said, "I'm beginning to like this Ed Clark." "He's a doll," Alis said. "He's about the only one in town who stands up to Father." "Does your father claim that he levitated Superior off the face of the Earth?" "Not to me he doesn't. I'm one of those banes of his existence, a skeptic. He gave up trying to magnolize me when I was sixteen. I had a science teacher in high school—not in Superior, incidentally—who gave me all kinds of embarrassing questions to ask Father. I asked them, being a natural-born needler, and Father has disowned me intellectually ever since." "How old are you, Miss Garet, if I may ask?" She sat up straight and tucked her sweater tightly into her skirt, emphasizing her good figure. To a male friend Don would have described the figure as outstanding. She had mocking eyes, a pert nose and a mouth of such moist red softness that it seemed perpetually waiting to be kissed. All in all she could have been the queen of a campus much more densely populated with co-eds than Cavalier was. "You may call me Alis," she said. "And I'm nineteen." Don grinned. "Going on?" "Three months past. How old are you , Mr. Cort?" "Don's the name I've had for twenty-six years. Please use it." "Gladly. And now, Don, unless you want another cup of coffee, I'll go with you to the end of the world." "On such short notice?" Don was intrigued. Last night the redhead from the club car had repelled an advance that hadn't been made, and this morning a blonde was apparently making an advance that hadn't been solicited. He wondered where Geneva Jervis was, but only vaguely. "I'll admit to the double entendre ," Alis said. "What I meant—for now—was that we can stroll out to where Superior used to be attached to the rest of Ohio and see how the Earth is getting along without us." "Delighted. But don't you have any classes?" "Sure I do. Non-Einsteinian Relativity 1, at nine o'clock. But I'm a demon class-cutter, which is why I'm still a Senior at my advanced age. On to the brink!" They walked south from the campus and came to the railroad track. The train was standing there with nowhere to go. It had been abandoned except for the conductor, who had dutifully spent the night aboard. "What's happening?" he asked when he saw them. "Any word from down there?" "Not that I know of," Don said. He introduced him to Alis Garet. "What are you going to do?" "What can I do?" the conductor asked. "You can go over to Cavalier and have breakfast," Alis said. "Nobody's going to steal your old train." The conductor reckoned as how he might just do that, and did. "You know," Don said, "I was half-asleep last night but before the train stopped I thought it was running alongside a creek for a while." "South Creek," Alis said. "That's right. It's just over there." "Is it still? I mean hasn't it all poured off the edge by now? Was that Superior's water supply?" Alis shrugged. "All I know is you turn on the faucet and there's water. Let's go look at the creek." They found it coursing along between the banks. "Looks just about the same," she said. "That's funny. Come on; let's follow it to the edge." The brink, as Alis called it, looked even more awesome by daylight. Everything stopped short. There were the remnants of a cornfield, with the withered stalks cut down, then there was nothing. There was South Creek surging along, then nothing. In the distance a clump of trees, with a few autumn leaves still clinging to their branches, simply ended. "Where is the water going?" Don asked. "I can't make it out." "Down, I'd say. Rain for the Earth-people." "I should think it'd be all dried up by now. I'm going to have a look." "Don't! You'll fall off!" "I'll be careful." He walked cautiously toward the edge. Alis followed him, a few feet behind. He stopped a yard from the brink and waited for a spell of dizziness to pass. The Earth was spread out like a topographer's map, far below. Don took another wary step, then sat down. "Chicken," said Alis. She laughed uncertainly, then she sat down, too. "I still can't see where the water goes," Don said. He stretched out on his stomach and began to inch forward. "You stay there." Finally he had inched to a point where, by stretching out a hand, he could almost reach the edge. He gave another wriggle and the fingers of his right hand closed over the brink. For a moment he lay there, panting, head pressed to the ground. "How do you feel?" Alis asked. "Scared. When I get my courage back I'll pick up my head and look." Alis put a hand out tentatively, then purposefully took hold of his ankle and held it tight. "Just in case a high wind comes along," she said. "Thanks. It helps. Okay, here we go." He lifted his head. "Damn." "What?" "It still isn't clear. Do you have a pocket mirror?" "I have a compact." She took it out of her bag with her free hand and tossed it to him. It rolled and Don had to grab to keep it from going over the edge. Alis gave a little shriek. Don was momentarily unnerved and had to put his head back on the ground. "Sorry," she said. Don opened the compact and carefully transferred it to his right hand. He held it out beyond the edge and peered into it, focusing it on the end of the creek. "Now I've got it. The water isn't going off the edge!" "It isn't? Then where is it going?" "Down, of course, but it's as if it's going into a well, or a vertical tunnel, just short of the edge." "Why? How?" "I can't see too well, but that's my impression. Hold on now. I'm coming back." He inched away from the edge, then got up and brushed himself off. He returned her compact. "I guess you know where we go next." "The other end of the creek?" "Exactly." South Creek did not bisect Superior, as Don thought it might, but flowed in an arc through a southern segment of it. They had about two miles to go, past South Creek Bridge—which used to lead to Ladenburg, Alis said—past Raleigh Country Club (a long drive would really put the ball out of play, Don thought) and on to the edge again. But as they approached what they were forced to consider the source of the creek, they found a wire fence at the spot. "This is new," Alis said. The fence, which had a sign on it, warning—electrified , was semicircular, with each end at the edge and tarpaulins strung behind it so they could see the mouth of the creek. The water flowed from under the tarp and fence. "Look how it comes in spurts," Alis said. "As if it's being pumped." Smaller print on the sign said: Protecting mouth of South Creek, one of two sources of water for Superior. Electrical charge in fence is sufficient to kill. It was signed: Vincent Grande, Chief of Police, Hector Civek, Mayor . "What's the other source, besides the faucet in your bathroom?" Don asked. "North Lake, maybe," Alis said. "People fish there but nobody's allowed to swim." "Is the lake entirely within the town limits?" "I don't know." "If it were on the edge, and if I took a rowboat out on it, I wonder what would happen?" "I know one thing—I wouldn't be there holding your ankle while you found out." She took his arm as they gazed past the electrified fence at the Earth below and to the west. "It's impressive, isn't it?" she said. "I wonder if that's Indiana way over there?" He patted her hand absent-mindedly. "I wonder if it's west at all. I mean, how do we know Superior is maintaining the same position up here as it used to down there?" "We could tell by the sun, silly." "Of course," he said, grinning at his stupidity. "And I guess we're not high enough to see very far. If we were we'd be able to see the Great Lakes—or Lake Erie, anyway." They were musing about the geography when a plane came out of a cloudbank and, a second later, veered sharply. They could make out UAL on the underside of a wing. As it turned they imagined they could see faces peering out of the windows. They waved and thought they saw one or two people wave back. Then the plane climbed toward the east and was gone. "Well," Don said as they turned to go back to Cavalier, "now we know that they know. Maybe we'll begin to get some answers. Or, if not answers, then transportation." "Transportation?" Alis squeezed the arm she was holding. "Why? Don't you like it here?" "If you mean don't I like you, the answer is yes, of course I do. But if I don't get out of this handcuff soon so I can take a bath and get into clean clothes, you're not going to like me." "You're still quite acceptable, if a bit whiskery." She stopped, still holding his arm, and he turned so they were face to face. "So kiss me," she said, "before you deteriorate." They were in the midst of an extremely pleasant kiss when the brief case at the end of Don's handcuff began to talk to him.
|
C. Ed Clark
|
What is the purpose of the Orthan taking over a human host?
A. To get the full human experience, and understand what makes the planet worthwhile.
B. To investigate the planet without vslling attention, and determine if it's worth colonizing.
C. To assimilate the human host into the Hord, and add to it their knowledge.
D. To examine the memories of the human host, and see what knowledge they have.
|
QUEST OF THIG By BASIL WELLS Thig of Ortha was the vanguard of the conquering "HORDE." He had blasted across trackless space to subdue a defenseless world—only to meet on Earth emotions that were more deadly than weapons. [Transcriber's Note: This etext was produced from Planet Stories Fall 1942. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] Thig carefully smoothed the dark sand and seaweed of the lonely beach over the metal lid of the flexible ringed tunnel that linked the grubby ship from another planet with the upper air. He looked out across the heaving waters of the Sound toward Connecticut. He stared appraisingly around at the luxuriant green growth of foliage further inland; and started toward the little stretch of trees and brush, walking carefully because of the lesser gravitation. Thig was shorter than the average Earthman—although on Ortha he was well above the average in height—but his body was thick and powerfully muscled. His skull was well-shaped and large; his features were regular, perhaps a trifle oversize, and his hair and eyes were a curiously matching blend of reddish brown. Oddest of all, he wore no garments, other than the necessary belt and straps to support his rod-like weapon of white metal and his pouches for food and specimens. The Orthan entered the narrow strip of trees and crossed to the little-used highway on the other side. Here he patiently sat down to wait for an Earthman or an Earthwoman to pass. His task now was to bring a native, intact if possible, back to the carefully buried space cruiser where his two fellows and himself would drain the creature's mentality of all its knowledge. In this way they could learn whether a planet was suited for colonization by later swarms of Orthans. Already they had charted over a hundred celestial bodies but of them all only three had proven worthy of consideration. This latest planet, however, 72-P-3 on the chart, appeared to be an ideal world in every respect. Sunlight, plenty of water and a dense atmospheric envelope made of 72-P-3 a paradise among planets. The explorer from another world crouched into the concealment of a leafy shrub. A creature was approaching. Its squat body was covered with baggy strips of bluish cloth and it carried a jointed rod of metal and wood in its paw. It walked upright as did the men of Ortha. Thig's cold eyes opened a trifle wider as he stared into the thing's stupid face. It was as though he was looking into a bit of polished metal at the reflection of himself! The Earthman was opposite now and he must waste no more precious time. The mighty muscles of the Orthan sent him hurtling across the intervening space in two prodigious bounds, and his hands clamped across the mouth and neck of the stranger.... Lewis Terry was going fishing. For a week the typewriter mill that had ground out a thousand assorted yarns of the untamed West and the frigid desolation of the Northwoods had been silent. Lewis wondered if he was going stale. He had sat every day for eight hours in front of that shiny-buttoned bane of the typist, but there were no results. Feebly he had punched a key two days ago and a $ sign had appeared. He hadn't dared touch the machine since. For Mr. Terry, that hard-hitting writer of two-gun action, had never been further west of Long Island than Elizabeth, and he had promised his wife, Ellen, that he would take the three children and herself on a trailer tour of the West that very summer. Since that promise, he could not write a word. Visions of whooping red-skinned Apaches and be-chapped outlaws raiding his little trailer home kept rolling up out of his subconscious. Yet he had to write at least three novelets and a fistful of short stories in the next two weeks to finance the great adventure—or the trip was off. So Lewis left the weathered old cottage in the early dawn and headed for his tubby old boat at the landing in an attempt to work out a salable yarn.... "Hey!" he shouted as a naked man sprang out of the bushes beside the road. "What's the trouble?" Then he had no time for further speech, the massive arms of the stranger had wound around him and two hamlike hands shut off his speech and his wind. He fought futilely against trained muscles. The hand clamping his throat relaxed for a moment and hacked along the side of his head. Blackness flooded the brain of Lewis, and he knew no more. "There it is," announced Thig, dropping the limp body of the captured Earthman to the metal deck-plates. "It is a male of the species that must have built the cities we saw as we landed." "He resembles Thig," announced Kam. "But for the strange covering he wears he might be Thig." "Thig will be this creature!" announced Torp. "With a psychic relay we will transfer the Earthman's memories and meager store of knowledge to the brain of Thig! He can then go out and scout this world without arousing suspicion. While he is gone, I will take Kam and explore the two inner planets." "You are the commander," said Thig. "But I wish this beast did not wear these clumsy sheathing upon his body. On Ortha we do not hamper the use of our limbs so." "Do not question the word of your commander," growled Torp, swelling out his thick chest menacingly. "It is for the good of our people that you disguise yourself as an Earthman." "For the good of the Horde," Thig intoned almost piously as he lifted Terry's body and headed for the laboratory. Service for the Horde was all that the men of Ortha knew. Carefully cultured and brought to life in the laboratories of their Horde, they knew neither father nor mother. Affection and love were entirely lacking in their early training and later life. They were trained antlike from childhood that only the growth and power of the Horde were of any moment. Men and women alike toiled and died like unfeeling robots of flesh and bone for the Horde. The Horde was their religion, their love-life, their everything! So it was that the bodies of the Earthman and the Orthan were strapped on two parallel tables of chill metal and the twin helmets, linked to one another by the intricacies of the psychic relay, put upon their heads. For ten hours or more the droning hum of the relay sucked Terry's brain dry of knowledge. The shock upon the nervous system of the Earthman proved too violent and his heart faltered after a time and stopped completely. Twice, with subtle drugs they restored pseudo-life to his body and kept the electrical impulses throbbing from his tortured brain, but after the third suspension of life Thig removed his helmet. "There is nothing more to learn," he informed his impassive comrades. "Now, let us get on with the plastic surgery that is required. My new body must return to its barbaric household before undue attention is aroused. And when I return I will take along some of the gleaming baubles we found on the red planet—these people value them highly." An hour later, his scars and altered cartilage already healed and painless, Thig again scraped sand over the entrance to the space ship and set out along the moonlit beach toward the nearest path running inland to his home. Memory was laying the country bare about him, Terry's own childhood memories of this particular section of Long Island. Here was the place where Jake and Ted had helped him dig for the buried treasure that old 'Notch-ear' Beggs had told them so exactly about. Remembrance of that episode gave Thig an idea about the little lump of jewels in his pocket. He had found them in a chest along the beach! He was coming up on the porch now and at the sound of his foot on the sagging boards the screen door burst open and three little Earth-creatures were hugging at his legs. An odd sensation, that his acquired memories labeled as pleasure, sent a warm glow upward from around his heart. Then he saw the slender red-haired shape of a woman, the mate of the dead man he knew, and confusion struck his well-trained brain. Men had no mates on Ortha, sex had been overthrown with all the other primitive impulses of barbarism; so he was incapable of understanding the emotions that swept through his acquired memory. Unsteadily he took her in his arms and felt her warm lips pressed, trembling, against his own. That same hot wave of pulsing blood choked achingly up into his throat. "Lew, dear," Ellen was asking, "where have you been all day? I called up at the landing but you were not there. I wanted to let you know that Saddlebag Publications sent a check for $50 for "Reversed Revolvers" and three other editors asked for shorts soon." "Shoulda got a hundred bucks for that yarn," grunted Thig, and gasped. For the moment he had been Lewis Terry and not Thig! So thoroughly had he acquired the knowledge of Terry that he found himself unconsciously adopting the thinking and mannerism of the other. All the better this way, he realized—more natural. "Sorry I was late," he said, digging into his pocket for the glittering baubles, "but I was poking around on the beach where we used to hunt treasure and I found an old chest. Inside it I found nothing but a handful of these." He flashed the jewels in front of Ellen's startled eyes and she clung, unbelieving, to his arm. "Why, Lew," she gasped, "they're worth a fortune! We can buy that new trailer now and have a rebuilt motor in the car. We can go west right away.... Hollywood, the Grand Canyon, cowboys!" "Uh huh," agreed the pseudo Lewis, memories of the ferocious savages and gunmen of his stories rendering him acutely unhappy. Sincerely he hoped that the west had reformed. "I saved some kraut and weiners," Ellen said. "Get washed up while I'm warming them up. Kids ate all the bread so I had to borrow some from the Eskoes. Want coffee, too?" "Mmmmmm," came from the depths of the chipped white wash-basin. "Home again," whispered Ellen as she stood beside Thig twelve weeks later and gazed tearfully at the weathered little gray house. She knelt beside the front stoop and reached for the key hidden beneath it. "The west was wonderful; tremendous, vast and beautiful," she went on as they climbed the steps, "but nowhere was there any place as beautiful as our own little strip of sky and water." Thig sank into a dusty old swing that hung on creaking chains from the exposed rafters of the porch roof. He looked down at the dusty gray car and the bulbous silvery bulk of the trailer that had been their living quarters for almost three months. Strange thoughts were afloat in the chaos of his cool Orthan brain. Tonight or tomorrow night at the latest he must contact his two fellows and report that Earth was a planetary paradise. No other world, including Ortha, was so well-favored and rich. An expeditionary force to wipe the grotesque civilizations of Earth out of existence would, of course, be necessary before the first units of new Hordes could be landed. And there Thig balked. Why must they destroy these people, imperfect though their civilization might be, to make room for the Hordes? Thig tried to tell himself that it was the transmitted thoughts of the dead Earthman that made him feel so, but he was not too sure. For three months he had lived with people who loved, hated, wept and sacrificed for reasons that he had never known existed. He had learned the heady glory of thinking for himself and making his own decisions. He had experienced the primitive joy of matching his wits and tongue against the wits of other unpredictable human beings. There was no abrupt division of men and women into definite classes of endeavor. A laborer thought the same thoughts that a governor might think. Uncertainty added zest to every day's life. The Orthan had come to question the sole devotion of the individual to the Horde to the exclusion of all other interests. What, he wondered, would one new world—or a hundred—populated by the Hordes add to the progress of humanity? For a hundred thousand years the Orthan civilization had remained static, its energies directed into certain well-defined channels. They were mindless bees maintaining their vast mechanical hives. There was that moment on the brink of the Grand Canyon when Ellen had caught his arm breathlessly at all the beauty spread away there beneath them. There were mornings in the desert when the sun painted in lurid red the peaks above the harsh black-and-whites of the sagebrush and cactus slopes. There was the little boy, his body burning with fever, who nestled trustingly against his tense man's body and slept—the son of Ellen and the man he had destroyed. Thig groaned. He was a weakling to let sentimentality so get the better of his judgment. He would go now to the space ship and urge them to blast off for Ortha. He sprang off the porch and strode away down the road toward the beach. The children ran to him; wanted to go along. He sent them away harshly but they smiled and waved their brown little hands. Ellen came to the door and called after him. "Hurry home, dear," she said. "I'll have a bite ready in about an hour." He dared not say anything, for his voice would have broken and she would have known something was wrong. She was a very wise sort of person when something was troubling him. He waved his stubby paw of a hand to show that he had heard, and blindly hurried toward the Sound. Oddly enough, as he hurried away along the narrow path through the autumn woods, his mind busied itself with a new epic of the west that lived no longer. He mentally titled it: "Rustlers' Riot" and blocked in the outlines of his plot. One section of his brain was that of the careless author of gunslinging yarns, a section that seemed to be sapping the life from his own brain. He knew that the story would never be written, but he toyed with the idea. So far had Thig the emotionless, robot-being from Ortha drifted from the unquestioning worship of the Horde! "You have done well," announced Torp when Thig had completed his report on the resources and temperatures of various sections of Terra. "We now have located three worlds fit for colonization and so we will return to Ortha at once. "I will recommend the conquest of this planet, 72-P-3 at once and the complete destruction of all biped life upon it. The mental aberrations of the barbaric natives might lead to endless complications if they were permitted to exist outside our ordered way of life. I imagine that three circuits of the planet about its primary should prove sufficient for the purposes of complete liquidation." "But why," asked Thig slowly, "could we not disarm all the natives and exile them on one of the less desirable continents, Antarctica for example or Siberia? They are primitive humans even as our race was once a race of primitives. It is not our duty to help to attain our own degree of knowledge and comfort?" "Only the good of the Horde matters!" shouted Torp angrily. "Shall a race of feeble-witted beasts, such as these Earthmen, stand in the way of a superior race? We want their world, and so we will take it. The Law of the Horde states that all the universe is ours for the taking." "Let us get back to Ortha at once, then," gritted out Thig savagely. "Never again do I wish to set foot upon the soil of this mad planet. There are forces at work upon Earth that we of Ortha have long forgotten." "Check the blood of Thig for disease, Kam," ordered Torp shortly. "His words are highly irrational. Some form of fever perhaps native to this world. While you examine him I will blast off for Ortha." Thig followed Kam into the tiny laboratory and found a seat beside the squat scientist's desk. His eyes roamed over the familiar instruments and gauges, each in its own precise position in the cases along the walls. His gaze lingered longest on the stubby black ugliness of a decomposition blaster in its rack close to the deck. A blast of the invisible radiations from that weapon's hot throat and flesh or vegetable fiber rotted into flaky ashes. The ship trembled beneath their feet; it tore free from the feeble clutch of the sand about it, and they were rocketing skyward. Thig's broad fingers bit deep into the unyielding metal of his chair. Suddenly he knew that he must go back to Earth, back to Ellen and the children of the man he had helped destroy. He loved Ellen, and nothing must stand between them! The Hordes of Ortha must find some other world, an empty world—this planet was not for them. "Turn back!" he cried wildly. "I must go back to Earth. There is a woman there, helpless and alone, who needs me! The Horde does not need this planet." Kam eyed him coldly and lifted a shining hypodermic syringe from its case. He approached Thig warily, aware that disease often made a maniac of the finest members of the Horde. "No human being is more important than the Horde," he stated baldly. "This woman of whom you speak is merely one unit of the millions we must eliminate for the good of the Horde." Then it was that Thig went berserk. His fists slashed into the thick jaw of the scientist and his fingers ripped at the hard cords overlying the Orthan's vital throat tubes. His fingers and thumb gouged deep into Kam's startled throat and choked off any cry for assistance before it could be uttered. Kam's hand swept down to the holster swung from his intricate harness and dragged his blaster from it. Thig's other hand clamped over his and for long moments they swayed there, locked together in silent deadly struggle. The fate of a world hung in the balance as Kam's other hand fought against that lone arm of Thig. The scales swung in favor of Kam. Slowly the flaring snout of his weapon tilted upward until it reached the level of Thig's waist. Thig suddenly released his grip and dragged his enemy toward him. A sudden reversal of pressure on Kam's gun hand sent the weapon swivelling about full upon its owner's thick torso. Thig's fingers pressed down upon Kam's button finger, down upon the stud set into the grip of the decomposition blaster, and Kam's muscles turned to water. He shrieked. Before Thig's eyes half of his comrade's body sloughed away into foul corruption that swiftly gave way to hardened blobs of dessicated matter. Horror for what he had done—that he had slain one of his own Horde—made his limbs move woodenly. All of his thoughts were dulled for the moment. Painfully slow, he turned his body around toward the control blister, turned around on leaden feet, to look full into the narrowed icy eyes of his commander. He saw the heavy barrel of the blaster slashing down against his skull but he could not swing a fraction of an inch out of the way. His body seemed paralyzed. This was the end, he thought as he waited stupidly for the blow to fall, the end for Ellen and the kids and all the struggling races of Earth. He would never write another cowboy yarn—they would all be dead anyhow soon. Then a thunderclap exploded against his head and he dropped endlessly toward the deck. Blows rained against his skull. He wondered if Torp would ever cease to hammer at him and turn the deadly ray of the weapon upon him. Blood throbbed and pounded with every blow.... Bam, Bam, Bam, the blood pounded in his ears. Like repeated blows of a hammer they shook his booming head. No longer was Torp above him. He was in the corner of the laboratory, a crumpled blood-smeared heap of bruised flesh and bone. He was unfettered and the blood was caked upon his skull and in his matted hair. Torp must have thought he had killed him with those savage blows upon the head. Even Torp, thought Thig ruefully, gave way to the primitive rage of his ancestors at times; but to that very bit of unconscious atavism he now owed his life. A cool-headed robot of an Orthan would have efficiently used the blaster to destroy any possibility of remaining life in his unconscious body. Thig rolled slowly over so that his eye found the door into the control room. Torp would be coming back again to dispose of their bodies through the refuse lock. Already the body of Kam was gone. He wondered why he had been left until last. Perhaps Torp wished to take cultures of his blood and tissues to determine whether a disease was responsible for his sudden madness. The cases of fragile instruments were just above his head. Association of memories brought him the flash of the heavy blaster in its rack beneath them. His hand went up and felt the welcome hardness of the weapon. He tugged it free. In a moment he was on his knees crawling across the plates of the deck toward the door. Halfway across the floor he collapsed on his face, the metal of the gun making a harsh clang. He heard the feet of Torp scuffle out of silence and a choked cry in the man's throat squalled out into a senseless whinny. Thig raised himself up on a quivering elbow and slid the black length of the blaster in front of him. His eyes sought the doorway and stared full into the glaring vacant orbs of his commander. Torp leaned there watching him, his breath gurgling brokenly through his deep-bitten lips. The clawing marks of nails, fingernails, furrowed his face and chest. He was a madman! The deadly attack of Thig; his own violent avenging of Kam's death, and now the apparent return of the man he had killed come to life had all served to jolt his rigidly trained brain from its accustomed groove. The shock had been too much for the established thought-processes of the Orthan. So Thig shot him where he stood, mercifully, before that vacant mad stare set him, too, to gibbering and shrieking. Then he stepped over the skeleton-thing that had been Torp, using the new strength that victory had given him to drive him along. He had saved a world's civilization from extinction! The thought sobered him; yet, somehow, he was pleased that he had done so. After all, it had been the Earthwoman and the children he had been thinking of while he battled Kam, a selfish desire to protect them all. He went to the desk where Torp had been writing in the ship's log and read the last few nervously scrawled lines: Planet 72-P-3 unfit for colonization. Some pernicious disease that strikes at the brain centers and causes violent insanity is existent there. Thig, just returned from a survey of the planet, went mad and destroyed Kam. In turn I was forced to slay him. But it is not ended. Already I feel the insidious virus of.... And there his writing ended abruptly. Thig nodded. That would do it. He set the automatic pilot for the planet Ortha. Unless a rogue asteroid or a comet crossed the ship's path she would return safely to Ortha with that mute warning of danger on 72-P-3. The body of Torp would help to confirm his final message. Then Thig crossed the cabin to the auxiliary life boat there, one of a half-dozen space ships in miniature nested within the great ship's hull, and cut free from the mother vessel. He flipped the drive lever, felt the thrumming of the rockets driving him from the parent ship. The sensation of free flight against his new body was strangely exhilerating and heady. It was the newest of the emotions he had experienced on Earth since that day, so many months before, when he had felt the warmness of Ellen's lips tight against his. Thig flipped the drive lever, felt the thrumming of the rockets driving him from the parent ship. He swung about to the port, watched the flaming drive-rockets of the great exploratory ship hurl it toward far-away Ortha, and there was no regret in his mind that he was not returning to the planet of his first existence. He thought of the dull greys and blacks of his planet, of the monotonous routine of existence that had once been his—and his heart thrilled to the memories of the starry nights and perfect exciting days he had spent on his three month trip over Earth. He made a brief salute to the existence he had known, turned with a tiny sigh, and his fingers made brief adjustments in the controls. The rocket-thrum deepened, and the thin whistle of tenuous air clutching the ship echoed through the hull-plates. He thought of many things in those few moments. He watched the roundness of Earth flatten out, then take on the cup-like illusion that all planets had for an incoming ship. He reduced the drive of his rockets to a mere whisper, striving to control the impatience that crowded his mind. He shivered suddenly, remembering his utter callousness the first time he had sent a space ship whipping down toward the hills and valleys below. And there was a sickness within him when he fully realized that, despite his acquired memory and traits, he was an alien from outer space. He fingered the tiny scars that had completely obliterated the slight differences in his appearance from an Earthman's, and his fingers trembled a bit, as he bent and stared through the vision port. He said a brief prayer in his heart to a God whose presence he now felt very deeply. There were tears in the depths of his eyes, then, and memories were hot, bitter pains. Earth was not far below him. As he let gravity suck him earthward, he heaved a gasp of relief. He was no longer Thig, a creature of a Horde's creation, but Lewis Terry, writer of lurid gun-smoking tales of the West. He must remember that always. He had destroyed the real Terry and now, for the rest of his life, he must make up to the dead man's family. The knowledge that Ellen's love was not really meant for him would be a knife twisting in his heart but for her sake he must endure it. Her dreams and happiness must never be shattered. The bulge of Earth was flattening out now and he could see the outlines of Long Island in the growing twilight. A new plot was growing in the brain of Lewis Terry, a yarn about a cowboy suddenly transported to another world. He smiled ironically. He had seen those other worlds. Perhaps some day he would write about them.... He was Lewis Terry! He must remember that!
|
B. To investigate the planet without vslling attention, and determine if it's worth colonizing.
|
Why doesn't Martha's description of Terry fit what the reporter considers to be the norm?
A. Terry is passionate about space exploration.
B. Terry didn't like football.
C. Terry is an only child.
D. Terry is shy. A bookworm, who doesn't play sports.
|
STAR MOTHER By ROBERT F. YOUNG A touching story of the most enduring love in all eternity. That night her son was the first star. She stood motionless in the garden, one hand pressed against her heart, watching him rise above the fields where he had played as a boy, where he had worked as a young man; and she wondered whether he was thinking of those fields now, whether he was thinking of her standing alone in the April night with her memories; whether he was thinking of the verandahed house behind her, with its empty rooms and silent halls, that once upon a time had been his birthplace. Higher still and higher he rose in the southern sky, and then, when he had reached his zenith, he dropped swiftly down past the dark edge of the Earth and disappeared from sight. A boy grown up too soon, riding round and round the world on a celestial carousel, encased in an airtight metal capsule in an airtight metal chariot ... Why don't they leave the stars alone? she thought. Why don't they leave the stars to God? The general's second telegram came early the next morning: Explorer XII doing splendidly. Expect to bring your son down sometime tomorrow . She went about her work as usual, collecting the eggs and allocating them in their cardboard boxes, then setting off in the station wagon on her Tuesday morning run. She had expected a deluge of questions from her customers. She was not disappointed. "Is Terry really way up there all alone, Martha?" "Aren't you scared , Martha?" "I do hope they can get him back down all right, Martha." She supposed it must have given them quite a turn to have their egg woman change into a star mother overnight. She hadn't expected the TV interview, though, and she would have avoided it if it had been politely possible. But what could she do when the line of cars and trucks pulled into the drive and the technicians got out and started setting up their equipment in the backyard? What could she say when the suave young man came up to her and said, "We want you to know that we're all very proud of your boy up there, ma'am, and we hope you'll do us the honor of answering a few questions." Most of the questions concerned Terry, as was fitting. From the way the suave young man asked them, though, she got the impression that he was trying to prove that her son was just like any other average American boy, and such just didn't happen to be the case. But whenever she opened her mouth to mention, say, how he used to study till all hours of the night, or how difficult it had been for him to make friends because of his shyness, or the fact that he had never gone out for football—whenever she started to mention any of these things, the suave young man was in great haste to interrupt her and to twist her words, by requestioning, into a different meaning altogether, till Terry's behavior pattern seemed to coincide with the behavior pattern which the suave young man apparently considered the norm, but which, if followed, Martha was sure, would produce not young men bent on exploring space but young men bent on exploring trivia. A few of the questions concerned herself: Was Terry her only child? ("Yes.") What had happened to her husband? ("He was killed in the Korean War.") What did she think of the new law granting star mothers top priority on any and all information relating to their sons? ("I think it's a fine law ... It's too bad they couldn't have shown similar humanity toward the war mothers of World War II.") It was late in the afternoon by the time the TV crew got everything repacked into their cars and trucks and made their departure. Martha fixed herself a light supper, then donned an old suede jacket of Terry's and went out into the garden to wait for the sun to go down. According to the time table the general had outlined in his first telegram, Terry's first Tuesday night passage wasn't due to occur till 9:05. But it seemed only right that she should be outside when the stars started to come out. Presently they did, and she watched them wink on, one by one, in the deepening darkness of the sky. She'd never been much of a one for the stars; most of her life she'd been much too busy on Earth to bother with things celestial. She could remember, when she was much younger and Bill was courting her, looking up at the moon sometimes; and once in a while, when a star fell, making a wish. But this was different. It was different because now she had a personal interest in the sky, a new affinity with its myriad inhabitants. And how bright they became when you kept looking at them! They seemed to come alive, almost, pulsing brilliantly down out of the blackness of the night ... And they were different colors, too, she noticed with a start. Some of them were blue and some were red, others were yellow ... green ... orange ... It grew cold in the April garden and she could see her breath. There was a strange crispness, a strange clarity about the night, that she had never known before ... She glanced at her watch, was astonished to see that the hands indicated two minutes after nine. Where had the time gone? Tremulously she faced the southern horizon ... and saw her Terry appear in his shining chariot, riding up the star-pebbled path of his orbit, a star in his own right, dropping swiftly now, down, down, and out of sight beyond the dark wheeling mass of the Earth ... She took a deep, proud breath, realized that she was wildly waving her hand and let it fall slowly to her side. Make a wish! she thought, like a little girl, and she wished him pleasant dreams and a safe return and wrapped the wish in all her love and cast it starward. Sometime tomorrow, the general's telegram had said— That meant sometime today! She rose with the sun and fed the chickens, fixed and ate her breakfast, collected the eggs and put them in their cardboard boxes, then started out on her Wednesday morning run. "My land, Martha, I don't see how you stand it with him way up there! Doesn't it get on your nerves ?" ("Yes ... Yes, it does.") "Martha, when are they bringing him back down?" ("Today ... Today !") "It must be wonderful being a star mother, Martha." ("Yes, it is—in a way.") Wonderful ... and terrible. If only he can last it out for a few more hours, she thought. If only they can bring him down safe and sound. Then the vigil will be over, and some other mother can take over the awesome responsibility of having a son become a star— If only ... The general's third telegram arrived that afternoon: Regret to inform you that meteorite impact on satellite hull severely damaged capsule-detachment mechanism, making ejection impossible. Will make every effort to find another means of accomplishing your son's return. Terry!— See the little boy playing beneath the maple tree, moving his tiny cars up and down the tiny streets of his make-believe village; the little boy, his fuzz of hair gold in the sunlight, his cherub-cheeks pink in the summer wind— Terry!— Up the lane the blue-denimed young man walks, swinging his thin tanned arms, his long legs making near-grownup strides over the sun-seared grass; the sky blue and bright behind him, the song of cicada rising and falling in the hazy September air— Terry ... —probably won't get a chance to write you again before take-off, but don't worry, Ma. The Explorer XII is the greatest bird they ever built. Nothing short of a direct meteorite hit can hurt it, and the odds are a million to one ... Why don't they leave the stars alone? Why don't they leave the stars to God? The afternoon shadows lengthened on the lawn and the sun grew red and swollen over the western hills. Martha fixed supper, tried to eat, and couldn't. After a while, when the light began to fade, she slipped into Terry's jacket and went outside. Slowly the sky darkened and the stars began to appear. At length her star appeared, but its swift passage blurred before her eyes. Tires crunched on the gravel then, and headlights washed the darkness from the drive. A car door slammed. Martha did not move. Please God , she thought, let it be Terry , even though she knew that it couldn't possibly be Terry. Footsteps sounded behind her, paused. Someone coughed softly. She turned then— "Good evening, ma'am." She saw the circlet of stars on the gray epaulet; she saw the stern handsome face; she saw the dark tired eyes. And she knew. Even before he spoke again, she knew— "The same meteorite that damaged the ejection mechanism, ma'am. It penetrated the capsule, too. We didn't find out till just a while ago—but there was nothing we could have done anyway ... Are you all right, ma'am?" "Yes. I'm all right." "I wanted to express my regrets personally. I know how you must feel." "It's all right." "We will, of course, make every effort to bring back his ... remains ... so that he can have a fitting burial on Earth." "No," she said. "I beg your pardon, ma'am?" She raised her eyes to the patch of sky where her son had passed in his shining metal sarcophagus. Sirius blossomed there, blue-white and beautiful. She raised her eyes still higher—and beheld the vast parterre of Orion with its central motif of vivid forget-me-nots, its far-flung blooms of Betelguese and Rigel, of Bellatrix and Saiph ... And higher yet—and there flamed the exquisite flower beds of Taurus and Gemini, there burgeoned the riotous wreath of the Crab; there lay the pulsing petals of the Pleiades ... And down the ecliptic garden path, wafted by a stellar breeze, drifted the ocher rose of Mars ... "No," she said again. The general had raised his eyes, too; now, slowly, he lowered them. "I think I understand, ma'am. And I'm glad that's the way you want it ... The stars are beautiful tonight, aren't they." "More beautiful than they've ever been," she said. After the general had gone, she looked up once more at the vast and variegated garden of the sky where her son lay buried, then she turned and walked slowly back to the memoried house. THE END Transcriber's Note: This etext was produced from Amazing Stories January 1959. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
D. Terry is shy. A bookworm, who doesn't play sports.
|
Where is Stinson from?
A. Montana
B. Missouri
C. Mississippi
D. Michigan
|
THE GOD NEXT DOOR By BILL DOEDE Illustrated by IVIE [Transcriber's Note: This etext was produced from Galaxy Magazine August 1961. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The sand-thing was powerful, lonely and strange. No doubt it was a god—but who wasn't? Stinson lay still in the sand where he fell, gloating over the success of his arrival. He touched the pencil-line scar behind his ear where the cylinder was buried, marveling at the power stored there, power to fling him from earth to this fourth planet of the Centaurian system in an instant. It had happened so fast that he could almost feel the warm, humid Missouri air, though he was light years from Missouri. He got up. A gray, funnel-shaped cloud of dust stood off to his left. This became disturbing, since there was scarcely enough wind to move his hair. He watched it, trying to recall what he might know about cyclones. But he knew little. Weather control made cyclones and other climatic phenomena on earth practically non-existent. The cloud did not move, though, except to spin on its axis rapidly, emitting a high-pitched, scarcely audible whine, like a high speed motor. He judged it harmless. He stood on a wide valley floor between two mountain ranges. Dark clouds capped one peak of the mountains on his left. The sky was deep blue. He tested the gravity by jumping up and down. Same as Earth gravity. The sun—no, not the sun. Not Sol. What should he call it, Alpha or Centaurus? Well, perhaps neither. He was here and Earth was somewhere up there. This was the sun of this particular solar system. He was right the first time. The sun burned fiercely, although he would have said it was about four o'clock in the afternoon, if this had been Earth. Not a tree, nor a bush, nor even a wisp of dry grass was in sight. Everywhere was desert. The funnel of sand had moved closer and while he watched it, it seemed to drift in the wind—although there was no wind. Stinson backed away. It stopped. It was about ten feet tall by three feet in diameter at the base. Then Stinson backed away again. It was changing. Now it became a blue rectangle, then a red cube, a violet sphere. He wanted to run. He wished Benjamin were here. Ben might have an explanation. "What am I afraid of?" he said aloud, "a few grains of sand blowing in the wind? A wind devil?" He turned his back and walked away. When he looked up the wind devil was there before him. He looked back. Only one. It had moved. The sun shone obliquely, throwing Stinson's shadow upon the sand. The wind devil also had a shadow, although the sun shone through it and the shadow was faint. But it moved when the funnel moved. This was no illusion. Again Stinson felt the urge to run, or to use the cylinder to project himself somewhere else, but he said, "No!" very firmly to himself. He was here to investigate, to determine if this planet was capable of supporting life. Life? Intelligence? He examined the wind devil as closely as he dared, but it was composed only of grains of sand. There was no core, no central place you could point to and say, here is the brain, or the nervous system. But then, how could a group of loosely spaced grains of sand possibly have a nervous system? It was again going through its paces. Triangle, cube, rectangle, sphere. He watched, and when it became a triangle again, he smoothed a place in the sand and drew a triangle with his forefinger. When it changed to a cube he drew a square, a circle for a sphere, and so on. When the symbols were repeated he pointed to each in turn, excitement mounting. He became so absorbed in doing this that he failed to notice how the wind devil drew closer and closer, but when he inhaled the first grains of sand, the realization of what was happening dawned with a flash of fear. Instantly he projected himself a thousand miles away. Now he was in an area of profuse vegetation. It was twilight. As he stood beside a small creek, a chill wind blew from the northwest. He wanted to cover himself with the long leaves he found, but they were dry and brittle, for here autumn had turned the leaves. Night would be cold. He was not a woodsman. He doubted if he could build a fire without matches. So he followed the creek to where it flowed between two great hills. Steam vapors rose from a crevice. A cave was nearby and warm air flowed from its mouth. He went inside. At first he thought the cave was small, but found instead that he was in a long narrow passageway. The current of warm air flowed toward him and he followed it, cautiously, stepping carefully and slowly. Then it was not quite so dark. Soon he stepped out of the narrow passageway into a great cavern with a high-vaulted ceiling. The light source was a mystery. He left no shadow on the floor. A great crystal sphere hung from the ceiling, and he was curious about its purpose, but a great pool of steaming water in the center of the cavern drew his attention. He went close, to warm himself. A stone wall surrounding the pool was inscribed with intricate art work and indecipherable symbols. Life. Intelligence. The planet was inhabited. Should he give up and return to earth? Or was there room here for his people? Warming his hands there over the great steaming pool he thought of Benjamin, and Straus, and Jamieson—all those to whom he had given cylinders, and who were now struggling for life against those who desired them. He decided it would not be just, to give up so easily. The wide plaza between the pool and cavern wall was smooth as polished glass. Statues lined the wall. He examined them. The unknown artist had been clever. From one angle they were animals, from another birds, from a third they were vaguely humanoid creatures, glowering at him with primitive ferocity. The fourth view was so shocking he had to turn away quickly. No definable form or sculptured line was visible, yet he felt, or saw—he did not know which senses told him—the immeasurable gulf of a million years of painful evolution. Then nothing. It was not a curtain drawn to prevent him from seeing more. There was no more. He stumbled toward the pool's wall and clutched for support, but his knees buckled. His hand slid down the wall, over the ancient inscriptions. He sank to the floor. Before he lost consciousness he wondered, fleetingly, if a lethal instrument was in the statue. He woke with a ringing in his ears, feeling drugged and sluggish. Sounds came to him. He opened his eyes. The cavern was crowded. These creatures were not only humanoid, but definitely human, although more slight of build than earth people. The only difference he could see at first sight was that they had webbed feet. All were dressed from the waist down only, in a shimmering skirt that sparkled as they moved. They walked with the grace of ballet dancers, moving about the plaza, conversing in a musical language with no meaning for Stinson. The men were dark-skinned, the women somewhat lighter, with long flowing hair, wide lips and a beauty that was utterly sensual. He was in chains! They were small chains, light weight, of a metal that looked like aluminum. But all his strength could not break them. They saw him struggling. Two of the men came over and spoke to him in the musical language. "My name is Stinson," he said, pointing to himself. "I'm from the planet Earth." They looked at each other and jabbered some more. "Look," he said, "Earth. E-A-R-T-H, Earth." He pointed upward, described a large circle, then another smaller, and showed how Earth revolved around the sun. One of the men poked him with a stick, or tube of some kind. It did not hurt, but angered him. He left the chains by his own method of travel, and reappeared behind the two men. They stared at the place where he had been. The chains tinkled musically. He grasped the shoulder of the offender, spun him around and slapped his face. A cry of consternation rose from the group, echoing in the high ceilinged cavern. "SBTL!" it said, "ZBTL ... XBTL ... zbtl." The men instantly prostrated themselves before him. The one who had poked Stinson with the stick rose, and handed it to him. Still angered, Stinson grasped it firmly, with half a notion to break it over his head. As he did so, a flash of blue fire sprang from it. The man disappeared. A small cloud of dust settled slowly to the floor. Disintegrated! Stinson's face drained pale, and suddenly, unaccountably, he was ashamed because he had no clothes. "I didn't mean to kill him!" he cried. "I was angry, and...." Useless. They could not understand. For all he knew, they might think he was threatening them. The object he had thought of as a stick was in reality a long metal tube, precisely machined, with a small button near one end. This weapon was completely out of place in a culture such as this. Or was it? What did he know of these people? Very little. They were humanoid. They had exhibited human emotions of anger, fear and, that most human of all characteristics, curiosity. But up to now the tube and the chain was the only evidence of an advanced technology, unless the ancient inscriptions in the stone wall of the pool, and the statues lining the wall were evidences. There was a stirring among the crowd. An object like a pallet was brought, carried by four of the women. They laid it at his feet, and gestured for him to sit. He touched it cautiously, then sat. Instantly he sprang to his feet. There, at the cavern entrance, the wind devil writhed and undulated in a brilliant harmony of colors. It remained in one spot, though, and he relaxed somewhat. One of the women came toward him, long golden hair flowing, firm breasts dipping slightly at each step. Her eyes held a language all their own, universal. She pressed her body against him and bore him to the pallet, her kisses fire on his face. Incongruously, he thought of Benjamin back on earth, and all the others with cylinders, who might be fighting for their lives at this moment. He pushed her roughly aside. She spoke, and he understood! Her words were still the same gibberish, but now he knew their meaning. Somehow he knew also that the wind devil was responsible for his understanding. "You do not want me?" she said sadly. "Then kill me." "Why should I kill you?" She shrugged her beautiful shoulders. "It is the way of the Gods," she said. "If you do not, then the others will." He took the tube-weapon in his hands, careful not to touch the button. "Don't be afraid. I didn't mean to kill the man. It was an accident. I will protect you." She shook her head. "One day they will find me alone, and they'll kill me." "Why?" She shrugged. "I have not pleased you." "On the contrary, you have. There is a time and place for everything, though." Suddenly a great voice sounded in the cavern, a voice with no direction. It came from the ceiling, the floor, the walls, the steaming pool. It was in the language of the web-footed people; it was in his own tongue. "No harm must come to this woman. The God with fingers on his feet has decreed this." Those in the cavern looked at the woman with fear and respect. She kissed Stinson's feet. Two of the men came and gave her a brilliant new skirt. She smiled at him, and he thought he had never seen a more beautiful face. The great, bodiless voice sounded again, but those in the cavern went about their activities. They did not hear. "Who are you?" Stinson looked at the wind devil, since it could be no one else speaking, and pointed to himself. "Me?" "Yes." "I am Stinson, of the planet Earth." "Yes, I see it in your mind, now. You want to live here, on this planet." "Then you must know where I came from, and how." "I do not understand how. You have a body, a physical body composed of atoms. It is impossible to move a physical body from one place to another by a mere thought and a tiny instrument, yet you have done so. You deserted me out in the desert." "I deserted you?" Stinson cried angrily, "You tried to kill me!" "I was attempting communication. Why should I kill you?" He was silent a moment, looking at the people in the cavern. "Perhaps because you feared I would become the God of these people in your place." Stinson felt a mental shrug. "It is of no importance. When they arrived on this planet I attempted to explain that I was not a God, but the primitive is not deeply buried in them. They soon resorted to emotion rather than reason. It is of no importance." "I'd hardly call them primitive, with such weapons." "The tube is not of their technology. That is, they did not make it directly. These are the undesirables, the incorrigibles, the nonconformists from the sixth planet. I permit them here because it occupies my time, to watch them evolve." "You should live so long." "Live?" the wind devil said. "Oh, I see your meaning. I'd almost forgotten. You are a strange entity. You travel by a means even I cannot fully understand, yet you speak of time as if some event were about to take place. I believe you think of death. I see your physical body has deteriorated since yesterday. Your body will cease to exist, almost as soon as those of the sixth planet peoples. I am most interested in you. You will bring your people, and live here." "I haven't decided. There are these web-footed people, who were hostile until they thought I was a God. They have destructive weapons. Also, I don't understand you. I see you as a cone of sand which keeps changing color and configuration. Is it your body? Where do you come from? Is this planet populated with your kind?" The wind devil hesitated. "Where do I originate? It seems I have always been. You see this cavern, the heated pool, the statues, the inscriptions. Half a million years ago my people were as you. That is, they lived in physical bodies. Our technology surpassed any you have seen. The tube these webfoots use is a toy by comparison. Our scientists found the ultimate nature of physical law. They learned to separate the mind from the body. Then my people set a date. Our entire race was determined to free itself from the confines of the body. The date came." "What happened?" "I do not know. I alone exist. I have searched all the levels of time and matter from the very beginning. My people are gone. Sometimes it almost comes to me, why they are gone. And this is contrary to the greatest law of all—that an entity, once in existence, can never cease to exist." Stinson was silent, thinking of the endless years of searching through the great gulf of time. His eyes caught sight of the woman, reclining now on the pallet. The men had left her and stood in groups, talking, glancing at him, apparently free of their awe and fear already. The woman looked at him, and she was not smiling. "Please ask the Sand God," she said, "to speak to my people again. Their fear of him does not last. When He is gone they will probably kill us." "As for the webfoots," the wind devil, or Sand God, said, "I will destroy them. You and your people will have the entire planet." "Destroy them?" Stinson asked, incredulously, "all these people? They have a right to live like any one else." "Right? What is it—'right?' They are entities. They exist, therefore they always will. My people are the only entities who ever died. To kill the body is unimportant." "No. You misunderstand. Listen, you spoke of the greatest law. Your law is a scientific hypothesis. It has to do with what comes after physical existence, not with existence itself. The greatest law is this, that an entity, once existing, must not be harmed in any way. To do so changes the most basic structure of nature." The Sand God did not reply. The great bodiless, directionless voice was silent, and Stinson felt as if he had been taken from some high place and set down in a dark canyon. The cone of sand was the color of wood ashes. It pulsed erratically, like a great heart missing a beat now and then. The web-footed people milled about restlessly. The woman's eyes pleaded. When he looked back, the Sand God was gone. Instantly a new note rose in the cavern. The murmur of unmistakable mob fury ran over the webfoots. Several of the men approached the woman with hatred in their voices. He could not understand the words now. But he understood her. "They'll kill me!" she cried. Stinson pointed the disintegrating weapon at them and yelled. They dropped back. "We'll have to get outside," he told her. "This mob will soon get out of hand. Then the tube won't stop them. They will rush in. I can't kill them all at once, even if I wanted to. And I don't." Together they edged toward the cavern entrance, ran quickly up the inclined passageway, and came out into crisp, cold air. The morning sun was reflected from a million tiny mirrors on the rocks, the trees and grass. A silver thaw during the night had covered the whole area with a coating of ice. Stinson shivered. The woman handed him a skirt she had thoughtfully brought along from the cavern. He took it, and they ran down the slippery path leading away from the entrance. From the hiding place behind a large rock they watched, as several web-footed men emerged into the sunlight. They blinked, covered their eyes, and jabbered musically among themselves. One slipped and fell on the ice. They re-entered the cave. Stinson donned the shimmering skirt, smiling as he did so. The others should see him now. Benjamin and Straus and Jamieson. They would laugh. And Ben's wife, Lisa, she would give her little-girl laugh, and probably help him fasten the skirt. It had a string, like a tobacco pouch, which was tied around the waist. It helped keep him warm. He turned to the woman. "I don't know what I'll do with you, but now that we're in trouble together, we may as well introduce ourselves. My name is Stinson." "I am Sybtl," she said. "Syb-tl." He tried to imitate her musical pronunciation. "A very nice name." She smiled, then pointed to the cavern. "When the ice is gone, they will come out and follow us." "We'd better make tracks." "No," she said, "we must run, and make no tracks." "Okay, Sis," he said. "Sis?" "That means, sister." "I am not your sister. I am your wife." " What? " "Yes. When a man protects a woman from harm, it is a sign to all that she is his chosen. Otherwise, why not let her die? You are a strange God." "Listen, Sybtl," he said desperately, "I am not a God and you are not my wife. Let's get that straight." "But...." "No buts. Right now we'd better get out of here." He took her hand and they ran, slid, fell, picked themselves up again, and ran. He doubted the wisdom of keeping her with him. Alone, the webfoots were no match for him. He could travel instantly to any spot he chose. But with Sybtl it was another matter; he was no better than any other man, perhaps not so good as some because he was forty, and never had been an athlete. How was he to decide if this planet was suitable for his people, hampered by a woman, slinking through a frozen wilderness like an Indian? But the woman's hand was soft. He felt strong knowing she depended on him. Anyway, he decided, pursuit was impossible. They left no tracks on the ice. They were safe, unless the webfoots possessed talents unknown to him. So they followed the path leading down from the rocks, along the creek with its tumbling water. Frozen, leafless willows clawed at their bodies. The sun shone fiercely in a cloudless sky. Already water ran in tiny rivulets over the ice. The woman steered him to the right, away from the creek. Stinson's bare feet were numb from walking on ice. Christ, he thought, what am I doing here, anyway? He glanced down at Sybtl and remembered the webfoots. He stopped, tempted to use his cylinder and move to a warmer, less dangerous spot. The woman pulled on his arm. "We must hurry!" He clutched the tube-weapon. "How many shots in this thing?" "Shots?" "How often can I use it?" "As often as you like. It is good for fifty years. Kaatr—he is the one you destroyed—brought it from the ship when we came. Many times he has used it unwisely." "When did you come?" "Ten years ago. I was a child." "I thought only criminals were brought here." She nodded. "Criminals, and their children." "When will your people come again?" She shook her head. "Never. They are no longer my people. They have disowned us." "And because of me even those in the cavern have disowned you." Suddenly she stiffened beside him. There, directly in their path, stood the Sand God. It was blood red now. It pulsed violently. The great voice burst forth. "Leave the woman!" it demanded angrily. "The webfoots are nearing your position." "I cannot leave her. She is helpless against them." "What form of primitive stupidity are you practicing now? Leave, or they will kill you." Stinson shook his head. The Sand God pulsed more violently than before. Ice melted in a wide area around it. Brown, frozen grass burned to ashes. "You will allow them to kill you, just to defend her life? What business is it of yours if she lives or dies? My race discarded such primitive logic long before it reached your level of development." "Yes," Stinson said, "and your race no longer exists." The Sand God became a sphere of blue flame. A wave of intense heat drove them backward. "Earthman," the great voice said, "go back to your Earth. Take your inconsistencies with you. Do not come here again to infect my planet with your primitive ideas. The webfoots are not as intelligent as you, but they are sane. If you bring your people here, I shall destroy you all." The sphere of blue fire screamed away across the frozen wilderness, and the thunder of its passing shook the ground and echoed among the lonely hills. Sybtl shivered against his arm. "The Sand God is angry," she said. "My people tell how he was angry once before, when we first came here. He killed half of us and burned the ship that brought us. That is how Kaatr got the tube-weapon. It was the only thing the Sand God didn't burn, that and the skirts. Then, when he had burned the ship, the Sand God went to the sixth planet and burned two of the largest cities, as a warning that no more of us must come here." Well, Stinson said to himself, that does it. We are better off on Earth. We can't fight a monster like him. Sybtl touched his arm. "Why did the Sand God come? He did not speak." "He spoke to me." "I did not hear." "Yes, I know now. His voice sounds like thunder in the sky, but it is a voice that speaks only in the mind. He said I must leave this planet." She glanced at him with suddenly awakened eyes, as if thinking of it for the first time. "Where is your ship?" "I have no ship." "Then he will kill you." She touched her fingers on his face. "I am sorry. It was all for me." "Don't worry. The Sand God travels without a ship, why shouldn't I?" "Now?" "As soon as you are safe. Come." Steam rose from the burned area, charred like a rocket launching pit. They stepped around it carefully. Stinson felt warm air, but there was no time, now, to warm cold feet or dwell on the vagaries of Sand Gods. Together they crossed the narrow valley. Sybtl led him toward a tall mound of rock. Here they came to the creek again, which flowed into a small canyon. They climbed the canyon wall. Far away, small figures moved. The webfoots were on their trail. She drew him into a small cave. It was heated, like the great cavern, but held no walled pool nor mysterious lighting. But it was warm, and the small entrance made an excellent vantage point for warding off attack. "They will not find us...." A high-pitched keening burst suddenly around them. Stinson knew they had heard, or felt the sound for some time, that now its frequency was in an audible range. "The Sand God," Sybtl said. "Sometimes he plays among the clouds. He makes it rain in a dry summer, or sometimes warms the whole world for days at a time in winter, so the snow melts and the grass begins to green. Then he tires and lets winter come back again. He is the loneliest God in the universe." "What makes you think he's lonely?" She shrugged her shoulders. "I just know. But he's an angry God now. See those clouds piling in the East? Soon they will hide the sun. Then he will make them churn and boil, like river whirlpools in spring. At least he does this when he plays. Who knows what he will do when he's angry?" "The Sand God isn't doing this," Stinson said. "It's only a storm." She covered his lips with her fingers. "Don't say that. He may hear you and be more angry." "But it is, don't you see? You give him powers he does not possess." Sybtl shook her head and stroked his face with her long, slim fingers. "Poor little God-with-fingers-on-his-feet," she said. "You do not understand. The Sand God is terrible, even when he plays. See the lightning? It is blue. The lightning of a storm that comes by itself is not blue. He is running around the world on feet like the rockets of space ships, and when he strikes the clouds, blue fire shoots away." The clouds continued to build on one another. Soon the blue flashes of lightning extended across the sky from horizon to horizon. The earth trembled. Sybtl moved closer, trembling also. "He never did this before," she said. "He never made the earth shake before." Great boulders crashed down the canyon walls and dropped into the creek. They dared not move from the cave, although death seemed certain if they stayed. "I'll leave for a moment," he said. "I'll be back soon." "You're leaving?" There was panic in her voice. "Only for a moment." "And you won't come back. You will go to your world." "No. I'll be back." "Promise? No, don't promise. The promises of Gods often are forgotten before the sounds die away." "I'll be back." He disappeared at once, giving her no chance to object again, and went to the desert of sand, where he had first arrived on the planet. He wanted to see if the storm were world-wide. Stinson had never been in a sand storm before, even on Earth. He could not breathe. He could not see. Bullets of sand stung his skin. Bullets of sand shot into his eyes. Clouds of sand howled around him. He fell, and the wind rolled him over and over in the sand like a tumbleweed. The skirt flew up around his face. He could not get up again. He returned to the cave. Soon after, while they sat huddled together, watching the chaos of tumbling rocks, lightning, and driving rain, the high-pitched keening came again. A sphere of blue fire appeared in the east. Its brilliance put the lightning to shame. It bore down on the cave swiftly, purposefully. Stinson prepared himself to leave. In spite of his desire to protect Sybtl, it was useless to get himself killed when he was powerless to help her. But at the last moment it veered off. "Fiend!" Stinson screamed the word, vaguely marvelling at his own fury. The blue sphere turned and came back. "Monster!" Again. "Murderer!" "Adolescent!" This time it kept going. The rain and wind ceased. Lightning stopped. Thunder rumbled distantly. Clouds disappeared. Stinson and Sybtl emerged from the cave. There was no longer a question of attack from the webfoots, the storm had taken care of that. The fierce sun began its work of drying rocks and throwing shadows and coaxing life out into the open again. Down in the canyon a bird sang, a lonely, cheerful twitter. "The Sand God is tired," Sybtl said. "He is not angry now. I'm glad. Perhaps he will let you stay." "No. Even if he allowed it, I couldn't stay. My people could never live here with a God who is half devil." The cone of sand suddenly appeared. It stood in the canyon, its base on a level with the cave. It was quiet. It was dull gray in color. It exuded impressions of death, of hopeful words solemnly spoken over lowered coffins, of cold earth and cold space, of dank, wet catacombs, of creeping, crawling nether things. The bird's twitter stopped abruptly. "Earthman," the Sand God said, as if he were about to make a statement. Stinson ignored him. He glanced down at Sybtl, who sensed that this was a time for good-bys. He thought, perhaps I can stay here alone with her. The webfoots might find us, or the Sand God might destroy us in one of his fits, but it might be worth it. "Don't go," she said. "Not yet." "Earthman, hear me." "I hear you." "Why does your mind shrink backward?" "I've decided not to bring my people here." " You decided?" "Certainly," Stinson said boldly. "Call it rationalization, if you wish. You ordered us away; and I have several good reasons for not coming here if the door was open." "I've changed my mind. You will be welcomed." "Listen to that, will you?" Stinson said angrily. "Just listen! You set yourself up as a God for the webfoots. You get them eating out of your hand. Then what do you do? You throw a fit. Yes, a fit! Like an adolescent. Worse." "Earthman, wait...." "No!" Stinson shot back. "You've owned this planet for a million years. You have brooded here alone since before my people discovered fire, and in all those ages you never learned self-control. I can't subject my people to the whims of an entity who throws a planetary fit when it pleases him." Stinson relaxed. He'd had his say. Sybtl trembled beside him. A small mammal, round, furry, hopped by, sniffing inquisitively. Sybtl said, "Is the Sand God happy?" She shook her head. "No, he is not happy. He is old, old, old. I can feel it. My people say that when one gets too old it is well to die. But Gods never die, do they? I would not like to be a God." "Stinson," the Sand God said. "You said I was adolescent. You are correct. Do you remember I told you how my people, the entire race, left their bodies at the same time? Do you imagine all of us were adults?" "I suppose not. Sounds reasonable. How old were you?" "Chronologically, by our standards, I was nine years old." "But you continued to develop after...." "No." Stinson tried to imagine it. At first there must have been a single voice crying into a monstrous emptiness, "Mother, where are you? MOTHER! Where is everyone ?" A frenzied searching of the planet, the solar system, the galaxy. Then a returning to the planet. Empty.... Change. Buildings, roads, bridges weathering slowly. Such a race would have built of durable metal. Durable? Centuries, eons passed. Buildings crumbled to dust, dust blew away. Bridges eroded, fell, decomposed into basic elements. The shape of constellations changed. All trace of civilization passed except in the cavern of the heated pool. Constellations disappeared, new patterns formed in the night sky. The unutterably total void of time—FIVE HUNDRED THOUSAND YEARS! And a nine-year-old child brooding over an empty world. "I don't understand why your development stopped," Stinson said. "Nor do I. But perhaps ... well, I sense that I would continue, if you brought your people here. You have already taught me the value of life. There is a oneness, a bond that ties each living thing to every other living thing. It is a lesson my people never knew. Select any portion of this planet that suits you. Take the web-footed woman for your wife. Have children. I promise never to harm you in any way." "The webfoots?" "You and they shall share the planet." The Sand God disappeared. Sybtl said; "Is the Sand God angry again?" "No, he is not angry." "I'm glad. You will leave now?" "No. This is my home." She laughed softly. "You are a strange God." "Listen," he said, "I am not a God. Get that through your head." She drew him into the cave. Her lips were cool and sweet. The cave was pleasantly warm.
|
B. Missouri
|
What is motivating the King's army to fight against the Turks? (territorial conquest, religious, gold/money, personal glory)
A. National pride
B. Religious faith
C. Personal glory
D. Territorial conquest
|
... After a Few Words ... by Seaton McKettrig Illustrated by Summer [Transcriber's Note: This etext was produced from Analog October 1962. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] This is a science-fiction story. History is a science; the other part is, as all Americans know, the most fictional field we have today. He settled himself comfortably in his seat, and carefully put the helmet on, pulling it down firmly until it was properly seated. For a moment, he could see nothing. Then his hand moved up and, with a flick of the wrist, lifted the visor. Ahead of him, in serried array, with lances erect and pennons flying, was the forward part of the column. Far ahead, he knew, were the Knights Templars, who had taken the advance. Behind the Templars rode the mailed knights of Brittany and Anjou. These were followed by King Guy of Jerusalem and the host of Poitou. He himself, Sir Robert de Bouain, was riding with the Norman and English troops, just behind the men of Poitou. Sir Robert turned slightly in his saddle. To his right, he could see the brilliant red-and-gold banner of the lion-hearted Richard of England— gules, in pale three lions passant guardant or . Behind the standard-bearer, his great war horse moving with a steady, measured pace, his coronet of gold on his steel helm gleaming in the glaring desert sun, the lions of England on his firm-held shield, was the King himself. Further behind, the Knights Hospitallers protected the rear, guarding the column of the hosts of Christendom from harassment by the Bedouins. "By our Lady!" came a voice from his left. "Three days out from Acre, and the accursed Saracens still elude us." Sir Robert de Bouain twisted again in his saddle to look at the knight riding alongside him. Sir Gaeton de l'Arc-Tombé sat tall and straight in his saddle, his visor up, his blue eyes narrowed against the glare of the sun. Sir Robert's lips formed a smile. "They are not far off, Sir Gaeton. They have been following us. As we march parallel to the seacoast, so they have been marching with us in those hills to the east." "Like the jackals they are," said Sir Gaeton. "They assail us from the rear, and they set up traps in our path ahead. Our spies tell us that the Turks lie ahead of us in countless numbers. And yet, they fear to face us in open battle." "Is it fear, or are they merely gathering their forces?" "Both," said Sir Gaeton flatly. "They fear us, else they would not dally to amass so fearsome a force. If, as our informers tell us, there are uncounted Turks to the fore, and if, as we are aware, our rear is being dogged by the Bedouin and the black horsemen of Egypt, it would seem that Saladin has at hand more than enough to overcome us, were they all truly Christian knights." "Give them time. We must wait for their attack, sir knight. It were foolhardy to attempt to seek them in their own hills, and yet they must stop us. They will attack before we reach Jerusalem, fear not." "We of Gascony fear no heathen Musselman," Sir Gaeton growled. "It's this Hellish heat that is driving me mad." He pointed toward the eastern hills. "The sun is yet low, and already the heat is unbearable." Sir Robert heard his own laugh echo hollowly within his helmet. "Perhaps 'twere better to be mad when the assault comes. Madmen fight better than men of cooler blood." He knew that the others were baking inside their heavy armor, although he himself was not too uncomfortable. Sir Gaeton looked at him with a smile that held both irony and respect. "In truth, sir knight, it is apparent that you fear neither men nor heat. Nor is your own blood too cool. True, I ride with your Normans and your English and your King Richard of the Lion's Heart, but I am a Gascon, and have sworn no fealty to him. But to side with the Duke of Burgundy against King Richard—" He gave a short, barking laugh. "I fear no man," he went on, "but if I had to fear one, it would be Richard of England." Sir Robert's voice came like a sword: steely, flat, cold, and sharp. "My lord the King spoke in haste. He has reason to be bitter against Philip of France, as do we all. Philip has deserted the field. He has returned to France in haste, leaving the rest of us to fight the Saracen for the Holy Land leaving only the contingent of his vassal the Duke of Burgundy to remain with us." "Richard of England has never been on the best of terms with Philip Augustus," said Sir Gaeton. "No, and with good cause. But he allowed his anger against Philip to color his judgment when he spoke harshly against the Duke of Burgundy. The Duke is no coward, and Richard Plantagenet well knows it. As I said, he spoke in haste." "And you intervened," said Sir Gaeton. "It was my duty." Sir Robert's voice was stubborn. "Could we have permitted a quarrel to develop between the two finest knights and warleaders in Christendom at this crucial point? The desertion of Philip of France has cost us dearly. Could we permit the desertion of Burgundy, too?" "You did what must be done in honor," the Gascon conceded, "but you have not gained the love of Richard by doing so." Sir Robert felt his jaw set firmly. "My king knows I am loyal." Sir Gaeton said nothing more, but there was a look in his eyes that showed that he felt that Richard of England might even doubt the loyalty of Sir Robert de Bouain. Sir Robert rode on in silence, feeling the movement of the horse beneath him. There was a sudden sound to the rear. Like a wash of the tide from the sea came the sound of Saracen war cries and the clash of steel on steel mingled with the sounds of horses in agony and anger. Sir Robert turned his horse to look. The Negro troops of Saladin's Egyptian contingent were thundering down upon the rear! They clashed with the Hospitallers, slamming in like a rain of heavy stones, too close in for the use of bows. There was only the sword against armor, like the sound of a thousand hammers against a thousand anvils. "Stand fast! Stand fast! Hold them off!" It was the voice of King Richard, sounding like a clarion over the din of battle. Sir Robert felt his horse move, as though it were urging him on toward the battle, but his hand held to the reins, keeping the great charger in check. The King had said "Stand fast!" and this was no time to disobey the orders of Richard. The Saracen troops were coming in from the rear, and the Hospitallers were taking the brunt of the charge. They fought like madmen, but they were slowly being forced back. The Master of the Hospitallers rode to the rear, to the King's standard, which hardly moved in the still desert air, now that the column had stopped moving. The voice of the Duke of Burgundy came to Sir Robert's ears. "Stand fast. The King bids you all to stand fast," said the duke, his voice fading as he rode on up the column toward the knights of Poitou and the Knights Templars. The Master of the Hospitallers was speaking in a low, urgent voice to the King: "My lord, we are pressed on by the enemy and in danger of eternal infamy. We are losing our horses, one after the other!" "Good Master," said Richard, "it is you who must sustain their attack. No one can be everywhere at once." The Master of the Hospitallers nodded curtly and charged back into the fray. The King turned to Sir Baldwin de Carreo, who sat ahorse nearby, and pointed toward the eastern hills. "They will come from there, hitting us in the flank; we cannot afford to amass a rearward charge. To do so would be to fall directly into the hands of the Saracen." A voice very close to Sir Robert said: "Richard is right. If we go to the aid of the Hospitallers, we will expose the column to a flank attack." It was Sir Gaeton. "My lord the King," Sir Robert heard his voice say, "is right in all but one thing. If we allow the Egyptians to take us from the rear, there will be no need for Saladin and his Turks to come down on our flank. And the Hospitallers cannot hold for long at this rate. A charge at full gallop would break the Egyptian line and give the Hospitallers breathing time. Are you with me?" "Against the orders of the King?" "The King cannot see everything! There are times when a man must use his own judgment! You said you were afraid of no man. Are you with me?" After a moment's hesitation, Sir Gaeton couched his lance. "I'm with you, sir knight! Live or die, I follow! Strike and strike hard!" "Forward then!" Sir Robert heard himself shouting. "Forward for St. George and for England!" "St. George and England!" the Gascon echoed. Two great war horses began to move ponderously forward toward the battle lines, gaining momentum as they went. Moving in unison, the two knights, their horses now at a fast trot, lowered their lances, picking their Saracen targets with care. Larger and larger loomed the Egyptian cavalrymen as the horses changed pace to a thundering gallop. The Egyptians tried to dodge, as they saw, too late, the approach of the Christian knights. Sir Robert felt the shock against himself and his horse as the steel tip of the long ash lance struck the Saracen horseman in the chest. Out of the corner of his eye, he saw that Sir Gaeton, too, had scored. The Saracen, impaled on Sir Robert's lance, shot from the saddle as he died. His lighter armor had hardly impeded the incoming spear-point, and now his body dragged it down as he dropped toward the desert sand. Another Moslem cavalryman was charging in now, swinging his curved saber, taking advantage of Sir Robert's sagging lance. There was nothing else to do but drop the lance and draw his heavy broadsword. His hand grasped it, and it came singing from its scabbard. The Egyptian's curved sword clanged against Sir Robert's helm, setting his head ringing. In return, the knight's broadsword came about in a sweeping arc, and the Egyptian's horse rode on with the rider's headless body. Behind him, Sir Robert heard further cries of "St. George and England!" The Hospitallers, taking heart at the charge, were going in! Behind them came the Count of Champagne, the Earl of Leister, and the Bishop of Beauvais, who carried a great warhammer in order that he might not break Church Law by shedding blood. Sir Robert's own sword rose and fell, cutting and hacking at the enemy. He himself felt a dreamlike detachment, as though he were watching the battle rather than participating in it. But he could see that the Moslems were falling back before the Christian onslaught. And then, quite suddenly, there seemed to be no foeman to swing at. Breathing heavily, Sir Robert sheathed his broadsword. Beside him, Sir Gaeton did the same, saying: "It will be a few minutes before they can regroup, sir knight. We may have routed them completely." "Aye. But King Richard will not approve of my breaking ranks and disobeying orders. I may win the battle and lose my head in the end." "This is no time to worry about the future," said the Gascon. "Rest for a moment and relax, that you may be the stronger later. Here—have an Old Kings ." He had a pack of cigarettes in his gauntleted hand, which he profferred to Sir Robert. There were three cigarettes protruding from it, one slightly farther than the others. Sir Robert's hand reached out and took that one. "Thanks. When the going gets rough, I really enjoy an Old Kings ." He put one end of the cigarette in his mouth and lit the other from the lighter in Sir Gaeton's hand. "Yes, sir," said Sir Gaeton, after lighting his own cigarette, " Old Kings are the greatest. They give a man real, deep-down smoking pleasure." "There's no doubt about it, Old Kings are a man's cigarette." Sir Robert could feel the soothing smoke in his lungs as he inhaled deeply. "That's great. When I want a cigarette, I don't want just any cigarette." "Nor I," agreed the Gascon. " Old Kings is the only real cigarette when you're doing a real man's work." "That's for sure." Sir Robert watched a smoke ring expand in the air. There was a sudden clash of arms off to their left. Sir Robert dropped his cigarette to the ground. "The trouble is that doing a real he-man's work doesn't always allow you to enjoy the fine, rich tobaccos of Old Kings right down to the very end." "No, but you can always light another later," said the Gascon knight. King Richard, on seeing his army moving suddenly toward the harassed rear, had realized the danger and had charged through the Hospitallers to get into the thick of the fray. Now the Turks were charging down from the hills, hitting—not the flank as he had expected, but the rear! Saladin had expected him to hold fast! Sir Robert and Sir Gaeton spurred their chargers toward the flapping banner of England. The fierce warrior-king of England, his mighty sword in hand, was cutting down Turks as though they were grain-stalks, but still the Saracen horde pressed on. More and more of the terrible Turks came boiling down out of the hills, their glittering scimitars swinging. Sir Robert lost all track of time. There was nothing to do but keep his own great broadsword moving, swinging like some gigantic metronome as he hacked down the Moslem foes. And then, suddenly, he found himself surrounded by the Saracens! He was isolated and alone, cut off from the rest of the Christian forces! He glanced quickly around as he slashed another Saracen from pate to breastbone. Where was Sir Gaeton? Where were the others? Where was the red-and-gold banner of Richard? He caught a glimpse of the fluttering banner far to the rear and started to fall back. And then he saw another knight nearby, a huge man who swung his sparkling blade with power and force. On his steel helm gleamed a golden coronet! Richard! And the great king, in spite of his prowess was outnumbered heavily and would, within seconds, be cut down by the Saracen horde! Without hesitation, Sir Robert plunged his horse toward the surrounded monarch, his great blade cutting a path before him. He saw Richard go down, falling from the saddle of his charger, but by that time his own sword was cutting into the screaming Saracens and they had no time to attempt any further mischief to the King. They had their hands full with Sir Robert de Bouain. He did not know how long he fought there, holding his charger motionless over the inert body of the fallen king, hewing down the screaming enemy, but presently he heard the familiar cry of "For St. George and for England" behind him. The Norman and English troops were charging in, bringing with them the banner of England! And then Richard was on his feet, cleaving the air about him with his own broadsword. Its bright edge, besmeared with Saracen blood, was biting viciously into the foe. The Turks began to fall back. Within seconds, the Christian knights were boiling around the embattled pair, forcing the Turks into retreat. And for the second time, Sir Robert found himself with no one to fight. And then a voice was saying: "You have done well this day, sir knight. Richard Plantagenet will not forget." Sir Robert turned in his saddle to face the smiling king. "My lord king, be assured that I would never forget my loyalty to my sovereign and liege lord. My sword and my life are yours whenever you call." King Richard's gauntleted hand grasped his own. "If it please God, I shall never ask your life. An earldom awaits you when we return to England, sir knight." And then the king mounted his horse and was running full gallop after the retreating Saracens. Robert took off his helmet. He blinked for a second to adjust his eyes to the relative dimness of the studio. After the brightness of the desert that the televicarion helmet had projected into his eyes, the studio seemed strangely cavelike. "How'd you like it, Bob?" asked one of the two producers of the show. Robert Bowen nodded briskly and patted the televike helmet. "It was O.K.," he said. "Good show. A little talky at the beginning, and it needs a better fade-out, but the action scenes were fine. The sponsor ought to like it—for a while, at least." "What do you mean, 'for a while'?" Robert Bowen sighed. "If this thing goes on the air the way it is, he'll lose sales." "Why? Commercial not good enough?" " Too good! Man, I've smoked Old Kings , and, believe me, the real thing never tasted as good as that cigarette did in the commercial!"
|
B. Religious faith
|
Why did Loy Chuk's people live underground?
A. Subterranean passages protect against desert sand storms.
B. Subterranean passages protect against larger predators.
C. Loy Chuk comes from a rodent species. Rodents usually live in underground burrows.
D. The temperature above ground at night is very cold.
|
THE ETERNAL WALL By RAYMOND Z. GALLUN A scream of brakes, the splash into icy waters, a long descent into alkaline depths ... it was death. But Ned Vince lived again—a million years later! "See you in half an hour, Betty," said Ned Vince over the party telephone. "We'll be out at the Silver Basket before ten-thirty...." Ned Vince was eager for the company of the girl he loved. That was why he was in a hurry to get to the neighboring town of Hurley, where she lived. His old car rattled and roared as he swung it recklessly around Pit Bend. There was where Death tapped him on the shoulder. Another car leaped suddenly into view, its lights glaring blindingly past a high, up-jutting mass of Jurassic rock at the turn of the road. Dazzled, and befuddled by his own rash speed, Ned Vince had only swift young reflexes to rely on to avoid a fearful, telescoping collision. He flicked his wheel smoothly to the right; but the County Highway Commission hadn't yet tarred the traffic-loosened gravel at the Bend. An incredible science, millions of years old, lay in the minds of these creatures. Ned could scarcely have chosen a worse place to start sliding and spinning. His car hit the white-painted wooden rail sideways, crashed through, tumbled down a steep slope, struck a huge boulder, bounced up a little, and arced outward, falling as gracefully as a swan-diver toward the inky waters of the Pit, fifty feet beneath.... Ned Vince was still dimly conscious when that black, quiet pool geysered around him in a mighty splash. He had only a dazing welt on his forehead, and a gag of terror in his throat. Movement was slower now, as he began to sink, trapped inside his wrecked car. Nothing that he could imagine could mean doom more certainly than this. The Pit was a tremendously deep pocket in the ground, spring-fed. The edges of that almost bottomless pool were caked with a rim of white—for the water, on which dead birds so often floated, was surcharged with alkali. As that heavy, natronous liquid rushed up through the openings and cracks beneath his feet, Ned Vince knew that his friends and his family would never see his body again, lost beyond recovery in this abyss. The car was deeply submerged. The light had blinked out on the dash-panel, leaving Ned in absolute darkness. A flood rushed in at the shattered window. He clawed at the door, trying to open it, but it was jammed in the crash-bent frame, and he couldn't fight against the force of that incoming water. The welt, left by the blow he had received on his forehead, put a thickening mist over his brain, so that he could not think clearly. Presently, when he could no longer hold his breath, bitter liquid was sucked into his lungs. His last thoughts were those of a drowning man. The machine-shop he and his dad had had in Harwich. Betty Moore, with the smiling Irish eyes—like in the song. Betty and he had planned to go to the State University this Fall. They'd planned to be married sometime.... Goodbye, Betty ... The ripples that had ruffled the surface waters in the Pit, quieted again to glassy smoothness. The eternal stars shone calmly. The geologic Dakota hills, which might have seen the dinosaurs, still bulked along the highway. Time, the Brother of Death, and the Father of Change, seemed to wait.... "Kaalleee! Tik!... Tik, tik, tik!... Kaalleee!..." The excited cry, which no human throat could quite have duplicated accurately, arose thinly from the depths of a powder-dry gulch, water-scarred from an inconceivable antiquity. The noon-day Sun was red and huge. The air was tenuous, dehydrated, chill. "Kaalleee!... Tik, tik, tik!..." At first there was only one voice uttering those weird, triumphant sounds. Then other vocal organs took up that trilling wail, and those short, sharp chuckles of eagerness. Other questioning, wondering notes mixed with the cadence. Lacking qualities identifiable as human, the disturbance was still like the babble of a group of workmen who have discovered something remarkable. The desolate expanse around the gulch, was all but without motion. The icy breeze tore tiny puffs of dust from grotesque, angling drifts of soil, nearly waterless for eons. Patches of drab lichen grew here and there on the up-jutting rocks, but in the desert itself, no other life was visible. Even the hills had sagged away, flattened by incalculable ages of erosion. At a mile distance, a crumbling heap of rubble arose. Once it had been a building. A gigantic, jagged mass of detritus slanted upward from its crest—red debris that had once been steel. A launching catapult for the last space ships built by the gods in exodus, perhaps it was—half a million years ago. Man was gone from the Earth. Glacial ages, war, decadence, disease, and a final scattering of those ultimate superhumans to newer worlds in other solar systems, had done that. "Kaalleee!... Tik, tik, tik!..." The sounds were not human. They were more like the chatter and wail of small desert animals. But there was a seeming paradox here in the depths of that gulch, too. The glint of metal, sharp and burnished. The flat, streamlined bulk of a flying machine, shiny and new. The bell-like muzzle of a strange excavator-apparatus, which seemed to depend on a blast of atoms to clear away rock and soil. Thus the gulch had been cleared of the accumulated rubbish of antiquity. Man, it seemed, had a successor, as ruler of the Earth. Loy Chuk had flown his geological expedition out from the far lowlands to the east, out from the city of Kar-Rah. And he was very happy now—flushed with a vast and unlooked-for success. He crouched there on his haunches, at the dry bottom of the Pit. The breeze rumpled his long, brown fur. He wasn't very different in appearance from his ancestors. A foot tall, perhaps, as he squatted there in that antique stance of his kind. His tail was short and furred, his undersides creamy. White whiskers spread around his inquisitive, pink-tipped snout. But his cranium bulged up and forward between shrewd, beady eyes, betraying the slow heritage of time, of survival of the fittest, of evolution. He could think and dream and invent, and the civilization of his kind was already far beyond that of the ancient Twentieth Century. Loy Chuk and his fellow workers were gathered, tense and gleeful, around the things their digging had exposed to the daylight. There was a gob of junk—scarcely more than an irregular formation of flaky rust. But imbedded in it was a huddled form, brown and hard as old wood. The dry mud that had encased it like an airtight coffin, had by now been chipped away by the tiny investigators; but soiled clothing still clung to it, after perhaps a million years. Metal had gone into decay—yes. But not this body. The answer to this was simple—alkali. A mineral saturation that had held time and change in stasis. A perfect preservative for organic tissue, aided probably during most of those passing eras by desert dryness. The Dakotas had turned arid very swiftly. This body was not a mere fossil. It was a mummy. "Kaalleee!" Man, that meant. Not the star-conquering demi-gods, but the ancestral stock that had built the first machines on Earth, and in the early Twenty-first Century, the first interplanetary rockets. No wonder Loy Chuk and his co-workers were happy in their paleontological enthusiasm! A strange accident, happening in a legendary antiquity, had aided them in their quest for knowledge. At last Loy Chuk gave a soft, chirping signal. The chant of triumph ended, while instruments flicked in his tiny hands. The final instrument he used to test the mummy, looked like a miniature stereoscope, with complicated details. He held it over his eyes. On the tiny screen within, through the agency of focused X-rays, he saw magnified images of the internal organs of this ancient human corpse. What his probing gaze revealed to him, made his pleasure even greater than before. In twittering, chattering sounds, he communicated his further knowledge to his henchmen. Though devoid of moisture, the mummy was perfectly preserved, even to its brain cells! Medical and biological sciences were far advanced among Loy Chuk's kind. Perhaps, by the application of principles long known to them, this long-dead body could be made to live again! It might move, speak, remember its past! What a marvelous subject for study it would make, back there in the museums of Kar-Rah! "Tik, tik, tik!..." But Loy silenced this fresh, eager chattering with a command. Work was always more substantial than cheering. With infinite care—small, sharp hand-tools were used, now—the mummy of Ned Vince was disengaged from the worthless rust of his primitive automobile. With infinite care it was crated in a metal case, and hauled into the flying machine. Flashing flame, the latter arose, bearing the entire hundred members of the expedition. The craft shot eastward at bullet-like speed. The spreading continental plateau of North America seemed to crawl backward, beneath. A tremendous sand desert, marked with low, washed-down mountains, and the vague, angular, geometric mounds of human cities that were gone forever. Beyond the eastern rim of the continent, the plain dipped downward steeply. The white of dried salt was on the hills, but there was a little green growth here, too. The dead sea-bottom of the vanished Atlantic was not as dead as the highlands. Far out in a deep valley, Kar-Rah, the city of the rodents, came into view—a crystalline maze of low, bubble-like structures, glinting in the red sunshine. But this was only its surface aspect. Loy Chuk's people had built their homes mostly underground, since the beginning of their foggy evolution. Besides, in this latter day, the nights were very cold, the shelter of subterranean passages and rooms was welcome. The mummy was taken to Loy Chuk's laboratory, a short distance below the surface. Here at once, the scientist began his work. The body of the ancient man was put in a large vat. Fluids submerged it, slowly soaking from that hardened flesh the alkali that had preserved it for so long. The fluid was changed often, until woody muscles and other tissues became pliable once more. Then the more delicate processes began. Still submerged in liquid, the corpse was submitted to a flow of restorative energy, passing between complicated electrodes. The cells of antique flesh and brain gradually took on a chemical composition nearer to that of the life that they had once known. At last the final liquid was drained away, and the mummy lay there, a mummy no more, but a pale, silent figure in its tatters of clothing. Loy Chuk put an odd, metal-fabric helmet on its head, and a second, much smaller helmet on his own. Connected with this arrangement, was a black box of many uses. For hours he worked with his apparatus, studying, and guiding the recording instruments. The time passed swiftly. At last, eager and ready for whatever might happen now, Loy Chuk pushed another switch. With a cold, rosy flare, energy blazed around that moveless form. For Ned Vince, timeless eternity ended like a gradual fading mist. When he could see clearly again, he experienced that inevitable shock of vast change around him. Though it had been dehydrated, his brain had been kept perfectly intact through the ages, and now it was restored. So his memories were as vivid as yesterday. Yet, through that crystalline vat in which he lay, he could see a broad, low room, in which he could barely have stood erect. He saw instruments and equipment whose weird shapes suggested alienness, and knowledge beyond the era he had known! The walls were lavender and phosphorescent. Fossil bone-fragments were mounted in shallow cases. Dinosaur bones, some of them seemed, from their size. But there was a complete skeleton of a dog, too, and the skeleton of a man, and a second man-skeleton that was not quite human. Its neck-vertebrae were very thick and solid, its shoulders were wide, and its skull was gigantic. All this weirdness had a violent effect on Ned Vince—a sudden, nostalgic panic. Something was fearfully wrong! The nervous terror of the unknown was on him. Feeble and dizzy after his weird resurrection, which he could not understand, remembering as he did that moment of sinking to certain death in the pool at Pit Bend, he caught the edge of the transparent vat, and pulled himself to a sitting posture. There was a muffled murmur around him, as of some vast, un-Earthly metropolis. "Take it easy, Ned Vince...." The words themselves, and the way they were assembled, were old, familiar friends. But the tone was wrong. It was high, shrill, parrot-like, and mechanical. Ned's gaze searched for the source of the voice—located the black box just outside of his crystal vat. From that box the voice seemed to have originated. Before it crouched a small, brownish animal with a bulging head. The animal's tiny-fingered paws—hands they were, really—were touching rows of keys. To Ned Vince, it was all utterly insane and incomprehensible. A rodent, looking like a prairie dog, a little; but plainly possessing a high order of intelligence. And a voice whose soothingly familiar words were more repugnant somehow, simply because they could never belong in a place as eerie as this. Ned Vince did not know how Loy Chuk had probed his brain, with the aid of a pair of helmets, and the black box apparatus. He did not know that in the latter, his language, taken from his own revitalized mind, was recorded, and that Loy Chuk had only to press certain buttons to make the instrument express his thoughts in common, long-dead English. Loy, whose vocal organs were not human, would have had great difficulty speaking English words, anyway. Ned's dark hair was wildly awry. His gaunt, young face held befuddled terror. He gasped in the thin atmosphere. "I've gone nuts," he pronounced with a curious calm. "Stark—starin'—nuts...." Loy's box, with its recorded English words and its sonic detectors, could translate for its master, too. As the man spoke, Loy read the illuminated symbols in his own language, flashed on a frosted crystal plate before him. Thus he knew what Ned Vince was saying. Loy Chuk pressed more keys, and the box reproduced his answer: "No, Ned, not nuts. Not a bit of it! There are just a lot of things that you've got to get used to, that's all. You drowned about a million years ago. I discovered your body. I brought you back to life. We have science that can do that. I'm Loy Chuk...." It took only a moment for the box to tell the full story in clear, bold, friendly terms. Thus Loy sought, with calm, human logic, to make his charge feel at home. Probably, though, he was a fool, to suppose that he could succeed, thus. Vince started to mutter, struggling desperately to reason it out. "A prairie dog," he said. "Speaking to me. One million years. Evolution. The scientists say that people grew up from fishes in the sea. Prairie dogs are smart. So maybe super-prairie-dogs could come from them. A lot easier than men from fish...." It was all sound logic. Even Ned Vince knew that. Still, his mind, tuned to ordinary, simple things, couldn't quite realize all the vast things that had happened to himself, and to the world. The scope of it all was too staggeringly big. One million years. God!... Ned Vince made a last effort to control himself. His knuckles tightened on the edge of the vat. "I don't know what you've been talking about," he grated wildly. "But I want to get out of here! I want to go back where I came from! Do you understand—whoever, or whatever you are?" Loy Chuk pressed more keys. "But you can't go back to the Twentieth Century," said the box. "Nor is there any better place for you to be now, than Kar-Rah. You are the only man left on Earth. Those men that exist in other star systems are not really your kind anymore, though their forefathers originated on this planet. They have gone far beyond you in evolution. To them you would be only a senseless curiosity. You are much better off with my people—our minds are much more like yours. We will take care of you, and make you comfortable...." But Ned Vince wasn't listening, now. "You are the only man left on Earth." That had been enough for him to hear. He didn't more than half believe it. His mind was too confused for conviction about anything. Everything he saw and felt and heard might be some kind of nightmare. But then it might all be real instead, and that was abysmal horror. Ned was no coward—death and danger of any ordinary Earthly kind, he could have faced bravely. But the loneliness here, and the utter strangeness, were hideous like being stranded alone on another world! His heart was pounding heavily, and his eyes were wide. He looked across this eerie room. There was a ramp there at the other side, leading upward instead of a stairway. Fierce impulse to escape this nameless lair, to try to learn the facts for himself, possessed him. He bounded out of the vat, and with head down, dashed for the ramp. He had to go most of the way on his hands and knees, for the up-slanting passage was low. Excited animal chucklings around him, and the occasional touch of a furry body, hurried his feverish scrambling. But he emerged at last at the surface. He stood there panting in that frigid, rarefied air. It was night. The Moon was a gigantic, pock-marked bulk. The constellations were unrecognizable. The rodent city was a glowing expanse of shallow, crystalline domes, set among odd, scrub trees and bushes. The crags loomed on all sides, all their jaggedness lost after a million years of erosion under an ocean that was gone. In that ghastly moonlight, the ground glistened with dry salt. "Well, I guess it's all true, huh?" Ned Vince muttered in a flat tone. Behind him he heard an excited, squeaky chattering. Rodents in pursuit. Looking back, he saw the pinpoint gleams of countless little eyes. Yes, he might as well be an exile on another planet—so changed had the Earth become. A wave of intolerable homesickness came over him as he sensed the distances of time that had passed—those inconceivable eons, separating himself from his friends, from Betty, from almost everything that was familiar. He started to run, away from those glittering rodent eyes. He sensed death in that cold sea-bottom, but what of it? What reason did he have left to live? He'd be only a museum piece here, a thing to be caged and studied.... Prison or a madhouse would be far better. He tried to get hold of his courage. But what was there to inspire it? Nothing! He laughed harshly as he ran, welcoming that bitter, killing cold. Nostalgia had him in its clutch, and there was no answer in his hell-world, lost beyond the barrier of the years.... Loy Chuk and his followers presently came upon Ned Vince's unconscious form, a mile from the city of Kar-Rah. In a flying machine they took him back, and applied stimulants. He came to, in the same laboratory room as before. But he was firmly strapped to a low platform this time, so that he could not escape again. There he lay, helpless, until presently an idea occurred to him. It gave him a few crumbs of hope. "Hey, somebody!" he called. "You'd better get some rest, Ned Vince," came the answer from the black box. It was Loy Chuk speaking again. "But listen!" Ned protested. "You know a lot more than we did in the Twentieth Century. And—well—there's that thing called time-travel, that I used to read about. Maybe you know how to make it work! Maybe you could send me back to my own time after all!" Little Loy Chuk was in a black, discouraged mood, himself. He could understand the utter, sick dejection of this giant from the past, lost from his own kind. Probably insanity looming. In far less extreme circumstances than this, death from homesickness had come. Loy Chuk was a scientist. In common with all real scientists, regardless of the species from which they spring, he loved the subjects of his researches. He wanted this ancient man to live and to be happy. Or this creature would be of scant value for study. So Loy considered carefully what Ned Vince had suggested. Time-travel. Almost a legend. An assault upon an intangible wall that had baffled far keener wits than Loy's. But he was bent, now, on the well-being of this anachronism he had so miraculously resurrected—this human, this Kaalleee.... Loy jabbed buttons on the black box. "Yes, Ned Vince," said the sonic apparatus. "Time-travel. Perhaps that is the only thing to do—to send you back to your own period of history. For I see that you will never be yourself, here. It will be hard to accomplish, but we'll try. Now I shall put you under an anesthetic...." Ned felt better immediately, for there was real hope now, where there had been none before. Maybe he'd be back in his home-town of Harwich again. Maybe he'd see the old machine-shop, there. And the trees greening out in Spring. Maybe he'd be seeing Betty Moore in Hurley, soon.... Ned relaxed, as a tiny hypo-needle bit into his arm.... As soon as Ned Vince passed into unconsciousness, Loy Chuk went to work once more, using that pair of brain-helmets again, exploring carefully the man's mind. After hours of research, he proceeded to prepare his plans. The government of Kar-Rah was a scientific oligarchy, of which Loy was a prime member. It would be easy to get the help he needed. A horde of small, grey-furred beings and their machines, toiled for many days. Ned Vince's mind swam gradually out of the blur that had enveloped it. He was wandering aimlessly about in a familiar room. The girders of the roof above were of red-painted steel. His tool-benches were there, greasy and littered with metal filings, just as they had always been. He had a tractor to repair, and a seed-drill. Outside of the machine-shop, the old, familiar yellow sun was shining. Across the street was the small brown house, where he lived. With a sudden startlement, he saw Betty Moore in the doorway. She wore a blue dress, and a mischievous smile curved her lips. As though she had succeeded in creeping up on him, for a surprise. "Why, Ned," she chuckled. "You look as though you've been dreaming, and just woke up!" He grimaced ruefully as she approached. With a kind of fierce gratitude, he took her in his arms. Yes, she was just like always. "I guess I was dreaming, Betty," he whispered, feeling that mighty sense of relief. "I must have fallen asleep at the bench, here, and had a nightmare. I thought I had an accident at Pit Bend—and that a lot of worse things happened.... But it wasn't true ..." Ned Vince's mind, over which there was still an elusive fog that he did not try to shake off, accepted apparent facts simply. He did not know anything about the invisible radiations beating down upon him, soothing and dimming his brain, so that it would never question or doubt, or observe too closely the incongruous circumstances that must often appear. The lack of traffic in the street without, for instance—and the lack of people besides himself and Betty. He didn't know that this machine-shop was built from his own memories of the original. He didn't know that this Betty was of the same origin—a miraculous fabrication of metal and energy-units and soft plastic. The trees outside were only lantern-slide illusions. It was all built inside a great, opaque dome. But there were hidden television systems, too. Thus Loy Chuk's kind could study this ancient man—this Kaalleee. Thus, their motives were mostly selfish. Loy, though, was not observing, now. He had wandered far out into cold, sad sea-bottom, to ponder. He squeaked and chatted to himself, contemplating the magnificent, inexorable march of the ages. He remembered the ancient ruins, left by the final supermen. "The Kaalleee believes himself home," Loy was thinking. "He will survive and be happy. But there was no other way. Time is an Eternal Wall. Our archeological researches among the cities of the supermen show the truth. Even they, who once ruled Earth, never escaped from the present by so much as an instant...." THE END PRINTED IN U. S. A. Transcriber's Note: This etext was produced from Amazing Stories April 1956 and was first published in Amazing Stories November 1942. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
D. The temperature above ground at night is very cold.
|
Why did Eddie’s mother forget to make dinner?
A. Eddie forgot to do some of his chores, so she had to do them for him.
B. Mr. Taylor was injured at work.
C. Mr. Taylor’s isotope was stolen
D. Eddie forgot was home earlier than expected, so sinner wasn’t ready yet.
|
YOUNG READERS Atom Mystery 11 CHAPTER ONE It was only a dream. Eddie Taylor would like to have finished it, but the bar of morning sunlight poking in under the window shade pried his eyes open. The dream fled. Eddie kicked off the sheet, swung his feet to the floor, and groped under the bed for his tennis shoes. He heard his father’s heavy footsteps in the hallway. They stopped outside of his bedroom door. “You awake, Eddie?” “I’m awake, Dad,” Eddie answered. “Breakfast’s ready. Get washed and dressed.” 12 “Be right there,” Eddie said. Then, remembering the dream, he added, “Oh, Dad, is it all right if I use the Geiger counter today?” Mr. Taylor opened the door. He was a big man, broad-shouldered and still thin-waisted. Eddie found it easy to believe the stories he had heard about his father being an outstanding football player in his time. Even his glasses and the gray hair at his temples didn’t add much age, although Eddie knew it had been eighteen years since his father had played his last game of college football. “You may use the Geiger counter any time you want, Eddie,” Mr. Taylor said, “as long as you take good care of it. You figured out where you can find some uranium ore?” Eddie smiled sheepishly. “I—I had a dream,” he said. “Plain as day. It was out on Cedar Point. I was walking along over some rocks. Suddenly the Geiger counter began clicking like everything.” 13 “Cedar Point?” his father asked. “I’ve never been out there. But, from what I hear, there are plenty of rock formations. Might be worth a try, at that. You never can tell where you might strike some radioactivity.” “Do you believe in dreams, Dad?” “Well, now, that’s a tough question, son. I can’t say that I really do. Still, one clue is as good as another when it comes to hunting uranium ore, I guess. But right now we’d better get out to breakfast before your mother scalps us. Hurry it up.” His father turned and went back down the hallway toward the kitchen. Eddie pulled on his trousers and T shirt and went into the bathroom. He washed hurriedly, knowing that even if he missed a spot or two, he was fairly safe. During the summer months his freckles got so thick and dark that it would take a magnifying glass to detect any small smudges of dirt hiding among them. He plastered some water on his dark-red hair, pushed a comb through it, and shrugged as it snapped back almost to its original position. Oh, well, he had tried. 14 He grinned into the mirror, reached a finger into his mouth, and unhooked the small rubber bands from his tooth braces. He dropped them into the waste basket. He’d put fresh ones in after breakfast. He brushed his teeth carefully, taking particular pains around the metal braces. The tooth-straightening orthodontist had warned him about letting food gather around the metal clamps. It could start cavities. Finished, Eddie went out to breakfast. “Good morning, dear,” his mother greeted him, handing him a plate of eggs. “Hi, Mom,” Eddie said. “Gotta hurry. Big day today.” “So your father says. But I’m afraid your big day will have to start with sorting out and tying up those newspapers and magazines that have been collecting in the garage.” “Aw, Mom—” “Eddie, I asked you to do it three days ago. Remember? And the Goodwill truck comes around today.” “But, Mom—” 15 “No arguments, son,” his father put in calmly but firmly. “School vacation doesn’t mean that your chores around here are on vacation, too. Get at it right away, and you’ll still have time to hunt your uranium. “Well,” Mr. Taylor added, excusing himself from the table, “I’d better be getting over to school. I’m expecting to receive shipment of a new radioisotope today.” The very word excited Eddie. In fact, anything having to do with atomic science excited him. He knew something about isotopes—pronounced eye-suh-tope . You couldn’t have a father who was head of the atomic-science department at Oceanview College without picking up a little knowledge along the way. Eddie knew that a radioisotope was a material which had been “cooked” in an atomic reactor until it was “hot” with radioactivity. When carefully controlled, the radiation stored up in such isotopes was used in many beneficial ways. 16 “Why don’t college professors get summer vacations, too?” Eddie asked. One reason for asking that particular question was to keep from prying deeper into the subject of the radioisotope. Much of his father’s work at Oceanview College was of a secret nature. Eddie had learned not to ask questions about it. His father usually volunteered any information he wanted known, so Eddie stuck to questions which could and would be answered. “We get vacations,” his father said. “But—well, my work is a little different, you know. At the speed atomic science is moving today, we simply can’t afford to waste time. But don’t worry. We’ll take a week or so off before school starts in the fall. Maybe head for the mountains with our tent and sleeping bags.” “And Geiger counter?” Eddie asked eagerly. “Wouldn’t think of leaving it home,” his father said, smiling. “By the way, I put new batteries in it the other day. Take it easy on them. Remember to switch it off when you’re not actually using it.” “I will,” Eddie promised. He had forgotten several times before, weakening the batteries. 17 It took Eddie over an hour to sort out the newspapers and magazines in the garage, tie them in neat bundles, and place them out on the front curb for the Goodwill pickup. By that time the sun was high overhead. It had driven off the coolness which the ocean air had provided during the earlier hours. “Anything else, Mom?” he asked, returning to the house and getting the Geiger counter out of the closet. He edged toward the back door before his mother had much time to think of something more for him to do. “I guess not, dear,” Mrs. Taylor said, smiling over his hasty retreat. “What are you going to do?” “Think I’ll do a little prospecting,” Eddie said. “Where?” “Probably in the hills beyond the college,” Eddie said. The more he thought about it, the more he realized it was a little late in the day to go to Cedar Point. The best way to get there was by rowboat across Moon Bay, and that was too long a row to be starting now. Besides, there were plenty of other places around the outskirts of Oceanview where likely looking rock formations invited search with a Geiger counter. 18 “Are you going alone?” his mother asked. “Oh, guess I’ll stop by and see if Teena wants to go,” Eddie answered casually. He tried to make it sound as though he would be doing Teena Ross a big favor. After all, she was only a girl. Eddie didn’t figure a girl would make a very good uranium prospecting partner, but most of the fellows he knew were away at camp, or vacationing with their folks, or something like that. “She’ll enjoy it, I’m sure,” his mother said. “I’ll take Sandy, too,” Eddie said. “He needs the exercise.” “That’s a good idea, dear. Be back in time for an early dinner.” Eddie let Sandy off his chain. The taffy-colored cocker spaniel yipped wildly over his freedom, racing back and forth as Eddie started down the street. 19 Christina Ross—whom everybody called Teena—lived at the far end of the block. Eddie went around to the side door of the light-green stucco house and knocked. “Oh, hi, Eddie,” Teena greeted him, appearing at the screen door. “I was hoping you’d come over.” “Well, I—I just happened to be going by,” Eddie said. “Thought you might want to watch me do a little prospecting with the Geiger counter. But maybe you’re too busy.” That’s how to handle it, Eddie thought. Don’t act anxious. Let Teena be anxious. Then maybe she’ll even offer to bring along a couple of sandwiches or some fruit. “Oh, I’d love to go,” Teena said eagerly, “but I’m just finishing the dishes. Come on in.” “I’m in kind of a hurry.” “I’ll only be a minute.” She pushed the screen door open for him. “I’ll make us some sandwiches.” “Stay here, Sandy,” Eddie said. “Sit.” The dog minded, although he looked a bit rebellious. 20 Eddie went inside and followed Teena to the kitchen. He felt triumphant about the sandwiches. Teena tossed him a dish towel. “You dry them,” she said. “Who, me?” “Why not? You’re in a hurry, aren’t you? I can make the sandwiches while you dry the silverware.” She smiled, putting tiny crinkles in her small, slightly upturned nose. She wore her hair in a pony tail. Even though her hair was blond all year long, it seemed even lighter in the summer. Eddie couldn’t tell whether the sun had faded it, or whether her deep summer tan simply made her hair look lighter by contrast. Maybe both. “Hello, Eddie,” Mrs. Ross said, coming into the kitchen. “Looks like Teena put you to work.” “She always does, Mrs. Ross,” Eddie said, pretending great injury. “Don’t know why I keep coming over here.” “I know,” Teena spoke up quickly. “It’s because we’re friends, that’s why.” 21 Eddie knew she was right. They were friends—good friends. They had been ever since Eddie’s family had moved to Oceanview and his father had become head of the college’s atomic-science department. In fact, their parents were close friends, also. Teena’s father was chief engineer for the Acme Aviation Company, one of the coast town’s largest manufacturing concerns. “Well, I’ll be glad to finish them, Eddie,” Mrs. Ross offered. “I know how boys detest doing dishes.” “Oh, I don’t really mind, Mrs. Ross,” Eddie said. “Besides, Teena’s making sandwiches to take with us.” “Another prospecting trip?” Teena’s mother glanced at the Geiger counter which Eddie had set carefully on the dinette table. “I still think there must be some uranium around here,” Eddie insisted. “And we can find it if anyone can.” “I agree,” Mrs. Ross said. “But even if you don’t find it, you both seem to enjoy your hikes.” 22 “Oh, yes, it’s fun, Mother,” Teena replied, wrapping wax paper around a sandwich. “Guess I’m ready. I’ve got a bone for Sandy, too.” “Don’t go too far out from town,” Mrs. Ross cautioned, as Eddie picked up the Geiger counter. “And stick near the main roads. You know the rules.” “We sure do, Mrs. Ross,” Eddie assured her. “And we’ll be back early.” They walked past the college campus, and toward the rocky foothills beyond. At various rock mounds and outcroppings, Eddie switched on the Geiger counter. The needle of the dial on the black box wavered slightly. A slow clicking came through the earphones, but Eddie knew these indicated no more than a normal background count. There were slight traces of radioactivity in almost all earth or rocks. It was in the air itself, caused by mysterious and ever-present cosmic rays, so there was always a mild background count when the Geiger counter was turned on; but to mean anything, the needle had to jump far ahead on the gauge, and the clicking through the earphones had to speed up until it sounded almost like bacon frying in a hot skillet. 23 There was none of that today. After they had hiked and searched most of the forenoon, Eddie said, “We might as well call it a day, Teena. Doesn’t seem to be anything out here.” “It’s all right with me,” Teena agreed, plucking foxtails from Sandy’s ears. “Pretty hot, anyway. Let’s eat our sandwiches and go back home.” “All right,” Eddie said. “You know, one of these days I’d like to go out to Cedar Point and scout around. Maybe we’ll find something there.” Then he told Teena about his dream. Teena smiled. “A dream sure isn’t much to go on,” she said, “but they say it’s pretty out on Cedar Point. I’ll go any time you want to, Eddie.” She handed him one of the sandwiches. It was midafternoon by the time they arrived back at Teena’s house. They worked a while on a new jigsaw puzzle Teena had received on a recent birthday. Then Eddie said good-by and went on down the street toward his own home. 24 After putting Sandy on his long chain and filling his water dish, Eddie went in the back door. He put the Geiger counter in the closet and went into the kitchen. “What’s for dinner, Mom?” he asked. Mrs. Taylor turned from the sink. Eddie knew at once, just seeing the expression on his mother’s face, that something was wrong. “Dinner?” his mother said absently. “It’s not quite four o’clock yet, Eddie. Besides, dinner may be a little late today.” “But this morning you said it would be early,” Eddie reminded her, puzzled. “This morning I didn’t know what might happen.” 25 Then Eddie heard the sound of his father’s voice coming from the den. There was a strange urgent tone in it. The door to the den was open. Eddie went through the dining room and glanced into the den. His father sat stiffly behind his homemade desk, talking rapidly into the telephone. Eddie caught only the last few sketchy words. Then his father placed the telephone in its cradle, glanced up, and saw Eddie. If there had been even the slightest doubt in Eddie’s mind about something being wrong, it vanished now. Mr. Taylor looked years older than he had that very morning. Worry lay deep in his eyes. He fumbled thoughtfully with a pencil, turning it end over end on his desk. “Hello, son,” he said. He didn’t even ask whether Eddie had discovered any uranium ore that day. Always before, he had shown genuine interest in Eddie’s prospecting trips. “Dad,” Eddie said anxiously, “what—what’s the matter?” “It shows that much, does it, son?” his father said tiredly. “What’s wrong, Dad?” Eddie prompted. “Or can’t you tell me?” Mr. Taylor leaned back. “Quite a bit’s wrong, Eddie,” he said, “and I guess there’s no reason why I shouldn’t tell you. It’ll be in the evening papers, anyway.” 26 “Evening papers?” “Eddie, you remember me mentioning this morning about that radioisotope shipment I was expecting today?” “I remember,” Eddie said. “Did it come?” “It did—and it didn’t,” his father said. “What does that mean, Dad?” Eddie asked, puzzled. “The delivery truck arrived at the school with it,” his father explained, “but while the driver was inquiring where to put it, the container disappeared.” “Disappeared?” “The radioisotope was stolen, Eddie,” his father said slowly. “Stolen right out from under our noses!” 27 CHAPTER TWO At the moment, Eddie didn’t pry for further information on the theft of the valuable radioactive isotope. His father had plenty on his mind, as it was. The main information was in the evening Globe , which Eddie rushed out to get as soon as he heard it plop onto the front porch. He took the newspaper to his father to read first. After having finished, Mr. Taylor handed the paper to Eddie and leaned back thoughtfully in his chair. 28 “They’ve got it pretty straight, at that,” Mr. Taylor said, “but I’m afraid this is going to stir up quite a bit of trouble.” “It wasn’t your fault, was it, Dad?” Eddie defended. “It was as much mine as anybody’s, son,” his father said. “Probably more so. After all, I am head of the department. I knew about the shipment. That should make it my responsibility to see that it was properly received and placed in our atomic-materials storage vault. But there is little point in trying to place the blame on anyone. I’m willing to accept that part of it. The important thing is that we recover that radioisotope. Not only is it of a secret nature, but it is also dangerously radioactive if improperly handled.” “But—but wasn’t it in a safe container?” Eddie asked. 29 “Of course,” his father said. “There were only two ounces of it in a fifty-pound lead capsule. As long as it remains in that capsule it’s safe. As you know, the lead prevents any radiation from escaping. Out of that capsule, however, those two ounces of radioisotope can be very dangerous.” “Fifty pounds,” Eddie said thoughtfully. “That’s a pretty big thing to steal, isn’t it?” “Not when it’s lead, son,” his father replied. “Not much bigger than a two-quart milk bottle, in fact.” “Even at that, no kid could have taken it,” Eddie said. “Kid?” His father smiled thinly. “We don’t think it was any kid, Eddie. Not by a long shot. The whole thing was carefully planned and carefully carried out. It was not the work of amateurs.” Eddie read the newspaper account. The small truck from Drake Ridge, where one of the country’s newest atomic reactors was located, had arrived earlier than expected at Oceanview College. It had backed up to the receiving dock where all of the college supplies were delivered. Since deliveries during vacation months were few, there was no one on the dock when the truck arrived. A half hour later, when the delivery was expected, there would have been. The truck’s early arrival had caught them unprepared. 30 The driver had left the truck and had gone around the building to the front office. It had taken him less than five minutes to locate the receiving-dock foreman. Together, they had returned through the small warehouse and opened the rear door onto the dock. During that short time someone had pried open the heavy padlock on the delivery truck’s rear door and had stolen the fifty-pound lead capsule containing the radioisotope. Dusty footprints on the pavement around the rear of the truck indicated that two men had carried out the theft. A heavy iron pry bar had been dropped at the rear of the truck after the lock was sprung. It was a common type used by carpenters. There were no fingerprints or other identifying marks on it. The footprints were barely visible and of no help other than to indicate that two men were involved in the crime. 31 “Dad,” Eddie asked, looking up from the paper, “how could anyone carry away something weighing fifty pounds without being noticed?” “Chances are they had their car parked nearby,” his father said. “As you know, there are no fences or gates around Oceanview College. People come and go as they please. As a matter of fact, there are always quite a few automobiles parked around the shipping and receiving building, and parking space is scarce even during summer sessions. Anyone could park and wait there unnoticed. Or they could walk around without attracting any undue attention.” “But, Dad,” Eddie continued, “how would the men know that the delivery truck would arrive a half hour early?” “They wouldn’t,” his father said. “They may have had another plan. The way things worked out, they didn’t need to use it. The early delivery and the business of leaving the truck unguarded for a few minutes probably gave them a better opportunity than they had expected. At least, they took quick advantage of it.” 32 “I don’t see what anyone would want with a radioisotope,” Eddie said. “Maybe they figured there was something else inside of that lead capsule.” “That’s unlikely, son,” Mr. Taylor said. “Believe me, it was no common theft. Nor were the thieves ordinary thieves. That isotope was a new one. A very secret one. Our job at the college was to conduct various tests with it in order to find out exactly how it could best be put to use as a cure for disease, or for sterilizing food, or even as a source of power.” “Power?” Eddie said. “Boy, it must have been a strong isotope.” He knew that the strength of radioisotopes could be controlled largely by the length of time they were allowed to “cook” in an atomic reactor and soak up radioactivity. 33 “We weren’t planning to run a submarine with it,” his father said. “It wasn’t that strong. Still, it doesn’t take so very much radioactivity to make two ounces of an isotope quite powerful—and quite deadly. I only hope whoever stole it knows what he’s doing. However, I’m sure he does.” “You mean he must have been an atomic scientist himself?” Eddie asked. “Let’s just say he—or both of them—have enough training in the subject to know how to handle that isotope safely,” Mr. Taylor said. “But, Dad,” Eddie wondered, “what could they do with it?” “They could study it,” his father explained. “At least, they could send it somewhere to be broken down and studied. Being a new isotope, the formula is of great value.” “What do you mean, send it somewhere?” Eddie asked. “Perhaps to some other country.” “Then—then you mean whoever stole it were spies!” Eddie exclaimed breathlessly. “That’s entirely possible,” his father said. “In fact, it’s the only logical explanation I can think of. People simply don’t go around stealing radioactive isotopes without a mighty important reason.” 34 “Dinner’s ready,” Eddie’s mother called from the kitchen. During dinner Eddie wasn’t sure just what he was eating. The idea of spies stealing atomic materials kept building up in his mind. By the time dessert was finished, he was anxious to talk with someone, yet he knew he shouldn’t bother his father with any more questions. He asked if he could go over and visit with Teena for a while. “Well, you were together most of the day,” his mother said, “but I guess it’s all right. Be back in about an hour, though.” It was a balmy evening. On such evenings, he and Teena sometimes walked along the beach barefoot, collecting sea shells. Today Eddie had no desire to do that. He ran down the block. Teena answered his knock. “Come on in, Eddie,” she invited, seeming surprised to see him. “Mother and I are just finishing dinner.” “Oh, I figured you’d be through by now,” Eddie apologized, following her inside. 35 “Hello, Eddie,” Mrs. Ross said, but she didn’t seem as cheerful as usual. “Good evening, Mrs. Ross,” Eddie said. “I—I hope I’m not making a pest of myself.” He looked around for Mr. Ross, but Teena’s father apparently hadn’t arrived home from Acme Aircraft yet. There wasn’t a place set for him at the table, either. “You’re never a pest, Eddie,” Mrs. Ross assured him. “I was going to call your mother in a little while about that newspaper write-up.” “Oh, you read it?” Eddie said. “How could anyone miss it?” Teena said. “Right on the front page.” “I suppose your father is quite concerned over it,” Teena’s mother said. “Oh, yes,” Eddie affirmed. “He was the one who ordered the isotope.” “What’s an isotope?” Teena asked. “I’m not sure I know, either,” Mrs. Ross said. “Maybe we could understand more of what it’s all about if you could explain what a radioisotope is, Eddie.” 36 “Well,” Eddie said slowly, “it’s not easy to explain, but I’ll try. You know how rare uranium is. There’s not nearly enough of it to fill all the needs for radioactive materials. Besides, pure uranium is so powerful and expensive and dangerous to handle that it’s not a very good idea to try using it in its true form. So they build an atomic reactor like the one at Drake Ridge.” “We’ve driven by it,” Mrs. Ross said. “My, it’s a big place.” “I’ll say,” Eddie agreed. “Of course, only one building holds the reactor itself. It’s the biggest building near the center.” “I remember it,” Teena said. “Well, the reactor is about four stories high,” Eddie went on. “They call it a uranium ‘pile.’ It’s made up of hundreds and hundreds of graphite bricks. That’s where they get the name ‘pile’—from brick pile. Anyway, scattered around in between the bricks are small bits of uranium. Uranium atoms are radioactive. That is, they keep splitting up and sending out rays.” “Why do they do that?” Teena asked. 37 “It’s just the way nature made uranium, I guess,” Eddie said. “Most atoms stay in one piece, although they move around lickety-split all of the time. Uranium atoms not only move around, but they break apart. They shoot out little particles called neutrons. These neutrons hit other atoms and split them apart, sending out more neutrons. It’s a regular chain reaction.” “I’ve heard of chain reactions,” Mrs. Ross said. “Well, with all of the splitting up and moving around of the uranium atoms,” Eddie went on, “an awful lot of heat builds up. If they don’t control it—well, you’ve seen pictures of atomic-bomb explosions. That’s a chain reaction out of control.” “Out of control is right,” Teena said. 38 “But the atomic piles control the reaction,” Eddie said. “The graphite bricks keep the splitting-up atoms apart so one neutron won’t go smashing into other atoms unless they want it to. They have ways of controlling it so that only as much radiation builds up as they want. You can even hear the reactor hum as the radioactive rays go tearing through it. But by careful tending, the scientists keep the atomic collisions far enough apart so the thing doesn’t blow up.” “Boy, that sounds dangerous,” Teena said. “Well, they know just how to do it,” Eddie replied. “Aren’t the rays dangerous?” Mrs. Ross asked. “I’ll say they’re dangerous,” Eddie said. “But the whole pile is covered by a shield of concrete about eight feet thick. That keeps the rays from getting out and injuring the workmen.” “Goodness. Eight feet is a lot of cement.” “It takes a lot to stop radioactive atomic particles,” Eddie explained. “Especially the gamma rays. They’re the fastest and most dangerous, and the hardest to stop. Alpha and beta rays are fairly easy to stop. But the gamma rays are regular high-velocity invisible bullets. They’ll go right through a stone wall unless it’s plenty thick. Of course, you can’t see them. Not with even the most powerful microscope in the world.” 39 “I wouldn’t want to work around a place where I might get shot at by—by dangerous rays you can’t even see,” Teena said. “I would,” Eddie said. “Everyone is carefully protected. They see to that. Well, anyway, if all of those uranium atoms were shooting radioactive rays around inside of that pile and doing nothing, there would be an awful lot of energy going to waste. So the atomic scientists take certain elements which aren’t radioactive, but can be made radioactive, and shove small pieces of them into holes drilled in the pile.” “Isn’t that dangerous?” Teena asked. “They don’t shove them in with their bare hands,” Eddie said, trying not to show exasperation. “They use long holders to push the small chunks of material into the holes in the reactor. Then, as those uranium atoms keep splitting up and shooting particles around inside of the pile, some of them smack into the chunks of material, and stick there. Most elements will soak up radiation, just like a sponge soaks up water.” 40 “My, that’s interesting, Eddie,” Mrs. Ross said. “I’ve seen them do it,” Eddie said proudly, then added, “from behind a protective shield, of course. When the material has soaked up enough radiation, they pull it back out. They say it’s ‘cooked.’” “You mean it’s hot?” Teena asked. “It’s hot,” Eddie said, “but not like if it came out of a stove. By hot, they mean it’s radioactive. If you touched it, or even got near it, you would get burned, but you probably wouldn’t even know it for a while. It would be a radiation burn. That’s a kind of burn you don’t feel, but it destroys your blood cells and tissues, and—well, you’ve had it.” “So that’s what a radioisotope is,” Mrs. Ross said. “It’s like a sponge. Only instead of soaking up water, it soaks up radiation.” 41 “That’s about it,” Eddie said. “My dad says that as more is learned about the ways to use isotopes, the whole world is going to be improved. You’ve heard of radiocobalt for curing cancer. Well, that’s an isotope. They make it by cooking cobalt in an atomic reactor. Oh, there are hundreds of different isotopes. Like I said, isotopes can be made of most of the elements. And there are over a hundred elements. Some soak up a lot of radioactivity, and are strong and dangerous. Others absorb only a little and are pretty safe to use. Depends, too, on how long they let them cook in the reactor.” “What kind was the one stolen from the college today?” Teena asked. “Dad didn’t say exactly,” Eddie answered, “except he did say that if whoever took it didn’t know what he was doing and opened up the lead capsule, it could kill him. Of course, even the mild isotopes are deadly if they’re not handled right.” “My goodness, it is a serious matter, isn’t it?” Mrs. Ross said. 42 Eddie nodded. It was even more serious than its threat of danger to anyone who handled it carelessly. It was a new isotope—a secret isotope. His father hadn’t said whether it had been developed for curing things or for destroying things. But many radioisotopes could do either; it depended on how they were used. Eddie assumed that anyone who would stoop to stealing isotopes more than likely would be interested in their ability to destroy rather than their ability to benefit mankind. “Well, I certainly do hope everything works out all right,” Teena’s mother said. “So do I,” Teena agreed. Eddie glanced at the kitchen clock. “Oh, boy,” he said, “I’d better be heading back home. I didn’t mean to come over here and talk so long.” “Oh, we’re glad you did, Eddie,” Mrs. Ross said. “I’m afraid too few of us know anything about this atom business.” 43 “That’s right, Mrs. Ross,” Eddie agreed. “People should talk more and read more about it. After all, this is an atomic age. We might as well face it. My father says that in horse-and-buggy days everyone knew how to feed a horse and grease a wagon wheel. They knew what was needed to get the work done. But now that atoms are being harnessed to do the work, not many people even bother to find out what an atom is.” Mrs. Ross smiled. “I guess you’re right, Eddie,” she said, “but I wouldn’t quite know how to go about feeding an atom.” “Or greasing one,” Teena added. Eddie laughed. “I sure wouldn’t want the job of trying to feed a herd of them the size of a period,” he said. “Did you know that there are about three million billion atoms of carbon in a single period printed at the end of a sentence. That’s how small atoms are.” “Three million billion is a lot of something,” a man’s voice spoke behind him. “What are we talking about, Eddie?” “Oh, hello, Mr. Ross,” Eddie said, turning around and standing up. “I didn’t hear you come in.” 44 Teena’s father was a medium-sized man with light-brown hair which was getting somewhat thin on top. He was usually quite cheerful and full of fun, but tonight his face seemed unusually drawn and sober. He stepped to the table, leaned over, and gave both Teena and Mrs. Ross a kiss on the cheek. “Eddie was telling us about atoms,” Teena’s mother said. “Did you know there were three million billion of them in a period?” “How many in a comma?” Mr. Ross said to Eddie, then added quickly, “forget it, Eddie. It wasn’t very funny. I—I’m afraid I don’t feel very funny tonight.” “Sit down, dear,” Mrs. Ross said. “I’ll warm your dinner. You didn’t sound very cheerful when you called to say you would be late. How did everything go at the plant today?” “Not so good,” Teena’s father said tiredly. “In fact, not good at all.” Problems. It seemed that everyone had problems, Eddie thought, as he started to leave.
|
C. Mr. Taylor’s isotope was stolen
|
What are the baseline models?
|
### Introduction
Humans deploy structure-sensitive expectations to guide processing during natural language comprehension BIBREF0. While it has been shown that neural language models show similar structure-sensitivity in their predictions about upcoming material BIBREF1, BIBREF2, previous work has focused on dependencies that are conditioned by features attached to a single word, such as subject number BIBREF3, BIBREF4 or wh-question words BIBREF5. There has been no systematic investigation into models' ability to compute phrase-level features—features that are attached to a set of words—and whether models can deploy these more abstract properties to drive downstream expectations. In this work, we assess whether state-of-the-art neural models can compute and employ phrase-level gender and number features of coordinated subject Noun Phrases (CoordNPs) with two nouns. Typical syntactic phrases are endocentric: they are headed by a single child, whose features determine the agreement requirements for the entire phrase. In Figure FIGREF1, for example, the word star heads the subject NP The star; since star is singular, the verb must be singular. CoordNPs lack endocentricity: neither conjunct NP solely determines the features of the NP as a whole. Instead, these feature values are determined by compositional rules sensitive to the features of the conjuncts and the identity of the coordinator. In Figure FIGREF1, because the coordinator is and, the subject NP number is plural even though both conjuncts (the star and the moon) are singular. As this case demonstrates, the agreement behavior for CoordNPs must be driven by more abstract, constituent-level representations, and cannot be reduced to features hosted on a single lexical item. We use four suites of experiments to assess whether neural models are able to build up phrase-level representations of CoordNPs on the fly and deploy them to drive humanlike behavior. First, we present a simple control experiment to show that models can represent number and gender features of non-coordinate NPs (Non-coordination Agreement). Second, we show that models modulate their expectations for downstream verb number based on the CoordNP's coordinating conjunction combined with the features of the coordinated nouns (Simple Coordination). We rule out the possibility that models are using simple heuristics by designing a set of stimuli where a simple heuristic would fail due to structural ambiguity (Complex Coordination). The striking success for all models in this experiment indicates that even neural models with no explicit hierarchical bias, trained on a relatively small amount of text are able to learn fine-grained and robust generalizations about the interaction between CoordNPs and local syntactic context. Finally, we use subject–auxiliary inversion to test whether an upstream lexical item modulates model expectation for the phrasal-level features of a downstream CoordNP (Inverted Coordination). Here, we find that all models are insensitive to the fine-grained features of this particular syntactic context. Overall, our results indicate that neural models can learn fine-grained information about the interaction of Coordinated NPs and local syntactic context, but their behavior remains unhumanlike in many key respects. ### Methods ::: Psycholinguistics Paradigm
To determine whether state-of-the-art neural architectures are capable of learning humanlike CoordNP/verb agreement properties, we adopt the psycholinguistics paradigm for model assessment. In this paradigm the models are tested using hand-crafted sentences designed to test underlying network knowledge. The assumption here is that if a model implicitly learns humanlike linguistic knowledge during training, its expectations for upcoming words should qualitatively match human expectations in novel contexts. For example, BIBREF1 and BIBREF6 assessed how well neural models had learned the subject/verb number agreement by feeding them with the prefix The keys to the cabinet .... If the models predicted the grammatical continuation are over the ungrammatical continuation is, they can be said to have learned the number agreement insofar as the number of the head noun and not the number of the distractor noun, cabinet, drives expectations about the number of the matrix verb. If models are able to robustly modulate their expectations based on the internal components of the CoordNP, this will provide evidence that the networks are building up a context-sensitive phrase-level representation. We quantify model expectations as surprisal values. Surprisal is the negative log-conditional probability $S(x_i) = -\log _2 p(x_i|x_1 \dots x_{i-1})$ of a sentence's $i^{th}$ word $x_i$ given the previous words. Surprisal tells us how strongly $x_i$ is expected in context and is known to correlate with human processing difficulty BIBREF7, BIBREF0, BIBREF8. In the CoordNP/Verb agreement studies presented here, cases where the proceeding context sets high expectation for a number-inflected verb form $w_i$, (e.g. singular `is') we would expect $S(w_i)$ to be lower than its number-mismatched counterpart (e.g. plural `are'). ### Methods ::: Models Tested ::: Recurrent Neural Network (RNN) Language Models
are trained to output the probability distribution of the upcoming word given a context, without explicitly representing the structure of the context BIBREF9, BIBREF10. We trained two two-layer recurrent neural language models with long short-term memory architecture BIBREF11 on a relatively small corpus. The first model, referred as `LSTM (PTB)' in the following sections, was trained on the sentences from Penn Treebank BIBREF12. The second model, referred as `LSTM (FTB)', was trained on the sentences from French Treebank BIBREF13. We set the size of input word embedding and LSTM hidden layer of both models as 256. We also compare LSTM language models trained on large corpora. We incorporate two pretrained English language models: one trained on the Billion Word benchmark (referred as `LSTM (1B)') from BIBREF14, and the other trained on English Wikipedia (referred as `LSTM (enWiki)') from BIBREF3. For French, we trained a large LSTM language model (referred as `LSTM (frWaC)') on a random subset (about 4 million sentences, 138 million word tokens) of the frWaC dataset BIBREF15. We set the size of the input embeddings and hidden layers to 400 for the LSTM (frWaC) model since it is trained on a large dataset. ### Methods ::: Models Tested ::: ActionLSTM
models the linearized bracketed tree structure of a sentence by learning to predict the next action required to construct a phrase-structure parse BIBREF16. The action space consists of three possibilities: open a new non-terminal node and opening bracket; generate a terminal node; and close a bracket. To compute surprisal values for a given token, we approximate $P(w_i|w_{1\cdots i-1)}$ by marginalizing over the most-likely partial parses found by word-synchronous beam search BIBREF17. ### Methods ::: Models Tested ::: Generative Recurrent Neural Network Grammars (RNNG)
jointly model the word sequence as well as the underlying syntactic structure BIBREF18. Following BIBREF19, we estimate surprisal using word-synchronous beam search BIBREF17. We use the same hyper-parameter settings as BIBREF18. The annotation schemes used to train the syntactically-supervised models differ slightly between French and English. In the PTB (English) CoordNPs are flat structures bearing an `NP' label. In FTB (French), CoordNPs are binary-branching, labeled as NPs, except for the phrasal node dominating the coordinating conjunction, which is labeled `COORD'. We examine the effects of annotation schemes on model performance in Appendix SECREF8. ### Experiment 1: Non-coordination Agreement
In order to provide a baseline for following experiments, here we assess whether the models tested have learned basic representations of number and gender features for non-coordinated Noun Phrases. We test number agreement in English and French as well as gender agreement in French. Both English and French have two grammatical number feature: singular (sg) and plural (pl). French has two grammatical gender features: masculine (m) and feminine (f). The experimental materials include sentences where the subject NPs contain a single noun which can either match with the matrix verb (in the case of number agreement) or a following predicative adjective (in the case of gender agreement). Conditions are given in Table TABREF9 and Table TABREF10. We measure model behavior by computing the plural expectation, or the surprisal of the singular continuation minus the surprisal of the plural continuation for each condition and took the average for each condition. We expect a positive plural expectation in the Npl conditions and a negative plural expectation in the Nsg conditions. For gender expectation we compute a gender expectation, which is S(feminine continuation) $-$ S(masculine continuation). We measure surprisal at the verbs and predicative adjectives themselves. The results for this experiment are in Figure FIGREF11, with the plural expectation and gender expectation on the y-axis and conditions on the x-axis. For this and subsequent experiments error bars represent 95% confidence intervals for across-item means. For number agreement, all the models in English and French show positive plural expectation when the head noun is plural and negative plural expectation when it is singular. For gender agreement, however, only the LSTM (frWaC) shows modulation of gender expectation based on the gender of the head noun. This is most likely due to the lower frequency of predicative adjectives compared to matrix verbs in the corpus. ### Experiment 2: Simple Coordination
In this section, we test whether neural language models can use grammatical features hosted on multiple components of a coordination phrase—the coordinated nouns as well as the coordinating conjunction—to drive downstream expectations. We test number agreement in both English and French and gender agreement in French. ### Experiment 2: Simple Coordination ::: Number Agreement
In simple subject/verb number agreement, the number features of the CoordNP are determined by the coordinating conjunction and the number features of the two coordinated NPs. CoordNPs formed by and are plural and thus require plural verbs; CoordNPs formed by or allow either plural or singular verbs, often with the number features of the noun linearly closest to the verb playing a more important role, although this varies cross-linguistically BIBREF20. Forced-choice preference experiments in BIBREF21 reveal that English native speakers prefer singular agreement when the closest conjunct in an or-CoordNP is singular and plural agreement when the closest conjunct is plural. In French, both singular and plural verbs are possible when two singular NPs are joined via disjunction BIBREF22. In order to assess whether the neural models learn the basic CoordNP licensing for English, we adapted 37 items from BIBREF21, along the 16 conditions outlined in Table TABREF14. Test items consist of the sentence preamble, followed by either the singular or plural BE verb, half the time in present tense (is/are) and half the time in past tense (was/were). We measured the plural expectation, following the procedure in Section SECREF3. We created 24 items using the same conditions as the English experiment to test the models trained in French, using the 3rd person singular and plural form of verb aller, `to go' (va, vont). Within each item, nouns match in gender; across all conditions half the nouns are masculine, half feminine. The results for this experiment can be seen in Figure FIGREF12, with the results for English on the left and French on the right. The results for and are on the top row, or on the bottom row. For all figures the y-axis shows the plural expectation, or the difference in surprisal between the singular condition and the plural condition. Turning first to English-and (Figure FIGREF12), all models show plural expectation (the bars are significantly greater than zero) in the pl_and_pl and sg_and_pl conditions, as expected. For the pl_and_sg condition, only the LSTM (enWiki) and ActionLSTM are greater than zero, indicating humanlike behavior. For the sg_and_sg condition, only the LSTM (enWiki) model shows the correct plural expectation. For the French-and (Figure FIGREF12), all models show positive plural expectation in all conditions, as expected, except for the LSTM (FTB) in the sg_and_sg condition. Examining the results for English-or, we find that all models demonstrate humanlike expectation in the pl_or_pl and sg_or_pl conditions. The LSTM (1B), LSTM (PTB), and RNNG models show zero or negative singular expectation for the pl_or_sg conditions, as expected. However the LSTM (enWiki) and ActionLSTM models show positive plural expectation in this condition, indicating that they have not learned the humanlike generalizations. All models show significantly negative plural expectation in the sg_or_sg condition, as expected. In the French-or cases, models show almost identical behavior to the and conditions, except the LSTM (frWaC) shows smaller plural expectation when singular nouns are linearly proximal to the verb. These results indicate moderate success at learning coordinate NP agreement, however this success may be the result of an overly simple heuristic. It appears that expectation for both plural and masculine continuations are driven by a linear combination of the two nominal number/gender features transferred into log-probability space, with the earlier noun mattering less than the later noun. A model that optimally captures human grammatical preferences should show no or only slight difference across conditions in the surprisal differential for the and conditions, and be greater than zero in all cases. Yet, all the models tested show gradient performance based on the number of plural conjuncts. ### Experiment 2: Simple Coordination ::: Gender Agreement
In French, if two nouns are coordinated with et (and-coordination), agreement must be masculine if there is one masculine element in the coordinate structure. If the nouns are coordinated with ou (or-coordination), both masculine and feminine agreement is acceptable BIBREF23, BIBREF24. Although linear proximity effects have been tested for a number of languages that employ grammatical gender, as in e.g. Slavic languages BIBREF25, there is no systematic study for French. To assess whether the French neural models learned humanlike gender agreement, we created 24 test items, following the examples in Table TABREF16, and measured the masculine expectation. In our test items, the coordinated subject NP is followed by a predicative adjective, which either takes on masculine or feminine gender morphology. Results from the experiment can be seen in Figure FIGREF17. No models shows qualitative difference based on the coordinator, and only the LSTM (frWaC) shows significant behavior difference between conditions. Here, we find positive masculine expectation in the m_and_m and f_and_m conditions, and negative masculine expectation in the f_and_f condition, as expected. However, in the m_and_f condition, the masculine expectation is not significantly different from zero, where we would expect it to be positive. In the or-coordination conditions, following our expectation, masculine expectation is positive when both conjuncts are masculine and negative when both are feminine. For the LSTM (FTB) and ActionLSTM models, the masculine expectation is positive (although not significantly so) in all conditions, consistent with results in Section SECREF3. ### Experiment 3: Complex Coordination
One possible explanation for the results presented in the previous section is that the models are using a `bag of features' approach to plural and masculine licensing that is opaque to syntactic context: Following a coordinating conjunction surrounded by nouns, models simply expect the following verb to be plural, proportionally to the number of plural nouns. In this section, we control for this potential confound by conducting two experiments: In the Complex Coordination Control experiments we assess models' ability to extend basic CoordNP licensing into sententially-embedded environments, where the CoordNP can serve as an embedded subject. In the Complex Coordination Critical experiments, we leverage the sentential embedding environment to demonstrate that when the CoordNPs cannot plausibly serve as the subject of the embedded phrase, models are able to suppress the previously-demonstrated expectations set up by these phrases. These results demonstrate that models are not following a simple strategy for predicting downstream number and gender features, but are building up CoordNP representations on the fly, conditioned on the local syntactic context. ### Experiment 3: Complex Coordination ::: Complex Coordination Control
Following certain sentential-embedding verbs, CoordNPs serve unambiguously as the subject of the verb's sentence complement and should trigger number agreement behavior in the main verb of the embedded clause, similar to the behavior presented in SECREF13. To assess this, we use the 37 test items in English and 24 items in French in section SECREF13, following the conditions in Table TABREF19 (for number agreement), testing only and coordination. For gender agreement, we use the same test items and conditions for and coordination in Section SECREF15, but with the Coordinated NPs embedded in a context similar to SECREF18. As before, we derived the plural expectation by measuring the difference in surprisal between the singular and plural continuations and the gender expectation by computing the difference in surprisal between the masculine and feminine predicates. . Je croyais que les prix et les dépenses étaient importants/importantes. I thought that the.pl price.mpl and the.pl expense.fpl were important.mpl/fpl I thought that the prices and the expenses were important. The results for the control experiments can be seen in Figure FIGREF20, with English number agreement on the top row, French number agreement in the middle row and French gender agreement on the bottom. The y-axis shows either plural or masculine expectation, with the various conditions along the x-axis. For English number agreement, we find that the models behave similarly as they do for simple coordination contexts. All models show significant plural expectation when the closest noun is plural, with only two models demonstrating plural expectation in the sg_and_sg case. The French number agreement tests show similar results, with all models except LSTM (FTB) demonstrating significant plural prediction in all cases. Turning to French gender agreement, only the LSTM (frWaC) shows sensitivity to the various conditions, with positive masculine expectation in the m_and_m condition and negative expectation in the f_and_f condition, as expected. These results indicate that the behavior shown in Section SECREF13 extends to more complex syntactic environments—in this case to sentential embeddings. Interestingly, for some models, such as the LSTM (1B), behavior is more humanlike when the CoordNP serves as the subject of an embedded sentence. This may be because the model, which has a large number of hidden states and may be extra sensitive to fine-grained syntactic information carried on lexical items BIBREF2, is using the complementizer, that, to drive more robust expectations. ### Experiment 3: Complex Coordination ::: Complex Coordination Critical
In order to assess whether the models' strategy for CoordNP/verb number agreement is sensitive to syntactic context, we contrast the results presented above to those from a second, critical experiment. Here, two coordinated nouns follow a verb that cannot take a sentential complement, as in the examples given in Table TABREF23. Of the two possible continuations—are or is—the plural is only grammatically licensed when the second of the two conjuncts is plural. In these cases, the plural continuation may lead to a final sentence where the first noun serves as the verb's object and the second introduces a second main clause coordinated with the first, as in I fixed the doors and the windows are still broken. For the same reason, the singular-verb continuation is only licensed when the noun immediately following and is singular. We created 37 test items in both English and French, and calculated the plural expectation. If the models were following a simple strategy to drive CoordNP/verb number agreement, then we should see either no difference in plural expectation across the four conditions or behavior no different from the control experiment. If, however, the models are sensitive to the licensing context, we should see a contrast based solely on the number features of the second conjunct, where plural expectation is positive when the second conjunct is plural, and negative otherwise. Experimental items for a critical gender test were created similarly, as in Example SECREF22. As with plural agreement, gender expectation should be driven solely by the second conjunct: For the f_and_m and m_and_m conditions, the only grammatical continuation is one where the adjectival predicate bears masculine gender morphology. Conversely, for the m_and_f or f_and_f conditions, the only grammatical continuation is one where the adjectival predicate bears feminine morphology. As in SECREF13, we created 24 test items and measured the gender expectation by calculating the difference in surprisal between the masculine and feminine continuations. . Nous avons accepté les prix et les dépenses étaient importants/importantes. we have accepted the.pl price.mpl and the expense.fpl were important.mpl/fpl We have accepted the prices and the expenses were important. The results from the critical experiments are in Figure FIGREF21, with the English number agreement on the top row, French number agreement in the middle and gender expectation on the bottom row. Here the y-axis shows either plural expectation or masculine expectation, with the various conditions are on the x-axis. The results here are strikingly different from those in the control experiments. For number agreement, all models in both languages show strong plural expectation in conditions where the second noun is plural (blue and green bars), as they do in the control experiments. Crucially, when the second noun is singular, the plural expectation is significantly negative for all models (save for the French LSTM (FTB) pl_and_sg condition). Turning to gender agreement, only the LSTM (frWaC) model shows differentiation between the four conditions tested. However, whereas the f_and_m and m_and_f gender expectations are not significantly different from zero in the control condition, in the critical condition they pattern with the purely masculine and purely feminine conditions, indicating that, in this syntactic context, the model has successfully learned to base gender expectation solely off of the second noun. These results are inconsistent with a simple `bag of features' strategy that is insensitive to local syntactic context. They indicate that both models can interpret the same string as either a coordinated noun phrase, or as an NP object and the start of a coordinated VP with the second NP as its subject. ### Experiment 4: Inverted Coordination
In addition to using phrase-level features to drive expectation about downstream lexical items, human processors can do the inverse—use lexical features to drive expectations about upcoming syntactic chunks. In this experiment, we assess whether neural models use number features hosted on a verb to modulate their expectations for upcoming CoordNPs. To assess whether neural language models learn inverted coordination rules, we adapted items from Section SECREF13 in both English (37 items) and French (24 items), following the paradigm in Table TABREF24. The first part of the phrase contains either a plural or singular verb and a plural or singular noun. In this case, we sample the surprisal for the continuations and (or is grammatical in all conditions, so it is omitted from this study). Our expectation is that `and' is less surprising in the Vpl_Nsg condition than in the Vsg_Nsg condition, where a CoordNP is not licensed by the grammar in either French or English (as in *What is the pig and the cat eating?). We also expect lower surprisal for and in the Vpl_Nsg condition, where it is obligatory for a grammatical continuation, than in the Vpl_Npl condition, where it is optional. For French experimental items, the question is embedded into a sentential-complement taking verb, following Example SECREF6, due to the fact that unembedded subject-verb inverted questions sound very formal and might be relatively rare in the training data. . Je me demande où vont le maire et I myself ask where go.3PL the.MSG mayor.MSG and The results for both languages are shown in Figure FIGREF25, with the surprisal at the coordinator on the y-axis and the various conditions on the x-axis. No model in either language shows a signficant difference in surprisal between the Vpl_Nsg and Vpl_Npl conditions or between the Vpl_Nsg and Vsg_Nsg conditions. The LSTM (1B) shows significant difference between the Vpl_Nsg and Vpl_Npl conditions, but in the opposite direction than expected, with the coordinator less surprising in the latter condition. These results indicate that the models are unable to use the fine-grained context sensitivity to drive expectations for CoordNPs, at least in the inversion setting. ### Discussion
The experiments presented here extend and refine a line of research investigating what linguistic knowledge is acquired by neural language models. Previous studies have demonstrated that sequential models trained on a simple regime of optimizing the next word can learn long-distance syntactic dependencies in impressive detail. Our results provide complimentary insights, demonstrating that a range of model architectures trained on a variety of datasets can learn fine-grained information about the interaction of CoordNPs and local syntactic context, but their behavior remains unhumanlike in many key ways. Furthermore, to our best knowledge, this work presents the first psycholinguistic analysis of neural language models trained on French, a high-resource language that has so far been under-investigated in this line of research. In the simple coordination experiment, we demonstrated that models were able to capture some of the agreement behaviors of humans, although their performance deviated in crucial aspects. Whereas human behavior is best modeled as a `percolation' process, the neural models appear to be using a linear combination of NP constituent number to drive CoordNP/verb number agreement, with the second noun weighted more heavily than the first. In these experiments, supervision afforded by the RNNG and ActionLSTM models did not translate into more robust or humanlike learning outcomes. The complex coordination experiments provided evidence that the neural models tested were not using a simple `bag of features' strategy, but were sensitive to syntactic context. All models tested were able to interpret material that had similar surface form in ways that corresponded to two different tree-structural descriptions, based on local context. The inverted coordination experiment provided a contrasting example, in which models were unable to modulate expectations based on subtleties in the syntactic environment. Across all our experiments, the French models performed consistently better on subject/verb number agreement than on subject/predicate gender agreement. Although there are likely more examples of subject/verb number agreement in the French training data, gender agreement is syntactically mandated and widespread in French. It remains an open question why all but one of the models tested were unable to leverage the numerous examples of gender agreement seen in various contexts during training to drive correct subject/predicate expectations. ### Acknowledgments
This project is supported by a grant of Labex EFL ANR-10-LABX-0083 (and Idex ANR-18-IDEX-0001) for AA and MIT–IBM AI Laboratory and the MIT–SenseTimeAlliance on Artificial Intelligence for RPL. We would like to thank the anonymous reviewers for their comments and Anne Abeillé for her advice and feedback. ### The Effect of Annotation Schemes
This section further investigates the effects of CoordNP annotation schemes on the behaviors of structurally-supervised models. We test whether an explicit COORD phrasal tag improves model performance. We trained two additional RNNG models on 38,546 sentences from the Penn Treebank annotated with two different schemes: The first, RNNG (PTB-control) was trained with the original Penn Treebank annotation. The second, RNNG (PTB-coord), was trained on the same sentences, but with an extended coordination annotation scheme, meant to employ the scheme employed in the FTB, adapted from BIBREF26. We stripped empty categories from their scheme and only kept the NP-COORD label for constituents inside a coordination structure. Figure FIGREF26 illustrates the detailed annotation differences between two datasets. We tested both models on all the experiments presented in Sections SECREF3-SECREF6 above. Turning to the results of these six experiments: We see little difference between the two models in the Non-coordination agreement experiment. For the Complex coordination control and Complex coordination critical experiments, both models are largely the same as well. However, in the Simple and-coordination and Simple or-coordination experiments the values for all conditions are shifted upwards for the RNNG PTB-coord model, indicating higher over-all preference for the plural continuation. Furthermore, the range of values is reduced in the RNNG PTB-coord model, compared to the RNNG PTB-control model. These results indicate that adding an explicit COORD phrasal label does not drastically change model performance: Both models still appear to be using a linear combination of number features to drive plural vs. singular expectation. However, the explicit representation has made the interior of the coordination phrase more opaque to the model (each feature matters less) and has slightly shifted model preference towards plural continuations. In this sense, the PTB-coord model may have learned a generalization about CoordNPs, but this generalization remains unlike the ones learned by humans. ### PTB/FTB Agreement Patterns
We present statistics of subject/predicate agreement patterns in the Penn Treebank (PTB) and French Treebank (FTB) in Table TABREF28 and TABREF29. Figure 1: Subject-verb agreement with (a) the head of a noun phrase structure, and (b) the coordination structure. Table 1: A summary of models tested. Table 2: Conditions of number agreement in Noncoordination Agreement experiment. Table 3: Conditions of gender agreement in Noncoordination Agreement experiment. Table 4: Conditions of number agreement in Simple Coordination experiment. Figure 2: Non-Coordination Agreement experiments for English (number) and French (number and gender). Figure 3: Comparison of models’ expectation preferences for singular vs. plural predicate in English and French Simple Coordination experiments. Table 5: Conditions for the and-coordination experiment. (Items for or-coordination are the same except that we change the coordinator to ou.) Figure 4: Comparison of models’ expectation preferences for Feminine v.s. Masculine predicative adjectives in French. Table 6: Conditions of number agreement in Complex Coordination Control experiment. Figure 5: Comparison of model’s expectation preferences in the Complex Coordination Control experiments. Figure 6: Comparison of model’s expectation preferences in the Complex Coordination Critical experiments. Table 7: Conditions of number agreement in Complex Coordination Critical experiment. Table 8: Conditions in Inverted Coordination experiment. Figure 7: Comparison of models’ surprisals of andcoordination in Inverted Coordination experiment. Figure 8: Comparison of annotation schemes of coordination structure. Table 9: Frequency of number agreement patterns in PTB and FTB. Table 10: Frequency of gender agreement patterns in FTB. Figure 9: Comparison between RNNGs trained on PTB data with original annotation vs. fine-grained annotation of coordination structure.
|
Recurrent Neural Network (RNN), ActionLSTM, Generative Recurrent Neural Network Grammars (RNNG)
|
What was the cause of Mr. John Chapman's admission to the clinic on 04/03/2017?
Choose the correct answer from the following options:
A. Motor vehicle accident
B. Fall from a height
C. Motocross accident
D. Assault
E. Stroke
|
### Patient Report 0
**Dear colleague, **
We are reporting on our shared patient, Mr. John Chapman, born on
11/16/1994, who received emergency treatment at our clinic on
04/03/2017.
**Diagnoses**:
- Severe open traumatic brain injury with fractures of the cranial
vault, mastoid, and skull base
- Dissection of the distal internal carotid artery on both sides
- Subarachnoid hemorrhage involving both hemispheres and extending
into the basal cisterns
- Aspiration pneumonia
**Other Diagnoses: **
- Status post rib fracture 2005
- Status post appendectomy 2006
- Status post distal radius fracture 2008
- Status post elbow fracture 20010
**Procedure**: External ventricular drain (EVD) placement.
**Medical History:** Admission through the emergency department as a
polytrauma alert. The patient was involved in a motocross accident,
where he jumped, fell, and landed face-first. He was intubated at the
scene, and either during or before intubation, aspiration occurred. No
issues with airway, breathing, or circulation (A, B, or C problems) were
noted. A CT scan performed in the emergency department revealed an open
traumatic brain injury with fractures of the cranial vault, mastoid, and
skull base, as well as dissection of both carotid arteries. Upon
admission, we encountered an intubated and sedated patient with a
Richmond Agitation-Sedation Scale (RASS) score of -4. He was
hemodynamically stable at all times.
**Current Recommendations:**
- Regular checks of vigilance, laboratory values and microbiological
findings.
- Careful balancing
### Patient Report 1
**Dear colleague, **
We report on Mr. John Chapman, born on 11/16/1994, who was admitted to
our Intensive Care Unit from 04/03/2017 to 05/01/2017.
**Diagnoses:**
- Open severe traumatic brain injury with fractures of the skull
vault, mastoid, and skull base
- Dissection of the distal ACI on both sides
- Subarachnoid hemorrhage involving both hemispheres and extending
into basal cisterns
- Infarct areas in the border zone between MCA-ACA on the right
frontal and left parietal sides
- Malresorptive hydrocephalus
<!-- -->
- Rhabdomyolysis
- Aspiration pneumonia
**Other Diagnoses: **
- Status post rib fracture in 2005
- Status post appendectomy in 2006
- Status post distal radius fracture in 2008
- Status post elbow fracture in 20010
**Surgical Procedures:**
- 04/03/2017: Placement of external ventricular drain
- 04/08/2017: Placement of an intracranial pressure monitoring
catheter
- 04/13/2017: Surgical tracheostomy
- 05/01/2017: Left ventriculoperitoneal shunt placement
**Medical History:** The patient was admitted through the emergency
department as a polytrauma alert. The patient had fallen while riding a
motocross bike, landing face-first after jumping. He was intubated at
the scene. Aspiration occurred either during or before intubation. No
problems with breathing or circulation were noted. The CT performed in
the emergency department showed an open traumatic brain injury with
fractures of the skull vault, mastoid, and skull base, as well as
dissection of the carotid arteries on both sides and bilateral
subarachnoid hemorrhage.
Upon admission, the patient was sedated and intubated, with a Richmond
Agitation-Sedation Scale (RASS) score of -4, and was hemodynamically
stable under controlled ventilation.
**Therapy and Progression:**
[Neurology]{.underline}: Following the patient\'s admission, an external
ventricular drain was placed. Reduction of sedation had to be
discontinued due to increased intracranial pressure. A right pupil size
greater than the left showed no intracranial correlate. With
persistently elevated intracranial pressure, intensive intracranial
pressure therapy was initiated using deeper sedation, administration of
hyperosmolar sodium, and cerebrospinal fluid drainage, which normalized
intracranial pressure. Intermittently, there were recurrent intracranial
pressure peaks, which could be treated conservatively. Transcranial
Doppler examinations showed normal flow velocities. Microbiological
samples from cerebrospinal fluid were obtained when the patient had
elevated temperatures, but no bacterial growth was observed. Due to the
inability to adequately monitor intracranial pressure via the external
ventricular drain, an intracranial pressure monitoring catheter was
placed to facilitate adequate intracranial pressure monitoring. In the
perfusion computed tomography, progressive edema with increasingly
obstructed external ventricular spaces and previously known infarcts in
the border zone area were observed. To ensure appropriate intracranial
pressure monitoring, a Tuohy drain was inserted due to cerebrospinal
fluid buildup on 04/21/2017. After the initiation of antibiotic therapy
for suspected ventriculitis, the intracranial pressure monitoring
catheter was removed on 04/20/2017. Subsequently, a liquorrhea
developed, leading to the placement of a Tuohy drain. After successful
antibiotic treatment of ventriculitis, a ventriculoperitoneal shunt was
placed on 05/01/2017 without complications, and the Tuohy drain was
removed. Radiological control confirmed the correct positioning. The
patient gradually became more alert. Both pupils were isochoric and
reacted to light. All extremities showed movement, although the patient
only intermittently responded to commands. On 05/01/2017, a VP shunt was
placed on the left side without complications. Currently, the patient is
sedated with continuous clonidine at 60µg/h.
**Hemodynamics**: To maintain cerebral perfusion pressure in the
presence of increased intracranial pressure, circulatory support with
vasopressors was necessary. Echocardiography revealed preserved cardiac
function without wall motion abnormalities or right heart strain,
despite the increasing need for noradrenaline support. As the patient
had bilateral carotid dissection, a therapy with Aspirin 100mg was
initiated. On 04/16/2017, clinical examination revealed right\>left leg
circumference difference and redness of the right leg. Utrasound
revealed a long-segment deep vein thrombosis in the right leg, extending
from the pelvis (proximal end of the thrombus not clearly delineated) to
the lower leg. Therefore, Heparin was increased to a therapeutic dose.
Heparin therapy was paused on postoperative day 1, and prophylactic
anticoagulation started, followed by therapeutic anticoagulation on
postoperative day 2. The patient was switched to subcutaneous Lovenox.
**Pulmonary**: Due to the history of aspiration in the prehospital
setting, a bronchoscopy was performed, revealing a moderately obstructed
bronchial system with several clots. As prolonged sedation was
necessary, a surgical tracheostomy was performed without complications
on 04/13/2017. Subsequently, we initiated weaning from mechanical
ventilation. The current weaning strategy includes 12 hours of
synchronized intermittent mandatory ventilation (SIMV) during the night,
with nighttime pressure support ventilation (DuoPAP: Ti high 1.3s,
respiratory rate 11/min, Phigh 11 mbar, PEEP 5 mbar, Psupport 5 mbar,
trigger 4l, ramp 50 ms, expiratory trigger sensitivity 25%).
**Abdomen**: FAST examinations did not reveal any signs of
intra-abdominal trauma. Enteral feeding was initiated via a gastric
tube, along with supportive parenteral nutrition. With forced bowel
movement measures, the patient had regular bowel movements. On
04/17/2017, a complication-free PEG (percutaneous endoscopic
gastrostomy) placement was performed due to the potential long-term need
for enteral nutrition. The PEG tube is currently being fed with tube
feed nutrition, with no bowel movement for the past four days.
Additionally, supportive parenteral nutrition is being provided.
**Kidney**: Initially, the patient had polyuria without confirming
diabetes insipidus, and subsequently, adequate diuresis developed.
Retention parameters were within the normal range. As crush parameters
increased, a therapy involving forced diuresis was initiated, resulting
in a significant reduction of crush parameters.
**Infection Course:** Upon admission, with elevated infection parameters
and intermittently febrile temperatures, empirical antibiotic therapy
was initiated for suspected pneumonia using Piperacillin/Tazobactam.
Staphylococcus capitis was identified in blood cultures, and
Staphylococcus aureus was found in bronchial lavage. Both microbes were
sensitive to the current antibiotic therapy, so treatment with
Piperacillin/Tazobactam continued. Additionally, Enterobacter cloacae
was identified in tracheobronchial secretions during the course, also
sensitive to the ongoing antibiotic therapy. On 05/17, the patient
experienced another fever episode with elevated infection parameters and
right lower lobe infiltrates in the chest X-ray. After obtaining
microbiological samples, antibiotic therapy was switched to Meropenem
for suspected pneumonia. Microbiological findings from cerebrospinal
fluid indicated gram-negative rods. Therefore, antibiotic therapy was
adjusted to Ciprofloxacin in accordance with susceptibility testing due
to suspected ventriculitis, and the Meropenem dose was increased. This
led to a reduction in infection parameters. Finally, microbiological
examination of cerebrospinal fluid, blood cultures, and urine revealed
no pathological findings. Infection parameters decreased. We recommend
continuing antibiotic therapy until 05/02/2017.
**Anti-Infective Course: **
- Piperacillin/Tazobactam 04/03/2017-04/16/2017: Staph. Capitis in
Blood Culture Staph. Aureus in Bronchial Lavage
- Meropenem 04/16/2017-present (increased dose since 04/18) CSF:
gram-negative rods in Blood Culture: Pseudomonas aeruginosa
Acinetobacter radioresistens
- Ciprofloxacin 04/18/2017-present CSF: gram-negative rods in Blood
Culture: Pseudomonas aeruginosa, Acinetobacter radioresistens
**Weaning Settings:** Weaning Stage 6: 12-hour synchronized intermittent
mandatory ventilation (SIMV) with DuoPAP during the night (Thigh 1.3s,
respiratory rate 11/min, Phigh 11 mbar, PEEP 5 mbar, Psupport 5 mbar,
trigger 4l, ramp 50 ms, expiratory trigger sensitivity 25%).
**Status at transfer:** Currently, Mr. Chapman is monosedated with
Clonidine. He spontaneously opens both eyes and spontaneously moves all
four extremities. Pupils are bilaterally moderately dilated, round and
sensitive to light. There is bulbar divergence. Circulation is stable
without catecholamine therapy. He is in the process of weaning,
currently spontaneous breathing with intermittent CPAP. Renal function
is sufficient, enteral nutrition via PEG with supportive parenteral
nutrition is successful.
**Current Medication:**
**Medication** **Dosage** **Frequency**
------------------------------------ ---------------- ---------------
Bisoprolol (Zebeta) 2.5 mg 1-0-0
Ciprofloxacin (Cipro) 400 mg 1-1-1
Meropenem (Merrem) 4 g Every 4 hours
Morphine Hydrochloride (MS Contin) 10 mg 1-1-1-1-1-1
Polyethylene Glycol 3350 (MiraLAX) 13.1 g 1-1-1
Acetaminophen (Tylenol) 1000 mg 1-1-1-1
Aspirin 100 mg 1-0-0
Enoxaparin (Lovenox) 30 mg (0.3 mL) 0-0-1
Enoxaparin (Lovenox) 70 mg (0.7 mL) 1-0-1
**Lab results:**
**Parameter** **Results** **Reference Range**
-------------------- ------------- ---------------------
Creatinine (Jaffé) 0.42 mg/dL 0.70-1.20 mg/dL
Urea 31 mg/dL 17-48 mg/dL
Total Bilirubin 0.35 mg/dL \< 1.20 mg/dL
Hemoglobin 7.6 g/dL 13.5-17.0 g/dL
Hematocrit 28% 39.5-50.5%
Red Blood Cells 3.5 M/uL 4.3-5.8 M/uL
White Blood Cells 10.35 K/uL 3.90-10.50 K/uL
Platelets 379 K/uL 150-370 K/uL
MCV 77.2 fL 80.0-99.0 fL
MCH 24.1 pg 27.0-33.5 pg
MCHC 32.5 g/dL 31.5-36.0 g/dL
MPV 11.3 fL 7.0-12.0 fL
RDW-CV 17.7% 11.5-15.0%
Quick 54% 78-123%
INR 1.36 0.90-1.25
aPTT 32.8 sec 25.0-38.0 sec
**Addition: Radiological Findings**
[Clinical Information and Justification:]{.underline} Suspected deep
vein thrombosis (DVT) on the right leg.
[Special Notes:]{.underline} Examination at the bedside in the intensive
care unit, no digital image archiving available.
[Findings]{.underline}: Confirmation of a long-segment deep venous
thrombosis in the right leg, starting in the pelvis (proximal end not
clearly delineated) and extending to the lower leg.
Visible Inferior Vena Cava without evidence of thrombosis.
The findings were communicated to the treating physician.
**Full-Body Trauma CT on 04/03/2017:**
[Clinical Information and Justification:]{.underline} Motocross
accident. Polytrauma alert. Consequences of trauma? Informed consent:
Emergency indication. Recommended monitoring of kidney and thyroid
laboratory parameters.
**Findings**: CCT: Dissection of the distal internal carotid artery on
both sides (left 2-fold).
Signs of generalized elevated intracranial pressure.
Open skull-brain trauma with intracranial air inclusions and skull base
fracture at the level of the roof of the ethmoidal/sphenoidal sinuses
and clivus (in a close relationship to the bilateral internal carotid
arteries) and the temporal
**CT Head on 04/16/2017:**
[Clinical Information and Justification:]{.underline} History of skull
fracture, removal of EVD (External Ventricular Drain). Inquiry about the
course.
[Findings]{.underline}: Regression of ventricular system width (distance
of SVVH currently 41 mm, previously 46 mm) with residual liquor caps,
indicative of regressed hydrocephalus. Interhemispheric fissure in the
midline. No herniation.
Complete regression of subdural hematoma on the left, tentorial region.
Known defect areas on the right frontal lobe where previous catheters
were inserted.
Progression of a newly hypodense demarcated cortical infarct on the
left, postcentral.
Known bilateral skull base fractures involving the petrous bone, with
secretion retention in the mastoid air cells bilaterally. Minimal
secretion also in the sphenoid sinuses.
Postoperative bone fragments dislocated intracranially after right
frontal trepanation.
**Chest X-ray on 04/24/2017.**
[Clinical Information and Justification:]{.underline} Mechanically
ventilated patient. Suspected pneumonia. Question about infiltrates.
[Findings]{.underline}: Several previous images for comparison, last one
from 08/20/2021.
Persistence of infiltrates in the right lower lobe. No evidence of new
infiltrates. Removal of the tracheal tube and central venous catheter
with a newly inserted tracheal cannula. No evidence of pleural effusion
or pneumothorax.
**CT Head on 04/25/2017:**
[Clinical Information and Justification:]{.underline} Severe traumatic
brain injury with brain edema, one External Ventricular Drain removed,
one parenchymal catheter removed; Follow-up.
[Findings]{.underline}: Previous images available, CT last performed on
04/09/17, and MRI on 04/16/17.
Massive cerebrospinal fluid (CSF) stasis supra- and infratentorially
with CSF pressure caps at the ventricular and cisternal levels with
completely depleted external CSF spaces, differential diagnosis:
malresorptive hydrocephalus. The EVD and parenchymal catheter have been
completely removed.
No evidence of fresh intracranial hemorrhage. Residual subdural hematoma
on the left, tentorial. Slight regression of the cerebellar tonsils.
Increasing hypodensity of the known defect zone on the right frontal
region, differential diagnosis: CSF diapedesis. Otherwise, the status is
the same as for the other defects.
Secretion in the sphenoid sinus and mastoid cells bilaterally, known
bilateral skull base fractures.
**Bedside Chest X-ray on 04/262017:**
[Clinical Information and Justification]{.underline}: Respiratory
insufficiency. Inquiry about cardiorespiratory status.
[Findings]{.underline}: Previous image from 08/17/2021.
Left Central Venous Catheter and gastric tube in unchanged position.
Persistent consolidation in the right para-hilar region, differential
diagnosis: contusion or partial atelectasis. No evidence of new
pulmonary infiltrates. No pleural effusion. No pneumothorax. No
pulmonary congestion.
**Brain MRI on 04/26/2017:**
[Clinical Information and Justification:]{.underline} Severe skull-brain
trauma with skull calvarium, mastoid, and skull base fractures.
Assessment of infarct areas/edema for rehabilitation planning.
[Findings:]{.underline} Several previous examinations available.
Persistent small sulcal hemorrhages in both hemispheres (left \> right)
and parenchymal hemorrhage on the left frontal with minimal perifocal
edema.
Narrow subdural hematoma on the left occipital extending tentorially (up
to 2 mm).
No current signs of hypoxic brain damage. No evidence of fresh ischemia.
Slightly regressed ventricular size. No herniation. Unchanged placement
of catheters on the right frontal side. Mastoid air cells blocked
bilaterally due to known bilateral skull base fractures, mucosal
swelling in the sphenoid and ethmoid sinuses. Polypous mucosal swelling
in the left maxillary sinus. Other involved paranasal sinuses and
mastoids are clear.
**Bedside Chest X-ray on 04/27/2017:**
[Clinical Information and Justification:]{.underline} Tracheal cannula
placement. Inquiry about the position.
[Findings]{.underline}: Images from 04/03/2017 for comparison.
Tracheal cannula with tip projecting onto the trachea. No pneumothorax.
Regressing infiltrate in the right lower lung field. No leaking pleural
effusions.
Left ubclavian central venous catheter with tip projecting onto the
superior vena cava. Gastric tube in situ.
**CT Head on 04/28/2017:**
[Clinical Information and Justification:]{.underline} Open head injury,
bilateral subarachnoid hemorrhage (SAH), EVD placement. Inquiry about
herniation.
[Findings]{.underline}: Comparison with the last prior examination from
the previous day.
Generalized signs of cerebral edema remain constant, slightly
progressing with a somewhat increasing blurred cortical border,
particularly high frontal.
Essentially constant transtentorial herniation of the midbrain and low
position of the cerebellar tonsils. Marked reduction of inner CSF spaces
and depleted external CSF spaces, unchanged position of the ventricular
drainage catheter with the tip in the left lateral ventricle.
Constant small parenchymal hemorrhage on the left frontal and constant
SDH at the tentorial edge on both sides. No evidence of new intracranial
space-occupying hemorrhage.
Slightly less distinct demarcation of the demarcated infarcts/defect
zones, e.g., on the right frontal region, differential diagnosis:
fogging.
**CT Head Angiography with Perfusion on 04/28/2017:**
[Clinical Information and Justification]{.underline}: Post-traumatic
head injury, rising intracranial pressure, bilateral internal carotid
artery dissection. Inquiry about intracranial bleeding, edema course,
herniation, brain perfusion.
[Emergency indication:]{.underline} Vital indication. Recommended
monitoring of kidney and thyroid laboratory parameters. Consultation
with the attending physician from and the neuroradiology service was
conducted.
[Technique]{.underline}: Native moderately of the neurocranium. CT
angiography of brain-supplying cervical intracranial vessels during
arterial contrast agent phase and perfusion imaging of the neurocranium
after intravenous injection of a total of 140 ml of Xenetix-350. DLP
Head 502.4 mGy*cm. DLP Body 597.4 mGy*cm.
[Findings]{.underline}: Previous images from 08/11/2021 and the last CTA
of the head/neck from 04/03/2017 for comparison.
[Brain]{.underline}: Constant bihemispheric and cerebellar brain edema
with a slit-like appearance of the internal and completely compressed
external ventricular spaces. Constant compression of the midbrain with
transtentorial herniation and a constant tonsillar descent.
Increasing demarcation of infarct areas in the border zone of MCA-ACA on
the right frontal, possibly also on the left frontal. Predominantly
preserved cortex-gray matter contrast, sometimes discontinuous on both
frontal sides, differential diagnosis: artifact-related, differential
diagnosis: disseminated infarct demarcations/contusions.
Unchanged placement of the ventricular drainage from the right frontal
with the catheter tip in the left lateral ventricle anterior horn.
Constant subdural hematoma tentorial and posterior falx. Increasingly
vague delineation of the small frontal parenchymal hemorrhage. No new
space-occupying intracranial bleeding.
No evidence of secondary dislocation of the skull base fracture with
constant fluid collections in the paranasal sinuses and mastoid air
cells. Hematoma possible, cerebrospinal fluid leakage possible.
[CT Angiography Head/Neck]{.underline}: Constant presentation of
bilateral internal carotid artery dissection.
No evidence of higher-grade vessel stenosis or occlusion of the
brain-supplying intracranial arteries.
Moderately dilated venous collateral circuits in the cranial soft
tissues on both sides, right \> left. Moderately dilated ophthalmic
veins on both sides, right \> left.
No evidence of sinus or cerebral venous thrombosis. Slight perfusion
deficits in the area of the described infarct areas and contusions.
No evidence of perfusion mismatches in the perfusion imaging.
Unchanged presentation of the other documented skeletal segments.
Additional Note: Discussion of findings with the responsible medical
colleagues on-site and by telephone, as well as with the neuroradiology
service by telephone, was conducted.
**CT Head on 04/30/2017:**
[Clinical Information and Justification]{.underline}: Open head injury
following a motorcycle accident.. Inquiry about rebleeding, edema, EVD
displacement.
[Findings and Assessment:]{.underline} CT last performed on 04/05/2017
for comparison.
Constant narrow subdural hematoma on both sides, tentorial and posterior
parasagittal. Constant small parenchymal hemorrhage on the left frontal.
No new intracranial bleeding.
Progressively demarcated infarcts on the right frontal and left
parietal.
Slightly progressive compression of the narrow ventricles as an
indication of progressive edema. Completely depleted external CSF spaces
with the ventricular drain catheter in the left lateral ventricle.
Increasing compression of the midbrain due to transtentorial herniation,
progressive tonsillar descent of 6 mm.
Fracture of the skull base and the petrous part of the temporal bone on
both sides without significant displacement. Hematoma in the mastoid and
sphenoid sinuses and the maxillary sinus.
**CT Head on 05/01/2017:**
[Clinical Information and Justification:]{.underline} Open skull-brain
trauma. Inquiry about CSF stasis, bleeding, edema.
[Findings]{.underline}: CT last performed on 04/05/17 for comparison.
Completely regressed subarachnoid hemorrhages on both sides. Minimal SDH
components on the tentorial edges bilaterally (left more than right,
with a 3 mm margin width). No new intracranial bleeding. Continuously
narrow inner ventricular system and narrow basal cisterns. The fourth
ventricle is unfolded. Narrow external CSF spaces and consistently
swollen gyration with global cerebral edema.
Better demarcated circumscribed hypodensity in the centrum semiovale on
the right (Series 3, Image 176) and left (Series 3, Image 203);
Differential diagnosis: fresh infarcts due to distal ACI dissections.
Consider repeat vascular imaging. No midline shift. No herniation.
Regressing intracranial air inclusions. Fracture of the skull base and
the petrous part of the temporal bone on both sides without significant
displacement. Hematoma in the maxillary, sphenoidal, and ethmoidal
sinuses.
**Consultation Reports:**
**1) Consultation with Ophthalmology on 04/03/2017**
[Patient Information:]{.underline}
- Motorbike accident, heavily contaminated eyes.
- Request for assessment.
**Diagnosis:** Motorbike accident
**Findings:** Patient intubated, unresponsive. In cranial CT, the
eyeball appears intact, no retrobulbar hematoma. Intraocular pressure:
Right/left within the normal range. Eyelid margins of both eyes crusty
with sand, inferiorly in the lower lid sac, and on the upper lid with
sand. Lower lid somewhat chemotic. Slight temporal hyperemia in the left
eyelid angle. Both eyes have erosions, small, multiple, superficial.
Lower conjunctival sac clean. Round pupils, anisocoria right larger than
left. Left iris hyperemia, no iris defects in the direct light. Lens
unremarkable. Reduced view of the optic nerve head due to miosis,
somewhat pale, rather sharp-edged, central neuroretinal rim present,
central vessels normal. Left eye, due to narrow pupil, limited view,
optic nerve head not visible, central vessels normal, no retinal
hemorrhages.
**Assessment:** Eyelid and conjunctival foreign bodies removed. Mild
erosions in the lower conjunctival sac. Right optic nerve head somewhat
pale, rather sharp-edged.
**Current Recommendations:**
- Antibiotic eye drops three times a day for both eyes.
- Ensure complete eyelid closure.
**2) Consultation with Craniomaxillofacial (CMF) Surgery on 04/05/2017**
**Patient Information:**
- Motorbike accident with severe open traumatic brain injury with
fractures of the cranial vault, mastoid, and skull base
<!-- -->
- Request for assessment.
- Patient with maxillary fracture.
**Findings:** According to the responsible attending physician,
\"minimal handling in case of decompensating intracranial pressure\" is
indicated. Therefore, currently, a cautious approach is suggested
regarding surgical intervention for the radiologically hardly displaced
maxillary fracture. Re-consultation is possible if there are changes in
the clinical outcome.
**Assessment:** Awaiting developments.
**3) Consultation with Neurology on 04/06/2017**
**Patient Information:**
- Brain edema following a severe open traumatic brain injury with
fractures of the cranial vault, mastoid, and skull base
<!-- -->
- Request for assessment.
- Traumatic subarachnoid hemorrhage, intracranial artery dissection,
and various other injuries.
**Findings:** Patient comatose, intubated, sedated. Isocoric pupils. No
light reaction in either eye. No reaction to pain stimuli for
vestibulo-ocular reflex and oculomotor responses. Babinski reflex
negative.
**Assessment:** Long-term ventilation due to a history of intracerebral
bleeding and skull base fracture. No response to pain stimuli or light
reactions in the eyes.
**Procedure/Therapy Suggestion:** Monitoring of patient condition.
**4) Consultation with ENT on 04/16/2017**
**Patient Information:** Tracheostomy tube change.
**Findings:** Tracheostomy tube change performed. Stoma unremarkable.
Trachea clear up to the bifurcation. Sutures in place.
**Assessment:** Re-consultation on 08/27/2021 for suture removal.
**5) Consultation with Neurology on 04/22/2017**
**Patient Information:** Adduction deficit., Request for assessment.
**Findings:** Long-term ventilation due to a history of intracerebral
bleeding and skull base fracture. Adduction deficit in the right eye and
horizontal nystagmus.
**Assessment:** Suspected mesencephalic lesion due to horizontal
nystagmus, but no diagnostic or therapeutic action required.
**6) Consultation with ENT on 04/23/2017**
**Patient Information:** Suture removal. Request for assessment.
**Findings:** Tracheostomy site unremarkable. Sutures trimmed, and skin
sutures removed.
**Assessment:** Procedure completed successfully.
Please note that some information is clinical and may not include
specific dates or recommendations for further treatment.
**Antibiogram:**
**Antibiotic** **Organism 1 (Pseudomonas aeruginosa)** **Organism 2 (Acinetobacter radioresistens)**
------------------------- ----------------------------------------- -----------------------------------------------
Aztreonam I (4.0) \-
Cefepime I (2.0) \-
Cefotaxime \- \-
Amikacin S (\<=2.0) S (4.0)
Ampicillin \- \-
Piperacillin I (\<=4.0) \-
Piperacillin/Tazobactam I (8.0) \-
Imipenem I (2.0) S (\<=0.25)
Meropenem S (\<=0.25) S (\<=0.25)
Ceftriaxone \- \-
Ceftazidime I (4.0) \-
Gentamicin . (\<=1.0) S (\<=1.0)
Tobramycin S (\<=1.0) S (\<=1.0)
Cotrimoxazole \- S (\<=20.0)
Ciprofloxacin I (\<=0.25) I (0.5)
Moxifloxacin \- \-
Fosfomycin \- \-
Tigecyclin \- \-
\"S\" means Susceptible
\"I\" means Intermediate
\".\" indicates not specified
\"-\" means Resistant
### Patient Report 2
**Dear colleague, **
We are reporting on our mutual patient, Mr. John Chapman, born on
11/16/1994, who presented himself to our Outpatient Clinic from
08/08/2018.
**Diagnoses**:
- Right abducens Nerve Palsy and Facial Nerve Palsy
- Lagophthalmos with corneal opacities due to eyelid closure deficit
- Left Abducens Nerve Palsy with slight compensatory head leftward
rotation and preferred leftward gaze
- Bilateral disc swelling
- Suspected left cavernous internal carotid artery aneurysm following
traumatic ICA dissection
- History of shunt explantation due to dysfunction and right-sided
re-implantation (Codman, current pressure setting 12 cm H2O)
- History of left VP shunt placement (programmable
ventriculoperitoneal shunt, initial pressure setting 5/25 cm H2O,
adjusted to 3 cm H2O before discharge)
- Malresorptive hydrocephalus
- History of severe open head injury in a motocross accident with
multiple skull fractures and distal dissection
**Procedure**: We conducted the following preoperative assessment:
- Visual acuity: Distant vision: Right eye: 0.5, Left eye: 0.8p
- Eye position: Fusion/Normal with significant esotropia in the right
eye; no fusion reflex observed
- Ocular deviation: After CT, at distance, esodeviation simulating
alternating 100 prism diopters (overcorrection); at near,
esodeviation simulating alternating 90 prism diopters
- Head posture: Fusion/Normal with leftward head turn of 5-10 degrees
- Correspondence: Bagolini test shows suppression at both distance and
near fixation
- Motility: Right eye abduction limited to 25 degrees from the
midline, abduction in up and down gaze limited to 30 degrees from
midline; left eye abduction limited to 30 degrees
- Binocular functions: Bagolini test shows suppression in the right
eye at both distance and near fixation; Lang I negative
**Current Presentation:** Mr. Chapman presented himself today in our
neurovascular clinic, providing an MRI of the head.
**Medical History:** The patient is known to have a pseudoaneurysm of
the cavernous left internal carotid artery following traumatic carotid
dissection in 04/2017, along with ipsilateral abducens nerve palsy.
**Physical Examination:** Patient in good general condition. Oriented in
all aspects. No cyanosis. No edema. Warm and dry skin. Normal nasal and
pharyngeal findings. Pupils round, equal, and react promptly to light
bilaterally. Moist tongue. Pharynx and buccal mucosa unremarkable. No
jugular vein distension. No carotid bruits heard. Palpation of lymph
nodes unremarkable. Palpation of the thyroid gland unremarkable, freely
movable. Lungs: Normal chest shape, moderately mobile, vesicular breath
sounds. Heart: Regular heart action, normal rate; heart sounds clear, no
pathological sounds. Abdomen: Peristalsis and bowel sounds normal in all
quadrants; soft abdomen, no tenderness, no palpable masses, liver and
spleen not palpable due to limited access, non-tender kidneys. Normal
peripheral pulses; joints freely movable. Strength, motor function, and
sensation are unremarkable.
**Therapy and Progression:** The pseudoaneurysm has shown slight
enlargement in the recent follow-up imaging and remains partially
thrombosed. The findings were discussed on during a neurovascular board
meeting, where a recommendation for endovascular treatment was made,
which the patient has not yet pursued. Since Mr. Chapman has not been
able to decide on treatment thus far, it is advisable to further
evaluate this still asymptomatic condition through a diagnostic
angiography. This examination would also help in better planning any
potential intervention. Mr. Chapman agreed to this course of action, and
we will provide him with a timely appointment for the angiography.
**Lab results upon Discharge:**
**Parameter** **Results** **Reference Range**
-------------------- ------------- ---------------------
Creatinine (Jaffé) 0.44 mg/dL 0.70-1.20 mg/dL
Urea 31 mg/dL 17-48 mg/dL
Total Bilirubin 0.35 mg/dL \< 1.20 mg/dL
Hemoglobin 7.8 g/dL 13.5-17.0 g/dL
Hematocrit 28% 39.5-50.5%
Red Blood Cells 3.5 M/uL 4.3-5.8 M/uL
White Blood Cells 10.35 K/uL 3.90-10.50 K/uL
Platelets 379 K/uL 150-370 K/uL
MCV 77.2 fL 80.0-99.0 fL
MCH 24.1 pg 27.0-33.5 pg
MCHC 32.5 g/dL 31.5-36.0 g/dL
MPV 11.3 fL 7.0-12.0 fL
RDW-CV 17.7% 11.5-15.0%
Quick 54% 78-123%
INR 1.36 0.90-1.25
aPTT 32.8 sec 25.0-38.0 sec
### Patient Report 3
**Dear colleague, **
We are reporting on our patient, Mr. John Chapman, born on 11/16/1994,
who was under our inpatient care from 05/25/2019 to 05/26/2019.
**Diagnoses: **
- Pseudoaneurysm of the cavernous left internal carotid artery
following traumatic carotid dissection
- Abducens nerve palsy.
- History of severe open head trauma with fractures of the cranial
vault, mastoid, and skull base. Distal ICA dissection bilaterally.
Bilateral hemispheric subarachnoid hemorrhage extending into the
basal cisterns.mInfarct areas in the MCA-ACA border zones, right
frontal, and left parietal. Malresorptive hydrocephalus.
<!-- -->
- Rhabdomyolysis.
- History of aspiration pneumonia.
- Suspected Propofol infusion syndrome.
**Current Presentation:** For cerebral digital subtraction angiography
of the intracranial vessels. The patient presented with stable
cardiopulmonary conditions.
**Medical History**: The patient was admitted for the evaluation of a
pseudoaneurysm of the supra-aortic vessels. Further medical history can
be assumed to be known.
**Physical Examination:** Patient in good general condition. Oriented in
all aspects. No cyanosis. No edema. Warm and dry skin. Normal nasal and
pharyngeal findings. Pupils round, equal, and react promptly to light
bilaterally. Moist tongue. Pharynx and buccal mucosa unremarkable. No
jugular vein distension. No carotid bruits heard. Palpation of lymph
nodes unremarkable. Palpation of the thyroid gland unremarkable, freely
movable. Lungs: Normal chest shape, moderately mobile, vesicular breath
sounds. Heart: Regular heart action, normal rate; heart sounds clear, no
pathological sounds. Abdomen: Peristalsis and bowel sounds normal in all
quadrants; soft abdomen, no tenderness, no palpable masses, liver and
spleen not palpable due to limited access, non-tender kidneys. Normal
peripheral pulses; joints freely movable. Strength, motor function, and
sensation are unremarkable.
**Supra-aortic angiography on 05/25/2019:**
[Clinical context, question, justifying indication:]{.underline}
Pseudoaneurysm of the left ICA. Written consent was obtained for the
procedure. Anesthesia, Medications: Procedure performed under local
anesthesia. Medications: 500 IU Heparin in 500 mL NaCl for flushing.
[Methodology]{.underline}: Puncture of the right common femoral artery
under local anesthesia. 4F sheath, 4F vertebral catheter. Serial
angiographies after selective catheterization of the internal carotid
arteries. Uncomplicated manual intra-arterial contrast medium injection
with a total of 50 mL of Iomeron 300. Post-interventional closure of the
puncture site by manual compression. Subsequent application of a
circular pressure bandage.
[Technique]{.underline}: Biplanar imaging technique, area dose product
1330 cGy x cm², fluoroscopy time 3:43 minutes.
[Findings]{.underline}: The perfused portion of the partially thrombosed
cavernous aneurysm of the left internal carotid artery measures 4 x 2
mm. No evidence of other vascular pathologies in the anterior
circulation.
[Recommendation]{.underline}: In case of post-procedural bleeding,
immediate manual compression of the puncture site and notification of
the on-call neuroradiologist are advised.
- Pressure bandage to be kept until 2:30 PM. Bed rest until 6:30 PM.
- Follow-up in our Neurovascular Clinic
**Addition: Doppler ultrasound of the right groin on 05/26/2019:**
[Clinical context, question, justifying indication:]{.underline} Free
fluid? Hematoma?
[Findings]{.underline}: A CT scan from 04/05/2017 is available for
comparison. No evidence of a significant hematoma or an aneurysm in the
right groin puncture site. No evidence of an arteriovenous fistula.
Normal flow profiles of the femoral artery and vein. No evidence of
thrombosis.
**Treatment and Progression:** Pre-admission occurred on 05/24/2019 due
to a medically justified increase in risk for DSA of intracranial
vessels. After appropriate preparation, the angiography was performed on
05/25/2019. The puncture site was managed with a pressure bandage. In
the color Doppler sonographic control the following day, neither a
puncture aneurysm nor an arteriovenous fistula was detected. On
05/25/2019, we discharged the patient in good subjective condition for
your outpatient follow-up care.
**Current Recommendations:** Outpatient follow-up
**Lab results:**
**Parameter** **Reference Range** **Result**
----------------------- --------------------- -------------
Sodium 136-145 mEq/L 141 mEq/L
Potassium 3.5-4.5 mEq/L 4.9 mEq/L
Chloride 98-107 mEq/L 100 mEq/L
Osmolality 280-300 mOsm/kg 290 mOsm/kg
Glucose in Fluoride 60-110 mg/dL 76 mg/dL
Creatinine (Jaffé) 0.70-1.20 mg/dL 0.98 mg/dL
CRP \< 5.0 mg/L 4.5 mg/L
Triglycerides \< 150 mg/dL 119 mg/dL
Creatine Kinase \< 190 U/L 142 U/L
Free Triiodothyronine 2.00-4.40 ng/L 3.25 ng/L
Free Thyroxine 9.30-17.00 ng/L 14.12 ng/L
TSH Basal 0.27-4.20 mU/L 1.65 mU/L
Hemoglobin 13.5-17.0 g/dL 14.3 g/dL
Hematocrit 39.5-50.5% 43.4%
Erythrocytes 4.3-5.8 M/uL 5.6 M/uL
Leukocytes 3.90-10.50 K/uL 10.25 K/uL
Platelets 150-370 K/uL 198 K/uL
MCV 80.0-99.0 fL 83.2 fL
MCH 27.0-33.5 pg 28.1 pg
MCHC 31.5-36.0 g/dL 33.4 g/dL
MPV 7.0-12.0 fL 11.6 fL
RDW-CV 11.5-15.0% 13.5%
Quick \> 78% 90%
INR \< 1.25 1.07
aPTT 25.0-38.0 sec 36.1 sec
|
Motocross accident
|
Approximately how long was Stinson on the planet before he decided it was home?
A. 12 hours
B. 24 hours
C. 36 hours
D. 48 hours
|
THE GOD NEXT DOOR By BILL DOEDE Illustrated by IVIE [Transcriber's Note: This etext was produced from Galaxy Magazine August 1961. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The sand-thing was powerful, lonely and strange. No doubt it was a god—but who wasn't? Stinson lay still in the sand where he fell, gloating over the success of his arrival. He touched the pencil-line scar behind his ear where the cylinder was buried, marveling at the power stored there, power to fling him from earth to this fourth planet of the Centaurian system in an instant. It had happened so fast that he could almost feel the warm, humid Missouri air, though he was light years from Missouri. He got up. A gray, funnel-shaped cloud of dust stood off to his left. This became disturbing, since there was scarcely enough wind to move his hair. He watched it, trying to recall what he might know about cyclones. But he knew little. Weather control made cyclones and other climatic phenomena on earth practically non-existent. The cloud did not move, though, except to spin on its axis rapidly, emitting a high-pitched, scarcely audible whine, like a high speed motor. He judged it harmless. He stood on a wide valley floor between two mountain ranges. Dark clouds capped one peak of the mountains on his left. The sky was deep blue. He tested the gravity by jumping up and down. Same as Earth gravity. The sun—no, not the sun. Not Sol. What should he call it, Alpha or Centaurus? Well, perhaps neither. He was here and Earth was somewhere up there. This was the sun of this particular solar system. He was right the first time. The sun burned fiercely, although he would have said it was about four o'clock in the afternoon, if this had been Earth. Not a tree, nor a bush, nor even a wisp of dry grass was in sight. Everywhere was desert. The funnel of sand had moved closer and while he watched it, it seemed to drift in the wind—although there was no wind. Stinson backed away. It stopped. It was about ten feet tall by three feet in diameter at the base. Then Stinson backed away again. It was changing. Now it became a blue rectangle, then a red cube, a violet sphere. He wanted to run. He wished Benjamin were here. Ben might have an explanation. "What am I afraid of?" he said aloud, "a few grains of sand blowing in the wind? A wind devil?" He turned his back and walked away. When he looked up the wind devil was there before him. He looked back. Only one. It had moved. The sun shone obliquely, throwing Stinson's shadow upon the sand. The wind devil also had a shadow, although the sun shone through it and the shadow was faint. But it moved when the funnel moved. This was no illusion. Again Stinson felt the urge to run, or to use the cylinder to project himself somewhere else, but he said, "No!" very firmly to himself. He was here to investigate, to determine if this planet was capable of supporting life. Life? Intelligence? He examined the wind devil as closely as he dared, but it was composed only of grains of sand. There was no core, no central place you could point to and say, here is the brain, or the nervous system. But then, how could a group of loosely spaced grains of sand possibly have a nervous system? It was again going through its paces. Triangle, cube, rectangle, sphere. He watched, and when it became a triangle again, he smoothed a place in the sand and drew a triangle with his forefinger. When it changed to a cube he drew a square, a circle for a sphere, and so on. When the symbols were repeated he pointed to each in turn, excitement mounting. He became so absorbed in doing this that he failed to notice how the wind devil drew closer and closer, but when he inhaled the first grains of sand, the realization of what was happening dawned with a flash of fear. Instantly he projected himself a thousand miles away. Now he was in an area of profuse vegetation. It was twilight. As he stood beside a small creek, a chill wind blew from the northwest. He wanted to cover himself with the long leaves he found, but they were dry and brittle, for here autumn had turned the leaves. Night would be cold. He was not a woodsman. He doubted if he could build a fire without matches. So he followed the creek to where it flowed between two great hills. Steam vapors rose from a crevice. A cave was nearby and warm air flowed from its mouth. He went inside. At first he thought the cave was small, but found instead that he was in a long narrow passageway. The current of warm air flowed toward him and he followed it, cautiously, stepping carefully and slowly. Then it was not quite so dark. Soon he stepped out of the narrow passageway into a great cavern with a high-vaulted ceiling. The light source was a mystery. He left no shadow on the floor. A great crystal sphere hung from the ceiling, and he was curious about its purpose, but a great pool of steaming water in the center of the cavern drew his attention. He went close, to warm himself. A stone wall surrounding the pool was inscribed with intricate art work and indecipherable symbols. Life. Intelligence. The planet was inhabited. Should he give up and return to earth? Or was there room here for his people? Warming his hands there over the great steaming pool he thought of Benjamin, and Straus, and Jamieson—all those to whom he had given cylinders, and who were now struggling for life against those who desired them. He decided it would not be just, to give up so easily. The wide plaza between the pool and cavern wall was smooth as polished glass. Statues lined the wall. He examined them. The unknown artist had been clever. From one angle they were animals, from another birds, from a third they were vaguely humanoid creatures, glowering at him with primitive ferocity. The fourth view was so shocking he had to turn away quickly. No definable form or sculptured line was visible, yet he felt, or saw—he did not know which senses told him—the immeasurable gulf of a million years of painful evolution. Then nothing. It was not a curtain drawn to prevent him from seeing more. There was no more. He stumbled toward the pool's wall and clutched for support, but his knees buckled. His hand slid down the wall, over the ancient inscriptions. He sank to the floor. Before he lost consciousness he wondered, fleetingly, if a lethal instrument was in the statue. He woke with a ringing in his ears, feeling drugged and sluggish. Sounds came to him. He opened his eyes. The cavern was crowded. These creatures were not only humanoid, but definitely human, although more slight of build than earth people. The only difference he could see at first sight was that they had webbed feet. All were dressed from the waist down only, in a shimmering skirt that sparkled as they moved. They walked with the grace of ballet dancers, moving about the plaza, conversing in a musical language with no meaning for Stinson. The men were dark-skinned, the women somewhat lighter, with long flowing hair, wide lips and a beauty that was utterly sensual. He was in chains! They were small chains, light weight, of a metal that looked like aluminum. But all his strength could not break them. They saw him struggling. Two of the men came over and spoke to him in the musical language. "My name is Stinson," he said, pointing to himself. "I'm from the planet Earth." They looked at each other and jabbered some more. "Look," he said, "Earth. E-A-R-T-H, Earth." He pointed upward, described a large circle, then another smaller, and showed how Earth revolved around the sun. One of the men poked him with a stick, or tube of some kind. It did not hurt, but angered him. He left the chains by his own method of travel, and reappeared behind the two men. They stared at the place where he had been. The chains tinkled musically. He grasped the shoulder of the offender, spun him around and slapped his face. A cry of consternation rose from the group, echoing in the high ceilinged cavern. "SBTL!" it said, "ZBTL ... XBTL ... zbtl." The men instantly prostrated themselves before him. The one who had poked Stinson with the stick rose, and handed it to him. Still angered, Stinson grasped it firmly, with half a notion to break it over his head. As he did so, a flash of blue fire sprang from it. The man disappeared. A small cloud of dust settled slowly to the floor. Disintegrated! Stinson's face drained pale, and suddenly, unaccountably, he was ashamed because he had no clothes. "I didn't mean to kill him!" he cried. "I was angry, and...." Useless. They could not understand. For all he knew, they might think he was threatening them. The object he had thought of as a stick was in reality a long metal tube, precisely machined, with a small button near one end. This weapon was completely out of place in a culture such as this. Or was it? What did he know of these people? Very little. They were humanoid. They had exhibited human emotions of anger, fear and, that most human of all characteristics, curiosity. But up to now the tube and the chain was the only evidence of an advanced technology, unless the ancient inscriptions in the stone wall of the pool, and the statues lining the wall were evidences. There was a stirring among the crowd. An object like a pallet was brought, carried by four of the women. They laid it at his feet, and gestured for him to sit. He touched it cautiously, then sat. Instantly he sprang to his feet. There, at the cavern entrance, the wind devil writhed and undulated in a brilliant harmony of colors. It remained in one spot, though, and he relaxed somewhat. One of the women came toward him, long golden hair flowing, firm breasts dipping slightly at each step. Her eyes held a language all their own, universal. She pressed her body against him and bore him to the pallet, her kisses fire on his face. Incongruously, he thought of Benjamin back on earth, and all the others with cylinders, who might be fighting for their lives at this moment. He pushed her roughly aside. She spoke, and he understood! Her words were still the same gibberish, but now he knew their meaning. Somehow he knew also that the wind devil was responsible for his understanding. "You do not want me?" she said sadly. "Then kill me." "Why should I kill you?" She shrugged her beautiful shoulders. "It is the way of the Gods," she said. "If you do not, then the others will." He took the tube-weapon in his hands, careful not to touch the button. "Don't be afraid. I didn't mean to kill the man. It was an accident. I will protect you." She shook her head. "One day they will find me alone, and they'll kill me." "Why?" She shrugged. "I have not pleased you." "On the contrary, you have. There is a time and place for everything, though." Suddenly a great voice sounded in the cavern, a voice with no direction. It came from the ceiling, the floor, the walls, the steaming pool. It was in the language of the web-footed people; it was in his own tongue. "No harm must come to this woman. The God with fingers on his feet has decreed this." Those in the cavern looked at the woman with fear and respect. She kissed Stinson's feet. Two of the men came and gave her a brilliant new skirt. She smiled at him, and he thought he had never seen a more beautiful face. The great, bodiless voice sounded again, but those in the cavern went about their activities. They did not hear. "Who are you?" Stinson looked at the wind devil, since it could be no one else speaking, and pointed to himself. "Me?" "Yes." "I am Stinson, of the planet Earth." "Yes, I see it in your mind, now. You want to live here, on this planet." "Then you must know where I came from, and how." "I do not understand how. You have a body, a physical body composed of atoms. It is impossible to move a physical body from one place to another by a mere thought and a tiny instrument, yet you have done so. You deserted me out in the desert." "I deserted you?" Stinson cried angrily, "You tried to kill me!" "I was attempting communication. Why should I kill you?" He was silent a moment, looking at the people in the cavern. "Perhaps because you feared I would become the God of these people in your place." Stinson felt a mental shrug. "It is of no importance. When they arrived on this planet I attempted to explain that I was not a God, but the primitive is not deeply buried in them. They soon resorted to emotion rather than reason. It is of no importance." "I'd hardly call them primitive, with such weapons." "The tube is not of their technology. That is, they did not make it directly. These are the undesirables, the incorrigibles, the nonconformists from the sixth planet. I permit them here because it occupies my time, to watch them evolve." "You should live so long." "Live?" the wind devil said. "Oh, I see your meaning. I'd almost forgotten. You are a strange entity. You travel by a means even I cannot fully understand, yet you speak of time as if some event were about to take place. I believe you think of death. I see your physical body has deteriorated since yesterday. Your body will cease to exist, almost as soon as those of the sixth planet peoples. I am most interested in you. You will bring your people, and live here." "I haven't decided. There are these web-footed people, who were hostile until they thought I was a God. They have destructive weapons. Also, I don't understand you. I see you as a cone of sand which keeps changing color and configuration. Is it your body? Where do you come from? Is this planet populated with your kind?" The wind devil hesitated. "Where do I originate? It seems I have always been. You see this cavern, the heated pool, the statues, the inscriptions. Half a million years ago my people were as you. That is, they lived in physical bodies. Our technology surpassed any you have seen. The tube these webfoots use is a toy by comparison. Our scientists found the ultimate nature of physical law. They learned to separate the mind from the body. Then my people set a date. Our entire race was determined to free itself from the confines of the body. The date came." "What happened?" "I do not know. I alone exist. I have searched all the levels of time and matter from the very beginning. My people are gone. Sometimes it almost comes to me, why they are gone. And this is contrary to the greatest law of all—that an entity, once in existence, can never cease to exist." Stinson was silent, thinking of the endless years of searching through the great gulf of time. His eyes caught sight of the woman, reclining now on the pallet. The men had left her and stood in groups, talking, glancing at him, apparently free of their awe and fear already. The woman looked at him, and she was not smiling. "Please ask the Sand God," she said, "to speak to my people again. Their fear of him does not last. When He is gone they will probably kill us." "As for the webfoots," the wind devil, or Sand God, said, "I will destroy them. You and your people will have the entire planet." "Destroy them?" Stinson asked, incredulously, "all these people? They have a right to live like any one else." "Right? What is it—'right?' They are entities. They exist, therefore they always will. My people are the only entities who ever died. To kill the body is unimportant." "No. You misunderstand. Listen, you spoke of the greatest law. Your law is a scientific hypothesis. It has to do with what comes after physical existence, not with existence itself. The greatest law is this, that an entity, once existing, must not be harmed in any way. To do so changes the most basic structure of nature." The Sand God did not reply. The great bodiless, directionless voice was silent, and Stinson felt as if he had been taken from some high place and set down in a dark canyon. The cone of sand was the color of wood ashes. It pulsed erratically, like a great heart missing a beat now and then. The web-footed people milled about restlessly. The woman's eyes pleaded. When he looked back, the Sand God was gone. Instantly a new note rose in the cavern. The murmur of unmistakable mob fury ran over the webfoots. Several of the men approached the woman with hatred in their voices. He could not understand the words now. But he understood her. "They'll kill me!" she cried. Stinson pointed the disintegrating weapon at them and yelled. They dropped back. "We'll have to get outside," he told her. "This mob will soon get out of hand. Then the tube won't stop them. They will rush in. I can't kill them all at once, even if I wanted to. And I don't." Together they edged toward the cavern entrance, ran quickly up the inclined passageway, and came out into crisp, cold air. The morning sun was reflected from a million tiny mirrors on the rocks, the trees and grass. A silver thaw during the night had covered the whole area with a coating of ice. Stinson shivered. The woman handed him a skirt she had thoughtfully brought along from the cavern. He took it, and they ran down the slippery path leading away from the entrance. From the hiding place behind a large rock they watched, as several web-footed men emerged into the sunlight. They blinked, covered their eyes, and jabbered musically among themselves. One slipped and fell on the ice. They re-entered the cave. Stinson donned the shimmering skirt, smiling as he did so. The others should see him now. Benjamin and Straus and Jamieson. They would laugh. And Ben's wife, Lisa, she would give her little-girl laugh, and probably help him fasten the skirt. It had a string, like a tobacco pouch, which was tied around the waist. It helped keep him warm. He turned to the woman. "I don't know what I'll do with you, but now that we're in trouble together, we may as well introduce ourselves. My name is Stinson." "I am Sybtl," she said. "Syb-tl." He tried to imitate her musical pronunciation. "A very nice name." She smiled, then pointed to the cavern. "When the ice is gone, they will come out and follow us." "We'd better make tracks." "No," she said, "we must run, and make no tracks." "Okay, Sis," he said. "Sis?" "That means, sister." "I am not your sister. I am your wife." " What? " "Yes. When a man protects a woman from harm, it is a sign to all that she is his chosen. Otherwise, why not let her die? You are a strange God." "Listen, Sybtl," he said desperately, "I am not a God and you are not my wife. Let's get that straight." "But...." "No buts. Right now we'd better get out of here." He took her hand and they ran, slid, fell, picked themselves up again, and ran. He doubted the wisdom of keeping her with him. Alone, the webfoots were no match for him. He could travel instantly to any spot he chose. But with Sybtl it was another matter; he was no better than any other man, perhaps not so good as some because he was forty, and never had been an athlete. How was he to decide if this planet was suitable for his people, hampered by a woman, slinking through a frozen wilderness like an Indian? But the woman's hand was soft. He felt strong knowing she depended on him. Anyway, he decided, pursuit was impossible. They left no tracks on the ice. They were safe, unless the webfoots possessed talents unknown to him. So they followed the path leading down from the rocks, along the creek with its tumbling water. Frozen, leafless willows clawed at their bodies. The sun shone fiercely in a cloudless sky. Already water ran in tiny rivulets over the ice. The woman steered him to the right, away from the creek. Stinson's bare feet were numb from walking on ice. Christ, he thought, what am I doing here, anyway? He glanced down at Sybtl and remembered the webfoots. He stopped, tempted to use his cylinder and move to a warmer, less dangerous spot. The woman pulled on his arm. "We must hurry!" He clutched the tube-weapon. "How many shots in this thing?" "Shots?" "How often can I use it?" "As often as you like. It is good for fifty years. Kaatr—he is the one you destroyed—brought it from the ship when we came. Many times he has used it unwisely." "When did you come?" "Ten years ago. I was a child." "I thought only criminals were brought here." She nodded. "Criminals, and their children." "When will your people come again?" She shook her head. "Never. They are no longer my people. They have disowned us." "And because of me even those in the cavern have disowned you." Suddenly she stiffened beside him. There, directly in their path, stood the Sand God. It was blood red now. It pulsed violently. The great voice burst forth. "Leave the woman!" it demanded angrily. "The webfoots are nearing your position." "I cannot leave her. She is helpless against them." "What form of primitive stupidity are you practicing now? Leave, or they will kill you." Stinson shook his head. The Sand God pulsed more violently than before. Ice melted in a wide area around it. Brown, frozen grass burned to ashes. "You will allow them to kill you, just to defend her life? What business is it of yours if she lives or dies? My race discarded such primitive logic long before it reached your level of development." "Yes," Stinson said, "and your race no longer exists." The Sand God became a sphere of blue flame. A wave of intense heat drove them backward. "Earthman," the great voice said, "go back to your Earth. Take your inconsistencies with you. Do not come here again to infect my planet with your primitive ideas. The webfoots are not as intelligent as you, but they are sane. If you bring your people here, I shall destroy you all." The sphere of blue fire screamed away across the frozen wilderness, and the thunder of its passing shook the ground and echoed among the lonely hills. Sybtl shivered against his arm. "The Sand God is angry," she said. "My people tell how he was angry once before, when we first came here. He killed half of us and burned the ship that brought us. That is how Kaatr got the tube-weapon. It was the only thing the Sand God didn't burn, that and the skirts. Then, when he had burned the ship, the Sand God went to the sixth planet and burned two of the largest cities, as a warning that no more of us must come here." Well, Stinson said to himself, that does it. We are better off on Earth. We can't fight a monster like him. Sybtl touched his arm. "Why did the Sand God come? He did not speak." "He spoke to me." "I did not hear." "Yes, I know now. His voice sounds like thunder in the sky, but it is a voice that speaks only in the mind. He said I must leave this planet." She glanced at him with suddenly awakened eyes, as if thinking of it for the first time. "Where is your ship?" "I have no ship." "Then he will kill you." She touched her fingers on his face. "I am sorry. It was all for me." "Don't worry. The Sand God travels without a ship, why shouldn't I?" "Now?" "As soon as you are safe. Come." Steam rose from the burned area, charred like a rocket launching pit. They stepped around it carefully. Stinson felt warm air, but there was no time, now, to warm cold feet or dwell on the vagaries of Sand Gods. Together they crossed the narrow valley. Sybtl led him toward a tall mound of rock. Here they came to the creek again, which flowed into a small canyon. They climbed the canyon wall. Far away, small figures moved. The webfoots were on their trail. She drew him into a small cave. It was heated, like the great cavern, but held no walled pool nor mysterious lighting. But it was warm, and the small entrance made an excellent vantage point for warding off attack. "They will not find us...." A high-pitched keening burst suddenly around them. Stinson knew they had heard, or felt the sound for some time, that now its frequency was in an audible range. "The Sand God," Sybtl said. "Sometimes he plays among the clouds. He makes it rain in a dry summer, or sometimes warms the whole world for days at a time in winter, so the snow melts and the grass begins to green. Then he tires and lets winter come back again. He is the loneliest God in the universe." "What makes you think he's lonely?" She shrugged her shoulders. "I just know. But he's an angry God now. See those clouds piling in the East? Soon they will hide the sun. Then he will make them churn and boil, like river whirlpools in spring. At least he does this when he plays. Who knows what he will do when he's angry?" "The Sand God isn't doing this," Stinson said. "It's only a storm." She covered his lips with her fingers. "Don't say that. He may hear you and be more angry." "But it is, don't you see? You give him powers he does not possess." Sybtl shook her head and stroked his face with her long, slim fingers. "Poor little God-with-fingers-on-his-feet," she said. "You do not understand. The Sand God is terrible, even when he plays. See the lightning? It is blue. The lightning of a storm that comes by itself is not blue. He is running around the world on feet like the rockets of space ships, and when he strikes the clouds, blue fire shoots away." The clouds continued to build on one another. Soon the blue flashes of lightning extended across the sky from horizon to horizon. The earth trembled. Sybtl moved closer, trembling also. "He never did this before," she said. "He never made the earth shake before." Great boulders crashed down the canyon walls and dropped into the creek. They dared not move from the cave, although death seemed certain if they stayed. "I'll leave for a moment," he said. "I'll be back soon." "You're leaving?" There was panic in her voice. "Only for a moment." "And you won't come back. You will go to your world." "No. I'll be back." "Promise? No, don't promise. The promises of Gods often are forgotten before the sounds die away." "I'll be back." He disappeared at once, giving her no chance to object again, and went to the desert of sand, where he had first arrived on the planet. He wanted to see if the storm were world-wide. Stinson had never been in a sand storm before, even on Earth. He could not breathe. He could not see. Bullets of sand stung his skin. Bullets of sand shot into his eyes. Clouds of sand howled around him. He fell, and the wind rolled him over and over in the sand like a tumbleweed. The skirt flew up around his face. He could not get up again. He returned to the cave. Soon after, while they sat huddled together, watching the chaos of tumbling rocks, lightning, and driving rain, the high-pitched keening came again. A sphere of blue fire appeared in the east. Its brilliance put the lightning to shame. It bore down on the cave swiftly, purposefully. Stinson prepared himself to leave. In spite of his desire to protect Sybtl, it was useless to get himself killed when he was powerless to help her. But at the last moment it veered off. "Fiend!" Stinson screamed the word, vaguely marvelling at his own fury. The blue sphere turned and came back. "Monster!" Again. "Murderer!" "Adolescent!" This time it kept going. The rain and wind ceased. Lightning stopped. Thunder rumbled distantly. Clouds disappeared. Stinson and Sybtl emerged from the cave. There was no longer a question of attack from the webfoots, the storm had taken care of that. The fierce sun began its work of drying rocks and throwing shadows and coaxing life out into the open again. Down in the canyon a bird sang, a lonely, cheerful twitter. "The Sand God is tired," Sybtl said. "He is not angry now. I'm glad. Perhaps he will let you stay." "No. Even if he allowed it, I couldn't stay. My people could never live here with a God who is half devil." The cone of sand suddenly appeared. It stood in the canyon, its base on a level with the cave. It was quiet. It was dull gray in color. It exuded impressions of death, of hopeful words solemnly spoken over lowered coffins, of cold earth and cold space, of dank, wet catacombs, of creeping, crawling nether things. The bird's twitter stopped abruptly. "Earthman," the Sand God said, as if he were about to make a statement. Stinson ignored him. He glanced down at Sybtl, who sensed that this was a time for good-bys. He thought, perhaps I can stay here alone with her. The webfoots might find us, or the Sand God might destroy us in one of his fits, but it might be worth it. "Don't go," she said. "Not yet." "Earthman, hear me." "I hear you." "Why does your mind shrink backward?" "I've decided not to bring my people here." " You decided?" "Certainly," Stinson said boldly. "Call it rationalization, if you wish. You ordered us away; and I have several good reasons for not coming here if the door was open." "I've changed my mind. You will be welcomed." "Listen to that, will you?" Stinson said angrily. "Just listen! You set yourself up as a God for the webfoots. You get them eating out of your hand. Then what do you do? You throw a fit. Yes, a fit! Like an adolescent. Worse." "Earthman, wait...." "No!" Stinson shot back. "You've owned this planet for a million years. You have brooded here alone since before my people discovered fire, and in all those ages you never learned self-control. I can't subject my people to the whims of an entity who throws a planetary fit when it pleases him." Stinson relaxed. He'd had his say. Sybtl trembled beside him. A small mammal, round, furry, hopped by, sniffing inquisitively. Sybtl said, "Is the Sand God happy?" She shook her head. "No, he is not happy. He is old, old, old. I can feel it. My people say that when one gets too old it is well to die. But Gods never die, do they? I would not like to be a God." "Stinson," the Sand God said. "You said I was adolescent. You are correct. Do you remember I told you how my people, the entire race, left their bodies at the same time? Do you imagine all of us were adults?" "I suppose not. Sounds reasonable. How old were you?" "Chronologically, by our standards, I was nine years old." "But you continued to develop after...." "No." Stinson tried to imagine it. At first there must have been a single voice crying into a monstrous emptiness, "Mother, where are you? MOTHER! Where is everyone ?" A frenzied searching of the planet, the solar system, the galaxy. Then a returning to the planet. Empty.... Change. Buildings, roads, bridges weathering slowly. Such a race would have built of durable metal. Durable? Centuries, eons passed. Buildings crumbled to dust, dust blew away. Bridges eroded, fell, decomposed into basic elements. The shape of constellations changed. All trace of civilization passed except in the cavern of the heated pool. Constellations disappeared, new patterns formed in the night sky. The unutterably total void of time—FIVE HUNDRED THOUSAND YEARS! And a nine-year-old child brooding over an empty world. "I don't understand why your development stopped," Stinson said. "Nor do I. But perhaps ... well, I sense that I would continue, if you brought your people here. You have already taught me the value of life. There is a oneness, a bond that ties each living thing to every other living thing. It is a lesson my people never knew. Select any portion of this planet that suits you. Take the web-footed woman for your wife. Have children. I promise never to harm you in any way." "The webfoots?" "You and they shall share the planet." The Sand God disappeared. Sybtl said; "Is the Sand God angry again?" "No, he is not angry." "I'm glad. You will leave now?" "No. This is my home." She laughed softly. "You are a strange God." "Listen," he said, "I am not a God. Get that through your head." She drew him into the cave. Her lips were cool and sweet. The cave was pleasantly warm.
|
B. 24 hours
|
What programming language is the tool written in?
|
### Introduction
Entity extraction is one of the most major NLP components. Most NLP tools (e.g., NLTK, Stanford CoreNLP, etc.), including commercial services (e.g., Google Cloud API, Alchemy API, etc.), provide entity extraction functions to recognize named entities (e.g., PERSON, LOCATION, ORGANIZATION, etc.) from texts. Some studies have defined fine-grained entity types and developed extraction methods BIBREF0 based on these types. However, these methods cannot comprehensively cover domain-specific entities. For instance, a real estate search engine needs housing equipment names to index these terms for providing fine-grained search conditions. There is a significant demand for constructing user-specific entity dictionaries, such as the case of cuisine and ingredient names for restaurant services. A straightforward solution is to prepare a set of these entity names as a domain-specific dictionary. Therefore, this paper focuses on the entity population task, which is a task of collecting entities that belong to an entity type required by a user. We develop LUWAK, a lightweight tool for effective interactive entity population. The key features are four-fold: We think these features are key components for effective interactive entity population. We choose an interactive user feedback strategy for entity population for LUWAK. A major approach to entity population is bootstrapping, which uses several entities that have been prepared as a seed set for finding new entities. Then, these new entities are integrated into the initial seed set to create a new seed set. The bootstrapping approach usually repeats the procedure until it has collected a sufficient number of entities. The framework cannot prevent the incorporation of incorrect entities that do not belong to the entity type unless user interaction between iterations. The problem is commonly called semantic drift BIBREF1 . Therefore, we consider user interaction, in which feedback is given to expanded candidates, as essential to maintaining the quality of an entity set. LUWAK implements fundamental functions for entity population, including (a) importing an initial entity set, (b) generating entity candidates, (c) obtaining user feedback, and (d) publishing populated entity dictionary. We aim to reduce the user’s total workload as a key metric of an entity population tool. That is, an entity population tool should provide the easiest and fastest solution to collecting entities of a particular entity type. User interaction cost is a dominant factor in the entire workload of an interactive tool. Thus, we carefully design the user interface for users to give feedbacks to the tool intuitively. Furthermore, we also consider the end-to-end user cost reduction. We adhere to the concept of developing installation-free software to distribute the tool among a wide variety of users, including nontechnical clusters. This lightweight design of LUWAK might speed up the procedure of the whole interactive entity population workflow. Furthermore, this advantage might be beneficial to continuously improve the whole pipeline of interactive entity population system. ### LUWAK: A lightweight tool for interactive entity population
Our framework adopts the interactive entity expansion approach. This approach organizes the collaboration of a human worker and entity expansion algorithms to generate a user-specific entity dictionary efficiently. We show the basic workflow of LUWAK in Figure 1 . (Step 1) LUWAK assumes that a user prepares an initial seed set manually. The seed set is shown in the Entity table. (Step 2) A user can send entities in the Entity table to an Expansion API for obtaining entity candidates. (Step 3) LUWAK shows the entity candidates in the Candidate table for user interaction. Then, the user checks accept/reject buttons to update the Entity table. After submitting the judgments, LUWAK shows the Entity table again. The user can directly add, edit, or delete entities in the table at any time. (Step 4) the user can also easily see how these entities stored in the Entity table appear in a document. (Step 5) After repeating the same procedure (Steps 2–4) for a sufficient time, the user can publish the Entity table as an output. ### Implementation
LUWAK is implemented in pure JavaScript code, and it uses the LocalStorage of a web browser. A user does not have to install any packages except for a web browser. The only thing the user must do is to download the LUWAK software and put it in a local directory. We believe the cost of installing the tool will keep away a good amount of potential users. This philosophy follows our empirical feeling of the curse of installing required packages/libraries. Moreover, LUWAK does not need users to consider the usage and maintenance of these additional packages. That is why we are deeply committed to making LUWAK a pure client-side tool in the off-the-shelf style. ### LUWAK Dashboard
LUWAK has a dashboard for quickly viewing an entity dictionary in progress. The dashboard consists of two tables: the Entity table and the Feedback table. The Entity table provides efficient ways to construct and modify an entity dictionary. Figure UID11 shows the screenshot of the Entity table. The table shows entities in the current entity set. Each row corresponds to an entity entry. Each entry has a label, which denotes whether the predefined entity type is a positive or a negative example, an original entity, which was used to find the entity, and the score, which denotes the confidence score. A user can directly edit the table by adding, renaming, and deleting entities. Moreover, the entity inactivation function allows a user to manually inactivate entities, so that entity expansion algorithms do not use the inactivated entities. The table implements a page switching function, a search function, and a sorting function to ensure visibility even when there is a large number of entities in the table. ### Entity Candidate Generation
We design the entity candidate generation module as an external API (Expansion API). The Expansion API receives a set of entities with positive labels. The Expansion API returns top- $k$ entity candidates. As an initial implementation, we used GloVe BIBREF2 as word embedding models for implementing an Expansion API. This API calculates the cosine similarity between a set of positive entities and entities candidates to generate a ranked list. We prepared models trained based on the CommonCrawl corpus and the Twitter corpus. Note that the specification of the expansion algorithm is not limited to the algorithm described in this paper, as LUWAK considers the Expansion API as an external function. Moreover, we also utilize the category-based expansion module, in which we used is-a relationship between the ontological category and each entity and expanded seeds via category-level. For example, if most of the entities already inserted in the dictionary share the same category, such as Programming Languages, the system suggests that "Programming Language" entities should be inserted in the dictionary when we develop a job skill name dictionary. Category-based entity expansion is helpful to avoid the candidate entity one by one. We used Yago BIBREF3 as an existing knowledge base. External API. In our design of LUWAK, Expansion APIs are placed as an external function outside LUWAK. There are three reasons why we adopt this design. First, we want LUWAK to remain a corpus-free tool. Users do not have to download any corpora or models to start using LUWAK, and it takes too much time to launch an Expansion API server. Second, LUWAK’s design allows external contributors to build their own expansion APIs that are compatible with LUWAK’s interface. We developed the initial version of the LUWAK package to contain an entity Expansion API so users can launch their expansion APIs internally. Third, the separation between LUWAK and the Expansion APIs enables Expansion APIs to use predetermined options for algorithms, including non-embedding-based methods (e.g., pattern-based methods). We can use more than one entity expansion model to find related entities. For instance, general embedding models, such as those built on Wikipedia, might be a good choice in early iterations, whereas more domain-specific models trained on domain-specific corpora might be helpful in later iterations. LUWAK is flexible to change and use more than one Expansion API. This design encourages us to continuously refine the entity expansion module easily. ### Example: Housing Equipment Entity Population
We show an example of populating house equipment entities using LUWAK for improving a real estate search engine. The preliminary step is to prepare seed entities that belong to the real estate house equipment entity type (e.g., kitchen, bath). In this case, a user is supposed to provide several entities ( $\sim $ 10) as an initial set of the category. LUWAK first asks the user to upload an initial seed set. The user can add, rename, and delete entities on the Entity table as he or she wants. The user can also choose a set of entity expansion models at any time. Figure 2 shows the entity dashboard in this example. When the user submits the current entity set by clicking the Expand Seed Set button (Figure UID11 ), LUWAK sends a request to the external Expansion APIs that are selected to obtain expanded entities. The returned values will be stored in the Feedback table, as Figure UID12 shows. The Feedback table provides a function to capture user feedback intuitively. The user can click the + or - buttons to assign positive or negative labels to the entity candidates. The score column stores the similarity score, which is calculated by the Expansion API as reference information for users. The user can also see how these entities are generated by looking at the original entities in the original column. The original entity information can be used to detect semantic drift. For instance, if the user finds the original entity of some entity candidates has negative labels, the user might consider inactivating the entity to prevent semantic drift. In the next step, the user reflects the feedback by clicking the Submit Feedback button. Then, the user will see the entity dashboard with the newly added entities as shown in Figure UID13 . The user can inactivate the entity by clicking the inactivate button. The user can sort rows by column values to take a brief look at the current entity set. Also, the entity dashboard provides a search function to find an entity for action. The user can also check how entities appear in a test document. As shown in Figure UID14 , LUWAK highlights these entities in the current entity set. After the user is satisfied with the amount of the current entity set in the table, the Export button allows the user to download the entire table, including inactivated entities. ### Related Work and Discussion
Entity population is one of the important practical problems in NLP. Generated entity dictionaries can be used in various applications, including search engines, named entity extraction, and entity linking. Iterative seed expansion is known to be an efficient approach to construct user-specific entity dictionaries. Previous studies have aimed to construct a high-quality entity dictionary from a small number of seed entities BIBREF4 , BIBREF5 , BIBREF6 , BIBREF7 . As we stated in "Entity Candidate Generation" , LUWAK is flexible with the types of algorithms used for entity population. A user can select any combinations of different methods once the Expansion API of the methods are available. Stanford Pattern-based Information Extraction and Diagnostics (SPIED) BIBREF8 is a pattern-based entity population system. SPIED requires not only an initial seed set but also document collection because it uses the pattern-based approach. After a user inputs initial seed entities, SPIED generates regular expression patterns to find entity candidates from a given document collection. This approach incurs a huge computational cost for calculating the scores of every regular expression pattern and every entity candidate in each iteration. Furthermore, SPIED adopts a bootstrapping approach, which does not involve user feedback for each iteration. This approach can easily result in semantic drift. Interactive Knowledge Extraction BIBREF9 (IKE) is an interactive bootstrapping tool for collecting relation-extraction patterns. IKE also provides a search-based entity extraction function and an embedding-based entity expansion function for entity population. A user can interactively add entity candidates generated by an embedding-based algorithm to an entity dictionary. LUWAK is a more lightweight tool than IKE, which only focuses on the entity population task. LUWAK has numerous features, such as the multiple entity expansion model choices, that are not implemented in IKE. Moreover, LUWAK is a corpus-free tool that does not require a document collection for entity population. Thus, we differentiate LUWAK from IKE, considering it a more lightweight entity population tool. ### Summary
This paper has presented LUWAK, a lightweight front-end tool for interactive entity population. LUWAK provides a set of basic functions such as entity expansion and user feedback assignment. We have implemented LUWAK in pure JavaScript with LocalStorage to make it an installation-free tool. We believe that LUWAK plays an important role in delivering the values of existing entity expansion techniques to potential users including nontechnical people without supposing a large amount of human cost. Moreover, we believe that this design makes it easy to compare performances between interactive entity population pipelines and develop more sophisticated ones. Figure 1. Basic workflow of LUWAK. Figure 2. Screenshots of the LUWAK Dashboard: (a) Entity table shows initial seed entities. (b) Feedback table shows entity candidates generated by the initial seed entities. (c) Entity table has added entities after the submission of generated entity candidates. (d) Entity highlighting function visually emphasizes the words in the current entity set.
|
JavaScript
|
What is the size of the dataset?
|
### Introduction
Sexual violence, including harassment, is a pervasive, worldwide problem with a long history. This global problem has finally become a mainstream issue thanks to the efforts of survivors and advocates. Statistics show that girls and women are put at high risk of experiencing harassment. Women have about a 3 in 5 chance of experiencing sexual harassment, whereas men have slightly less than 1 in 5 chance BIBREF0, BIBREF1, BIBREF2. While women in developing countries are facing distinct challenges with sexual violence BIBREF3, however sexual violence is ubiquitous. In the United States, for example, there are on average >300,000 people who are sexually assaulted every year BIBREF4. Additionally, these numbers could be underestimated, due to reasons like guilt, blame, doubt and fear, which stopped many survivors from reporting BIBREF5. Social media can be a more open and accessible channel for those who have experienced harassment to be empowered to freely share their traumatic experiences and to raise awareness of the vast scale of sexual harassment, which then allows us to understand and actively address abusive behavior as part of larger efforts to prevent future sexual harassment. The deadly gang rape of a medical student on a Delhi bus in 2012 was a catalyst for protest and action, including the development of Safecity, which uses online and mobile technology to work towards ending sexual harassment and assault. More recently, the #MeToo and #TimesUp movements, further demonstrate how reporting personal stories on social media can raise awareness and empower women. Millions of people around the world have come forward and shared their stories. Instead of being bystanders, more and more people become up-standers, who take action to protest against sexual harassment online. The stories of people who experienced harassment can be studied to identify different patterns of sexual harassment, which can enable solutions to be developed to make streets safer and to keep women and girls more secure when navigating city spaces BIBREF6. In this paper, we demonstrated the application of natural language processing (NLP) technologies to uncover harassment patterns from social media data. We made three key contributions: 1. Safecity is the largest publicly-available online forum for reporting sexual harassment BIBREF6. We annotated about 10,000 personal stories from Safecity with the key elements, including information of harasser (i.e. the words describing the harasser), time, location and the trigger words (i.e. the phrases indicate the harassment that occurred). The key elements are important for studying the patterns of harassment and victimology BIBREF5, BIBREF7. Furthermore, we also associated each story with five labels that characterize the story in multiple dimensions (i.e. age of harasser, single/multiple harasser(s), type of harasser, type of location and time of day). The annotation data are available online. 2. We proposed joint learning NLP models that use convolutional neural network (CNN) BIBREF8 and bi-directional long short-term memory (BiLSTM) BIBREF9, BIBREF10 as basic units. Our models can automatically extract the key elements from the sexual harassment stories and at the same time categorize the stories in different dimensions. The proposed models outperformed the single task models, and achieved higher than previously reported accuracy in classifications of harassment forms BIBREF6. 3. We uncovered significant patterns from the categorized sexual harassment stories. ### Related Work
Conventional surveys and reports are often used to study sexual harassment, but harassment on these is usually under-reported BIBREF2, BIBREF5. The high volume of social media data available online can provide us a much larger collection of firsthand stories of sexual harassment. Social media data has already been used to analyze and predict distinct societal and health issues, in order to improve the understanding of wide-reaching societal concerns, including mental health, detecting domestic abuse, and cyberbullying BIBREF11, BIBREF12, BIBREF13, BIBREF14. There are a very limited number of studies on sexual harassment stories shared online. Karlekar and Bansal karlekar2018safecity were the first group to our knowledge that applied NLP to analyze large amount ( $\sim $10,000) of sexual harassment stories. Although their CNN-RNN classification models demonstrated high performance on classifying the forms of harassment, only the top 3 majority forms were studied. In order to study the details of the sexual harassment, the trigger words are crucial. Additionally, research indicated that both situational factors and person (or individual difference) factors contribute to sexual harassment BIBREF15. Therefore, the information about perpetrators needs to be extracted as well as the location and time of events. Karlekar and Bansal karlekar2018safecity applied several visualization techniques in order to capture such information, but it was not obtained explicitly. Our preliminary research demonstrated automatic extraction of key element and story classification in separate steps BIBREF16. In this paper, we proposed joint learning NLP models to directly extract the information of the harasser, time, location and trigger word as key elements and categorize the harassment stories in five dimensions as well. Our approach can provide an avenue to automatically uncover nuanced circumstances informing sexual harassment from online stories. ### Data Collection and Annotation
We obtained 9,892 stories of sexual harassment incidents that was reported on Safecity. Those stories include a text description, along with tags of the forms of harassment, e.g. commenting, ogling and groping. A dataset of these stories was published by Karlekar and Bansal karlekar2018safecity. In addition to the forms of harassment, we manually annotated each story with the key elements (i.e. “harasser", “time", “location", “trigger"), because they are essential to uncover the harassment patterns. An example is shown in Figure FIGREF3. Furthermore, we also assigned each story classification labels in five dimensions (Table TABREF4). The detailed definitions of classifications in all dimensions are explained below. Age of Harasser: Individual difference such as age can affect harassment behaviors. Therefore, we studied the harassers in two age groups, young and adult. Young people in this paper refer to people in the early 20s or younger. Single/Multiple Harasser(s): Harassers may behave differently in groups than they do alone. Type of Harasser: Person factors in harassment include the common relationships or titles of the harassers. Additionally, the reactions of people who experience harassment may vary with the harassers' relations to themselves BIBREF5. We defined 10 groups with respects to the harassers' relationships or titles. We put conductors and drivers in one group, as they both work on the public transportation. Police and guards are put in the same category, because they are employed to provide security. Manager, supervisors, and colleagues are in the work-related group. The others are described by their names. Type of Location: It will be helpful to reveal the places where harassment most frequently occurs BIBREF7, BIBREF6. We defined 14 types of locations. “Station/stop” refers to places where people wait for public transportation or buy tickets. Private places include survivors' or harassers' home, places of parties and etc. The others are described by their names. Time of Day: The time of an incident may be reported as “in evening” or at a specific time, e.g. “10 pm”. We considered that 5 am to 6 pm as day time, and the rest of the day as the night. Because many of the stories collected are short, many do not contain all of the key elements. For example, “A man came near to her tried to be physical with her .”. The time and location are unknown from the story. In addition, the harassers were strangers to those they harassed in many cases. For instance, “My friend was standing in the queue to pay bill and was ogled by a group of boys.”, we can only learn that there were multiple young harassers, but the type of harasser is unclear. The missing information is hence marked as “unspecified”. It is different from the label “other", which means the information is provided but the number of them is too small to be represented by a group, for example, a “trader”. All the data were labeled by two annotators with training. Inter-rater agreement was measured by Cohen's kappa coefficient, ranging from 0.71 to 0.91 for classifications in different dimensions and 0.75 for key element extraction (details can refer to Table 1 in supplementary file). The disagreements were reviewed by a third annotator and a final decision was made. ### Proposed Models
The key elements can be very informative when categorizing the incidents. For instance, in Figure 1, with identified key elements, one can easily categorize the incident in dimensions of “age of harasser” (adult), “single/multiple harasser(s)” (single), “type of harasser” (unspecified), “type of location” (park) , “time of day” (day time). Therefore, we proposed two joint learning schemes to extract the key elements and categorize the incidents together. In the models' names, “J”, “A”, “SA” stand for joint learning, attention, and supervised attention, respectively. ### Proposed Models ::: CNN Based Joint Learning Models
In Figure FIGREF6, the first proposed structure consists of two layers of CNN modules. J-CNN: To predict the type of key element, it is essential for the CNN model to capture the context information around each word. Therefore, the word along with its surrounding context of a fixed window size was converted into a context sequence. Assuming a window size of $2l + 1$ around the target word $w_0$, the context sequence is $[(w_{-l}, w_{-l+1},...w_0, ...w_{l-1},w_l)]$, where $w_i (i \in [-l,l])$ stands for the $ith$ word from $w_0$. Because the context of the two consecutive words in the original text are only off by one position, it will be difficult for the CNN model to detect the difference. Therefore, the position of each word in this context sequence is crucial information for the CNN model to make the correct predictions BIBREF17. That position was embedded as a $p$ dimensional vector, where $p$ is a hyperparameter. The position embeddings were learned at the training stage. Each word in the original text was then converted into a sequence of the concatenation of word and position embeddings. Such sequence was fed into the CNN modules in the first layer of the model, which output the high level word representation ($h_i, i\in [0,n-1]$, where n is the number of input words). The high level word representation was then passed into a fully connected layer, to predict the key element type for the word. The CNN modules in this layer share the same parameters. We input the sequence of high level word representations ($h_i$) from the first layer into another layer of multiple CNN modules to categorize the harassment incident in each dimension (Figure FIGREF6). Inside each CNN module, the sequence of word representations were first passed through a convolution layer to generate a sequence of new feature vectors ($C =[c_0,c_1,...c_q]$). This vector sequence ($C$) was then fed into a max pooling layer. This is followed by a fully connected layer. Modules in this layer do not share parameters across classification tasks. J-ACNN: We also experimented with attentive pooling, by replacing the max pooling layer. The attention layer aggregates the sequence of feature vectors ($C$) by measuring the contribution of each vector to form the high level representation of the harassment story. Specifically, That is, a fully connected layer with non-linear activation was applied to each vector $c_{i}$ to get its hidden representation $u_{i}$. The similarity of $u_{i}$ with a context vector $u_{w}$ was measured and get normalized through a softmax function, as the importance weight $\alpha _{i}$. The final representation of the incident story $v$ was an aggregation of all the feature vectors weighted by $\alpha _{i}$. $W_{\omega }$, $b_{\omega }$ and $u_{w}$ were learned during training. The final representation ($v$) was passed into one fully connected layer for each classification task. We also applied different attention layers for different classifications, because the classification modules categorize the incident in different dimensions, their focuses vary. For example, to classify “time of day”, one needs to focus on the time phrases, but pays more attention to harassers when classifying “age of harasser”. J-SACNN: To further exploit the information of the key elements, we applied supervision BIBREF18 to the attentive pooling layer, with the annotated key element types of the words as ground truth. For instance, in classification of “age of harasser”, the ground truth attention labels for words with key element types of “harasser” are 1 and others are 0. To conform to the CNN structure, we applied convolution to the sequence of ground truth attention labels, with the same window size ($w$) that was applied to the word sequence (Eq. DISPLAY_FORM11). where $\circ $ is element-wise multiplication, $e_t$ is the ground truth attention label, and the $W \in R^{w\times 1}$ is a constant matrix with all elements equal to 1. $\alpha ^{*}$ was normalized through a softmax function and used as ground truth weight values of the vector sequence ($C$) output from the convolution layer. The loss was calculated between learned attention $\alpha $ and $\alpha ^{*}$ (Eq. DISPLAY_FORM12), and added to the total loss. ### Proposed Models ::: BiLSTM Based Joint Learning Models
J-BiLSTM: The model input the sequence of word embeddings to the BiLSTM layer. To extract key elements, the hidden states from the forward and backward LSTM cells were concatenated and used as word representations to predict the key element types. To classify the harassment story in different dimensions, concatenation of the forward and backward final states of BiLSTM layer was used as document level representation of the story. J-ABiLSTM: We also experimented on BiLSTM model with the attention layer to aggregate the outputs from BiLSTM layer (Figure FIGREF7). The aggregation of the outputs was used as document level representation. J-SABiLSTM: Similarly, we experimented with the supervised attention. In all the models, softmax function was used to calculate the probabilities at the prediction step, and the cross entropy losses from extraction and classification tasks were added together. In case of supervised attention, the loss defined in Eq. DISPLAY_FORM12 was added to the total loss as well. We applied the stochastic gradient descent algorithm with mini-batches and the AdaDelta update Rule (rho=0.95 and epsilon=1e-6) BIBREF19, BIBREF20. The gradients were computed using back-propagation. During training, we also optimized the word and position embeddings. ### Experiments and Results ::: Experimental Settings
Data Splits: We used the same splits of train, develop, and test sets used by Karlekar and Bansal BIBREF6, with 7201, 990 and 1701 stories, respectively. In this study, we only considered single label classifications. Baseline Models: CNN and BiLSTM models that perform classification and extraction separately were used as baseline models. In classification, we also experimented with BiLSTM with the attention layer. To demonstrate that the improvement came from joint learning structure rather the two layer structure in J-CNN, we investigated the same model structure without training on key element extraction. We use J-CNN* to denote it. Preprocess: All the texts were converted to lowercase and preprocessed by removing non-alphanumeric characters, excluding “. ! ? ” . The word embeddings were pre-trained using fastText BIBREF21 with dimension equaling 100. Hyperparameters: For the CNN model, the filter size was chosen to be (1,2,3,4), with 50 filters per filter size. Batch size was set to 50 and the dropout rate was 0.5. The BiLSTM model comprises two layers of one directional LSTM. Every LSTM cell has 50 hidden units. The dropout rate was 0.25. Attention size was 50. ### Experiments and Results ::: Results and Discussions
We compared joint learning models with the single task models. Results are averages from five experiments. Although not much improvement was achieved in key element extraction (Figure TABREF16), classification performance improved significantly with joint learning schemes (Table TABREF17). Significance t-test results are shown in Table 2 in the supplementary file. BiLSTM Based Models: Joint learning BiLSTM with attention outperformed single task BiLSTM models. One reason is that it directed the attention of the model to the correct part of the text. For example, S1: “ foogreen!1.7003483371809125 foowhen foogreen!3.4324652515351772 fooi foogreen!10.76661329716444 foowas foogreen!20.388443022966385 fooreturning foogreen!9.704475291073322 foomy foogreen!6.052316632121801 foohome foogreen!2.477810252457857 fooafter foogreen!3.5612427163869143 foofinishing foogreen!4.7736018896102905 foomy foogreen!4.634172189980745 fooclass foogreen!0.6899426807649434 foo. foogreen!0.35572052001953125 fooi foogreen!0.3427551419008523 foowas foogreen!0.293194578262046 fooin foogreen!0.2028885210165754 fooqueue foogreen!0.10553237370913848 footo foogreen!0.19472737039905041 fooget foogreen!0.44946340494789183 fooon foogreen!0.5511227645911276 foothe foogreen!2.056689700111747 foomicro foogreen!2.597035141661763 foobus foogreen!2.5683704297989607 fooand foogreen!4.6382867731153965 foothere foogreen!9.827975183725357 foowas foogreen!21.346069872379303 fooa foogreen!22.295180708169937 foogirl foogreen!11.672522872686386 fooopposite foogreen!8.892465382814407 footo foogreen!18.20233091711998 foome foogreen!13.192926533520222 foojust foogreen!26.24184638261795 foothen foogreen!40.2555949985981 fooa foogreen!30.108729377388954 fooyoung foogreen!115.02625793218613 fooman foogreen!93.40204298496246 footried foogreen!58.68498980998993 footo foogreen!144.01434361934662 footouch foogreen!108.82275551557541 fooher foogreen!80.9452086687088 fooon foogreen!47.26015031337738 foothe foogreen!47.71501570940018 foobreast foogreen!19.392695277929306 foo.” S2: “ foogreen!0.2212507533840835 foowhen foogreen!0.26129744946956635 fooi foogreen!0.3014186804648489 foowas foogreen!0.314583390718326 fooreturning foogreen!0.23829322890378535 foomy foogreen!0.018542312318459153 foohome foogreen!0.06052045864635147 fooafter foogreen!0.3865368489641696 foofinishing foogreen!0.5127551266923547 foomy foogreen!0.569560332223773 fooclass foogreen!0.037081812479300424 foo. foogreen!0.061129467212595046 fooi foogreen!0.12043083552271128 foowas foogreen!0.2053432835964486 fooin foogreen!0.038308095099637285 fooqueue foogreen!0.05270353358355351 footo foogreen!0.07939991337480024 fooget foogreen!0.14962266141083091 fooon foogreen!0.11444976553320885 foothe foogreen!0.013002995729038958 foomicro foogreen!0.016201976904994808 foobus foogreen!0.14046543219592422 fooand foogreen!0.12413455988280475 foothere foogreen!0.18423641449771821 foowas foogreen!0.3394613158889115 fooa foogreen!1.0372470133006573 foogirl foogreen!0.20553644571918994 fooopposite foogreen!0.2821453963406384 footo foogreen!0.5574009846895933 foome foogreen!0.2709480468183756 foojust foogreen!0.2582515007816255 foothen foogreen!0.9223996312357485 fooa foogreen!788.9420390129089 fooyoung foogreen!199.1765946149826 fooman foogreen!0.39259070763364434 footried foogreen!0.27069455245509744 footo foogreen!0.5092779756523669 footouch foogreen!0.7033208385109901 fooher foogreen!0.6793316570110619 fooon foogreen!0.5892394692637026 foothe foogreen!0.4084075626451522 foobreast foogreen!0.14951340563129634 foo.” S3: “ foogreen!0.23944019631017 foowhen foogreen!0.16698541003279388 fooi foogreen!0.3381385176908225 foowas foogreen!0.21315943740773946 fooreturning foogreen!0.3222442464902997 foomy foogreen!0.8483575657010078 foohome foogreen!0.10339960863348097 fooafter foogreen!0.2440519310766831 foofinishing foogreen!0.39699181797914207 foomy foogreen!1.2218113988637924 fooclass foogreen!0.1232976937899366 foo. foogreen!0.10928708070423454 fooi foogreen!0.2562549489084631 foowas foogreen!0.8099888218566775 fooin foogreen!2.9650430660694838 fooqueue foogreen!0.507337914314121 footo foogreen!0.727736041881144 fooget foogreen!0.7367140497080982 fooon foogreen!0.711284636054188 foothe foogreen!194.2763775587082 foomicro foogreen!786.8869304656982 foobus foogreen!0.4422159108798951 fooand foogreen!0.43104542419314384 foothere foogreen!0.4694198723882437 foowas foogreen!0.5085613229312003 fooa foogreen!0.4430979897733778 foogirl foogreen!0.36199347232468426 fooopposite foogreen!0.31067250529304147 footo foogreen!0.2927705936599523 foome foogreen!0.24646619567647576 foojust foogreen!0.23911069729365408 foothen foogreen!0.11775700113503262 fooa foogreen!0.002219072712250636 fooyoung foogreen!0.0019248132048232947 fooman foogreen!0.32698659924790263 footried foogreen!0.3118939639534801 footo foogreen!0.5727249081246555 footouch foogreen!0.5670131067745388 fooher foogreen!0.7104063988663256 fooon foogreen!0.6698771030642092 foothe foogreen!0.4756081907544285 foobreast foogreen!0.26600153069011867 foo.” In S1, the regular BiLSTM with attention model for classification on “age of harasser” put some attention on phrases other than the harasser, and hence aggregated noise. This could explain why the regular BiLSTM model got lower performance than the CNN model. However, when training with key element extractions, it put almost all attention on the harasser “young man” (S2), which helped the model make correct prediction of “young harasser”. When predicting the “type of location” (S3), the joint learning model directed its attention to “micro bus”. CNN Based Models: Since CNN is efficient for capturing the most useful information BIBREF22, it is quite suitable for the classification tasks in this study. It achieved better performance than the BiLSTM model. The joint learning method boosted the performance even higher. This is because the classifications are related to the extracted key elements, and the word representation learned by the first layer of CNNs (Figure FIGREF6) is more informative than word embedding. By plotting of t-SNEs BIBREF23 of the two kinds of word vectors, we can see the word representations in the joint learning model made the words more separable (Figure 1 in supplementary file). In addition, no improvement was found with the J-CNN* model, which demonstrated the joint learning with extraction is essential for the improvement. With supervised attentive pooling, the model can get additional knowledge from key element labels. It helped the model in cases when certain location phrases were mentioned but the incidents did not happen at those locations. For instance, “I was followed on my way home .”, max pooling will very likely to predict it as “private places”. But, it is actually unknown. In other cases, with supervised attentive pooling, the model can distinguish “metro” and “metro station”, which are “transportation” and “stop/station” respectively. Therefore, the model further improved on classifications on “type of location” with supervised attention in terms of macro F1. For some tasks, like “time of day”, there are fewer cases with such disambiguation and hence max pooling worked well. Supervised attention improved macro F1 in location and harasser classifications, because it made more correct predictions in cases that mentioned location and harasser. But the majority did not mention them. Therefore, the accuracy of J-SACNN did not increase, compared with the other models. Classification on Harassment Forms: In Table TABREF18, we also compared the performance of binary classifications on harassment forms with the results reported by Karlekar and Bansal karlekar2018safecity. Joint learning models achieved higher accuracy. In some harassment stories, the whole text or a span of the text consists of trigger words of multiple forms, such as “stare, whistles, start to sing, commenting”. The supervised attention mechanism will force the model to look at all such words rather than just the one related to the harassment form for classification and hence it can introduce noise. This can explain why J-SACNN got lower accuracy in two of the harassment form classifications, compared to J-ACNN. In addition, J-CNN model did best in “ogling” classification. ### Patterns of Sexual Harassment
We plotted the distribution of harassment incidents in each categorization dimension (Figure FIGREF19). It displays statistics that provide important evidence as to the scale of harassment and that can serve as the basis for more effective interventions to be developed by authorities ranging from advocacy organizations to policy makers. It provides evidence to support some commonly assumed factors about harassment: First, we demonstrate that harassment occurred more frequently during the night time than the day time. Second, it shows that besides unspecified strangers (not shown in the figure), conductors and drivers are top the list of identified types of harassers, followed by friends and relatives. Furthermore, we uncovered that there exist strong correlations between the age of perpetrators and the location of harassment, between the single/multiple harasser(s) and location, and between age and single/multiple harasser(s) (Figure FIGREF20). The significance of the correlation is tested by chi-square independence with p value less than 0.05. Identifying these patterns will enable interventions to be differentiated for and targeted at specific populations. For instance, the young harassers often engage in harassment activities as groups. This points to the influence of peer pressure and masculine behavioral norms for men and boys on these activities. We also found that the majority of young perpetrators engaged in harassment behaviors on the streets. These findings suggest that interventions with young men and boys, who are readily influenced by peers, might be most effective when education is done peer-to-peer. It also points to the locations where such efforts could be made, including both in schools and on the streets. In contrast, we found that adult perpetrators of sexual harassment are more likely to act alone. Most of the adult harassers engaged in harassment on public transportation. These differences in adult harassment activities and locations, mean that interventions should be responsive to these factors. For example, increasing the security measures on transit at key times and locations. In addition, we also found that the correlations between the forms of harassment with the age, single/multiple harasser, type of harasser, and location (Figure FIGREF21). For example, young harassers are more likely to engage in behaviors of verbal harassment, rather than physical harassment as compared to adults. It was a single perpetrator that engaged in touching or groping more often, rather than groups of perpetrators. In contrast, commenting happened more frequently when harassers were in groups. Last but not least, public transportation is where people got indecently touched most frequently both by fellow passengers and by conductors and drivers. The nature and location of the harassment are particularly significant in developing strategies for those who are harassed or who witness the harassment to respond and manage the everyday threat of harassment. For example, some strategies will work best on public transport, a particular closed, shared space setting, while other strategies might be more effective on the open space of the street. These results can provide valuable information for all members of the public. Sharing stories of harassment has been found by researchers to shift people’s cognitive and emotional orientation towards their traumatic experiences BIBREF24. Greater awareness of patterns and scale of harassment experiences promises to ensure those who have been subjected to this violence that they are not alone, empowering others to report incidents, and ensuring them that efforts are being made to prevent others from experiencing the same harassment. These results also provide various authorities tools to identify potential harassment patterns and to make more effective interventions to prevent further harassment incidents. For instance, the authorities can increase targeted educational efforts at youth and adults, and be guided in utilizing limited resources the most effectively to offer more safety measures, including policing and community-based responses. For example, focusing efforts on highly populated public transportation during the nighttime, when harassment is found to be most likely to occur. ### Conclusions
We provided a large number of annotated personal stories of sexual harassment. Analyzing and identifying the social patterns of harassment behavior is essential to changing these patterns and social tolerance for them. We demonstrated the joint learning NLP models with strong performances to automatically extract key elements and categorize the stories. Potentiality, the approaches and models proposed in this study can be applied to sexual harassment stories from other sources, which can process and summarize the harassment stories and help those who have experienced harassment and authorities to work faster, such as by automatically filing reports BIBREF6. Furthermore, we discovered meaningful patterns in the situations where harassment commonly occurred. The volume of social media data is huge, and the more we can extract from these data, the more powerful we can be as part of the efforts to build a safer and more inclusive communities. Our work can increase the understanding of sexual harassment in society, ease the processing of such incidents by advocates and officials, and most importantly, raise awareness of this urgent problem. ### Acknowledgments
We thank the Safecity for granting the permission of using the data. Table 1: Definition of classes in different dimensions about sexual harassment. Figure 2: CNN based Joint learning Model. WL and WR are the left and right context around each word. Figure 3: BiLSM based Joint Learning Model. Here we use an input of five words as an example. Table 2: Key element extraction results. Table 3: Classification accuracy and macro F1 of the models. The best scores are in bold. Table 4: Harassment form classification accuracy of models. * Reported by Karlekar and Bansal (2018) Figure 4: Distributions of incidents. A) Distributions over age of harasser, B) over single/multiple harasser(s), C) over time of day, D) over type of harasser. E) over type of location. Figure 5: Distributions of incidents over two dimensions. A) Distributions of incidents A) with young/adult harassers at each location, B) with single/multiple harasser(s) at each location, C) across young/adult harassers and single/multiple harasser(s) Figure 6: Distributions of incidents with harassment forms and different dimensions. Distributions of harassment forms A) within each age group, B) within single/multiple harasser(s), C) over locations, D) within each harasser type.
|
9,892 stories of sexual harassment incidents
|
How much labeled data is available for these two languages?
|
### Introduction
Named Entity Recognition (NER) is a classification task that identifies words in a text that refer to entities (such as dates, person, organization and location names). It is a core task of natural language processing and a component for many downstream applications like search engines, knowledge graphs and personal assistants. For high-resource languages like English, this is a well-studied problem with complex state-of-the-art systems reaching close to or above 90% F1-score on the standard datasets CoNLL03 BIBREF0 and Ontonotes BIBREF1. In recent years, research has been extended to a larger pool of languages including those of developing countries BIBREF2, BIBREF3, BIBREF4, BIBREF5. Often, for these languages (like Hausa and Yorùbá studied here), there exists a large population with access to digital devices and internet (and therefore digital text), but natural language processing (NLP) tools do not support them. One key reason is the absence of labeled training data required to train these systems. While manually labeled, gold-standard data is often only available in small quantities, it tends to be much easier to obtain large amounts of unlabeled text. Distant and weak supervision methods can then be used to create labeled data in a (semi-) automatic way. Using context BIBREF6, BIBREF7, external knowledge and resources BIBREF8, BIBREF9, expert rules BIBREF10, BIBREF11 or self-training BIBREF12, BIBREF13, a corpus or dataset can be labeled quickly and cheaply. Additionally, a variety of noise-handling methods have been proposed to circumvent the negative effects that errors in this automatic annotation might have on the performance of a machine learning classifier. In this work, we study two methods of distant supervision for NER: Automatic annotation rules and matching of lists of entities from an external knowledge source. While distant supervision has been successfully used for high resource languages, it is not straight forward that these also work in low-resource settings where the amount of available external information might be much lower. The knowledge graph of Wikidata e.g. contains 4 million person names in English while only 32 thousand such names are available in Yorùbá, many of which are Western names. Orthogonally to distant supervision, the pre-training of word embeddings is a key component for training many neural NLP models. A vector representation for words is built in an unsupervised fashion, i.e. on unlabeled text. Standard embedding techniques include Word2Vec BIBREF14, GloVe BIBREF15 and FastText BIBREF16. In the last two years, contextualized word embeddings have been proposed BIBREF17, BIBREF18, BIBREF19. At the cost of having a much larger model size, these vector representations take the context of words into account and have been shown to outperform other embeddings in many tasks. In this study, we evaluate both types of representations. The key questions we are interested in this paper are: How do NER models perform for Hausa and Yorùbá, two languages from developing countries? Are distant-supervision techniques relying on external information also useful in low-resource settings? How do simple and contextual word embeddings trade-off in model size and performance? ### Background & Methods ::: Languages
Hausa language is the second most spoken indigenous language in Africa with over 40 million native speakers BIBREF20, and one of the three major languages in Nigeria, along with Igbo and Yorùbá. The language is native to the Northern part of Nigeria and the southern part of Niger, and it is widely spoken in West and Central Africa as a trade language in eight other countries: Benin, Ghana, Cameroon, Togo, Côte d'Ivoire, Chad, Burkina Faso, and Sudan. Hausa has several dialects but the one regarded as standard Hausa is the Kananci spoken in the ancient city of Kano in Nigeria. Kananci is the dialect popularly used in many local (e.g VON news) and international news media such as BBC, VOA, DW and Radio France Internationale. Hausa is a tone language but the tones are often ignored in writings, the language is written in a modified Latin alphabet. Despite the popularity of Hausa as an important regional language in Africa and it's popularity in news media, it has very little or no labelled data for common NLP tasks such as text classification, named entity recognition and question answering. Yorùbá language is the third most spoken indigenous language in Africa after Swahilli and Hausa with over 35 million native speakers BIBREF20. The language is native to the South-western part of Nigeria and the Southern part of Benin, and it is also spoken in other countries like Republic of Togo, Ghana, Côte d'Ivoire, Sierra Leone, Cuba and Brazil. Yorùbá has several dialects but the written language has been standardized by the 1974 Joint Consultative Committee on Education BIBREF21, it has 25 letters without the Latin characters (c, q, v, x and z) and with additional characters (ẹ, gb, ṣ , ọ). Yorùbá is a tone language and the tones are represented as diacritics in written text, there are three tones in Yorùbá namely low ( \), mid (“$-$”) and high ($/$). The mid tone is usually ignored in writings. Often time articles written online including news articles like BBC and VON ignore diacritics. Ignoring diacritics makes it difficult to identify or pronounce words except they are in a context. For example, owó (money), ọw (broom), òwò (business), w (honour), ọw (hand), and w (group) will be mapped to owo without diacritics. Similar to the Hausa language, there are few or no labelled datasets for NLP tasks. ### Background & Methods ::: Datasets & Embeddings
The Hausa data used in this paper is part of the LORELEI language pack. It consists of Broad Operational Language Translation (BOLT) data gathered from news sites, forums, weblogs, Wikipedia articles and twitter messages. We use a split of 10k training and 1k test instances. Due to the Hausa data not being publicly available at the time of writing, we could only perform a limited set of experiments on it. The Yorùbá NER data used in this work is the annotated corpus of Global Voices news articles recently released by BIBREF22. The dataset consists of 1,101 sentences (26k tokens) divided into 709 training sentences, 113 validation sentences and 279 test sentences based on 65%/10%/25% split ratio. The named entities in the dataset are personal names (PER), organization (ORG), location (LOC) and date & time (DATE). All other tokens are assigned a tag of "O". For the Yorùbá NER training, we make use of Yorùbá FastText embeddings BIBREF22 and multilingual-BERT that was trained on 104 languages including Yorùbá. Instead of the original FastText embeddings BIBREF16, we chose FastText embeddings trained on a multi-domain and high-quality dataset BIBREF22 because it gave better word similarity scores. ### Background & Methods ::: Distant and Weak Supervision
In this work, we rely on two sources of distant supervision chosen for its ease of application: Rules allow to apply the knowledge of domain experts without the manual effort of labeling each instance. They are especially suited for entities that follow specific patterns, like time phrases in text (see also BIBREF23). We use them for the DATE entity. In Yoruba, date expressions are written with the keywords of “ọj” (day), “oṣù” (month), and “ọdn” (year). Similarly, time expressions are written with keywords such as “wákàtí” (hour), “ìṣjú (minute) and “ìṣjú-aaya (seconds). Relative date and time expressions are also written with keywords “ḷodn” (in the year), “loṣù” (in the month), “lọs” (in the week), “lọj” (in the day). An example of a date expression is: “8th of December, 2018” in Yorùbá translates to “ọj 8 oṣù Ọp, ọdún 2018” Lists of Entities can be obtained from a variety of sources like gazetteers, dictionaries, phone books, census data and Wikipedia categories BIBREF24. In recent years, knowledge bases like Freebase and Wikidata have become another option to retrieve entity lists in a structured way. An entity list is created by extracting all names of that type from a knowledge source (e.g. all person names from Wikidata). If a word or token from the unlabeled text matches an entry in an entity list, it is assigned the corresponding label. Experts can add heuristics to this automatic labeling that improve the matching BIBREF25. These include e.g. normalizing the grammatical form of words or filtering common false positives. Another popular method for low-resource NER is the use of cross-lingual information BIBREF26. Alternatives to distant supervision are crowd-sourcing BIBREF27 and non-expert annotations BIBREF28. ### Background & Methods ::: Learning With Noisy Labels
The labels obtained through distant and weak supervision methods tend to contain a high amount of errors. In the Food101N dataset BIBREF29 around 20% of the automatically obtained labels are incorrect while for Clothing1M BIBREF30 the noise rate is more than 60%. Learning with this additional, noisily labeled data can result in lower classification performance compared to just training on a small set of clean labels (cf. e.g. BIBREF31). A variety of techniques have been proposed to handle label noise like modelling the underlying noise process BIBREF32 and filtering noisy instances BIBREF33, BIBREF34. BIBREF35 gives an in-depth introduction into this field and BIBREF36 survey more recent approaches, focusing on the vision domain. In this work, we experiment with three noise handling techniques. The approach by BIBREF37 estimates a noise channel using the EM algorithm. It treats all labels as possibly noisy and does not distinguish between a clean and a noisy part of the data. In contrast, the method by BIBREF38 leverages the existence of a small set of gold standard labels, something that - in our experience - is often available even in low resource settings. Having such a small set of clean labels is beneficial both for the main model itself as well as for the noise handling technique. Both approaches model the relationship between clean and noisy labels using a confusion matrix. This allows adapting the noisy to the clean label distribution during training. For a setting with 5 labels, it only requires $5^2=25$ additional parameters to estimate which could be beneficial when only few training data is available. The technique by BIBREF39 (adapted to NER by BIBREF38) learns a more complex neural network to clean the noisy labels before training with them. It also takes the features into account when cleaning the noise and it might, therefore, be able to model more complex noise processes. All three techniques can be easily added to the existing standard neural network architectures for NER. ### Models & Experimental Settings
Hausa Distant supervision on Hausa was performed using lists of person names extracted from Wikipedia data. Since we had limited access to the data, we tested a simplified binary NER-tagging setting (PERSON-tags only). As a base model, we used a Bi-LSTM model developed for Part-of-Speech tagging BIBREF40. For noise handling, we apply the Noise Channel model by BIBREF37. Yorùbá For Yorùbá, the entity lists were created by extracting person, location and organization entities from Wikidata in English and Yorùbá. Additionally, a list of person names in Nigeria was obtained from a Yorùbá Name website (8,365 names) and list of popular Hausa, Igbo, Fulani and Yorùbá people on Wikipedia (in total 9,241 names). As manual heuristic, a minimum name length of 2 was set for extraction of PER (except for Nigerian names), LOC and ORG. The Nigerian names were set to include names with a minimum length of 3. For the DATE label, a native Yorùbá speaker wrote some annotation rules using 11 “date keywords” (“ọj”, “ọs”, “os”, “ọdn”, “wákàtí” , “ḷodn”, “ḷodn-un”, “ọdn-un” “lọs” , “lọj”, “ aago”) following these two criteria: (1) A token is a date keyword or follows a date keyword in a sequence. (2) A token is a digit. For Yorùbá, we evaluate four settings with different amounts of clean data, namely 1k, 2k, 4k and the full dataset. As distantly supervised data with noisy labels, the full dataset is used. Additionally, 19,559 words from 18 articles of the Global News Corpus (different from the articles in the training corpus) were automatically annotated. The Bi-LSTM model consists of a Bi-LSTM layer followed by a linear layer to extract input features. The Bi-LSTM layer has a 300-dimensional hidden state for each direction. For the final classification, an additional linear layer is added to output predicted class distributions. For noise handling, we experiment with the Confusion Matrix model by BIBREF38 and the Cleaning model by BIBREF39. We repeat all the Bi-LSTM experiments 20 times and report the average F1-score (following the approach by BIBREF41) and the standard error. The BERT model is obtained by fine-tuning the pre-trained BERT embeddings on NER data with an additional untrained CRF classifier. We fine-tuned all the parameters of BERT including that of the CRF end-to-end. This has been shown to give better performance than using word features extracted from BERT to train a classifier BIBREF19. The evaluation result is obtained as an average of 5 runs, we report the F1-score and the standard error in the result section. ### Results
The results for Hausa are given in Table TABREF14. Training with a mix of 50% clean and 50% distantly-supervised data performs 15 F1-score points below using the whole 100% clean data which is to be expected due to the lower quality of the distantly-supervised labels. Using the Noise Channel closes half of this gap. Due to the limited availability of the dataset, we could unfortunately not investigate this further, but it shows already the benefits that are possible through noise-handling. An evaluation of the distant supervision for Yorùbá is given in Table TABREF14. The quality of the automatically annotated labels differs between the classes. Locations perform better than person and organization names, probably due to locations being less diverse and better covered in Wikidata. With simple date rules, we obtain already a 48% F1-score. This shows the importance of leveraging the knowledge of native speakers in automatic annotations. Overall a decent annotation can be obtained by the distant supervision and it even outperforms some of the actual machine learning models in the low-resource setting. Table TABREF14 compares using only Wikidata as data source versus adding additional, manually obtained lists of person names. While adding a list of Yorùbá names only improves recall slightly, the integration of Nigerian names helps to boost recall by 13 points. The experimental results for Yorùbá are given in Figure FIGREF11. The setting differs from the experiments with Hausa in that there is a small clean training set and additional, distantly-supervised data. For the Bi-LSTM model, adding distantly-supervised labels always helps. In the low-resource settings with 1k and 2k labeled data, it more than doubles the performance. Handling the noise in the distant supervision can result in slight improvements. The noise-cleaning approach struggles somewhat while the confusion matrix architecture does give better results in the majority of the scenarios. Training on 5k labeled data with distantly supervised data and noise handling, one can obtain a performance close to using the full 17k manually labeled token. The Bi-LSTM model has 1.50 million parameters (1.53 million for the cleaning model), while BERT has 110 million parameters. There is a clear trade-off between model size and performance. The BERT model is 70 times larger and obtains consistently better results due to its more complex, contextual embeddings pretrained on more data. Still, the F1-score also drops nearly half for the BERT model in the 1k setting compared to the full dataset. For 1k and 2k labeled data, the distant supervision helps to improve the model's performance. However, once the model trained only on clean data reaches a higher F1-score than the distant supervision technique, the model trained on clean and distantly-supervised data deteriorates. This suggests that the BERT model overfits too much on the noise in the distant supervision. ### Conclusion
In this study, we analysed distant supervision techniques and label-noise handling for NER in Hausa and Yorùbá, two languages from developing countries. We showed that they can be successfully leveraged in a realistic low-resource scenario to double a classifier's performance. If model size is not a constraint, the more complex BERT model clearly outperforms the smaller Bi-LSTM architecture. Nevertheless, there is still a large gap between the best performing model on Yorùbá with 66 F1-score and the state-of-the-art in English around 90. We see several interesting follow-ups to these evaluations. In the future, we want to evaluate if noise handling methods can also allow the more complex BERT model to benefit from distant supervision. Regarding the model complexity, it would be interesting to experiment with more compact models like DistilBERT BIBREF42 that reach a similar performance with a smaller model size for high-resource settings. In general, we want to investigate more in-depth the trade-offs between model complexity and trainability in low-resource scenarios. ### Acknowledgments
The experiments on Hausa were possible thanks to the collaboration with Florian Metze and CMU as part of the LORELEI project. Gefördert durch die Deutsche Forschungsgemeinschaft (DFG) – Projektnummer 232722074 – SFB 1102 / Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project-ID 232722074 – SFB 1102, the EU-funded Horizon 2020 projects ROXANNE under grant agreement No. 833635 and COMPRISE (http://www.compriseh2020.eu/) under grant agreement No. 3081705. Figure 1: F1-scores and standard error for Yorùbá.
|
10k training and 1k test, 1,101 sentences (26k tokens)
|
Regarding laboratory findings in Mrs. Sample, which parameter was elevated beyond the reference range in january 2017?
Choose the correct answer from the following options:
A. Sodium
B. Potassium
C. Alkaline Phosphatase
D. Creatinine
E. Total Bilirubin
|
### Patient Report 0
**Dear colleague, **
We wish to provide an update regarding Mrs. Anna Sample, born on
01.01.1970. She was admitted to our clinic from 01/01/2017 to
01/02/2017.
**Diagnosis:** Diffuse large B-cell lymphoma of germinal center type; ID
01/2017
- Ann-Arbor: Stage IV
- R-IPI: 2 (LDH, stage)
- CNS-IPI: 2
- Histology: Aggressive B-NHL (DLBCL, NOS); no evidence of t(14;18)
translocation. Ki-67 at 40%. Positive reaction to MUM1, numerous
CD68-positive macrophages. Negative reaction to ALK1 and TdT.
- cMRI: Chronic inflammatory lesions suggestive of Multiple Sclerosis (MS)
- CSF: no evidence of malignancy
- Bone marrow aspiration: no infiltration from the pre-existing
lymphoma.
**Current treatment: **
Initiated R-Pola-CHP regimen q21
- Polatuzumab vedotin: 1.8mg/kg on Day 1.
- Rituximab: 375mg/m² on Day 0.
- Cyclophosphamide: 750mg/m² on Day 1.
- Doxorubicin: 50mg/m on Day 1.
- Prednisone: 100mg orally from Day 1-5.
**Previous therapy and course**
From 12/01/2016: Discomfort in the dorsal calf and thoracic spine,
weakness in the arms with limited ability to lift and grasp, occasional
dizziness.
12/19/2016 cMRI: chronic inflammatory marks indicative of MS.
12/20/2016 MRI: thoracic/lumbar spinal cord: Indication of a
metastatic mass starting from the left pedicle T1 with a significant
extraosseous tumor element and full spinal narrowing at the level of
T10-L1 with pressure on the myelon and growth into the neuroforamen
T11/T12 on the right and T12/L1 left. More lesions suggestive for
metastasis are L2, L3, and L4, once more with extraosseous tumor element
and invasion of the left pedicle.
12/21/2016 Fixed dorsal support T8-9 to L3-4. Decompression via
laminectomy T10 and partial laminectomy lumbar vertebra 3.
12/24/2016 CT chest/abdomen/pelvis: Magnified left axillary lymph node.
In the ventral left upper lobe, indication of a round, loose, cloudy
deposit, i.e., of inflammatory origin, follow-up in 5-7 weeks.
Nodule-like deposit in the upper inner quadrant of the right breast,
senological examination suggested.
**Pathology**: Aggressive B-NHL (DLBCL, NOS); no evidence of t(14;18)
translocation. Ki-67 staining was at 40%. Positive reaction to MUM1.
Numerous CD68-positive macrophages. No reaction to ALK1 and TdT.
**Other diagnoses**
- Primary progressive type of multiple sclerosis (ID 03/02)
- Mood disorder.
- 2-vessel CHD
**Medical History**
Mrs. Sample was transferred inpatient from DC for the initiation of
chemotherapy (R-Pola-CHP) for her DLBCL. In the context of her
pre-existing ALS, she presented on 12/19/2016 with acute pains and
restricted mobility in her upper limbs. After her admission to HK
Flowermoon, an MRI was performed which revealed a thoracic neoplastic
growth especially at the level of T10-L1, but also affecting lumbar
vertebra 3, L4 and L6. Surgical intervention on 12/21/2016 at DC
resulted in symptom relief. Presently, her complaints are restricted to
post-operative spine discomfort, shoulder hypoesthesia, and intermittent
hand numbness. She reported a weight loss of 5 kg during her
hospitalization. She denied having respiratory symptoms, infections,
systemic symptoms, or gastrointestinal complaints. Mrs. Sample currently
has a urinary catheter in place.
**Physical examination on admission**
General: The patient has a satisfactory nutritional status, normal
weight, and is dependent on a walker. Her functional status is evaluated
as ECOG 2. Cardiovascular: Regular heart rhythm at a normal rate. Heart
sounds are clear with no detected murmurs. Respiratory: Normal alveolar
breath sounds. No wheezing, stridor, or other abnormal sounds.
Abdominal: The abdomen is soft, non-tender, and non-distended with
normal bowel sounds in all quadrants. There is no palpable enlargement
of the liver or spleen, and the kidneys are not palpable.
Musculoskeletal: Tenderness noted in the cervical and thoracic spine
area, but no other remarkable findings. This is consistent with her
post-operative status. Lymphatic: No enlargement detected in the
temporal, occipital, cubital, or popliteal lymph nodes. Oral: The oral
mucosa is moist and well-perfused. The oropharynx is unremarkable, and
the tongue appears normal. Peripheral Vascular: Pulses in the hands are
strong and regular. No edema observed. Neurological: Cranial nerves are
intact. There is numbness in both hands and mild hypoesthesia in the
shoulders. Motor strength is 3/5 in the right arm, attributed to her
known ALS diagnosis. No other motor or sensory deficits noted.
Occasional bladder incontinence and intermittent gastrointestinal
disturbances are reported.
**Medications on admission**
Acetylsalicylic acid (Aspirin®) 100 mg: Take 1 tablet in the morning.
Atorvastatin (Lipitor®) 40 mg: Take 1 tablet in the evening. Fingolimod
(Gilenya®) 0.5 mg: Take 1 capsule in the evening. Sertraline (Zoloft®)
50 mg: Take 2 tablets in the morning. Hydromorphone (Dilaudid® or
Exalgo® for extended-release) 2 mg: Take 1 capsule in the morning and 1
in the evening. Lorazepam (Ativan®) 1 mg: 1 tablet as needed.
**Radiology/Nuclear Medicine**
**MR Head 3D unenhanced + contrast from 12/19/2016 10:30 AM**
**Technique:** Sequences obtained include 3D FLAIR, 3D DIR, 3D T2, SWI,
DTI/DWI, plain MPRAGE, and post-contrast MPRAGE. All images are of good
quality. Imaging area: Brain.
There are 20 FLAIR hyperintense lesions in the brain parenchyma,
specifically located periventricularly and in the cortical/juxtacortical
regions (right and left frontal, left temporal, and right and left
insular). No contrast-enhancing lesions are identified. There are also
subcortical/nonspecific lesions present, with some lesions appearing
confluent. The spinal cord is visualized up to the C4 level. No spinal
lesions are noted.
[Incidental findings:]{.underline}
- Brain volume assessment: no indication of reduced brain volume.
- CSF space: age-appropriate usual wide, moderate, and symmetric CSF
spacing with no signs of CSF flow abnormalities.
- Cortical-Subcortical Differentiation: Clear cortical-subcortical
distinction.
- RML-characteristic alterations: none detected.
- Eye socket: appears normal.
- Nasal cavities: Symmetric mucosal thickening with a focus on the
right ethmoidal sinus.
- Pituitary and peri-auricular region: no abnormalities.
- Subcutaneous lesion measuring 14.4 x 21.3 mm, right parietal likely
representing an inflamed cyst or abscess, differential includes soft
tissue growth.
[Evaluation]{.underline}
Dissemination: MRI standards for spatial distribution are satisfied. MRI
criteria for temporal distribution are unfulfilled. Comprehensive
neurological review: The findings are consistent with a chronic
inflammatory CNS disease in the sense of Multiple Sclerosis.
**MR Spine plain + post-contrast from 12/20/2016 10:00 AM**
**Technique:** GE 3T MRI Scanner
MRI was conducted under anesthesia due to claustrophobia.
**Sequences**: Holospinal T2 Dixon sagittal, T1 pre-contrast, T1 fs
post-contrast. The spine is visualized from the craniocervical junction
to S2.
**Thoracic spine: **
On T2-STIR and T2, there is a hyperintense signal of vertebral bodies T5
and T6 with inconsistent delineation of the vertebral endplates,
indicative of age-related changes. There is a reduction in the height of
the disc spaces T4/5 and T5/6 with subligamentous disc protrusion
leading to a spongy appearance of the spinal cord at this location.
Myelon atrophy is noted at T5/6, along with a T2 bright lesion
suggestive of MS at the level of T3 and also T4/5. Spine: A large
intraspinal mass extends from T10-L1, causing an anterior spongy
appearance of the spinal cord and resulting in complete spinal canal
stenosis at this level. On fat-only imaging, there is almost total
replacement of the marrow space of vertebral body T11 with external
tumor extension and infiltration into the lateral structures (more on
the left than the right) and neural foramen T11 on both sides. There is
mild disc herniation at T8/9 with slight sponginess of the spinal cord.
MS-characteristic spinal cord lesions are noted at segments T5 and T8/9.
**Lumbar spine: **
T2-DIXON shows bright signal intensity of the anterior part of lumbar
vertebra 1, a patchy appearance of lumbar vertebra 2, and lumbar
vertebra 4. Almost the entire marrow space is replaced in the fat-only
imaging. There is an external tumor mass posterior to lumbar vertebra 4
without significant spinal canal stenosis, which involves the left
lateral structure and a pronounced appearance of the cauda equina at
lumbar vertebra 1. A call to communicate the results was made at 11:15
a.m. to the on-duty orthopedic surgeon and to colleagues in neurology.
Evaluation Evidence of a metastatic lesion originating from the left
pedicle of T10 with a significant extramedullary tumor mass and full
spinal canal narrowing at the level of T10 with compression of the
spinal cord and extension into the neural foramen T11-T12 on the right,
and T12-L1 on the left. Additional sites suggestive of metastasis
include L2, L3, and L4, again with extramedullary tumor components and
invasion of the left lateral structure. Contrast enhancement of the
distal cord is noted. There are MS-characteristic spinal cord lesions at
the levels of T3, T4-5, T5, and T8-9. The conus medullaris is not
visualized due to spinal cord displacement.
**CT Thoracic Spine from 01/03/2017**
[Clinical Findings]{.underline}
Lateral and medial alignment is stable. No sign of vertebral column
damage. Multiple segment degenerative alterations in the spine. No
indications of mineralization in the recognized space at the level of
T10/L2. Invasion of T10 and L4 with composite osteolytic-osteoblastic
defects starting from the left pedicle into the vertebral column. More
cortical inconsistency with enhanced sclerosis at the endplate of L2.
Review with prior MRI indicative of a different composite defect. Defect
pit at the endplate of lumbar vertebra 2.
Minor pericardial effusion with nearby superior ventilation. Intubation
tube placed. Mild cardiomediastinum. Splenomegaly. Standard display of
the tissue organs of the mid-abdomen, as naturally observed. Normal
spleen. Thin adrenals. Tightly raised kidney bowls and leading ureter
from both aspects, e.g., upon entry during the exhalation period after
gadolinium inclusion in the earlier MRI. No bowel obstruction.
Intestinal stasis. No sign of abnormally magnified lymphatic vessels.
Remaining pin holes in the femoral head on both sides.
[Evaluation]{.underline}
Composite osteolytic-osteoblastic defects starting from the left pedicle
in T10 and lumbar vertebra 4, and at the endplate of lumbar vertebra 2.
**CT Thoracic Spine from 01/04/2017 **
Intraoperative CT imaging for enhanced guidance.
Two intraoperative CT scans were undertaken in total.
On the concluding CT scan, recently implanted non-radiopaque pedicle
screws T8-9 to L2-L3 at tumor band T10. Regular screw placement. No
evident sign of material breakage.
Apart from this, no notable alteration in findings from CT of
01/03/2017.
Evaluation Intraoperative CT imaging for better guidance. Recently
inserted pedicle screws T8/T9 and L2/L3 in tumor indication T10,
ultimate standard screw positioning done transpedicular.
**CT Chest/Abdomen/Pelvis + Contrast from 01/09/2017**
Results: After uneventful intravenous administration of Omnipaque 320, a
multi-slice helical CT of the chest, abdomen, and pelvis was performed
during the venous phase of contrast enhancement. Additional oral
contrast was given using Gastrografin (diluted 1:35). Thin slice
reconstructions were obtained, along with secondary coronal and sagittal
reconstructions.
[Thorax]{.underline}:
Uniform presentation of the apical thoracic sections when included. No
evidence of subclavian lymphadenopathy. Uniform visualization of the pectoral
tissues. No evidence of mediastinal lymphadenopathy. The anterior segment
of the left upper lobe (series 205, image 88 of 389) shows a subpleural
ground-glass opaque solid consolidation. There is an enlarged lymph node
in the left hilar region measuring approximately 1.2 cm laterally. Otherwise,
there are no signs of suspicious intrapulmonary markings, no new inflammatory
infiltrates, no pneumothorax, no pericardial effusion. In the upper inner
quadrant of the right breast there is an oval mass, DD cystadenoma,
DD glandular cluster (measuring 1.2 cm).
[Abdomen/pelvis: ]{.underline}
Dominant display of the gastrocolic junction; absence of oral contrast
in this zone prevents more detailed analysis. Uniformly displayed
hepatic tissue with no signs of focal, density-varied lesions. Portal
and liver veins are well filled. Liver with minor auxiliary liver.
Adrenal nodes thin on both sides. Natural kidneys on both sides. Urinary
sac with placed transurethral tube and intravesical gas pockets.
Gallbladder typical. Paravertebral and within vertebral and in the
domain of the superior hepatic artery multiple pronounced lymph nodes,
these up to a maximum of 8 mm. Typical representation of the organs in
the pelvic region.
[Skeleton: ]{.underline}
Condition post dorsal reinforcement (T8-T9-L2-L3). After surgery,
epidermal air pockets and bloated tissue inflammation in the access path
zone. Signs of a resin in a pre-spinal vessel anterior to T8 and T9.
Known mixed osteoblastic/osteolytic bony metastasis of lumbar vertebra 4
and the cap plate of lumbar vertebra 2. State post-cutting of the
pedicle of T10. L5 also with slightly multiple solidified core
osteolytic defects.
[Evaluation:]{.underline}
- No sign of primary malignancy in the previously mentioned mixed
osteoblastic/osteolytic lesions in the vertebra (to be deemed
suspicious in coordination with the MR review of 12/20/2016).
- A magnified lymph node exists in the left hilar territory. In the
anterior left upper lobe, evidence of a solid cloudy consolidation,
i.e., of inflammatory origin, revisitation in 5-7 weeks recommended.
- Rounded consolidation in the upper inner quadrant of the right
breast, further breast examination advised.
**Functional Diagnostics**
Extended Respiratory Function (Diffusion) from 01/15/2017
[Evaluation]{.underline}
Patient cooperation: satisfactory. No detectable obstructive ventilation
issue. No pulmonary over-expansion after RV/TLC. No identified
restrictive ventilation impairment. Standard O2 diffusion ability. No
evidence of low oxygen levels, no blockage.
[Consultations / Therapy Reports]{.underline}
Psychological Support Consultation from 01/22/2017
[Current Situation/History:]{.underline}
The patient initially discussed \"night episodes\" in the calves, which
over time manifested during the day and were coupled with discomfort in
the cervical region. Prior, she had visited the Riverside Medical Center
multiple times before an MRI was executed. A \"mass in the neck\" was
identified. Since she suffers from fear of heights and fear of crowds,
the MRI could only be done under mild sedation. The phobias emerged
abruptly in 2011 with no apparent cause, leading to multiple hospital
visits. She is now in outpatient care. Additionally, she battles with
MS, with the most recent flare-up in 2012. She declined a procedure,
which was set for the MRI is day, because \"two sedatives in one day
felt excessive.\" She anticipates avoiding a repeated procedure.
Currently, however, she still experiences spasms in her right hand and a
numbing sensation in her fingers. She still encounters discomfort (NRS
5/10). She was previously informed that relief might be gradual, but she
is \"historically been restless\". Therefore, \"resting and inactivity\"
negatively impact her spirits and rest.
[Medical background:]{.underline}
Several in-patient and day clinic admissions since 2011.
Now, from 2015, continuous outpatient psychological counseling (CBT),
somatic therapy, particular sessions for driving anxieties. Also
undergoing outpatient psychiatric care (fluoxetine 90mg).
[Psychopathological Observations:]{.underline}
Patient appears well-groomed, responsive and clear-minded, talkative and
forthright. Aware of location, date, and identity. Adequate focus,
recall, and concentration. Mental organization is orderly. No evidence
of delusional beliefs or identity disturbances. No compulsions, mentions
fear of expansive spaces and fear of water. Emotional responsiveness
intact, heightened psychomotor activity. Mood swings between despondent
and irritable, lowered motivation. Diminished appetite, issues with
sleep initiation and maintenance. Firm and believable denial of
immediate suicidal thoughts, patient appears cooperative. No current
signs of self-harm or threat to others.
[Handling the Condition, Strengths:]{.underline}
Currently, her coping strategy seems to be proactive with some restless
elements. Ms. S. says she remains \"optimistic\" and is well-backed by
her communal links. Notably, she shares a close bond with her
80-year-old aunt. Her other social bonds primarily arise from her
association with a hockey enthusiasts club. Hockey has been a crucial
support for her from a young age.
[Evaluation Diagnoses:]{.underline}
Adjustment disorder: anxiety and depressive reaction mixed
Agoraphobia
Acrophobia
[Interventions, approaches:]{.underline}
An evaluative and assistive discussion was conducted. The patient has a
dependable therapeutic community for post-hospitalization. Additionally,
she was provided the contact of the psychological support outpatient
center. She mentioned finding the therapeutic conversation comforting,
prompting an arranged check-in the subsequent week. We also suggest
guidance in self-initiated physical activities to aid her recovery and
temper restlessness.
**NC: Consultation of 01/15/2017**
[Examination findings:]{.underline}
Patient alert, fully oriented. Articulate and spontaneous speech.
Cranial nerve evaluation normal. No evident sensorimotor abnormalities.
BDK with voiding challenges. Sphincter response diminished, but fecal
control maintained. KPPS at 85%. Wound site clean and non-irritated,
except for the lower central portion.
[Procedure:]{.underline}
Neurosurgical intervention not required; no reassessment of the lower
wound needed. Advise return if neurological symptoms intensify.
The patient, diagnosed with relapsing-remitting multiple sclerosis that
initially manifested aggressively, has been relapse-free on fingolimod
since 2009 and was generally well, barring a slight imbalance when
walking due to minor weakness in her left leg. She later experienced
numbness and weakness in her legs, reaching up to the hip, persisting
for several days and then faced challenges with urination and bowel
movements approximately 7 weeks prior. During a home examination, a
lesion was identified in the T10 which was surgically addressed by our
in-house neurosurgery team. Histology identified it as a DLBCL, leading
to a chemotherapy plan inclusive of Rituximab. Post-surgery, her
symptoms have subsided somewhat, but the patient still has BDK and
relies on a wheelchair.
On clinical neurological assessment, a mild paraparesis was noted in her
left leg, accompanied by heightened reflex response and sporadic left
foot spasms, which were intense but temporary.
To conclude, the new neurological manifestations are not a recurrence of
the formerly stable multiple sclerosis. As Rituximab is also an
effective third-phase drug for MS treatment and is essential,
discontinuing fingolimod (second phase) was discussed with the patient.
After a span of approximately 4-5 months following the last Rituximab
treatment, a radiological (cMRI) and clinical review is suggested. Based
on results, either resuming fingolimod or, if no adverse effects
present, potentially continuing Rituximab treatment is recommended (for
this, reach our neuroimmunology outpatient department). The primary
neurologist was unavailable for comments.
**Boards**
Oncology tumor board as of 01/22/2017
6 cycles of R-Pola-CHP
[Pathology]{.underline}
Pathology. Findings from 01/05/2017
[Clinical information/question:]{.underline}
Tumor cuff T10. dignity? Entity? Macroscopy:
1st lamina T10: fixed. some assembled 0.7 x 0.5 x 0.2 cm calcareous
tissue fragments. Complete embedding. Decalcify in EDTA. 2nd ligament:
Fix. some assembled 0.9 x 0.7 x 0.2 cm, coarse, partly also calcareous
tissue fragments. Complete embedding. Decalcify overnight in EDTA.
3\. epidural tumor: Numerous beige-colored tissue fragments, 3.8 x 2.8 x
0.6 cm. Embedding of exemplary sections after lamellation.
[Processing]{.underline}: 1 block, HE. Microscopy:
1\. and 2. (lamina T10 and ligament) are still being decalcified.
3rd epidural tumor: Paravertebral soft tissue with infiltrates of a
partly lymphoid, partly blastic neoplasia. The tumor cells are diffuse,
sometimes nodular in the tissue and have hyperchromatic nuclei with
coarse-grained chromatin and a narrow cytoplasmic border. There are also
blastic cells with enlarged nuclei, vesicular chromatin, and sometimes
prominent nucleoli. The stroma is loose and vacuolated. Clearly
pronounced crush artifacts.
Preliminary report of critical findings:
3\. epidural tumor: paravertebral soft tissue with infiltrates of
lymphoid and blastic cells compatible with hematologic neoplasia.
Additional immunohistochemical staining is being performed to further
characterize the tumor. In addition, material 1 (lamina T10) and
material 2 (ligament) are still undergoing decalcification. A follow-up
report will be provided.
Processing: 2 blocks, decalcification, HE, Giemsa, IHC: CD20, PAX5,
Bcl2, Bcl6, CD5, CD3, CD23, CD21, Kappa, Lambda, CD10, c-Myc, CyclinD1,
CD30, MIB1, EBV/EBER.
Molecular pathology: testing for B-cell clonal expansion and IgH/Bcl2
translocation.
[Microscopy]{.underline}:
1\. Ligament: Scarred connective tissue and fragmented bone tissue
without evidence of the tumor described in the preliminary findings
under 3.
2\. Lamina T10: Bone tissue without evidence of the tumor described in
the preliminary findings under 3.
3\. Epidural tumor: Immunohistochemically, blastic tumor cells show a
positive reaction after incubation with antibodies against CD20, PAX5
and BCL2. Partially positive reaction against Bcl-6 (\<20%). Some
isolated blastic cells staining positive for CD30. Lymphoid cells
positive for CD3 and CD5. Some residual germinal centers with positive
reaction to CD23 and CD21. Predominantly weak positive reaction of
blasts and lymphoid cells to CD10. Some solitary cells with positive
reaction to kappa, rather unspecific, flat reaction to lambda. No
overexpression of c-Myc or cyclinD1. No
No reaction to EBV/EBER. The Ki-67 proliferation index is 40%, related
to blastic tumor cells \> 90%.
Significantly limited evaluability of immunohistochemical staining due
to severe squeezing artifacts of the material.
[Molecular pathology report:]{.underline}
Examination for clonal B-cell expansion and t(14;18) translocation
Methodology:
DNA was isolated from the sent tissues and used in duplicate in specific
PCRs (B-cell clonality analysis with Biomed-2 primer sets: IGHG1 gene:
scaffold 2 and 3). The size distribution of the PCR products was further
analyzed by fragment analysis.
To detect a BCL2/IgH fusion corresponding to a t(14;18) translocation,
DNA was inserted into a specific PCR (according to Stetler-Stevenson et
al. Blood. 1998;72:1822-25).
Results:
Amplification of isolated DNA: good. B cell clonality analyses
IGHG1 fragment 2: polyclonal signal pattern.
IGHG1 frame 3: reproducible clonal signal at approximately 115/116 bp.
t(14;18) translocation: negative.
[Molecular pathology report:]{.underline}
Molecular pathologic evidence of clonal B-cell expansion. No evidence of
t(14;18) translocation in test material with normal control reactions.
Preliminary critical findings report:
1\. Lamina TH 10: tumor-free bone tissue.
2\. Ligament: Tumor-free, scarred connective tissue and fragmented bone
tissue.
3\. Epidural tumor: aggressive B non-Hodgkin\'s lymphoma.
Findings (continued)
Additional findings from 01/06/2017
Immunohistochemical processing: MUM1, ALK1, CD68, TdT. Microscopy:
3\. Immunohistochemically, blastic tumor cells are positive for MUM1.
Numerous CD68-positive macrophages. No reaction to ALK1 and TdT.
Critical findings report:
1\. Lamina T10: Tumor-free bone tissue. 2: Tumor free, scarred connective
tissue and fragmented bone tissue.
3\. epidural tumor: aggressive B-non-Hodgkin lymphoma, morphologically
and immunohistochemically most compatible with diffuse large B-cell
lymphoma (DLBCL, NOS) of germinal center type according to Hans
classifier (GCB).
**Path. Findings from 01/05/2017**
Clinical Findings
Clinical data:
Initial diagnosis of DLBCL with spinal involvement.
Puncture Site(s): 1
Collection date: 01/04/2017
Arrival at cytology lab: 01/04/2017, 8 PM. Material:
1 Liquid Material: 2 mL colorless, clear Processing:
MGG staining Microscopic:
ZTA:
Liquid precipitate Erythrocytes
(+) Lymphocytes (+) Granulocytes
Eosinophils Histiocytes Siderophages
\+ Monocytes
Others: Isolated evidence of fewer monocytes. No evidence of atypical
cells. Critical report of findings:
CSF sediment without evidence of inflammation or malignancy. Diagnostic
Grading:
Negative
Therapy and course
Mrs. S was admitted from the neurosurgical department for chemotherapy
(R-POLA-CHP) of suspected DLBCL with spinal/vertebral manifestations.
After exclusion of clinical and laboratory contraindications,
antineoplastic therapy was started on 01/08/2017. This was well
tolerated under the usual supportive measures. There were no acute
complications.
During her hospitalization, Ms. S reported numbness in both vascular
hemispheres. A neurosurgical and neurological presentation was made
without acute need for action. In consultation with the neurology
department, the existing therapy with fingolimod should be discontinued
due to the concomitant use of rituximab and the associated risk of PML.
If necessary, re-exposure to fingolimod may be considered after
completion of oncologic therapy.
On 01/07/2017, a port placement was performed by our vascular surgery
department without complications.
On 01/19/2017, a single administration of Pegfilgrastim 6 mg s.c. was
performed. With a latency of 10 days, G-CSF should not be repeated in
the meantime.
We are able to transfer Mrs. S to the Mountain Hospital Center
(Neurological Initial Therapy & Recovery) on 02/01/2017. We thank you
for accommodating the patient and are available for any additional
inquiries.
**Medications at Discharge**
**Aspirin (Aspirin®)** - 100mg, 1 tablet in the morning
**Atorvastatin - 40mg -** 1 tablet at bedtime
**Sertraline - 50mg** - 2 tablets in the morning
**Lorazepam (Tavor®)** - 1mg, as needed
**Fingolimod** - 0.5mg, 1 capsule at bedtime, Note: Take a break as
directed
**Hydromorphone hydrochloride** - 2mg (extended-release), 2 capsules in
the morning and 2 capsules at bedtime
**Melatonin -** 2mg (sustained-release), 1 tablet at bedtime
**Baclofen (Lioresal®) -** 10mg, 1 tablet three times a day
**Pregabalin -** 75mg, 1 capsule in the morning and 1 capsule at bedtime
**MoviCOL® (Macrogol, Sodium chloride, Potassium chloride) -** 1 packet
three times a day, mixed with water for oral intake
**Pantoprazole -** 40mg, 1 tablet in the morning
**Colecalciferol (Vitamin D3) -** 20000 I.U., 1 capsule on Monday and
Thursday
**Co-trimoxazole -** 960mg, 1 tablet on Monday, Wednesday, and Friday
**Valaciclovir -** 500mg, 1 tablet in the morning and 1 tablet at
bedtime
**Prednisolone -** 50mg, 2 tablets in the morning, Continue through
02/19/2017
**Enoxaparin sodium (Clexane®) -** 40mg (4000 I.U.), 1 injection at
bedtime, Note: Continue in case of immobility
**Dimenhydrinate (Vomex A®)** - 150mg (sustained-release), as needed for
nausea, up to 2 capsules daily.
**Procedure**
**Oncology board decision: 6 cycles of R-Pola-CHP.**
- Fingolimod pause, re-evaluation in 4-5 months.
- Continuation of therapy near residence in the clinic as of
02/28/2017
- Bi-Weekly laboratory tests (electrolytes, blood count, kidney and
liver function tests)
- In case of fever \>38.3 °C please report immediately to our
emergency room
- Immediate gynecological examination for nodular mass in the left
breast
Dates:
- From 03/01/2017 third cycle of R-Pola-CHP in the clinic. The patient
will be informed of the date by telephone.
If symptoms persist or exacerbate, we advocate for an urgent revisit.
Outside standard working hours, emergencies can also be addressed at the
emergency hub.
During discharge management, the patient was extensively educated and
assisted, and equipped with required appliances, medication scripts, and
absence from work notices.
All observations were thoroughly deliberated upon. Multiple alternate
therapy notions were considered before making a treatment proposition.
The opportunity for a second viewpoint and recommendation to our
facility was also emphasized.
**Lab values at discharge: **
**Metabolic Panel**
**Parameter** **Results** **Reference Range**
---------------------------------- ------------- ---------------------
Sodium 136 mEq/L 135 - 145 mEq/L
Potassium 3.9 mEq/L 3.5 - 5.0 mEq/L
Creatinine 1.2 mg/dL 0.7 - 1.3 mg/dL
BUN (Blood Urea Nitrogen) 19 mg/dL 7 - 18 mg/dL
Alkaline Phosphatase 138 U/L 40 - 129 U/L
Total Bilirubin 0.3 mg/dL \< 1.2 mg/dL
GGT (Gamma-Glutamyl Transferase) 82 U/L \< 66 U/L
ALT (Alanine Aminotransferase) 42 U/L 10 - 50 U/L
AST (Aspartate Aminotransferase) 34 U/L 10 - 50 U/L
LDH (Lactate Dehydrogenase) 366 U/L \< 244 U/L
Uric Acid 4.1 mg/dL 3 - 7 mg/dL
Calcium 9.0 mg/dL 8.8 - 10.6 mg/dL
**Kidney Function**
**Parameter** **Results** **Reference Range**
------------------------------- ------------- ---------------------
GFR (MDRD) \>60 mL/min \> 60 mL/min
GFR (CKD-EPI with Creatinine) 64 mL/min \> 90 mL/min
**Inflammatory Markers**
**Parameter** **Results** **Reference Range**
-------------------------- ------------- ---------------------
CRP (C-Reactive Protein) 2.5 mg/dL \< 0.5 mg/dL
**Coagulation Panel**
**Parameter** **Results** **Reference Range**
--------------- ------------- ---------------------
PT Percentage 103% 70 - 120%
INR 1.0 N/A
aPTT 25 sec 26 - 37 sec
**Complete Blood Count**
**Parameter** **Results** **Reference Range**
--------------- ---------------- ---------------------
WBC 12.71 x10\^9/L 4.0 - 9.0 x10\^9/L
RBC 2.9 x10\^12/L 4.5 - 6.0 x10\^12/L
Hemoglobin 8.1 g/dL 14 - 18 g/dL
Hematocrit 24.7% 40 - 48%
MCH 28 pg 27 - 32 pg
MCV 86 fL 82 - 92 fL
MCHC 32.8 g/dL 32 - 36 g/dL
Platelets 257 x10\^9/L 150 - 450 x10\^9/L
**Differential**
**Parameter** **Results** **Reference Range**
--------------- ------------- ---------------------
Neutrophils 77% 40 - 70%
Lymphocytes 4% 25 - 40%
Monocytes 18% 4 - 10%
Eosinophils 0% 2 - 4%
Basophils 0% 0 - 1%
### Patient Report 1
**Dear colleague, **
I am writing to provide a follow-up report on our mutual patient, Mrs.
Anna Sample, born on January 1st, 1970, post her recent visit to our
clinic on October 9th, 2017.
Upon assessment, Mrs. Sample reported experiencing a moderate
improvement in symptoms since the initiation of the R-Pola-CHP regimen.
The discomfort in her dorsal calf and thoracic spine has notably
reduced, and her arm strength has seen gradual improvement, though she
occasionally still encounters difficulty in grasping objects.
She has been undergoing physiotherapy to aid in the recovery of her arm
strength.
**Physical Examination:** No palpable lymphadenopathy. Her neurological
examination was stable with no new deficits.
**Laboratory Findings:** Most recent blood counts and biochemistry
panels showed a trend towards normalization, with liver enzymes within
the reference range.
**Imaging:**
-Ultrasound of the abdomen was conducted.
-A follow-up MRI conducted showed a reduction in the size of the
previously noted metastatic masses. There\'s a decreased impingement on
the myelon at the levels of T10-L1. The lesions in L2, L3, and L4 also
showed signs of regression.
-PET scan was performed: Favourable response. Increased FDG avidity in
the liver: Liver MRI recommended.
-Liver MRI: No pathology of the liver.
**Senological Examination:** The nodule-like deposit in the right breast
was found to be benign.
**Medication on admission:** Aspirin (Aspirin®) - 100mg, 1 tablet in the
morning Atorvastatin - 40mg - 1 tablet at bedtime Sertraline - 50mg - 2
tablets in the morning Lorazepam (Tavor®) - 1mg, as needed Fingolimod -
0.5mg, 1 capsule at bedtime, Note: Take a break as directed
Hydromorphone hydrochloride - 2mg (extended-release), 2 capsules in the
morning and 2 capsules at bedtime Melatonin - 2mg (sustained-release), 1
tablet at bedtime Baclofen (Lioresal®) - 10mg, 1 tablet three times a
day Pregabalin - 75mg, 1 capsule in the morning and 1 capsule at bedtime
MoviCOL® (Macrogol, Sodium chloride, Potassium chloride) - 1 packet
three times a day, mixed with water for oral intake Pantoprazole - 40mg,
1 tablet in the morning Colecalciferol (Vitamin D3) - 20000 I.U., 1
capsule on Monday and Thursday Co-trimoxazole - 960mg, 1 tablet on
Monday, Wednesday, and Friday Valaciclovir - 500mg, 1 tablet in the
morning and 1 tablet at bedtime Prednisolone - 50mg, 2 tablets in the
morning, Continue through 02/19/2017 Enoxaparin sodium (Clexane®) - 40mg
(4000 I.U.), 1 injection at bedtime, Note: Continue in case of
immobility Dimenhydrinate (Vomex A®) - 150mg (sustained-release), as
needed for nausea, up to 2 capsules daily.
**Physician\'s report for ultrasound on 10/05/2017:**
Liver: The liver is large with 18.1 cm in the MCL, 18.5 cm in the CCD
and 20.2 cm in the AL. The internal structure is not compacted. Focal
changes are not seen. Orthograde flow in the portal vein (vmax 16 cm/s).
Gallbladder: the gallbladder is 9.0 x 2.9 cm, the lumen is free of
stones.
Biliary tract: The intra- and extrahepatic bile ducts are not
obstructed, DHC 5 mm, DC 3 mm.
Pancreas: The pancreas is approximately 3.2/1.5/3.0 cm in size, the
internal structure is moderately echo-rich.
Spleen: The spleen is 28.0 x 9.6 cm, the parenchyma is homogeneous.
Kidneys: The right kidney is 9.8/2.0 cm, the pelvis is not congested.
The left kidney is 12.4/1.2 cm, the pelvis is not congested. Vessels
retroperitoneal: the aorta is normal in width in the partially visible
area.
Stomach/intestine: The gastric corpus wall is up to 14 mm thick. No
evidence of free fluid in the abdominal cavity.
Bladder/genitals: The prostate is orientationally about 3.8 x 4.8 x 3.1
cm, the urinary bladder is moderately full.
**MR Spine plain + post-contrast from 10/06/2017**
**Study:** Magnetic Resonance Imaging (MRI) of the thoracolumbar spine
**Clinical Information:** Follow-up MRI post treatment for previously
noted metastatic masses.
**Technique:** Standard T1-weighted, T2-weighted, and post-contrast
enhanced sequences of the thoracolumbar spine were obtained in sagittal
and axial planes.
**Findings:** There is a reduction in the size of the previously noted
metastatic masses when compared to prior MRI studies. A reduced mass
effect is observed at the levels of T10-L1. Notably, there is decreased
impingement on the myelon at these levels. This indicates a significant
improvement, suggesting a positive response to the recent therapy. The
lesion noted in the previous study at the level of L2 has shown signs of
regression in both size and intensity. Similar regression is noted for
the lesion at the L3 level. The lesion at the L4 level has also
decreased in size as compared to previous imaging. The intervertebral
discs show preserved hydration. No significant disc protrusions or
herniations are observed. The vertebral bodies do not show any
significant collapse or deformity. Bone marrow signal is otherwise
normal, apart from the aforementioned lesions. The spinal canal
maintains a normal caliber throughout, and there is no significant canal
stenosis. The conus medullaris and cauda equina nerve roots appear
unremarkable without evidence of displacement or compression.
**Impression:** Reduction in the size of previously noted metastatic
masses, indicating a positive therapeutic response. Decreased
impingement on the myelon at the levels of T10-L1, suggesting
significant regression of the previously observed mass effect.
Regression of lesions at L2, L3, and L4 levels, further indicating the
positive response to treatment.
**Positron Emission Tomography (PET)/CT from 10/09/2017:**
**Indication:** Follow-up evaluation of Diffuse large B-cell lymphoma of
germinal center type diagnosed in 01/2017.
**Technique:** Whole-body FDG-PET/CT was performed from the base of the
skull to the mid-thighs.
**Findings:** Liver: There is increased FDG uptake in the liver,
predominantly in the anterolateral segment. The size of the liver is
consistent with the previous ultrasound report, measuring 18.1 cm in the
MCL, 18.5 cm in the CCD, and 20.2 cm in the AL. The SUV max is 5.5.
Lymph Nodes: There is no pathological FDG uptake in the previously noted
left axillary lymph node, suggesting a therapeutic response. Lungs:
Previously noted deposit in the ventral left upper lobe now demonstrates
reduced FDG avidity. No other FDG-avid nodules or masses. Bone: There\'s
no FDG uptake in the spine, including the previously described
metastatic lesion, indicating a positive response to treatment.
**Impression:** Overall, the findings demonstrate a marked metabolic
improvement in the sites of lymphoma previously noted, particularly in
the left axillary lymph node and the vertebral bone lesions. The liver,
however, presents with increased FDG avidity, especially in the
anterolateral segment. This uptake might represent active lymphomatous
involvement or could be due to an inflammatory process. Given the
differential, and to ascertain the etiology, further diagnostic
evaluation, such as a liver MRI or biopsy, is recommended.
**Liver MRI from 10/11/2017:**
**Clinical Indication:** Evaluation of increased FDG uptake in the liver
as noted on the recent PET scan. Concern for active lymphomatous
involvement or an inflammatory process.
**Technique:** MRI of the liver was performed using a 3T scanner.
Sequences included T1-weighted (in-phase and out-of-phase), T2-weighted,
diffusion-weighted imaging (DWI), and post-contrast dynamic imaging
after the administration of gadolinium-based contrast agent.
**Detailed Findings:** The liver demonstrates enlargement with
measurements consistent with the recent ultrasound: 18.1 cm in the
mid-clavicular line (MCL), 18.5 cm in the maximum cranial-caudal
diameter (CCD), and 20.2 cm along the anterior line (AL).
The liver parenchyma is mostly homogenous. However, there is a region in
the anterolateral segment demonstrating T2 hyperintensity and
hypointensity on T1-weighted images. The aforementioned region in the
anterolateral segment demonstrates restricted diffusion, suggestive of
increased cellular density. After gadolinium administration, there is
peripheral enhancement of the lesion in the arterial phase, followed by
progressive central filling in portal venous and delayed phases. This
pattern is suggestive of a focal nodular hyperplasia (FNH) or atypical
hemangioma. The intrahepatic and extrahepatic bile ducts are not
dilated. No evidence of any obstructing lesion. The hepatic arteries,
portal vein, and hepatic veins appear patent with no evidence of
thrombosis or stenosis. The gallbladder, pancreas, spleen, and adjacent
segments of the bowel appear normal. No lymphadenopathy is noted in the
porta hepatis or celiac axis.
**Impression:** Enlarged liver with a suspicious lesion in the
anterolateral segment demonstrating characteristics that might be
consistent with focal nodular hyperplasia or atypical hemangioma. No
indication of lymphomatous involvement of the liver.
**Discussion: **
Given her positive response to the treatment so far, we intend to
continue with the current regimen, with careful monitoring of her side
effects and symptomatology.
We deeply appreciate your continued involvement in Mrs. Sample\'s
healthcare journey. Collaborative care is paramount, especially in cases
as complex as hers. Should you have any recommendations, insights, or if
you require additional information, please do not hesitate to reach out.
**Medication at discharge: **
Aspirin 100mg: Take 1 tablet in the morning; Atorvastatin 40mg: Take 1
tablet at bedtime; Sertraline 50mg: Take 2 tablets in the morning;
Lorazepam 1mg: Take as needed; Melatonin (sustained-release) 2mg: Take 1
tablet at bedtime; Fingolimod 0.5mg: Take 1 capsule at bedtime/take a
break as directed; Hydromorphone hydrochloride (extended-release) 2mg:
Take 2 capsules in the morning and 2 capsules at bedtime; Pregabalin
75mg: Take 1 capsule in the morning and 1 capsule at bedtime; Baclofen
10mg: Take 1 tablet three times a day; MoviCOL®: Mix 1 packet with water
and take orally three times a day; Pantoprazole 40mg: Take 1 tablet in
the morning; Colecalciferol (Vitamin D3) 20000 I.U.: Take 1 capsule on
Monday and Thursday; Dimenhydrinate (sustained-release) 150mg: Take as
needed for nausea, up to 2 capsules daily.
**Metabolic Panel**
**Parameter** **Results** **Reference Range**
---------------------------------- ------------- ---------------------
Sodium 138 mEq/L 135 - 145 mEq/L
Potassium 4.1 mEq/L 3.5 - 5.0 mEq/L
Creatinine 1.1 mg/dL 0.7 - 1.3 mg/dL
BUN (Blood Urea Nitrogen) 17 mg/dL 7 - 18 mg/dL
Alkaline Phosphatase 124 U/L 40 - 129 U/L
Total Bilirubin 0.4 mg/dL \< 1.2 mg/dL
GGT (Gamma-Glutamyl Transferase) 75 U/L \< 66 U/L
ALT (Alanine Aminotransferase) 39 U/L 10 - 50 U/L
AST (Aspartate Aminotransferase) 36 U/L 10 - 50 U/L
LDH (Lactate Dehydrogenase) 342 U/L \< 244 U/L
Uric Acid 3.8 mg/dL 3 - 7 mg/dL
Calcium 9.12 mg/dL 8.8 - 10.6 mg/dL
**Kidney Function**
**Parameter** **Results** **Reference Range**
------------------------------- ------------- ---------------------
GFR (MDRD) \>62 mL/min \> 60 mL/min
GFR (CKD-EPI with Creatinine) 67 mL/min \> 90 mL/min
**Inflammatory Markers**
**Parameter** **Results** **Reference Range**
-------------------------- ------------- ---------------------
CRP (C-Reactive Protein) 1.8 mg/dL \< 0.5 mg/dL
**Coagulation Panel**
**Parameter** **Results** **Reference Range**
--------------- ------------- ---------------------
PT Percentage 105% 70 - 120%
INR 0.98 N/A
aPTT 28 sec 26 - 37 sec
**Complete Blood Count**
**Parameter** **Results** **Reference Range**
--------------- --------------- ---------------------
WBC 11.9 x10\^9/L 4.0 - 9.0 x10\^9/L
RBC 3.1 x10\^12/L 4.5 - 6.0 x10\^12/L
Hemoglobin 8.4 g/dL 14 - 18 g/dL
Hematocrit 26% 40 - 48%
MCH 27.8 pg 27 - 32 pg
MCV 84 fL 82 - 92 fL
MCHC 33 g/dL 32 - 36 g/dL
Platelets 263 x10\^9/L 150 - 450 x10\^9/L
**Differential**
**Parameter** **Results** **Reference Range**
--------------- ------------- ---------------------
Neutrophils 73% 40 - 70%
Lymphocytes 7% 25 - 40%
Monocytes 16% 4 - 10%
Eosinophils 1% 2 - 4%
Basophils 0.5% 0 - 1%
|
Alkaline Phosphatase
|
Which language(s) are found in the WSD datasets?
|
### Introduction
Word sense disambiguation (WSD) automatically assigns a pre-defined sense to a word in a text. Different senses of a word reflect different meanings a word has in different contexts. Identifying the correct word sense given a context is crucial in natural language processing (NLP). Unfortunately, while it is easy for a human to infer the correct sense of a word given a context, it is a challenge for NLP systems. As such, WSD is an important task and it has been shown that WSD helps downstream NLP tasks, such as machine translation BIBREF0 and information retrieval BIBREF1. A WSD system assigns a sense to a word by taking into account its context, comprising the other words in the sentence. This can be done through discrete word features, which typically involve surrounding words and collocations trained using a classifier BIBREF2, BIBREF3, BIBREF4, BIBREF5. The classifier can also make use of continuous word representations of the surrounding words BIBREF6, BIBREF7. Neural WSD systems BIBREF8, BIBREF9 feed the continuous word representations into a neural network that captures the whole sentence and the word representation in the sentence. However, in both approaches, the word representations are independent of the context. Recently, pre-trained contextualized word representations BIBREF10, BIBREF11, BIBREF12, BIBREF13 have been shown to improve downstream NLP tasks. Pre-trained contextualized word representations are obtained through neural sentence encoders trained on a huge amount of raw texts. When the resulting sentence encoder is fine-tuned on the downstream task, such as question answering, named entity recognition, and sentiment analysis, with much smaller annotated training data, it has been shown that the trained model, with the pre-trained sentence encoder component, achieves new state-of-the-art results on those tasks. While demonstrating superior performance in downstream NLP tasks, pre-trained contextualized word representations are still reported to give lower accuracy compared to approaches that use non-contextualized word representations BIBREF10, BIBREF12 when evaluated on WSD. This seems counter-intuitive, as a neural sentence encoder better captures the surrounding context that serves as an important cue to disambiguate words. In this paper, we explore different strategies of integrating pre-trained contextualized word representations for WSD. Our best strategy outperforms prior methods of incorporating pre-trained contextualized word representations and achieves new state-of-the-art accuracy on multiple benchmark WSD datasets. The following sections are organized as follows. Section SECREF2 presents related work. Section SECREF3 describes our pre-trained contextualized word representation. Section SECREF4 proposes different strategies to incorporate the contextualized word representation for WSD. Section SECREF5 describes our experimental setup. Section SECREF6 presents the experimental results. Section SECREF7 discusses the findings from the experiments. Finally, Section SECREF8 presents the conclusion. ### Related Work
Continuous word representations in real-valued vectors, or commonly known as word embeddings, have been shown to help improve NLP performance. Initially, exploiting continuous representations was achieved by adding real-valued vectors as classification features BIBREF14. BIBREF6 fine-tuned non-contextualized word embeddings by a feed-forward neural network such that those word embeddings were more suited for WSD. The fine-tuned embeddings were incorporated into an SVM classifier. BIBREF7 explored different strategies of incorporating word embeddings and found that their best strategy involved exponential decay that decreased the contribution of surrounding word features as their distances to the target word increased. The neural sequence tagging approach has also been explored for WSD. BIBREF8 proposed bidirectional long short-term memory (LSTM) BIBREF15 for WSD. They concatenated the hidden states of the forward and backward LSTMs and fed the concatenation into an affine transformation followed by softmax normalization, similar to the approach to incorporate a bidirectional LSTM adopted in sequence labeling tasks such as part-of-speech tagging and named entity recognition BIBREF16. BIBREF9 proposed a self-attention layer on top of the concatenated bidirectional LSTM hidden states for WSD and introduced multi-task learning with part-of-speech tagging and semantic labeling as auxiliary tasks. However, on average across the test sets, their approach did not outperform SVM with word embedding features. Subsequently, BIBREF17 proposed the incorporation of glosses from WordNet in a bidirectional LSTM for WSD, and reported better results than both SVM and prior bidirectional LSTM models. A neural language model (LM) is aimed at predicting a word given its surrounding context. As such, the resulting hidden representation vector captures the context of a word in a sentence. BIBREF10 designed context2vec, which is a one-layer bidirectional LSTM trained to maximize the similarity between the hidden state representation of the LSTM and the target word embedding. BIBREF12 designed ELMo, which is a two-layer bidirectional LSTM language model trained to predict the next word in the forward LSTM and the previous word in the backward LSTM. In both models, WSD was evaluated by nearest neighbor matching between the test and training instance representations. However, despite training on a huge amount of raw texts, the resulting accuracies were still lower than those achieved by WSD approaches with pre-trained non-contextualized word representations. End-to-end neural machine translation (NMT) BIBREF18, BIBREF19 learns to generate an output sequence given an input sequence, using an encoder-decoder model. The encoder captures the contextualized representation of the words in the input sentence for the decoder to generate the output sentence. Following this intuition, BIBREF11 trained an encoder-decoder model on parallel texts and obtained pre-trained contextualized word representations from the encoder. ### Pre-Trained Contextualized Word Representation
The contextualized word representation that we use is BERT BIBREF13, which is a bidirectional transformer encoder model BIBREF20 pre-trained on billions of words of texts. There are two tasks on which the model is trained, i.e., masked word and next sentence prediction. In both tasks, prediction accuracy is determined by the ability of the model to understand the context. A transformer encoder computes the representation of each word through an attention mechanism with respect to the surrounding words. Given a sentence $x^n_1$ of length $n$, the transformer computes the representation of each word $x_i$ through a multi-head attention mechanism, where the query vector is from $x_i$ and the key-value vector pairs are from the surrounding words $x_{i^{\prime }}$ ($1 \le i^{\prime } \le n$). The word representation produced by the transformer captures the contextual information of a word. The attention mechanism can be viewed as mapping a query vector $\mathbf {q}$ and a set of key-value vector pairs $(\mathbf {k}, \mathbf {v})$ to an output vector. The attention function $A(\cdot )$ computes the output vector which is the weighted sum of the value vectors and is defined as: where $\mathbf {K}$ and $\mathbf {V}$ are matrices, containing the key vectors and the value vectors of the words in the sentence respectively, and $\alpha (\mathbf {q}, \mathbf {k}, \rho )$ is a scalar attention weight between $\mathbf {q}$ and $\mathbf {k}$, re-scaled by a scalar $\rho $. Two building blocks for the transformer encoder are the multi-head attention mechanism and the position-wise feed-forward neural network (FFNN). The multi-head attention mechanism with $H$ heads leverages the attention function in Equation DISPLAY_FORM1 as follows: where $\oplus $ denotes concatenation of vectors, $\mathbf {W}_\text{MH} \in \mathbb {R}^{d_\text{model} \times Hd_\mathbf {v}}$, $\mathbf {W}^\mathbf {Q}_\eta , \mathbf {W}^\mathbf {K}_\eta \in \mathbb {R}^{d_\mathbf {k} \times d_\text{model}}$, and $ \mathbf {W}^\mathbf {V}_\eta \in \mathbb {R}^{d_\mathbf {v} \times d_\text{model}}$. The input vector $\mathbf {q} \in \mathbb {R}^{d_\text{model}}$ is the hidden vector for the ambiguous word, while input matrices $\mathbf {K}, \mathbf {V} \in \mathbb {R}^{d_\text{model} \times n}$ are the concatenation of the hidden vectors of all words in the sentence. For each attention head, the dimension of both the query and key vectors is $d_\mathbf {k}$ while the dimension of the value vector is $d_\mathbf {v}$. The encoder model dimension is $d_\text{model}$. The position-wise FFNN performs a non-linear transformation on the attention output corresponding to each input word position as follows: in which the input vector $\mathbf {u} \in \mathbb {R}^{d_\text{model}}$ is transformed to the output vector with dimension $d_\text{model}$ via a series of linear projections with the ReLU activation function. For the hidden layer $l$ ($1 \le l \le L$), the self-attention sub-layer output $\mathbf {f}^l_i$ is computed as follows: where LayerNorm refers to layer normalization BIBREF21 and the superscript $l$ and subscript $\mathbf {h}$ indicate that each encoder layer $l$ has an independent set of multi-head attention weight parameters (see Equations DISPLAY_FORM2 and ). The input for the first layer is $\mathbf {h}^0_i = \mathbf {E}(x_i)$, which is the non-contextualized word embedding of $x_i$. The second sub-layer consists of the position-wise fully connected FFNN, computed as: where, similar to self-attention, an independent set of weight parameters in Equation DISPLAY_FORM3 is defined in each layer. ### Incorporating Pre-Trained Contextualized Word Representation
As BERT is trained on the masked word prediction task, which is to predict a word given the surrounding (left and right) context, the pre-trained model already captures the context. In this section, we describe different techniques of leveraging BERT for WSD, broadly categorized into nearest neighbor matching and linear projection of hidden layers. ### Incorporating Pre-Trained Contextualized Word Representation ::: Nearest Neighbor Matching
A straightforward way to disambiguate word sense is through 1-nearest neighbor matching. We compute the contextualized representation of each word in the training data and the test data through BERT. Given a hidden representation $\mathbf {h}^L_{i}$ at the $L$-th layer for word $x_i$ in the test data, nearest neighbor matching finds a vector $\mathbf {h^*}$ in the $L$-th layer from the training data such that where the sense assigned to $x_i$ is the sense of the word whose contextualized representation is $\mathbf {h^*}$. This WSD technique is adopted in earlier work on contextualized word representations BIBREF10, BIBREF12. ### Incorporating Pre-Trained Contextualized Word Representation ::: Linear Projection of Hidden Layers
Apart from nearest neighbor matching, we can perform a linear projection of the hidden vector $\mathbf {h}_i$ by an affine transformation into an output sense vector, with its dimension equal to the number of senses for word $x_i$. The output of this affine transformation is normalized by softmax such that all its values sum to 1. Therefore, the predicted sense $\mathbf {s}_i$ of word $x_i$ is formulated as where $\mathbf {s}_i$ is a vector of predicted sense distribution for word $x_i$, while $\mathbf {W}^{\text{lexelt}(x_i)}$ and $\mathbf {b}^{\text{lexelt}(x_i)}$ are respectively the projection matrix and bias vector specific to the lexical element (lexelt) of word $x_i$, which consists of its lemma and optionally its part-of-speech tag. We choose the sense corresponding to the element of $\mathbf {s}_i$ with the maximum value. Training the linear projection model is done by the back-propagation algorithm, which updates the model parameters to minimize a cost function. Our cost function is the negative log-likelihood of the softmax output value that corresponds to the tagged sense in the training data. In addition, we propose two novel ways of incorporating BERT's hidden representation vectors to compute the sense output vector, which are described in the following sub-subsections. ### Incorporating Pre-Trained Contextualized Word Representation ::: Linear Projection of Hidden Layers ::: Last Layer Projection
Similar to the nearest neighbor matching model, we can feed the hidden vector of BERT in the last layer, $\mathbf {h}^L_i$, into an affine transformation followed by softmax. That is, $\mathbf {h}_i$ in Equation DISPLAY_FORM10 is instantiated by $\mathbf {h}^L_i$. The last layer projection model is illustrated in Figure FIGREF7(a). ### Incorporating Pre-Trained Contextualized Word Representation ::: Linear Projection of Hidden Layers ::: Weighted Sum of Hidden Layers
BERT consists of multiple layers stacked one after another. Each layer carries a different representation of word context. Taking into account different hidden layers may help the WSD system to learn from different context information encoded in different layers of BERT. To take into account all layers, we compute the weighted sum of all hidden layers, $\mathbf {h}^l_i$, where $1 \le l \le L$, corresponding to a word position $i$, through attention mechanism. That is, $\mathbf {h}_i$ in Equation DISPLAY_FORM10 is replaced by the following equation: where $\mathbf {m} \in \mathbb {R}^{d_\text{model}}$ is a projection vector to obtain scalar values with the key vectors. The model with weighted sum of all hidden layers is illustrated in Figure FIGREF7(b). ### Incorporating Pre-Trained Contextualized Word Representation ::: Linear Projection of Hidden Layers ::: Gated Linear Unit
While the contextualized representations in the BERT hidden layer vectors are features that determine the word sense, some features are more useful than the others. As such, we propose filtering the vector values by a gating vector whose values range from 0 to 1. This mechanism is known as the gated linear unit (GLU) BIBREF22, which is formulated as where $\mathbf {W}^\mathbf {h}$ and $\mathbf {W}^\mathbf {g}$ are separate projection matrices and $\mathbf {b}^\mathbf {h}$ and $\mathbf {b}^\mathbf {g}$ are separate bias vectors. The symbols $\sigma (\cdot )$ and $\odot $ denote the sigmoid function and element-wise vector multiplication operation respectively. GLU transforms the input vector $\mathbf {h}$ by feeding it to two separate affine transformations. The second transformation is used as the sigmoid gate to filter the input vector, which is summed with the vector after the first affine transformation. ### Experimental Setup
We conduct experiments on various WSD tasks. The description and the statistics for each task are given in Table . For English, a lexical element (lexelt) is defined as a combination of lemma and part-of-speech tag, while for Chinese, it is simply the lemma, following the OntoNotes setup. We exploit English BERT$_\text{BASE}$ for the English tasks and Chinese BERT for the Chinese task. We conduct experiments with different strategies of incorporating BERT as described in Section SECREF4, namely 1-nearest neighbor matching (1-nn) and linear projection. In the latter technique, we explore strategies including simple last layer projection, layer weighting (LW), and gated linear unit (GLU). In the linear projection model, we train the model by the Adam algorithm BIBREF23 with a learning rate of $10^{-3}$. The model parameters are updated per mini-batch of 16 sentences. As update progresses, we pick the best model parameters from a series of neural network updates based on accuracy on a held-out development set, disjoint from the training set. The state-of-the-art supervised WSD approach takes into account features from the neighboring sentences, typically one sentence to the left and one to the right apart from the current sentence that contains the ambiguous words. We also exploit this in our model, as BERT supports inputs with multiple sentences separated by a special [SEP] symbol. For English all-words WSD, we train our WSD model on SemCor BIBREF24, and test it on Senseval-2 (SE2), Senseval-3 (SE3), SemEval 2013 task 12 (SE13), and SemEval 2015 task 13 (SE15). This common benchmark, which has been annotated with WordNet-3.0 senses BIBREF25, has recently been adopted in English all-words WSD. Following BIBREF9, we choose SemEval 2007 Task 17 (SE07) as our development data to pick the best model parameters after a number of neural network updates, for models that require back-propagation training. We also evaluate on Senseval-2 and Senseval-3 English lexical sample tasks, which come with pre-defined training and test data. For each word type, we pick 20% of the training instances to be our development set and keep the remaining 80% as the actual training data. Through this development set, we determine the number of epochs to use in training. We then re-train the model with the whole training dataset using the number of epochs identified in the initial training step. While WSD is predominantly evaluated on English, we are also interested in evaluating our approach on Chinese, to evaluate the effectiveness of our approach in a different language. We use OntoNotes Release 5.0, which contains a number of annotations including word senses for Chinese. We follow the data setup of BIBREF26 and conduct an evaluation on four genres, i.e., broadcast conversation (BC), broadcast news (BN), magazine (MZ), and newswire (NW), as well as the concatenation of all genres. While the training and development datasets are divided into genres, we train on the concatenation of all genres and test on each individual genre. For Chinese WSD evaluation, we train IMS BIBREF5 on the Chinese OntoNotes dataset as our baseline. We also incorporate pre-trained non-contextualized Chinese word embeddings as IMS features BIBREF6, BIBREF7. The pre-trained word embeddings are obtained by training the word2vec skip-gram model on Chinese Gigaword Fifth Edition, which after automatic word segmentation contains approximately 2 billion words. Following BIBREF6, we incorporate the embedding features of words within a window surrounding the target ambiguous word. In our experiments, we take into account 5 words to the left and right. ### Results
We present our experimental results and compare them with prior baselines. ### Results ::: English All-Words Tasks
For English all-words WSD, we compare our approach with three categories of prior approaches. Firstly, we compare our approach with the supervised SVM classifier approach, namely IMS BIBREF5. We compare our approach with both the original IMS without word embedding features and IMS with non-contextualized word embedding features, that is, word2vec with exponential decay BIBREF7. We also compare with SupWSD BIBREF27, which is an alternative implementation of IMS. Secondly, we compare our approach with the neural WSD approaches that leverage bidirectional LSTM (bi-LSTM). These include the bi-LSTM model with attention trained jointly with lexical semantic labeling task BIBREF9 (BiLSTMatt+LEX) and the bi-LSTM model enhanced with gloss representation from WordNet (GAS). Thirdly, we show comparison with prior contextualized word representations for WSD, pre-trained on a large number of texts, namely context2vec BIBREF10 and ELMo BIBREF12. In these two models, WSD is treated as nearest neighbor matching as described in Section SECREF4. Table shows our WSD results in F1 measure. It is shown in the table that with the nearest neighbor matching model, BERT outperforms context2vec and ELMo. This shows the effectiveness of BERT's pre-trained contextualized word representation. When we include surrounding sentences, one to the left and one to the right, we get improved F1 scores consistently. We also show that linear projection to the sense output vector further improves WSD performance, by which our best results are achieved. While BERT has been shown to outperform other pre-trained contextualized word representations through the nearest neighbor matching experiments, its potential can be maximized through linear projection to the sense output vector. It is worthwhile to note that our more advanced linear projection, by means of layer weighting (§SECREF12 and gated linear unit (§SECREF14) gives the best F1 scores on all test sets. All our BERT WSD systems outperform gloss-enhanced neural WSD, which has the best overall score among all prior systems. ### Results ::: English Lexical Sample Tasks
For English lexical sample tasks, we compare our approach with the original IMS BIBREF5 and IMS with non-contextualized word embedding features. The embedding features incorporated into IMS include CW embeddings BIBREF28, obtained from a convolutional language model, fine-tuned (adapted) to WSD BIBREF6 (+adapted CW), and word2vec skip-gram BIBREF29 with exponential decay BIBREF7 (+w2v+expdecay). We also compare our approach with the bi-LSTM, on top of which sense classification is defined BIBREF8, and context2vec BIBREF10, which is a contextualized pre-trained bi-LSTM model trained on 2B words of text. Finally, we also compare with a prior multi-task and semi-supervised WSD approach learned through alternating structure optimization (ASO) BIBREF3, which also utilizes unlabeled data for training. As shown in Table , our BERT-based WSD approach with linear projection model outperforms all prior approaches. context2vec, which is pre-trained on a large amount of texts, performs worse than the prior semi-supervised ASO approach on Senseval-3, while our best result outperforms ASO by a large margin. Neural bi-LSTM performs worse than IMS with non-contextualized word embedding features. Our neural model with pre-trained contextualized word representations outperforms the best result achieved by IMS on both Senseval-2 and Senseval-3. ### Results ::: Chinese OntoNotes WSD
We compare our approach with IMS without and with word embedding features as the baselines. The results are shown in Table . Across all genres, BERT outperforms the baseline IMS with word embedding (non-contextualized word representation) features BIBREF6. The latter also improves over the original IMS without word embedding features BIBREF5. Linear projection approaches consistently outperform nearest neighbor matching by a significant margin, similar to the results on English WSD tasks. The best overall result for the Chinese OntoNotes test set is achieved by the models with simple projection and hidden layer weighting. ### Discussion
Across all tasks (English all-words, English lexical sample, and Chinese OntoNotes), our experiments demonstrate the effectiveness of BERT over various prior WSD approaches. The best results are consistently obtained by linear projection models, which project the last hidden layer or the weighted sum of all hidden layers to an output sense vector. We can view the BERT hidden layer outputs as contextual features, which serve as useful cues in determining the word senses. In fact, the attention mechanism in transformer captures the surrounding words. In prior work like IMS BIBREF5, these contextual cues are captured by the manually-defined surrounding word and collocation features. The features obtained by the hidden vector output are shown to be more effective than the manually-defined features. We introduced two advanced linear projection techniques, namely layer weighting and gated linear unit (GLU). While BIBREF12 showed that the second biLSTM layer results in better WSD accuracy compared to the first layer (nearer to the individual word representation), we showed that taking into account different layers by means of the attention mechanism is useful for WSD. GLU as an activation function has been shown to be effective for better convergence and to overcome the vanishing gradient problem in the convolutional language model BIBREF22. In addition, the GLU gate vector, with values ranging from 0 to 1, can be seen as a filter for the features from the hidden layer vector. Compared with two prior contextualized word representations models, context2vec BIBREF10 and ELMo BIBREF12, BERT achieves higher accuracy. This shows the effectiveness of the attention mechanism used in the transformer model to represent the context. Our BERT WSD models outperform prior neural WSD models by a large margin. These prior neural WSD models perform comparably with IMS with embeddings as classifier features, in addition to the discrete features. While neural WSD approaches BIBREF8, BIBREF9, BIBREF17 exploit non-contextualized word embeddings which are trained on large texts, the hidden layers are trained only on a small amount of labeled data. ### Conclusion
For the WSD task, we have proposed novel strategies of incorporating BERT, a pre-trained contextualized word representation which effectively captures the context in its hidden vectors. Our experiments show that linear projection of the hidden vectors, coupled with gating to filter the values, gives better results than the prior state of the art. Compared to prior neural and feature-based WSD approaches that make use of non-contextualized word representations, using pre-trained contextualized word representation with our proposed incorporation strategy achieves significantly higher scores. Figure 1: Illustration of WSD models by linear projection of (a) the last layer and (b) the weighted sum of all layers. Table 1: Statistics of the datasets used for the English all-words task, English lexical sample task, and Chinese OntoNotes WSD task in terms of the number of instances and the number of distinct lexelts. For Chinese WSD task, “All” refers to the concatenation of all genres BC, BN, MZ, and NW. Table 2: English all-words task results in F1 measure (%), averaged over three runs. SemEval 2007 Task 17 (SE07) test set is used as the development set. We show the results of nearest neighbor matching (1nn) and linear projection, by simple last layer linear projection, layer weighting (LW), and gated linear units (GLU). Apart from BERT representation of one sentence (1sent), we also show BERT representation of one sentence plus one surrounding sentence to the left and one to the right (1sent+1sur). The best result in each dataset is shown in bold. Statistical significance tests by bootstrap resampling (∗: p < 0.05) compare 1nn (1sent+1sur) with each of Simple (1sent+1sur), LW (1sent+1sur), GLU (1sent+1sur), and GLU+LW (1sent+1sur). Table 3: English lexical sample task results in accuracy (%), averaged over three runs. Best accuracy in each dataset is shown in bold. Statistical significance tests by bootstrap resampling (∗: p < 0.05) compare 1nn (1sent+1sur) with each of Simple (1sent+1sur), LW (1sent+1sur), GLU (1sent+1sur), and GLU+LW (1sent+1sur). Table 4: Chinese OntoNotes WSD results in accuracy (%), averaged over three runs, for each genre. All BERT results in this table are obtained from the representation of one sentence plus one surrounding sentence to the left and to the right (1sent+1sur). We show results of various BERT incorporation strategy, namely nearest neighbor matching (1nn), simple projection, projection with layer weighting (LW) and gated linear unit (GLU). Best accuracy in each genre is shown in bold. Statistical significance tests by bootstrap resampling (∗: p < 0.05) compare 1nn with each of Simple, LW, GLU, and GLU+LW.
|
WSD is predominantly evaluated on English, we are also interested in evaluating our approach on Chinese
|
Why don't Harry's parents want him to join the UN?
A. Harry's parents think he is too lazy to succeed in the UN.
B. Harry's parents want him to go to trade school.
C. Harry's parents feel that joining the UN means he is turning his back on America.
D. Harry's parents don't want him to be a soldier.
|
Transcriber's Note: This etext was produced from Analog, January 1961. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. THE GREEN BERET By TOM PURDOM It's not so much the decisions a man does make that mark him as a Man—but the ones he refrains from making. Like the decision "I've had enough!" Illustrated by Schoenherr Read locked the door and drew his pistol. Sergeant Rashid handed Premier Umluana the warrant. "We're from the UN Inspector Corps," Sergeant Rashid said. "I'm very sorry, but we have to arrest you and bring you in for trial by the World Court." If Umluana noticed Read's gun, he didn't show it. He read the warrant carefully. When he finished, he said something in Dutch. "I don't know your language," Rashid said. "Then I'll speak English." Umluana was a small man with wrinkled brow, glasses and a mustache. His skin was a shade lighter than Read's. "The Inspector General doesn't have the power to arrest a head of state—especially the Premier of Belderkan. Now, if you'll excuse me, I must return to my party." In the other room people laughed and talked. Glasses clinked in the late afternoon. Read knew two armed men stood just outside the door. "If you leave, Premier, I'll have to shoot you." "I don't think so," Umluana said. "No, if you kill me, all Africa will rise against the world. You don't want me dead. You want me in court." Read clicked off the safety. "Corporal Read is very young," Rashid said, "but he's a crack shot. That's why I brought him with me. I think he likes to shoot, too." Umluana turned back to Rashid a second too soon. He saw the sergeant's upraised hand before it collided with his neck. "Help! Kidnap. " Rashid judo chopped him and swung the inert body over his shoulders. Read pulled a flat grenade from his vest pocket. He dropped it and yellow psycho gas hissed from the valve. "Let's be off," Rashid said. The door lock snapped as they went out the window. Two men with rifles plunged into the gas; sighing, they fell to the floor in a catatonic trance. A little car skimmed across the lawn. Bearing the Scourge of Africa, Rashid struggled toward it. Read walked backward, covering their retreat. The car stopped, whirling blades holding it a few inches off the lawn. They climbed in. "How did it go?" The driver and another inspector occupied the front seat. "They'll be after us in half a minute." The other inspector carried a light machine gun and a box of grenades. "I better cover," he said. "Thanks," Rashid said. The inspector slid out of the car and ran to a clump of bushes. The driver pushed in the accelerator. As they swerved toward the south, Read saw a dozen armed men run out of the house. A grenade arced from the bushes and the pursuers recoiled from the cloud that rose before them. "Is he all right?" the driver asked. "I don't think I hurt him." Rashid took a syrette from his vest pocket. "Well, Read, it looks like we're in for a fight. In a few minutes Miaka Station will know we're coming. And God knows what will happen at the Game Preserve." Read wanted to jump out of the car. He could die any minute. But he had set his life on a well-oiled track and he couldn't get off until they reached Geneva. "They don't know who's coming," he said. "They don't make them tough enough to stop this boy." Staring straight ahead, he didn't see the sergeant smile. Two types of recruits are accepted by the UN Inspector Corps: those with a fanatic loyalty to the ideals of peace and world order, and those who are loyal to nothing but themselves. Read was the second type. A tall, lanky Negro he had spent his school days in one of the drab suburbs that ring every prosperous American city. It was the home of factory workers, clerks, semiskilled technicians, all who do the drudge work of civilization and know they will never do more. The adults spent their days with television, alcohol and drugs; the young spent their days with gangs, sex, television and alcohol. What else was there? Those who could have told him neither studied nor taught at his schools. What he saw on the concrete fields between the tall apartment houses marked the limits of life's possibilities. He had belonged to a gang called The Golden Spacemen. "Nobody fools with me," he bragged. "When Harry Read's out, there's a tiger running loose." No one knew how many times he nearly ran from other clubs, how carefully he picked the safest spot on the battle line. "A man ought to be a man," he once told a girl. "He ought to do a man's work. Did you ever notice how our fathers look, how they sleep so much? I don't want to be like that. I want to be something proud." He joined the UN Inspector Corps at eighteen, in 1978. The international cops wore green berets, high buttonless boots, bush jackets. They were very special men. For the first time in his life, his father said something about his ambitions. "Don't you like America, Harry? Do you want to be without a country? This is the best country in the world. All my life I've made a good living. Haven't you had everything you ever wanted? I've been a king compared to people overseas. Why, you stay here and go to trade school and in two years you'll be living just like me." "I don't want that," Read said. "What do you mean, you don't want that?" "You could join the American Army," his mother said. "That's as good as a trade school. If you have to be a soldier." "I want to be a UN man. I've already enlisted. I'm in! What do you care what I do?" The UN Inspector Corps had been founded to enforce the Nuclear Disarmament Treaty of 1966. Through the years it had acquired other jobs. UN men no longer went unarmed. Trained to use small arms and gas weapons, they guarded certain borders, bodyguarded diplomats and UN officials, even put down riots that threatened international peace. As the UN evolved into a strong world government, the UN Inspector Corps steadily acquired new powers. Read went through six months training on Madagascar. Twice he nearly got expelled for picking fights with smaller men. Rather than resign, he accepted punishment which assigned him to weeks of dull, filthy extra labor. He hated the restrictions and the iron fence of regulations. He hated boredom, loneliness and isolation. And yet he responded with enthusiasm. They had given him a job. A job many people considered important. He took his turn guarding the still disputed borders of Korea. He served on the rescue teams that patrol the busy Polar routes. He mounted guard at the 1980 World's Fair in Rangoon. "I liked Rangoon," he even told a friend. "I even liked Korea. But I think I liked the Pole job best. You sit around playing cards and shooting the bull and then there's a plane crash or something and you go out and win a medal. That's great for me. I'm lazy and I like excitement." One power implied in the UN Charter no Secretary General or Inspector General had ever tried to use. The power to arrest any head of state whose country violated international law. Could the World Court try and imprison a politician who had conspired to attack another nation? For years Africa had been called "The South America of the Old World." Revolution followed revolution. Colonies became democracies. Democracies became dictatorships or dissolved in civil war. Men planted bases on the moon and in four years, 1978-82, ringed the world with matter transmitters; but the black population of Africa still struggled toward political equality. Umluana took control of Belderkan in 1979. The tiny, former Dutch colony, had been a tottering democracy for ten years. The very day he took control the new dictator and his African party began to build up the Belderkan Army. For years he had preached a new Africa, united, free of white masters, the home of a vigorous and perfect Negro society. His critics called him a hypocritical racist, an opportunist using the desires of the African people to build himself an empire. He began a propaganda war against neighboring South Africa, promising the liberation of that strife-torn land. Most Negro leaders, having just won representation in the South African Parliament, told him to liberate his own country. They believed they could use their first small voice in the government to win true freedom for their people. But the radio assault and the arms buildup continued. Early in 1982, South Africa claimed the Belderkan Army exceeded the size agreed to in the Disarmament Treaty. The European countries and some African nations joined in the accusation. China called the uproar a vicious slur on a new African nation. The United States and Russia, trying not to get entangled, asked for more investigation by the UN. But the evidence was clear. Umluana was defying world law. If he got away with it, some larger and more dangerous nation might follow his precedent. And the arms race would begin again. The Inspector General decided. They would enter Belderkan, arrest Umluana and try him by due process before the World Court. If the plan succeeded, mankind would be a long step farther from nuclear war. Read didn't know much about the complicated political reasons for the arrest. He liked the Corp and he liked being in the Corp. He went where they sent him and did what they told him to do. The car skimmed above the tree-tops. The driver and his two passengers scanned the sky. A plane would have been a faster way to get out of the country. But then they would have spent hours flying over Africa, with Belderkan fighters in hot pursuit, other nations joining the chase and the world uproar gaining volume. By transmitter, if all went well, they could have Umluana in Geneva in an hour. They were racing toward Miaka, a branch transmitter station. From Miaka they would transmit to the Belderkan Preserve, a famous tourist attraction whose station could transmit to any point on the globe. Even now a dozen inspectors were taking over the Game Preserve station and manning its controls. They had made no plans to take over Miaka. They planned to get there before it could be defended. "There's no military base near Miaka," Rashid said. "We might get there before the Belderkans." "Here comes our escort," Read said. A big car rose from the jungle. This one had a recoilless rifle mounted on the roof. The driver and the gunner waved and fell in behind them. "One thing," Read said, "I don't think they'll shoot at us while he's in the car." "Don't be certain, corporal. All these strong-arm movements are alike. I'll bet Umluana's lieutenants are hoping he'll become a dead legend. Then they can become live conquerors." Sergeant Rashid came from Cairo. He had degrees in science and history from Cambridge but only the Corp gave him work that satisfied his conscience. He hated war. It was that simple. Read looked back. He saw three spots of sunlight about two hundred feet up and a good mile behind. "Here they come, Sarge." Rashid turned his head. He waved frantically. The two men in the other car waved back. "Shall I duck under the trees?" the driver asked. "Not yet. Not until we have to." Read fingered the machine gun he had picked up when he got in the car. He had never been shot at. Twice he had faced an unarmed mob, but a few shots had sent them running. Birds flew screaming from their nests. Monkeys screeched and threw things at the noisy, speeding cars. A little cloud of birds surrounded each vehicle. The escort car made a sharp turn and charged their pursuers. The big rifle fired twice. Read saw the Belderkan cars scatter. Suddenly machine-gun bullets cracked and whined beside him. "Evade," Rashid said. "Don't go down." Without losing any forward speed, the driver took them straight up. Read's stomach bounced. A shell exploded above them. The car rocked. He raised his eyes and saw a long crack in the roof. "Hit the floor," Rashid said. They knelt on the cramped floor. Rashid put on his gas mask and Read copied him. Umluana breathed like a furnace, still unconscious from the injection Rashid had given him. I can't do anything , Read thought. They're too far away to shoot back. All we can do is run. The sky was clear and blue. The jungle was a noisy bazaar of color. In the distance guns crashed. He listened to shells whistle by and the whipcrack of machine-gun bullets. The car roller-coastered up and down. Every time a shell passed, he crawled in waves down his own back. Another explosion, this time very loud. Rashid raised his eyes above the seat and looked out the rear window. "Two left. Keep down, Read." "Can't we go down?" Read said. "They'll get to Miaka before us." He shut his eyes when he heard another loud explosion. Sergeant Rashid looked out the window again. He swore bitterly in English and Egyptian. Read raised his head. The two cars behind them weren't fighting each other. A long way back the tree-tops burned. "How much farther?" Rashid said. The masks muffled their voices. "There it is now. Shall I take us right in?" "I think you'd better." The station was a glass diamond in a small clearing. The driver slowed down, then crashed through the glass walls and hovered by the transmitter booth. Rashid opened the door and threw out two grenades. Read jumped out and the two of them struggled toward the booth with Umluana. The driver, pistol in hand, ran for the control panel. There were three technicians in the station and no passengers. All three panicked when the psycho gas enveloped them. They ran howling for the jungle. Through the window of his mask, Read saw their pursuers land in the clearing. Machine-gun bullets raked the building. They got Umluana in the booth and hit the floor. Read took aim and opened fire on the largest car. "Now, I can shoot back," he said. "Now we'll see what they do." "Are you ready, Rashid?" yelled the driver. "Man, get us out of here!" The booth door shut. When it opened, they were at the Game Preserve. The station jutted from the side of a hill. A glass-walled waiting room surrounded the bank of transmitter booths. Read looked out the door and saw his first battlefield. Directly in front of him, his head shattered by a bullet, a dead inspector lay behind an overturned couch. Read had seen dozens of training films taken during actual battles or after atomic attacks. He had laughed when other recruits complained. "That's the way this world is. You people with the weak stomachs better get used to it." Now he slid against the rear wall of the transmitter booth. A wounded inspector crawled across the floor to the booth. Read couldn't see his wound, only the pain scratched on his face and the blood he deposited on the floor. "Did you get Umluana?" he asked Sergeant Rashid. "He's in the booth. What's going on?" Rashid's Middle East Oxford seemed more clipped than ever. "They hit us with two companies of troops a few minutes ago. I think half our men are wounded." "Can we get out of here?" "They machine-gunned the controls." Rashid swore. "You heard him, Read! Get out there and help those men." He heard the screams of the wounded, the crack of rifles and machine guns, all the terrifying noise of war. But since his eighteenth year he had done everything his superiors told him to do. He started crawling toward an easy-chair that looked like good cover. A bullet cracked above his head, so close he felt the shock wave. He got up, ran panicky, crouched, and dove behind the chair. An inspector cracked the valve on a smoke grenade. A white fog spread through the building. They could see anyone who tried to rush them but the besiegers couldn't pick out targets. Above the noise, he heard Rashid. "I'm calling South Africa Station for a copter. It's the only way out of here. Until it comes, we've got to hold them back." Read thought of the green beret he had stuffed in his pocket that morning. He stuck it on his head and cocked it. He didn't need plain clothes anymore and he wanted to wear at least a part of his uniform. Bullets had completely shattered the wall in front of him. He stared through the murk, across the broken glass. He was Corporal Harry Read, UN Inspector Corps—a very special man. If he didn't do a good job here, he wasn't the man he claimed to be. This might be the only real test he would ever face. He heard a shout in rapid French. He turned to his right. Men in red loincloths ran zigzagging toward the station. They carried light automatic rifles. Half of them wore gas masks. "Shoot the masks," he yelled. "Aim for the masks." The machine gun kicked and chattered on his shoulder. He picked a target and squeezed off a burst. Tensely, he hunted for another mask. Three grenades arced through the air and yellow gas spread across the battlefield. The attackers ran through it. A few yards beyond the gas, some of them turned and ran for their own lines. In a moment only half a dozen masked men still advanced. The inspectors fired a long, noisy volley. When they stopped only four attackers remained on their feet. And they were running for cover. The attackers had come straight up a road that led from the Game Preserve to the station. They had not expected any resistance. The UN men had already taken over the station, chased out the passengers and technicians and taken up defense positions; they had met the Belderkans with a dozen grenades and sent them scurrying for cover. The fight so far had been vicious but disorganized. But the Belderkans had a few hundred men and knew they had wrecked the transmitter controls. The first direct attack had been repulsed. They could attack many more times and continue to spray the building with bullets. They could also try to go around the hill and attack the station from above; if they did, the inspectors had a good view of the hill and should see them going up. The inspectors had taken up good defensive positions. In spite of their losses, they still had enough firepower to cover the area surrounding the station. Read surveyed his sector of fire. About two hundred yards to his left, he saw the top of a small ditch. Using the ditch for cover, the Belderkans could sneak to the top of the hill. Gas grenades are only three inches long. They hold cubic yards of gas under high pressure. Read unclipped a telescoping rod from his vest pocket. He opened it and a pair of sights flipped up. A thin track ran down one side. He had about a dozen grenades left, three self-propelling. He slid an SP grenade into the rod's track and estimated windage and range. Sighting carefully, not breathing, muscles relaxed, the rod rock steady, he fired and lobbed the little grenade into the ditch. He dropped another grenade beside it. The heavy gas would lie there for hours. Sergeant Rashid ran crouched from man to man. He did what he could to shield the wounded. "Well, corporal, how are you?" "Not too bad, sergeant. See that ditch out there? I put a little gas in it." "Good work. How's your ammunition?" "A dozen grenades. Half a barrel of shells." "The copter will be here in half an hour. We'll put Umluana on, then try to save ourselves. Once he's gone, I think we ought to surrender." "How do you think they'll treat us?" "That we'll have to see." An occasional bullet cracked and whined through the misty room. Near him a man gasped frantically for air. On the sunny field a wounded man screamed for help. "There's a garage downstairs," Rashid said. "In case the copter doesn't get here on time, I've got a man filling wine bottles with gasoline." "We'll stop them, Sarge. Don't worry." Rashid ran off. Read stared across the green land and listened to the pound of his heart. What were the Belderkans planning? A mass frontal attack? To sneak in over the top of the hill? He didn't think, anymore than a rabbit thinks when it lies hiding from the fox or a panther thinks when it crouches on a branch above the trail. His skin tightened and relaxed on his body. "Listen," said a German. Far down the hill he heard the deep-throated rumble of a big motor. "Armor," the German said. The earth shook. The tank rounded the bend. Read watched the squat, angular monster until its stubby gun pointed at the station. It stopped less than two hundred yards away. A loud-speaker blared. ATTENTION UN SOLDIERS. ATTENTION UN SOLDIERS. YOU MAY THINK US SAVAGES BUT WE HAVE MODERN WEAPONS. WE HAVE ATOMIC WARHEADS, ALL GASES, ROCKETS AND FLAME THROWERS. IF YOU DO NOT SURRENDER OUR PREMIER, WE WILL DESTROY YOU. "They know we don't have any big weapons," Read said. "They know we have only gas grenades and small arms." He looked nervously from side to side. They couldn't bring the copter in with that thing squatting out there. A few feet away, sprawled behind a barricade of tables, lay a man in advanced shock. His deadly white skin shone like ivory. They wouldn't even look like that. One nuclear shell from that gun and they'd be vaporized. Or perhaps the tank had sonic projectors; then the skin would peel off their bones. Or they might be burned, or cut up by shrapnel, or gassed with some new mist their masks couldn't filter. Read shut his eyes. All around him he heard heavy breathing, mumbled comments, curses. Clothes rustled as men moved restlessly. But already the voice of Sergeant Rashid resounded in the murky room. "We've got to knock that thing out before the copter comes. Otherwise, he can't land. I have six Molotov cocktails here. Who wants to go hunting with me?" For two years Read had served under Sergeant Rashid. To him, the sergeant was everything a UN inspector should be. Rashid's devotion to peace had no limits. Read's psych tests said pride alone drove him on. That was good enough for the UN; they only rejected men whose loyalties might conflict with their duties. But an assault on the tank required something more than a hunger for self-respect. Read had seen the inspector who covered their getaway. He had watched their escort charge three-to-one odds. He had seen another inspector stay behind at Miaka Station. And here, in this building, lay battered men and dead men. All UN inspectors. All part of his life. And he was part of their life. Their blood, their sacrifice, and pain, had become a part of him. "I'll take a cocktail, Sarge." "Is that Read?" "Who else did you expect?" "Nobody. Anybody else?" "I'll go," the Frenchman said. "Three should be enough. Give us a good smoke screen." Rashid snapped orders. He put the German inspector in charge of Umluana. Read, the Frenchman and himself, he stationed at thirty-foot intervals along the floor. "Remember," Rashid said. "We have to knock out that gun." Read had given away his machine gun. He held a gas-filled bottle in each hand. His automatic nestled in its shoulder holster. Rashid whistled. Dozens of smoke grenades tumbled through the air. Thick mist engulfed the tank. Read stood up and ran forward. He crouched but didn't zigzag. Speed counted most here. Gunfire shook the hill. The Belderkans couldn't see them but they knew what was going on and they fired systematically into the smoke. Bullets ploughed the ground beside him. He raised his head and found the dim silhouette of the tank. He tried not to think about bullets ploughing through his flesh. A bullet slammed into his hip. He fell on his back, screaming. "Sarge. Sarge. " "I'm hit, too," Rashid said. "Don't stop if you can move." Listen to him. What's he got, a sprained ankle? But he didn't feel any pain. He closed his eyes and threw himself onto his stomach. And nearly fainted from pain. He screamed and quivered. The pain stopped. He stretched out his hands, gripping the wine bottles, and inched forward. Pain stabbed him from stomach to knee. "I can't move, Sarge." "Read, you've got to. I think you're the only—" "What?" Guns clattered. Bullets cracked. "Sergeant Rashid! Answer me." He heard nothing but the lonely passage of the bullets in the mist. "I'm a UN man," he mumbled. "You people up there know what a UN man is? You know what happens when you meet one?" When he reached the tank, he had another bullet in his right arm. But they didn't know he was coming and when you get within ten feet of a tank, the men inside can't see you. He just had to stand up and drop the bottle down the gun barrel. That was all—with a broken hip and a wounded right arm. He knew they would see him when he stood up but he didn't think about that. He didn't think about Sergeant Rashid, about the complicated politics of Africa, about crowded market streets. He had to kill the tank. That was all he thought about. He had decided something in the world was more important than himself, but he didn't know it or realize the psychologists would be surprised to see him do this. He had made many decisions in the last few minutes. He had ceased to think about them or anything else. With his cigarette lighter, he lit the rag stuffed in the end of the bottle. Biting his tongue, he pulled himself up the front of the tank. His long arm stretched for the muzzle of the gun. He tossed the bottle down the dark throat. As he fell, the machine-gun bullets hit him in the chest, then in the neck. He didn't feel them. He had fainted the moment he felt the bottle leave his hand. The copter landed ten minutes later. Umluana left in a shower of bullets. A Russian private, the ranking man alive in the station, surrendered the survivors to the Belderkans. His mother hung the Global Medal above the television set. "He must have been brave," she said. "We had a fine son." "He was our only son," her husband said. "What did he volunteer for? Couldn't somebody else have done it?" His wife started to cry. Awkwardly, he embraced her. He wondered what his son had wanted that he couldn't get at home. THE END
|
C. Harry's parents feel that joining the UN means he is turning his back on America.
|
What is the significance of the conference that Dr. Lessing is invited to?
A. Invitations are the primary source of imposter syndrome for scientists in this field
B. It shows that Dr. Melrose has more control in the field that we realize
C. It offers a chance for Dr. Lessing to get feedback on the parts of his theories he's not certain of
D. It serves as an opportunity for Dr. Lessing to publicize his book
|
BRAMBLE BUSH BY ALAN E. NOURSE [Transcriber's Note: This etext was produced from Worlds of If Science Fiction, August 1957. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] There was a man in our town, and he was wondrous wise; He jumped into a bramble bush and scratched out both his eyes. And when he saw what he had done, with all his might and main He jumped into another bush and scratched them in again. MOTHER GOOSE Dr. David Lessing found Jack Dorffman and the boy waiting in his office when he arrived at the Hoffman Center that morning. Dorffman looked as though he'd been running all night. There were dark pouches under his eyes; his heavy unshaven face seemed to sag at every crease. Lessing glanced sharply at his Field Director and sank down behind his desk with a sigh. "All right, Jack—what's wrong?" "This kid is driving me nuts," said Dorffman through clenched teeth. "He's gone completely hay-wire. Nobody's been able to get near him for three weeks, and now at six o'clock this morning he decides he's leaving the Farm. I talk to him, I sweat him down, I do everything but tie him to the bed, and I waste my time. He's leaving the Farm. Period." "So you bring him down here," said Lessing sourly. "The worst place he could be, if something's really wrong." He looked across at the boy. "Tommy? Come over and sit down." There was nothing singular about the boy's appearance. He was thin, with a pale freckled face and the guileless expression of any normal eight-year-old as he blinked across the desk at Lessing. The awkward grey monitor-helmet concealed a shock of sandy hair. He sat with a mute appeal in his large grey eyes as Lessing flipped the reader-switch and blinked in alarm at the wildly thrashing pattern on the tape. The boy was terrorized. He was literally pulsating with fear. Lessing sat back slowly. "Tell me about it, Tommy," he said gently. "I don't want to go back to the Farm," said the boy. "Why?" "I just don't. I hate it there." "Are you frightened?" The boy bit his lip and nodded slowly. "Of me? Of Dr. Dorffman?" "No. Oh, no!" "Then what?" Again the mute appeal in the boy's eyes. He groped for words, and none came. Finally he said, "If I could only take this off—" He fingered the grey plastic helmet. "You think that would make you feel better?" "It would, I know it would." Lessing shook his head. "I don't think so, Tommy. You know what the monitor is for, don't you?" "It stops things from going out." "That's right. And it stops things from going in. It's an insulator. You need it badly. It would hurt you a great deal if you took it off, away from the Farm." The boy fought back tears. "But I don't want to go back there—" The fear-pattern was alive again on the tape. "I don't feel good there. I never want to go back." "Well, we'll see. You can stay here for a while." Lessing nodded at Dorffman and stepped into an adjoining room with him. "You say this has been going on for three weeks ?" "I'm afraid so. We thought it was just a temporary pattern—we see so much of that up there." "I know, I know." Lessing chewed his lip. "I don't like it. We'd better set up a battery on him and try to spot the trouble. And I'm afraid you'll have to set it up. I've got that young Melrose from Chicago to deal with this morning—the one who's threatening to upset the whole Conference next month with some crazy theories he's been playing with. I'll probably have to take him out to the Farm to shut him up." Lessing ran a hand through sparse grey hair. "See what you can do for the boy downstairs." "Full psi precautions?" asked Dorffman. "Certainly! And Jack—in this case, be sure of it. If Tommy's in the trouble I think he's in, we don't dare risk a chance of Adult Contact now. We could end up with a dead boy on our hands." Two letters were waiting on Lessing's desk that morning. The first was from Roberts Bros., announcing another shift of deadline on the book, and demanding the galley proofs two weeks earlier than scheduled. Lessing groaned. As director of psionic research at the Hoffman Medical Center, he had long since learned how administrative detail could suck up daytime hours. He knew that his real work was at the Farm—yet he hadn't even been to the Farm in over six weeks. And now, as the book approached publication date, Lessing wondered if he would ever really get back to work again. The other letter cheered him a bit more. It bore the letterhead of the International Psionics Conference: Dear Dr. Lessing: In recognition of your position as an authority on human Psionic behavior patterns, we would be gratified to schedule you as principle speaker at the Conference in Chicago on October 12th. A few remarks in discussion of your forthcoming book would be entirely in order— They were waiting for it, then! He ran the galley proofs into the scanner excitedly. They knew he had something up his sleeve. His earlier papers had only hinted at the direction he was going—but the book would clear away the fog. He scanned the title page proudly. "A Theory of Psionic Influence on Infant and Child Development." A good title—concise, commanding, yet modest. They would read it, all right. And they would find it a light shining brightly in the darkness, a guide to the men who were floundering in the jungle of a strange and baffling new science. For they were floundering. When they were finally forced to recognize that this great and powerful force did indeed exist in human minds, with unimaginable potential if it could only be unlocked, they had plunged eagerly into the search, and found themselves in a maddening bramble bush of contradictions and chaos. Nothing worked, and everything worked too well. They were trying to study phenomena which made no sense, observing things that defied logic. Natural laws came crashing down about their ears as they stood sadly by and watched things happen which natural law said could never happen. They had never been in this jungle before, nor in any jungle remotely like it. The old rules didn't work here, the old methods of study failed. And the more they struggled, the thicker and more impenetrable the bramble bush became— But now David Lessing had discovered a pathway through that jungle, a theory to work by— At his elbow the intercom buzzed. "A gentleman to see you," the girl said. "A Dr. Melrose. He's very impatient, sir." He shut off the scanner and said, "Send him in, please." Dr. Peter Melrose was tall and thin, with jet black hair and dark mocking eyes. He wore a threadbare sport coat and a slouch. He offered Lessing a bony hand, then flung himself into a chair as he stared about the office in awe. "I'm really overwhelmed," he said after a moment. "Within the stronghold of psionic research at last. And face to face with the Master in the trembling flesh!" Lessing frowned. "Dr. Melrose, I don't quite understand—" "Oh, it's just that I'm impressed," the young man said airily. "Of course, I've seen old dried-up Authorities before—but never before a brand spanking new one, just fresh out of the pupa, so to speak!" He touched his forehead in a gesture of reverence. "I bow before the Oracle. Speak, oh Motah, live forever! Cast a pearl at my feet!" "If you've come here to be insulting," Lessing said coldly, "you're just wasting time." He reached for the intercom switch. "I think you'd better wait before you do that," Melrose said sharply, "because I'm planning to take you apart at the Conference next month unless I like everything I see and hear down here today. And if you don't think I can do it, you're in for quite a dumping." Lessing sat back slowly. "Tell me—just what, exactly, do you want?" "I want to hear this fairy tale you're about to publish in the name of 'Theory'," Melrose said. "I want to see this famous Farm of yours up in Connecticut and see for myself how much pressure these experimental controls you keep talking about will actually bear. But mostly, I want to see just what in psionic hell you're so busy making yourself an Authority about." There was no laughter in the man's sharp brown eyes. "You couldn't touch me with a ten foot pole at this conference," snapped Lessing. The other man grinned. "Try me! We shook you up a little bit last year, but you didn't seem to get the idea." "Last year was different." Lessing scowled. "As for our 'fairy tale', we happen to have a staggering body of evidence that says that it's true." "If the papers you've already published are a preview, we think it's false as Satan." "And our controls are above suspicion." "So far, we haven't found any way to set up logical controls," said Melrose. "We've done a lot of work on it, too." "Oh, yes—I've heard about your work. Not bad, really. A little misdirected, is all." "According to your Theory, that is." "Wildly unorthodox approach to psionics—but at least you're energetic enough." "We haven't been energetic enough to find an orthodox approach that got us anywhere. We doubt if you have, either. But maybe we're all wrong." Melrose grinned unpleasantly. "We're not unreasonable, your Majesty. We just ask to be shown. If you dare, that is." Lessing slammed his fist down on the desk angrily. "Have you got the day to take a trip?" "I've got 'til New Year." Lessing shouted for his girl. "Get Dorffman up here. We're going to the Farm this afternoon." The girl nodded, then hesitated. "But what about your lunch?" "Bother lunch." He gave Melrose a sidelong glare. "We've got a guest here who's got a lot of words he's going to eat for us...." Ten minutes later they rode the elevator down to the transit levels and boarded the little shuttle car in the terminal below the Hoffman Center. They sat in silence as the car dipped down into the rapid-transit channels beneath the great city, swinging northward in the express circuit through Philadelphia and Camden sectors, surfacing briefly in Trenton sector, then dropping underground once again for the long pull beneath Newark, Manhattan and Westchester sectors. In less than twenty minutes the car surfaced on a Parkway channel and buzzed north and east through the verdant Connecticut countryside. "What about Tommy?" Lessing asked Dorffman as the car sped along through the afternoon sun. "I just finished the prelims. He's not cooperating." Lessing ground his teeth. "I should be running him now instead of beating the bushes with this—" He broke off to glare at young Melrose. Melrose grinned. "I've heard you have quite a place up here." "It's—unconventional, at any rate," Lessing snapped. "Well, that depends on your standards. Sounds like a country day school, from what I've heard. According to your papers, you've even used conventional statistical analysis on your data from up here." "Until we had to throw it out. We discovered that what we were trying to measure didn't make sense in a statistical analysis." "Of course, you're sure you were measuring something ." "Oh, yes. We certainly were." "Yet you said that you didn't know what." "That's right," said Lessing. "We don't." "And you don't know why your instruments measure whatever they're measuring." The Chicago man's face was thoughtful. "In fact, you can't really be certain that your instruments are measuring the children at all. It's not inconceivable that the children might be measuring the instruments , eh?" Lessing blinked. "It's conceivable." "Mmmm," said Melrose. "Sounds like a real firm foundation to build a theory on." "Why not?" Lessing growled. "It wouldn't be the first time the tail wagged the dog. The psychiatrists never would have gotten out of their rut if somebody hadn't gotten smart and realized that one of their new drugs worked better in combatting schizophrenia when the doctor took the medicine instead of the patient. That was quite a wall to climb." "Yes, wasn't it," mused Melrose, scratching his bony jaw. "Only took them seventy years to climb it, thanks to a certain man's theories. I wonder how long it'll take psionics to crawl out of the pit you're digging for it?" "We're not digging any pit," Lessing exploded angrily. "We're exploring—nothing more. A phenomenon exists. We've known that, one way or another, for centuries. The fact that it doesn't seem to be bound by the same sort of natural law we've observed elsewhere doesn't mean that it isn't governed by natural law. But how can we define the law? How can we define the limits of the phenomenon, for that matter? We can't work in the dark forever—we've got to have a working hypothesis to guide us." "So you dreamed up this 'tadpole' idea," said Melrose sourly. "For a working hypothesis—yes. We've known for a long time that every human being has extrasensory potential to one degree or another. Not just a few here and there—every single one. It's a differentiating quality of the human mind. Just as the ability to think logically in a crisis instead of giving way to panic is a differentiating quality." "Fine," said Melrose. "Great. We can't prove that, of course, but I'll play along." Lessing glared at him. "When we began studying this psi-potential, we found out some curious things. For one thing, it seemed to be immensely more powerful and active in infants and children than in adults. Somewhere along the line as a child grows up, something happens. We don't know what. We do know that the child's psi-potential gradually withdraws deeper and deeper into his mind, burying itself farther and farther out of reach, just the way a tadpole's tail is absorbed deeper and deeper into the growing frog until there just isn't any tail any more." Lessing paused, packing tobacco into his pipe. "That's why we have the Farm—to try to discover why. What forces that potential underground? What buries it so deeply that adult human beings can't get at it any more?" "And you think you have an answer," said Melrose. "We think we might be near an answer. We have a theory that explains the available data." The shuttle car bounced sharply as it left the highway automatics. Dorffman took the controls. In a few moments they were skimming through the high white gates of the Farm, slowing down at the entrance to a long, low building. "All right, young man—come along," said Lessing. "I think we can show you our answer." In the main office building they donned the close-fitting psionic monitors required of all personnel at the Farm. They were of a hard grey plastic material, with a network of wiring buried in the substance, connected to a simple pocket-sized power source. "The major problem," Lessing said, "has been to shield the children from any external psionic stimuli, except those we wished to expose them to. Our goal is a perfectly controlled psi environment. The monitors are quite effective—a simple Renwick scrambler screen." "It blocks off all types of psi activity?" asked Melrose. "As far as we can measure, yes." "Which may not be very far." Jack Dorffman burst in: "What Dr. Lessing is saying is that they seem effective for our purposes." "But you don't know why," added Melrose. "All right, we don't know why. Nobody knows why a Renwick screen works—why blame us?" They were walking down the main corridor and out through an open areaway. Behind the buildings was a broad playground. A baseball game was in progress in one corner; across the field a group of swings, slides, ring bars and other playground paraphernalia was in heavy use. The place was teeming with youngsters, all shouting in a fury of busy activity. Occasionally a helmeted supervisor hurried by; one waved to them as she rescued a four-year-old from the parallel bars. They crossed into the next building, where classes were in progress. "Some of our children are here only briefly," Lessing explained as they walked along, "and some have been here for years. We maintain a top-ranking curriculum—your idea of a 'country day school' wasn't so far afield at that—with scholarships supported by Hoffman Center funds. Other children come to us—foundlings, desertees, children from broken homes, children of all ages from infancy on. Sometimes they stay until they have reached college age, or go on to jobs. As far as psionics research is concerned, we are not trying to be teachers. We are strictly observers. We try to place the youngsters in positions where they can develope what potential they have— without the presence of external psionic influences they would normally be subject to. The results have been remarkable." He led them into a long, narrow room with chairs and ash trays, facing a wide grey glass wall. The room fell into darkness, and through the grey glass they could see three children, about four years old, playing in a large room. "They're perfectly insulated from us," said Lessing. "A variety of recording instruments are working. And before you ask, Dr. Melrose, they are all empirical instruments, and they would all defy any engineer's attempts to determine what makes them go. We don't know what makes them go, and we don't care—they go. That's all we need. Like that one, for instance—" In the corner a flat screen was flickering, emitting a pale green fluorescent light. It hung from the wall by two plastic rods which penetrated into the children's room. There was no sign of a switch, nor a power source. As the children moved about, the screen flickered. Below it, a recording-tape clicked along in little spurts and starts of activity. "What are they doing?" Melrose asked after watching the children a few moments. "Those three seem to work as a team, somehow. Each one, individually, had a fairly constant recordable psi potential of about seventeen on the arbitrary scale we find useful here. Any two of them scale in at thirty-four to thirty-six. Put the three together and they operate somewhere in the neighborhood of six hundred on the same scale." Lessing smiled. "This is an isolated phenomenon—it doesn't hold for any other three children on the Farm. Nor did we make any effort to place them together—they drew each other like magnets. One of our workers spent two weeks trying to find out why the instruments weren't right. It wasn't the instruments, of course." Lessing nodded to an attendant, and peered around at Melrose. "Now, I want you to watch this very closely." He opened a door and walked into the room with the children. The fluorescent screen continued to flicker as the children ran to Lessing. He inspected the block tower they were building, and stooped down to talk to them, his lips moving soundlessly behind the observation wall. The children laughed and jabbered, apparently intrigued by the game he was proposing. He walked to the table and tapped the bottom block in the tower with his thumb. The tower quivered, and the screen blazed out with green light, but the tower stood. Carefully Lessing jogged all the foundation blocks out of place until the tower hung in midair, clearly unsupported. The children watched it closely, and the foundation blocks inched still further out of place.... Then, quite casually, Lessing lifted off his monitor. The children continued staring at the tower as the screen gave three or four violent bursts of green fire and went dark. The block tower fell with a crash. Moments later Lessing was back in the observation room, leaving the children busily putting the tower back together. There was a little smile on his lips as he saw Melrose's face. "Perhaps you're beginning to see what I'm driving at," he said slowly. "Yes," said Melrose. "I think I'm beginning to see." He scratched his jaw. "You think that it's adult psi-contact that drives the child's potential underground—that somehow adult contact acts like a damper, a sort of colossal candle-snuffer." "That's what I think," said Lessing. "How do you know those children didn't make you take off your monitor?" Lessing blinked. "Why should they?" "Maybe they enjoy the crash when the blocks fall down." "But that wouldn't make any difference, would it? The blocks still fall down." Melrose paced down the narrow room. "This is very good," he said suddenly, his voice earnest. "You have fine facilities here, good workers. And in spite of my flippancy, Dr. Lessing, I have never imagined for a moment that you were not an acute observer and a careful, highly imaginative worker. But suppose I told you, in perfect faith, that we have data that flatly contradicts everything you've told me today. Reproducible data, utterly incompatable with yours. What would you say to that?" "I'd say you were wrong," said Lessing. "You couldn't have such data. According to the things I am certain are true, what you're saying is sheer nonsense." "And you'd express that opinion in a professional meeting?" "I would." "And as an Authority on psionic behavior patterns," said Melrose slowly, "you would kill us then and there. You would strangle us professionally, discredit anything we did, cut us off cold." The tall man turned on him fiercely. "Are you blind, man? Can't you see what danger you're in? If you publish your book now, you will become an Authority in a field where the most devastating thing that could possibly happen would be— the appearance of an Authority ." Lessing and Dorffman rode back to the Hoffman Center in grim silence. At first Lessing pretended to work; finally he snapped off the tape recorder in disgust and stared out the shuttle-car window. Melrose had gone on to Idlewild to catch a jet back to Chicago. It was a relief to see him go, Lessing thought, and tried to force the thin, angry man firmly out of his mind. But somehow Melrose wouldn't force. "Stop worrying about it," Dorffman urged. "He's a crackpot. He's crawled way out on a limb, and now he's afraid your theory is going to cut it off under him. Well, that's his worry, not yours." Dorffman's face was intense. "Scientifically, you're on unshakeable ground. Every great researcher has people like Melrose sniping at him. You just have to throw them off and keep going." Lessing shook his head. "Maybe. But this field of work is different from any other, Jack. It doesn't follow the rules. Maybe scientific grounds aren't right at all, in this case." Dorffman snorted. "Surely there's nothing wrong with theorizing—" "He wasn't objecting to the theory. He's afraid of what happens after the theory." "So it seems. But why?" "Have you ever considered what makes a man an Authority?" "He knows more about his field than anybody else does." "He seems to, you mean. And therefore, anything he says about it carries more weight than what anybody else says. Other workers follow his lead. He developes ideas, formulates theories—and then defends them for all he's worth ." "But why shouldn't he?" "Because a man can't fight for his life and reputation and still keep his objectivity," said Lessing. "And what if he just happens to be wrong? Once he's an Authority the question of what's right and what's wrong gets lost in the shuffle. It's what he says that counts." "But we know you're right," Dorffman protested. "Do we?" "Of course we do! Look at our work! Look at what we've seen on the Farm." "Yes, I know." Lessing's voice was weary. "But first I think we'd better look at Tommy Gilman, and the quicker we look, the better—" A nurse greeted them as they stepped off the elevator. "We called you at the Farm, but you'd already left. The boy—" She broke off helplessly. "He's sick, Doctor. He's sicker than we ever imagined." "What happened?" "Nothing exactly—happened. I don't quite know how to describe it." She hurried them down the corridor and opened a door into a large children's playroom. "See what you think." The boy sat stolidly in the corner of the room. He looked up as they came in, but there was no flicker of recognition or pleasure on his pale face. The monitor helmet was still on his head. He just sat there, gripping a toy fire engine tightly in his hands. Lessing crossed the room swiftly. "Tommy," he said. The boy didn't even look at him. He stared stupidly at the fire engine. "Tommy!" Lessing reached out for the toy. The boy drew back in terror, clutching it to his chest. "Go away," he choked. "Go away, go away—" When Lessing persisted the boy bent over swiftly and bit him hard on the hand. Lessing sat down on the table. "Tommy, listen to me." His voice was gentle. "I won't try to take it again. I promise." "Go away." "Do you know who I am?" Tommy's eyes shifted haltingly to Lessing's face. He nodded. "Go away." "Why are you afraid, Tommy?" "I hurt. My head hurts. I hurt all over. Go away." "Why do you hurt?" "I—can't get it—off," the boy said. The monitor , Lessing thought suddenly. Something had suddenly gone horribly wrong—could the boy really be sensing the source of the trouble? Lessing felt a cold knot gather in the pit of his stomach. He knew what happened when adult psi-contact struck a psi-high youngster's mind. He had seen it a hundred times at the Farm. But even more—he had felt it in his own mind, bursting from the child. Like a violent physical blow, the hate and fear and suspicion and cruelty buried and repressed in the adult mind, crushing suddenly into the raw receptors of the child's mind like a smothering fog—it was a fearful thing. A healthy youngster could survive it, even though the scar remained. But this youngster was sick— And yet an animal instinctively seeks its own protection . With trembling fingers Lessing reached out and opened the baffle-snap on the monitor. "Take it off, Tommy," he whispered. The boy blinked in amazement, and pulled the grey helmet from his head. Lessing felt the familiar prickly feeling run down his scalp as the boy stared at him. He could feel deep in his own mind the cold chill of terror radiating from the boy. Then, suddenly, it began to fade. A sense of warmth—peace and security and comfort—swept in as the fear faded from the boy's face. The fire engine clattered to the floor. They analyzed the tapes later, punching the data cards with greatest care, filing them through the machines for the basic processing and classification that all their data underwent. It was late that night when they had the report back in their hands. Dorffman stared at it angrily. "It's obviously wrong," he grated. "It doesn't fit. Dave, it doesn't agree with anything we've observed before. There must be an error." "Of course," said Lessing. "According to the theory. The theory says that adult psi-contact is deadly to the growing child. It smothers their potential through repeated contact until it dries up completely. We've proved that, haven't we? Time after time. Everything goes according to the theory—except Tommy. But Tommy's psi-potential was drying up there on the Farm, until the distortion was threatening the balance of his mind. Then he made an adult contact, and we saw how he bloomed." Lessing sank down to his desk wearily. "What are we going to do, Jack? Formulate a separate theory for Tommy?" "Of course not," said Dorffman. "The instruments were wrong. Somehow we misread the data—" "Didn't you see his face ?" Lessing burst out. "Didn't you see how he acted ? What do you want with an instrument reading?" He shook his head. "It's no good, Jack. Something different happened here, something we'd never counted on. It's something the theory just doesn't allow for." They sat silently for a while. Then Dorffman said: "What are you going to do?" "I don't know," said Lessing. "Maybe when we fell into this bramble bush we blinded ourselves with the urge to classify—to line everything up in neat rows like pins in a paper. Maybe we were so blind we missed the path altogether." "But the book is due! The Conference speech—" "I think we'll make some changes in the book," Lessing said slowly. "It'll be costly—but it might even be fun. It's a pretty dry, logical presentation of ideas, as it stands. Very austere and authoritarian. But a few revisions could change all that—" He rubbed his hands together thoughtfully. "How about it, Jack? Do we have nerve enough to be laughed at? Do you think we could stand a little discredit, making silly asses of ourselves? Because when I finish this book, we'll be laughed out of existence. There won't be any Authority in psionics for a while—and maybe that way one of the lads who's really sniffing out the trail will get somebody to listen to him! "Get a pad, get a pencil! We've got work to do. And when we finish, I think we'll send a carbon copy out Chicago way. Might even persuade that puppy out there to come here and work for me—"
|
D. It serves as an opportunity for Dr. Lessing to publicize his book
|
Over which datasets/corpora is this work evaluated?
|
### Introduction
Currently, voice-controlled smart devices are widely used in multiple areas to fulfill various tasks, e.g. playing music, acquiring weather information and booking tickets. The SLU system employs several modules to enable the understanding of the semantics of the input speeches. When there is an incoming speech, the ASR module picks it up and attempts to transcribe the speech. An ASR model could generate multiple interpretations for most speeches, which can be ranked by their associated confidence scores. Among the $n$-best hypotheses, the top-1 hypothesis is usually transformed to the NLU module for downstream tasks such as domain classification, intent classification and named entity recognition (slot tagging). Multi-domain NLU modules are usually designed hierarchically BIBREF0. For one incoming utterance, NLU modules will firstly classify the utterance as one of many possible domains and the further analysis on intent classification and slot tagging will be domain-specific. In spite of impressive development on the current SLU pipeline, the interpretation of speech could still contain errors. Sometimes the top-1 recognition hypothesis of ASR module is ungrammatical or implausible and far from the ground-truth transcription BIBREF1, BIBREF2. Among those cases, we find one interpretation exact matching with or more similar to transcription can be included in the remaining hypotheses ($2^{nd}- n^{th}$). To illustrate the value of the $2^{nd}- n^{th}$ hypotheses, we count the frequency of exact matching and more similar (smaller edit distance compared to the 1st hypothesis) to transcription for different positions of the $n$-best hypotheses list. Table TABREF1 exhibits the results. For the explored dataset, we only collect the top 5 interpretations for each utterance ($n = 5$). Notably, when the correct recognition exists among the 5 best hypotheses, 50% of the time (sum of the first row's percentages) it occurs among the $2^{nd}-5^{th}$ positions. Moreover, as shown by the second row in Table TABREF1, compared to the top recognition hypothesis, the other hypotheses can sometimes be more similar to the transcription. Over the past few years, we have observed the success of reranking the $n$-best hypotheses BIBREF1, BIBREF3, BIBREF4, BIBREF5, BIBREF6, BIBREF7, BIBREF8, BIBREF9, BIBREF10 before feeding the best interpretation to the NLU module. These approaches propose the reranking framework by involving morphological, lexical or syntactic features BIBREF8, BIBREF9, BIBREF10, speech recognition features like confidence score BIBREF1, BIBREF4, and other features like number of tokens, rank position BIBREF1. They are effective to select the best from the hypotheses list and reduce the word error rate (WER) BIBREF11 of speech recognition. Those reranking models could benefit the first two cases in Table TABREF2 when there is an utterance matching with transcription. However, in other cases like the third row, it is hard to integrate the fragmented information in multiple hypotheses. This paper proposes various methods integrating $n$-best hypotheses to tackle the problem. To the best of our knowledge, this is the first study that attempts to collectively exploit the $n$-best speech interpretations in the SLU system. This paper serves as the basis of our $n$-best-hypotheses-based SLU system, focusing on the methods of integration for the hypotheses. Since further improvements of the integration framework require considerable setup and descriptions, where jointly optimized tasks (e.g. transcription reconstruction) trained with multiple ways (multitask BIBREF12, multistage learning BIBREF13) and more features (confidence score, rank position, etc.) are involved, we leave those to a subsequent article. This paper is organized as follows. Section SECREF2 introduces the Baseline, Oracle and Direct models. Section SECREF3 describes proposed ways to integrate $n$-best hypotheses during training. The experimental setup and results are described in Section SECREF4. Section SECREF5 contains conclusions and future work. ### Baseline, Oracle and Direct Models ::: Baseline and Oracle
The preliminary architecture is shown in Fig. FIGREF4. For a given transcribed utterance, it is firstly encoded with Byte Pair Encoding (BPE) BIBREF14, a compression algorithm splitting words to fundamental subword units (pairs of bytes or BPs) and reducing the embedded vocabulary size. Then we use a BiLSTM BIBREF15 encoder and the output state of the BiLSTM is regarded as a vector representation for this utterance. Finally, a fully connected Feed-forward Neural Network (FNN) followed by a softmax layer, labeled as a multilayer perceptron (MLP) module, is used to perform the domain/intent classification task based on the vector. For convenience, we simplify the whole process in Fig.FIGREF4 as a mapping $BM$ (Baseline Mapping) from the input utterance $S$ to an estimated tag's probability $p(\tilde{t})$, where $p(\tilde{t}) \leftarrow BM(S)$. The $Baseline$ is trained on transcription and evaluated on ASR 1st best hypothesis ($S=\text{ASR}\ 1^{st}\ \text{best})$. The $Oracle$ is trained on transcription and evaluated on transcription ($S = \text{Transcription}$). We name it Oracle simply because we assume that hypotheses are noisy versions of transcription. ### Baseline, Oracle and Direct Models ::: Direct Models
Besides the Baseline and Oracle, where only ASR 1-best hypothesis is considered, we also perform experiments to utilize ASR $n$-best hypotheses during evaluation. The models evaluating with $n$-bests and a BM (pre-trained on transcription) are called Direct Models (in Fig. FIGREF7): Majority Vote. We apply the BM model on each hypothesis independently and combine the predictions by picking the majority predicted label, i.e. Music. Sort by Score. After parallel evaluation on all hypotheses, sort the prediction by the corresponding confidence score and choose the one with the highest score, i.e. Video. Rerank (Oracle). Since the current rerank models (e.g., BIBREF1, BIBREF3, BIBREF4) attempt to select the hypothesis most similar to transcription, we propose the Rerank (Oracle), which picks the hypothesis with the smallest edit distance to transcription (assume it is the $a$-th best) during evaluation and uses its corresponding prediction. ### Integration of N-BEST Hypotheses
All the above mentioned models apply the BM trained on one interpretation (transcription). Their abilities to take advantage of multiple interpretations are actually not trained. As a further step, we propose multiple ways to integrate the $n$-best hypotheses during training. The explored methods can be divided into two groups as shown in Fig. FIGREF11. Let $H_1, H_2,..., H_n $ denote all the hypotheses from ASR and $bp_{H_k, i} \in BPs$ denotes the $i$-th pair of bytes (BP) in the $k^{th}$ best hypothesis. The model parameters associated with the two possible ways both contain: embedding $e_{bp}$ for pairs of bytes, BiLSTM parameters $\theta $ and MLP parameters $W, b$. ### Integration of N-BEST Hypotheses ::: Hypothesized Text Concatenation
The basic integration method (Combined Sentence) concatenates the $n$-best hypothesized text. We separate hypotheses with a special delimiter ($<$SEP$>$). We assume BPE totally produces $m$ BPs (delimiters are not split during encoding). Suppose the $n^{th}$ hypothesis has $j$ pairs. The entire model can be formulated as: In Eqn. DISPLAY_FORM13, the connected hypotheses and separators are encoded via BiLSTM to a sequence of hidden state vectors. Each hidden state vector, e.g. $h_1$, is the concatenation of forward $h_{1f}$ and backward $h_{1b}$ states. The concatenation of the last state of the forward and backward LSTM forms the output vector of BiLSTM (concatenation denoted as $[,]$). Then, in Eqn. DISPLAY_FORM14, the MLP module defines the probability of a specific tag (domain or intent) $\tilde{t}$ as the normalized activation ($\sigma $) output after linear transformation of the output vector. ### Integration of N-BEST Hypotheses ::: Hypothesis Embedding Concatenation
The concatenation of hypothesized text leverages the $n$-best list by transferring information among hypotheses in an embedding framework, BiLSTM. However, since all the layers have access to both the preceding and subsequent information, the embedding among $n$-bests will influence each other, which confuses the embedding and makes the whole framework sensitive to the noise in hypotheses. As the second group of integration approaches, we develop models, PoolingAvg/Max, on the concatenation of hypothesis embedding, which isolate the embedding process among hypotheses and summarize the features by a pooling layer. For each hypothesis (e.g., $i^{th}$ best in Eqn. DISPLAY_FORM16 with $j$ pairs of bytes), we could get a sequence of hidden states from BiLSTM and obtain its final output state by concatenating the first and last hidden state ($h_{output_i}$ in Eqn. DISPLAY_FORM17). Then, we stack all the output states vertically as shown in Eqn. SECREF15. Note that in the real data, we will not always have a fixed size of hypotheses list. For a list with $r$ ($<n$) interpretations, we get the embedding for each of them and pad with the embedding of the first best hypothesis until a fixed size $n$. When $r\ge n$, we only stack the top $n$ embeddings. We employ $h_{output_1}$ for padding to enhance the influence of the top 1 hypothesis, which is more reliable. Finally, one unified representation could be achieved via Pooling (Max/Avg pooling with $n$ by 1 sliding window and stride 1) on the concatenation and one score could be produced per possible tag for the given task. ### Experiment ::: Dataset
We conduct our experiments on $\sim $ 8.7M annotated anonymised user utterances. They are annotated and derived from requests across 23 domains. ### Experiment ::: Performance on Entire Test Set
Table TABREF24 shows the relative error reduction (RErr) of Baseline, Oracle and our proposed models on the entire test set ($\sim $ 300K utterances) for multi-class domain classification. We can see among all the direct methods, predicting based on the hypothesis most similar to the transcription (Rerank (Oracle)) is the best. As for the other models attempting to integrate the $n$-bests during training, PoolingAvg gets the highest relative improvement, 14.29%. It as well turns out that all the integration methods outperform direct models drastically. This shows that having access to $n$-best hypotheses during training is crucial for the quality of the predicted semantics. ### Experiment ::: Performance Comparison among Various Subsets
To further detect the reason for improvements, we split the test set into two parts based on whether ASR first best agrees with transcription and evaluate separately. Comparing Table TABREF26 and Table TABREF27, obviously the benefits of using multiple hypotheses are mainly gained when ASR 1st best disagrees with the transcription. When ASR 1st best agrees with transcription, the proposed integration models can also keep the performance. Under that condition, we can still improve a little (3.56%) because, by introducing multiple ASR hypotheses, we could have more information and when the transcription/ASR 1st best does not appear in the training set's transcriptions, its $n$-bests list may have similar hypotheses included in the training set's $n$-bests. Then, our integration model trained on $n$-best hypotheses as well has clue to predict. The series of comparisons reveal that our approaches integrating the hypotheses are robust to the ASR errors and whenever the ASR model makes mistakes, we can outperform more significantly. ### Experiment ::: Improvements on Different Domains and Different Numbers of Hypotheses
Among all the 23 domains, we choose 8 popular domains for further comparisons between the Baseline and the best model of Table TABREF24, PoolingAvg. Fig. FIGREF29 exhibits the results. We could find the PoolingAvg consistently improves the accuracy for all 8 domains. In the previous experiments, the number of utilized hypotheses for each utterance during evaluation is five, which means we use the top 5 interpretations when the size of ASR recognition list is not smaller than 5 and use all the interpretations otherwise. Changing the number of hypotheses while evaluation, Fig. FIGREF30 shows a monotonic increase with the access to more hypotheses for the PoolingAvg and PoolingMax (Sort by Score is shown because it is the best achievable direct model while the Rerank (Oracle) is not realistic). The growth becomes gentle after four hypotheses are leveraged. ### Experiment ::: Intent Classification
Since another downstream task, intent classification, is similar to domain classification, we just show the best model in domain classification, PoolingAvg, on domain-specific intent classification for three popular domains due to space limit. As Table TABREF32 shows, the margins of using multiple hypotheses with PoolingAvg are significant as well. ### Conclusions and Future Work
This paper improves the SLU system robustness to ASR errors by integrating $n$-best hypotheses in different ways, e.g. the aggregation of predictions from hypotheses or the concatenation of hypothesis text or embedding. We can achieve significant classification accuracy improvements over production-quality baselines on domain and intent classifications, 14% to 25% relative gains. The improvement is more significant for a subset of testing data where ASR first best is different from transcription. We also observe that with more hypotheses utilized, the performance can be further improved. In the future, we aim to employ additional features (e.g. confidence scores for hypotheses or tokens) to integrate $n$-bests more efficiently, where we can train a function $f$ to obtain a weight for each hypothesis embedding before pooling. Another direction is using deep learning framework to embed the word lattice BIBREF16 or confusion network BIBREF17, BIBREF18, which can provide a compact representation of multiple hypotheses and more information like times, in the SLU system. ### Acknowledgements
We would like to thank Junghoo (John) Cho for proofreading. Fig. 3: Integration of n-best hypotheses with two possible ways: 1) concatenate hypothesized text and 2) concatenate hypothesis embedding. Table 3: Micro and Macro F1 score for multi-class domain classification. Table 4: Performance comparison for the subset (∼ 19%) where ASR first best disagrees with transcription. Table 5: Performance comparison for the subset (∼ 81%) where ASR first best agrees with transcription. Fig. 5: The influence of different amount of hypotheses. Table 6: Intent classification for three important domains. Fig. 4: Improvements on important domains.
|
on $\sim $ 8.7M annotated anonymised user utterances
|
What object is found by the guards, giving away the six Venusian conspirators? Who does it belong to?
A. A spy ray. It belongs to the six insurgents who plan to blow up the Earth ship.
B. A Venus-tobacco cigarette. It belongs to the Exec officer, who the six insurgents killed when breaking into the Earth ship.
C. An atomite bomb. It belongs to the guard they killed just before breaking into where the Earth ship is kept.
D. A rifle. It belongs to the guard they killed just before breaking into where the Earth ship is kept.
|
DOUBLECROSS by JAMES Mac CREIGH Revolt was brewing on Venus, led by the descendant of the first Earthmen to land. Svan was the leader making the final plans—plotting them a bit too well. [Transcriber's Note: This etext was produced from Planet Stories Winter 1944. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The Officer of the Deck was pleased as he returned to the main lock. There was no reason why everything shouldn't have been functioning perfectly, of course, but he was pleased to have it confirmed, all the same. The Executive Officer was moodily smoking a cigarette in the open lock, staring out over the dank Venusian terrain at the native town. He turned. "Everything shipshape, I take it!" he commented. The OD nodded. "I'll have a blank log if this keeps up," he said. "Every man accounted for except the delegation, cargo stowed, drivers ready to lift as soon as they come back." The Exec tossed away his cigarette. " If they come back." "Is there any question?" The Exec shrugged. "I don't know, Lowry," he said. "This is a funny place. I don't trust the natives." Lowry lifted his eyebrows. "Oh? But after all, they're human beings, just like us—" "Not any more. Four or five generations ago they were. Lord, they don't even look human any more. Those white, flabby skins—I don't like them." "Acclimation," Lowry said scientifically. "They had to acclimate themselves to Venus's climate. They're friendly enough." The Exec shrugged again. He stared at the wooden shacks that were the outskirts of the native city, dimly visible through the ever-present Venusian mist. The native guard of honor, posted a hundred yards from the Earth-ship, stood stolidly at attention with their old-fashioned proton-rifles slung over their backs. A few natives were gazing wonderingly at the great ship, but made no move to pass the line of guards. "Of course," Lowry said suddenly, "there's a minority who are afraid of us. I was in town yesterday, and I talked with some of the natives. They think there will be hordes of immigrants from Earth, now that we know Venus is habitable. And there's some sort of a paltry underground group that is spreading the word that the immigrants will drive the native Venusians—the descendants of the first expedition, that is—right down into the mud. Well—" he laughed—"maybe they will. After all, the fittest survive. That's a basic law of—" The annunciator over the open lock clanged vigorously, and a metallic voice rasped: "Officer of the Deck! Post Number One! Instruments reports a spy ray focused on the main lock!" Lowry, interrupted in the middle of a word, jerked his head back and stared unbelievingly at the tell-tale next to the annunciator. Sure enough, it was glowing red—might have been glowing for minutes. He snatched at the hand-phone dangling from the wall, shouted into it. "Set up a screen! Notify the delegation! Alert a landing party!" But even while he was giving orders, the warning light flickered suddenly and went out. Stricken, Lowry turned to the Exec. The Executive Officer nodded gloomily. He said, "You see!" "You see?" Svan clicked off the listening-machine and turned around. The five others in the room looked apprehensive. "You see?" Svan repeated. "From their own mouths you have heard it. The Council was right." The younger of the two women sighed. She might have been beautiful, in spite of her dead-white skin, if there had been a scrap of hair on her head. "Svan, I'm afraid," she said. "Who are we to decide if this is a good thing? Our parents came from Earth. Perhaps there will be trouble at first, if colonists come, but we are of the same blood." Svan laughed harshly. " They don't think so. You heard them. We are not human any more. The officer said it." The other woman spoke unexpectedly. "The Council was right," she agreed. "Svan, what must we do?" Svan raised his hand, thoughtfully. "One moment. Ingra, do you still object?" The younger woman shrank back before the glare in his eyes. She looked around at the others, found them reluctant and uneasy, but visibly convinced by Svan. "No," she said slowly. "I do not object." "And the rest of us? Does any of us object?" Svan eyed them, each in turn. There was a slow but unanimous gesture of assent. "Good," said Svan. "Then we must act. The Council has told us that we alone will decide our course of action. We have agreed that, if the Earth-ship returns, it means disaster for Venus. Therefore, it must not return." An old man shifted restlessly. "But they are strong, Svan," he complained. "They have weapons. We cannot force them to stay." Svan nodded. "No. They will leave. But they will never get back to Earth." "Never get back to Earth?" the old man gasped. "Has the Council authorized—murder?" Svan shrugged. "The Council did not know what we would face. The Councilmen could not come to the city and see what strength the Earth-ship has." He paused dangerously. "Toller," he said, "do you object?" Like the girl, the old man retreated before his eyes. His voice was dull. "What is your plan?" he asked. Svan smiled, and it was like a dark flame. He reached to a box at his feet, held up a shiny metal globe. "One of us will plant this in the ship. It will be set by means of this dial—" he touched a spot on the surface of the globe with a pallid finger—"to do nothing for forty hours. Then—it will explode. Atomite." He grinned triumphantly, looking from face to face. The grin faded uncertainly as he saw what was in their eyes—uncertainty, irresolution. Abruptly he set the bomb down, savagely ripped six leaves off a writing tablet on the table next him. He took a pencil and made a mark on one of them, held it up. "We will let chance decide who is to do the work," he said angrily. "Is there anyone here who is afraid? There will be danger, I think...." No answer. Svan jerked his head. "Good," he said. "Ingra, bring me that bowl." Silently the girl picked up an opaque glass bowl from the broad arm of her chair. It had held Venus-tobacco cigarettes; there were a few left. She shook them out and handed the bowl to Svan, who was rapidly creasing the six fatal slips. He dropped them in the bowl, stirred it with his hand, offered it to the girl. "You first, Ingra," he said. She reached in mechanically, her eyes intent on his, took out a slip and held it without opening it. The bowl went the rounds, till Svan himself took the last. All eyes were on him. No one had looked at their slips. Svan, too, had left his unopened. He sat at the table, facing them. "This is the plan," he said. "We will go, all six of us, in my ground car, to look at the Earth-ship. No one will suspect—the whole city has been to see it already. One will get out, at the best point we can find. It is almost dusk now. He can hide, surely, in the vegetation. The other five will start back. Something will go wrong with the car—perhaps it will run off the road, start to sink in the swamp. The guards will be called. There will be commotion—that is easy enough, after all; a hysterical woman, a few screams, that's all there is to it. And the sixth person will have his chance to steal to the side of the ship. The bomb is magnetic. It will not be noticed in the dark—they will take off before sunrise, because they must travel away from the sun to return—in forty hours the danger is removed." There was comprehension in their eyes, Svan saw ... but still that uncertainty. Impatiently, he crackled: "Look at the slips!" Though he had willed his eyes away from it, his fingers had rebelled. Instinctively they had opened the slip, turned it over and over, striving to detect if it was the fatal one. They had felt nothing.... And his eyes saw nothing. The slip was blank. He gave it but a second's glance, then looked up to see who had won the lethal game of chance. Almost he was disappointed. Each of the others had looked in that same second. And each was looking up now, around at his neighbors. Svan waited impatiently for the chosen one to announce it—a second, ten seconds.... Then gray understanding came to him. A traitor! his subconscious whispered. A coward! He stared at them in a new light, saw their indecision magnified, became opposition. Svan thought faster than ever before in his life. If there was a coward, it would do no good to unmask him. All were wavering, any might be the one who had drawn the fatal slip. He could insist on inspecting every one, but—suppose the coward, cornered, fought back? In fractions of a second, Svan had considered the evidence and reached his decision. Masked by the table, his hand, still holding the pencil, moved swiftly beneath the table, marked his own slip. In the palm of his hand, Svan held up the slip he had just marked in secret. His voice was very tired as he said, "I will plant the bomb." The six conspirators in Svan's old ground car moved slowly along the main street of the native town. Two Earth-ship sailors, unarmed except for deceptively flimsy-looking pistols at their hips, stood before the entrance to the town's Hall of Justice. "Good," said Svan, observing them. "The delegation is still here. We have ample time." He half turned in the broad front seat next to the driver, searching the faces of the others in the car. Which was the coward? he wondered. Ingra? Her aunt? One of the men? The right answer leaped up at him. They all are , he thought. Not one of them understands what this means. They're afraid. He clamped his lips. "Go faster, Ingra," he ordered the girl who was driving. "Let's get this done with." She looked at him, and he was surprised to find compassion in her eyes. Silently she nodded, advanced the fuel-handle so that the clumsy car jolted a trace more rapidly over the corduroy road. It was quite dark now. The car's driving light flared yellowishly in front of them, illuminating the narrow road and the pale, distorted vegetation of the jungle that surrounded them. Svan noticed it was raining a little. The present shower would deepen and intensify until midnight, then fall off again, to halt before morning. But before then they would be done. A proton-bolt lanced across the road in front of them. In the silence that followed its thunderous crash, a man's voice bellowed: "Halt!" The girl, Ingra, gasped something indistinguishable, slammed on the brakes. A Venusian in the trappings of the State Guard advanced on them from the side of the road, proton-rifle held ready to fire again. "Where are you going?" he growled. Svan spoke up. "We want to look at the Earth-ship," he said. He opened the door beside him and stepped out, careless of the drizzle. "We heard it was leaving tonight," he continued, "and we have not seen it. Is that not permitted?" The guard shook his head sourly. "No one is allowed near the ship. The order was just issued. It is thought there is danger." Svan stepped closer, his teeth bared in what passed for a smile. "It is urgent," he purred. His right hand flashed across his chest in a complicated gesture. "Do you understand?" Confusion furrowed the guard's hairless brows, then was replaced by a sudden flare of understanding—and fear. "The Council!" he roared. "By heaven, yes, I understand! You are the swine that caused this—" He strove instinctively to bring the clumsy rifle up, but Svan was faster. His gamble had failed; there was only one course remaining. He hurled his gross white bulk at the guard, bowled him over against the splintery logs of the road. The proton-rifle went flying, and Svan savagely tore at the throat of the guard. Knees, elbows and claw-like nails—Svan battered at the astonished man with every ounce of strength in his body. The guard was as big as Svan, but Svan had the initial advantage ... and it was only a matter of seconds before the guard lay unconscious, his skull a mass of gore at the back where Svan had ruthlessly pounded it against the road. Svan grunted as his fingers constricted brutally. Svan rose, panting, stared around. No one else was in sight, save the petrified five and the ground car. Svan glared at them contemptuously, then reached down and heaved on the senseless body of the guard. Over the shoulder of the road the body went, onto the damp swampland of the jungle. Even while Svan watched the body began to sink. There would be no trace. Svan strode back to the car. "Hurry up," he gasped to the girl. "Now there is danger for all of us, if they discover he is missing. And keep a watch for other guards." Venus has no moon, and no star can shine through its vast cloud layer. Ensign Lowry, staring anxiously out through the astro-dome in the bow of the Earth-ship, cursed the blackness. "Can't see a thing," he complained to the Exec, steadily writing away at the computer's table. "Look—are those lights over there?" The Exec looked up wearily. He shrugged. "Probably the guards. Of course, you can't tell. Might be a raiding party." Lowry, stung, looked to see if the Exec was smiling, but found no answer in his stolid face. "Don't joke about it," he said. "Suppose something happens to the delegation?" "Then we're in the soup," the Exec said philosophically. "I told you the natives were dangerous. Spy-rays! They've been prohibited for the last three hundred years." "It isn't all the natives," Lowry said. "Look how they've doubled the guard around us. The administration is co-operating every way they know how. You heard the delegation's report on the intercom. It's this secret group they call the Council." "And how do you know the guards themselves don't belong to it?" the Exec retorted. "They're all the same to me.... Look, your light's gone out now. Must have been the guard. They're on the wrong side to be coming from the town, anyhow...." Svan hesitated only a fraction of a second after the girl turned the lights out and stopped the car. Then he reached in the compartment under the seat. If he took a little longer than seemed necessary to get the atomite bomb out of the compartment, none of the others noticed. Certainly it did not occur to them that there had been two bombs in the compartment, though Svan's hand emerged with only one. He got out of the car, holding the sphere. "This will do for me," he said. "They won't be expecting anyone to come from behind the ship—we were wise to circle around. Now, you know what you must do?" Ingra nodded, while the others remained mute. "We must circle back again," she parroted. "We are to wait five minutes, then drive the car into the swamp. We will create a commotion, attract the guards." Svan, listening, thought: It's not much of a plan. The guards would not be drawn away. I am glad I can't trust these five any more. If they must be destroyed, it is good that their destruction will serve a purpose. Aloud, he said, "You understand. If I get through, I will return to the city on foot. No one will suspect anything if I am not caught, because the bomb will not explode until the ship is far out in space. Remember, you are in no danger from the guards." From the guards , his mind echoed. He smiled. At least, they would feel no pain, never know what happened. With the amount of atomite in that bomb in the compartment, they would merely be obliterated in a ground-shaking crash. Abruptly he swallowed, reminded of the bomb that was silently counting off the seconds. "Go ahead," he ordered. "I will wait here." "Svan." The girl, Ingra, leaned over to him. Impulsively she reached for him, kissed him. "Good luck to you, Svan," she said. "Good luck," repeated the others. Then silently the electric motor of the car took hold. Skilfully the girl backed it up, turned it around, sent it lumbering back down the road. Only after she had traveled a few hundred feet by the feel of the road did she turn the lights on again. Svan looked after them. The kiss had surprised him. What did it mean? Was it an error that the girl should die with the others? There was an instant of doubt in his steel-shackled mind, then it was driven away. Perhaps she was loyal, yet certainly she was weak. And since he could not know which was the one who had received the marked slip, and feared to admit it, it was better they all should die. He advanced along the midnight road to where the ground rose and the jungle plants thinned out. Ahead, on an elevation, were the rain-dimmed lights of the Earth-ship, set down in the center of a clearing made by its own fierce rockets. Svan's mist-trained eyes spotted the circling figures of sentries, and knew that these would be the ship's own. They would not be as easily overcome as the natives, not with those slim-shafted blasters they carried. Only deceit could get him to the side of the ship. Svan settled himself at the side of the road, waiting for his chance. He had perhaps three minutes to wait; he reckoned. His fingers went absently to the pouch in his wide belt, closed on the slip of paper. He turned it over without looking at it, wondering who had drawn the first cross, and been a coward. Ingra? One of the men? He became abruptly conscious of a commotion behind him. A ground car was racing along the road. He spun around and was caught in the glare of its blinding driving-light, as it bumped to a slithering stop. Paralyzed, he heard the girl's voice. "Svan! They're coming! They found the guard's rifle, and they're looking for us! Thirty Earthmen, Svan, with those frightful guns. They fired at us, but we got away and came for you. We must flee!" He stared unseeingly at the light. "Go away!" he croaked unbelievingly. Then his muscles jerked into action. The time was almost up—the bomb in the car— "Go away!" he shrieked, and turned to run. His fists clenched and swinging at his side, he made a dozen floundering steps before something immense pounded at him from behind. He felt himself lifted from the road, sailing, swooping, dropping with annihilating force onto the hard, charred earth of the clearing. Only then did he hear the sound of the explosion, and as the immense echoes died away he began to feel the pain seeping into him from his hideously racked body.... The Flight Surgeon rose from beside him. "He's still alive," he said callously to Lowry, who had just come up. "It won't last long, though. What've you got there?" Lowry, a bewildered expression on his beardless face, held out the two halves of a metallic sphere. Dangling ends of wires showed where a connection had been broken. "He had a bomb," he said. "A magnetic-type, delayed-action atomite bomb. There must have been another in the car, and it went off. They—they were planning to bomb us." "Amazing," the surgeon said dryly. "Well, they won't do any bombing now." Lowry was staring at the huddled, mutilated form of Svan. He shuddered. The surgeon, seeing the shudder, grasped his shoulder. "Better them than us," he said. "It's poetic justice if I ever saw it. They had it coming...." He paused thoughtfully, staring at a piece of paper between his fingers. "This is the only part I don't get," he said. "What's that?" Lowry craned his neck. "A piece of paper with a cross on it? What about it?" The surgeon shrugged. "He had it clenched in his hand," he said. "Had the devil of a time getting it loose from him." He turned it over slowly, displayed the other side. "Now what in the world would he be doing carrying a scrap of paper with a cross marked on both sides?"
|
D. A rifle. It belongs to the guard they killed just before breaking into where the Earth ship is kept.
|
What does Peet seem to care about the most?
A. keeping all of his power and money
B. the safety of all citizens on Mercury
C. getting off of Mercury
D. the people that work for him
|
Red Witch of Mercury By EMMETT McDOWELL Death was Jaro Moynahan's stock in trade, and every planet had known his touch. But now, on Mercury, he was selling his guns into the weirdest of all his exploits—gambling his life against the soft touch of a woman's lips. [Transcriber's Note: This etext was produced from Planet Stories Summer 1945. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] On the stage of Mercury Sam's Garden , a tight-frocked, limber-hipped, red-head was singing " The Lady from Mars ." The song was a rollicking, ribald ditty, a favorite of the planters and miners, the space pilots and army officers who frequented the garden. The girl rendered it with such gusto that the audience burst into a roar of applause. She bent her head in acknowledgment so that her bronze red hair fell down about her face. There was perspiration on her upper lip and temples. Her crimson mouth wore a fixed smile. Her eyes were frightened. The man, who had accompanied the singer on the piano, sat at the foot of the stage, his back to the crowded tables. He did not look up at the singer but kept his pale, immature face bent over the keys, while his fingers lightly, automatically picked out the tune. Sweat trickled down the back of his neck, plastered his white coat to his back. Without looking up, he said: "Have you spotted him?" His voice was pitched to reach the singer alone. The girl, with an almost imperceptible gesture, shook her head. The night was very hot; but then it is always hot on Mercury, the newest, the wildest, the hottest of Earth's frontiers. Fans spaced about the garden's walls sluggishly stirred the night air, while the men and women sitting at the tables drank heavily of Latonka, the pale green wine of Mercury. Only the native waiters, the enigmatic, yellow-eyed Mercurians, seemed unaffected by the heat. They didn't sweat at all. Up on the stage the singer was about to begin another number when she stiffened. "Here he is," she said to the pianist without moving her lips. The pianist swung around on his stool, lifted his black eyes to the gate leading to the street. Just within the entrance, a tall, thin man was standing. He looked like a gaunt gray wolf loitering in the doorway. His white duraloes suit hung faultlessly. His black hair was close-cropped, his nose thin and aquiline. For a moment he studied the crowded garden before making his way to a vacant table. "Go on," said the pianist in a flat voice. The red-head shivered. Stepping from the stage she picked her way through the tables until she came to the one occupied by the newcomer. "May I join you?" she asked in a low voice. The man arose. "Of course. I was expecting you. Here, sit down." He pulled out a chair, motioned for the waiter. The Mercurian, his yellow incurious eyes like two round topazes, sidled up. "Bring us a bottle of Latonka from the Veederman region, well iced." The waiter slipped away. "So," said the red-head; "you have come. I did not think you would be in time." Her hands were clenched in her lap. The knuckles were white. The man said nothing. "I did not want to call you in, Jaro Moynahan." It was the first time she had used his name. "You have the reputation of being unpredictable. I don't trust you, but since...." She stopped as the waiter placed glasses on the table and deftly poured the pale green wine. The man, Jaro Moynahan, raised his glass. "Here's to the revolution," he said. His low voice carried an odd, compelling note. His eyes, light blue and amused, were pale against his brown face. The girl drew in her breath. "No! Mercury is not ready for freedom. Only a handful of fanatics are engineering the revolution. The real Mercurian patriots are against it, but they are afraid to protest. You've got to believe me. The revolution is scheduled to break during the Festival of the Rains. If it does, the Terrestrials here will be massacred. The Mercurians hate them. We haven't but a handful of troops." Jaro Moynahan wiped the sweat from his forehead with a fine duraweb handkerchief. "I had forgotten how abominably hot it can be here." The girl ignored the interruption. "There is one man; he is the leader, the very soul of the revolution. The Mercurians worship him. They will do whatever he says. Without him they would be lost. He is the rebel, Karfial Hodes. I am to offer you ten thousand Earth notes to kill Karfial Hodes." Jaro Moynahan refilled their empty glasses. He was a big man, handsome in a gaunt fashion. Only his eyes were different. They were flat and a trifle oblique with straight brows. The pupils were a pale and penetrating blue that could probe like a surgeon's knife. Now he caught the girl's eyes and held them with his own as a man spears a fish. "Why call me all the way from Mars for that? Why not have that gunman at the piano rub Hodes out?" The girl started, glanced at the pianist, said with a shiver: "We can't locate Karfial Hodes. Don't look at me that way, Jaro. You frighten me. I'm telling the truth. We can't find him. That's why we called you. You've got to find him, Jaro. He's stirring up all Mercury." "Who's putting up the money?" "I can't tell you." "Ah," said Jaro Moynahan; "so that's the way it is." "That's the way it is." "There isn't much time," he said after a moment. "The Rains are due any day now." "No," the girl replied. "But we think he's here in the city." "Why? What makes you think that?" "He was seen," she began, then stopped with a gasp. The lights had gone out. It was as unexpected as a shot in the back. One moment the garden was glowing in light, the next the hot black night swooped down on the revelers, pressing against their eyes like dark wool. The fans about the walls slowed audibly and stopped. It grew hotter, closer. Jaro Moynahan slipped sideways from the table. He felt something brush his sleeve. Somewhere a girl giggled. "What's coming off here?" growled a petulant male voice. Other voices took up the plaint. Across the table from Jaro there was the feel of movement; he could sense it. An exclamation was suddenly choked off as if a hand had been clamped over the girl's mouth. "Red!" said Jaro in a low voice. There was no answer. "Red!" he repeated, louder. Unexpectedly, the deep, ringing voice of Mercury Sam boomed out from the stage. "It's all right. The master fuse blew out. The lights will be on in a moment." On the heels of his speech the lights flashed on, driving the night upward. The fans recommenced their monotonous whirring. Jaro Moynahan glanced at the table. The red-headed singer was gone. So was the pianist. Jaro Moynahan sat quietly back down and poured himself another glass of Latonka. The pale green wine had a delicate yet exhilarating taste. It made him think of cool green grapes beaded with dew. On the hot, teeming planet of Mercury it was as refreshing as a cold plunge. He wondered who was putting up the ten thousand Earth notes? Who stood to lose most in case of a revolution? The answer seemed obvious enough. Who, but Albert Peet. Peet controlled the Latonka trade for which there was a tremendous demand throughout the Universe. And what had happened to the girl. Had the rebels abducted her. If so, he suspected that they had caught a tartar. The Red Witch had the reputation of being able to take care of herself. He beckoned a waiter, paid his bill. As the Mercurian started to leave, a thought struck Jaro. These yellow-eyed Mercurians could see as well in the dark as any alley-prowling cat. For centuries they had lived most their lives beneath ground to escape the terrible rays of the sun. Only at night did they emerge to work their fields and ply their trades. He peeled off a bill, put it in the waiter's hands. "What became of the red-headed singer?" The Mercurian glanced at the bill, then back at the Earthman. There was no expression in his yellow eyes. "She and the man, the queer white one who plays the piano, slipped out the gate to the street." Jaro shrugged, dismissed the waiter. He had not expected to get much information from the waiter, but he was not a man to overlook any possibility. If the girl had been abducted, only Mercurians could have engineered it in the dark; and the Mercurians were a clannish lot. Back on the narrow alley-like street Jaro Moynahan headed for his hostelry. By stretching out his arms he could touch the buildings on either side: buildings with walls four feet thick to keep out the heat of the sun. Beneath his feet, he knew, stretched a labyrinth of rooms and passages. Somewhere in those rat-runs was Karfial Hodes, the revolutionist, and the girl. At infrequent intervals green globes cut a hole in the night, casting a faint illumination. He had just passed one of these futile street lamps when he thought he detected a footfall behind him. It was only the whisper of a sound, but as he passed beyond the circle of radiation, he flattened himself in a doorway. Nothing stirred. There was no further sound. Again he started forward, but now he was conscious of shadows following him. They were never visible, but to his trained ears there came stealthy, revealing noises: the brush of cloth against the baked earth walls, the sly shuffle of a step. He ducked down a bisecting alley, faded into a doorway. Immediately all sounds of pursuit stopped. But as soon as he emerged he was conscious again of the followers. In the dense, humid night, he was like a blind man trying to elude the cat-eyed Mercurians. Jaro Moynahan In the East a sullen red glow stained the heavens like the reflection of a fire. The Mercurian dawn was about to break. With an oath, he set out again for his hostelry. He made no further effort to elude the followers. Once back in his room, Jaro Moynahan stripped off his clothes, unbuckled a shoulder holster containing a compressed air slug gun, stepped under the shower. His body was lean and brown as his face and marked with innumerable scars. There were small round puckered scars and long thin ones, and his left shoulder bore the unmistakable brownish patch of a ray burn. Stepping out of the shower, he dried, rebuckled on the shoulder holster, slipped into pajamas. The pajamas were blue with wide gaudy stripes. Next he lit a cigarette and stretching out on the bed began to contemplate his toes with singular interest. He had, he supposed, killed rather a lot of men. He had fought in the deadly little wars of the Moons of Jupiter for years, then the Universal Debacle of 3368, after that the Martian Revolution as well as dozens of skirmishes between the Federated Venusian States. No, there was little doubt but that he had killed quite a number of men. But this business of hunting a man through the rat-runs beneath the city was out of his line. Furthermore, there was something phony about the entire set up. The Mercurians, he knew, had been agitating for freedom for years. Why, at this time when the Earth Congress was about to grant them self-government, should they stage a revolution? A loud, authoritative rapping at the door interrupted further speculation. He swung his bare feet over the edge of the bed, stood up and ground out his cigarette. Before he could reach the door the rapping came again. Throwing off the latch, he stepped back, balancing on the balls of his feet. "Come in," he called. The door swung open. A heavy set man entered, shut and locked the door, then glanced around casually. His eyes fastened on Jaro. He licked his lips. "Mr. Moynahan, the—ah—professional soldier, I believe." His voice was high, almost feminine. "I'm Albert Peet." He held out a fat pink hand. Jaro said nothing. He ignored the hand, waited, poised like a cat. Mr. Peet licked his lips again. "I have come, Mr. Moynahan, on a matter of business, urgent business. I had not intended to appear in this matter. I preferred to remain behind the scenes, but the disappearance of Miss Mikail has—ah—forced my hand." He paused. Jaro still said nothing. Miss Mikail must be the red-headed singer, whom at different times he had known under a dozen different aliases. He doubted that even she remembered her right name. "Miss Mikail made you a proposition?" Albert Peet's voice was tight. "Yes," said Jaro. "You accepted?" "Why, no. As it happened she was abducted before I had the chance." Mr. Peet licked his lips. "But you will, surely you will. Unless Karfial Hodes is stopped immediately there will be a bloody uprising all over the planet during the Festival of the Rains. Earth doesn't realize the seriousness of the situation." "Then I was right; it is you who are putting up the ten thousand Earth notes." "Not entirely," said Peet uncomfortably. "There are many of us here, Mercurians as well as Earthmen, who recognize the danger. We have—ah—pooled our resources." "But you stand to lose most in case of a successful revolution?" "Perhaps. I have a large interest in the Latonka trade. It is—ah—lucrative." Jaro Moynahan lit a cigarette, sat down on the edge of the bed. "Why beat about the bush," he asked with a sudden grin. "Mr. Peet, you've gained control of the Latonka trade. Other Earthmen are in control of the mines and the northern plantations. Together you form perhaps the strongest combine the Universe has ever seen. You actually run Mercury, and you've squeezed out every possible penny. Every time self-government has come before the Earth Congress you've succeeded in blocking it. You are, perhaps, the most cordially-hated group anywhere. I don't wonder that you are afraid of a revolution." Mr. Peet took out a handkerchief and mopped his forehead. "Fifteen thousand Earth notes I can offer you. But no more. That is as high as I can go." Jaro laughed. "How did you know Red had been kidnapped?" "We have a very efficient information system. I had the report of Miss Mikail's abduction fifteen minutes after the fact." Jaro raised his eyebrows. "Perhaps then you know where she is?" Mr. Peet shook his head. "No. Karfial Hodes' men abducted her." A second rapping at the door caused them to exchange glances. Jaro went to the door, opened it. The pianist at the gardens was framed in the entrance. His black eyes burned holes in his pale boyish face. His white suit was blotched with sweat and dirt. "They told me Mr. Peet was here," he said. "It's for you," said Jaro over his shoulder. Mr. Peet came to the door. "Hello, Stanley. I thought Hodes had you? Where's Miss Mikail?" "I got away. Look, Mr. Peet, I got to see you alone." Albert Peet said, "Would you excuse me, Mr. Moynahan?" He licked his lips. "I'll just step out into the hall a moment." He went out, drawing the door shut after him. Jaro lit a cigarette. He padded nervously back and forth across the room, his bare feet making no noise. He sat down on the edge of the bed. He got up and ground out the cigarette. He went to the door, but did not open it. Instead, he took another turn about the room. Again he came to a halt before the door, pressed his ear against the panel. For a long time he listened but could distinguish no murmur of voices. With an oath he threw open the door. The hall was empty. II Jaro returned to his room, stripped off his pajamas, climbed back into his suit. He tested the slug gun. It was a flat, ugly weapon which hurled a slug the size of a quarter. He preferred it because, though he seldom shot to kill, it stopped a man like a well placed mule's hoof. He adjusted the gun lightly in its holster in order that it wouldn't stick if he were called upon to use it in a hurry. Then he went out into the hall. At the desk he inquired if any messages had come for him. There were none, but the clerk had seen Mr. Peet with a young fellow take the incline to the underground. Above the clerk's head a newsograph was reeling off the current events almost as soon as they happened. Jaro read: " Earth Congress suspends negotiations on Mercurian freedom pending investigation of rumored rebellion. Terrestrials advised to return to Earth. Karfial Hodes, Mercurian patriot, being sought. " Jaro descended the incline to the network of burrows which served as streets during the flaming days. Here in the basements and sub-basements were located the shops and dram houses where the Mercurians sat around little tables drinking silently of the pale green Latonka. The burrows were but poorly lit, the natives preferring the cool gloom, and Jaro had to feel his way, rubbing shoulders with the strange, silent populace. But when he reached the Terrestrial quarter of the city, bright radoxide lights took the place of the green globes, and there was a sprinkling of Colonial guards among the throng. Jaro halted before a door bearing a placard which read: "LATONKA TRUST" He pushed through the door into a rich carpeted reception room. At the far end was a second door beside which sat a desk, door and desk being railed off from the rest of the office. The door into Albert Peet's inner sanctum was ajar. Jaro could distinguish voices; then quite clearly he heard Albert Peet say in a high girlish tone: "Stanley, I thought I left you in the native quarter. Why did you follow me? How many times have I told you never to come here?" The reply was unintelligible. Then the pale-faced young man came through the door shutting it after himself. At the sight of Jaro Moynahan he froze. "What're you sneaking around here for?" Jaro settled himself warily, his light blue eyes flicking over the youth. "Let's get this straight," he said mildly. "I've known your kind before. Frankly, ever since I saw you I've had to repress a desire to step on you as I might a spider." The youth's black eyes were hot as coals, his fingers twitching. His hands began to creep upward. "You dirty ..." he began, but he got no further. Jaro Moynahan shot him in the shoulder. The compressed air slug gun had seemed to leap into Jaro's hand. The big slug, smacked the gunman's shoulder with a resounding thwack, hurled him against the wall. Jaro vaulted the rail, deftly relieved him of two poisoned needle guns. "I'll get you for this," said Stanley, his mouth twisted in pain. "You've broken my shoulder. I'll kill you." The door to the inner sanctum swung open. "What's happened?" cried Albert Peet in distress. "What's wrong with you, Stanley?" "This dirty slob shot me in the shoulder." "But how badly?" Peet was wringing his hands. "Nothing serious," said Jaro. "He'll have his arm in a sling for a while. That's all." "Stanley," said Mr. Peet. "You're bleeding all over my carpet. Why can't you go in the washroom. There's a tile floor in there. If you hadn't disobeyed this wouldn't have happened. You and your fights. Has anyone called a doctor? Where's Miss Webb? Miss Webb! Oh, Miss Webb! That girl. Miss Webb!" Stanley climbed to his feet, swayed a moment drunkenly, then wobbled out a door on the left just as a tall brunette hurried in from the right. She had straight black hair which hung not quite to her shoulders, and dark brown eyes, and enough of everything else to absorb Jaro's attention. "Oh!" exclaimed Miss Webb as she caught sight of the blood staining the carpet. Joan Webb "There's been an—ah—accident," said Mr. Peet, and he licked his lips. "Call a doctor, Miss Webb." Miss Webb raised an eyebrow, went to the visoscreen. In a moment she had tuned in the prim starched figure of a nurse seated at a desk. "Could Dr. Baer rush right over here? There's been an accident." "Rush over where?" said the girl in the visoscreen. "These gadgets aren't telepathic, honey." "Oh," said Miss Webb, "the offices of the Latonka Trust." The girl in the visoscreen thawed like ice cream in the sun. "I'm sure Dr. Baer can come. He'll be there in a moment." "Thank you," said Miss Webb. She flicked the machine off, then added: "You trollop." Mr. Peet regarded Jaro Moynahan with distress. "Really, Mr. Moynahan, was it necessary to shoot Stanley? Isn't that—ah—a little extreme? I'm afraid it might incapacitate him, and I had a job for him." "Oh," cried Miss Webb, her brown eyes crackling. "Did you shoot that poor boy? Aren't you the big brave man?" "Poor boy?" said Jaro mildly. "Venomous little rattlesnake. I took these toys away from him." He held out the poisoned dart guns. "You take them, Mr. Peet. Frankly, they give me the creeps. They might go off. A scratch from one of those needles would be enough." Mr. Peet accepted the guns gingerly. He held them as if they might explode any minute. He started to put them in his pocket, thought better of it, glanced around helplessly. "Here, Miss Webb," he said, "do something with these. Put them in my desk." Miss Webb's eyes grew round as marbles. "I wouldn't touch one of those nasty little contraptions for all the Latonka on Mercury." "Here, I'll take them," said Stanley coming back into the room. He had staunched the flow of blood. His face was even whiter, if possible. Jaro eyed him coldly as with his good hand the youth dropped the dart guns back into their holsters. "Act like you want to use those and I'll put a slug in your head next time." "Now, Mr. Moynahan." Mr. Peet licked his lips nervously. "Stanley, go into my office. The doctor will be here in a moment. Miss Webb, you may go home. I'll have no more work for you today." Albert Peet led Stanley through the door. Jaro and Miss Webb were alone. With his eye on the door, Jaro said: "When you go out, turn left toward the native quarter. Wait for me in the first grog shop you come to." Miss Webb raised her eyebrows. "What's this? A new technique?" "Look," began Jaro annoyed. "My eyes are practically popping out of my head now," she interrupted. "Another morning like this and I take the first space liner back to Earth." She jammed her hat on backward, snatched her bag from the desk drawer. "I'm not trying to pick you up. This is...." "How disappointing." Jaro began again patiently. "Wait for me in the first grog shop. There's something I must know. It's important." He cleared his throat. "Don't you find the heat rather uncomfortable, Miss Webb. But perhaps you've become accustomed to it." Mr. Peet came back into the room. "Why, no, I mean yes," replied Miss Webb, a blank expression in her eyes. "Goodbye, Miss Webb," said Mr. Peet firmly. Jaro grinned and winked at her. Miss Webb tottered out of the room. As the door closed behind the girl, Albert Peet licked his lips, said: "Mr. Moynahan, I suppose my disappearance back at your room requires some explanation. But the fact is that Stanley brought an important bit of news." He paused. Jaro said nothing. "You might be interested to know that Miss Mikail is quite safe. Karfial Hodes has her, but Stanley assures me she will be quite safe." Again he paused. As Jaro remained silent, his neck mottled up pinkly. "The fact is, Mr. Moynahan, that we won't need you after all. I realize that we've put you to considerable trouble and we're prepared to pay you whatever you believe your time is worth. Say five hundred Earth notes?" "That's fair enough," replied Jaro. Albert Peet sighed. "I have the check made out." "Only," continued Jaro coldly, "I'm not ready to be bought off. I think I'll deal myself a hand in this game." Mr. Peet's face fell. "You won't reconsider?" "Sorry," said Jaro; "but I've got a date. I'm late now." He started to leave. "Stanley!" called Albert Peet. The pale-faced young man appeared in the doorway, the dart gun in his good hand. Jaro Moynahan dropped on his face, jerking out his slug gun as he fell. There was a tiny plop like a cap exploding. He heard the whisper of the poisoned dart as it passed overhead. Then he fired from the floor. The pale-faced young man crumpled like an empty sack. Jaro got up, keeping an eye on Albert Peet, brushed off his knees. "You've killed him," said Peet. "If I were you, Mr. Moynahan, I would be on the next liner back to Earth." Without answering, Jaro backed watchfully from the room. Once Jaro Moynahan had regained the street, he mopped his forehead with his handkerchief. Whatever was going on, these boys played for keeps. Warily he started down the passage toward the native quarter. At the first basement grog shop he turned in. His eyes swept the chamber, then he grinned. At a corner table, a tall glass of Latonka before her, sat Miss Webb. Her hat was still on backwards, and she was perched on the edge of her chair as if ready to spring up and away like a startled faun. " Bang! " said Jaro coming up behind her and poking a long brown finger in the small of her back. Miss Webb uttered a shriek, jerked so violently that her hat tilted over one eye. She regarded him balefully from beneath the brim. "Never a dull moment," she gritted. Still grinning, Jaro sat down. "I'm Jaro Moynahan, Miss Webb. I think Albert Peet forgot to introduce us. There's some skullduggery going on here that I'm particularly anxious to get to the bottom of. I thought you might be able to help me." "Yes," replied Miss Webb sweetly. A native waiter, attracted no doubt by her scream, came over and took Jaro's order. "All right," Jaro smiled, but his pale blue eyes probed the girl thoughtfully. "I'll have to confide certain facts which might be dangerous for you to know. Are you game, Miss Webb?" "Since we're going to be so chummy," she replied; "you might begin by calling me Joan. You make me feel downright ancient." "Well then," he said. "In the first place, I just killed that baby-faced gunman your boss had in his office." " Awk! " said Joan, choking on the Latonka. "It was self-defense," he hastened to assure her. "He took a pot shot at me with that poisoned dart gun." "But the police!" she cried, as she caught her breath. "There'll never be an investigation. Albert Peet will see to that. I was called here on what I supposed was a legitimate revolution. Instead I was offered ten thousand Earth notes to assassinate the leader of the revolution." "What revolution? I'm going around in circles." "The Mercurians, of course." "I don't believe it," said the girl. "The Mercurians are the most peaceable people in the Universe. They've been agitating for freedom, yes. But they believe in passive resistance. I don't believe you could induce a Mercurian to kill, even in self-protection. That's why Albert Peet and the rest of the combine had such an easy time gaining control of the Latonka trade." "Score one," breathed Jaro, "I begin to see light. Miss Webb—ah, Joan—I've a notion that we're going to be a great team. How do you happen to be Albert Peet's private secretary?" "A gal's gotta eat. But the truth is, I was quitting. The Latonka Trust is almost on the rocks. Their stock has been dropping like a meteor." Jaro Moynahan raised his oblique brows but did not interrupt. "Albert Peet," she continued, "has been trying to sell out but nobody will touch the stock, not since it looks as if the Earth Congress is going to grant the Mercurians their freedom. Everybody knows that the first thing the Mercurians will do, will be to boot out the Latonka Trust." "What about this Karfial Hodes?" said Jaro. "I've heard that he's inciting the Mercurians to rebellion. The newscaster had a line about the revolution too. The government has advised all Terrestrials to return to Earth." "It's not true," Joan flared. "It's all a pack of lies invented by the Latonka Trust. I know." "But I should think rumors like that would run down the Latonka stock."
|
A. keeping all of his power and money
|
What off-the-shelf QA model was used to answer sub-questions?
|
### Introduction
Question answering (QA) systems have become remarkably good at answering simple, single-hop questions but still struggle with compositional, multi-hop questions BIBREF0, BIBREF1. In this work, we examine if we can answer hard questions by leveraging our ability to answer simple questions. Specifically, we approach QA by breaking a hard question into a series of sub-questions that can be answered by a simple, single-hop QA system. The system's answers can then be given as input to a downstream QA system to answer the hard question, as shown in Fig. FIGREF1. Our approach thus answers the hard question in multiple, smaller steps, which can be easier than answering the hard question all at once. For example, it may be easier to answer “What profession do H. L. Mencken and Albert Camus have in common?” when given the answers to the sub-questions “What profession does H. L. Mencken have?” and “Who was Albert Camus?” Prior work in learning to decompose questions into sub-questions has relied on extractive heuristics, which generalizes poorly to different domains and question types, and requires human annotation BIBREF2, BIBREF3. In order to scale to any arbitrary question, we would require sophisticated natural language generation capabilities, which often relies on large quantities of high-quality supervised data. Instead, we find that it is possible to learn to decompose questions without supervision. Specifically, we learn to map from the distribution of hard questions to the distribution of simpler questions. First, we automatically construct a noisy, “pseudo-decomposition” for each hard question by retrieving relevant sub-question candidates based on their similarity to the given hard question. We retrieve candidates from a corpus of 10M simple questions that we extracted from Common Crawl. Second, we train neural text generation models on that data with (1) standard sequence-to-sequence learning and (2) unsupervised sequence-to-sequence learning. The latter has the advantage that it can go beyond the noisy pairing between questions and pseudo-decompositions. Fig. FIGREF2 overviews our decomposition approach. We use decompositions to improve multi-hop QA. We first use an off-the-shelf single-hop QA model to answer decomposed sub-questions. We then give each sub-question and its answer as additional input to a multi-hop QA model. We test our method on HotpotQA BIBREF0, a popular multi-hop QA benchmark. Our contributions are as follows. First, QA models relying on decompositions improve accuracy over a strong baseline by 3.1 F1 on the original dev set, 11 F1 on the multi-hop dev set from BIBREF4, and 10 F1 on the out-of-domain dev set from BIBREF3. Our most effective decomposition model is a 12-block transformer encoder-decoder BIBREF5 trained using unsupervised sequence-to-sequence learning, involving masked language modeling, denoising, and back-translation objectives BIBREF6. Second, our method is competitive with state-of-the-art methods SAE BIBREF7 and HGN BIBREF8 which leverage strong supervision. Third, we show that our approach automatically learns to generate useful decompositions for all 4 question types in HotpotQA, highlighting the general nature of our approach. In our analysis, we explore how sub-questions improve multi-hop QA, and we provide qualitative examples that highlight how question decomposition adds a form of interpretability to black-box QA models. Our ablations show that each component of our pipeline contributes to QA performance. Overall, we find that it is possible to successfully decompose questions without any supervision and that doing so improves QA. ### Method
We now formulate the problem and overview our high-level approach, with details in the following section. We aim to leverage a QA model that is accurate on simple questions to answer hard questions, without using supervised question decompositions. Here, we consider simple questions to be “single-hop” questions that require reasoning over one paragraph or piece of evidence, and we consider hard questions to be “multi-hop.” Our aim is then to train a multi-hop QA model $M$ to provide the correct answer $a$ to a multi-hop question $q$ about a given a context $c$ (e.g., several paragraphs). Normally, we would train $M$ to maximize $\log p_M(a | c, q)$. To help $M$, we leverage a single-hop QA model that may be queried with sub-questions $s_1, \dots , s_N$, whose “sub-answers” to each sub-question $a_1, \dots , a_N$ may be provided to the multi-hop QA model. $M$ may then instead maximize the (potentially easier) objective $\log p_M(a | c, q, [s_1, a_1], \dots , [a_N, s_N])$. Supervised decomposition models learn to map each question $q \in Q$ to a decomposition $d = [s_1; \dots ; s_N]$ of $N$ sub-questions $s_n \in S$ using annotated $(q, d)$ examples. In this work, we do not assume access to strong $(q, d)$ supervision. To leverage the single-hop QA model without supervision, we follow a three-stage approach: 1) map a question $q$ into sub-questions $s_1, \dots , s_N$ via unsupervised techniques, 2) find sub-answers $a_1, \dots , a_N$ with the single-hop QA model, and 3) provide $s_1, \dots , s_N$ and $a_1, \dots , a_N$ to help predict $a$. ### Method ::: Unsupervised Question Decomposition
To train a decomposition model, we need appropriate training data. We assume access to a hard question corpus $Q$ and a simple question corpus $S$. Instead of using supervised $(q, d)$ training examples, we design an algorithm that constructs pseudo-decompositions $d^{\prime }$ to form $(q, d^{\prime })$ pairs from $Q$ and $S$ using an unsupervised approach (§SECREF4). We then train a model to map $q$ to a decomposition. We explore learning to decompose with standard and unsupervised sequence-to-sequence learning (§SECREF6). ### Method ::: Unsupervised Question Decomposition ::: Creating Pseudo-Decompositions
For each $q \in Q$, we construct a pseudo-decomposition set $d^{\prime } = \lbrace s_1; \dots ; s_N\rbrace $ by retrieving simple question $s$ from $S$. We concatenate all $N$ simple questions in $d^{\prime }$ to form the pseudo-decomposition used downstream. $N$ may be chosen based on the task or vary based on $q$. To retrieve useful simple questions for answering $q$, we face a joint optimization problem. We want sub-questions that are both (i) similar to $q$ according to some metric $f$ and (ii) maximally diverse: ### Method ::: Unsupervised Question Decomposition ::: Learning to Decompose
Having now retrieved relevant pseudo-decompositions, we examine different ways to learn to decompose (with implementation details in the following section): ### Method ::: Unsupervised Question Decomposition ::: Learning to Decompose ::: No Learning
We use pseudo-decompositions directly, employing retrieved sub-questions in downstream QA. ### Method ::: Unsupervised Question Decomposition ::: Learning to Decompose ::: Sequence-to-Sequence (Seq2Seq)
We train a Seq2Seq model with parameters $\theta $ to maximize $\log p_{\theta }(d^{\prime } | q)$. ### Method ::: Unsupervised Question Decomposition ::: Learning to Decompose ::: Unsupervised Sequence-to-Sequence (USeq2Seq)
We start with paired $(q, d^{\prime })$ examples but do not learn from the pairing, because the pairing is noisy. We use unsupervised sequence-to-sequence learning to learn a $q \rightarrow d$ mapping instead of training directly on the noisy pairing. ### Method ::: Answering Sub-Questions
To answer the generated sub-questions, we use an off-the-shelf QA model. The QA model may answer sub-questions using any free-form text (i.e., a word, phrase, sentence, etc.). Any QA model is suitable, so long as it can accurately answer simple questions in $S$. We thus leverage good accuracy on questions in $S$ to help QA models on questions in $Q$. ### Method ::: QA using Decompositions
Downstream QA systems may use sub-questions and sub-answers in various ways. We add sub-questions and sub-answers as auxiliary input for a downstream QA model to incorporate in its processing. We now describe the implementation details of our approach outlined above. ### Experimental Setup ::: Question Answering Task
We test unsupervised decompositions on HotpotQA BIBREF0, a standard benchmark for multi-hop QA. We use HotpotQA's “Distractor Setting,” which provides 10 context paragraphs from Wikipedia. Two (or more) paragraphs contain question-relevant sentences called “supporting facts,” and the remaining paragraphs are irrelevant, “distractor paragraphs.” Answers in HotpotQA are either yes, no, or a span of text in an input paragraph. Accuracy is measured with F1 and Exact Match (EM) scores between the predicted and gold spans. ### Experimental Setup ::: Unsupervised Decomposition ::: Question Data
We use HotpotQA questions as our initial multi-hop, hard question corpus $Q$. We use SQuAD 2 questions as our initial single-hop, simple question corpus $S$. However, our pseudo-decomposition corpus should be large, as the corpus will be used to train neural Seq2Seq models, which are data hungry. A larger $|S|$ will also improve the relevance of retrieved simple questions to the hard question. Thus, we take inspiration from work in machine translation on parallel corpus mining BIBREF9, BIBREF10 and in unsupervised QA BIBREF11. We augment $Q$ and $S$ by mining more questions from Common Crawl. We choose sentences which start with common “wh”-words and end with “?” Next, we train a FastText classifier BIBREF12 to classify between 60K questions sampled from Common Crawl, SQuAD 2, and HotpotQA. Then, we classify Common Crawl questions, adding questions classified as SQuAD 2 questions to $S$ and questions classified as HotpotQA questions to $Q$. Question mining greatly increases the number of single-hop questions (130K $\rightarrow $ 10.1M) and multi-hop questions (90K $\rightarrow $ 2.4M). Thus, our unsupervised approach allows us to make use of far more data than supervised counterparts. ### Experimental Setup ::: Unsupervised Decomposition ::: Creating Pseudo-Decompositions
To create pseudo-decompositions, we set the number $N$ of sub-questions per question to 2, as questions in HotpotQA usually involve two reasoning hops. In Appendix §SECREF52, we discuss how our method works when $N$ varies per question. ### Experimental Setup ::: Unsupervised Decomposition ::: Creating Pseudo-Decompositions ::: Similarity-based Retrieval
To retrieve question-relevant sub-questions, we embed any text $t$ into a vector $\mathbf {v}_t$ by summing the FastText vectors BIBREF13 for words in $t$. We use cosine similarity as our similarity metric $f$. Let $q$ be a multi-hop question used to retrieve pseudo-decomposition $(s_1^*, s_2^*)$, and let $\hat{\mathbf {v}}$ be the unit vector of $\mathbf {v}$. Since $N=2$, Eq. DISPLAY_FORM5 reduces to: The last term requires $O(|S|^2)$ comparisons, which is expensive as $|S|$ is large ($>$10M). Instead of solving Eq. (DISPLAY_FORM19) exactly, we find an approximate pseudo-decomposition $(s_1^{\prime }, s_2^{\prime })$ by computing Eq. (DISPLAY_FORM19) over $S^{\prime } = \operatornamewithlimits{topK}_{\lbrace s \in S\rbrace }\left[ \mathbf {\hat{v}}_{q}^{\top } \mathbf {\hat{v}}_s\right]$, using $K=1000$. We use FAISS BIBREF14 to efficiently build $S^{\prime }$. ### Experimental Setup ::: Unsupervised Decomposition ::: Creating Pseudo-Decompositions ::: Random Retrieval
For comparison, we test random pseudo-decompositions, where we randomly retrieve $s_1, \dots , s_N$ by sampling from $S$. USeq2Seq trained on random $d^{\prime } = [s_1; \dots ; s_N]$ should at minimum learn to map $q$ to multiple simple questions. ### Experimental Setup ::: Unsupervised Decomposition ::: Creating Pseudo-Decompositions ::: Editing Pseudo-Decompositions
Since the sub-questions are retrieval-based, the sub-questions are often not about the same entities as $q$. As a post-processing step, we replace entities in $(s^{\prime }_1, s^{\prime }_2)$ with entities from $q$. We find all entities in $(s^{\prime }_1, s^{\prime }_2)$ that do not appear in $q$ using spaCy BIBREF15. We replace these entities with a random entity from $q$ with the same type (e.g., “Date” or “Location”) if and only if one exists. We use entity replacement on pseudo-decompositions from both random and similarity-based retrieval. ### Experimental Setup ::: Unsupervised Decomposition ::: Unsupervised Decomposition Models ::: Pre-training
Pre-training is a key ingredient for unsupervised Seq2Seq methods BIBREF16, BIBREF17, so we initialize all decomposition models with the same pre-trained weights, regardless of training method (Seq2Seq or USeq2Seq). We warm-start our pre-training with the pre-trained, English Masked Language Model (MLM) from BIBREF6, a 12-block decoder-only transformer model BIBREF5 trained to predict masked-out words on Toronto Books Corpus BIBREF18 and Wikipedia. We train the model with the MLM objective for one epoch on the augmented corpus $Q$ (2.4 M questions), while also training on decompositions $D$ formed via random retrieval from $S$. For our pre-trained encoder-decoder, we initialize a 6-block encoder with the first 6 MLM blocks, and we initialize a 6-block decoder with the last 6 MLM blocks, randomly initializing the remaining weights as in BIBREF6. ### Experimental Setup ::: Unsupervised Decomposition ::: Unsupervised Decomposition Models ::: Seq2Seq
We fine-tune the pre-trained encoder-decoder using maximum likelihood. We stop training based on validation BLEU BIBREF19 between generated decompositions and pseudo-decompositions. ### Experimental Setup ::: Unsupervised Decomposition ::: Unsupervised Decomposition Models ::: USeq2Seq
We follow the approach by BIBREF6 in unsupervised translation. Training follows two stages: (1) MLM pre-training on the training corpora (described above), followed by (2) training simultaneously with denoising and back-translation objectives. For denoising, we produce a noisy input $\hat{d}$ by randomly masking, dropping, and locally shuffling tokens in $d \sim D$, and we train a model with parameters $\theta $ to maximize $\log p_{\theta }(d | \hat{d})$. We likewise maximize $\log p_{\theta }(q | \hat{q})$. For back-translation, we generate a multi-hop question $\hat{q}$ for a decomposition $d \sim D$, and we maximize $\log p_{\theta }(d | \hat{q})$. Similarly, we maximize $\log p_{\theta }(q | \hat{d})$. To stop training without supervision, we use a modified version of round-trip BLEU BIBREF17 (see Appendix §SECREF56 for details). We train with denoising and back-translation on smaller corpora of HotpotQA questions ($Q$) and their pseudo-decompositions ($D$). ### Experimental Setup ::: Single-hop Question Answering Model
We train our single-hop QA model following prior work from BIBREF3 on HotpotQA. ### Experimental Setup ::: Single-hop Question Answering Model ::: Model Architecture
We fine-tune a pre-trained model to take a question and several paragraphs and predicts the answer, similar to the single-hop QA model from BIBREF21. The model computes a separate forward pass on each paragraph (with the question). For each paragraph, the model learns to predict the answer span if the paragraph contains the answer and to predict “no answer” otherwise. We treat yes and no predictions as spans within the passage (prepended to each paragraph), as in BIBREF22 on HotpotQA. During inference, for the final softmax, we consider all paragraphs as a single chunk. Similar to BIBREF23, we subtract a paragraph's “no answer” logit from the logits of all spans in that paragraph, to reduce or increase span probabilities accordingly. In other words, we compute the probability $p(s_p)$ of each span $s_p$ in a paragraph $p \in \lbrace 1, \dots , P \rbrace $ using the predicted span logit $l(s_p)$ and “no answer” paragraph logit $n(p)$ as follows: We use $\textsc {RoBERTa}_{\textsc {LARGE}}$ BIBREF24 as our pre-trained initialization. Later, we also experiment with using the $\textsc {BERT}_{\textsc {BASE}}$ ensemble from BIBREF3. ### Experimental Setup ::: Single-hop Question Answering Model ::: Training Data and Ensembling
Similar to BIBREF3, we train an ensemble of 2 single-hop QA models using data from SQuAD 2 and HotpotQA questions labeled as “easy” (single-hop). To ensemble, we average the logits of the two models before predicting the answer. SQuAD is a single-paragraph QA task, so we adapt SQuAD to the multi-paragraph setting by retrieving distractor paragraphs from Wikipedia for each question. We use the TFIDF retriever from DrQA BIBREF25 to retrieve 2 distractor paragraphs, which we add to the input for one model in the ensemble. We drop words from the question with a 5% probability to help the model handle any ill-formed sub-questions. We use the single-hop QA ensemble as a black-box model once trained, never training the model on multi-hop questions. ### Experimental Setup ::: Single-hop Question Answering Model ::: Returned Text
We have the single-hop QA model return the sentence containing the model's predicted answer span, alongside the sub-questions. Later, we compare against alternatives, i.e., returning the predicted answer span without its context or not returning sub-questions. ### Experimental Setup ::: Multi-hop Question Answering Model
Our multi-hop QA architecture is identical to the single-hop QA model, but the multi-hop QA model also uses sub-questions and sub-answers as input. We append each (sub-question, sub-answer) pair in order to the multi-hop question along with separator tokens. We train one multi-hop QA model on all of HotpotQA, also including SQuAD 2 examples used to train the single-hop QA model. Later, we experiment with using $\textsc {BERT}_{\textsc {LARGE}}$ and $\textsc {BERT}_{\textsc {BASE}}$ instead of $\textsc {RoBERTa}_{\textsc {LARGE}}$ as the multi-hop QA model. All reported error margins show the mean and std. dev. across 5 multi-hop QA training runs using the same decompositions. ### Results on Question Answering
We compare variants of our approach that use different learning methods and different pseudo-aligned training sets. As a baseline, we compare RoBERTa with decompositions to a RoBERTa model that does not use decompositions but is identical in all other respects. We train the baseline for 2 epochs, sweeping over batch size $\in \lbrace 64, 128\rbrace $, learning rate $\in \lbrace 1 \times 10^{-5}, 1.5 \times 10^{-5}, 2 \times 10^{-5}, 3 \times 10^{-5}\rbrace $, and weight decay $\in \lbrace 0, 0.1, 0.01, 0.001\rbrace $; we choose the hyperparameters that perform best on our dev set. We then use the best hyperparameters for the baseline to train our RoBERTa models with decompositions. We report results on 3 versions of the dev set: (1) the original version, (2) the multi-hop version from BIBREF4 which created some distractor paragraphs adversarially to test multi-hop reasoning, and (3) the out-of-domain version from BIBREF3 which retrieved distractor paragraphs using the same procedure as the original version, but excluded paragraphs in the original version. ### Results on Question Answering ::: Main Results
Table shows how unsupervised decompositions affect QA. Our RoBERTa baseline performs quite well on HotpotQA (77.0 F1), despite processing each paragraph separately, which prohibits inter-paragraph reasoning. The result is in line with prior work which found that a version of our baseline QA model using BERT BIBREF26 does well on HotpotQA by exploiting single-hop reasoning shortcuts BIBREF21. We achieve significant gains over our strong baseline by leveraging decompositions from our best decomposition model, trained with USeq2Seq on FastText pseudo-decompositions; we find a 3.1 F1 gain on the original dev set, 11 F1 gain on the multi-hop dev set, and 10 F1 gain on the out-of-domain dev set. Unsupervised decompositions even match the performance of using (within our pipeline) supervised and heuristic decompositions from DecompRC (i.e., 80.1 vs. 79.8 F1 on the original dev set). More generally, all decomposition methods improve QA over the baseline by leveraging the single-hop QA model (“1hop” in Table ). Using FastText pseudo-decompositions as sub-questions directly improves QA over using random sub-questions on the multi-hop set (72.4 vs. 70.9 F1) and out-of-domain set (72.0 vs. 70.7 F1). USeq2Seq on random pseudo-decompositions also improves over the random sub-question baseline (e.g., 79.8 vs. 78.4 F1 on HotpotQA). However, we only find small improvements when training USeq2Seq on FastText vs. Random pseudo-decompositions (e.g., 77.1 vs. 76.5 F1 on the out-of-domain dev set). The best decomposition methods learn with USeq2Seq. Using Seq2Seq to generate decompositions gives similar QA accuracy as the “No Learning” setup, e.g. both approaches achieve 78.9 F1 on the original dev set for FastText pseudo-decompositions. The results are similar perhaps since supervised learning is directly trained to place high probability on pseudo-decompositions. USeq2Seq may improve over Seq2Seq by learning to align hard questions and pseudo-decompositions while ignoring the noisy pairing. After our experimentation, we chose USeq2Seq trained on FastText pseudo-decompositions as the final model, and we submitted the model for hidden test evaluation. Our approach achieved a test F1 of 79.34 and Exact Match (EM) of 66.33. Our approach is competitive with concurrent, state-of-the-art systems SAE BIBREF7 and HGN BIBREF8, which both (unlike our approach) learn from additional, strong supervision about which sentences are necessary to answer the question. ### Results on Question Answering ::: Question Type Breakdown
To understand where decompositions help, we break down QA performance across 4 question types from BIBREF3. “Bridge” questions ask about an entity not explicitly mentioned in the question (“When was Erik Watts' father born?”). “Intersection” questions ask to find an entity that satisfies multiple separate conditions (“Who was on CNBC and Fox News?”). “Comparison” questions ask to compare a property of two entities (“Which is taller, Momhil Sar or K2?”). “Single-hop” questions are likely answerable using single-hop shortcuts or single-paragraph reasoning (“Where is Electric Six from?”). We split the original dev set into the 4 types using the supervised type classifier from BIBREF3. Table shows F1 scores for RoBERTa with and without decompositions across the 4 types. Unsupervised decompositions improve QA across all question types. Our single decomposition model generates useful sub-questions for all question types without special case handling, unlike earlier work from BIBREF3 which handled each question type separately. For single-hop questions, our QA approach does not require falling back to a single-hop QA model and instead learns to leverage decompositions to better answer questions with single-hop shortcuts (76.9 vs. 73.9 F1 without decompositions). ### Results on Question Answering ::: Answers to Sub-Questions are Crucial
To measure the usefulness of sub-questions and sub-answers, we train the multi-hop QA model with various, ablated inputs, as shown in Table . Sub-answers are crucial to improving QA, as sub-questions with no answers or random answers do not help (76.9 vs. 77.0 F1 for the baseline). Only when sub-answers are provided do we see improved QA, with or without sub-questions (80.1 and 80.2 F1, respectively). It is important to provide the sentence containing the predicted answer span instead of the answer span alone (80.1 vs. 77.8 F1, respectively), though the answer span alone still improves over the baseline (77.0 F1). ### Results on Question Answering ::: How Do Decompositions Help?
Decompositions help to answer questions by retrieving important supporting evidence to answer questions. Fig. FIGREF41 shows that multi-hop QA accuracy increases when the sub-answer sentences are the “supporting facts” or sentences needed to answer the question, as annotated by HotpotQA. We retrieve supporting facts without learning to predict them with strong supervision, unlike many state-of-the-art models BIBREF7, BIBREF8, BIBREF22. ### Results on Question Answering ::: Example Decompositions
To illustrate how decompositions help QA, Table shows example sub-questions from our best decomposition model with predicted sub-answers. Sub-questions are single-hop questions relevant to the multi-hop question. The single-hop QA model returns relevant sub-answers, sometimes in spite of grammatical errors (Q1, SQ$_1$) or under-specified questions (Q2, SQ$_1$). The multi-hop QA model then returns an answer consistent with the predicted sub-answers. The decomposition model is largely extractive, copying from the multi-hop question rather than hallucinating new entities, which helps generate relevant sub-questions. To better understand our system, we analyze the model for each stage: decomposition, single-hop QA, and multi-hop QA. ### Analysis ::: Unsupervised Decomposition Model ::: Intrinsic Evaluation of Decompositions
We evaluate the quality of decompositions on other metrics aside from downstream QA. To measure the fluency of decompositions, we compute the likelihood of decompositions using the pre-trained GPT-2 language model BIBREF27. We train a classifier on the question-wellformedness dataset of BIBREF28, and we use the classifier to estimate the proportion of sub-questions that are well-formed. We measure how abstractive decompositions are by computing (i) the token Levenstein distance between the multi-hop question and its generated decomposition and (ii) the ratio between the length of the decomposition and the length of the multi-hop question. We compare our best decomposition model against the supervised+heuristic decompositions from DecompRC BIBREF3 in Table . Unsupervised decompositions are both more natural and well-formed than decompositions from DecompRC. Unsupervised decompositions are also closer in edit distance and length to the multi-hop question, consistent with our observation that our decomposition model is largely extractive. ### Analysis ::: Unsupervised Decomposition Model ::: Quality of Decomposition Model
Another way to test the quality of the decomposition model is to test if the model places higher probability on decompositions that are more helpful for downstream QA. We generate $N=5$ hypotheses from our best decomposition model using beam search, and we train a multi-hop QA model to use the $n$th-ranked hypothesis as a question decomposition (Fig. FIGREF46, left). QA accuracy decreases as we use lower probability decompositions, but accuracy remains relatively robust, at most decreasing from 80.1 to 79.3 F1. The limited drop suggests that decompositions are still useful if they are among the model's top hypotheses, another indication that our model is trained well for decomposition. ### Analysis ::: Single-hop Question Answering Model ::: Sub-Answer Confidence
Figure FIGREF46 (right) shows that the model's sub-answer confidence correlates with downstream multi-hop QA performance for all HotpotQA dev sets. A low confidence sub-answer may be indicative of (i) an unanswerable or ill-formed sub-question or (ii) a sub-answer that is more likely to be incorrect. In both cases, the single-hop QA model is less likely to retrieve the useful supporting evidence to answer the multi-hop question. ### Analysis ::: Single-hop Question Answering Model ::: Changing the Single-hop QA Model
We find that our approach is robust to the single-hop QA model that answers sub-questions. We use the $\textsc {BERT}_{\textsc {BASE}}$ ensemble from BIBREF3 as the single-hop QA model. The model performs much worse compared to our $\textsc {RoBERTa}_{\textsc {LARGE}}$ single-hop ensemble when used directly on HotpotQA (56.3 vs. 66.7 F1). However, the model results in comparable QA when used to answer single-hop sub-questions within our larger system (79.9 vs. 80.1 F1 for our $\textsc {RoBERTa}_{\textsc {LARGE}}$ ensemble). ### Analysis ::: Multi-hop Question Answering Model ::: Varying the Base Model
To understand how decompositions impact performance as the multi-hop QA model gets stronger, we vary the base pre-trained model. Table shows the impact of adding decompositions to $\textsc {BERT}_{\textsc {BASE}}$ , $\textsc {BERT}_{\textsc {LARGE}}$ , and finally $\textsc {RoBERTa}_{\textsc {LARGE}}$ (see Appendix §SECREF64 for hyperparameters). The gain from using decompositions grows with strength of the multi-hop QA model. Decompositions improve QA by 1.2 F1 for a $\textsc {BERT}_{\textsc {BASE}}$ model, by 2.6 F1 for the stronger $\textsc {BERT}_{\textsc {LARGE}}$ model, and by 3.1 F1 for our best $\textsc {RoBERTa}_{\textsc {LARGE}}$ model. ### Related Work
Answering complicated questions has been a long-standing challenge in natural language processing. To this end, prior work has explored decomposing questions with supervision or heuristic algorithms. IBM Watson BIBREF29 decomposes questions into sub-questions in multiple ways or not at all. DecompRC BIBREF3 largely frames sub-questions as extractive spans of a multi-hop question, learning to predict span-based sub-questions via supervised learning on human annotations. In other cases, DecompRC decomposes a multi-hop question using a heuristic algorithm, or DecompRC does not decompose at all. Watson and DecompRC use special case handling to decompose different questions, while our algorithm is fully automated and requires minimal hand-engineering. More traditional, semantic parsing methods map questions to compositional programs, whose sub-programs can be viewed as question decompositions in a formal language BIBREF2, BIBREF30. Examples include classical QA systems like SHRDLU BIBREF31 and LUNAR BIBREF32, as well as neural Seq2Seq semantic parsers BIBREF33 and neural module networks BIBREF34, BIBREF35. Such methods usually require strong, program-level supervision to generate programs, as in visual QA BIBREF36 and on HotpotQA BIBREF37. Some models use other forms of strong supervision, e.g. predicting the “supporting evidence” to answer a question annotated by HotpotQA. Such an approach is taken by SAE BIBREF7 and HGN BIBREF8, whose methods may be combined with our approach. Unsupervised decomposition complements strongly and weakly supervised decomposition approaches. Our unsupervised approach enables methods to leverage millions of otherwise unusable questions, similar to work on unsupervised QA BIBREF11. When decomposition examples exist, supervised and unsupervised learning can be used in tandem to learn from both labeled and unlabeled examples. Such semi-supervised methods outperform supervised learning for tasks like machine translation BIBREF38. Other work on weakly supervised question generation uses a downstream QA model's accuracy as a signal for learning to generate useful questions. Weakly supervised question generation often uses reinforcement learning BIBREF39, BIBREF40, BIBREF41, BIBREF42, BIBREF43, where an unsupervised initialization can greatly mitigate the issues of exploring from scratch BIBREF44. ### Conclusion
We proposed an algorithm that decomposes questions without supervision, using 3 stages: (1) learning to decompose using pseudo-decompositions without supervision, (2) answering sub-questions with an off-the-shelf QA system, and (3) answering hard questions more accurately using sub-questions and their answers as additional input. When evaluated on HotpotQA, a standard benchmark for multi-hop QA, our approach significantly improved accuracy over an equivalent model that did not use decompositions. Our approach relies only on the final answer as supervision but works as effectively as state-of-the-art methods that rely on strong supervision, such as supporting fact labels or example decompositions. Qualitatively, we found that unsupervised decomposition resulted in fluent sub-questions whose answers often match the annotated supporting facts in HotpotQA. Our unsupervised decompositions are largely extractive, which is effective for compositional, multi-hop questions but not all complex questions, showing room for future work. Overall, this work opens up exciting avenues for leveraging methods in unsupervised learning and natural language generation to improve the interpretability and generalization of machine learning systems. ### Acknowledgements
EP is supported by the NSF Graduate Research Fellowship. KC is supported by Samsung Advanced Institute of Technology (Next Generation Deep Learning: from pattern recognition to AI) and Samsung Research (Improving Deep Learning using Latent Structure). KC also thanks eBay and NVIDIA for their support. We thank Paul Christiano, Sebastian Riedel, He He, Jonathan Berant, Alexis Conneau, Jiatao Gu, Sewon Min, Yixin Nie, Lajanugen Logeswaran, and Adam Fisch for helpful feedback, as well as Yichen Jiang and Peng Qi for help with evaluation. ### Pseudo-Decompositions
Tables - show examples of pseudo-decompositions and learned decompositions from various models. ### Pseudo-Decompositions ::: Variable Length Pseudo-Decompositions
In §SECREF15, we leveraged domain knowledge about the task to fix the pseudo-decomposition length $N=2$. A general algorithm for creating pseudo-decompositions should find a suitable $N$ for each question. We find that Eq. DISPLAY_FORM5 in SECREF4 always results in decompositions of length $N=2$, as the regularization term grows quickly with $N$. Thus, we test another formulation based on Euclidean distance: We create pseudo-decompositions in an similar way as before, first finding a set of candidate sub-questions $S^{\prime } \in S$ with high cosine similarity to $\mathbf {v}_q$, then performing beam search up to a maximum value of $N$. We test pseudo-decomposition formulations by creating synthetic compositional questions by combining 2-3 single-hop questions with “and.” We then measure the ranking of the correct decomposition (a concatenation of the single-hop questions). For $N=2$, both methods perform well, but Eq. DISPLAY_FORM5 does not work for decompositions where $N=3$, whereas Eq. DISPLAY_FORM53 does, achieving a mean reciprocal rank of 30%. However, Eq. DISPLAY_FORM5 outperforms Eq. DISPLAY_FORM53 on HotpotQA, e.g., achieving 79.9 vs. 79.4 F1 when using the $\textsc {BERT}_{\textsc {BASE}}$ ensemble from BIBREF3 to answer sub-questions. Eq. DISPLAY_FORM5 is also faster to compute and easier to scale. Moreover, Eq. DISPLAY_FORM53 requires an embedding space where summing sub-question representations is meaningful, whereas Eq. DISPLAY_FORM5 only requires embeddings that encode semantic similarity. Thus, we adopt Eq. DISPLAY_FORM5 for our main experiments. Table contains an example where the variable length decomposition method mentioned above produces a three-subquestion decomposition whereas the other methods are fixed to two subquestions. ### Pseudo-Decompositions ::: Impact of Question Corpus Size
In addition to our previous results on FastText vs. Random pseudo-decompositions, we found it important to use a large question corpus to create pseudo-decompositions. QA F1 increased from 79.2 to 80.1 when we trained decomposition models on pseudo-decompositions comprised of questions retrieved from Common Crawl ($>$10M questions) rather than only SQuAD 2 ($\sim $130K questions), using an appropriately larger beam size (100 $\rightarrow $ 1000). ### Pseudo-Decompositions ::: Pseudo-Decomposition Retrieval Method
Table shows QA results with pseudo-decompositions retrieved using sum-bag-of-word representations from FastText, TFIDF, $\textsc {BERT}_{\textsc {LARGE}}$ first layer hidden states. We also vary the learning method and include results Curriculum Seq2Seq (CSeq2Seq), where we initialize the USeq2Seq approach with the Seq2Seq model trained on the same data. ### Unsupervised Decomposition Model ::: Unsupervised Stopping Criterion
To stop USeq2Seq training, we use an unsupervised stopping criterion to avoid relying on a supervised validation set of decompositions. We generate a decomposition $\hat{d}$ for a multi-hop question $q$, and we measure BLEU between $q$ and the model-generated question $\hat{q}$ for $\hat{d}$, similar to round-trip BLEU in unsupervised translation BIBREF17. We scale round-trip BLEU score by the fraction of “good” decompositions, where a good decomposition has (1) 2 sub-questions (question marks), (2) no sub-question which contains all words in the multi-hop question, and (3) no sub-question longer than the multi-hop question. Without scaling, decomposition models achieve perfect round-trip BLEU by copying the multi-hop question as the decomposition. We measure scaled BLEU across multi-hop questions in HotpotQA dev, and we stop training when the metric does not increase for 3 consecutive epochs. It is possible to stop training the decomposition model based on downstream QA accuracy. However, training a QA model on each decomposition model checkpoint (1) is computationally expensive and (2) ties decompositions to a specific, downstream QA model. In Figure FIGREF57, we show downstream QA results across various USeq2Seq checkpoints when using the $\textsc {BERT}_{\textsc {BASE}}$ single-hop QA ensemble from BIBREF3. The unsupervised stopping criterion does not significantly hurt downstream QA compared to using a weakly-supervised stopping criterion. ### Unsupervised Decomposition Model ::: Training Hyperparameters ::: MLM Pre-training
We pre-train our encoder-decoder distributed across 8 DGX-1 machines, each with 8, 32GB NVIDIA V100 GPUs interconnected by Infiniband. We pre-train using the largest possible batch size (1536), and we choose the best learning rate ($3 \times 10^{-5}$) based on training loss after a small number of iterations. We chose a maximum sequence length of 128. We keep other hyperparameters identical to those from BIBREF6 used in unsupervised translation. ### Unsupervised Decomposition Model ::: Training Hyperparameters ::: USeq2Seq
We train each decomposition model with distributed training across 8, 32GB NVIDIA V100 GPUs. We chose the largest possible batch size (256) and then the largest learning rate which resulted in stable training ($3 \times 10^{-5}$). Other hyperparameters are the same as BIBREF6. ### Unsupervised Decomposition Model ::: Training Hyperparameters ::: Seq2Seq
We use a large batch size (1024) and chose the largest learning rate which resulted in stable training across many pseudo-decomposition training corpora ($1 \times 10^{-4}$). We keep other training settings and hyperparameters the same as for USeq2Seq. ### Multi-hop QA Model ::: Varying the Number of Training Examples
To understand how decompositions impact performance given different amounts of training data, we vary the number of multi-hop training examples. We use the “medium” and “hard” level labels in HotpotQA to determine which examples are multi-hop. We consider training setups where the multi-hop QA model does or does not use data augmentation via training on hotpot “easy”/single-hop questions and SQuAD 2 questions. Fig. FIGREF63 shows the results. Decompositions improve QA, so long as the multi-hop QA model has enough training data, either via single-hop QA examples or enough multi-hop QA examples. ### Multi-hop QA Model ::: Training Hyperparameters
To train $\textsc {RoBERTa}_{\textsc {LARGE}}$ , we fix the number of training epochs to 2, as training longer did not help. We sweep over batch size $\in \lbrace 64, 128\rbrace $, learning rate $\in \lbrace 1 \times 10^{-5}, 1.5 \times 10^{-5}, 2 \times 10^{-5}, 3 \times 10^{-5}\rbrace $, and weight decay $\in \lbrace 0, 0.1, 0.01, 0.001\rbrace $, similar to the ranges used in the original paper BIBREF24. We chose the hyperparameters that did best for the baseline QA model (without decompositions) on our validation set: batch size 64, learning rate $1.5 \times 10^{-5}$, and weight decay $0.01$. Similarly, for the experiments with BERT, we fix the number of epochs to 2 and choose hyperparameters by sweeping over the recommended ranges from BIBREF26 for learning rate ($\lbrace 2 \times 10^{-5}, 3 \times 10^{-5}, 5 \times 10^{-5}\rbrace $) and batch size ($\lbrace 16, 32\rbrace $). For $\textsc {BERT}_{\textsc {BASE}}$ , we thus choose learning rate $2 \times 10^{-5}$ and batch size 16, and for $\textsc {BERT}_{\textsc {LARGE}}$ , we use the whole-word masking model with learning rate $2 \times 10^{-5}$ and batch size 32. We train all QA models with mixed precision floating point arithmetic BIBREF45, distributing training across 8, 32GB NVIDIA V100 GPUs. ### Multi-hop QA Model ::: Improvements across Detailed Question Types
To better understand where decompositions improve QA, we show the improvement across various fine-grained splits of the evaluation sets in Figures FIGREF66-FIGREF70. Figure 1. Overview: Using unsupervised learning, we decompose a multi-hop question into single-hop sub-questions, whose predicted answers are given to a downstream question answering model. Figure 2. Unsupervised Decomposition: Step 1: We create a corpus of pseudo-decompositions D by finding candidate subquestions from a simple question corpus S which are similar to a multi-hop question in Q. Step 2: We learn to map multi-hop questions to decompositions using Q and D as training data, via either standard or unsupervised sequence-to-sequence learning. Table 1. Unsupervised decompositions significantly improve the F1 on HOTPOTQA over the baseline. We achieve comparable F1 to methods which use supporting fact supervision (†). (*) We use supervised and heuristic decompositions from Min et al. (2019b). (‡) Scores are approximate due to mismatched Wikipedia dumps. Table 2. F1 scores on 4 types of questions in HOTPOTQA. Unsupervised decompositions improves QA for all types. Table 3. Ablation Study: QA model F1 when trained with different sub-answers: the sentence containing the predicted subanswer, the predicted sub-answer span, and a random entity from the context. We also train QA models with (X) or without (7) sub-questions and sub-answers. Figure 3. Multi-hop QA is better when the single-hop QA model answers with the ground truth “supporting fact” sentences. We plot mean and std. across 5 random QA training runs. Table 5. Analysis of sub-questions produced by our method vs. the supervised+heuristic method of Min et al. (2019b). From left-toright: Negative Log-Likelihood (NLL) according to GPT2 (lower is better), % Well-Formed according to a classifier, Edit Distance between decomposition and multi-hop question, and token-wise Length Ratio between decomposition and multi-hop question. Figure 4. Left: We decode from the decomposition model with beam search and use nth-ranked hypothesis as a question decomposition. We plot the F1 of a multi-hop QA model trained to use the nth-ranked decomposition. Right: Multi-hop QA is better when the single-hop QA model places high probability on its sub-answer. Table 4. Example sub-questions generated by our model, along with predicted sub-answer sentences (answer span underlined) and final predicted answer. Table 6. Stronger QA models benefit more from decompositions. Table 7. QA F1 scores for all combinations of learning methods and pseudo-decomposition retrieval methods that we tried. Figure 5. How multi-hop QA accuracy varies over the course of decomposition model training, for one training run of USeq2Seq on FastText pseudo-decompositions. Our unsupervised stopping criterion selects the epoch 3 checkpoint, which performs roughly as well as the best checkpoint (epoch 5). Figure 6. Performance of downstream, multi-hop QA model, with and without decompositions, when varying the amount of training data. We also assess the impact of removing single-hop training data (SQUAD 2.0 and HOTPOTQA“easy” questions). Figure 7. Performance difference for various answer entity types when the QA model does vs. does not use decompositions. We see the largest, consistent gains for entity-centric answers. Figure 9. Performance difference for yes/no and span answer types for comparison questions when the QA model does vs. does not use decompositions. Decompositions are roughly as helpful for yes/no questions as for span-based questions. Figure 8. Performance difference for bridge and comparison questions when the QA model does vs. does not use decompositions. Here, we use the original bridge/comparison splits from HOTPOTQA, which does not have a one-hop category and categorizes intersection questions as bridge. For the original dev set, the improvement with decompositions is greater for comparison questions than bridge questions. The multi-hop set does not alter comparison questions from the original version, so these scores do not change much. Figure 10. Performance difference for various multi-hop “wh”words when the QA model does vs. does not use decompositions. Improvements by question word vary across dev sets. Figure 11. Performance difference between a model when the QA model does vs. does not use decompositions, stratified by whether the gold final answer is in a sub-answer sentence. We find a larger improvement over the baseline when the gold answer contained in a sub-answer sentence. Table 8. Various decomposition methods for the question “What is the name of the singer who’s song was released as the lead single from the album “Confessions,” and that had popular song stuck behind for eight consecutive weeks?” Here, the Variable USeq2Seq model has decomposed the question into three subquestions rather than two. Table 9. Various decomposition methods for the question “Are both Coldplay and Pierre Bouvier from the same country?” Table 10. Various decomposition methods for the question “Who is older, Annie Morton or Terry Richardson?” Table 11. Various decomposition methods for the question “In which year was the King who made the 1925 Birthday Honours born?” Table 12. Various decomposition methods for the question “Where are Teide National Park and Garajonay National Park located? Table 13. Various decomposition methods for the question “Since 2 June 2017, The Leader of Fine Gael had been held by which Irish Fine Gael politician who has served as Taoiseach and Minister for Defence?”
|
$\textsc {BERT}_{\textsc {BASE}}$ ensemble from BIBREF3
|
Raymond sees the two hundred years between his time and Martin's time _____. Martin sees the time span as _____.
A. as time to allow for people to realize that everyone is expendable.
just enough time to dull the perceptions of an entire society.
B. as time to lose something, though he was unable to define it, that was important to society as a whole.
time to refine people.
C. as time to refine people.
time to lose something, though he was unable to define it, that was important to society as a whole.
D. as just enough time to dull the perceptions of an entire society.
time to allow for people to realize that everyone is expendable.
|
THE MAN OUTSIDE By EVELYN E. SMITH Illustrated by DILLON [Transcriber's Note: This etext was produced from Galaxy Science Fiction August 1957. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] No one, least of all Martin, could dispute that a man's life should be guarded by his kin—but by those who hadn't been born yet? Nobody in the neighborhood was surprised when Martin's mother disappeared and Ninian came to take care of him. Mothers had a way of disappearing around those parts and the kids were often better off without them. Martin was no exception. He'd never had it this good while he was living with his old lady. As for his father, Martin had never had one. He'd been a war baby, born of one of the tides of soldiers—enemies and allies, both—that had engulfed the country in successive waves and bought or taken the women. So there was no trouble that way. Sometimes he wondered who Ninian really was. Obviously that story about her coming from the future was just a gag. Besides, if she really was his great-great-grand-daughter, as she said, why would she tell him to call her " Aunt Ninian "? Maybe he was only eleven, but he'd been around and he knew just what the score was. At first he'd thought maybe she was some new kind of social worker, but she acted a little too crazy for that. He loved to bait her, as he had loved to bait his mother. It was safer with Ninian, though, because when he pushed her too far, she would cry instead of mopping up the floor with him. "But I can't understand," he would say, keeping his face straight. "Why do you have to come from the future to protect me against your cousin Conrad?" "Because he's coming to kill you." "Why should he kill me? I ain't done him nothing." Ninian sighed. "He's dissatisfied with the current social order and killing you is part of an elaborate plan he's formulated to change it. You wouldn't understand." "You're damn right. I don't understand. What's it all about in straight gas?" "Oh, just don't ask any questions," Ninian said petulantly. "When you get older, someone will explain the whole thing to you." So Martin held his peace, because, on the whole, he liked things the way they were. Ninian really was the limit, though. All the people he knew lived in scabrous tenement apartments like his, but she seemed to think it was disgusting. "So if you don't like it, clean it up," he suggested. She looked at him as if he were out of his mind. "Hire a maid, then!" he jeered. And darned if that dope didn't go out and get a woman to come clean up the place! He was so embarrassed, he didn't even dare show his face in the streets—especially with the women buttonholing him and demanding to know what gave. They tried talking to Ninian, but she certainly knew how to give them the cold shoulder. One day the truant officer came to ask why Martin hadn't been coming to school. Very few of the neighborhood kids attended classes very regularly, so this was just routine. But Ninian didn't know that and she went into a real tizzy, babbling that Martin had been sick and would make up the work. Martin nearly did get sick from laughing so hard inside. But he laughed out of the other side of his mouth when she went out and hired a private tutor for him. A tutor—in that neighborhood! Martin had to beat up every kid on the block before he could walk a step without hearing "Fancy Pants!" yelled after him. Ninian worried all the time. It wasn't that she cared what these people thought of her, for she made no secret of regarding them as little better than animals, but she was shy of attracting attention. There were an awful lot of people in that neighborhood who felt exactly the same way, only she didn't know that, either. She was really pretty dumb, Martin thought, for all her fancy lingo. "It's so hard to think these things out without any prior practical application to go by," she told him. He nodded, knowing what she meant was that everything was coming out wrong. But he didn't try to help her; he just watched to see what she'd do next. Already he had begun to assume the detached role of a spectator. When it became clear that his mother was never going to show up again, Ninian bought one of those smallish, almost identical houses that mushroom on the fringes of a city after every war, particularly where intensive bombing has created a number of desirable building sites. "This is a much better neighborhood for a boy to grow up in," she declared. "Besides, it's easier to keep an eye on you here." And keep an eye on him she did—she or a rather foppish young man who came to stay with them occasionally. Martin was told to call him Uncle Raymond. From time to time, there were other visitors—Uncles Ives and Bartholomew and Olaf, Aunts Ottillie and Grania and Lalage, and many more—all cousins to one another, he was told, all descendants of his. Martin was never left alone for a minute. He wasn't allowed to play with the other kids in the new neighborhood. Not that their parents would have let them, anyway. The adults obviously figured that if a one-car family hired private tutors for their kid, there must be something pretty wrong with him. So Martin and Ninian were just as conspicuous as before. But he didn't tip her off. She was grown up; she was supposed to know better than he did. He lived well. He had food to eat that he'd never dreamed of before, warm clothes that no one had ever worn before him. He was surrounded by more luxury than he knew what to do with. The furniture was the latest New Grand Rapids African modern. There were tidy, colorful Picasso and Braque prints on the walls. And every inch of the floor was modestly covered by carpeting, though the walls were mostly unabashed glass. There were hot water and heat all the time and a freezer well stocked with food—somewhat erratically chosen, for Ninian didn't know much about meals. The non-glass part of the house was of neat, natural-toned wood, with a neat green lawn in front and a neat parti-colored garden in back. Martin missed the old neighborhood, though. He missed having other kids to play with. He even missed his mother. Sure, she hadn't given him enough to eat and she'd beaten him up so hard sometimes that she'd nearly killed him—but then there had also been times when she'd hugged and kissed him and soaked his collar with her tears. She'd done all she could for him, supporting him in the only way she knew how—and if respectable society didn't like it, the hell with respectable society. From Ninian and her cousins, there was only an impersonal kindness. They made no bones about the fact that they were there only to carry out a rather unpleasant duty. Though they were in the house with him, in their minds and in their talk they were living in another world—a world of warmth and peace and plenty where nobody worked, except in the government service or the essential professions. And they seemed to think even that kind of job was pretty low-class, though better than actually doing anything with the hands. In their world, Martin came to understand, nobody worked with hands; everything was done by machinery. All the people ever did was wear pretty clothes and have good times and eat all they wanted. There was no devastation, no war, no unhappiness, none of the concomitants of normal living. It was then that Martin began to realize that either the whole lot of them were insane, or what Ninian had told him at first was the truth. They came from the future. When Martin was sixteen, Raymond took him aside for the talk Ninian had promised five years before. "The whole thing's all my brother Conrad's fault. You see, he's an idealist," Raymond explained, pronouncing the last word with distaste. Martin nodded gravely. He was a quiet boy now, his brief past a dim and rather ridiculous memory. Who could ever imagine him robbing a grocery store or wielding a broken bottle now? He still was rather undersized and he'd read so much that he'd weakened his eyes and had to wear glasses. His face was pallid, because he spent little time in the sun, and his speech rather overbred, his mentors from the future having carefully eradicated all current vulgarities. "And Conrad really got upset over the way Earth has been exploiting the not so intelligent life-forms on the other planets," Raymond continued. "Which is distressing—though, of course, it's not as if they were people. Besides, the government has been talking about passing laws to do away with the—well, abuses and things like that, and I'm sure someday everything will come out all right. However, Conrad is so impatient." "I thought, in your world, machines did all the work," Martin suggested. "I've told you—our world is precisely the same as this one!" Raymond snapped. "We just come a couple of centuries or so later, that's all. But remember, our interests are identical. We're virtually the same people ... although it is amazing what a difference two hundred odd years of progress and polish can make in a species, isn't it?" He continued more mildly: "However, even you ought to be able to understand that we can't make machinery without metal. We need food. All that sort of thing comes from the out-system planets. And, on those worlds, it's far cheaper to use native labor than to ship out all that expensive machinery. After all, if we didn't give the natives jobs, how would they manage to live?" "How did they live before? Come to think of it, if you don't work, how do you live now?... I don't mean in the now for me, but the now for you," Martin explained laboriously. It was so difficult to live in the past and think in the future. "I'm trying to talk to you as if you were an adult," Raymond said, "but if you will persist in these childish interruptions—" "I'm sorry," Martin said. But he wasn't, for by now he had little respect left for any of his descendants. They were all exceedingly handsome and cultivated young people, with superior educations, smooth ways of speaking and considerable self-confidence, but they just weren't very bright. And he had discovered that Raymond was perhaps the most intelligent of the lot. Somewhere in that relatively short span of time, his line or—more frightening—his race had lost something vital. Unaware of the near-contempt in which his young ancestor held him, Raymond went on blandly: "Anyhow, Conrad took it upon himself to feel particularly guilty, because, he decided, if it hadn't been for the fact that our great-grandfather discovered the super-drive, we might never have reached the stars. Which is ridiculous—his feeling guilty, I mean. Perhaps a great-grandfather is responsible for his great-grandchildren, but a great-grandchild can hardly be held accountable for his great-grandfather." "How about a great-great-grandchild?" Martin couldn't help asking. Raymond flushed a delicate pink. "Do you want to hear the rest of this or don't you?" "Oh, I do!" Martin said. He had pieced the whole thing together for himself long since, but he wanted to hear how Raymond would put it. "Unfortunately, Professor Farkas has just perfected the time transmitter. Those government scientists are so infernally officious—always inventing such senseless things. It's supposed to be hush-hush, but you know how news will leak out when one is always desperate for a fresh topic of conversation." Anyhow, Raymond went on to explain, Conrad had bribed one of Farkas' assistants for a set of the plans. Conrad's idea had been to go back in time and "eliminate!" their common great-grandfather. In that way, there would be no space-drive, and, hence, the Terrestrials would never get to the other planets and oppress the local aborigines. "Sounds like a good way of dealing with the problem," Martin observed. Raymond looked annoyed. "It's the adolescent way," he said, "to do away with it, rather than find a solution. Would you destroy a whole society in order to root out a single injustice?" "Not if it were a good one otherwise." "Well, there's your answer. Conrad got the apparatus built, or perhaps he built it himself. One doesn't inquire too closely into such matters. But when it came to the point, Conrad couldn't bear the idea of eliminating our great-grandfather—because our great-grandfather was such a good man, you know." Raymond's expressive upper lip curled. "So Conrad decided to go further back still and get rid of his great-grandfather's father—who'd been, by all accounts, a pretty worthless character." "That would be me, I suppose," Martin said quietly. Raymond turned a deep rose. "Well, doesn't that just go to prove you mustn't believe everything you hear?" The next sentence tumbled out in a rush. "I wormed the whole thing out of him and all of us—the other cousins and me—held a council of war, as it were, and we decided it was our moral duty to go back in time ourselves and protect you." He beamed at Martin. The boy smiled slowly. "Of course. You had to. If Conrad succeeded in eliminating me, then none of you would exist, would you?" Raymond frowned. Then he shrugged cheerfully. "Well, you didn't really suppose we were going to all this trouble and expense out of sheer altruism, did you?" he asked, turning on the charm which all the cousins possessed to a consternating degree. Martin had, of course, no illusions on that score; he had learned long ago that nobody did anything for nothing. But saying so was unwise. "We bribed another set of plans out of another of the professor's assistants," Raymond continued, as if Martin had answered, "and—ah—induced a handicraft enthusiast to build the gadget for us." Induced , Martin knew, could have meant anything from blackmail to the use of the iron maiden. "Then we were all ready to forestall Conrad. If one of us guarded you night and day, he would never be able to carry out his plot. So we made our counter-plan, set the machine as far back as it would go—and here we are!" "I see," Martin said. Raymond didn't seem to think he really did. "After all," he pointed out defensively, "whatever our motives, it has turned into a good thing for you. Nice home, cultured companions, all the contemporary conveniences, plus some handy anachronisms—I don't see what more you could ask for. You're getting the best of all possible worlds. Of course Ninian was a ninny to locate in a mercantile suburb where any little thing out of the way will cause talk. How thankful I am that our era has completely disposed of the mercantiles—" "What did you do with them?" Martin asked. But Raymond rushed on: "Soon as Ninian goes and I'm in full charge, we'll get a more isolated place and run it on a far grander scale. Ostentation—that's the way to live here and now; the richer you are, the more eccentricity you can get away with. And," he added, "I might as well be as comfortable as possible while I suffer through this wretched historical stint." "So Ninian's going," said Martin, wondering why the news made him feel curiously desolate. Because, although he supposed he liked her in a remote kind of way, he had no fondness for her—or she, he knew, for him. "Well, five years is rather a long stretch for any girl to spend in exile," Raymond explained, "even though our life spans are a bit longer than yours. Besides, you're getting too old now to be under petticoat government." He looked inquisitively at Martin. "You're not going to go all weepy and make a scene when she leaves, are you?" "No...." Martin said hesitantly. "Oh, I suppose I will miss her. But we aren't very close, so it won't make a real difference." That was the sad part: he already knew it wouldn't make a difference. Raymond clapped him on the shoulder. "I knew you weren't a sloppy sentimentalist like Conrad. Though you do have rather a look of him, you know." Suddenly that seemed to make Conrad real. Martin felt a vague stirring of alarm. He kept his voice composed, however. "How do you plan to protect me when he comes?" "Well, each one of us is armed to the teeth, of course," Raymond said with modest pride, displaying something that looked like a child's combination spaceman's gun and death ray, but which, Martin had no doubt, was a perfectly genuine—and lethal—weapon. "And we've got a rather elaborate burglar alarm system." Martin inspected the system and made one or two changes in the wiring which, he felt, would increase its efficiency. But still he was dubious. "Maybe it'll work on someone coming from outside this house , but do you think it will work on someone coming from outside this time ?" "Never fear—it has a temporal radius," Raymond replied. "Factory guarantee and all that." "Just to be on the safe side," Martin said, "I think I'd better have one of those guns, too." "A splendid idea!" enthused Raymond. "I was just about to think of that myself!" When it came time for the parting, it was Ninian who cried—tears at her own inadequacy, Martin knew, not of sorrow. He was getting skillful at understanding his descendants, far better than they at understanding him. But then they never really tried. Ninian kissed him wetly on the cheek and said she was sure everything would work out all right and that she'd come see him again. She never did, though, except at the very last. Raymond and Martin moved into a luxurious mansion in a remote area. The site proved a well-chosen one; when the Second Atomic War came, half a dozen years later, they weren't touched. Martin was never sure whether this had been sheer luck or expert planning. Probably luck, because his descendants were exceedingly inept planners. Few people in the world then could afford to live as stylishly as Martin and his guardian. The place not only contained every possible convenience and gadget but was crammed with bibelots and antiques, carefully chosen by Raymond and disputed by Martin, for, to the man from the future, all available artifacts were antiques. Otherwise, Martin accepted his new surroundings. His sense of wonder had become dulled by now and the pink pseudo-Spanish castle—"architecturally dreadful, of course," Raymond had said, "but so hilariously typical"—impressed him far less than had the suburban split-level aquarium. "How about a moat?" Martin suggested when they first came. "It seems to go with a castle." "Do you think a moat could stop Conrad?" Raymond asked, amused. "No," Martin smiled, feeling rather silly, "but it would make the place seem safer somehow." The threat of Conrad was beginning to make him grow more and more nervous. He got Raymond's permission to take two suits of armor that stood in the front hall and present them to a local museum, because several times he fancied he saw them move. He also became an adept with the ray gun and changed the surrounding landscape quite a bit with it, until Raymond warned that this might lead Conrad to them. During those early years, Martin's tutors were exchanged for the higher-degreed ones that were now needful. The question inevitably arose of what the youth's vocation in that life was going to be. At least twenty of the cousins came back through time to hold one of their vigorous family councils. Martin was still young enough to enjoy such occasions, finding them vastly superior to all other forms of entertainment. "This sort of problem wouldn't arise in our day, Martin," Raymond commented as he took his place at the head of the table, "because, unless one specifically feels a call to some profession or other, one just—well, drifts along happily." "Ours is a wonderful world," Grania sighed at Martin. "I only wish we could take you there. I'm sure you would like it." "Don't be a fool, Grania!" Raymond snapped. "Well, Martin, have you made up your mind what you want to be?" Martin affected to think. "A physicist," he said, not without malice. "Or perhaps an engineer." There was a loud, excited chorus of dissent. He chuckled inwardly. "Can't do that," Ives said. "Might pick up some concepts from us. Don't know how; none of us knows a thing about science. But it could happen. Subconscious osmosis, if there is such a thing. That way, you might invent something ahead of time. And the fellow we got the plans from particularly cautioned us against that. Changing history. Dangerous." "Might mess up our time frightfully," Bartholomew contributed, "though, to be perfectly frank, I can't quite understand how." "I am not going to sit down and explain the whole thing to you all over again, Bart!" Raymond said impatiently. "Well, Martin?" "What would you suggest?" Martin asked. "How about becoming a painter? Art is eternal. And quite gentlemanly. Besides, artists are always expected to be either behind or ahead of their times." "Furthermore," Ottillie added, "one more artist couldn't make much difference in history. There were so many of them all through the ages." Martin couldn't hold back his question. "What was I, actually, in that other time?" There was a chilly silence. "Let's not talk about it, dear," Lalage finally said. "Let's just be thankful we've saved you from that !" So drawing teachers were engaged and Martin became a very competent second-rate artist. He knew he would never be able to achieve first rank because, even though he was still so young, his work was almost purely intellectual. The only emotion he seemed able to feel was fear—the ever-present fear that someday he would turn a corridor and walk into a man who looked like him—a man who wanted to kill him for the sake of an ideal. But the fear did not show in Martin's pictures. They were pretty pictures. Cousin Ives—now that Martin was older, he was told to call the descendants cousin —next assumed guardianship. Ives took his responsibilities more seriously than the others did. He even arranged to have Martin's work shown at an art gallery. The paintings received critical approval, but failed to evoke any enthusiasm. The modest sale they enjoyed was mostly to interior decorators. Museums were not interested. "Takes time," Ives tried to reassure him. "One day they'll be buying your pictures, Martin. Wait and see." Ives was the only one of the descendants who seemed to think of Martin as an individual. When his efforts to make contact with the other young man failed, he got worried and decided that what Martin needed was a change of air and scenery. "'Course you can't go on the Grand Tour. Your son hasn't invented space travel yet. But we can go see this world. What's left of it. Tourists always like ruins best, anyway." So he drew on the family's vast future resources and bought a yacht, which Martin christened The Interregnum . They traveled about from sea to ocean and from ocean to sea, touching at various ports and making trips inland. Martin saw the civilized world—mostly in fragments; the nearly intact semi-civilized world and the uncivilized world, much the same as it had been for centuries. It was like visiting an enormous museum; he couldn't seem to identify with his own time any more. The other cousins appeared to find the yacht a congenial head-quarters, largely because they could spend so much time far away from the contemporary inhabitants of the planet and relax and be themselves. So they never moved back to land. Martin spent the rest of his life on The Interregnum . He felt curiously safer from Conrad there, although there was no valid reason why an ocean should stop a traveler through time. More cousins were in residence at once than ever before, because they came for the ocean voyage. They spent most of their time aboard ship, giving each other parties and playing an avant-garde form of shuffleboard and gambling on future sporting events. That last usually ended in a brawl, because one cousin was sure to accuse another of having got advance information about the results. Martin didn't care much for their company and associated with them only when not to have done so would have been palpably rude. And, though they were gregarious young people for the most part, they didn't court his society. He suspected that he made them feel uncomfortable. He rather liked Ives, though. Sometimes the two of them would be alone together; then Ives would tell Martin of the future world he had come from. The picture drawn by Raymond and Ninian had not been entirely accurate, Ives admitted. True, there was no war or poverty on Earth proper, but that was because there were only a couple of million people left on the planet. It was an enclave for the highly privileged, highly interbred aristocracy, to which Martin's descendants belonged by virtue of their distinguished ancestry. "Rather feudal, isn't it?" Martin asked. Ives agreed, adding that the system had, however, been deliberately planned, rather than the result of haphazard natural development. Everything potentially unpleasant, like the mercantiles, had been deported. "Not only natives livin' on the other worlds," Ives said as the two of them stood at the ship's rail, surrounded by the limitless expanse of some ocean or other. "People, too. Mostly lower classes, except for officials and things. With wars and want and suffering," he added regretfully, "same as in your day.... Like now, I mean," he corrected himself. "Maybe it is worse, the way Conrad thinks. More planets for us to make trouble on. Three that were habitable aren't any more. Bombed. Very thorough job." "Oh," Martin murmured, trying to sound shocked, horrified—interested, even. "Sometimes I'm not altogether sure Conrad was wrong," Ives said, after a pause. "Tried to keep us from getting to the stars, hurting the people—I expect you could call them people—there. Still—" he smiled shamefacedly—"couldn't stand by and see my own way of life destroyed, could I?" "I suppose not," Martin said. "Would take moral courage. I don't have it. None of us does, except Conrad, and even he—" Ives looked out over the sea. "Must be a better way out than Conrad's," he said without conviction. "And everything will work out all right in the end. Bound to. No sense to—to anything, if it doesn't." He glanced wistfully at Martin. "I hope so," said Martin. But he couldn't hope; he couldn't feel; he couldn't even seem to care. During all this time, Conrad still did not put in an appearance. Martin had gotten to be such a crack shot with the ray pistol that he almost wished his descendant would show up, so there would be some excitement. But he didn't come. And Martin got to thinking.... He always felt that if any of the cousins could have come to realize the basic flaw in the elaborate plan they had concocted, it would have been Ives. However, when the yacht touched at Tierra del Fuego one bitter winter, Ives took a severe chill. They sent for a doctor from the future—one of the descendants who had been eccentric enough to take a medical degree—but he wasn't able to save Ives. The body was buried in the frozen ground at Ushuaia, on the southern tip of the continent, a hundred years or more before the date of his birth. A great many of the cousins turned up at the simple ceremony. All were dressed in overwhelming black and showed a great deal of grief. Raymond read the burial service, because they didn't dare summon a clerical cousin from the future; they were afraid he might prove rather stuffy about the entire undertaking. "He died for all of us," Raymond concluded his funeral eulogy over Ives, "so his death was not in vain." But Martin disagreed. The ceaseless voyaging began again. The Interregnum voyaged to every ocean and every sea. Some were blue and some green and some dun. After a while, Martin couldn't tell one from another. Cousin after cousin came to watch over him and eventually they were as hard for him to tell apart as the different oceans. All the cousins were young, for, though they came at different times in his life, they had all started out from the same time in theirs. Only the young ones had been included in the venture; they did not trust their elders. As the years went by, Martin began to lose even his detached interest in the land and its doings. Although the yacht frequently touched port for fuel or supplies—it was more economical to purchase them in that era than to have them shipped from the future—he seldom went ashore, and then only at the urging of a newly assigned cousin anxious to see the sights. Most of the time Martin spent in watching the sea—and sometimes he painted it. There seemed to be a depth to his seascapes that his other work lacked. When he was pressed by the current cousin to make a land visit somewhere, he decided to exhibit a few of his sea paintings. That way, he could fool himself into thinking that there was some purpose to this journey. He'd come to believe that perhaps what his life lacked was purpose, and for a while he kept looking for meaning everywhere, to the cousin's utter disgust. "Eat, drink and be merry, or whatever you Romans say when you do as you do," the cousin—who was rather woolly in history; the descendants were scraping bottom now—advised. Martin showed his work in Italy, so that the cousin could be disillusioned by the current crop of Romans. He found that neither purpose nor malice was enough; he was still immeasurably bored. However, a museum bought two of the paintings. Martin thought of Ives and felt an uncomfortable pang of a sensation he could no longer understand. "Where do you suppose Conrad has been all this time?" Martin idly asked the current cousin—who was passing as his nephew by now. The young man jumped, then glanced around him uncomfortably. "Conrad's a very shrewd fellow," he whispered. "He's biding his time—waiting until we're off guard. And then—pow!—he'll attack!" "Oh, I see," Martin said. He had often fancied that Conrad would prove to be the most stimulating member of the whole generation. But it seemed unlikely that he would ever have a chance for a conversation with the young man. More than one conversation, anyhow. "When he does show up, I'll protect you," the cousin vowed, touching his ray gun. "You haven't a thing to worry about." Martin smiled with all the charm he'd had nothing to do but acquire. "I have every confidence in you," he told his descendant. He himself had given up carrying a gun long ago. There was a war in the Northern Hemisphere and so The Interregnum voyaged to southern waters. There was a war in the south and they hid out in the Arctic. All the nations became too drained of power—fuel and man and will—to fight, so there was a sterile peace for a long time. The Interregnum roamed the seas restlessly, with her load of passengers from the future, plus one bored and aging contemporary. She bore big guns now, because of the ever-present danger of pirates.
|
C. as time to refine people.
time to lose something, though he was unable to define it, that was important to society as a whole.
|
How is octave convolution concept extended to multiple resolutions and octaves?
|
### Introduction
Deep convolutional neural networks (CNNs) with 2D convolutions and small kernels BIBREF1, have achieved state-of-the-art results for several speech recognition tasks BIBREF2, BIBREF3, BIBREF4, BIBREF5, BIBREF6. The accuracy of those models grows with their complexity, leading to redundant latent representations. Several approaches have been proposed in the literature to reduce this redundancy BIBREF7, BIBREF8, BIBREF9, BIBREF10, BIBREF11, and therefore to improve their efficiency. Octave convolutional layers BIBREF0 address the problem of spatial redundancy in feature maps by learning feature representations at high and low resolutions. The low resolution processing path increases the size of the receptive field in the original input space, which is a plausible explanation of the improved performance for image classification. We extend the octave convolution concept to multi-scale octave convolutional layers, which include lower resolution feature maps with a higher compression rate (reduction by more than one octave), and the use of more than two feature map tensor groups in order to be learn representations at multiple scales. Multi-scale processing have been previously proposed for a variety of speech recognition tasks BIBREF12, BIBREF13, BIBREF14, BIBREF15, BIBREF16. In deep CNN acoustic models, some of the feature maps may need to represent information which varies at a lower rate, such as the characteristics of the speaker or background noise, compared to the information necessary for phonetic discrimination. Spatial average pooling in a low resolution group of feature maps can be interpreted as a form of low-pass filtering, providing smoothed representations of the observed data, potentially leading to improved performance. We investigate the use of multi-scale octave convolutional layers for robust speech recognition, and attempt to shed more light on the explainability of the models by evaluating the robustness of the learned representations using an affine transformation loss to measure the similarity between clean and noisy encodings. ### Multi-scale octave convolutions
An octave convolutional layer BIBREF0 factorizes the output feature maps of a convolutional layer into two groups. The resolution of the low-frequency feature maps is reduced by an octave – height and width dimensions are divided by 2. In this work, we explore spatial reduction by up to 3 octaves – dividing by $2^t$, where $t=1,2,3$ – and for up to 4 groups. We refer to such a layer as a multi-octave convolutional (MultiOctConv) layer, and an example with three groups and reductions of one and two octaves is depicted in Fig. FIGREF1. In a vanilla CNN the convolutions have the same spatial resolution throughout the network. An octave convolutional (OctConv) layer is divided into high- and low-frequency feature maps and a multi-octave convolutional (MultiOctConv) layer has feature maps reduced by multiple octaves. Let the input feature tensor be $X \in \mathbb {R}^{c_{in} \times h \times w}$, where $c_{in}$ denotes the number of input channels and $h$ and $w$ correspond to the spatial dimensions. In a MultiOctConv layer working at 3 resolutions, $X$ is factorized along the channel dimension into $X = \lbrace X^1, X^2, X^3\rbrace $. The first tensor group tensor, $X^1$, is a representation at the same spatial scale as $X$. The spatial dimensions of the second and third group tensors, $X^2$ and $X^3$, are reduced by one and two octaves respectively. The dimensions of the input tensors $X^1$, $X^2$ and $X^3$ are described in Fig. FIGREF1. The fraction of the channels for each group is denoted with $\alpha _{n} \in [0,1]$, where $\sum _{n=1}^{N} \alpha _{n} = 1$ for $N$ resolution groups in the MultiOctConv layer. For simplicity, we use the same $\alpha _{n}$ for input and output representations within the same scale group. Similarly, the output tensors are also factorized into $Y = \lbrace Y^1, Y^2, Y^3\rbrace $. Their dimensions are analogous to the dimensions of the input tensors and are described in Fig. FIGREF1. To compute $Y^1$, $Y^2$ and $Y^3$ we operate directly on the factorized input tensors $X^1$, $X^2$ and $X^3$. Inter-frequency information update is implemented as a sum of feature maps from different resolution groups. To be able to sum those representations for a desired output scale, the spatial dimensions of the input tensors must be the same. For this reason, two operations are employed: spatial average pooling pool($X, p$) and bilinear interpolation upsample($X, u$), where $p$ is the kernel size and stride for the the 2D pooling layer and $u$ is the upsampling factor. The output MultiOctConv representations are therefore computed as where $f(.)$ is the convolution function and $W^{n_{in}\rightarrow {n_{out}}}\in \mathbb {R}^{c_{in} \times k \times k \times c_{out}}$ is the convolution filter for a $k \times k$ kernel. We call the information update “intra-frequency” when $n_{in} = n_{out}$, and “inter-frequency” when $n_{in} \ne n_{out}$. It is important to note that the convolution $f(.)$ operates on the tensors compressed with average pooling and on the tensors before upsampling, making the design more efficient. The number of parameters in the MultiOctConv layer is the same as in a vanilla convolutional layer. ### Multi-scale octave convolutions ::: Robustness of learned representations
To evaluate the robustness of learned representations, we compare the projections of clean and noisy Aurora-4 samples. The similarity between them is measured using the mean squared error (MSE) loss of an affine projection $y$ of $N$ clean to noisy samples (Eq. DISPLAY_FORM3), to take into account permutations of hidden representations and to ensure invariance of the metric to affine transformations of the encodings. The number of units in layer $y$ and the dimensionality $D$ of $\mathbf {x}_{h}$ is 1024. We use the Aurora-4 test sets and compare clean encodings $\mathbf {x}_{h,clean}$ with noisy encodings $\mathbf {x}_{h,noise}$, obtained as the activations from the last convolutional layer with a forward pass through a trained model. Both hidden representations were obtained for CNN and octave CNN (OctCNN) models in order to compare representations between the models. Also, for intra-model comparison, we evaluate the loss with the encodings from high and low-resolution groups (paths $Y^{1\rightarrow 1}$ and $Y^{2\rightarrow 1}$). This analysis aims to evaluate if the low-resolution groups for noisy samples are indeed more similar to the clean ones than the high-resolution encodings, suggesting more robust representations. We optimize the parameters of $y$ with back-propagation using a fixed number of 3 epochs and we report the validation loss for Aurora-4 test sets. ### Experimental setup
Aurora-4 BIBREF17: We evaluate our models on the simulated multi-condition Aurora-4 dataset, consisting of $\sim $15h of audio for training and $\sim $9h for testing. The test set is divided into 4 subsets: A, B, C, and D. Subset A contains clean-condition recordings, subset B has 6 noise types added to the recordings (car, babble, restaurant, street, airport, train), subset C is recorded with a mismatched microphone, and subset D is recorded with a mismatched microphone and with noise added. In our experiments, we use multi-condition GMM-HMM forced alignments as targets for CNN training. The number of CD states for Aurora-4 is 3422. AMI BIBREF18: AMI contains $\sim $100h of meeting recordings, captured by an independent headset microphone (IHM), single distant microphone (SDM), and multiple distant microphones (MDM), where the mics are combined using the BeamformIt BIBREF19 toolkit. We train our models using the MDM data and evaluate the models for all 3 types of recordings to analyze the effect of mismatched training/testing conditions. We use the suggested train/dev/eval data split BIBREF20, and we evaluate the models on both dev and eval sets. The number of CD states for AMI is 3984. Features: In our experiments, we use 40-dimension mel-scaled filterbank (FBANK) features with {-5,..,5} context for splicing, resulting in a $40\times 11$ input feature map. Models: Our baseline CNN model BIBREF21 consists of 15 convolutional and one fully-connected layer. We use $3\times 3$ kernels throughout the network. We start with 64 output channels in the first layer and double them after 3 and 9 layers. We use batch normalization in every convolutional layer, and ReLU afterwards (unless a reverse order is noted). The initial learning rate is 0.001. We use early stopping for training. ### Results
We present our results in terms of accuracy and robustness on Aurora-4 and AMI, as well as in terms of the computational cost, which is calculated as the number of multiply-accumulate operations (MACCs) performed for a single input feature map. The cost reduction when using octave convolutions stems from reduced dimensions $c_{in}$, $c_{out}$, $h$, and $w$ compared to a vanilla convolutional layer. Aurora-4: Results for Aurora-4 are presented in Table TABREF4. We replace vanilla convolutional layers of our baseline model (CNN) with OctConv and MultiOctConv layers. We first evaluate which layers can be replaced and find that all but the first layer, operating directly on the input representation, should be replaced for the best performance. This approach (L2-L15) is also the least costly. Reducing the ratio of low-resolution representations to 0.125 improves the WER for the mismatched microphone scenario C, but not for all test conditions. Applying batch normalization after ReLU is beneficial for test set C and D. For OctCNN models, the WER for test set D dropped by $\sim 0.4\%$ with a compression by one octave, and by another $\sim 0.4\%$ with a reversed batch normalization and ReLU order. The biggest differences between the MultiOctCNN models can be observed for test set D. The models with the lowest WERs are the ones with a spatial reduction by 2 or 3 octaves, and with 2 or 3 groups. This indicates that multi-scale octave convolutions seem to be an effective, as well as an efficient design for processing speech with background noise and channel mismatch. For MultiOctCNNs, batch normalization after ReLU also gives a performance boost for test set D, with a drop to $13.57\%$. To further evaluate the robustness of the latent representations we measured the MSE between the (projected) representations, described above (Fig. FIGREF5). The loss for the activations at the output of Conv15 ("all") is similar for CNN and OctCNN models for test sets B and C, but lower for test set D for OctCNN, indicating that the learned representations are more robust, contributing to lower WERs. As expected, within-model comparison of the loss show that the representations at low resolution are more similar to the clean encodings from test set A than the ones at high resolution. We believe that this effect improves the robustness of latent representations and results in a decreased WER. AMI: Results for AMI are presented in Table TABREF6. In contrast to the Aurora-4 findings, better performance was achieved with an all OctCNN model (L1-L15). This is an interesting finding, and we believe that the multi-scale processing of the input feature space is beneficial for AMI MDM because of the reverberation in the data. The reverbarated input time$\times $freq representation can be viewed as a spatially redundant one, therefore the OctConv layer applied to the input representation is effective. Unfortunately, the only MultiOctConv model superior to the baseline CNN is the one with 3 groups with a spatial reduction by 1 and 2 octaves. This result indicates that the spatial redundancy for this architecture for AMI MDM is not degrading the performance. However, in terms of the computational cost, we can reduce the #MACCs by a factor of 1.8 with only a small WER increase for a model with 4 resolution groups. ### Conclusions
We have presented multi-scale octave CNN models for robust and efficient speech recognition. We build on Chen et al BIBREF0, applying the method to robust ASR and extending it to multiple resolution groups with a spatial reduction of more than one octave. Our experiments confirm that multi-scale processing of the hidden representations is not only more computationally efficient, but also improves the recognition. Similarity measures between clean and noisy encodings indicates that multi-scale processing in a deep CNN acoustic model improves the robustness of learned representations, especially in the additive noise and mismatched microphone scenario. The gain of the octave convolutions was also observed for AMI MDM data with significant reverberation, when applied to the input feature space. However, the model performance for AMI MDM was not improved with multi-octave convolutions. More careful tuning of the $\alpha $ hyperparameter could improve the results, as it controls the ratio of multi-scale feature maps in the model, enabling both learning of fine-grained representations preserving the details necessary for phonetic discrimination, as well as smoothed, more invariant representations improving the robustness of the model. It would also be possible to set $\alpha $ layer-by-layer to enable the fractions of channels at different resolutions to vary according to the depth of the representation. We proposed a single projection layer MSE loss to measure the affine relationship of clean and noisy hidden representations. With this approach, we evaluated the robustness of the encodings and improved the explainability of our models. More thorough analysis of the representations learned is an interesting future direction. We confirmed that the noisy lower resolution representations are more similar to the clean counterparts than high resolution representations, and thus are more robust. However, we did not investigate the reason for the increased similarity, leaving future work to ascertain if the lower resolution group corresponds to better speaker or noise characteristics, or more invariant phonetic representations. Fig. 1. Multi-octave convolution scheme for 3 resolution groups. Red and green arrows show the connections for the initial and final MultiOctConv layers, respectively. N corresponds to the total number of groups in the MultiOctConv layer (N = 3 in this example). αn is a fraction of channels corresponding to group n. h and w are spatial dimensions. Fig. 2. Proposed method to measure the robustness of learned representations. Fig. 3. MSE affine transformation loss to measure the similarity of ”clean” and ”noisy” encodings (xh,clean and xh,noisy). ”all” corresponds to the output of the last convolutional layer (Conv15), ”high” and ”low” correspond to its Y 1→1 and Y 2→1 branch, respectively. Table 1. WERs [%] for Aurora-4 test sets A, B, C and D for octave and multi-octave CNNs. ”OctConv” column indicates where a Conv layer was replaced with an OctConv or MultiOctConv. 21, 22 and 23 correspond to the factor of spatial dimension reduction. Models with batch normalization after ReLU are denoted by †. Table 2. WERs [%] for models trained on AMI MDM and evaluated on IHM, SDM and MDM conditions. ”OctConv” column indicates where a Conv layer was replaced with an OctConv or MultiOctConv. 21, 22 and 23 correspond to the factor of spatial dimension reduction. Models with batch normalization after ReLU are denoted by †.
|
The resolution of the low-frequency feature maps is reduced by an octave – height and width dimensions are divided by 2. In this work, we explore spatial reduction by up to 3 octaves – dividing by $2^t$, where $t=1,2,3$ – and for up to 4 groups. We refer to such a layer as a multi-octave convolutional (MultiOctConv) layer,
|
What is binary variational dropout?
|
### Introduction
Recurrent neural networks (RNNs) are among the most powerful models for natural language processing, speech recognition, question-answering systems and other problems with sequential data BIBREF2 , BIBREF3 , BIBREF4 , BIBREF5 , BIBREF6 . For complex tasks such as machine translation BIBREF5 or speech recognition BIBREF3 modern RNN architectures incorporate a huge number of parameters. To use these models on portable devices with limited memory, for instance, smartphones, the model compression is desired. High compression level may also lead to an acceleration of RNNs. In addition, compression regularizes RNNs and helps to avoid overfitting. There are a lot of RNNs compression methods based on specific weight matrix representations BIBREF7 , BIBREF8 or sparsification via pruning BIBREF9 . In this paper we focus on RNNs compression via sparsification. Most of the methods from this group are heuristic and require time-consuming hyperparameters tuning. Recently Molchanov et. al. dmolch proposed a principled method based on variational dropout for sparsification of fully connected and convolutional networks. A probabilistic model was described in which parameters controlling sparsity are tuned automatically during neural network training. This model called Sparse Variational Dropout (Sparse VD) leads to extremely sparse solutions without a significant quality drop. However, this technique was not previously investigated for RNNs. In this paper we apply Sparse VD to recurrent neural networks. To take into account the specifics of RNNs we rely on some insights underlined in the paper by Gal & Ghahramani gal where they explain the proper way to use binary dropout in RNNs from the Bayesian point of view. In the experiments we show that LSTMs with Sparse VD yield high sparsity level with just a slight drop in quality. We achieved 99.5% sparsity level on sentiment analysis task and up to 87.6% in character level language modeling experiment. ### Bayesian Neural Networks
Consider a neural network with weights $\omega $ modeling the dependency of the target variables $y=\lbrace y^1, \dots , y^\ell \rbrace $ on the corresponding input objects $X = \lbrace x^1, \dots , x^\ell \rbrace $ . In a Bayesian neural network the weights $\omega $ are treated as random variables. With the prior distribution $p(\omega )$ we search for the posterior distribution $p(\omega |X, y)$ that will help to find expected target value during inference. In the case of neural networks, true posterior is usually intractable but it can be approximated by some parametric distribution $q_\lambda (\omega )$ . The quality of this approximation is measured by the KL-divergence $KL(q_\lambda (\omega )||p(\omega |X, y))$ . The optimal parameter $\lambda $ can be found by maximization of the variational lower bound w.r.t. $\lambda $ : $$\mathcal {L}=\sum _{i=1}^\ell \mathbb {E}_{q_\lambda (\omega )} \log p(y^i|x^i, \omega ) - KL(q_\lambda (\omega )||p(\omega ))$$ (Eq. 2) The expected log-likelihood term in ( 2 ) is usually approximated by Monte-Carlo sampling. To make the MC estimation unbiased, the weights are parametrized by a deterministic function: $\omega = g(\lambda , \xi )$ , where $\xi $ is sampled from some non-parametric distribution (the reparameterization trick BIBREF10 ). The KL-divergence term in ( 2 ) acts as a regularizer and is usually computed or approximated analytically. ### Sparse Variational Dropout
Dropout BIBREF11 is a standard technique for regularization of neural networks. It implies that inputs of each layer are multiplied by a randomly generated noise vector. The elements of this vector are usually sampled from Bernoulli or Gaussian distribution with the parameters tuned using cross-validation. Kingma et al. kingma interpreted Gaussian dropout from a Bayesian perspective that allowed to tune dropout rate automatically during model training. Later this model was extended to sparsify fully connected and convolutional neural networks resulting in a model called Sparse Variational Dropout (Sparse VD) BIBREF0 . Consider one dense layer of a feed-forward neural network with an input of the size $n$ , an output of the size $m$ and a weight matrix $W$ . Following Kingma et al. kingma, in Sparse VD the prior on the weights is a fully factorized log-uniform distribution $p(|w_{ij}|) \propto \frac{1}{|w_{ij}|}$ and the posterior is searched in the form of fully factorized normal distribution: $$q(w_{ij}|m_{ij}, \alpha _{ij}) = \mathcal {N}(m_{ij}, \alpha _{ij} m^2_{ij}).$$ (Eq. 4) Employment of such form of the posterior distribution is equivalent to putting multiplicative BIBREF12 or additive BIBREF0 normal noise on the weights in the following manner: $$w_{ij} = m_{ij} \xi _{ij}, \quad \xi _{ij}\sim \mathcal {N}(1, \alpha _{ij}),$$ (Eq. 5) $$w_{ij} = m_{ij} + \epsilon _{ij}, \quad \epsilon _{ij}\sim \mathcal {N}(0, \sigma ^2_{ij}), \quad \alpha _{ij} = \frac{\sigma ^2_{ij}}{m^2_{ij}}.$$ (Eq. 6) The representation ( 6 ) is called additive reparameterization BIBREF0 . It reduces the variance of the gradients of $\mathcal {L}$ w. r. t. $m_{ij}$ . Moreover, since a sum of normal distributions is a normal distribution with computable parameters, the noise may be applied to the preactivation (input vector times weight matrix $W$ ) instead of $W$ . This trick is called the local reparameterization trick BIBREF13 , BIBREF12 and it reduces the variance of the gradients even further and makes training more efficient. In Sparse VD optimization of the variational lower bound ( 2 ) is performed w. r. t. $\lbrace M, \log \sigma \rbrace $ . The KL-divergence factorizes over the weights and its terms depend only on $\alpha _{ij}$ because of the specific choice of the prior BIBREF12 : $$KL(q(w_{ij}|m_{ij}, \alpha _{ij})||p(w_{ij}))=k(\alpha _{ij}).$$ (Eq. 7) Each term can be approximated as follows BIBREF0 : $${\begin{array}{c}k(\alpha ) \approx 0.64 \sigma (1.87 + 1.49\log \alpha )-\\
\:\:\:\,- 0.5 \log (1 + \alpha ^{-1}) + C.
\end{array}}$$ (Eq. 8) KL-divergence term encourages large values of $\alpha _{ij}$ . If $\alpha _{ij} \rightarrow \infty $ for a weight $w_{ij}$ , the posterior over this weight is a high-variance normal distribution and it is beneficial for model to put $m_{ij} = 0$ as well as $\sigma _{ij}=\alpha _{ij} m^2_{ij}=0$ to avoid inaccurate predictions. As a result, the posterior over $w_{ij}$ approaches zero-centered $\delta $ -function, the weight does not affect the network's output and can be ignored. ### Dropout for Recurrent Neural Networks
Yet another Bayesian model was proposed by Gal & Ghahramani bindrop to explain the binary dropout. On this base, a recipe how to apply a binary dropout to the RNNs properly was proposed by Gal & Ghahramani gal. The recurrent neural network takes a sequence $x = [x_0, \dots , x_T]$ , $x_t\in \mathbb {R}^n$ as an input and maps it into the sequence of hidden states: $${\begin{array}{c}h_{t} = f_h(x_t, h_{t-1}) = g_h(x_{t} W^x + h_{t-1} W^h + b_1)\\
h_i \in \mathbb {R}^m, \, h_0 = \bar{0}
\end{array}}$$ (Eq. 10) Throughout the paper, we assume that the output of the RNN depends only on the last hidden state: $$y = f_y(h_T) = g_y(h_T W^y + b_2).$$ (Eq. 11) Here $g_h$ and $g_y$ are some nonlinear functions. However, all the techniques we discuss further can be easily applied to the more complex setting, e. g. language model with several outputs for one input sequence (one output for each time step). Gal & Ghahramani gal considered RNNs as Bayesian networks. The prior on the recurrent layer weights $\omega =\lbrace W^x, W^h\rbrace $ is a fully factorized standard normal distribution. The posterior is factorized over the rows of weights, and each factor is searched as a mixture of two normal distributions: $
q(w^x_k|m^x_k) = p^x \mathcal {N}(0, \sigma ^2 I) + (1-p^x) \mathcal {N}(m^x_k, \sigma ^2 I),\quad \:
$ $$q(w^h_j|m^h_j) = p^h \mathcal {N}(0, \sigma ^2 I) + (1-p^h) \mathcal {N}(m^h_j, \sigma ^2 I),$$ (Eq. 12) Under assumption $\sigma \approx 0$ sampling the row of weights from such posterior means putting all the weights from this row either to 0 (drop the corresponding input neuron) or to some learned values. Thus this model is a probabilistic analog of binary dropout with dropout rates $p^x$ and $p^h$ . After unfolding the recurrence in the network, the maximization of the variational lower bound for such model looks as follows: $$\sum _{i=1}^\ell \int q(\omega |M) \log \Bigl (y^i\big |f_y\bigl (f_h(x^i_T, f_h(\dots f_h (x^i_1, h^i_0))\bigr )\Bigr ) d \omega - \\
-KL\Bigl (q(\omega |M)\big \Vert p(\omega )\Bigr ) \rightarrow \max _{M}$$ (Eq. 13) Each integral in the first part of ( 13 ) is estimated with MC integration with a single sample $\hat{\omega }_i \sim q(\omega |M)$ . To make this estimation unbiased: (a) the weights sample $\hat{\omega }_i$ should remain the same for all time steps $t=\overline{1, T}$ for a fixed object; (b) dropout rates $p^x$ and $p^h$ should be fixed because the distribution we are sampling from depends on them. The KL-divergence term from ( 13 ) is approximately equivalent to $L_2$ regularization of the variational parameters $M$ . Finally, this probabilistic model leads to the following dropout application in RNNs: we sample a binary mask for the input and hidden neurons, one mask per object for all moments of time, and optimize the $L_2$ -regularized log-likelihood with the dropout rates and the weight of $L_2$ -regularization chosen using cross-validation. Also, the same dropout technique may be applied to forward connections in RNNs, for example in embedding and dense layers BIBREF1 . The same technique can be applied to more complex architectures like LSTM in which the information flow between input and hidden units is controlled by the gate elements: $$i = sigm(h_{t-1}W^h_i + x_t W^x_i) \quad o = sigm(h_{t-1}W^h_o + x_t W^x_o)$$ (Eq. 14) $$f = sigm(h_{t-1}W^h_f + x_t W^x_f) \quad
g = tanh(h_{t-1} W^h_g + x_t W^x_g)$$ (Eq. 15) Here binary dropout masks for input and hidden neurons are generated 4 times: individually for each of the gates $i,o,f$ and input modulation $g$ . ### Variational Dropout for RNN sparsification
Dropout for RNNs proposed by Gal & Ghahramani gal helps to avoid overfitting but is very sensitive to the choice of the dropout rates. On the other hand, Sparse VD allows automatic tuning of the Gaussian dropout parameters individually for each weight which results in the model sparsification. We combine these two techniques to sparsify and regularize RNNs. Following Molchanov et al. dmolch, we use the fully factorized log-uniform prior and approximate the posterior with a fully factorized normal distribution over the weights $\omega =\lbrace W^x, W^h\rbrace $ : $${\begin{array}{c}q(w^x_{ki}|m^x_{ki}, \sigma ^x_{ki}) = \mathcal {N}\bigl (m^x_{ki}, {\sigma ^x_{ki}}^2\bigr ), \:\\
q(w^h_{ji}|m^h_{ji}, \sigma ^h_{ji}) = \mathcal {N}\bigl (m^h_{ji}, {\sigma ^h_{ji}}^2\bigr ), \end{array}}$$ (Eq. 17) where $\sigma ^x_{ki}$ and $\sigma ^h_{ji}$ have the same meaning as in additive reparameterization ( 6 ). To train the model, we maximize the variational lower bound approximation $$\sum _{i=1}^\ell \int q(\omega |M, \sigma ) \log \Bigl (y^i\big |f_y\bigl (f_h(x^i_T, f_h(\dots f_h (x^i_1, h^i_0))\bigr )\Bigr ) d \omega - \\
- \sum _{k,i=1}^{n,m} k\biggl (\frac{{\sigma ^x_{ki}}^2}{{m^x_{ki}}^2}\biggr ) - \sum _{j,i=1}^{m,m} k\biggl (\frac{{\sigma ^h_{ji}}^2}{{m^h_{ji}}^2}\biggr )$$ (Eq. 18) w. r. t. $\lbrace M, \log \sigma \rbrace $ using stochastic mini-batch methods. Here the recurrence in the expected log-likelihood term is unfolded as in ( 13 ) and the KL is approximated using ( 8 ). The integral in ( 18 ) is estimated with a single sample $\hat{\omega }_i \sim q(\omega |M, \alpha )$ per input sequence. We use the reparameterization trick (for unbiased integral estimation) and additive reparameterization (for gradients variance reduction) to sample both input-to-hidden and hidden-to-hidden weight matrices $\widehat{W}^x, \widehat{W}^h$ . To reduce the variance of the gradients and for more computational efficiency we also apply the local reparameterization trick to input-to-hidden matrix $\widehat{W}^x$ moving the noise from the weights to the preactivations: $${\begin{array}{c}(x_t \widehat{W}^x)_j = \sum _{k=1}^n x_{t,k} m^x_{kj} +
\epsilon _j \sqrt{\sum _{k=1}^n x^2_{t,k} {\sigma ^x_{kj}}^2}\:, \\
\epsilon _j \sim \mathcal {N}(0, 1).
\end{array}}$$ (Eq. 19) As a result, only 2-dimensional noise on input-to-hidden connections is required for each mini-batch: we generate one noise vector of length $m$ for each object in a mini-batch. The local reparameterization trick cannot be applied to the hidden-to-hidden matrix $W^h$ . We use the same sample $\widehat{W}^h$ for all moments of time, therefore in the multiplication $h_{t-1} \widehat{W}^h$ the vector $h_{t-1}$ depends on $\widehat{W}^h$ and the rule about the sum of normally distributed random variables cannot be applied. Since usage of 3-dimensional noise (2 dimensions of $\widehat{W}^h$ and a mini-batch size) is too resource-consuming we sample one noise matrix for all objects in a mini-batch for efficiency: $$\hat{w}^h_{ji}=m^h_{ji}+\sigma ^j_{ji}\epsilon ^h_{ji},\quad \epsilon ^h_{ji} \sim \mathcal {N}(0, 1).$$ (Eq. 20) The final framework works as follows: we sample Gaussian additive noise on the input-to-hidden preactivations (one per input sequence) and hidden-to-hidden weight matrix (one per mini-batch), optimize the variational lower bound ( 18 ) w. r. t. $\lbrace M, \log \sigma \rbrace $ , and for many weights we obtain the posterior in the form of a zero-centered $\delta $ -function because the KL-divergence encourages sparsity. These weights can then be safely removed from the model. In LSTM the same prior-posterior pair is consisered for all input-to-hidden and hidden-to-hidden matrices and all computations stay the same. The noise matrices for input-to-hidden and hidden-to-hidden connections are generated individually for each of the gates $i,o,f$ and input modulation $g$ . ### Experiments
We perform experiments with LSTM as the most popular recurrent architecture nowadays. We use Theano BIBREF14 and Lasagne BIBREF15 for implementation. The source code will be available soon at https://github.com/tipt0p/SparseBayesianRNN. We demonstrate the effectiveness of our approach on two diverse problems: Character Level Language Modeling and Sentiment Analysis. Our results show that Sparse Variational Dropout leads to a high level of sparsity in recurrent models without a significant quality drop. We use the dropout technique of Gal & Ghahramani gal as a baseline because it is the most similar dropout technique to our approach and denote it VBD (variational binary dropout). According to Molchanov et al. dmolch, training neural networks with Sparse Variational Dropout from a random initialization is troublesome, as a lot of weights may become pruned away before they could possibly learn something useful from the data. We observe the same effect in our experiments with LSTMs, especially with more complex models. LSTM trained from a random initialization may have high sparsity level, but also have a noticeable quality drop. To overcome this issue we start from pre-trained models that we obtain by training networks without Sparse Variational Dropout for several epochs. Weights in models with Sparse Variational Dropout cannot converge exactly to zero because of the stochastic nature of the training procedure. To obtain sparse networks we explicitly put weights with high corresponding dropout rates to 0 during testing as in Molchanov et al. dmolch. We use the value $\log \alpha = 3$ as a threshold. For all weights that we sparsify using Sparse Variational Dropout, we initialize $\log {\sigma ^2}$ with -6. We optimize our networks using Adam BIBREF16 . Networks without any dropout overfit for both our tasks, therefore, we present results for them with early stopping. Throughout experiments we use the mean values of the weights to evaluate the model quality (we do not sample weights from posterior on the evaluating phase). This is a common practice when working with dropout. ### Sentiment Analysis
Following Gal & Ghahramani gal we evaluated our approach on the sentiment analysis regression task. The dataset is constructed based on Cornell film reviews corpus collected by Pang & Lee regrdata. It consists of approximately 10 thousands non-overlapping segments of 200 words from the reviews. The task is to predict corresponding film scores from 0 to 1. We use the provided train and test partitions. We use networks with one embedding layer of 128 units, one LSTM layer of 128 hidden units, and finally, a fully connected layer applied to the last output of the LSTM (resulting in a scalar output). All weights are initialized in the same way as in Gal & Ghahramani gal. We train our networks using batches of size 128 and a learning rate of 0.001 for 1000 epochs. We also clip the gradients with threshold 0.1. For all layers with VBD we use dropout rate 0.3 and weight decay $10^{-3}$ (these parameters are chosen using cross validation). As a baseline, we train the network without any dropout and with VBD on all layers. In this experiment, our goal is to check the applicability of Sparse VD for recurrent networks, therefore we apply it only to LSTM layer. For embedding and dense layers we use VBD. We try both start training of the network with Sparse VD from random initialization and from two different pre-trained models. The first pre-trained model is obtained after 4 epochs of training of the network without any dropout. The second one is obtained after 200 epochs of training of the network with VBD on all layers. We choose number of pretraining epochs using models quality on cross-validation. The results are shown in Table 1 . In this task our approach achieves extremely high sparsity level both from random initialization and from pre-trained models. Sparse VD networks trained from pre-trained models achieve even better quality than baselines. Note that models already have this sparsity level after approximately 20 epochs. ### Character Level Language Modeling
Following Mikolov et al. mikolov11 we use the Penn Treebank Corpus to train our Language Model (LM). The dataset contains approximately 6 million characters and a vocabulary of 50 characters. We use the provided train, validation and test partitions. We use networks with one LSTM layer of 1000 hidden units to solve the character level LM task. All weight matrices of the networks are initialized orthogonally and all biases are initialized with zeros. Initial values of hidden and cell elements are trainable and also initialized with zeros. We train our networks on non-overlapping sequences of 100 characters in batches of 64 using a learning rate of 0.002 for 50 epochs, and clip gradients with threshold 1. For all layers with VBD we use dropout rate 0.25 and do not use weight decay (these parameters are chosen using quality of VDB model on validation set). As a baseline, we train the network without any dropout and with VBD only on recurrent weights (hidden-to-hidden). Semeniuta et al. semeniuta16 showed that for this particular task applying dropout for feed-forward connections additionally to VBD on recurrent ones does not improve the network quality. We observe the same effect in our experiments. In this experiment we try to sparsify both LSTM and dense layers therefore we apply Sparse VD for all layers. We try both start training of the network with Sparse VD from random initialization and from two different pre-trained models. The first pre-trained model is obtained after 11 epochs of training of the network without any dropout. The second one is obtained after 50 epochs of training of the network with VBD on recurrent connections. We choose the number of pretraining epochs using models quality on validation set. The results are shown in Table 2 . Here we do not achieve such extreme sparsity level as in the previous experiment. This effect may be a consequence of the higher complexity of the task. Also in LM problem we have several outputs for one input sequence (one output for each time step) instead of one output in Sentiment regression. As a result the log-likelihood part of the loss function is much stronger for LM task and regularizer can not sparsify the network so effectively. Here we see that the balance between the likelihood and the regularizer varies a lot for different tasks with RNNs and should be explored futher. Fig. 1 and 2 show the progress of test quality and network sparsity level through the training process. Sparse VD network trained from random initialization underfits and therefore has a slight quality drop in comparison to baseline network without regularization. Sparse VD networks trained from pre-trained models achieve much higher quality but have lower sparsity levels than the one trained from random initialization. Better pretrained models are harder to sparsify. The quality of the model pretrained with VBD drops on the first epoches while the sparsity grows, and the model does not fully recover later. ### Regularization of RNNs
Deep neural networks often suffer from overfitting, and different regularization techniques are used to improve their generalization ability. Dropout BIBREF11 is a popular method of neural networks regularization. The first successful implementations of this method for RNNs BIBREF17 , BIBREF18 applied dropout only for feed-forward connections and not recurrent ones. Introducing dropout in recurrent connections may lead to a better regularization technique but its straightforward implementation may results in underfitting and memory loss through time BIBREF19 . Several ways of dropout application for recurrent connections in LSTM were proposed recently BIBREF20 , BIBREF1 , BIBREF19 . These methods inject binary noise into different parts of LSTM units. Semeniuta et al. semeniuta16 shows that proper implementation of dropout for recurrent connections is important not only for effective regularization but also to avoid vanishing gradients. Bayer et al. bayer13 successfully applied fast dropout BIBREF13 , a deterministic approximation of dropout, to RNNs. Krueger et al. zoneout introduced zoneout which forces some hidden units to maintain their previous values, like in feedforward stochastic depth networks BIBREF21 . ### Compression of RNNs
Reducing RNN size is an important and rapidly developing area of research. One possible concept is to represent large RNN weight matrix by some approximation of the smaller size. For example, Tjandra et. al. tjandra use Tensor Train decomposition of the weight matrices and Le et al. kroneker approximate this matrix with Kronecker product. Hubara et. al itay limit the weights and activations to binary values proposing a way how to compute gradients w. r. t. them. Another concept is to start with a large network and to reduce its size during or after training. The most popular approach here is pruning: the weights of the RNN are cut off on some threshold. Narang et al. pruning choose threshold using several hyperparameters that control the frequency, the rate and the duration of the weights eliminating. ### Discussion and future work
When applying Sparse VD to RNNs we rely on the dropout for RNNs proposed by Gal & Ghahramani gal. The reason is that this dropout technique for RNNs is the closest one to Sparse VD approach. However, there are several other dropout methods for recurrent networks that outperform this baseline BIBREF19 , BIBREF22 . Comparison with them is our future work. Combining Sparse VD with these latest dropout recipes is also an interesting research direction. The challenge here is that the noise should be put on the neurons or gates instead of the weights as in our model. However, there are several recent papers BIBREF23 , BIBREF24 where group sparsity methods are proposed for fully connected and convolutional networks. These methods can be used to solve the underlined problem. The comparison of our approach with other RNN sparsification techniques is still a work-in-progress. It would be interesting to perform this comparison on larger networks, for example, for speech recognition task. One more curious direction of the research is to sparsify not only recurrent layer but an embedding layer too. It may have a lot of parameters in the tasks with large dictionary, such as word based language modeling. ### Acknowledgements
We would like to thank Dmitry Molchanov and Arsenii Ashukha for valuable feedback. Nadezhda Chirkova has been supported by Russian Academic Excellence Project `5-100', and Ekaterina Lobacheva has been supported by Russian Science Foundation grant 17-71-20072. We would also like to thank the Department of Algorithms and Theory of Programming, Faculty of Innovation and High Technology in Moscow Institute of Physics and Technology for provided computational resources. Table 1. Results on sentiment regression task. Prediction quality is reported in MSE (the lower the better). Sparsity levels reported for W x and Wh separately in percents of zero weights. For Sparse VD methods initialization types are reported in brackets. Table 2. Results on character level Language Modelling task. Prediction quality is reported in bits-per-character (lower is better). Sparsity levels reported for W x, Wh and W y separately in percents of zero weights. For Sparse VD methods initialization types are reported in brackets. Figure 1. Results on character level Language Modelling task: prediction quality on the test set in bits-per-character (lower is better). Left — training from random initialization, right — training from pretrained initialization. Stars correspond to pretrained models. For Sparse VD methods initialization types are reported in brackets. Figure 2. Results on character level Language Modelling task: sparsity levels for W y,W x,Wh in LSTM. Initialization types are reported in brackets.
|
the dropout technique of Gal & Ghahramani gal
|
How does the author feel about Princess Mononoke?
A. It is wonderfully strange.
B. It is a world that draws you in and takes your breath away. The only distraction is poor voice casting.
C. It is a powerful vision of the apocalypse.
D. It is technically dazzling.
|
Machines in the Garden In the animated ecological epic Princess Mononoke , the camera travels over landscapes with a clear, steady gaze, like a Zen hang glider. The images have none of the comin'-at-ya pop-surrealism of American cartoons, many of which have characters that spring out of the frame like jack-in-the-boxes. The Japanese director, Hayao Miyazaki, who spent three years on Princess Mononoke and is reported to have done 70 percent of its paintings himself, seems to work from the outside in: to begin with the curve of the earth, then the mossy hills, the watercolor foliage, the nubby stones, the whorls on the wood, the meticulous carvings on a teacup. He captures the texture of light and the currents of air. You could almost settle down in this landscape. A view of nature that some would call "tree-hugging" doesn't feel softheaded when the trees are rendered in such brilliant and robust detail. But then, "soft" is not a word you can apply to Princess Mononoke , however pantheistic its worldview. The film, which is rated PG-13, is full of splattery carnage. If Miyazaki in long shot is contemplative, in close-up he's ferocious. He's both inside and outside the action: He knows when to rock your world and when to induce a state of sorrowful detachment. According to the New York Times , Toy Story animators screened reels of his work when their imaginations flagged, and writers for Star Trek named an alien species after one of his features. Watching Princess Mononoke --which has been dubbed to Disney/Miramax specifications by American and English stars but retains its two-hour-plus length, its gory beheadings, and its grim, near-apocalyptic finale--you can understand their worship. It isn't that Miyazaki's work is technically so dazzling in this age of digitized miracles; it's that everything is sublimely in proportion. The movie has a scope that makes Hollywood's homiletic, follow-your-dream fables look even more solipsistic. Miyazaki is after nothing less than the moment in our history (the film is set in the 14 th and 15 th centuries) when the power shifted from a "natural" world to one shaped by human technology. It's the beginning of what Bill McKibben called "the end of nature"--that is, when nature became no longer an autonomous, self-regulating force but one touched (and, in Miyazaki's view, poisoned) by human industry. The hero, Ashitaka, a warrior from the isolationist Emishi clan, is forced in the first scene to kill a marauding boar--a god turned into a demon (covered in roiling, corrosive worms) by an iron ball lodged in its body. Infected, destined to be consumed by--and to die of--rage, Ashitaka leaves his village in search of the iron ball's source. He discovers a fortress-cum-arms-manufacturing plant called Irontown, presided over by one of the most complex villains in modern film: the regal Lady Eboshi. On one hand, she's a benevolent industrialist who presides over a warmly matriarchal society; on the other, she wants to destroy the forest, harness its resources, and exterminate its animal deities--chiefly the Spirit of the Forest, a magnificent deer god whose touch brings instant life or death, and who transforms at dusk into the towering Night Walker. P rincess Mononoke builds to a full-scale war between humans and the animal kingdom--which does not, by the way, consist of your father's cartoon critters. In fact, the boars and apes have little patience with Ashitaka's call for nature and mankind to live together in harmony; they'd like to eat him. The wolf god, Moro, is slightly more sympathetic, but that's because her adopted "daughter," San (a k a Princess Mononoke), is human. San is first seen sucking a wound of her huge wolf mother, then, as the gore drips from her mouth, training her dark eyes on Ashitaka with feral hatred. Her second appearance--a lone attack on Irontown to assassinate Lady Eboshi--is one of the movie's high points. It's Miyazaki's use of sound--and silence--that takes your breath away: the determined tap of the wolf princess's shoes as she scuttles over the fortress's rooftops; the silence of Eboshi and her army as they stare at this tiny yet formidable tomboy against the black sky. Their battle is so furious that the blades streak and lose definition--it's almost subliminal. It's a shame that the wolf princess warms up to Ashitaka and spends the rest of the film either saving him or being saved by him. She loses that punk-bitch allure. The voice of Claire Danes doesn't help. When Danes says, "I'd do anything to get you humans out of my forest," she sounds like a Valley Girl peeved over lack of parking spaces at the mall. (San needs a more ragged voice--I'd be interested to hear the original Japanese actress.) Billy Crudup is just as Disneyfied (Miramaxed?), but that doesn't hurt as much because Ashitaka is conceived from the start as a rather bland ingénu. Gillian Anderson's growling Moro sounds silly (she doesn't have the breath control), and the fey-hick tones of Billy Bob Thornton are too recognizable as the Akim Tamiroff-like mercenary, Jigo. But Minnie Driver--coming off a triumphantly dizzy Jane in Tarzan --once again provides a voice that the animators deserve. "Bring the strange-ah to me late-ah," she commands in sexy Martian Queen cadences that will stir the loins of Flash Gordon fans everywhere. "I would like to thank him puh-sonally." The overfamiliar voices nudge Princess Mononoke closer to its American counterparts--but not by a lot. There's always something wondrously strange. The "kodamas" are little tree spirits on doughboy bodies. They cock their trapezoidal dice heads and emit a series of clicks; then their heads pop back with a conclusive rattle. Something about them seems just right; I could watch them for hours. (Miyazaki limits their appearances to seconds--he doesn't wear out their mystery the way that, say, George Lucas would.) And no Hollywood animated feature would end with such a powerful vision of apocalypse, as the land is bestridden by a colossus dropping a thick, caustic, tarlike gel that recalls the post-Hiroshima "black rain." Can you take the kids? I think so. As Miyazaki said at a New York Film Festival press conference, "Children understand intuitively that the world they have been born into is not a blessed world." Princess Mononoke , at least, can tell them why. "A special smile ... a certain touch ..." So begins the elevator-music theme song of Music of the Heart ... "I never had a lot that I loved so much." The credits had just started and I was already looking for a barf bag. Did Miramax and director Wes Craven have to work so hard to schlockify the story of Roberta Guaspari (played here by Meryl Streep), whose violin courses in East Harlem elementary schools have become a beacon for such programs nationwide? A fabled taskmaster (her story was told in the 1996 documentary Small Wonders ), Guaspari used music as a way to teach self-discipline--along with the healthy self-respect that follows in its wake. When the New York school board cut the funding for her program, she proved a marvel of self-promotion, attracting features in all the major dailies and ending up along with her best students at Carnegie Hall for a benefit "Fiddlefest"--along with Itzhak Perlman, Isaac Stern, and other legendary "fiddlers." Streep has said that she spent so much of the time on the set learning the violin (she doesn't play any instruments) that she didn't bring the full force of her acting technique to bear on Roberta. Maybe that's why the performance seems so natural. Let her always learn an instrument on the set! Still, she doesn't make much sense of Guaspari. The script, by Pamela Gray ( A Walk on the Moon ), has her students complain of her nastiness and perfectionism, but Streep--who has made herself look dumpy, thick-waisted, and bedraggled--is so busy telegraphing her vulnerability that all we get is dippy niceness. Instead of a monument to an individual's iron will, Music of the Heart becomes the story of a woman so helpless that she arouses the kindness of strangers. Directors of violent genre pieces like Craven (who got this mainstream gig in return for doing the Scream sequels) or Carl Franklin or Sam Raimi sometimes want so badly to belong to Establishment Hollywood--to go to the Academy Awards--that they neuter themselves. Bending over backward to show how sensitive they can be, they forget that violence--even if it's just emotional violence--belongs in "ordinary" dramas, too. Craven does good work with the young actors in the classroom scenes, but the film has a reticence common to most biopics and a mushy, TV-movie humanism that blands out its texture. OK, I was a puddle after some scenes, like the one where Guaspari pushes a student to get her to improve her posture and discovers that the girl is wearing a leg brace. But how much more emotional the Carnegie Hall climax would have been if instead of suddenly seeing these East Harlem kids on stage with Perlman, Stern, Joshua Bell, etc., we'd seen them rehearsing first and struggling to keep up. There's too much music of the heart and not enough music of the callused fingers. In outline, The Limey is a lean little B-movie revenge melodrama about a felonious Brit (Terence Stamp) who's newly sprung from prison and flies to Southern California to get to the bottom of his beautiful daughter's death: "My name's Wilson ... Who dunnit?" The film, directed by Steven Soderbergh, would be worth seeing just for Stamp's performance, at once rock-hard and goofily blinkered, and for Peter Fonda's wittily self-parodic turn as the suspected killer, a music producer who coasts on '60s counterculture easiness while his lackeys do the dirty work. ("Oh, man," he says, the fear finally seeping through the ether. "This is getting all too close to me.") But the picture's glory is its layered and intricate syntax. The dialogue moves ahead--there are great gobs of exposition--but the images continually double back: to Stamp and Lesley Ann Warren, as his daughter's acting teacher, simply gazing at each other; or to Stamp sitting on a plane, remembering his daughter as a girl on the beach, the lens of his home movie camera creating an eerily bright--almost supernatural--spot that dances over her face. The film's most violent act happens well off screen. (You hear the distant "pop-pop-pop-pop-pop" of the hero's gun.) The rest is only half-glimpsed, fantasized, or saturated by memory--or is the present the memory? Is all of The Limey a temporal hiccup? Some, including the critic at Time , have questioned Soderbergh's sanity. (But of course--Soderbergh flouts time!) I see a method to his madness. Less grandiosely than Harmony Korine in Julien Donkey-Boy , Soderbergh pores over every scene in search of its essential dramatic gesture. He's saying: This --not all that other stuff--is what's important. He telegraphs the ending--you know the Limey will somehow be at the root of his daughter's death--but it's still an emotional wow. The climax justifies the technique. It says the point of this odyssey isn't revenge but regret--for irredeemably blown chances and a tragic waste of love. Soderbergh is one of those rare filmmakers who learn on the job. Working within a tight genre structure, he's discovering hundreds of ways of editing a given scene that can give it the richness of a novel. Is he totally successful? No; he misses now and then, which is why the technique sticks out. But what a fantastic effort. See it and weep for what's missing in most other movies.
|
B. It is a world that draws you in and takes your breath away. The only distraction is poor voice casting.
|
How does Ben feel about Laura?
A. Ben loves Laura, but not enough to give up space travel.
B. Ben thinks Laura is the one.
C. Ben likes Laura but they only met 40 days ago. It's not that serious.
D. Ben thinks Laura got too serious, too fast. It's only been 40 days.
|
Spacemen Die at Home By EDWARD W. LUDWIG Illustrated by THORNE [Transcriber's Note: This etext was produced from Galaxy Science Fiction October 1951. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] One man's retreat is another's prison ... and it takes a heap of flying to make a hulk a home! Forty days of heaven and forty nights of hell. That's the way it's been, Laura. But how can I make you understand? How can I tell you what it's like to be young and a man and to dream of reaching the stars? And yet, at the same time, to be filled with a terrible, gnawing fear—a fear locked in my mind during the day and bursting out like an evil jack-in-the-box at night. I must tell you, Laura. Perhaps if I start at the beginning, the very beginning.... It was the Big Day. All the examinations, the physicals and psychos, were over. The Academy, with its great halls and classrooms and laboratories, lay hollow and silent, an exhausted thing at sleep after spawning its first-born. For it was June in this year of 1995, and we were the graduating class of the U. S. Academy of Interplanetary Flight. The first graduating class, Laura. That's why it was so important, because we were the first . We sat on a little platform, twenty-five of us. Below us was a beach of faces, most of them strange, shining like pebbles in the warm New Mexican sunlight. They were the faces of mothers and fathers and grandparents and kid brothers and sisters—the people who a short time ago had been only scrawled names on letters from home or words spoken wistfully at Christmas. They were the memory-people who, to me, had never really existed. But today they had become real, and they were here and looking at us with pride in their eyes. A voice was speaking, deep, sure, resonant. "... these boys have worked hard for six years, and now they're going to do a lot of big things. They're going to bring us the metals and minerals that we desperately need. They're going to find new land for our colonists, good rich land that will bear food and be a home for our children. And perhaps most important of all, they'll make other men think of the stars and look up at them and feel humility—for mankind needs humility." The speaker was Robert Chandler, who'd brought the first rocket down on Mars just five years ago, who'd established the first colony there, and who had just returned from his second hop to Venus. Instead of listening to his words, I was staring at his broad shoulders and his dark, crew-cut hair and his white uniform which was silk-smooth and skin-tight. I was worshiping him and hating him at the same time, for I was thinking: He's already reached Mars and Venus. Let him leave Jupiter and the others alone! Let us be the first to land somewhere! Let us be the first! Mickey Cameron, sitting next to me, dug an elbow into my ribs. "I don't see 'em, Ben," he whispered. "Where do you suppose they are?" I blinked. "Who?" "My folks." That was something I didn't have to worry about. My parents had died in a strato-jet crash when I was four, so I hadn't needed many of those "You are cordially invited" cards. Just one, which I'd sent to Charlie Taggart. Stardust Charlie, we called him, although I never knew why. He was a veteran of Everson's first trip to the Moon nearly twenty-five years ago, and he was still at it. He was Chief Jetman now on the Lunar Lady , a commercial ore ship on a shuttle between Luna City and White Sands. I remembered how, as a kid, I'd pestered him in the Long Island Spaceport, tagging after him like a puppy, and how he'd grown to like me until he became father, mother, and buddy all in one to me. And I remembered, too, how his recommendation had finally made me a cadet. My gaze wandered over the faces, but I couldn't find Charlie's. It wasn't surprising. The Lunar Lady was in White Sands now, but liberties, as Charlie said, were as scarce as water on Mars. It doesn't matter , I told myself. Then Mickey stiffened. "I see 'em, Ben! There in the fifth row!" Usually Mickey was the same whether in a furnace-hot engine room or a garden party, smiling, accepting whatever the world offered. But now a tenseness and an excitement had gripped even him. I was grateful that he was beside me; we'd been a good team during those final months at the Academy and I knew we'd be a good team in space. The Universe was mighty big, but with two of us to face it together, it would be only half as big. And then it seemed that all the proud faces were looking at us as if we were gods. A shiver went through my body. Though it was daytime, I saw the stars in my mind's vision, the great shining balls of silver, each like a voice crying out and pleading to be explored, to be touched by the sons of Earth. They expect a lot from us. They expect us to make a new kind of civilization and a better place out of Earth. They expect all this and a hell of a lot more. They think there's nothing we can't do. I felt very small and very humble. I was scared. Damned scared. At last it was over, and the proud faces descended upon us in a huge, babbling wave. Then I saw him. Good old Stardust Charlie. His wizened little body was shuffling down an aisle, his eyes shining like a child's. He'd been sandwiched, evidently, in one of the rear rows. But he wasn't the Charlie I'd seen a year ago. He'd become gaunt and old, and he walked with an unnatural stiffness. He looked so old that it was hard to believe he'd once been young. He scratched his mop of steel-gray hair and grinned. "You made it, boy," he chortled, "and by Jupiter, we'll celebrate tonight. Yes, siree, I got twenty-four hours, and we'll celebrate as good spacemen should!" Then Mickey strode up to us. He was his normal, boyish self again, walking lightly, his blond, curly-haired skull swaying as if in rhythm with some silent melody. And you, Laura, were with him. "Meet the Brat," he said. "My sister Laura." I stared almost rudely. You were like a doll lost in the immensity of your fluffy pink dress. Your hair was long and transformed into a golden froth where sunlight touched it. But your eyes were the eyes of a woman, glowing like dark stars and reflecting a softness, a gentleness that I'd never seen in eyes before. "I'm happy to meet you, Ben," you said. "I've heard of no one else for the past year." A tide of heat crept up from my collar. I stuttered through an introduction of Charlie. You and Mickey looked strangely at Charlie, and I realized that old Stardust was not a cadet's notion of the ideal spaceman. Charlie scorned the skin-tight uniforms of the government service and wore a shiny black suit that was a relic of Everson's early-day Moon Patrol. His tie was clumsily knotted, and a button on his coat was missing. And the left side of his face was streaked with dark scar tissue, the result of an atomic blowup on one of the old Moon ships. I was so accustomed to the scars, I was seldom aware of them; but others, I knew, would find them ugly. You were kind. You shook hands and said, softly: "It's a privilege to meet you, Charlie. Just think—one of Everson's men, one of the first to reach the Moon!" Charlie gulped helplessly, and Mickey said: "Still going to spend the weekend with us, aren't you, Ben?" I shook my head. "Charlie has only twenty-four hours liberty. We're planning to see the town tonight." "Why don't you both come with us?" you asked. "Our folks have their own plane, so it would be no problem. And we've got a big guest room. Charlie, wouldn't you like a home-cooked meal before going back to the Moon?" Charlie's answer was obscured by a sudden burst of coughing. I knew that he'd infinitely prefer to spend his liberty sampling Martian fizzes and Plutonian zombies. But this night seemed too sacred for Charlie's kind of celebration. "We'd really like to come," I said. On our way to the 'copter parking field, Dean Dawson passed us. He was a tall, willowy man, spectacled, looking the way an academy professor should look. "Ben," he called, "don't forget that offer. Remember you've got two months to decide." "No, thanks," I answered. "Better not count on me." A moment later Mickey said, frowning, "What was he talking about, Ben? Did he make you an offer?" I laughed. "He offered me a job here at the Academy teaching astrogation. What a life that would be! Imagine standing in a classroom for forty years when I've got the chance to—" I hesitated, and you supplied the right words: "When you've got the chance to be the first to reach a new planet. That's what most of you want, isn't it? That's what Mickey used to want." I looked at you as if you were Everson himself, because you seemed to understand the hunger that could lie in a man's heart. Then your last words came back and jabbed me: "That's what Mickey used to want." " Used to want?" I asked. "What do you mean?" You bit your lip, not answering. "What did she mean, Mickey?" Mickey looked down at his feet. "I didn't want to tell you yet, Ben. We've been together a long time, planning to be on a rocket. But—" "Yes?" "Well, what does it add up to? You become a spaceman and wear a pretty uniform. You wade through the sands of Mars and the dust of Venus. If you're lucky, you're good for five, maybe ten years. Then one thing or another gets you. They don't insure rocketmen, you know." My stomach was full of churning, biting ice. "What are you trying to say, Mickey?" "I've thought about it a long time. They want me for Cargo Supervisor of White Sands Port." He raised his hand to stop me. "I know. It's not so exciting. I'll just live a lot longer. I'm sorry, Ben." I couldn't answer. It was as if someone had whacked the back of my knees with the blast of a jet. "It doesn't change anything, Ben—right now, I mean. We can still have a good weekend." Charlie was muttering under his breath, smoldering like a bomb about to reach critical mass. I shook my head dazedly at him as we got to the 'copter. "Sure," I said to Mickey, "we can still have a good weekend." I liked your folks, Laura. There was no star-hunger in them, of course. They were simple and solid and settled, like green growing things, deep-rooted, belonging to Earth. They were content with a home that was cool on this warm summer night, with a 'copter and a tri-dimensional video, and a handsome automatic home that needed no servants or housework. Stardust Charlie was as comfortable as a Martian sand-monkey in a shower, but he tried courageously to be himself. At the dinner table he stared glassily at nothing and grated, "Only hit Mars once, but I'll never forget the kid who called himself a medic. Skipper started coughing, kept it up for three days. Whoopin' cough, the medic says, not knowin' the air had chemicals that turned to acid in your lungs. I'd never been to Mars before, but I knew better'n that. Hell, I says, that ain't whoopin' cough, that's lung-rot." That was when your father said he wasn't so hungry after all. Afterward, you and I walked onto the terrace, into the moonlit night, to watch for crimson-tailed continental rockets that occasionally streaked up from White Sands. We gazed for a few seconds up into the dark sky, and then you said: "Charlie is funny, isn't he? He's nice and I'm glad he's here, but he's sort of funny." "He's an old-time spaceman. You didn't need much education in those days, just a lot of brawn and a quick mind. It took guts to be a spaceman then." "But he wasn't always a spaceman. Didn't he ever have a family?" I smiled and shook my head. "If he had, he never mentioned it. Charlie doesn't like to be sentimental, at least not on the outside. As far as I know, his life began when he took off for the Moon with Everson." You stared at me strangely, almost in a sacred kind of way. I knew suddenly that you liked me, and my heart began to beat faster. There was silence. You were lovely, your soft hair like strands of gold, and there were flecks of silver in your dark eyes. Somehow I was afraid. I had the feeling that I shouldn't have come here. You kept looking at me until I had to ask: "What are you thinking, Laura?" You laughed, but it was a sad, fearful laugh. "No, I shouldn't be thinking it. You'd hate me if I told you, and I wouldn't want that." "I could never hate you." "It—it's about the stars," you said very softly. "I understand why you want to go to them. Mickey and I used to dream about them when we were kids. Of course I was a girl, so it was just a game to me. But once I dreamed of going to England. Oh, it was going to be so wonderful. I lived for months, just thinking about it. "One summer we went. I had fun. I saw the old buildings and castles, and the spaceports and the Channel Tube. But after it was over, I realized England wasn't so different from America. Places seem exciting before you get to them, and afterward they're not really." I frowned. "And you mean it might be the same with the stars? You think maybe I haven't grown up yet?" Anxiety darkened your features. "No, it'd be good to be a spaceman, to see the strange places and make history. But is it worth it? Is it worth the things you'd have to give up?" I didn't understand at first, and I wanted to ask, "Give up what ?" Then I looked at you and the promise in your eyes, and I knew. All through the years I'd been walking down a single, narrow path. Government boarding school, the Academy, my eyes always upward and on the stars. Now I'd stumbled into a cross-roads, beholding a strange new path that I'd never noticed before. You can go into space , I thought, and try to do as much living in ten years as normal men do in fifty. You can be like Everson, who died in a Moon crash at the age of 36, or like a thousand others who lie buried in Martian sand and Venusian dust. Or, if you're lucky, like Charlie—a kind of human meteor streaking through space, eternally alone, never finding a home. Or there's the other path. To stay on this little prison of an Earth in cool, comfortable houses. To be one of the solid, rooted people with a wife and kids. To be one of the people who live long enough to grow old, who awake to the song of birds instead of rocket grumblings, who fill their lungs with the clean rich air of Earth instead of poisonous dust. "I'm sorry," you said. "I didn't mean to make you sad, Ben." "It's all right," I said, clenching my fists. "You made sense—a lot of sense." The next morning Charlie said good-bye in our room. He rubbed his scarred face nervously as he cleared his throat with a series of thin, tight coughs. Then he pointed to a brown, faded tin box lying on the bed. "I'm leavin' that for you. It's full of old stuff, souvenirs mostly. Thought maybe you'd like to have 'em." I scowled, not understanding. "Why, Charlie? What for?" He shrugged as if afraid he might be accused of sentimentality. "Oh, it's just that I've been dodgin' meteors now for twenty-five years. That's a long time, boy. Ain't one spaceman in a thousand that lucky. Some of these days, I won't be so lucky." I tried to laugh. "You're good for another twenty-five years, Charlie." He shook his head stiffly, staring at nothing. "Maybe. Anyway, I'm gonna get off the Shuttle this time, make one more trip to Mars. Tell you what. There's a little stone cafe on Mars, the Space Rat , just off Chandler Field on the Grand Canal. When you get to Mars, take a look inside. I'll probably be there." He coughed again, a deep, rasping cough that filled his eyes with tears. "Not used to this Earth air," he muttered. "What I need's some Martian climate." Suddenly that cough frightened me. It didn't seem normal. I wondered, too, about his stiff movements and glassy stare. It was as if he were drugged. I shook the thought away. If Charlie was sick, he wouldn't talk about going to Mars. The medics wouldn't let him go even as far as Luna. We watched him leave, you and Mickey and I. "When will you be back?" you asked. Charlie's hard face contorted itself into a gargoylish grin. "Maybe a couple of months, maybe a couple of years. You know spacemen." Then he waved and strode away, a strange, gray, withered gnome of a man. I wanted him to say something, to tell me the secret that would kill the doubt worming through my brain. But he rounded a corner, still grinning and waving, and then he was gone. That afternoon Mickey showed me his room. It was more like a boy's room than a spaceman's. In it were all the little things that kids treasure—pennants, models of Everson's two ships, a tennis trophy, books, a home-made video. I began to realize how important a room like this could be to a boy. I could imagine, too, the happiness that parents felt as they watched their children grow to adulthood. I'd missed something. My folks were shadow-people, my impressions of them drawn half from ancient photos, half from imagination. For me, it had been a cold, automatic kind of life, the life of dormitories and routines and rules. I'd been so blinded by the brilliancy of my dreams, I hadn't realized I was different. My folks were killed in a rocket crash. If it weren't for rockets, I'd have lived the kind of life a kid should live. Mickey noticed my frown. "What's the matter, Ben? Still sore? I feel like a heel, but I'm just not like you and Charlie, I guess. I—" "No, I understand, Mickey. I'm not sore, really." "Listen, then. You haven't accepted any offer yet, have you?" "No. I got a couple of possibilities. Could get a berth on the Odyssey , the new ship being finished at Los Angeles. They want me, too, for the Moon Patrol, but that's old stuff, not much better than teaching. I want to be in deep space." "Well, how about staying with us till you decide? Might as well enjoy Earth life while you can. Okay?" I felt like running from the house, to forget that it existed. I wanted someone to tell me one of the old stories about space, a tale of courage that would put fuel on dying dreams. But I wanted, also, to be with you, Laura, to see your smile and the flecks of silver in your eyes and the way your nose turned upward ever so slightly when you laughed. You see, I loved you already, almost as much as I loved the stars. And I said, slowly, my voice sounding unfamiliar and far away, "Sure, I'll stay, Mickey. Sure." Forty days of joy, forty nights of fear and indecision. We did all the little things, like watching the rockets land at White Sands and flying down to the Gulf to swim in cool waters. You tried, unsuccessfully, to teach me to dance, and we talked about Everson and Charlie and the Moon and the stars. You felt you had to give the stars all the beauty and promise of a child's dream, because you knew that was what I wanted. One morning I thought, Why must I make a choice? Why can't I have both you and the stars? Would that be asking too much? All day the thought lay in my mind like fire. That evening I asked you to marry me. I said it very simply: "Laura, I want you to be my wife." You looked up at Venus, and you were silent for a long while, your face flushed. Then you murmured, "I—I want to marry you, Ben, but are you asking me to marry a spaceman or a teacher?" "Can't a spaceman marry, too?" "Yes, a spaceman can marry, but what would it be like? Don't you see, Ben? You'd be like Charlie. Gone for maybe two months, maybe two years. Then you'd have a twenty-four hour liberty—and I'd have what?" Somehow I'd expected words like these, but still they hurt. "I wouldn't have to be a spaceman forever. I could try it for a couple of years, then teach." "Would you, Ben? Would you be satisfied with just seeing Mars? Wouldn't you want to go on to Jupiter and Saturn and Uranus and on and on?" Your voice was choked, and even in the semi-darkness I saw tears glittering in your eyes. "Do you think I'd dare have children, Ben? Mickey told me what happened on the Cyclops . There was a leak in the atomic engines. The ship was flooded with radiation—just for a second. It didn't seem serious. The men had no burns. But a year later the captain had a child. And it was—" "I know, Laura. Don't say it." You had to finish. "It was a monster." That night I lay awake, the fears and doubts too frantic to let me sleep. You've got to decide now , I told myself. You can't stay here. You've got to make a choice. The teaching job was still open. The spot on the Odyssey was still open—and the big ship, it was rumored, was equipped to make it all the way to Pluto. You can take Dean Dawson's job and stay with Laura and have kids and a home and live to see what happens in this world sixty years from now. Or you can see what's on the other side of the mountain. You can be a line in a history book. I cursed. I knew what Charlie would say. He'd say, "Get the hell out of there, boy. Don't let a fool woman make a sucker out of you. Get out there on the Odyssey where you belong. We got a date on Mars, remember? At the Space Rat , just off Chandler Field on the Grand Canal." That's what he'd say. And yet I wanted you, Laura. I wanted to be with you, always. "Oh God," I moaned, "what shall I do?" Next morning the door chimes pealed, and you went to the door and brought back the audiogram. It was addressed to me; I wondered who could be sending me a message. I pressed the stud on the little gray cylinder, and a rasping, automatic voice droned: "Luna City, Luna, July 27, 1995. Regret to inform you of death of Charles Taggart, Chief Jetman...." Then there was a Latin name which was more polite than the word "lung-rot" and the metallic phrase, "This message brought to you by courtesy of United Nations Earth-Luna Communication Corps." I stood staring at the cylinder. Charles Taggart was dead. Charles Taggart was Charlie. Stardust Charlie. My heart thudded crazily against my chest. It couldn't be! Not Charlie! The audiogram had lied! I pressed the stud again. "... regret to inform you of death of Charles ..." I hurled the cylinder at the wall. It thudded, fell, rolled. The broken voice droned on. You ran to it, shut it off. "I'm sorry, Ben, so terribly—" Without answering, I walked into my room. I knew it was true now. I remembered Charlie's coughing, his gaunt features, his drugged gaze. The metallic words had told the truth. I sat for a long time on my bed, crying inside, but staring dry-eyed at Charlie's faded tin box. Then, finally, I fingered his meager possessions—a few wrinkled photos, some letters, a small black statue of a forgotten Martian god, a gold service medal from the Moon Patrol. This was what remained of Charlie after twenty-five years in space. It was a bitter bargain. A statue instead of a wife, yellowed letters instead of children, a medal instead of a home. It'd be a great future , I thought. You'd dream of sitting in a dingy stone dive on the Grand Canal with sand-wasps buzzing around smoky, stinking candles. A bottle of luchu juice and a couple of Martian girls with dirty feet for company. And a sudden cough that would be the first sign of lung-rot. To hell with it! I walked into your living room and called Dean Dawson on the visiphone. I accepted that job teaching. And now, Laura, it's nearly midnight. You're in your room, sleeping, and the house is silent. It's hard to tell you, to make you understand, and that is why I am writing this. I looked through Charlie's box again, more carefully this time, reading the old letters and studying the photographs. I believe now that Charlie sensed my indecision, that he left these things so that they could tell me what he could not express in words. And among the things, Laura, I found a ring. A wedding ring. In that past he never talked about, there was a woman—his wife. Charlie was young once, his eyes full of dreams, and he faced the same decision that I am facing. Two paths were before him, but he tried to travel both. He later learned what we already know—that there can be no compromise. And you know, too, which path he finally chose. Do you know why he had to drug himself to watch me graduate? So he could look at me, knowing that I would see the worlds he could never live to see. Charlie didn't leave just a few trinkets behind him. He left himself, Laura, for he showed me that a boy's dream can also be a man's dream. He made his last trip to Luna when he knew he was going to die. Heaven knows how he escaped a checkup. Maybe the captain understood and was kind—but that doesn't matter now. Do you know why he wanted to reach Mars? Do you know why he didn't want to die in the clean, cool air of Earth? It was because he wanted to die nearer home. His home, Laura, was the Universe, where the ship was his house, the crew his father, mother, brothers, the planets his children. You say that the beauty of the other side of the mountain vanishes after you reach it. But how can one ever be sure until the journey is made? Could I or Charlie or the thousand before us bear to look upon a star and think, I might have gone there; I could have been the first ? We said, too, that the life of a spaceman is lonely. Yet how could one be lonely when men like Charlie roam the spaceways? Charlie wanted me to himself that night after graduation. He wanted us to celebrate as spacemen should, for he knew that this would be his last night on Earth. It might have seemed an ugly kind of celebration to you, but he wanted it with all his heart, and we robbed him of it. Because of these things, Laura, I will be gone in the morning. Explain the best you can to Mickey and to your parents and Dean Dawson. Right now I've got a date that I'm going to keep—at a dingy stone cafe on Mars, the Space Rat , just off Chandler Field on the Grand Canal. Stardust Charlie will be there; he'll go with me in memory to whatever part of the Galaxy I may live to reach. And so will you, Laura. I have two wedding rings with me—his wife's ring and yours.
|
A. Ben loves Laura, but not enough to give up space travel.
|
What metrics are used to measure performance of models?
|
### Introduction
Task-oriented dialogue system is an important tool to build personal virtual assistants, which can help users to complete most of the daily tasks by interacting with devices via natural language. It's attracting increasing attention of researchers, and lots of works have been proposed in this area BIBREF0, BIBREF1, BIBREF2, BIBREF3, BIBREF4, BIBREF5, BIBREF6, BIBREF7. The existing task-oriented dialogue systems usually consist of four components: (1) natural language understanding (NLU), it tries to identify the intent of a user; (2) dialogue state tracker (DST), it keeps the track of user goals and constraints in every turn; (3) dialogue policy maker (DP), it aims to generate the next available dialogue action; and (4) natural language generator (NLG), it generates a natural language response based on the dialogue action. Among the four components, dialogue policy maker plays a key role in order to complete dialogues effectively, because it decides the next dialogue action to be executed. As far as we know, the dialogue policy makers in most existing task-oriented dialogue systems just use the classifiers of the predefined acts to obtain dialogue policy BIBREF0, BIBREF2, BIBREF4, BIBREF8, BIBREF9. The classification-based dialogue policy learning methods can assign either only a dialogue act and its corresponding parameters BIBREF10, BIBREF2, BIBREF0 or multiple dialogue acts without their corresponding parameters for a dialogue action BIBREF11. However, all these existing methods cannot obtain multiple dialogue acts and their corresponding parameters for a dialogue action at the same time. Intuitively, it will be more reasonable to construct multiple dialogue acts and their corresponding parameters for a dialogue action at the same time. For example, it can be shown that there are 49.4% of turns in the DSTC2 dataset and 61.5% of turns in the Maluuba dataset have multiple dialogue acts and their corresponding parameters as the dialogue action. If multiple dialogue acts and their corresponding parameters can be obtained at the same time, the final response of task-oriented dialogue systems will become more accurate and effective. For example, as shown in Figure FIGREF3, a user wants to get the name of a cheap french restaurant. The correct dialogue policy should generate three acts in current dialogue turn: offer(name=name_slot), inform(food=french) and inform(food=cheap). Thus, the user's real thought may be: “name_slot is a cheap french restaurant". If losing the act offer, the system may generate a response like “There are some french restaurants", which will be far from the user's goal. To address this challenge, we propose a Generative Dialogue Policy model (GDP) by casting the dialogue policy learning problem as a sequence optimization problem. The proposed model generates a series of acts and their corresponding parameters by the learned dialogue policy. Specifically, our proposed model uses a recurrent neural network (RNN) as action decoder to construct dialogue policy maker instead of traditional classifiers. Attention mechanism is used to help the decoder decode dialogue acts and their corresponding parameters, and then the template-based natural language generator uses the results of the dialogue policy maker to choose an appropriate sentence template as the final response to the user. Extensive experiments conducted on two benchmark datasets verify the effectiveness of our proposed method. Our contributions in this work are three-fold. The existing methods cannot construct multiple dialogue acts and their corresponding parameters at the same time. In this paper, We propose a novel generative dialogue policy model to solve the problem. The extensive experiments demonstrate that the proposed model significantly outperforms the state-of-the-art baselines on two benchmarks. We publicly release the source code. ### Related Work
Usually, the existing task-oriented dialogue systems use a pipeline of four separate modules: natural language understanding, dialogue belief tracker, dialogue policy and natural language generator. Among these four modules, dialogue policy maker plays a key role in task-oriented dialogue systems, which generates the next dialogue action. As far as we know, nearly all the existing approaches obtain the dialogue policy by using the classifiers of all predefined dialogue acts BIBREF12, BIBREF13. There are usually two kinds of dialogue policy learning methods. One constructs a dialogue act and its corresponding parameters for a dialogue action. For example, BIBREF0 constructs a simple classifier for all the predefined dialogue acts. BIBREF2 build a complex classifier for some predefined dialogue acts, addtionally BIBREF2 adds two acts for each parameter: one to inform its value and the other to request it. The other obtains the dialogue policy by using multi-label classification to consider multiple dialogue acts without their parameters. BIBREF11 performs multi-label multi-class classification for dialogue policy learning and then the multiple acts can be decided based on a threshold. Based on these classifiers, the reinforcement learning can be used to further update the dialogue policy of task-oriented dialogue systems BIBREF3, BIBREF14, BIBREF9. In the real scene, an correct dialogue action usually consists of multiple dialogue acts and their corresponding parameters. However, it is very hard for existing classification-based dialogue policy maker to achieve this goal. Thus, in this paper we propose a novel generative dialogue policy maker to address this issue by casting the dialogue policy learning problem as a sequence optimization problem. ### Technical Background ::: Encoder-Decoder Seq2Seq Models
Seq2Seq model was first introduced by BIBREF15 for statistical machine translation. It uses two recurrent neural networks (RNN) to solve the sequence-to-sequence mapping problem. One called encoder encodes the user utterance into a dense vector representing its semantics, the other called decoder decodes this vector to the target sentence. Now Seq2Seq framework has already been used in task-oriented dialog systems such as BIBREF4 and BIBREF1, and shows the challenging performance. In the Seq2Seq model, given the user utterance $Q=(x_1, x_2, ..., x_n)$, the encoder squeezes it into a context vector $C$ and then used by decoder to generate the response $R=(y_1, y_2, ..., y_m)$ word by word by maximizing the generation probability of $R$ conditioned on $Q$. The objective function of Seq2Seq can be written as: In particular, the encoder RNN produces the context vector $C$ by doing calculation below: The $h_t$ is the hidden state of the encoder RNN at time step $t$ and $f$ is the non-linear transformation which can be a long-short term memory unit LSTM BIBREF16 or a gated recurrent unit GRU BIBREF15. In this paper, we implement $f$ by using GRU. The decoder RNN generates each word in reply conditioned on the context vector $C$. The probability distribution of candidate words at every time step $t$ is calculated as: The $s_t$ is the hidden state of decoder RNN at time step $t$ and $y_{t-1}$ is the generated word in the reply at time $t-1$ calculated by softmax operations. ### Technical Background ::: Attention Mechanism
Attention mechanisms BIBREF17 have been proved to improved effectively the generation quality for the Seq2Seq framework. In Seq2Seq with attention, each $y_i$ corresponds to a context vector $C_i$ which is calculated dynamically. It is a weighted average of all hidden states of the encoder RNN. Formally, $C_i$ is defined as $C_i=\sum _{j=1}^{n} \alpha _{ij}h_j$, where $\alpha _{ij}$ is given by: where $s_{i-1}$ is the last hidden state of the decoder, the $\eta $ is often implemented as a multi-layer-perceptron (MLP) with tanh as the activation function. ### Generative Dialogue Policy
Figure FIGREF13 shows the overall system architecture of the proposed GDP model. Our model contains five main components: (1) utterance encoder; (2) dialogue belief tracker; (3) dialogue policy maker; (4) knowledge base; (5) template-based natural language generator. Next, we will describe each component of our proposed GDP model in detail. ### Generative Dialogue Policy ::: Notations and Task Formulation
Given the user utterance $U_t$ at turn $t$ and the dialogue context $C_{t-1}$ which contains the result of the dialogue belief tracker at turn $t-1$, the task-oriented dialog system needs to generate user's intents $C_t$ by dialogue belief tracker and then uses this information to get the knowledge base query result $k_t \in \mathbb {R}^k$. Then the model needs to generate the next dialogue action $A_t$ based on $k_t$, $U_t$ and $C_t$. The natural language generator provides the template-based response $R_t$ as the final reply by using $A_t$. The $U_t$ and $C_t$ are the sequences, $k_t$ is a one-hot vector representing the number of the query results. For baselines, in this paper, the $A_t$ is the classification result of the next dialogue action, but in our proposed model it's a sequence which contains multiple acts and their corresponding parameters. ### Generative Dialogue Policy ::: Utterance Encoder
A bidirectional GRU is used to encode the user utterance $U_t$, the last turn response $R_{t-1}$ made by the system and the dialogue context $C_{t-1}$ into a continuous representation. The vector is generated by concatenating the last forward and backward GRU states. $U_t = (w_1, w_2, ..., w_{T_m})$ is the user utterance at turn $t$. $C_{t-1}=(c_1, c_2, ..., c_{T_n})$ is the dialogue context made by dialogue belief tracker at $t-1$ turn. $R_{t-1}$ is the response made by our task-oriented dialogue system at last turn. Then the words of $[C_{t-1}, R_{t-1}, U_t]$ are firstly mapped into an embedding space and further serve as the inputs of each step to the bidirectional GRU. Let $n$ denotes the number of words in the sequence $[C_{t-1}, R_{t-1}, U_t]$. The $\overrightarrow{h_{t^{\prime }}^u}$ and $\overleftarrow{h_{t^{\prime }}^u}$ represent the forward and backward GRU state outputs at time step $t^{\prime }$. The encoder output of timestep $i$ denote as $\overline{h_i^u}$. where $e([C_{t-1}, R_{t-1}, U_t])$ is the embedding of the input sequence, $d_h$ is the hidden size of the GRU. $H_u$ contains the encoder hidden state of each timestep, which will be used by attention mechanism in dialogue policy maker. ### Generative Dialogue Policy ::: Dialogue State Tracker
Dialogue state tracker maintains the state of a conversation and collects the user's goals during the dialogue. Recent work successfully represents this component as discriminative classifiers. BIBREF5 verified that the generation is a better way to model the dialogue state tracker. Specifically, we use a GRU as the generator to decode the $C_t$ of current turn. In order to capture user intent information accurately, the basic attention mechanism is calculated when the decoder decodes the $C_t$ at each step, which is the same as the Eq. (DISPLAY_FORM12). where $m$ is the length of $C_t$, $e(y_i)$ is the embedding of the token, $d_h$ is the hidden size of the GRU and the hidden state at $i$ timestep of the RNN in dialogue state tracker denote as $h_i^d$. The decoded token at step $i$ denotes as $y_i^d$. ### Generative Dialogue Policy ::: Knowledge Base
Knowledge base is a database that stores information about the related task. For example, in the restaurant reservation, a knowledge base stores the information of all the restaurants, such as location and price. After dialogue belief tracker, the $C_t$ will be used as the constraints to search the results in knowledge base. Then the one-hot vector $k_t$ will be produced when the system gets the number of the results. The search result $k_t$ has a great influence on dialogue policy. For example, if the result has multiple matches, the system should request more constraints of the user. In practice, let $k_t$ be an one-hot vector of 20 dimensions to represent the number of query results. Then $k_t$ will be used as the cue for dialogue policy maker. ### Generative Dialogue Policy ::: Dialogue Policy Maker
In task-oriented dialogue systems, supervised classification is a straightforward solution for dialogue policy modeling. However, we observe that classification cannot hold enough information for dialogue policy modeling. The generative approach is another way to model the dialogue policy maker for task-oriented dialogue systems, which generates the next dialogue acts and their corresponding parameters based on the dialogue context word by word. Thus the generative approach converts the dialogue policy learning problem into a sequence optimization problem. The dialogue policy maker generates the next dialogue action $A_t$ based on $k_t$ and $[H_u, H_d]$. Our proposed model uses the GRU as the action decoder to decode the acts and their parameters for the response. Particularly, at step $i$, for decoding $y_i^p$ of $A_t$, the decoder GRU takes the embedding of $y_{i-1}^p$ to generate a hidden vector $h_i^p$. Basic attention mechanism is calculated. where $e$ is the embedding of the token, $c_u$ is the context vector of the input utterance and $c_d$ is the context vector of the dialogue state tracker. $h_i^p$ is the hidden state of the GRU in dialogue policy maker at $i$ timestep. where $y_i^p$ is the token decoded at $i$ timestep. And the final results of dialogue policy maker denote as $A_t$, and the $k$ is the length of it. In our proposed model, the dialogue policy maker can be viewed as a decoder of the seq2seq model conditioned on $[C_{t-1},R_{t-1},U_t]$ and $k_t$. ### Generative Dialogue Policy ::: Nature Language Generator
After getting the dialogue action $A_t$ by the learned dialogue policy maker, the task-oriented dialogue system needs to generate an appropriate response $R_t$ for users. We construct the natural language generator by using template sentences. For each dataset, we extract all the system responses, then we manually modify responses to construct the sentence templates for task-oriented dialogue systems. In our proposed model, the sequence of the acts and parameters $A_t$ will be used for searching appropriate template. However, the classification-based baselines use the categories of acts and their corresponding parameters to search the corresponding template. ### Generative Dialogue Policy ::: Training
In supervised learning, because our proposed model is built in a seq2seq way, the standard cross entropy is adopted as our objective function to train dialogue belief tracker and dialogue policy maker. After supervised learning, the dialogue policy can be further updated by using reinforcement learning. In the context of reinforcement learning, the decoder of dialogue policy maker can be viewed as a policy network, denoted as $\pi _{\theta }(y_j)$ for decoding $y_j$, $\theta $ is the parameters of the decoder. Accordingly, the hidden state created by GRU is the corresponding state, and the choice of the current token $y_j$ is an action. Reward function is also very important for reinforcement learning when decoding every token. To encourage our policy maker to generate correct acts and their corresponding parameters, we set the reward function as follows: once the dialogue acts and their parameters are decoded correctly, the reward is 2; otherwise, the reward is -5; only the label of the dialogue act is decoded correctly but parameters is wrong, the reward is 1; $\lambda $ is a decay parameter. More details are shown in Sec SECREF41. In our proposed model, rewards can only be obtained at the end of decoding $A_t$. In order to get the rewards at each decoding step, we sample some results $A_t$ after choosing $y_j$, and the reward of $y_j$ is set as the average of all the sampled results' rewards. In order to ensure that the model's performance is stable during the fine-tuning phase of reinforcement learning, we freeze the parameters of user utterance and dialogue belief tracker, only the parameters of the dialogue policy maker will be optimized by reinforcement learning. Policy gradient algorithm REINFORCE BIBREF18 is used for pretrained dialogue policy maker: where the $m$ is the length of the decoded action. The objective function $J$ can be optimized by gradient descent. ### Experiments
We evaluate the performance of the proposed model in three aspects: (1) the accuracy of the dialogue state tracker, it aims to show the impact of the dialogue state tracker on the dialogue policy maker; (2) the accuracy of dialogue policy maker, it aims to explain the performance of different methods of constructing dialogue policy; (3) the quality of the final response, it aims to explain the impact of the dialogue policy on the final dialogue response. The evaluation metrics are listed as follows: BPRA: Belief Per-Response Accuracy (BPRA) tests the ability to generate the correct user intents during the dialogue. This metric is used to evaluate the accuracy of dialogue belief tracker BIBREF1. APRA: Action Per-Response Accuracy (APRA) evaluates the per-turn accuracy of the dialogue actions generated by dialogue policy maker. For baselines, APRA evaluates the classification accuracy of the dialogue policy maker. But our model actually generates each individual token of actions, and we consider a prediction to be correct only if every token of the model output matches the corresponding token in the ground truth. BLEU BIBREF19: The metric evaluates the quality of the final response generated by natural language generator. The metric is usually used to measure the performance of the task-oriented dialogue system. We also choose the following metrics to evaluate the efficiency of training the model: $\mathbf {Time_{full}}$: The time for training the whole model, which is important for industry settings. $\mathbf {Time_{DP}}$: The time for training the dialogue policy maker in a task-oriented dialogue system. ### Experiments ::: Datasets
We adopt the DSTC2 BIBREF20 dataset and Maluuba BIBREF21 dataset to evaluate our proposed model. Both of them are the benchmark datasets for building the task-oriented dialog systems. Specifically, the DSTC2 is a human-machine dataset in the single domain of restaurant searching. The Maluuba is a very complex human-human dataset in travel booking domain which contains more slots and values than DSTC2. Detailed slot information in each dataset is shown in Table TABREF34. ### Experiments ::: Baselines
For comparison, we choose two state-of-the-art baselines and their variants. E2ECM BIBREF11: In dialogue policy maker, it adopts a classic classification for skeletal sentence template. In our implement, we construct multiple binary classifications for each act to search the sentence template according to the work proposed by BIBREF11. CDM BIBREF10: This approach designs a group of classifications (two multi-class classifications and some binary classifications) to model the dialogue policy. E2ECM+RL: It fine tunes the classification parameters of the dialogue policy by REINFORCE BIBREF18. CDM+RL: It fine tunes the classification of the act and corresponding parameters by REINFORCE BIBREF18. In order to verify the performance of the dialogue policy maker, the utterance encoder and dialogue belief tracker of our proposed model and baselines is the same, only dialogue policy maker is different. ### Experiments ::: Parameters settings
For all models, the hidden size of dialogue belief tracker and utterance encoder is 350, and the embedding size $d_{emb}$ is set to 300. For our proposed model, the hidden size of decoder in dialogue policy maker is 150. The vocabulary size $|V|$ is 540 for DSTC2 and 4712 for Maluuba. And the size of $k_t$ is set to 20. An Adam optimizer BIBREF22 is used for training our models and baselines, with a learning rate of 0.001 for supervised training and 0.0001 for reinforcement learning. In reinforcement learning, the decay parameter $\lambda $ is set to 0.8. The weight decay is set to 0.001. And early stopping is performed on developing set. ### Experiments ::: Experimental Results
The experimental results of the proposed model and baselines will be analyzed from the following aspects. BPRA Results: As shown in Table TABREF35, most of the models have similar performance on BPRA on these two datasets, which can guarantee a consistent impact on the dialogue policy maker. All the models perform very well in BPRA on DSTC2 dataset. On Maluuba dataset, the BPRA decreases because of the complex domains. We can notice that BPRA of CDM is slightly worse than other models on Maluuba dataset, the reason is that the CDM's dialogue policy maker contains lots of classifications and has the bigger loss than other models because of complex domains, which affects the training of the dialogue belief tracker. APRA Results: Compared with baselines, GDP achieves the best performance in APRA on two datasets. It can be noted that we do not compare with the E2ECM baseline in APRA. E2ECM only uses a simple classifier to recognize the label of the acts and ignores the parameters information. In our experiment, APRA of E2ECM is slightly better than our method. Considering the lack of parameters of the acts, it's unfair for our GDP method. Furthermore, the CDM baseline considers the parameters of the act. But GDP is far better than CDM in supervised learning and reinforcement learning. BLEU Results: GDP significantly outperforms the baselines on BLEU. As mentioned above, E2ECM is actually slightly better than GDP in APRA. But in fact, we can find that the language quality of the response generated by GDP is still better than E2ECM, which proves that lack of enough parameters information makes it difficult to find the appropriate sentence template in NLG. It can be found that the BLEU of all models is very poor on Maluuba dataset. The reason is that Maluuba is a human-human task-oriented dialogue dataset, the utterances are very flexible, the natural language generator for all methods is difficult to generate an accurate utterance based on the context. And DSTC2 is a human-machine dialog dataset. The response is very regular so the effectiveness of NLG will be better than that of Maluuba. But from the results, the GDP is still better than the baselines on Maluuba dataset, which also verifies that our proposed method is more accurate in modeling dialogue policy on complex domains than the classification-based methods. Time and Model Size: In order to obtain more accurate and complete dialogue policy for task-oriented dialogue systems, the proposed model has more parameters on the dialogue policy maker than baselines. As shown in Figure FIGREF44, E2ECM has the minimal dialogue policy parameters because of the simple classification. It needs minimum training time, but the performance of E2ECM is bad. The number of parameters in the CDM model is slightly larger than E2ECM. However, because both of them are classification methods, they all lose some important information about dialogue policy. Therefore, we can see from the experimental results that the quality of CDM's dialogue policy is as bad as E2ECM. The number of dialogue policy maker's parameters in GDP model is much larger than baselines. Although the proposed model need more time to be optimized by supervised learning and reinforcement learning, the performance is much better than all baselines. ### Experiments ::: Case Study
Table TABREF43 illustrates an example of our proposed model and baselines on DSTC2 dataset. In this example, a user's goal is to find a cheap restaurant in the east part of the town. In the current turn, the user wants to get the address of the restaurant. E2ECM chooses the inform and offer acts accurately, but the lack of the inform's parameters makes the final output deviate from the user's goal. CDM generates the parameters of offer successfully, but the lack of the information of inform also leads to a bad result. By contrast, the proposed model GDP can generate all the acts and their corresponding parameters as the dialogue action. Interestingly, the final result of GDP is exactly the same as the ground truth, which verifies that the proposed model is better than the state-of-the-art baselines. ### Conclusion
In this paper, we propose a novel model named GDP. Our proposed model treats the dialogue policy modeling as the generative task instead of the discriminative task which can hold more information for dialogue policy modeling. We evaluate the GDP on two benchmark task-oriented dialogue datasets. Extensive experiments show that GDP outperforms the existing classification-based methods on both action accuracy and BLEU. Figure 1: The examples in DSTC2 dataset, our proposed model can hold more information about dialogue policy than the classification models mentioned above. “MA, w/o P” is the model that chooses multiple acts without corresponding parameters during dialogue police modeling, “w/o MA, P” is the model that chooses only one act and its parameters. Figure 2: GDP overview. The utterance encoder encodes the user utterance, the dialogue context and the last reply of the systems into the dense vector. As for dialogue belief tracker, we use the approach of Lei et al. (2018) to generate dialogue context. Then this information will be used to search the knowledge base. Based on the user’s intents and query results, dialogue policy maker generates the next dialogue action by using our RNN-based proposed method. Table 1: The details of DSTC2 and Maluuba dataset. The Maluuba dataset is more complex than DSTC2, and has some continuous value space such as time and price which is hard to solve for classification model. Table 2: The performance of baselines and proposed model on DSTC2 and Maluuba dataset. T imefull is the time spent on training the whole model, T imeDP is the time spent on training the dialogue policy maker. Table 3: Case Study on DSTC2 dataset. The first column is the Dialogue Context of this case, it contains three parts: (1) Inf is the user’s intent captured by dialogue state tracker; (2) sys is the system response at last turn; (3) user is the user utterance in this turn. The second column to the fifth column has two rows, above is the action made by the learned dialogue policy maker below is the final response made by template-based generator. Figure 3: The number of the parameters. GDP has the bigger model size and more dialogue policy parameters because of the RNN-based dialogue policy maker.
|
BPRA, APRA, BLEU
|
What are two models' architectures in proposed solution?
|
### Introduction
NLP tasks that require multi-hop reasoning have recently enjoyed rapid progress, especially on multi-hop question answering BIBREF0, BIBREF1, BIBREF2. Advances have benefited from rich annotations of supporting evidence, as in the popular multi-hop QA and relation extraction benchmarks, e.g., HotpotQA BIBREF3 and DocRED BIBREF4, where the evidence sentences for the reasoning process were labeled by human annotators. Such evidence annotations are crucial for modern model training, since they provide finer-grained supervision for better guiding the model learning. Furthermore, they allow a pipeline fashion of model training, with each step, such as passage ranking and answer extraction, trained as a supervised learning sub-task. This is crucial from a practical perspective, in order to reduce the memory usage when handling a large amount of inputs with advanced, large pre-trained models BIBREF5, BIBREF6, BIBREF7. Manual evidence annotation is expensive, so there are only a few benchmarks with supporting evidence annotated. Even for these datasets, the structures of the annotations are still limited, as new model designs keep emerging and they may require different forms of evidence annotations. As a result, the supervision from these datasets can still be insufficient for training accurate models. Taking question answering with multi-hop reasoning as an example, annotating only supporting passages is not sufficient to show the reasoning processes due to the lack of necessary structural information (Figure FIGREF1). One example is the order of annotated evidence, which is crucial in logic reasoning and the importance of which has also been demonstrated in text-based QA BIBREF8. The other example is how the annotated evidence pieces are connected, which requires at least the definition of arguments, such as a linking entity, concept, or event. Such information has proved useful by the recently popular entity-centric methods BIBREF9, BIBREF10, BIBREF11, BIBREF12, BIBREF0, BIBREF2 and intuitively will be a benefit to these methods if available. We propose a cooperative game approach to recovering the reasoning chains with the aforementioned necessary structural information for multi-hop QA. Each recovered chain corresponds to a list of ordered passages and each pair of adjacent passages is connected with a linking entity. Specifically, we start with a model, the Ranker, which selects a sequence of passages arriving at the answers, with the restriction that each adjacent passage pair shares at least an entity. This is essentially an unsupervised task and the selection suffers from noise and ambiguity. Therefore we introduce another model, the Reasoner, which predicts the exact linking entity that points to the next passage. The two models play a cooperative game and are rewarded when they find a consistent chain. In this way, we restrict the selection to satisfy not only the format constraints (i.e., ordered passages with connected adjacencies) but also the semantic constraints (i.e., finding the next passage given that the partial selection can be effectively modeled by a Reasoner). Therefore, the selection can be less noisy. We evaluate the proposed method on datasets with different properties, i.e., HotpotQA and MedHop BIBREF13, to cover cases with both 2-hop and 3-hop reasoning. We created labeled reasoning chains for both datasets. Experimental results demonstrate the significant advantage of our proposed approach. ### Task Definition
Reasoning Chains Examples of reasoning chains in HotpotQA and MedHop are shown in Figure FIGREF1. Formally, we aim at recovering the reasoning chain in the form of $(p_1 \rightarrow e_{1,2} \rightarrow p_2 \rightarrow e_{2,3} \rightarrow \cdots \rightarrow e_{n-1,n} \rightarrow p_n)$, where each $p_i$ is a passage and each $e_{i,i+1}$ is an entity that connects $p_i$ and $p_{i+1}$, i.e., appearing in both passages. The last passage $p_n$ in the chain contains the correct answer. We say $p_i$ connects $e_{i-1,i}$ and $e_{i,i+1}$ in the sense that it describes a relationship between the two entities. Our Task Given a QA pair $(q,a)$ and all its candidate passages $\mathcal {P}$, we can extract all possible candidate chains that satisfy the conditions mentioned above, denoted as $\mathcal {C}$. The goal of reasoning chain recovery is to extract the correct chains from all the candidates, given $q,a$ and $\mathcal {P}$ as inputs. Related Work Although there are recent interests on predicting reasoning chains for multi-hop QA BIBREF0, BIBREF14, BIBREF2, they all consider a fully supervised setting; i.e., annotated reasoning chains are available. Our work is the first to recover reasoning chains in a more general unsupervised setting, thus falling into the direction of denoising over distant supervised signals. From this perspective, the most relevant studies in the NLP field includes BIBREF15, BIBREF16 for evidence identification in open-domain QA and BIBREF17, BIBREF18, BIBREF19 for rationale recovery. ### Method
The task of recovering reasoning chains is essentially an unsupervised problem, as we have no access to annotated reasoning chains. Therefore, we resort to the noisy training signal from chains obtained by distant supervision. We first propose a conditional selection model that optimizes the passage selection by considering their orders (Section SECREF4). We then propose a cooperative Reasoner-Ranker game (Section SECREF12) in which the Reasoner recovers the linking entities that point to the next passage. This enhancement encourages the Ranker to select the chains such that their distribution is easier for a linking entity prediction model (Reasoner) to capture. Therefore, it enables our model to denoise the supervision signals while recovering chains with entity information. Figure FIGREF3 gives our overall framework, with a flow describing how the Reasoner passes additional rewards to the Ranker. ### Method ::: Passage Ranking Model
The key component of our framework is the Ranker model, which is provided with a question $q$ and $K$ passages $\mathcal {P} = \lbrace p_1, p_2 ... p_K\rbrace $ from a pool of candidates, and outputs a chain of selected passages. ### Method ::: Passage Ranking Model ::: Passage Scoring
For each step of the chain, the Ranker estimates a distribution of the selection of each passage. To this end we first encode the question and passage with a 2-layer bi-directional GRU network, resulting in an encoded question $\mathbf {Q} = \lbrace \vec{\mathbf {q}_0}, \vec{\mathbf {q}_1}, ..., \vec{\mathbf {q}_N}\rbrace $ and $\mathbf {H}_i = \lbrace \vec{\mathbf {h}_{i,0}}, \vec{\mathbf {h}_{i,1}}, ..., \vec{\mathbf {h}_{i,M_i}}\rbrace $ for each passage $p_i \in P$ of length $M_i$. Then we use the MatchLSTM model BIBREF20 to get the matching score between $\mathbf {Q}$ and each $\mathbf {H}_i$ and derive the distribution of passage selection $P(p_i|q)$ (see Appendix SECREF6 for details). We denote $P(p_i|q)=\textrm {MatchLSTM}(\mathbf {H}_i, \mathbf {Q})$ for simplicity. ### Method ::: Passage Ranking Model ::: Conditional Selection
To model passage dependency along the chain of reasoning, we use a hard selection model that builds a chain incrementally. Provided with the $K$ passages, at each step $t$ the Ranker computes $P^t(p_i|\mathbf {Q}^{t-1}), i = 0, ..., K$, which is the probability of selecting passage $p_i$ conditioned on the query and previous states representation $\mathbf {Q}^{t-1}$. Then we sample one passage $p^t_{\tau }$ according to the predicted selection probability. The first step starts with the original question $\mathbf {Q}^0$. A feed-forward network is used to project the concatenation of query encoding and selected passage encoding $\tilde{\mathbf {m}}^t_{p_{\tau }}$ back to the query space, and the new query $\mathbf {Q}^{t+1}$ is used to select the next passage. ### Method ::: Passage Ranking Model ::: Reward via Distant Supervision
We use policy gradient BIBREF21 to optimize our model. As we have no access to annotated reasoning chains during training, the reward comes from distant supervision. Specifically, we reward the Ranker if a selected passage appears as the corresponding part of a distant supervised chain in $\mathcal {C}$. The model receives immediate reward at each step of selection. In this paper we only consider chains consist of $\le 3$ passages (2-hop and 3-hop chains). For the 2-hop cases, our model predicts a chain of two passages from the candidate set $\mathcal {C}$ in the form of $p_h\rightarrow e \rightarrow p_t$. Each candidate chain satisfies that $p_t$ contains the answer, while $p_h$ and $p_t$ contain a shared entity $e$. We call $p_h$ the head passage and $p_t$ the tail passage. Let $\mathcal {P}_{T}/\mathcal {P}_{H}$ denote the set of all tail/head passages from $\mathcal {C}$. Our model receives rewards $r_h, r_t$ according to its selections: For the 3-hop cases, we need to select an additional intermediate passage $p_m$ between $p_h$ and $p_t$. If we reward any $p_m$ selection that appears in the middle of a chain in candidate chain set $\mathcal {C}$, the number of feasible options can be very large. Therefore, we make our model first select the head passage $p_h$ and the tail passage $p_t$ independently and then select $p_m$ conditioned on $(p_h,p_t)$. We further restrict that each path in $\mathcal {C}$ must have the head passage containing an entity from $q$. Then the selected $p_m$ is only rewarded if it appears in a chain in $\mathcal {C}$ that starts with $p_h$ and ends with $p_t$: ### Method ::: Cooperative Reasoner
To alleviate the noise in the distant supervision signal $\mathcal {C}$, in addition to the conditional selection, we further propose a cooperative Reasoner model, also implemented with the MatchLSTM architecture (see Appendix SECREF6), to predict the linking entity from the selected passages. Intuitively, when the Ranker makes more accurate passage selections, the Reasoner will work with less noisy data and thus is easier to succeed. Specifically, the Reasoner learns to extract the linking entity from chains selected by a well-trained Ranker, and it benefits the Ranker training by providing extra rewards. Taking 2-hop as an example, we train the Ranker and Reasoner alternatively as a cooperative game: Reasoner Step: Given the first passage $p_t$ selected by the trained Ranker, the Reasoner predicts the probability of each entity $e$ appearing in $p_t$. The Reasoner is trained with the cross-entropy loss: Ranker Step: Given the Reasoner's top-1 predicted linking entity $e$, the reward for Ranker at the $2^{\textrm {nd}}$ step is defined as: The extension to 3-hop cases is straightforward; the only difference is that the Reasoner reads both the selected $p_h$ and $p_t$ to output two entities. The Ranker receives one extra reward if the Reasoner picks the correct linking entity from $p_h$, so does $p_t$. ### Experiments ::: Settings ::: Datasets
We evaluate our path selection model on HotpotQA bridge type questions and on the MedHop dataset. In HotpotQA, the entities are pre-processed Wiki anchor link objects and in MedHop they are drug/protein database identifiers. For HotpotQA, two supporting passages are provided along with each question. We ignore the support annotations during training and use them to create ground truth on development set: following BIBREF8, we determine the order of passages according to whether a passage contains the answer. We discard ambiguous instances. For MedHop, there is no evidence annotated. Therefore we created a new evaluation dataset by manually annotating the correct paths for part of the development set: we first extract all candidate paths in form of passage triplets $(p_h, p_m, p_t)$, such that $p_h$ contains the query drug and $p_t$ contains the answer drug, and $p_h/p_m$ and $p_m/p_t$ are connected by shared proteins. We label a chain as positive if all the drug-protein or protein-protein interactions are described in the corresponding passages. Note that the positive paths are not unique for a question. During training we select chains based on the full passage set $\mathcal {P}$; at inference time we extract the chains from the candidate set $\mathcal {C}$ (see Section SECREF2). ### Experiments ::: Settings ::: Baselines and Evaluation Metric
We compare our model with (1) random baseline, which randomly selects a candidate chain from the distant supervision chain set $\mathcal {C}$; and (2) distant supervised MatchLSTM, which uses the same base model as ours but scores and selects the passages independently. We use accuracy as our evaluation metric. As HotpotQA does not provide ground-truth linking entities, we only evaluate whether the supporting passages are fully recovered (yet our model still output the full chains). For MedHop we evaluate whether the whole predicted chain is correct. More details can be found in Appendix SECREF7. We use BIBREF24 as word embedding for HotpotQA, and BIBREF25 for MedHop. ### Experiments ::: Results ::: HotpotQA
We first evaluate on the 2-hop HotpotQA task. Our best performed model first selects the tail passage $p_t$ and then the head passage $p_h$, because the number of candidates of tail is smaller ($\sim $2 per question). Table TABREF21 shows the results. First, training a ranker with distant supervision performs significantly better than the random baseline, showing that the training process itself has a certain degree of denoising ability to distinguish the more informative signals from distant supervision labels. By introducing additional inductive bias of orders, the conditional selection model further improves with a large margin. Finally, our cooperative game gives the best performance, showing that a trained Reasoner has the ability of ignoring entity links that are irrelevant to the reasoning chain. Table TABREF22 demonstrates the effect of selecting directions, together with the methods' recall on head passages and tail passages. The latter is evaluated on a subset of bridge-type questions in HotpotQA which has no ambiguous support annotations in passage orders; i.e., among the two human-labeled supporting passages, only one contains the answer and thus must be a tail. The results show that selecting tail first performs better. The cooperative game mainly improves the head selection. ### Experiments ::: Results ::: MedHop
Results in table TABREF21 show that recovering chains from MedHop is a much harder task: first, the large number of distant supervision chains in $\mathcal {C}$ introduce too much noise so the Distant Supervised Ranker improves only 3%; second, the dependent model leads to no improvement because $\mathcal {C}$ is strictly ordered given our data construction. Our cooperative game manages to remain effective and gives further improvement. ### Conclusions
In this paper we propose the problem of recovering reasoning chains in multi-hop QA from weak supervision signals. Our model adopts an cooperative game approach where a ranker and a reasoner cooperate to select the most confident chains. Experiments on the HotpotQA and MedHop benchmarks show the effectiveness of the proposed approach. ### Details of MatchLSTMs for Passage Scoring and Reasoner ::: MatchLSTM for Passage Scoring
Given the embeddings $\mathbf {Q} = \lbrace \vec{\mathbf {q}_0}, \vec{\mathbf {q}_1}, ..., \vec{\mathbf {q}_N}\rbrace $ of the question $q$, and $\mathbf {H}_i = \lbrace \vec{\mathbf {h}_{i,0}}, \vec{\mathbf {h}_{i,1}}, ..., \vec{\mathbf {h}_{i,M_i}}\rbrace $ of each passage $p_i \in P$, we use the MatchLSTM BIBREF20 to match $\mathbf {Q}$ and $\mathbf {H}_i$ as follows: The final vector $\tilde{\mathbf {m}}_i$ represents the matching state between $q$ and $p_i$. All the $\tilde{\mathbf {m}}_i$s are then passed to a linear layer that outputs the ranking score of each passage. We apply softmax over the scores to get the probability of passage selection $P(p_i|q)$. We denote the above computation as $P(p_i|q)=\textrm {MatchLSTM}(\mathbf {H}_i, \mathbf {Q})$ for simplicity. ### Details of MatchLSTMs for Passage Scoring and Reasoner ::: MatchLSTM for Reasoner
Given the question embedding $\mathbf {Q}^r = \lbrace \vec{\mathbf {q}^r_0}, \vec{\mathbf {q}^r_1}, ..., \vec{\mathbf {q}^r_N}\rbrace $ and the input passage embedding $\mathbf {H}^r = \lbrace \vec{\mathbf {h}^r_{0}}, \vec{\mathbf {h}^r_{1}}, ..., \vec{\mathbf {h}^r_{M}}\rbrace $ of $p$, the Reasoner predicts the probability of each entity in the passage being the linking entity of the next passage in the chain. We use a reader model similar to BIBREF3 as our Reasoner network. We first describe an attention sub-module. Given input sequence embedding $\mathbf {A} = \lbrace \vec{\mathbf {a}_0}, \vec{\mathbf {a}_1}, ..., \vec{\mathbf {a}_N}\rbrace $ and $\mathbf {B} = \lbrace \vec{\mathbf {b}_{0}}, \vec{\mathbf {b}_{1}}, ..., \vec{\mathbf {b}_{M}}\rbrace $, we define $\tilde{\mathcal {M}} = \text{Attention}(\mathbf {A}, \mathbf {B})$: where FFN denotes a feed forward layer which projects the concatenated embedding back to the original space. The Reasoner network consists of multiple attention layers, together with a bidirectional GRU encoder and skip connection. For each token $e_k, k = 0, 1,..., M$ represented by $h^r_{p,k}$ at the corresponding location, we have: where $g$ is the classification layer, softmax is applied across all entities to get the probability. We denote the computation above as $P^r(e_k| \mathbf {p}) = \textrm {MatchLSTM.Reader}(e_k, \mathbf {p})$ for simplicity. ### Definition of Chain Accuracy
In HotpotQA, on average we can find 6 candidate chains (2-hop) in a instance, and the human labeled true reasoning chain is unique. A predicted chain is correct if the chain only contains all supporting passages (exact match of passages). In MedHop, on average we can find 30 candidate chains (3-hop). For each candidate chain our human annotators labeled whether it is correct or not, and the correct reasoning chain is not unique. A predicted chain is correct if it is one of the chains that human labeled as correct. The accuracy is defined as the ratio: Figure 1: An example of reasoning chains in HotpotQA (2- hop) and MedHop (3-hop). HotpotQA provides only supporting passages {P3, P9}, without order and linking information. Figure 2: Model overview. The cooperative Ranker and Reasoner are trained alternatively. The Ranker selects a passage p at each step conditioned on the question q and history selection, and receives reward r1 if p is evidence. Conditioned on q, the Reasoner predicts which entity from p links to the next evidence passage. The Ranker receives extra reward r2 if its next selection is connected by the entity predicted by the Reasoner. Both q and answer a are model inputs. While q is fed to the Ranker/Reasoner as input, empirically the best way of using a is for constructing the candidate set thus computing the reward r1. We omit the flow from q/a for simplicity. Table 1: Reasoning Chain selection results. Table 2: Ablation test on HotpotQA.
|
Reasoner model, also implemented with the MatchLSTM architecture, Ranker model
|
How does Bertrand Malloy feel about sending the team of man in his place?
A. He wishes he could do the job himself but knows they are the best for the job.
B. He is relieved to not have to go but wishes he could have found better replacements.
C. He is glad he does not have to go and things they will do a better job anyway.
D. He is fairly certain they are going to mess up the peace talks.
|
IN CASE OF FIRE By RANDALL GARRETT There are times when a broken tool is better than a sound one, or a twisted personality more useful than a whole one. For instance, a whole beer bottle isn't half the weapon that half a beer bottle is ... Illustrated by Martinez In his office apartment, on the top floor of the Terran Embassy Building in Occeq City, Bertrand Malloy leafed casually through the dossiers of the four new men who had been assigned to him. They were typical of the kind of men who were sent to him, he thought. Which meant, as usual, that they were atypical. Every man in the Diplomatic Corps who developed a twitch or a quirk was shipped to Saarkkad IV to work under Bertrand Malloy, Permanent Terran Ambassador to His Utter Munificence, the Occeq of Saarkkad. Take this first one, for instance. Malloy ran his finger down the columns of complex symbolism that showed the complete psychological analysis of the man. Psychopathic paranoia. The man wasn't technically insane; he could be as lucid as the next man most of the time. But he was morbidly suspicious that every man's hand was turned against him. He trusted no one, and was perpetually on his guard against imaginary plots and persecutions. Number two suffered from some sort of emotional block that left him continually on the horns of one dilemma or another. He was psychologically incapable of making a decision if he were faced with two or more possible alternatives of any major importance. Number three ... Malloy sighed and pushed the dossiers away from him. No two men were alike, and yet there sometimes seemed to be an eternal sameness about all men. He considered himself an individual, for instance, but wasn't the basic similarity there, after all? He was—how old? He glanced at the Earth calendar dial that was automatically correlated with the Saarkkadic calendar just above it. Fifty-nine next week. Fifty-nine years old. And what did he have to show for it besides flabby muscles, sagging skin, a wrinkled face, and gray hair? Well, he had an excellent record in the Corps, if nothing else. One of the top men in his field. And he had his memories of Diane, dead these ten years, but still beautiful and alive in his recollections. And—he grinned softly to himself—he had Saarkkad. He glanced up at the ceiling, and mentally allowed his gaze to penetrate it to the blue sky beyond it. Out there was the terrible emptiness of interstellar space—a great, yawning, infinite chasm capable of swallowing men, ships, planets, suns, and whole galaxies without filling its insatiable void. Malloy closed his eyes. Somewhere out there, a war was raging. He didn't even like to think of that, but it was necessary to keep it in mind. Somewhere out there, the ships of Earth were ranged against the ships of the alien Karna in the most important war that Mankind had yet fought. And, Malloy knew, his own position was not unimportant in that war. He was not in the battle line, nor even in the major production line, but it was necessary to keep the drug supply lines flowing from Saarkkad, and that meant keeping on good terms with the Saarkkadic government. The Saarkkada themselves were humanoid in physical form—if one allowed the term to cover a wide range of differences—but their minds just didn't function along the same lines. For nine years, Bertrand Malloy had been Ambassador to Saarkkad, and for nine years, no Saarkkada had ever seen him. To have shown himself to one of them would have meant instant loss of prestige. To their way of thinking, an important official was aloof. The greater his importance, the greater must be his isolation. The Occeq of Saarkkad himself was never seen except by a handful of picked nobles, who, themselves, were never seen except by their underlings. It was a long, roundabout way of doing business, but it was the only way Saarkkad would do any business at all. To violate the rigid social setup of Saarkkad would mean the instant closing off of the supply of biochemical products that the Saarkkadic laboratories produced from native plants and animals—products that were vitally necessary to Earth's war, and which could be duplicated nowhere else in the known universe. It was Bertrand Malloy's job to keep the production output high and to keep the materiel flowing towards Earth and her allies and outposts. The job would have been a snap cinch in the right circumstances; the Saarkkada weren't difficult to get along with. A staff of top-grade men could have handled them without half trying. But Malloy didn't have top-grade men. They couldn't be spared from work that required their total capacity. It's inefficient to waste a man on a job that he can do without half trying where there are more important jobs that will tax his full output. So Malloy was stuck with the culls. Not the worst ones, of course; there were places in the galaxy that were less important than Saarkkad to the war effort. Malloy knew that, no matter what was wrong with a man, as long as he had the mental ability to dress himself and get himself to work, useful work could be found for him. Physical handicaps weren't at all difficult to deal with. A blind man can work very well in the total darkness of an infrared-film darkroom. Partial or total losses of limbs can be compensated for in one way or another. The mental disabilities were harder to deal with, but not totally impossible. On a world without liquor, a dipsomaniac could be channeled easily enough; and he'd better not try fermenting his own on Saarkkad unless he brought his own yeast—which was impossible, in view of the sterilization regulations. But Malloy didn't like to stop at merely thwarting mental quirks; he liked to find places where they were useful . The phone chimed. Malloy flipped it on with a practiced hand. "Malloy here." "Mr. Malloy?" said a careful voice. "A special communication for you has been teletyped in from Earth. Shall I bring it in?" "Bring it in, Miss Drayson." Miss Drayson was a case in point. She was uncommunicative. She liked to gather in information, but she found it difficult to give it up once it was in her possession. Malloy had made her his private secretary. Nothing—but nothing —got out of Malloy's office without his direct order. It had taken Malloy a long time to get it into Miss Drayson's head that it was perfectly all right—even desirable—for her to keep secrets from everyone except Malloy. She came in through the door, a rather handsome woman in her middle thirties, clutching a sheaf of papers in her right hand as though someone might at any instant snatch it from her before she could turn it over to Malloy. She laid them carefully on the desk. "If anything else comes in, I'll let you know immediately, sir," she said. "Will there be anything else?" Malloy let her stand there while he picked up the communique. She wanted to know what his reaction was going to be; it didn't matter because no one would ever find out from her what he had done unless she was ordered to tell someone. He read the first paragraph, and his eyes widened involuntarily. "Armistice," he said in a low whisper. "There's a chance that the war may be over." "Yes, sir," said Miss Drayson in a hushed voice. Malloy read the whole thing through, fighting to keep his emotions in check. Miss Drayson stood there calmly, her face a mask; her emotions were a secret. Finally, Malloy looked up. "I'll let you know as soon as I reach a decision, Miss Drayson. I think I hardly need say that no news of this is to leave this office." "Of course not, sir." Malloy watched her go out the door without actually seeing her. The war was over—at least for a while. He looked down at the papers again. The Karna, slowly being beaten back on every front, were suing for peace. They wanted an armistice conference—immediately. Earth was willing. Interstellar war is too costly to allow it to continue any longer than necessary, and this one had been going on for more than thirteen years now. Peace was necessary. But not peace at any price. The trouble was that the Karna had a reputation for losing wars and winning at the peace table. They were clever, persuasive talkers. They could twist a disadvantage to an advantage, and make their own strengths look like weaknesses. If they won the armistice, they'd be able to retrench and rearm, and the war would break out again within a few years. Now—at this point in time—they could be beaten. They could be forced to allow supervision of the production potential, forced to disarm, rendered impotent. But if the armistice went to their own advantage ... Already, they had taken the offensive in the matter of the peace talks. They had sent a full delegation to Saarkkad V, the next planet out from the Saarkkad sun, a chilly world inhabited only by low-intelligence animals. The Karna considered this to be fully neutral territory, and Earth couldn't argue the point very well. In addition, they demanded that the conference begin in three days, Terrestrial time. The trouble was that interstellar communication beams travel a devil of a lot faster than ships. It would take more than a week for the Earth government to get a vessel to Saarkkad V. Earth had been caught unprepared for an armistice. They objected. The Karna pointed out that the Saarkkad sun was just as far from Karn as it was from Earth, that it was only a few million miles from a planet which was allied with Earth, and that it was unfair for Earth to take so much time in preparing for an armistice. Why hadn't Earth been prepared? Did they intend to fight to the utter destruction of Karn? It wouldn't have been a problem at all if Earth and Karn had fostered the only two intelligent races in the galaxy. The sort of grandstanding the Karna were putting on had to be played to an audience. But there were other intelligent races throughout the galaxy, most of whom had remained as neutral as possible during the Earth-Karn war. They had no intention of sticking their figurative noses into a battle between the two most powerful races in the galaxy. But whoever won the armistice would find that some of the now-neutral races would come in on their side if war broke out again. If the Karna played their cards right, their side would be strong enough next time to win. So Earth had to get a delegation to meet with the Karna representatives within the three-day limit or lose what might be a vital point in the negotiations. And that was where Bertrand Malloy came in. He had been appointed Minister and Plenipotentiary Extraordinary to the Earth-Karn peace conference. He looked up at the ceiling again. "What can I do?" he said softly. On the second day after the arrival of the communique, Malloy made his decision. He flipped on his intercom and said: "Miss Drayson, get hold of James Nordon and Kylen Braynek. I want to see them both immediately. Send Nordon in first, and tell Braynek to wait." "Yes, sir." "And keep the recorder on. You can file the tape later." "Yes, sir." Malloy knew the woman would listen in on the intercom anyway, and it was better to give her permission to do so. James Nordon was tall, broad-shouldered, and thirty-eight. His hair was graying at the temples, and his handsome face looked cool and efficient. Malloy waved him to a seat. "Nordon, I have a job for you. It's probably one of the most important jobs you'll ever have in your life. It can mean big things for you—promotion and prestige if you do it well." Nordon nodded slowly. "Yes, sir." Malloy explained the problem of the Karna peace talks. "We need a man who can outthink them," Malloy finished, "and judging from your record, I think you're that man. It involves risk, of course. If you make the wrong decisions, your name will be mud back on Earth. But I don't think there's much chance of that, really. Do you want to handle small-time operations all your life? Of course not. "You'll be leaving within an hour for Saarkkad V." Nordon nodded again. "Yes, sir; certainly. Am I to go alone?" "No," said Malloy, "I'm sending an assistant with you—a man named Kylen Braynek. Ever heard of him?" Nordon shook his head. "Not that I recall, Mr. Malloy. Should I have?" "Not necessarily. He's a pretty shrewd operator, though. He knows a lot about interstellar law, and he's capable of spotting a trap a mile away. You'll be in charge, of course, but I want you to pay special attention to his advice." "I will, sir," Nordon said gratefully. "A man like that can be useful." "Right. Now, you go into the anteroom over there. I've prepared a summary of the situation, and you'll have to study it and get it into your head before the ship leaves. That isn't much time, but it's the Karna who are doing the pushing, not us." As soon as Nordon had left, Malloy said softly: "Send in Braynek, Miss Drayson." Kylen Braynek was a smallish man with mouse-brown hair that lay flat against his skull, and hard, penetrating, dark eyes that were shadowed by heavy, protruding brows. Malloy asked him to sit down. Again Malloy went through the explanation of the peace conference. "Naturally, they'll be trying to trick you every step of the way," Malloy went on. "They're shrewd and underhanded; we'll simply have to be more shrewd and more underhanded. Nordon's job is to sit quietly and evaluate the data; yours will be to find the loopholes they're laying out for themselves and plug them. Don't antagonize them, but don't baby them, either. If you see anything underhanded going on, let Nordon know immediately." "They won't get anything by me, Mr. Malloy." By the time the ship from Earth got there, the peace conference had been going on for four days. Bertrand Malloy had full reports on the whole parley, as relayed to him through the ship that had taken Nordon and Braynek to Saarkkad V. Secretary of State Blendwell stopped off at Saarkkad IV before going on to V to take charge of the conference. He was a tallish, lean man with a few strands of gray hair on the top of his otherwise bald scalp, and he wore a hearty, professional smile that didn't quite make it to his calculating eyes. He took Malloy's hand and shook it warmly. "How are you, Mr. Ambassador?" "Fine, Mr. Secretary. How's everything on Earth?" "Tense. They're waiting to see what is going to happen on Five. So am I, for that matter." His eyes were curious. "You decided not to go yourself, eh?" "I thought it better not to. I sent a good team, instead. Would you like to see the reports?" "I certainly would." Malloy handed them to the secretary, and as he read, Malloy watched him. Blendwell was a political appointee—a good man, Malloy had to admit, but he didn't know all the ins and outs of the Diplomatic Corps. When Blendwell looked up from the reports at last, he said: "Amazing! They've held off the Karna at every point! They've beaten them back! They've managed to cope with and outdo the finest team of negotiators the Karna could send." "I thought they would," said Malloy, trying to appear modest. The secretary's eyes narrowed. "I've heard of the work you've been doing here with ... ah ... sick men. Is this one of your ... ah ... successes?" Malloy nodded. "I think so. The Karna put us in a dilemma, so I threw a dilemma right back at them." "How do you mean?" "Nordon had a mental block against making decisions. If he took a girl out on a date, he'd have trouble making up his mind whether to kiss her or not until she made up his mind for him, one way or the other. He's that kind of guy. Until he's presented with one, single, clear decision which admits of no alternatives, he can't move at all. "As you can see, the Karna tried to give us several choices on each point, and they were all rigged. Until they backed down to a single point and proved that it wasn't rigged, Nordon couldn't possibly make up his mind. I drummed into him how important this was, and the more importance there is attached to his decisions, the more incapable he becomes of making them." The Secretary nodded slowly. "What about Braynek?" "Paranoid," said Malloy. "He thinks everyone is plotting against him. In this case, that's all to the good because the Karna are plotting against him. No matter what they put forth, Braynek is convinced that there's a trap in it somewhere, and he digs to find out what the trap is. Even if there isn't a trap, the Karna can't satisfy Braynek, because he's convinced that there has to be—somewhere. As a result, all his advice to Nordon, and all his questioning on the wildest possibilities, just serves to keep Nordon from getting unconfused. "These two men are honestly doing their best to win at the peace conference, and they've got the Karna reeling. The Karna can see that we're not trying to stall; our men are actually working at trying to reach a decision. But what the Karna don't see is that those men, as a team, are unbeatable because, in this situation, they're psychologically incapable of losing." Again the Secretary of State nodded his approval, but there was still a question in his mind. "Since you know all that, couldn't you have handled it yourself?" "Maybe, but I doubt it. They might have gotten around me someway by sneaking up on a blind spot. Nordon and Braynek have blind spots, but they're covered with armor. No, I'm glad I couldn't go; it's better this way." The Secretary of State raised an eyebrow. " Couldn't go, Mr. Ambassador?" Malloy looked at him. "Didn't you know? I wondered why you appointed me, in the first place. No, I couldn't go. The reason why I'm here, cooped up in this office, hiding from the Saarkkada the way a good Saarkkadic bigshot should, is because I like it that way. I suffer from agoraphobia and xenophobia. "I have to be drugged to be put on a spaceship because I can't take all that empty space, even if I'm protected from it by a steel shell." A look of revulsion came over his face. "And I can't stand aliens!" THE END Transcriber's Note: This etext was produced from Astounding Science Fiction March 1960. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
C. He is glad he does not have to go and things they will do a better job anyway.
|
Which immunoglobulin class showed a complete deficiency in Mrs. Sanders?
Choose the correct answer from the following options:
A. IgE
B. IgD
C. IgG
D. IgB
E. None of the above
|
### Patient Report 0
**Dear colleague, **
We are writing to provide an update on the examination results of our
patient Mrs. Hilary Sanders, born on 08/24/1976, who presented to our
outpatient clinic on 10/09/2016.
**Diagnoses:**
- Common Variable Immunodeficiency Syndrome (CVID)
- Complete IgG deficiency
- Complete IgM deficiency
- Complete IgA deficiency
- Recurrent respiratory infections
- Idiopathic thombocytopenic purpura
- Arterial hypertension
- Initiation of subcutaneous immunoglobulin substitution therapy with
human immunoglobulin
**Medical History:** Mrs. Sanders presented with suspected previously
undiagnosed immunodeficiency. There were no reports of frequent
infections during childhood and adolescence. No increased herpes
infections. No history of pneumonia, meningitis, or other serious
infections.
**Current Presentation:** Mrs. Sanders has experienced recurrent
respiratory infections (bronchitis, pharyngitis) for about 3 years.
**Physical Examination:** She reported joint pain in the left knee and
numbness below the shoulder blade. A tendency to bruise easily. No
mucosal lesions, recurrent axillary lymph node swelling. No recurrent
fevers. No B-symptoms. No resting dyspnea, no subjective heart rhythm
disturbances, no syncope, no peripheral edema, or other signs of
cardiopulmonary decompensation.
**Immunological Diagnostics:**
- Immunoglobulins including subclasses: IgA, IgG, IgM, and all IgG
subclasses were reduced.
- Numerically unremarkable monocytes and granulocytes, lymphocytopenia
with reduced B- and NK-cells, normal CD4/CD8 ratio.
- B-lymphocyte subpopulation with numerically reduced B-cells.
- Monocytic HLA-DR expression (immune competence marker) within the
normal range.
- No evidence of acute or chronic T-cell activation.
- IL-6, LBP (Lipopolysaccharide-Binding Protein), and IL-8
post-erylysis were unremarkable, elevated s-IL-2.
- Monocytic TNF-alpha secretion after 4h LPS stimulation was
unremarkable.
- T-cell function after 24h polyvalent ConA stimulation: TNF-alpha,
IFN-gamma, IL-2, IL-4 unremarkable
**Assessment**: In the immunological diagnostics, as in previous
outpatient findings, a reduction in all major immunoglobulin classes and
subclasses was observed. Cellular immune status revealed lymphocytopenia
with reduced B- and natural killer-cells.
Further cellular immune status, including the complement system and
soluble mediators, showed no significant abnormalities except for an
elevated soluble IL-2 receptor. Given the unremarkable monocytic
TNF-alpha secretion after LPS stimulation, a significant Toll-like
Receptor 4 defect is unlikely. An antibody response to Tetanus Toxoid
was demonstrated in a vaccine titer test. Protective
pneumococcal-specific antibodies could not be detected. There were no
abnormalities in autoimmune diagnostics.
Immunofixation showed no evidence of monoclonal gammopathy.
Hypogammaglobulinemia due to enteral or renal protein loss is unlikely
in the presence of normal albumin.
Overall, the picture is consistent with Common Variable
Immunodeficiency. Formally, CVID is defined by a reduction in the major
immunoglobulin class IgG, with accompanying reduction in IgA and/or IgM,
in the absence of normal or impaired vaccine response. Due to very low
immunoglobulin levels and planned travel, determination of vaccine
response was currently omitted in the absence of therapeutic
consequence. After stable substitution, specific vaccine antibody levels
can be determined before or after vaccination, with the assumption that
stable antibody concentrations exist due to continuous immunoglobulin
substitution.
According to B-cell differentiation, it corresponds to Type Ib according
to the Freiburg Classification and Type B+smB-CD21lo according to the
Euro Classification. The classification is clinically relevant, as Type
Ia is associated with increased immunocytopenias (especially ITP and
AIH) and splenomegaly. In CVID with a high proportion (\>10%) of CD-21
low B-cells, increased granulomatous diseases and splenomegaly have also
been observed.
The indication for immunoglobulin substitution therapy exists because of
recurrent infections. The form of substitution therapy (intravenous. vs.
subcutaneous) is primarily based on patient preferences, but also on
medical conditions (concomitant diseases such as thrombocytopenia,
convenience, insurance, etc.).
**Current Recommendations:**
We propose to initiate immunoglobulin substitution therapy with Hizentra
20% (subcutaneous) at a dose of 200 ml once a week on Tuesdays. Further
information and training on subcutaneous immunoglobulin substitution
therapy will be provided by a home care nursing service.
Mrs. Sanders will remain under regular medical supervision with close
monitoring of clinical symptoms, laboratory parameters, and the
effectiveness of immunoglobulin substitution therapy. Any unexpected
side effects or changes in her condition should be reported immediately.
**Lab results:**
**Parameter** **Results** **Reference Range**
--------------------------------------- --------------- ---------------------
Sodium 141 mEq/L 132-146 mEq/L
Potassium 4.2 mEq/L 3.4-4.5 mEq/L
Calcium 2.41 mg/dL 2.15-2.50 mg/dL
Inorganic Phosphate 1.00 mg/dL 0.87-1.45 mg/dL
Selenium 0.79 µmol/L 0.60-1.50 µmol/L
Zinc 10.1 µmol/L 9.0-22.0 µmol/L
Creatinine 0.75 mg/dL 0.50-0.90 mg/dL
Estimated GFR (eGFR CKD-EPI) \>90 mL/min \>90 mL/min
Total Bilirubin 0.37 mg/dL \< 1.20 mg/dL
Albumin 4.55 g/dL 3.50-5.20 g/dL
Total Protein 6.3 g/dL 6.4-8.3 g/dL
Albumin Fraction 71.8% 55.8-66.1%
A1-Globulin 5.1% 2.9-4.9%
A2-Globulin in Serum 10.7% 7.1-11.8%
ß-Globulin in Serum 9.2% 8.4-13.1%
Gamma-Globulin in Serum 3.2% 11.1-18.8%
Immunoglobulin G 514 mg/dL 700-1600 mg/dL
Immunoglobulin A 14 mg/dL 70-400 mg/dL
Immunoglobulin M 19 mg/dL 40-230 mg/dL
Immunoglobulin E 90 kU/L 0.0-100.0 kU/L
IgG 1 299.5 mg/dL 280-800 mg/dL
IgG 2 162.7 mg/dL 115-570 mg/dL
IgG 3 49.1 mg/dL 24-125 mg/dL
IgG 4 4.0 mg/dL 5.2-125 mg/dL
Serum Immunofixation
CRP 4.8 mg/L \< 5.0 mg/L
C3 Complement 980 mg/L 900-1800 mg/L
C4 Complement 120 mg/L 100-400 mg/L
ß-2-Microglobulin 3.6 mg/L 0.8-2.2 mg/L
HBs Antigen Negative
HBc Antibody Negative
HBs Antibody Negative
Ferritin 56 µg/L 13-140 µg/L
ALT (GPT) 33 U/L \< 31 U/L
AST (GOT) 29 U/L \< 35 U/L
Alkaline Phosphatase 84 U/L 35-105 U/L
Creatine Kinase 90 U/L \< 167 U/L
CK-MB 8.3 U/L \< 24.0 U/L
Gamma-GT 40 U/L 5-36 U/L
LDH 204 U/L 135-214 U/L
Lipase 50 U/L 13-60 U/L
Cortisol 306.6 nmol/L 64.0-327.0 nmol/L
25-OH-Vitamin D3 65.3 nmol/L 50.0-150.0 nmol/L
1.25-OH-Vitamin D3 134 pmol/L 18.0-155.0 pmol/L
TSH 1.42 mU/L 0.27-4.20 mU/L
Vitamin B12 770 pg/mL 191-663 pg/mL
Folic Acid 14.6 ng/mL 4.6-18.7 ng/mL
Hemoglobin 13.9 g/dL 12.0-15.6 g/dL
Hematocrit 41.0% 35.5-45.5%
Erythrocytes 5.2 M/uL 3.9-5.2 M/uL
Leukocytes 4.13 K/uL 3.90-10.50 K/uL
Platelets 174 K/uL 150-370 K/uL
MCV 80.0 fL 80.0-99.0 fL
MCH 26.7 pg 27.0-33.5 pg
MCHC 33.6 g/dL 31.5-36.0 g/dL
RDW-CV 13.7% 11.5-15.0%
Absolute Neutrophils 2.87 K/uL 1.50-7.70 K/uL
Absolute Immature Granulocytes 0.010 K/uL \< 0.050 K/uL
Absolute Lymphocytes 0.71 K/uL 1.10-4.50 K/uL
Absolute Monocytes 0.42 K/uL 0.10-0.90 K/uL
Absolute Eosinophils 0.09 K/uL 0.02-0.50 K/uL
Absolute Basophils 0.03 K/uL 0.00-0.20 K/uL
HbA1c 4.9% \< 6.0%
HbA1c (IFCC) 30.1 mmol/mol \< 42.0
HBV Serology Result Negative
HIV1/2 Antibodies, P24 Antigen Negative
Hepatitis C Virus Antibodies in Serum Negative
**Dear colleague,**
We report the examination results of Mrs. Hilary Sanders, born on
08/24/1976 who presented at our outpatient clinic on 03/04/2017.
**Diagnoses:**
- Common Variable Immunodeficiency Syndrome (CVID)
- Complete IgG deficiency
- Complete IgM deficiency
- Complete IgA deficiency
- Recurrent respiratory infections
- Idiopathic thombocytopenic purpura
- Arterial hypertension
- Initiation of subcutaneous immunoglobulin substitution therapy human
immunoglobulin
**Immunological Diagnostics:**
- Immunoglobulins including subclasses: IgA, IgG, IgM, and all
IgG-Subclasses were reduced.
- Numerically unremarkable monocytes and granulocytes, lymphocytopenia
with reduced B- and natural killer-cells, normal CD4/CD8 ratio.
- B-lymphocyte subpopulation with numerically reduced B cells.
- Monocytic HLA-DR expression within the normal range.
- No evidence of acute or chronic T-cell activation.
- IL-6, LBP (Lipopolysaccharide-Binding Protein), and IL-8
post-erylysis were unremarkable, elevated s-IL-2.
- Monocytic TNF-alpha secretion after 4h LPS stimulation was
unremarkable.
**Assessment**: In the immunological diagnostics, as in previous
outpatient findings, a reduction in all major immunoglobulin classes and
subclasses was observed. Cellular immune status revealed lymphocytopenia
with reduced B- and natural killer-cells. The further cellular immune
status, including the complement system and soluble mediators, showed no
significant abnormalities except for an elevated soluble IL-2 receptor.
**Current Presentation:** Mrs. Sanders was again provided with detailed
information about her condition and the planned course of action. We
scheduled an appointment to initiate regular subcutaneous immunoglobulin
therapy.
**Medical History:** Mrs. Sanders received her first dose of Hizentra
20% subcutaneously as immunoglobulin substitution therapy for CVID. The
administration was well-tolerated initially, with no evidence of
significant local or systemic side effects. Mrs. Sanders was once again
informed about possible risks (especially hypersensitivity reactions)
and advised to contact us immediately in case of questions,
uncertainties, or any abnormalities. The dosing for the first four weeks
was 3x20mL Hizentra 20% subcutaneously, and from the fifth week onward,
it was changed to either 1x40mL or 2x20mL Hizentra 20% subcutaneously
per week.
In the past days, Mrs. Sanders has been experiencing a cold: runny nose,
cough (green-yellow), difficulty clearing mucus, slight fever, sinus
inflammation, sore throat, difficulty speaking, and swallowing problems.
There was no improvement.
**Physical Examination:** Reddened throat, no exudates, non-swollen
cervical lymph nodes, lung examination showed bronchitis-like breathing
sounds, no rales.
**Therapy and Progression**: Today\'s CRP is not elevated. IgGs are
still below normal. We recommended increasing immunoglobulin
substitution during the infection. The patient had difficulty finding a
suitable injection site on her abdomen. However, she reported that the
secretions were gradually becoming lighter, so she decided to wait with
the antibiotic and only use it if there was no improvement.
The patient has been receiving 3x20mL Hizentra 20% per week since her
last visit. She complained of developing skin hardening at the injection
sites, so a slower infusion time was discussed. She has been
experiencing a strong cough for several weeks without fever. No rales or
signs of pleuritis were detected on auscultation. No abnormalities were
observed on the chest X-ray. Laboratory results now show normal IgG
levels, so the dose was reduced to 2x20mL per week. A CT scan of the
thorax and abdominal ultrasound were requested.
**Chest X-ray in two planes from 03/04/2017:**
[Findings/Assessment:]{.underline} No previous images are available for
comparison. Upper mediastinum and heart appear normal, with no central
congestion. No pneumothorax, effusions, confluent infiltrates, or
significant focal lesions.
**Abdominal ultrasound on 03/04/2017:**
Hepatosplenomegaly and retroperitoneal lymphadenopathy up to 26mm.
**CT Chest/Abdomen/ from 03/04/2017:**
[Methodology]{.underline}: Digital overview radiographs. After
intravenous injection of contrast agent a 16-row CT scan of the thorax
and entire abdomen was performed in the venous contrast phase, with
primary data set reconstruction at a thickness of 1.25 mm. Multiplanar
reconstructions were created.
[Findings]{.underline}: A conventional radiographic pre-image from
11/18/2014 is available for comparison.
[Thorax]{.underline}: Normal lung parenchyma with normal vascular
markings. Small, sometimes hazy, sometimes nodular densities measuring
up to 4mm in both lower lobes and the left upper lobe. Small
pleura-adjacent density in the right lower lobe. No evidence of
confluent infiltrates. No pleural effusion or pneumothorax. Normal heart
size and configuration. Normal diameter of the thoracic aorta and
pulmonary trunk. Increased number and enlarged retroclavicular lymph
nodes on the right and left, axillary on both sides measuring up to 30mm
in diameter. Trachea and esophagus displayed normally. No hiatus hernia.
Thyroid and neck soft tissues were unremarkable, as far as depicted.
Normal thoracic soft tissue mantle. No soft tissue emphysema.
[Abdomen]{.underline}: Hepatomegaly with morphologically normal liver
parenchyma. No portal vein thrombosis. Gallbladder is unremarkable with
no calculi. Intrahepatic and extrahepatic bile ducts are not dilated.
Pancreas is normally lobulated and structured, with no dilation of the
pancreatic duct. Splenomegaly. Accessory spleen measuring approximately
20 mm in diameter. Splenic parenchyma is homogeneously contrasted in the
venous phase. Kidneys are orthotopically positioned, normal size with no
side differences, and contrasted equally on both sides. Two regularly
configured hypodense lesions in the left kidney, suggestive of
uncomplicated renal cysts. No dilation of the urinary tract, and no
evidence of stones. Adrenal glands are not visualized. Increased and
enlarged mesenteric, pararaortic, parailiacal, and inguinal lymph nodes
up to 30 mm in size. Gastrointestinal tract is displayed normally, as
far as assessable. Normal representation of major abdominal vessels. No
free intraperitoneal fluid or air.
[Osseous structures:]{.underline} No evidence of suspicious osseous
destruction. Normal soft tissue mantle.
[Assessment:]{.underline} Intrapulmonary multifocal, sometimes hazy,
sometimes nodular densities, differential diagnosis includes atypical
pneumonia. Thoracoabdominal lymphadenopathy. Hepatosplenomegaly without
suspicious lesions.
**Current Recommendations:**
- Outpatient follow-up for discussion of findings
- Continue regular subcutaneous immunoglobulin administration with
current regimen of Hizentra 20% 2x20mL/week
- Lung function test
- Gastroscopy
- In case of acute infection: increase immunoglobulin administration
- Abdominal ultrasound: annually
- H. pylori testing, e.g., breath test or H. pylori antigen in stool:
annually Seasonal influenza vaccination: annually
**Lab results upon discharge:**
**Parameter** **Results** **Reference Range**
---------------------- ------------- ---------------------
Total Protein 6.3 g/dL 6.4-8.3 g/dL
Albumin Fraction 71.8% 55.8-66.1%
A1-Globulin 5.1% 2.9-4.9%
Gamma-Globulin 3.2% 11.1-18.8%
Immunoglobulin G 188 mg/dL 700-1600 mg/dL
Immunoglobulin A 11 mg/dL 70-400 mg/dL
Immunoglobulin M 12 mg/dL 40-230 mg/dL
IgG Subclass 1 113 mg/dL 280-800 mg/dL
IgG Subclass 2 49.1 mg/dL 115-570 mg/dL
IgG Subclass 4 \<0.0 mg/dL 5.2-125 mg/dL
aPCP-IgG 7.32 mg/dL 10.00-191.20 mg/dL
aPCP-IgG2 2.74 mg/dL 4.70-89.40 mg/dL
ß-2-Microglobulin 3.6 mg/L 0.8-2.2 mg/L
LDH 224 U/L 135-214 U/L
Vitamin B12 708 pg/mL 191-663 pg/mL
Erythrocytes 5.3 M/uL 3.9-5.2 M/uL
Platelets 129 K/uL 150-370 K/uL
MCV 78.0 fL 80.0-99.0 fL
MCH 25.1 pg 27.0-33.5 pg
Absolute Lymphocytes 0.91 K/uL 1.10-4.50 K/uL
### Patient Report 1
**Dear colleague, **
We are reporting on Mrs. Hilary Sanders, born on 08/24/1976, who
presented to our Immunodeficiency Clinic on 10/06/2017.
**Diagnoses:**
- Common Variable Immunodeficiency Syndrome (CVID)
- Complete IgG deficiency
- Complete IgM deficiency
- Complete IgA deficiency
- Leukopenia and lymphopenia
- Recurrent respiratory infections
- Idiopathic thombocytopenic purpura
- Hepatosplenomegaly
- Thoracoabdominal, inguinal, and axillary lymphadenopathy
- Arterial hypertension
- Initiation of subcutaneous immunoglobulin substitution therapy human
immunoglobulin
**Medical History:** Mrs. Sanders first presented herself to our clinic,
with suspected undiagnosed immunodeficiency. Regular subcutaneous
immunoglobulin therapy with Hizentra 20% (2x20mL/week) has been
well-tolerated. Initially, there were frequent upper respiratory tract
infections with sore throat and cough. In the absence of fever, a
one-time course of Cotrim was prescribed for 7 days due to sinusitis. We
discussed Mrs. Sanders' medical history in detail, including the recent
CT findings. She has been informed about the necessity of vigilance in
case of unclear and especially persistent lymph node swellings.
Regarding the inguinal and axillary lymph nodes measuring up to 30mm in
diameter found on CT, we recommend an observational approach with
regular sonographic monitoring. There have been no significant changes
in laboratory parameters, with good IgG levels during ongoing
substitution therapy and known moderate leukopenia and lymphopenia.
During the next appointment, an additional lung function test, including
diffusion measurement, will be conducted
**Current Recommendations:**
- Outpatient follow-up, including lung function test
- Continue regular subcutaneous immunoglobulin administration with
current regimen of Hizentra 20% (2x20mL/week).
- Current gastroscopy.
<!-- -->
- In case of acute infection: increase immunoglobulin administration.
- Administer targeted, sufficiently long, and high-dose antibiotic
therapy if bacterial infections require treatment.
- Ideally, obtain material for microbiological diagnostics.
- In case of increasing diarrhea, consider outpatient stool
examinations, including Giardia lamblia and Cryptosporidium.
- Abdominal ultrasound: annually.
- Lung function test, including diffusion measurement: annually.
- H. pylori testing, e.g., breath test or H. pylori antigen in stool:
annually.
- Gastroscopy: approximately every 2-3 years, depending on previous
findings or H. pylori testing
- Chest X-ray or CT thorax: if clinical symptoms or lung function
abnormalities are observed.
- Seasonal influenza vaccination: annually.
**Lab results upon Discharge:**
**Parameter** **Results** **Reference Range**
-------------------------------- ------------- ---------------------
Sodium 141 mEq/L 132-146 mEq/L
Potassium 4.1 mEq/L 3.4-4.5 mEq/L
Creatinine (Jaffé) 0.82 mg/dL 0.50-0.90 mg/dL
Estimated GFR (eGFR CKD-EPI) \>90 \-
Total Bilirubin 0.21 mg/dL \< 1.20 mg/dL
Albumin 4.09 g/dL 3.5-5.2 g/dL
Immunoglobulin G 1025 mg/dL 700-1600 mg/dL
Immunoglobulin A 16 mg/dL 70-400 mg/dL
Immunoglobulin M 28 mg/dL 40-230 mg/dL
Free Lambda Light Chains 5.86 5.70-26.30
Free Kappa Light Chains 6.05 3.30-19.40
Kappa/Lambda Ratio 1.03 0.26-1.65
IgG Subclass 1 580.9 mg/dL 280-800 mg/dL
IgG Subclass 2 340.7 mg/dL 115-570 mg/dL
IgG Subclass 3 50.9 mg/dL 24-125 mg/dL
IgG Subclass 4 5.7 mg/dL 5.2-125 mg/dL
CRP 7.3 mg/L \< 5.0 mg/L
Haptoglobin 108 mg/dL 30-200 mg/dL
Ferritin 24 µg/L 13-140 µg/L
ALT 24 U/L \< 31 U/L
AST 37 U/L \< 35 U/L
Gamma-GT 27 U/L 5-36 U/L
Lactate Dehydrogenase 244 U/L 135-214 U/L
25-OH-Vitamin D3 91.7 nmol/L 50.0-150.0 nmol/L
Hemoglobin 13.1 g/dL 12.0-15.6 g/dL
Hematocrit 40.0% 35.5-45.5%
Red Blood Cells 5.5 M/uL 3.9-5.2 M/uL
White Blood Cells 2.41 K/uL 3.90-10.50 K/uL
Platelets 142 K/uL 150-370 K/uL
MCV 73.0 fL 80.0-99.0 fL
MCH 23.9 pg 27.0-33.5 pg
MCHC 32.7 g/dL 31.5-36.0 g/dL
MPV 10.7 fL 7.0-12.0 fL
RDW-CV 14.8% 11.5-15.0%
Absolute Neutrophils 1.27 K/uL 1.50-7.70 K/uL
Absolute Immature Granulocytes 0.000 K/uL \< 0.050 K/uL
Absolute Lymphocytes 0.67 K/uL 1.10-4.50 K/uL
Absolute Monocytes 0.34 K/uL 0.10-0.90 K/uL
Absolute Eosinophils 0.09 K/uL 0.02-0.50 K/uL
Absolute Basophils 0.04 K/uL 0.00-0.20 K/uL
Free Hemoglobin 5.00 mg/dL \< 20.00 mg/dL
**Abdominal Ultrasound on 10/06/2017:**
[Liver]{.underline}: Measures 19 cm in the MCL, homogeneous parenchyma,
no focal lesions.
[Gallbladder/Biliary Tract:]{.underline} No evidence of calculi, no
signs of inflammation, no congestion.
[Spleen]{.underline}: Measures 14 cm in diameter, homogeneous. Accessory
spleen measures 16 mm at the hilus.
[Pancreas]{.underline}: Morphologically unremarkable, as far as visible
due to intestinal gas overlay, no evidence of space-occupying processes.
Retroperitoneum: No signs of aneurysms. Enlarged retroperitoneal and
iliac lymph nodes, measuring up to approximately 2.5 cm in diameter.
[Kidneys]{.underline}: Both kidneys are of normal size (right 4.3 x 11.8
cm, left 4.6 cm x 11.9 cm). No congestion, no evidence of calculi
(stones), no evidence of space-occupying processes.
[Bladder]{.underline}: Smoothly defined and normally configured.
Minimally filled.
[Uterus]{.underline}: Size within the normal range, homogeneous.
No ascites.
[Assessment:]{.underline} Evidence of enlarged lymph nodes up to 2.5 cm
retroperitoneal and iliac. Compared to previous findings, a slight
decrease in splenomegaly.
### Patient Report 2
**Dear colleague, **
We are reporting on the examination results of our patient, Mrs. Hilary
Sanders, born on 08/24/1976, who presented herself in our
Immunodeficiency Clinic on 02/10/2018.
**Diagnoses:**
- Common Variable Immunodeficiency Syndrome (CVID)
- Complete IgG deficiency
- Complete IgM deficiency
- Complete IgA deficiency
- Recurrent respiratory infections
- Idiopathic thombocytopenic purpura
- Arterial hypertension
- Initiation of subcutaneous immunoglobulin substitution therapy human
immunoglobulin
**Medical History:** For a detailed medical history, please refer to our
previous medical records.
**Therapy and Progression:** Ongoing diarrhea in the morning, often
recurring in the afternoon. No melena, no fresh blood. Resolving
respiratory infection, positive influenza.
Currently, IgG levels remain within the target range. An increased need
for immunoglobulins is expected, especially in the third trimester of
pregnancy. Therefore, we recommend close monitoring with us during
pregnancy. Ferritin levels have further declined, indicating the need
for iron substitution. Anamnestically, there is an intolerance to oral
iron preparations.
**Recommendations:**
- Outpatient follow-up
- Early follow-up in case of infections or persistent diarrhea
- Continue regular subcutaneous immunoglobulin therapy, currently with
Hizentra 20% 2x20mL/week
- In case of increasing diarrhea, conduct outpatient stool
examinations, including testing for Giardia lamblia and
Cryptosporidium
- Pulmonary function tests including diffusion measurement: annually
- Helicobacter pylori (HP) testing: e.g., breath test or HP antigen in
stool: annually
- Gastroscopy: approximately every 2-3 years, depending on previous
findings and HP testing
- Chest X-ray or chest CT: in case of abnormal clinical presentation
or pulmonary function
- Annual seasonal influenza vaccination
### Patient Report 3
**Dear colleague, **
We are writing to provide an update on Mrs. Hilary Sanders, born on
08/24/1976, who presented to our outpatient Immunodeficiency Clinic on
04/12/2018.
**Diagnoses:**
- Common Variable Immunodeficiency Syndrome (CVID)
- Complete IgG deficiency
- Complete IgM deficiency
- Complete IgA deficiency
- Recurrent respiratory infections
- Idiopathic thombocytopenic purpura
- Arterial hypertension
- Initiation of subcutaneous immunoglobulin substitution therapy human
immunoglobulin
- Suspected CVID Enteropathy
**Medical History:** For a detailed medical history, please refer to our
previous medical records.
**Therapy and Progression:** Respiratory infection with symptoms for 3-4
weeks. No antibiotics. No significant infections since then. Hizentra
3x20 mL with good tolerance.
IgG levels remain within the target range; therefore, we recommend
continuing the current treatment unchanged.
Since the last visit, mild upper respiratory tract infections. No fever
(except for one episode of sinusitis), no antibiotics. SCIG treatment
unchanged with 3x20mL/week of Hizentra ®.
Mrs. Sanders continues to experience watery diarrhea about 5-7 times
daily. No blood in stools, no pain, no vomiting, no nausea. There has
been no clear association with specific foods observed. Current weight:
69kg.
We discussed further diagnostic steps. Initially, outpatient endoscopic
diagnostics should be performed.
**Current Recommendations:**
- Outpatient follow-up in three months
- Continue SCIG treatment as is
- External upper gastrointestinal endoscopy and colonoscopy (please
return with findings)
- In case of increasing diarrhea, conduct outpatient stool
examinations, including testing for Giardia lamblia and
Cryptosporidium
- Abdominal ultrasound: annually
- Pulmonary function tests including diffusion measurement: annually
- Helicobacter pylori testing: e.g., breath test or Helicobacter
pylori antigen in stool: annually
- Gastroscopy: approximately every 2-3 years, depending on previous
findings and Helicobacter pylori testing
- Chest X-ray or chest CT: in case of abnormal clinical presentation
or pulmonary function
- Annual seasonal influenza vaccination
### Patient Report 4
**Dear colleague, **
We are writing to provide an update on Mrs. Hilary Sanders, born on
08/24/1976, who presented to our outpatient Immunodeficiency Clinic on
02/18/2019.
**Diagnoses:**
- Common Variable Immunodeficiency Syndrome (CVID)
- Complete IgG deficiency
- Complete IgM deficiency
- Complete IgA deficiency
- Recurrent respiratory infections
- Idiopathic thombocytopenic purpura
- Arterial hypertension
- Initiation of subcutaneous immunoglobulin substitution therapy human
immunoglobulin
- Suspected CVID Enteropathy
**Medical History:** For a detailed medical history, please refer to our
previous medical records.
**Therapy and Progression:** Respiratory infections with symptoms for 7
weeks. No antibiotics. No significant infections since then. Hizentra
3x20 mL with good tolerance. Continued diarrhea, approximately 6 times a
day, without weight loss. IgG levels remain within the target range;
therefore, we recommend continuing the current treatment unchanged.
We discussed further diagnostic steps. Initially, outpatient endoscopic
diagnostics should be performed.
**Current Recommendations:**
- Outpatient follow-up in three months
- Continue treatment as is
- External upper gastrointestinal endoscopy (and colonoscopy (please
return with findings)
- In case of increasing diarrhea, conduct outpatient stool
examinations, including testing for Giardia lamblia and
Cryptosporidium
- Abdominal ultrasound: annually
- Pulmonary function tests including diffusion measurement: annually
- Helicobacter pylori (HP) testing: e.g., breath test or HP antigen in
stool: annually
- Gastroscopy: approximately every 2-3 years, depending on previous
findings and HP testing
- Chest X-ray or chest CT: in case of abnormal clinical presentation
or pulmonary function
- Annual seasonal influenza vaccination
### Patient Report 5
**Dear colleague, **
We are writing to provide summary on the clinical course of Mrs. Hilary
Sanders, born on 08/24/1976, who presented at our outpatient
Immunodeficiency Clinic.
**Diagnoses:**
- Common Variable Immunodeficiency Syndrome (CVID)
- Complete IgG deficiency
- Complete IgM deficiency
- Complete IgA deficiency
- Recurrent respiratory infections
- Idiopathic thombocytopenic purpura
- Arterial hypertension
- Initiation of subcutaneous immunoglobulin substitution therapy human
immunoglobulin
- Suspected CVID Enteropathy
- Iron-deficiency anemia
**Medical History:** For a detailed medical history, please refer to our
previous medical records.
**Therapy and Progression:** Overall stable condition. No longer
experiencing cough. Persistent fatigue. Upcoming appointment with the
Gastroenterology department next week. There is again an indication for
iron substitution.
**Update on 11/15/2019: Laboratory results from 11/15/2019:**
Transaminase elevation, Protein 18, markedly elevated BNP. However, IgA
is at 0.5 (otherwise not detectable), IgG subclasses within normal
range. Findings do not align. Patient informed by phone, returning for
further evaluation today; also screening for Hepatitis A, B, C, and E,
EBV, CMV, TSH, coagulation. No shortness of breath, no edema, no
abdominal enlargement, stable weight at 69 kg. In case of worsening
symptoms, shortness of breath, or fever, immediate referral to the
emergency department recommended.
**02/12/2020:** The patient is doing reasonably well. She has had a mild
cold for about 2 weeks, no fever, but nasal congestion and
yellowish-green sputum. No other infections. No antibiotics prescribed.
She has adapted to her gastrointestinal issues. An appointment with the
Gastroenterology department. She is currently working from home.
Medication: no new medications, only Cuvitru 20mL 3x weekly. Weight
remains stable at 67 kg. The last lung function test was in the summer
of this year and was within normal limits. Imaging has not been
performed recently. Gastroscopy and colonoscopy have not been conducted
for some time.
**04/14/2020:** Referral to Gastroenterology at is recommended for
persistent abdominal symptoms.
**10/24/2020:** The patient has mostly avoided social contacts due to
the pandemic. She continues to experience digestive problems (food
intolerances, diarrhea, flatulence). She has less stamina. Few
infections in the past year, at most a minor cold. No significant
infections. Hizentra injections remain unchanged at 20 mL 3 times a
week.
**03/22/2021:** Constant colds since December 2020. One-time antibiotic
treatment in October 2019. Subcutaneous Immunoglobulin therapy remains
unchanged at 20 mL 3 times weekly.
**09/19/2021:** She feels disoriented and very tired, more so than
usual. Difficulty maintaining a steady gaze. No steroid therapy was
administered. CT showed enlarged lymph nodes. Diarrhea, especially in
the morning, 3-4 times a day, additional bowel movements with meals,
sometimes watery. No fever, no infections. Hizentra injections continued
unchanged.
**Summary**: IgG levels are currently within the target range, so we
recommend continuing immunoglobulin substitution therapy without
changes. The antibody response (SARS-CoV-2 (S-Ag) IgG ELISA) to the
Covid-19 vaccination is, as expected, negative. However, there is a
positive detection of SARS-CoV-2 (N-Ag) IgG ELISA, as expected in the
case of viral contact (not vaccination). We consider this to be an
unspecific reaction and recommend further monitoring at the next
follow-up appointment. With a platelet count currently at 55 K/uL, we
recommend a short-term blood count check with us or your primary care
physician.
Due to the immunodeficiency, a lack of antibody response to vaccination
was expected. In the medium term, passive protection through
immunoglobulin substitution therapy will play a role. This is contingent
on a significant portion of plasma donors having antibodies against
SARS-CoV2. There is a multi-month delay from the time of donation to the
release of the preparations, so we anticipate that meaningful protection
through immunoglobulin products will not be expected. An exact prognosis
in this regard is not possible.
**Current Recommendations:**
- Outpatient follow-up in three months
- Consultation with Gastroenterology
- Continue SCIG treatment as is
- External upper gastrointestinal endoscopy and colonoscopy (please
return with findings)
- In case of increasing diarrhea, conduct outpatient stool
examinations, including testing for Giardia lamblia and
Cryptosporidium
- Abdominal ultrasound: annually
- Pulmonary function tests including diffusion measurement: annually
- Helicobacter pylori (HP) testing: e.g., breath test or HP antigen in
stool: annually
- Gastroscopy: approximately every 2-3 years, depending on previous
findings and HP testing
- Chest X-ray or chest CT: in case of abnormal clinical presentation
or pulmonary function
- Annual seasonal influenza vaccination
### Patient Report 6
**Dear colleague, **
We are providing you with an update regarding our patient Mrs. Hilary
Sanders, born on 08/24/1976. She was under our inpatient care from
03/29/2023 to 04/05/2023.
**Diagnoses:**
- Suspected CVID-Associated enteropathy
- Known hepatosplenomegaly with a borderline enlarged portal vein, no
significant portocaval shunts. Multiple liver lesions, possibly
hemangiomas further evaluation if not already done.
- Known retroperitoneal and iliac lymphadenopathy, likely related to
the underlying condition.
- Known changes in the lower lung bases, likely associated with the
underlying condition, e.g., ILD. Refer to previous examinations.
- Capsule endoscopy: Incomplete capsule enteroscopy with no evidence
of inflammatory changes. Some hyperemia and blurry vascular pattern
observed in the visible colon.
- CVID-Associated Hepatopathy in the Form of Nodular Regenerative
Hyperplasia
**Other Diagnoses:** Common Variable Immunodeficiency Syndrome (CVID)
with:
- Complete IgG deficiency
- Complete IgM deficiency
- Complete IgA deficiency
- Leukopenia and lymphopenia
- Initiation of subcutaneous immunoglobulin substitution therapy with
Hizentra 20%
- Infectious manifestations: Frequent respiratory tract infections
- Non-Infectious manifestations:
- ITP (Immune Thrombocytopenia)
- Hepatosplenomegaly
- Lymphadenopathy in supraclavicular, infraclavicular,
thoracoabdominal, inguinal, and axillary regions
- Suspected Granulomatous-Lymphocytic Interstitial Lung Disease in
CVID
<!-- -->
- Iron-deficiency anemia
**Pysical Examination:** Patient in normal general condition and
nutritional status (175 cm, 65.8 kg. No resting dyspnea.
[Neuro (grossly orienting):]{.underline} awake, oriented to
time/place/person/situation, No evidence of focal neurological deficit.
No meningism.
[Head/neck]{.underline}: pharynx non-irritable. Moist, rosy mucous
membranes. Tongue occupied.
[Skin]{.underline}: intact, turgor normal, no icterus, no cyanosis.
[Thorax]{.underline}: normal configuration, no spinal palpitation, renal
bed clear.
[Lung]{.underline}: vesicular breath sound bds, no accessory sounds,
sonorous tapping sound bds.
[Cor]{.underline}: Cardiac action pure, rhythmic, no vitia typical
murmurs.
[Abdomen]{.underline}: regular bowel sounds, soft abdominal wall, no
tenderness, no resistances, no hepatosplenomegaly.
[Extremities]{.underline}: no edema. Feet warm. Dorsalis pedis +/+ and
posterior tibial artery +/+.
**Current Presentation:** The patient was admitted for further
evaluation of suspected CVID-associated enteropathy, as she had been
experiencing chronic diarrhea for the past three years. On admission,
the patient reported an overall good general and nutritional condition.
She described her current subjective well-being as good but mentioned
having chronic diarrhea for the past three years, with up to 7 bowel
movements per day. The stools were watery without any signs of blood.
There were no indications of infection, such as fever, chills, dysuria,
hematuria, cough, sputum, or dyspnea. She also experienced intermittent
left-sided upper abdominal pain, primarily postprandially. She had a
good appetite.
On the day of admission, an esophagogastroduodenoscopy was performed,
which revealed erythematous antral gastritis. Additionally, there was an
approximately 1 cm irregular mucosal area at the corpus-antrum junction
on the greater curvature side. A magnetic resonance imaging scan showed
no evidence of inflamed bowel loops, ruling out chronic inflammatory
bowel disease or celiac disease. To further investigate, a capsule
endoscopy was performed, with results pending at the time of discharge.
Hypovitaminosis B12 and folate deficiency were ruled out. However,
iron-deficiency anemia was confirmed, and the patient had already
scheduled an outpatient appointment for iron substitution. Serum levels
of vitamin B6 and zinc were pending at discharge.
Due to a moderate increase in transaminases and evidence of
hepatosplenomegaly, we decided, after detailed explanation and with the
patient\'s consent, to perform a sonographically guided liver biopsy in
addition to the planned endoscopy. The differential diagnosis included
CVID-associated hepatopathy. The biopsy was successfully conducted ,
without any post-interventional bleeding. Histology revealed mild acute
hepatitis and nodular regenerative hyperplasia.This finding could be
consistent with changes in CVID-associated hepatopathy. Granulomas were
not observed. With only slightly elevated liver values, a trial therapy
with budesonide was initiated, and clinical (diarrhea?) and laboratory
(transaminases?) follow-up will be performed in the outpatient setting.
We discharged Mrs. Sanders in a cardiopulmonarily stable condition.
[Current Recommendations:]{.underline}
- Follow-up in the gastroenterological outpatient clinic
**Esophagogastroduodenoscopy (EGD) on 04/01/2023:** Introduction of the
gastroscope in a left lateral position. Visualized up to the descending
part of the duodenum. Unremarkable upper esophageal sphincter. Normal
motility and mucosa in the upper, middle, and distal esophagus. The
Z-line is sharply demarcated in the hiatus. The cardia closes
sufficiently. The stomach expands normally in all parts under air
insufflation. Multiple glandular cysts \< 8 mm in size in the fundus and
corpus. Approximately 1 cm irregular mucosal area at the corpus-antrum
junction on the greater curvature side. Streaky redness of the mucosa in
the antrum. Unremarkable mucosa in the bulb. Unremarkable mucosa in the
descending part of the duodenum. Step biopsies performed.
[Summary]{.underline}: Erythematous antral gastritis. Approximately 1 cm
irregular mucosal area at the corpus-antrum junction on the greater
curvature side, suggestive of inflammation. Multiple glandular cysts
observed in the fundus and corpus.
[Abdominal MRI on 04/02/2023:]{.underline}
[Clinical information, questions, and justification for the
exam]{.underline}: Chronic diarrhea, suspected CVID-associated
enteropathy, differential diagnosis of celiac disease, and inflammatory
bowel disease (IBD). Assessment of malignancy.
Technique: After oral administration of mannitol solution and injection
of 40 mg Buscopan, a 3-Tesla abdominal MRI was performed.
[Findings]{.underline}: Multiple nodular consolidations and opacities
detected in the lower basal lung segments, measuring 7 x 4 mm, for
example, in the right lateral lower lobe (Series 18, Image 3).
Additionally, streaky-reticular changes observed. Left diaphragmatic
elevation. Liver globally enlarged and smooth-bordered with several
lesions showing mild to moderately hyperintense signals in T2-weighted
images and hypointense signals in T1-weighted images. These lesions
demonstrated increased enhancement in the early contrast phases,
especially those at the periphery, and more diffuse enhancement in the
late phases. For example, a lesion measuring 12 x 11 mm in Segment 2, a
lesion measuring 8 mm in Segment 8 and a lesion measuring 21 x 13 mm in
Segment 7. The portal vein measures borderline wide, up to 15 mm in
diameter. Gallbladder is unremarkable without evidence of stones. Intra-
and extrahepatic bile ducts are not dilated. Spleen significantly
enlarged, measuring 14 cm in pole-to-pole distance and 7.2 cm in
transverse diameter, homogeneous enhancement in native phases and late
contrast phase. Large accessory spleen located hilarly. Bilateral
adrenal glands appear slender. Pancreas displays typical appearance with
no ductal dilatation. Both kidneys are in orthotopic position, with
unremarkable cortical cysts on the right side. No signs of urinary
obstruction. The urinary bladder is moderately filled. No free fluid.
Adequate dilation of small bowel loops. No evidence of significant bowel
obstruction. No thickened bowel walls or increased post-contrast signal
in the bowel loops. Cystic lesion in the right ovary measuring 17 x 11
mm consistent with a corpus luteum cyst. Multiple enlarged
retroperitoneal lymph nodes observed, for example, paracaval node with a
short-axis diameter of 14 mm and right iliacoexternal node with a
short-axis diameter of 14.5 mm No evidence of enlarged mesenteric or
inguinal lymph nodes.
|
IgG
|
What is the reason behind the drop in performance using BERT for some popular task?
|
### Introduction
In recent times, pre-trained contextual language models have led to significant improvement in the performance for many NLP tasks. Among the family of these models, the most popular one is BERT BIBREF0, which is also the focus of this work. The strength of the BERT model FIGREF2 stems from its transformerBIBREF1 based encoder architectureFIGREF1. While it is still not very clear as to why BERT along with its embedding works so well for downstream tasks when it is fine tuned, there has been some work in this direction that that gives some important cluesBIBREF2, BIBREF3. At a high level, BERT’s pipelines looks as follows: given a input sentence, BERT tokenizes it using wordPiece tokenizerBIBREF4. The tokens are then fed as input to the BERT model and it learns contextualized embeddings for each of those tokens. It does so via pre-training on two tasks - Masked Language Model (MLM)BIBREF0 and Next Sentence Prediction (NSP)BIBREF0. The focus of this work is to understand the issues that a practitioner can run into while trying to use BERT for building NLP applications in industrial settings. It is a well known fact that NLP applications in industrial settings often have to deal with the noisy data. There are different kinds of possible noise namely non-canonical text such as spelling mistakes, typographic errors, colloquialisms, abbreviations, slang, internet jargon, emojis, embedded metadata (such as hashtags, URLs, mentions), non standard syntactic constructions and spelling variations, grammatically incorrect text, mixture of two or more languages to name a few. Such noisy data is a hallmark of user generated text content and commonly found on social media, chats, online reviews, web forums to name a few. Owing to this noise a common issue that NLP models have to deal with is Out Of Vocabulary (OOV) words. These are words that are found in test and production data but not part of training data. In this work we highlight how BERT fails to handle Out Of Vocabulary(OOV) words, given its limited vocabulary. We show that this negatively impacts the performance of BERT when working with user generated text data and evaluate the same. This evaluation is motivated from the business use case we are solving where we are building a dialogue system to screen candidates for blue collar jobs. Our candidate user base, coming from underprivileged backgrounds, are often high school graduates. This coupled with ‘fat finger’ problem over a mobile keypad leads to a lot of typos and spelling mistakes in the responses sent to the dialogue system. Hence, for this work we focus on spelling mistakes as the noise in the data. While this work is motivated from our business use case, our findings are applicable across various use cases in industry - be it be sentiment classification on twitter data or topic detection of a web forum. To simulate noise in the data, we begin with a clean dataset and introduce spelling errors in a fraction of words present in it. These words are chosen randomly. We will explain this process in detail later. Spelling mistakes introduced mimic the typographical errors in the text introduced by our users. We then use the BERT model for tasks using both clean and noisy datasets and compare the results. We show that the introduction of noise leads to a significant drop in performance of the BERT model for the task at hand as compared to clean dataset. We further show that as we increase the amount of noise in the data, the performance degrades sharply. ### Related Work
In recent years pre-trained language models ((e.g. ELMoBIBREF5, BERTBIBREF0) have made breakthroughs in several natural language tasks. These models are trained over large corpora that are not human annotated and are easily available. Chief among these models is BERTBIBREF0. The popularity of BERT stems from its ability to be fine-tuned for a variety of downstream NLP tasks such as text classification, regression, named-entity recognition, question answeringBIBREF0, machine translationBIBREF6 etc. BERT has been able to establish State-of-the-art (SOTA) results for many of these tasks. People have been able to show how one can leverage BERT to improve searchBIBREF7. Owing to its success, researchers have started to focus on uncovering drawbacks in BERT, if any. BIBREF8 introduce TEXTFOOLER, a system to generate adversarial text. They apply it to NLP tasks of text classification and textual entailment to attack the BERT model. BIBREF9 evaluate three models - RoBERTa, XLNet, and BERT in Natural Language Inference (NLI) and Question Answering (QA) tasks for robustness. They show that while RoBERTa, XLNet and BERT are more robust than recurrent neural network models to stress tests for both NLI and QA tasks; these models are still very fragile and show many unexpected behaviors. BIBREF10 discuss length-based and sentence-based misclassification attacks for the Fake News Detection task trained using a context-aware BERT model and they show 78% and 39% attack accuracy respectively. Our contribution in this paper is to answer that can we use large language models like BERT directly over user generated data. ### Experiment
For our experiments, we use pre-trained BERT implementation as given by huggingface transformer library. We use the BERTBase uncased model. We work with three datasets namely - IMDB movie reviewsBIBREF11, Stanford Sentiment Treebank (SST-2) BIBREF12 and Semantic Textual Similarity (STS-B) BIBREF13. IMDB dataset is a popular dataset for sentiment analysis tasks, which is a binary classification problem with equal number of positive and negative examples. Both STS-B and SST-2 datasets are a part of GLUE benchmark[2] tasks . In STS-B too, we predict positive and negative sentiments. In SST-2 we predict textual semantic similarity between two sentences. It is a regression problem where the similarity score varies between 0 to 5. To evaluate the performance of BERT we use standard metrics of F1-score for imdb and STS-B, and Pearson-Spearman correlation for SST-2. In Table TABREF5, we give the statistics for each of the datasets. We take the original datasets and add varying degrees of noise (i.e. spelling errors to word utterances) to create datasets for our experiments. From each dataset, we create 4 additional datasets each with varying percentage levels of noise in them. For example from IMDB, we create 4 variants, each having 5%, 10%, 15% and 20% noise in them. Here, the number denotes the percentage of words in the original dataset that have spelling mistakes. Thus, we have one dataset with no noise and 4 variants datasets with increasing levels of noise. Likewise, we do the same for SST-2 and STS-B. All the parameters of the BERTBase model remain the same for all 5 experiments on the IMDB dataset and its 4 variants. This also remains the same across other 2 datasets and their variants. For all the experiments, the learning rate is set to 4e-5, for optimization we use Adam optimizer with epsilon value 1e-8. We ran each of the experiments for 10 and 50 epochs. ### Results
Let us discuss the results from the above mentioned experiments. We show the plots of accuracy vs noise for each of the tasks. For IMDB, we fine tune the model for the sentiment analysis task. We plot F1 score vs % of error, as shown in Figure FIGREF6. Figure FIGREF6imdba shows the performance after fine tuning for 10 epochs, while Figure FIGREF6imdbb shows the performance after fine tuning for 50 epochs. Similarly, Figure FIGREF9ssta and Figure FIGREF9sstb) shows F1 score vs % of error for Sentiment analysis on SST-2 dataset after fine tuning for 10 and 50 epochs respectively. Figure FIGREF12stsa and FIGREF12stsb shows Pearson-Spearman correlation vs % of error for textual semantic similarity on STS-B dataset after fine tuning for 10 and 50 epochs respectively. ### Results ::: Key Findings
It is clear from the above plots that as we increase the percentage of error, for each of the three tasks, we see a significant drop in BERT’s performance. Also, from the plots it is evident that the reason for this drop in performance is introduction of noise (spelling mistakes). After all we get very good numbers, for each of the three tasks, when there is no error (0.0 % error). To understand the reason behind the drop in performance, first we need to understand how BERT processes input text data. BERT uses WordPiece tokenizer to tokenize the text. WordPiece tokenizer utterances based on the longest prefix matching algorithm to generate tokens . The tokens thus obtained are fed as input of the BERT model. When it comes to tokenizing noisy data, we see a very interesting behaviour from WordPiece tokenizer. Owing to the spelling mistakes, these words are not directly found in BERT’s dictionary. Hence WordPiece tokenizer tokenizes noisy words into subwords. However, it ends up breaking them into subwords whose meaning can be very different from the meaning of the original word. Often, this changes the meaning of the sentence completely, therefore leading to substantial dip in the performance. To understand this better, let us look into two examples, one each from the IMDB and STS-B datasets respectively, as shown below. Here, (a) is the sentence as it appears in the dataset ( before adding noise) while (b) is the corresponding sentence after adding noise. The mistakes are highlighted with italics. The sentences are followed by the corresponding output of the WordPiece tokenizer on these sentences: In the output ‘##’ is WordPiece tokenizer’s way of distinguishing subwords from words. ‘##’ signifies subwords as opposed to words. Example 1 (imdb example): “that loves its characters and communicates something rather beautiful about human nature” (0% error) “that loves 8ts characters abd communicates something rathee beautiful about human natuee” (5% error) Output of wordPiece tokenizer: ['that', 'loves', 'its', 'characters', 'and', 'communicate', '##s', 'something', 'rather', 'beautiful', 'about', 'human','nature'] (0% error IMDB example) ['that', 'loves', '8', '##ts', 'characters', 'abd', 'communicate','##s', 'something','rat', '##hee', 'beautiful', 'about', 'human','nat', '##ue', '##e'] (5% error IMDB example) Example 2(STS example): “poor ben bratt could n't find stardom if mapquest emailed himpoint-to-point driving directions.” (0% error) “poor ben bratt could n't find stardom if mapquest emailed him point-to-point drivibg dirsctioge.” (5% error) Output of wordPiece tokenizer: ['poor', 'ben', 'brat', '##t', 'could', 'n', "'", 't', 'find','star', '##dom', 'if', 'map', '##quest', 'email', '##ed', 'him','point', '-', 'to', '-', 'point', 'driving', 'directions', '.'] (0% error STS example) ['poor', 'ben', 'brat', '##t', 'could', 'n', "'", 't', 'find','star', '##dom', 'if', 'map', '##quest', 'email', '##ed', 'him', 'point', '-', 'to', '-', 'point', 'dr', '##iv', '##ib','##g','dir','##sc', '##ti', '##oge', '.'] (5% error STS example) In example 1, the tokenizer splits communicates into [‘communicate’, ‘##s’] based on longest prefix matching because there is no exact match for “communicates” in BERT vocabulary. The longest prefix in this case is “communicate” and left over is “s” both of which are present in the vocabulary of BERT. We have contextual embeddings for both “communicate” and “##s”. By using these two embeddings, one can get an approximate embedding for “communicates”. However, this approach goes for a complete toss when the word is misspelled. In example 1(b) the word natuee (‘nature’ is misspelled) is split into ['nat', '##ue', '##e'] based on the longest prefix match. Combining the three embeddings one cannot approximate the embedding of nature. This is because the word nat has a very different meaning (it means ‘a person who advocates political independence for a particular country’). This misrepresentation in turn impacts the performance of downstream subcomponents of BERT bringing down the overall performance of BERT model. Hence, as we systematically introduce more errors, the quality of output of the tokenizer degrades further, resulting in the overall performance drop. Our results and analysis shows that one cannot apply BERT blindly to solve NLP problems especially in industrial settings. If the application you are developing gets data from channels that are known to introduce noise in the text, then BERT will perform badly. Examples of such scenarios are applications working with twitter data, mobile based chat system, user comments on platforms like youtube, reddit to name a few. The reason for the introduction of noise could vary - while for twitter, reddit it's often deliberate because that is how users prefer to write, while for mobile based chat it often suffers from ‘fat finger’ typing error problem. Depending on the amount of noise in the data, BERT can perform well below expectations. We further conducted experiments with different tokenizers other than WordPiece tokenizer. For this we used stanfordNLP WhiteSpace BIBREF14 and Character N-gram BIBREF15 tokenizers. WhiteSpace tokenizer splits text into tokens based on white space. Character N-gram tokenizer splits words that have more than n characters in them. Thus, each token has at most n characters in them. The resultant tokens from the respective tokenizer are fed to BERT as inputs. For our case, we work with n = 6. Results of these experiments are presented in Table TABREF25. Even though wordPiece tokenizer has the issues stated earlier, it is still performing better than whitespace and character n-gram tokenizer. This is primarily because of the vocabulary overlap between STS-B dataset and BERT vocabulary. ### Conclusion and Future Work
In this work we systematically studied the effect of noise (spelling mistakes) in user generated text data on the performance of BERT. We demonstrated that as the noise increases, BERT’s performance drops drastically. We further investigated the BERT system to understand the reason for this drop in performance. We show that the problem lies with how misspelt words are tokenized to create a representation of the original word. There are 2 ways to address the problem - either (i) preprocess the data to correct spelling mistakes or (ii) incorporate ways in BERT architecture to make it robust to noise. The problem with (i) is that in most industrial settings this becomes a separate project in itself. We leave (ii) as a future work to fix the issues. Figure 1: BERT architecture [1] Figure 2: The Transformer model architecture [2] Table 1: Number of utterances in each datasets Figure 3: F1 score vs % of error for Sentiment analysis on IMDB dataset Figure 4: F1 score vs % of error for Sentiment analysis on SST-2 data Figure 5: Pearson-Spearman correlation vs % of error for textual semantic similarity on STS-B dataset Table 2: Comparative results on STS-B dataset with different tokenizers
|
Hence WordPiece tokenizer tokenizes noisy words into subwords. However, it ends up breaking them into subwords whose meaning can be very different from the meaning of the original word. Often, this changes the meaning of the sentence completely, therefore leading to substantial dip in the performance.
|
Why was each inhabitant of the moon-town only referred to as their specific species rather than a distinct name?
A. They were all distinct by their light, and only needed to be referred to as their species.
B. The population was much too large to name each creature.
C. The humans of moon-town felt no need to waste time in naming each living creature as they died off too quickly.
D. There was only one of each, therefore, they were called by their species.
|
IT WAS A DULL, ROUTINE LITTLE WORLD. IT DIDN'T EVEN HAVE A CITY. EVERYTHING IT HAD WAS IN THE GARDEN BY R. A. LAFFERTY [Transcriber's Note: This etext was produced from Worlds of If Science Fiction, March 1961. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The protozoic recorder chirped like a bird. Not only would there be life traces on that little moon, but it would be a lively place. So they skipped several steps in the procedure. The chordata discerner read Positive over most of the surface. There was spinal fluid on that orb, rivers of it. So again they omitted several tests and went to the cognition scanner. Would it show Thought on the body? Naturally they did not get results at once, nor did they expect to; it required a fine adjustment. But they were disappointed that they found nothing for several hours as they hovered high over the rotation. Then it came—clearly and definitely, but from quite a small location only. "Limited," said Steiner, "as though within a pale. As though there were but one city, if that is its form. Shall we follow the rest of the surface to find another, or concentrate on this? It'll be twelve hours before it's back in our ken if we let it go now." "Let's lock on this one and finish the scan. Then we can do the rest of the world to make sure we've missed nothing," said Stark. There was one more test to run, one very tricky and difficult of analysis, that with the Extraordinary Perception Locator. This was designed simply to locate a source of superior thought. But this might be so varied or so unfamiliar that often both the machine and the designer of it were puzzled as to how to read the results. The E. P. Locator had been designed by Glaser. But when the Locator had refused to read Positive when turned on the inventor himself, bad blood developed between machine and man. Glaser knew that he had extraordinary perception. He was a much honored man in his field. He told the machine so heatedly. The machine replied, with such warmth that its relays chattered, that Glaser did not have extraordinary perception; he had only ordinary perception to an extraordinary degree. There is a difference , the machine insisted. It was for this reason that Glaser used that model no more, but built others more amenable. And it was for this reason also that the owners of Little Probe had acquired the original machine so cheaply. And there was no denying that the Extraordinary Perception Locator (or Eppel) was a contrary machine. On Earth it had read Positive on a number of crack-pots, including Waxey Sax, a jazz tootler who could not even read music. But it had also read Positive on ninety per cent of the acknowledged superior minds of the Earth. In space it had been a sound guide to the unusual intelligences encountered. Yet on Suzuki-Mi it had read Positive on a two-inch-long worm, only one of them out of billions. For the countless identical worms no trace of anything at all was shown by the test. So it was with mixed expectations that Steiner locked onto the area and got a flick. He then narrowed to a smaller area (apparently one individual, though this could not be certain) and got very definite action. Eppel was busy. The machine had a touch of the ham in it, and assumed an air of importance when it ran these tests. Finally it signaled the result, the most exasperating result it ever produces: the single orange light. It was the equivalent of the shrug of the shoulders in a man. They called it the "You tell me light." So among the intelligences there was at least one that might be extraordinary, though possibly in a crackpot way. It is good to be forewarned. "Scan the remainder of the world, Steiner," said Stark, "and the rest of us will get some sleep. If you find no other spot then we will go down on that one the next time it is in position under us, in about twelve hours." "You don't want to visit any of the other areas first? Somewhere away from the thoughtful creature?" "No. The rest of the world may be dangerous. There must be a reason that thought is in one spot only. If we find no others then we will go down boldly and visit this." So they all, except Steiner, went off to their bunks then: Stark, the Captain; Gregory Gilbert, the executive officer; Wolfgang Langweilig, the engineer; Casper Craig, super-cargo, tycoon and 51% owner of the Little Probe, and F. R. Briton, S.J., a Jesuit priest who was linguist and checker champion of the craft. Dawn did not come to the moon-town. The Little Probe hovered stationary in the light and the moon-town came up under the dawn. Then the Probe went down to visit whatever was there. "There's no town," said Steiner. "Not a building. Yet we're on the track of the minds. There's nothing but a meadow and some boscage, a sort of fountain or pool, and four streams coming out of it." "Keep on towards the minds," said Stark. "They're our target." "Not a building, not two sticks or stones placed together. That looks like an Earth-type sheep there. And that looks like an Earth-lion, I'm almost afraid to say. And those two ... why, they could well be Earth-people. But with a difference. Where is that bright light coming from?" "I don't know, but they're right in the middle of it. Land here. We'll go to meet them at once. Timidity has never been an efficacious tool with us." Well, they were people. And one could only wish that all people were like them. There was a man and a woman, and they were clothed either in very bright garments or in no garments at all, but only in a very bright light. "Talk to them, Father Briton," said Stark. "You are the linguist." "Howdy," said the priest. He may or may not have been understood, but the two of them smiled at him, so he went on. "Father Briton from Philadelphia," he said, "on detached service. And you, my good man, what is your handle, your monicker, your tag?" "Ha-Adamah," said the man. "And your daughter, or niece?" It may be that the shining man frowned momentarily at this; but the woman smiled, proving that she was human. "The woman is named Hawwah," said the man. "The sheep is named sheep, the lion is named lion, the horse is named horse and the hoolock is named hoolock." "I understand. It is possible that this could go on and on. How is it that you use the English tongue?" "I have only one tongue; but it is given to us to be understood by all; by the eagle, by the squirrel, by the ass, by the English." "We happen to be bloody Yankees, but we use a borrowed tongue. You wouldn't have a drink on you for a tubful of thirsty travellers, would you?" "The fountain." "Ah—I see." But the crew all drank of the fountain to be sociable. It was water, but water that excelled, cool and with all its original bubbles like the first water ever made. "What do you make of them?" asked Stark. "Human," said Steiner. "It may even be that they are a little more than human. I don't understand that light that surrounds them. And they seem to be clothed, as it were, in dignity." "And very little else," said Father Briton, "though that light trick does serve a purpose. But I'm not sure they'd pass in Philadelphia." "Talk to them again," said Stark. "You're the linguist." "That isn't necessary here, Captain. Talk to them yourself." "Are there any other people here?" Stark asked the man. "The two of us. Man and woman." "But are there any others?" "How would there be any others? What other kind of people could there be than man and woman?" "But is there more than one man or woman?" "How could there be more than one of anything?" The captain was a little puzzled by this, but he went on doggedly: "Ha-Adamah, what do you think that we are? Are we not people?" "You are not anything till I name you. But I will name you and then you can be. You are named Captain. He is named Priest. He is named Engineer. He is named Flunky." "Thanks a lot," said Steiner. "But are we not people?" persisted Captain Stark. "No. We are the people. There are no people but two. How could there be other people?" "And the damnest thing about it," muttered Langweilig, "is, how are you going to prove him wrong? But it does give you a small feeling." "Can we have something to eat?" asked the Captain. "Pick from the trees," said Ha-Adamah, "and then it may be that you will want to sleep on the grass. Being not of human nature (which does not need sleep or rest), it may be that you require respite. But you are free to enjoy the garden and its fruits." "We will," said Captain Stark. They wandered about the place, but they were uneasy. There were the animals. The lion and lioness were enough to make one cautious, though they offered no harm. The two bears had a puzzling look, as though they wanted either to frolic with you or to mangle you. "If there are only two people here," said Casper Craig, "then it may be that the rest of the world is not dangerous at all. It looked fertile wherever we scanned it, though not so fertile as this central bit. And those rocks would bear examining." "Flecked with gold, and possibly with something else," said Stark. "A very promising site." "And everything grows here," added Steiner. "Those are Earth-fruits and I never saw finer. I've tasted the grapes and plums and pears. The figs and dates are superb, the quince is as flavorsome as a quince can be, the cherries are excellent. And I never did taste such oranges. But I haven't yet tried the—" and he stopped. "If you're thinking what I'm afraid to think," said Gilbert, "then it will be the test at least: whether we're having a pleasant dream or whether this is reality. Go ahead and eat one." "I won't be the first to eat one. You eat." "Ask him first. You ask him." "Ha-Adamah, is it allowed to eat the apples?" "Certainly. Eat. It is the finest fruit in the garden." "Well, the analogy breaks down there," said Stark. "I was almost beginning to believe in the thing. But if it isn't that, then what. Father Briton, you are the linguist, but in Hebrew does not Ha-Adamah and Hawwah mean—?" "Of course they do. You know that as well as I." "I was never a believer. But would it be possible for the exact same proposition to maintain here as on Earth?" "All things are possible." And it was then that Ha-Adamah, the shining man, gave a wild cry: "No, no. Do not approach it. It is not allowed to eat of that one!" It was the pomegranate tree, and he was warning Langweilig away from it. "Once more, Father," said Stark, "you should be the authority; but does not the idea that it was the apple that was forbidden go back only to a medieval painting?" "It does. The name of the fruit is not mentioned in Genesis. In Hebrew exegesis, however, the pomegranate is usually indicated." "I thought so. Question the man further, Father. This is too incredible." "It is a little odd. Adam, old man, how long have you been here?" "Forever less six days is the answer that has been given to me. I never did understand the answer, however." "And have you gotten no older in all that time?" "I do not understand what 'older' is. I am as I have been from the beginning." "And do you think that you will ever die?" "To die I do not understand. I am taught that it is a property of fallen nature to die, and that does not pertain to me or mine." "And are you completely happy here?" "Perfectly happy according to my preternatural state. But I am taught that it might be possible to lose that happiness, and then to seek it vainly through all the ages. I am taught that sickness and ageing and even death could come if this happiness were ever lost. I am taught that on at least one other unfortunate world it has actually been lost." "Do you consider yourself a knowledgeable man?" "Yes, since I am the only man, and knowledge is natural to man. But I am further blessed. I have a preternatural intellect." Then Stark cut in once more: "There must be some one question you could ask him, Father. Some way to settle it. I am becoming nearly convinced." "Yes, there is a question that will settle it. Adam, old man, how about a game of checkers?" "This is hardly the time for clowning," said Stark. "I'm not clowning, Captain. How about it, Adam? I'll give you choice of colors and first move." "No. It would be no contest. I have a preternatural intellect." "Well, I beat a barber who was champion of Germantown. And I beat the champion of Morgan County, Tennessee, which is the hottest checker center on Earth. I've played against, and beaten, machines. But I never played a preternatural mind. Let's just set up the board, Adam, and have a go at it." "No. It would be no contest. I would not like to humble you." They were there for three days. They were delighted with the place. It was a world with everything, and it seemed to have only two inhabitants. They went everywhere except into the big cave. "What is there, Adam?" asked Captain Stark. "The great serpent lives there. I would not disturb him. He has long been cranky because plans he had for us did not materialize. But we are taught that should ever evil come to us, which it cannot if we persevere, it will come by him." They learned no more of the real nature of the sphere in their time there. Yet all but one of them were convinced of the reality when they left. And they talked of it as they took off. "A crowd would laugh if told of it," said Stark, "but not many would laugh if they had actually seen the place, or them. I am not a gullible man, but I am convinced of this: that this is a pristine and pure world and that ours and all the others we have visited are fallen worlds. Here are the prototypes of our first parents before their fall. They are garbed in light and innocence, and they have the happiness that we have been seeking for centuries. It would be a crime if anyone disturbed that happiness." "I too am convinced," said Steiner. "It is Paradise itself, where the lion lies down with the lamb, and where the serpent has not prevailed. It would be the darkest of crimes if we or others should play the part of the serpent, and intrude and spoil." "I am probably the most skeptical man in the world," said Casper Craig the tycoon, "but I do believe my eyes. I have been there and seen it. It is indeed an unspoiled Paradise; and it would be a crime calling to the wide heavens for vengeance for anyone to smirch in any way that perfection. "So much for that. Now to business. Gilbert, take a gram: Ninety Million Square Miles of Pristine Paradise for Sale or Lease. Farming, Ranching, exceptional opportunities for Horticulture. Gold, Silver, Iron, Earth-Type Fauna. Terms. Special Rates for Large Settlement Parties. Write, Gram, or call in person at any of our planetary offices as listed below. Ask for Brochure—Eden Acres Unlimited." Down in the great cave that Old Serpent, a two-legged one among whose names were "Snake-Oil Sam," spoke to his underlings: "It'll take them fourteen days to get back with the settlers. We'll have time to overhaul the blasters. We haven't had any well-equipped settlers for six weeks. It used to be we'd hardly have time to strip and slaughter and stow before there was another batch to take care of." "I think you'd better write me some new lines," said Adam. "I feel like a goof saying those same ones to each bunch." "You are a goof, and therefore perfect for the part. I was in show business long enough to know never to change a line too soon. I did change Adam and Eve to Ha-Adamah and Hawwah, and the apple to the pomegranate. People aren't becoming any smarter—but they are becoming better researched, and they insist on authenticity. "This is still a perfect come-on here. There is something in human nature that cannot resist the idea of a Perfect Paradise. Folks will whoop and holler to their neighbors to come in droves to spoil and mar it. It isn't greed or the desire for new land so much—though that is strong too. Mainly it is the feverish passion to befoul and poison what is unspoiled. Fortunately I am sagacious enough to take advantage of this trait. And when you start to farm a new world on a shoestring you have to acquire your equipment as you can." He looked proudly around at the great cave with its mountains and tiers of materials, heavy machinery of all sorts, titanic crates of foodstuff space-sealed; wheeled, tracked, propped, vaned and jetted vehicles; and power packs to run a world. He looked at the three dozen space ships stripped and stacked, and at the rather large pile of bone-meal in one corner. "We will have to have another lion," said Eve. "Bowser is getting old, and Marie-Yvette abuses him and gnaws his toes. And we do have to have a big-maned lion to lie down with the lamb." "I know it, Eve. The lion is a very important prop. Maybe one of the crackpot settlers will bring a new lion." "And can't you mix another kind of shining paint? This itches. It's hell." "I'm working on it." Casper Craig was still dictating the gram: "Amazing quality of longevity seemingly inherent in the locale. Climate ideal. Daylight or half-light. All twenty-one hours from Planet Delphina and from Sol. Pure water for all industrial purposes. Scenic and storied. Zoning and pre-settlement restrictions to insure congenial neighbors. A completely planned globular settlement in a near arm of our own galaxy. Low taxes and liberal credit. Financing our specialty—" "And you had better have an armed escort when you return," said Father Briton. "Why in cosmos would we want an armed escort?" "It's as phony as a seven-credit note!" "You, a man of the cloth doubt it? And us ready skeptics convinced by our senses? Why do you doubt?" "It is only the unbelieving who believe so easily in obvious frauds. Theologically unsound, dramaturgically weak, philologically impossible, zoologically rigged, salted conspicuously with gold and shot through with anachronisms. And moreover he was afraid to play me at checkers." "What?" "If I have a preternatural intellect I wouldn't be afraid of a game of checkers with anyone. Yet there was an unusual mind there somewhere; it was just that he chose not to make our acquaintance personally." "They looked at the priest thoughtfully. "But it was Paradise in one way," said Steiner at last. "How?" "All the time we were there the woman did not speak."
|
D. There was only one of each, therefore, they were called by their species.
|
When was Atezolizumab/Bevacizumab first administered to Mr. Wells?
Choose the correct answer from the following options:
A. 07/27/2019
B. 08/24/2019
C. 01/25/2022
D. 08/18/2019
E. 10/26/2021
|
### Patient Report 0
**Dear colleague, **
We report to you on Mr. Paul Wells, born on 04/02/1953, who was in our
inpatient treatment from 07/26/2019 to 07/28/2019.
**Diagnoses:** Suspected multifocal HCC segment IV, VII/VIII, first
diagnosed: 07/19.
- COPD, current severity level Gold III.
- Pulmonary emphysema, respiratory partial insufficiency with home
oxygen.
- Postnasal drip syndrome
**Current Presentation:** The elective presentation of Mr. Wells was
made in accordance with the decision of the interdisciplinary liver
board of 07/20/2019 for further diagnostics in the case of multiple
malignoma-specific hepatic space demands.
**Medical History: **In brief, Mr. Wells presented to the Medical Center
St. Luke's with persistent right-sided pain in the upper abdomen.
Computer tomography showed multiple intrahepatic masses of the right
liver lobe (SIV, SVII/VIII). For diagnostic clarification of the
malignoma-specific findings, the patient was presented to our liver
outpatient clinic. The tumor marker diagnostics have not been
conclusive. Analogous to the recommendation of the liver board, a liver
puncture, staging, and endoscopic exclusion of a primary in the
gastrointestinal tract should be initiated.
**Physical Examination:** Physical examination reveals an alert patient.
- Oral mucosa: Moist and rosy, no plaques typical of thrush, no
plaques typical of herpes.
- Hear: Heart sounds pure, rhythmic, normofrequency.
- Lungs: Laterally attenuated breath sound with wheezing.
- Abdomen: Abdomen soft, regular bowel sounds over all 4 quadrants, no
defensive tension, no resistances, diffuse pressure pain over the
upper abdomen. No renal tap pain, no spinal tap pain. Spleen
palpable under the costal arch.
- Extremities: No edema, freely movable
- Neurology: GCS 15, pupils directly and indirectly reactive to light,
no flapping tremor. No meningism.
**Therapy and Progression:** Mr. Wells presented an age-appropriate
general status and cardiopulmonary stability. Anamnestically, there was
no evidence of an acute infection. Skin or scleral icterus and pruritus
were denied. No B symptoms. No stool changes, no dysuria. There would be
regular alcohol consumption of about 3-4 beers a day, as well as
nicotine abuse (120 PY). The general performance in COPD Gold grade III
was strongly limited, with a walking distance reduced to 100m due to
dyspnea. He had a home oxygen demand with 4L/min O2 during the day, up
to 6L/min under load. At night, 2L/min O2. The last colonoscopy was
performed 4 years ago, with no anamnestic abnormalities. No known
allergies. Family history is positive for colorectal cancer (mother).
Clinical examination revealed the typical auscultation findings of
advanced COPD with attenuated breath sounds bilateral, with
hyperinflation and clear wheezing. Otherwise, there were no significant
findings. Laboratory chemistry did not reveal any higher-grade
abnormalities. On the day of admission, after detailed clarification,
the patient was able to undergo the complication-free sonographically
guided puncture of the liver cavity in SIV. Thereby, two punch cylinders
were preserved for histopathological processing. Histologically, the
findings presented as infiltrates of a macrotrabecular and
pseudoglandular growing, well-differentiated hepatocellular carcinoma
(G1). The postinterventional course was unremarkable. In particular, no
clinical or laboratory signs were found for bleeding.
CT staging revealed a size constant known in the short term.
Hypervascularized hepatic space demands in both lobes of the liver
without further malignancy suspect thoracoabdominal tumor detection and
without metastasis aspects. MR also revealed the large, partly exophytic
growing, partly centrally hemorrhaged HCC lesions in S3/4 and S7/8 to
the illustration. In addition, complete enforcement of the left lobe of
the liver was evident with smaller satellites and macroinvasion of the
left portal vein branch. There was a low cholestasis of the left biliary
system. Gastroscopy and colonoscopy were also performed. Here, a reflux
esophagitis, sigmoid diverticulosis, multiple colonic diverticula, and a
4mm polyp were removed from the sigmoid colon to prevent bleeding; a
hemoclip was applied. Histologically, no adenoma was found. An
appointment to discuss the findings in our HCC outpatient clinic has
been arranged. We recommend further therapy preparation and the
performance of an echocardiography.
We were able to discharge Mr. Wells on 7/28/19.
**Addition:**
**Ultrasound on 07/26/2019 10:15 AM:**
- Indication: Targeted liver puncture for suspected metastatic liver
malignancy
- Organ puncture: Quick: 114%, PTT: 28 s, and platelets: 475 G/L. A
valid declaration of consent is available. According to the patient,
he does not receive antiplatelet drugs.
- In segment IV, an approximately 8.3 x 6 cm echo-depleted mass with
central cystic fusion is accessible in the dorsal position of a
sonographically guided puncture at 6.5 cm puncture depth. After
extensive skin disinfection, local anesthesia with 10 mL Mecaine 1%
and puncture incision with a scalpel. Repeated puncture with 18 G
Magnum needles is performed. Two approximately 1 cm fragile whitish
cylinders obtained for histologic examination. Band-aid dressing.
- **Assessment:** Hepatic space demand
**MRI of the liver plain + contrast agent from 07/26/2019 1:15 PM:**
**Technique**: Coronary and axial T2 weighted sequences, axial
diffusion-weighted EPI sequence with ADC map (b: 0, 50, 300 and 600
s/mmÇ), axial dynamic T1 weighted sequences with Dixon fat suppression
and (liver-specific) contrast agent (Dotagraf/Primovist); slice
thickness: 4 mm. Premedication with 2 mL Buscopan.
**Liver**: Centrally hemorrhagic masses observed in liver segments 4, 7,
and 8 demonstrate T2 hyperintensity, marked diffusion restriction,
arterial phase enhancement, and venous phase washout. These
characteristics are congruent with histopathological diagnosis of
hepatocellular carcinoma. The largest lesion in segment 4 exhibits
pronounced exophytic growth but no evidence of organ invasion. Notably,
branches of the mammary arteries penetrate directly into the tumor.
Diffusion-weighted imaging further reveals disseminated foci throughout
the entire left hepatic lobe. Disruption of the peripheral left portal
vein branch indicative of macrovascular invasion, accompanied by
peripheral cholestasis in the left biliary system.
**Biliary Tract:** Bile ducts are emphasized on both left and right
sides, with no evidence of mechanical obstruction in drainage. The
common hepatic duct remains non-dilated.
**Pancreas and Spleen:** Both organs exhibit no abnormalities.
**Kidneys:** Normal signal characteristics observed.
**Bone Marrow:** Signal behavior is within normal limits.
Assessment: Radiological features highly suggestive of hepatocellular
carcinoma in liver segments 4, 7, and 8, with evidence of macrovascular
invasion and peripheral cholestasis in the left biliary system. No signs
of organ invasion or biliary obstruction. Pancreas, spleen, kidneys, and
bone marrow appear unremarkable.
**Assessment:**
Large liver lesions, some exophytic and some centrally hemorrhagic, are
observed in segments 3/4 and 7/8.
In addition, the left lobe of the liver is completely involved with
smaller satellite lesions and macroinvasion of the left portal branch.
Mild cholestasis of the left biliary system is noted.
Dilated bile ducts are also found on the right side with no apparent
mechanical obstruction to outflow.
**CT Chest/Abdomen/Pelvis with contrast agent from 07/27/2019 2:00 PM:**
**Clinical Indication:** Evaluation of an unclear liver lesion
(approximately 9 cm) in a patient with severe COPD. No prior
liver-related medical history.
**Question:** Are there any suspicious lesions in the liver?
**Pre-recordings:** Previous external CT abdomen dated 09/13/2021.
**Findings:**
**Technique:** CT imaging involved a multi-line spiral CT through the
chest, abdomen, and pelvis in the venous contrast phase. Oral contrast
agent with Gastrolux 1:33 in water was administered. Thin-layer
reconstructions and coronary and sagittal secondary reconstructions were
performed.
**Chest:** No axillary or mediastinal lymphadenopathy is observed. There
is marked coronary sclerosis, as well as calcification of the aortic and
mitral valves. Nonspecific nodules smaller than 2 mm are noted in the
posterolateral lower lobe on the right side and lateral middle lobe. No
pneumonic infiltrates are observed. There is reduced aeration with
presumed additional scarring changes at the base of the lung
bilaterally, along with centrilobular emphysema.
**Abdomen:** Known exophytic liver lesions are confirmed, with
involvement in segment III extending to the subhepatic region (0.1 cm
extension) and a 6 cm lesion in segment VIII. Further spotty
hypervascularized lesions are observed throughout the left lobe of the
liver. No pathological dilatation of intra- or extrahepatic bile ducts
is seen, and there is no evidence of portal vein thrombosis. There are
no pathologically enlarged lymph nodes at the hepatic portal,
retroperitoneal, or inguinal regions. No ascites or pneumoperitoneum is
noted. There is no pancreatic duct congestion, and the spleen is not
enlarged. Additionally, there is a Bosniak 1 left renal cyst measuring
3.6 cm. Pronounced sigmoid diverticulosis is observed, with no evidence
of other masses in the gastrointestinal tract. Skeletal imaging reveals
no malignancy-specific osteodestructions but shows ventral pontifying
spondylophytes of the thoracic spine with no fractures.
**Assessment:**
Short-term size-constant known hypervascularized hepatic space lesions
are present in both lobes of the liver.
No other malignancy-susceptible thoracoabdominal tumor evidence is
found, and there are no metastasis-specific lymph nodes.
**Gastroscopy from 07/28/2019**
**Findings:**
**Esophagus:** Unobstructed intubation of the esophageal orifice under
visualization. Mucosa appears inconspicuous, with the Z-line at 37 cm
and measuring less than 5 mm. Small mucosal lesions are observed but do
not straddle mucosal folds.
**Stomach:** The gastric lumen is completely distended under air
insufflation. There are streaky changes in the antrum, while the fundus
and cardia appear regular on inversion. The pylorus is inconspicuous and
passable.
**Duodenum:** Good development of the bulbus duodeni is noted, with good
insight into the pars descendens duodeni. The mucosa appears overall
inconspicuous.
**Assessment:** Findings suggest reflux esophagitis (Los Angeles
Classification Grade A) and antrum gastritis.
**Colonoscopy from 07/28/2019**
**Findings:**
**Colon:** Some residual fluid contamination is noted in the sigmoid
(Boston Bowel Preparation Scale \[BBPS\] 8). There is pronounced sigmoid
diverticulosis, along with multiple colonic diverticula. A 4mm polyp in
the lower sigma (Paris IIa, NICE 1) is observed and ablated with a cold
snare, with hemoclip application for bleeding prophylaxis. Other mucosal
findings appear inconspicuous, with normal vascular markings. There is
no indication of inflammatory or malignant processes.
**Maximum Insight:** Terminal ileum.
**Anus:** Inspection of the anal region reveals no pathological
findings. Palpation is inconspicuous, and the mucosa is smooth and
displaceable, with no resistance and no blood on the glove.
**Assessment:** Polypectomy was performed for sigmoid diverticulosis and
a colonic diverticulum, with histology revealing minimally hyperplastic
colorectal mucosa and no evidence of malignancy.
**Pathology from 08/27/2019**
**Clinical Information/Question:**
**Macroscopy:** Unclear liver tumor: numerous tissue samples up to a
maximum of 0.7 cm in size. Complete embedding.
Processing: One tissue block processed and stained with Hematoxylin and
Eosin (H&E), Gomori\'s trichrome, Iron stain, Diastase Periodic
Acid-Schiff (D-PAS), and Van Gieson stain.
**Microscopic Findings:**
- Liver architecture is presented in fragmented liver core biopsies
with observable lobular structures and two included portal fields.
- Hepatic trabeculae are notably wider than the typical 2-3 cell
width, featuring the formation of druse-like luminal structures.
- Sinusoidal dilatation is markedly observed.
- Hepatocytes show mildly enlarged nuclei with minimal cytologic
atypia and isolated mitotic figures.
- Gomori staining reveals a notable, partial loss of the fine
reticulin fiber network.
- Adjacent areas show fibrosed liver parenchyma containing hemosiderin
pigmentation.
- No significant evidence of parenchymal fatty degeneration is
observed.
**Assessment**: Histologic features indicative of marked sinusoidal
dilatation, trabecular widening, and partial loss of reticulin network,
alongside minimally atypical hepatocytes and fibrosed parenchyma with
hemosiderin pigment. No significant hepatic fat degeneration noted.
### Patient Report 1
**Dear colleague, **
We would like to report on Paul Wells, born on 04/02/1953, who was under
our outpatient treatment on 08/24/2019.
**Diagnoses:**
- Multifocal HCC (Hepatocellular Carcinoma) involving segments IV,
VII/VIII, with portal vein invasion, classified as BCLC C, diagnosed
in July 2019.
- Extensive HCC lesions, some exophytic and others centrally
hemorrhagic, in segments S3/4 and S7/8, complete involvement of the
left liver lobe with smaller satellite lesions, and macrovascular
invasion of the left portal vein.
- Histology from 07/27/2019: A well-differentiated hepatocellular
carcinoma (G1) with a macrotrabecular and pseudoglandular growth
pattern.
- Decision from the Liver Tumor Board on 08/18/2019: Recommending
systemic therapy.
- Initiation of Atezolizumab/Bevacizumab on 08/24/2019
- Liver fibrosis: Elevated alcohol consumption (3-4 beers/day).
**Other Diagnoses:**
- COPD with a current severity level of Gold III.
- Pulmonary emphysema.
- Respiratory partial insufficiency requiring home oxygen therapy.
- Postnasal Drip Syndrome.
- History of nicotine use (120 pack-years).
- Hypertension (high blood pressure).
**Medical History:** Mr. Wells presented with persistent right upper
abdominal pain and was initially treated at St. Luke\'s Medical Center.
CT scans revealed multiple intrahepatic lesions in the right liver lobe
(SIV, SVII/VIII). Short-term follow-up CT staging revealed a known,
size-stable, hypervascularized hepatic lesion in both lobes of the
liver, with no evidence of other thoracoabdominal malignancies or
suspicious lymph nodes. MRI also confirmed the presence of large HCC
lesions, some exophytic and others centrally hemorrhagic, in segments
S3/4 and S7/8, along with complete infiltration of the left liver lobe
with smaller satellite lesions and macroinvasion of the left portal
vein. There was mild cholestasis in the left biliary system.
**Current Recommendations: **
- Liver function remains good based on laboratory tests.
- Mr. Wells has been extensively informed about systemic therapy
options with Atezolizumab/Bevacizumab and the possibility of
alternative therapy with a tyrosine kinase inhibitor.
- The decision has been made to initiate standard first-line therapy
with Atezolizumab/Bevacizumab. Detailed information regarding
potential side effects has been provided, with particular emphasis
on the need for immediate medical evaluation in case of signs of
gastrointestinal bleeding (blood in stool, black tarry stool, or
vomiting blood) or worsening pulmonary symptoms.
- The patient has been strongly advised to abstain from alcohol
completely.
- A follow-up evaluation through liver MRI and CT has been scheduled
for January 4, 2020, at our HCC (Hepatocellular Carcinoma) clinic.
The exact appointment time will be communicated to the patient
separately.
- We are available for any questions or concerns.
- In case of persistent or worsening symptoms, we recommend an
immediate follow-up appointment.
### Patient Report 2
**Dear colleague, **
We would like to provide an update regarding Mr. Paul Wells, born on
04/02/1953, who was under our inpatient care from 08/13/2020 to
08/14/2020.
**Medical History:**
We assume familiarity with Mr. Wells\'s comprehensive medical history as
described in the previous referral letter. At the time of admission, he
reported significantly reduced physical performance due to his known
severe COPD. Following the consensus of the Liver Board, we admitted Mr.
Wells for a SIRT simulation.
**Current Presentation:** Mr. Wells is a 66-year-old patient with normal
consciousness and reduced general condition. He is largely compensated
on 3 liters of oxygen per minute. His abdomen is soft with regular
peristalsis. A palpable tumor mass in the right upper abdomen is noted.
**DSA Coeliac-Mesenteric on 08/13/2020:**
- Uncomplicated SIRT simulation.
- Catheter position 1: Right hepatic artery.
- Catheter position 2: Left hepatic artery.
- Catheter position 3: Liver segment arteries 4a/4b.
- Uncomplicated and technically successful embolization of parasitic
tumor supply from the inferior and superior epigastric arteries.
**Perfusion Scintigraphy of the Liver and Lungs, including SPECT/CT on
08/13/2020:**
- The liver/lung shunt volume is 9.4%.
- There is intense radioactivity accumulation in multiple lesions in
both the right and left liver lobes.
**Therapy and Progression:** On 08/13/2020, we performed a DSA
coeliac-mesenteric angiography on Mr. Wells, administering a total of
approximately 159 MBq Tc99m-MAA into the liver\'s arterial circulation
(simulation). This procedure revealed that a significant portion of
radioactivity would reach the lung parenchyma during therapy, posing a
risk of worsening his already compromised lung function. In view of
these comorbidities, SIRT was not considered a viable treatment option.
Therefore, an interdisciplinary decision was made during the conference
to recommend systemic therapy. With an uneventful course, we discharged
Mr. Wells in stable general condition on 08/14/2020.
### Patient Report 3
**Dear colleague, **
We are reporting on Paul Wells, born on 04/02/1953, who presented to our
interdisciplinary clinic for Hepato- and Cholangiocellular Tumors on
10/24/2020.
**Diagnoses:**
- Multifocal HCC Segment with portal vein invasion, BCLC C, first
diagnosed 07/19
- Large, partly exophytic, partly centrally hemorrhagic HCC lesions in
S3/4 and S7/8, complete infiltration of the left lateral lobe with
smaller satellites, macrovascular invasion of the left portal vein.
- Histology on 07/27/2019: Macrotrabecular and pseudoglandular growth
of well-differentiated hepatocellular carcinoma (G1).
- Histology from 07/27/2019: A well-differentiated hepatocellular
carcinoma (G1) with a macrotrabecular and pseudoglandular growth
pattern.
- Decision from the Liver Tumor Board on 08/18/2019: Recommending
systemic therapy.
- Initiation of Atezolizumab/Bevacizumab on 08/24/2019.
- Liver fibrosis: Elevated alcohol consumption (3-4 beers/day).
- CT in 01/2020: Very good tumor response.
- Re-administration of Atezolizumab/Bevacizumab on 01/25/2022 and
02/16/2022, followed by a treatment pause due to limited tolerance.
- CT from 02/2020 to 08/2020: Continuously regressing tumor findings.
- Liver fibrosis: Increased C2 consumption (3-4 beers/day).
**Other Diagnoses:**
- Suspected Polyneuropathy or Restless Legs Syndrome
- COPD, current severity Gold III.
- Pulmonary emphysema
- Respiratory partial insufficiency with home oxygen
- Postnasal-Drip Syndrome
- History of nicotine abuse (120 py)
- Transient worsening of lung function with steroid requirement after
Atezolizumab/Bevacizumab administrations
- History of severe pneumonia (Medical Center St. Luke's) in 10/2019
- Pneumogenic sepsis with detection of Streptococcus pneumoniae
- Arterial hypertension
- Atrial fibrillation
- Treatment with Apixaban
- Reflux esophagitis Grade A (Esophagogastroduodenoscopy in 08/2019).
**Current Presentation**: Mr. Wells presented to discuss follow-up after
systemic therapy with Atezolizumab/Bevacizumab due to his impaired
general condition.
**Medical History:** For detailed medical history, please refer to the
previous medical reports. In summary, Mr. Wells presented in 07/2019
with persistent right upper abdominal pain. A CT scan showed multiple
intrahepatic lesions in the right liver lobe (SIV, SVII/VIII). MR
imaging also revealed large, partly exophytic, partly centrally
hemorrhagic HCC lesions in S3/4 and S7/8. There was complete
infiltration of the left liver lobe with smaller satellites and
macroinvasion of the left portal vein branch. Histology confirmed a
well-differentiated hepatocellular carcinoma (G1). There is no known
underlying liver disease, but peritumoral liver fibrosis was observed
histologically. Mr. Wells reported increased alcohol consumption of 3-4
beers per day.
Due to comorbidities and a large tumor with a relatively high liver-lung
shunt, SIRT simulation was initially attempted but found to be an
unsuitable treatment option. Therefore, our interdisciplinary liver
tumor board recommended systemic therapy. After comprehensive
counseling, treatment with Atezolizumab/Bevacizumab commenced on
08/24/2019.
The therapy had to be paused after a single administration due to a
substantial increase in transaminases (GPT 164 U/L, GOT 151 U/L),
suspected to be associated with immunotherapy-induced hepatitis. With
only minimal improvement in transaminases, Prednisolone therapy was
initiated on and tapered successfully after significant transaminase
regression. However, before the next planned administration, the patient
experienced severe pneumonic sepsis, requiring hospitalization on
10/2019. Following discharge, there was a recurrent infection requiring
inpatient antibiotic therapy.
Staging examinations in 01/2020 showed a very good tumor response.
Subsequently, Atezolizumab/Bevacizumab was re-administered on 01/23/2020
and 02/14/2020. However, in the following days, the patient experienced
significant side effects, including oral burning, appetite and weight
loss, low blood pressure, and worsening pulmonary status. Steroid
treatment improved the pulmonary situation, but due to poor tolerance,
therapy was paused after 02/14/2020.
Currently, Mr. Wells reports a satisfactory general condition, although
his pulmonary function remains limited but stable.
**Summary:** Laboratory results from external testing on 01/02/2020
indicate excellent liver function, with transaminases within normal
range. The latest CT examination shows continued tumor regression.
However, MRI quality is limited due to the patient\'s inability to hold
their breath adequately. Given the excellent tumor response and previous
significant side effects, it was decided to continue the treatment pause
until the next tumor staging.
**Current Recommendations:** A follow-up imaging appointment has been
scheduled for four months from now. We kindly request you send the
latest CT images (Chest/Abdomen/Pelvis, including dynamic liver CT) and
current blood values to our HCC clinic. Due to limited assessability,
another MRI is not advisable.
We remain at your disposal for any further inquiries. In case of
persistent or worsened symptoms, we recommend prompt reevaluation.
**Medication upon discharge:**
**Medication** **Dosage** **Frequency**
------------------------------------- ------------ -------------------------
Ipratropium/Fenoterol (Combivent) As needed As needed
Beclomethasone/Formoterol (Fostair) 6+200 mcg 2-0-2
Tiotropium (Spiriva) 2.5 mcg 2-0-0
Prednisolone (Prelone) 5 mg 2-0-0 (or as necessary)
Pantoprazole (Protonix) 40 mg 1-0-0
Fenoterol 0.1 mg As needed
Apixaban (Eliquis) 5 mg On hold
Olmesartan (Benicar) 20 mg 1-0-0
Lab results upon Discharge:
**Parameter** **Results** **Reference Range**
----------------------------- ------------- ---------------------
Sodium (Na) 144 mEq/L 134-145 mEq/L
Potassium (K) 3.7 mEq/L 3.4-5.2 mEq/L
Calcium (Ca) 2.37 mEq/L 2.15-2.65 mEq/L
Chloride (Cl) 106 mEq/L 95-112 mEq/L
Inorganic Phosphate (PO4) 0.93 mEq/L 0.8-1.5 mEq/L
Transferrin Saturation 20 % 16-45 %
Magnesium 0.78 mEq/L 0.75-1.06 mEq/L
Creatinine 1.88 mg/dL \<1.2 mg/dL
GFR 36 mL/min \<90 mL/min
BUN 60 mg/dL 14-46 mg/dL
Uric Acid 4.6 mg/dL 3.0-6.9 mg/dL
Total Bilirubin 0.5 mg/dL \<1 mg/dL
Albumin 4.0 g/dL 3.6-5.0 g/dL
Total Protein 6.8 g/dL 6.5-8.7 g/dL
CRP 0.19 mg/dL \<0.5 mg/dL
Transferrin 269 mg/dL 200-360 mg/dL
Ferritin 110 mcg/L 30-300 mcg/L
ALT 339 U/L \<45 U/L
AST 424 U/L \<50 U/L
GGT 904 U/L \<55 U/L
Lipase 61 U/L \<70 U/L
Thyroid-Stimulating Hormone 0.54 mIU/L 0.27-4.20 mIU/L
Hemoglobin 14.5 g/dL 14.0-17.5 g/dL
Hematocrit 43 % 40-52 %
Red Blood Cells 4.60 M/µL 4.6-6.2 M/µL
White Blood Cells 8.78 K/µL 4.5-11.0 K/µL
Platelets 205 K/µL 150-400 K/µL
MCV 94 fL 81-100 fL
MCH 31.5 pg 27-34 pg
MCHC 33.5 g/dL 32.4-35.0 g/dL
MPV 11 fL 7-12 fL
RDW 14.8 % 11.9-14.5 %
Neutrophils 3.72 K/µL 1.8-7.7 K/µL
Lymphocytes 2.37 K/µL 1.4-3.7 K/µL
Monocytes 0.93 K/µL 0.2-1.0 K/µL
Eosinophils 1.67 K/µL \<0.7 K/µL
Basophils 0.09 K/µL 0.01-0.10 K/µL
Erythroblasts Negative \<0.01 K/µL
Antithrombin Activity 85 % 80-120 %
### Patient Report 4
**Dear colleague, **
We are reporting an update of the medical condition of Mr. Paul Wells
born on 04/02/1953, who presented for a follow up in our outpatient
clinic on 11/20/2020.
**Diagnoses:**
- Multifocal HCC Segment with portal vein invasion, BCLC C, first
diagnosed 07/19
- Large, partly exophytic, partly centrally hemorrhagic HCC lesions in
S3/4 and S7/8, complete infiltration of the left lateral lobe with
smaller satellites, macrovascular invasion of the left portal vein.
- Histology on 07/27/2019: Macrotrabecular and pseudoglandular growth
of well-differentiated hepatocellular carcinoma (G1).
- SIRT simulation: No feasible SIRT.
- Liver Tumor Board decision on 08/18/2019: Systemic therapy.
- Atezolizumab/Bevacizumab since 10/26/2021, with a pause starting on
09/17/2019, due to transaminase elevation.
- CT in 01/2020: Very good tumor response.
- Re-administration of Atezolizumab/Bevacizumab
- CT from 02/2020 to 08/2020: Continuously regressing tumor findings.
- Liver fibrosis: Increased alcohol consumption (3-4 beers/day).
**Other diagnoses:**
- COPD, current severity Gold III.
- Pulmonary emphysema.
- Respiratory partial insufficiency with home oxygen.
- Postnasal-Drip Syndrome.
- History of nicotine abuse (120 py).
- Transient worsening of lung function with steroid requirement after
Atezolizumab/Bevacizumab administrations
- Pneumogenic sepsis with detection of Streptococcus pneumonia
- Arterial hypertension.
- Atrial fibrillation
- Treatment with Apixaban.
- Reflux esophagitis LA Grade A (Esophagogastroduodenoscopy in
08/2019).
**Medical History:** Mr. Wells initially presented with right upper
abdominal pain, which led to the discovery of multiple intrahepatic
masses in liver segments IV, VII/VIII. Subsequent investigations
confirmed the diagnosis of HCC. He also suffers from chronic obstructive
pulmonary disease (COPD), emphysema, and respiratory insufficiency
requiring home oxygen therapy. Previous investigations and treatments
were documented in detail in our previous medical records.
**Physical Examination:**
- General Appearance: Alert, cooperative, and oriented.
- Vital Signs: Stable blood pressure, heart rate, respiratory rate,
and temperature. Oxygen Saturation (SpO2): Within the normal range.
- Respiratory System: Normal chest symmetry, no accessory muscle use.
Clear breath sounds, no wheezing or crackles. Regular respiratory
rate.
- Cardiovascular System: Regular heart rate and rhythm, no murmurs.
Strong radial and pedal pulses bilaterally. No lower extremity
edema.
- Gastrointestinal System: Soft, nontender abdomen. Bowel sounds
present in all quadrants. Spleen palpable under the costal arch.
- Neurological Examination: Alert and oriented. Cranial nerves, motor,
sensory, reflexes, coordination and gait normal. No focal
neurological deficits.
- Skin and Mucous Membranes: Intact skin, no rashes or lesions. Moist
oral mucosa without lesions.
- Extremities: No edema. Full range of motion in all joints. Normal
capillary refill.
- Lymphatic System:
- No palpable lymphadenopathy.
**MRI Liver (plain + contrast agent) on 11/20/2020 09:01 AM.**
- Imaging revealed stable findings in the liver. The previously
identified HCC lesions in segments IV, VII/VIII, including their
size and characteristics, remained largely unchanged. There was no
evidence of new lesions or metastases. Detailed MRI imaging provided
valuable insight into the nature of the lesions, their vascularity,
and possible effects on adjacent structures.
**CT Chest/Abdomen/Pelvis with contrast agent on 11/20/2020 12:45 PM.**
- Thoracoabdominal CT scan showed the same results as the previous
examination. Known space-occupying lesions in the liver remained
stable, and there was no evidence of malignancy or metastasis
elsewhere in the body. The examination also included a thorough
evaluation of the thoracic and pelvic regions to rule out possible
metastasis.
**Gastroscopy on 11/20/2020 13:45 PM.**
- Gastroscopy follow-up confirmed the previous diagnosis of reflux
esophagitis (Los Angeles classification grade A) and antral
gastritis. These findings were consistent with previous
investigations. It is important to note that while these findings
are unrelated to HCC, they contribute to Mr. Wells\' overall medical
profile and require ongoing treatment.
**Colonoscopy on 11/20/2020 15:15 PM.**
- Colonoscopy showed that the sigmoid colon polyp, which had been
removed during the previous examination, had not recurred. No new
abnormalities or malignancies were detected in the gastrointestinal
tract. This examination provides assurance that there is no
concurrent colorectal malignancy complicating Mr. Wells\' medical
condition.
**Pulmonary Function Testing:**
Mr. Wells\' COPD, emphysema, and respiratory insufficiency were
evaluated in detail. Pulmonary function tests confirmed his current
severity score of Gold III, indicating advanced COPD. Despite the
chronic nature of his disease, there has been no significant
deterioration since the last assessment.
**Oxygen Therapy:**
As previously documented, Mr. Wells requires home oxygen therapy. His
oxygen requirements have been constant, with no significant increase in
oxygen requirements during daily activities or at rest. This stability
in his oxygen demand is encouraging and indicates effective management
of his respiratory disease.
**Overall Assessment:** Based on the results of recent follow-up, Mr.
Paul Wells\' hepatocellular carcinoma (HCC) has not progressed
significantly. The previously noted HCC lesions have remained stable in
terms of size and characteristics. In addition, there is no evidence of
malignancy elsewhere in his thoracoabdominal region.
Mr. Wells\' COPD, emphysema, and respiratory insufficiency, which is
being treated with home oxygen therapy, have also not changed
significantly during this follow-up period. His cardiopulmonary
condition remains well controlled, with no acute deterioration.
Psychosocially, Mr. Wells continues to demonstrate resilience and
actively participates in his care. His strong support system continues
to contribute to his overall well-being.
Additional monitoring and follow-up appointments have been scheduled to
ensure continued management of Mr. Wells\' health. In addition,
discussions continue regarding potential treatment options and
interventions to provide him with the best possible care.
**Current Recommendations:** In light of the stability observed in Mr.
Wells\' HCC and overall medical condition, we recommend the following
steps for his continued care:
1. Regular Follow-up: Maintain a schedule of regular follow-up
appointments to monitor the status of the HCC, cardiopulmonary
function, and other associated conditions.
2. Lifestyle-Modification
### Patient Report 5
**Dear colleague, **
We report to you about Mr. Paul Wells born on 04/02/1953 who received
inpatient treatment from 02/04/2021 to 02/12/2021.
**Diagnosis**: Community-Acquired Pneumonia (CAP)
**Previous Diagnoses and Treatment:**
- Multifocal HCC Segment with portal vein invasion, BCLC C, first
diagnosed 07/19
- Large, partly exophytic, partly centrally hemorrhagic HCC lesions in
S3/4 and S7/8, complete infiltration of the left lateral lobe with
smaller satellites, macrovascular invasion of the left portal vein.
- Histology on 07/27/2019: Macrotrabecular and pseudoglandular growth
of well-differentiated hepatocellular carcinoma (G1).
- SIRT simulation attempt on 08/13/2019: No feasible SIRT.
- Liver Tumor Board decision on 08/18/2019: Systemic therapy.
- Atezolizumab/Bevacizumab since 10/26/2021, with a pause starting on
09/17/2019, due to transaminase elevation (up to 4x ULN).
- CT in 01/2020: Very good tumor response.
- Re-administration of Atezolizumab/Bevacizumab on 01/25/2022 and
02/16/2022, followed by a treatment pause due to limited tolerance.
- CT from 02/2020 to 08/2020: Continuously regressing tumor findings.
- Liver fibrosis: Increased C2 consumption (3-4 beers/day).
- Suspected PNP DD RLS (Restless Legs Syndrome).
<!-- -->
- COPD, current severity Gold III.
- Pulmonary emphysema.
- Respiratory partial insufficiency with home oxygen.
- Postnasal-Drip Syndrome.
- History of nicotine abuse (120 py).
- Transient worsening of lung function with steroid requirement after
Atezolizumab/Bevacizumab administrations
- Pneumogenic sepsis with Streptococcus pneumoniae detection.
- History of unclear infection vs. pneumonia in 10/2019-01/2020.
- Arterial hypertension.
- Atrial fibrillation
- Treatment with Apixaban.
- Reflux esophagitis LA Grade A (Esophagogastroduodenoscopy in
08/2019).
**Medical History:** For detailed medical history, please refer to the
previous medical reports. In summary, Mr. Wells presented in 07/2019
with persistent right upper abdominal pain. A CT scan showed multiple
intrahepatic lesions in the right liver lobe (SIV, SVII/VIII). MR
imaging also revealed large, partly exophytic, partly centrally
hemorrhagic HCC lesions in S3/4 and S7/8. There was complete
infiltration of the left liver lobe with smaller satellites and
macroinvasion of the left portal vein branch. Histology confirmed a
well-differentiated hepatocellular carcinoma (G1). There is no known
underlying liver disease, but peritumoral liver fibrosis was observed
histologically. Mr. Wells reported increased alcohol consumption of 3-4
beers per day.
Due to comorbidities and a large tumor with a relatively high liver-lung
shunt, SIRT simulation was initially attempted but found to be an
unsuitable treatment option. Therefore, our interdisciplinary liver
tumor board recommended systemic therapy. After comprehensive
counseling, treatment with Atezolizumab/Bevacizumab commenced on
08/24/2019.
Currently, Mr. Wells complains about progressively worsening respiratory
symptoms, which included shortness of breath, productive cough with
yellow-green sputum, pleuritic chest pain, fever, and chills, spanning a
period of five days.
**Physical Examination:**
Temperature: 38.6°C, Blood Pressure: 140/80 mm Hg, Heart Rate: 110 beats
per minute Respiratory Rate: 30 breaths per minute, Oxygen Saturation
(SpO2): 88% on room air
Breath Sounds: Auscultation revealed diminished breath sounds and coarse
crackles, notably in the right lower lobe.
The patient further reported pleuritic chest pain localized to the right
lower chest.
**Therapy and Progression:**
During his hospitalization, Mr. Wells was in stable cardiopulmonary
condition. We initiated an empiric antibiotic therapy with intravenous
Ceftriaxone and Azithromycin to treat community-acquired pneumonia
(CAP). Oxygen supplementation was provided to maintain adequate oxygen
saturation levels, and pain management strategies were implemented to
alleviate pleuritic chest pain. Additionally, pulmonary hygiene measures
and chest physiotherapy were applied to facilitate sputum clearance.
Frequent respiratory treatments with bronchodilators were administered
to mitigate airway obstruction, and continuous monitoring of vital
signs, oxygen saturation, and respiratory status was carried out.
Throughout his hospital stay, Mr. Wells exhibited gradual clinical
improvement, marked by several positive developments. These included the
resolution of fever, improved oxygen saturation levels, and a follow-up
chest X-ray demonstrating the resolution of the right lower lobe
consolidation. Furthermore, antibiotic therapy was adjusted based on
sputum culture results, which identified Streptococcus pneumoniae as the
causative pathogen. Mr. Wells continued to receive supportive care and
respiratory interventions.
We were thus able to discharge Mr. Wells in a good general condition.
|
08/24/2019
|
Of the following, which is the best plausible explanation for the behavior of the Earthen ships?
A. Since the Quest III trip promised to locate more planets, the current Earthens didn't trust them when they learned of their success because of how unlikely it was.
B. Since the Quest III trip promised to locate more planets, the current Earthens didn't trust them when they learned of their failure.
C. Since the Quest III trip promised to locate more natural resources, the current Earthens didn't trust them when they learned of their failure.
D. 900 years passed on Earth. The populations were different enough that the Quest III Earthens scared the current Earthen population.
|
THE GIANTS RETURN By ROBERT ABERNATHY Earth set itself grimly to meet them with corrosive fire, determined to blast them back to the stars. But they erred in thinking the Old Ones were too big to be clever. [Transcriber's Note: This etext was produced from Planet Stories Fall 1949. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] In the last hours the star ahead had grown brighter by many magnitudes, and had changed its color from a dazzling blue through white to the normal yellow, of a GO sun. That was the Doppler effect as the star's radial velocity changed relative to the Quest III , as for forty hours the ship had decelerated. They had seen many such stars come near out of the galaxy's glittering backdrop, and had seen them dwindle, turn red and go out as the Quest III drove on its way once more, lashed by despair toward the speed of light, leaving behind the mockery of yet another solitary and lifeless luminary unaccompanied by worlds where men might dwell. They had grown sated with the sight of wonders—of multiple systems of giant stars, of nebulae that sprawled in empty flame across light years. But now unwonted excitement possessed the hundred-odd members of the Quest III's crew. It was a subdued excitement; men and women, they came and stood quietly gazing into the big vision screens that showed the oncoming star, and there were wide-eyed children who had been born in the ship and had never seen a planet. The grownups talked in low voices, in tones of mingled eagerness and apprehension, of what might lie at the long journey's end. For the Quest III was coming home; the sun ahead was the Sun, whose rays had warmed their lives' beginning. Knof Llud, the Quest III's captain, came slowly down the narrow stair from the observatory, into the big rotunda that was now the main recreation room, where most of the people gathered. The great chamber, a full cross-section of the vessel, had been at first a fuel hold. At the voyage's beginning eighty per cent of the fifteen-hundred-foot cylinder had been engines and fuel; but as the immense stores were spent and the holds became radioactively safe, the crew had spread out from its original cramped quarters. Now the interstellar ship was little more than a hollow shell. Eyes lifted from the vision screens to interrogate Knof Llud; he met them with an impassive countenance, and announced quietly, "We've sighted Earth." A feverish buzz arose; the captain gestured for silence and went on, "It is still only a featureless disk to the telescope. Zost Relyul has identified it—no more." But this time the clamor was not to be settled. People pressed round the screens, peering into them as if with the naked eye they could pick out the atom of reflected light that was Earth, home. They wrung each other's hands, kissed, shouted, wept. For the present their fears were forgotten and exaltation prevailed. Knof Llud smiled wryly. The rest of the little speech he had been about to make didn't matter anyway, and it might have spoiled this moment. He turned to go, and was halted by the sight of his wife, standing at his elbow. His wry smile took on warmth; he asked, "How do you feel, Lesra?" She drew an uncertain breath and released it in a faint sigh. "I don't know. It's good that Earth's still there." She was thinking, he judged shrewdly, of Knof Jr. and Delza, who save from pictures could not remember sunlit skies or grassy fields or woods in summer.... He said, with a touch of tolerant amusement, "What did you think might have happened to Earth? After all, it's only been nine hundred years." "That's just it," said Lesra shakily. "Nine hundred years have gone by— there —and nothing will be the same. It won't be the same world we left, the world we knew and fitted in...." The captain put an arm round her with comforting pressure. "Don't worry. Things may have changed—but we'll manage." But his face had hardened against registering the gnawing of that same doubtful fear within him. He let his arm fall. "I'd better get up to the bridge. There's a new course to be set now—for Earth." He left her and began to climb the stairway again. Someone switched off the lights, and a charmed whisper ran through the big room as the people saw each other's faces by the pale golden light of Earth's own Sun, mirrored and multiplied by the screens. In that light Lesra's eyes gleamed with unshed tears. Captain Llud found Navigator Gwar Den looking as smug as the cat that ate the canary. Gwar Den was finding that the actual observed positions of the planets thus far located agreed quite closely with his extrapolations from long unused charts of the Solar System. He had already set up on the calculator a course that would carry them to Earth. Llud nodded curt approval, remarking, "Probably we'll be intercepted before we get that far." Den was jolted out of his happy abstraction. "Uh, Captain," he said hesitantly. "What kind of a reception do you suppose we'll get?" Llud shook his head slowly. "Who knows? We don't know whether any of the other Quests returned successful, or if they returned at all. And we don't know what changes have taken place on Earth. It's possible—not likely, though—that something has happened to break civilization's continuity to the point where our expedition has been forgotten altogether." He turned away grim-lipped and left the bridge. From his private office-cabin, he sent a message to Chief Astronomer Zost Relyul to notify him as soon as Earth's surface features became clear; then he sat idle, alone with his thoughts. The ship's automatic mechanisms had scant need of tending; Knof Llud found himself wishing that he could find some back-breaking task for everyone on board, himself included, to fill up the hours that remained. There was an extensive and well-chosen film library in the cabin, but he couldn't persuade himself to kill time that way. He could go down and watch the screens, or to the family apartment where he might find Lesra and the children—but somehow he didn't want to do that either. He felt empty, drained—like his ship. As the Quest III's fuel stores and the hope of success in man's mightiest venture had dwindled, so the strength had gone out of him. Now the last fuel compartment was almost empty and Captain Knof Llud felt tired and old. Perhaps, he thought, he was feeling the weight of his nine hundred Earth years—though physically he was only forty now, ten years older than when the voyage had begun. That was the foreshortening along the time axis of a space ship approaching the speed of light. Weeks and months had passed for the Quest III in interstellar flight while years and decades had raced by on the home world. Bemusedly Llud got to his feet and stood surveying a cabinet with built-in voice recorder and pigeonholes for records. There were about three dozen film spools there—his personal memoirs of the great expedition, a segment of his life and of history. He might add that to the ship's official log and its collections of scientific data, as a report to whatever powers might be on Earth now—if such powers were still interested. Llud selected a spool from among the earliest. It was one he had made shortly after leaving Procyon, end of the first leg of the trip. He slid it onto the reproducer. His own voice came from the speaker, fresher, more vibrant and confident than he knew it was now. "One light-day out from Procyon, the thirty-third day by ship's time since leaving Earth. "Our visit to Procyon drew a blank. There is only one huge planet, twice the size of Jupiter, and like Jupiter utterly unfit to support a colony. "Our hopes were dashed—and I think all of us, even remembering the Centaurus Expedition's failure, hoped more than we cared to admit. If Procyon had possessed a habitable planet, we could have returned after an absence of not much over twenty years Earth time. "It is cheering to note that the crew seems only more resolute. We go on to Capella; its spectrum, so like our own Sun's, beckons. If success comes there, a century will have passed before we can return to Earth; friends, relatives, all the generation that launched the Quest ships will be long since dead. Nevertheless we go on. Our generation's dream, humanity's dream, lives in us and in the ship forever...." Presently Knof Llud switched off that younger voice of his and leaned back, an ironic smile touching his lips. That fervent idealism seemed remote and foreign to him now. The fanfares of departure must still have been ringing in his ears. He rose, slipped the record back in its niche and picked out another, later, one. "One week since we passed close enough to Aldebaran to ascertain that that system, too, is devoid of planets. "We face the unpleasant realization that what was feared is probably true—that worlds such as the Sun's are a rare accident, and that we may complete our search without finding even one new Earth. "It makes no difference, of course; we cannot betray the plan.... This may be man's last chance of escaping his pitiful limitation to one world in all the Universe. Certainly the building of this ship and its two sisters, the immense expenditure of time and labor and energy stores that went into them, left Earth's economy drained and exhausted. Only once in a long age does mankind rise to such a selfless and transcendent effort—the effort of Egypt that built the pyramids, or the war efforts of the nations in the last great conflicts of the twentieth century. "Looked at historically, such super-human outbursts of energy are the result of a population's outgrowing its room and resources, and therefore signalize the beginning of the end. Population can be limited, but the price is a deadly frustration, because growth alone is life.... In our day the end of man's room for growth on the Earth was in sight—so we launched the Quests . Perhaps our effort will prove as futile as pyramid-building, less practical than orgies of slaughter to reduce pressure.... In any case, it would be impossible to transport very many people to other stars; but Earth could at least go into its decline with the knowledge that its race went onward and upward, expanding limitlessly into the Universe.... "Hopeless, unless we find planets!" Knof Llud shook his head sorrowfully and took off the spool. That was from the time when he had grown philosophical after the first disappointments. He frowned thoughtfully, choosing one more spool that was only four years old. The recorded voice sounded weary, yet alive with a strange longing.... "We are in the heart of Pleiades; a hundred stars show brilliant on the screens, each star encircled by a misty halo like lights glowing through fog, for we are traversing a vast diffuse nebula. "According to plan, the Quest III has reached its furthest point from Earth. Now we turn back along a curve that will take us past many more stars and stellar systems—but hope is small that any of those will prove a home for man, as have none of the thousands of stars examined already. "But what are a few thousand stars in a galaxy of billions? We have only, as it were, visited a handful of the outlying villages of the Universe, while the lights of its great cities still blaze far ahead along the Milky Way. "On flimsy excuses I have had Zost Relyul make observations of the globular cluster Omega Centauri. There are a hundred thousand stars there in a volume of space where one finds a few dozen in the Sun's neighborhood; there if anywhere must circle the planets we seek! But Omega Centauri is twenty thousand light years away.... "Even so—by expending its remaining fuel freely, the Quest III could achieve a velocity that would take us there without dying of senility of aging too greatly. It would be a one-way journey—even if enough fuel remained, there would be little point in returning to Earth after more than forty thousand years. By then our civilization certainly, and perhaps the human race itself, would have perished from memory. "That was why the planners limited our voyage, and those of the other Quests , to less than a thousand years Earth time. Even now, according to the sociodynamic predictions made then, our civilization—if the other expeditions failed also—will have reached a dangerously unstable phase, and before we can get back it may have collapsed completely from overpopulation. "Why go back, then with the news of our failure? Why not forget about Earth and go on to Omega Centauri? What use is quixotic loyalty to a decree five thousand years old, whose makers are dead and which may be forgotten back there? "Would the crew be willing? I don't know—some of them still show signs of homesickness, though they know with their minds that everything that was once 'home' has probably been swept away.... "It doesn't matter. Today I gave orders to swing the ship." Savagely Knof Llud stabbed the button that shut off the speaker. Then he sat for a time with head resting in his hands, staring into nothing. The memory of that fierce impulse to go on still had power to shake him. A couple of lines of poetry came into his head, as he read them once in translation from the ancient English.... ... for my purpose holds To sail beyond the sunset, and the baths Of all the western stars, until I die. Llud sighed. He still couldn't say just why he had given the order to turn back. The stars had claimed his heart—but he was still a part of Earth, and not even nine hundred years of space and time had been able to alter that. He wondered if there would still be a quiet stream and a green shady place beside it where a death-weary man, relieved at last of responsibility, could rest and dream no more.... Those things went on, if men didn't change them. And a pine forest where he and young Knof could go camping, and lie on their backs at night and gaze at the glittering constellations, far away, out of reach.... He wasn't sure he would want to do that, though. Suddenly a faint cushioned jar went through the great ship; it seemed to falter one moment in flight. The captain was on his feet instantly, but then his movements became unhurried. Whatever it had been was past, and he had a good idea what it had been—a meteoroid, nothing unusual in the vicinity of the Sun, though in interstellar space and around planetless stars such collisions were rare to the vanishing point. No harm could have been done. The Quest III's collision armor was nonmaterial and for practical purposes invulnerable. Just as he took his finger off the button that opened the door, the intercommunication phone shrilled imperatively. Knof Llud wheeled, frowning—surely a meteoroid impact wasn't that serious. Coincidence, maybe—it might be Zost Relyul calling as instructed. He reached the phone at the moment when another, heavier jolt shook the vessel. Llud snatched up the receiver with the speed of a scalded cat. "Captain?" It was Gwar Den's voice, stammering a little. "Captain, we're being attacked!" "Sound the alarm. Emergency stations." He had said it automatically, then felt a curious detached relief at the knowledge that after all these years he could still respond quickly and smoothly to a crisis. There was a moment's silence, and he heard the alarm start—three short buzzes and repeat, ringing through all the great length of the interstellar ship. Knowing that Gwar Den was still there, he said, "Now—attacked by what?" "Ships," said Gwar Den helplessly. "Five of them so far. No, there's a sixth now." Repeated blows quivered the Quest III's framework. The navigator said, obviously striving for calm, "They're light craft, not fifty feet long, but they move fast. The detectors hardly had time to show them before they opened up. Can't get a telescope beam on them long enough to tell much." "If they're that small," said Knof Llud deliberately, "they can't carry anything heavy enough to hurt us. Hold to course. I'll be right up." In the open doorway he almost fell over his son. Young Knof's eyes were big; he had heard his father's words. "Something's happened," he judged with deadly twelve-year-old seriousness and, without wasting time on questions, "Can I go with you, huh, Dad?" Llud hesitated, said, "All right. Come along and keep out of the way." He headed for the bridge with strides that the boy could not match. There were people running in the corridors, heading for their posts. Their faces were set, scared, uncomprehending. The Quest III shuddered, again and again, under blows that must have had millions of horsepower behind them; but it plunged on toward Earth, its mighty engines still steadily braking its interstellar velocity. To a man, the ship's responsible officers were already on the bridge, most of them breathless. To a man they looked appeal at Captain Knof Llud. "Well?" he snapped. "What are they doing?" Gwar Den spoke. "There are thirteen of them out there now, sir, and they're all banging away at us." The captain stared into the black star-strewn depths of a vision screen where occasional blue points of light winked ominously, never twice from the same position. Knof Jr. flattened himself against the metal wall and watched silently. His young face was less anxious than his elders'; he had confidence in his father. "If they had anything heavier," surmised the captain, "they'd have unlimbered it by now. They're out to get us. But at this rate, they can't touch us as long as our power lasts—or until they bring up some bigger stuff." The mild shocks went on—whether from projectiles or energy-charges, would be hard to find out and it didn't matter; whatever was hitting the Quest III's shell was doing it at velocities where the distinction between matter and radiation practically ceases to exist. But that shell was tough. It was an extension of the gravitic drive field which transmitted the engines' power equally to every atom of the ship; forces impinging on the outside of the field were similarly transmitted and rendered harmless. The effect was as if the vessel and all space inside its field were a single perfectly elastic body. A meteoroid, for example, on striking it rebounded—usually vaporized by the impact—and the ship, in obedience to the law of equal and opposite forces, rebounded too, but since its mass was so much greater, its deflection was negligible. The people in the Quest III would have felt nothing at all of the vicious onslaught being hurled against them, save that their inertialess drive, at its normal thrust of two hundred gravities, was intentionally operated at one half of one per cent efficiency to provide the illusion of Earthly gravitation. One of the officers said shakily, "It's as if they've been lying in wait for us. But why on Earth—" "That," said the captain grimly, "is what we have to find out. Why—on Earth. At least, I suspect the answer's there." The Quest III bored steadily on through space, decelerating. Even if one were no fatalist, there seemed no reason to stop decelerating or change course. There was nowhere else to go and too little fuel left if there had been; come what might, this was journey's end—perhaps in a more violent and final way than had been anticipated. All around wheeled the pigmy enemies, circling, maneuvering, and attacking, always attacking, with the senseless fury of maddened hornets. The interstellar ship bore no offensive weapons—but suddenly on one of the vision screens a speck of light flared into nova-brilliance, dazzling the watchers for the brief moment in which its very atoms were torn apart. Knof Jr. whooped ecstatically and then subsided warily, but no one was paying attention to him. The men on the Quest III's bridge looked questions at each other, as the thought of help from outside flashed into many minds at once. But Captain Llud said soberly, "It must have caught one of their own shots, reflected. Maybe its own, if it scored too direct a hit." He studied the data so far gathered. A few blurred pictures had been got, which showed cylindrical space ships much like the Quest III , except that they were rocket-propelled and of far lesser size. Their size was hard to ascertain, because you needed to know their distance and speed—but detector-beam echoes gave the distance, and likewise, by the Doppler method, the velocity of directly receding or approaching ships. It was apparent that the enemy vessels were even smaller than Gwar Den had at first supposed—not large enough to hold even one man. Tiny, deadly hornets with a colossal sting. "Robot craft, no doubt," said Knof Llud, but a chill ran down his spine as it occurred to him that perhaps the attackers weren't of human origin. They had seen no recognizable life in the part of the galaxy they had explored, but one of the other Quests might have encountered and been traced home by some unhuman race that was greedy and able to conquer. It became evident, too, that the bombardment was being kept up by a constant arrival of fresh attackers, while others raced away into space, presumably returning to base to replenish their ammunition. That argued a planned and prepared interception with virulent hatred behind it. Elsuz Llug, the gravitic engineer, calculated dismally, "At the rate we're having to shed energy, the fuel will be gone in six or eight hours." "We'll have reached Earth before then," Gwar Den said hopefully. "If they don't bring out the heavy artillery first." "We're under the psychological disadvantage," said the captain, "of not knowing why we're being attacked." Knof Jr. burst out, spluttering slightly with the violence of a thought too important to suppress, "But we're under a ps-psychological advantage, too!" His father raised an eyebrow. "What's that? I don't seem to have noticed it." "They're mad and we aren't, yet," said the boy. Then, seeing that he hadn't made himself clear, "In a fight, if a guy gets mad he starts swinging wild and then you nail him." Smiles splintered the ice of tension. Captain Llud said, "Maybe you've got something there. They seem to be mad, all right. But we're not in a position to throw any punches." He turned back to the others. "As I was going to say—I think we'd better try to parley with the enemy. At least we may find out who he is and why he's determined to smash us." And now instead of tight-beam detectors the ship was broadcasting on an audio carrier wave that shifted through a wide range of frequencies, repeating on each the same brief recorded message: "Who are you? What do you want? We are the interstellar expedition Quest III ...." And so on, identifying themselves and protesting that they were unarmed and peaceful, that there must be some mistake, and querying again, "Who are you ?" There was no answer. The ship drove on, its fuel trickling away under multiplied demands. Those outside were squandering vastly greater amounts of energy in the effort to batter down its defenses, but converting that energy into harmless gravitic impulses was costing the Quest III too. Once more Knof Llud had the insidious sense of his own nerves and muscles and will weakening along with the power-sinews of his ship. Zost Relyul approached him apologetically. "If you have time, Captain—I've got some data on Earth now." Eagerly Llud took the sheaf of photographs made with the telescope. But they told him nothing; only the continental outlines were clear, and those were as they had been nine hundred years ago.... He looked up inquiringly at Zost Relyul. "There are some strange features," said the astronomer carefully. "First of all—there are no lights on the night side. And on the daylight face, our highest magnification should already reveal traces of cities, canals, and the like—but it does not. "The prevailing color of the land masses, you see, is the normal green vegetation. But the diffraction spectrum is queer. It indicates reflecting surfaces less than one-tenth millimeter wide—so the vegetation there can't be trees or grass, but must be more like a fine moss or even a coarse mold." "Is that all?" demanded Llud. "Isn't it enough?" said Zost Relyul blankly. "Well—we tried photography by invisible light, of course. The infra-red shows nothing and likewise the ultraviolet up to the point where the atmosphere is opaque to it." The captain sighed wearily. "Good work," he said. "Keep it up; perhaps you can answer some of these riddles before—" " We know who you are ," interrupted a harshly crackling voice with a strange accent, " and pleading will do you no good. " Knof Llud whirled to the radio apparatus, his weariness dropping from him once more. He snapped, "But who are you?" and the words blended absurdly with the same words in his own voice on the still repeating tape. He snapped off the record; as he did so the speaker, still crackling with space static, said, "It may interest you to know that you are the last. The two other interstellar expeditions that went out have already returned and been destroyed, as you will soon be—the sooner, if you continue toward Earth." Knof Llud's mind was clicking again. The voice—which must be coming from Earth, relayed by one of the midget ships—was not very smart; it had already involuntarily told him a couple of things—that it was not as sure of itself as it sounded he deduced from the fact it had deigned to speak at all, and from its last remark he gathered that the Quest III's ponderous and unswerving progress toward Earth had somehow frightened it. So it was trying to frighten them. He shoved those facts back for future use. Just now he had to know something, so vitally that he asked it as a bald question, " Are you human? " The voice chuckled sourly. "We are human," it answered, "but you are not." The captain was momentarily silent, groping for an adequate reply. Behind him somebody made a choked noise, the only sound in the stunned hush, and the ship jarred slightly as a thunderbolt slammed vengefully into its field. "Suppose we settle this argument about humanity," said Knof Llud woodenly. He named a vision frequency. "Very well." The tone was like a shrug. The voice went on in its language that was quite intelligible, but alien-sounding with the changes that nine hundred years had wrought. "Perhaps, if you realize your position, you will follow the intelligent example of the Quest I's commander." Knof Llud stiffened. The Quest I , launched toward Arcturus and the star cloud called Berenice's Hair, had been after the Quest III the most hopeful of the expeditions—and its captain had been a good friend of Llud's, nine hundred years ago.... He growled, "What happened to him?" "He fought off our interceptors, which are around you now, for some time," said the voice lightly. "When he saw that it was hopeless, he preferred suicide to defeat, and took his ship into the Sun." A short pause. "The vision connection is ready." Knof Llud switched on the screen at the named wavelength, and a picture formed there. The face and figure that appeared were ugly, but undeniably a man's. His features and his light-brown skin showed the same racial characteristics possessed by those aboard the Quest III , but he had an elusive look of deformity. Most obviously, his head seemed too big for his body, and his eyes in turn too big for his head. He grinned nastily at Knof Llud. "Have you any other last wishes?" "Yes," said Llud with icy control. "You haven't answered one question. Why do you want to kill us? You can see we're as human as you are." The big-headed man eyed him with a speculative look in his great eyes, behind which the captain glimpsed the flickering raw fire of a poisonous hatred. "It is enough for you to know that you must die."
|
D. 900 years passed on Earth. The populations were different enough that the Quest III Earthens scared the current Earthen population.
|
What were Purnie's friends like in this story?
A. Self-interested and ignorant
B. Intelligent and caring
C. Malicious and blunt
D. Sweet and charming
|
BEACH SCENE By MARSHALL KING Illustrated by WOOD [Transcriber's Note: This etext was produced from Galaxy Magazine October 1960. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] It was a fine day at the beach for Purnie's game—but his new friends played very rough! Purnie ran laughing and shouting through the forest until he could run no more. He fell headlong into a patch of blue moss and whooped with delight in having this day free for exploring. He was free to see the ocean at last. When he had caught his breath, he looked back through the forest. No sign of the village; he had left it far behind. Safe from the scrutiny of brothers and parents, there was nothing now to stop him from going to the ocean. This was the moment to stop time. "On your mark!" he shouted to the rippling stream and its orange whirlpools. He glanced furtively from side to side, pretending that some object might try to get a head start. "Get set!" he challenged the thin-winged bees that hovered over the abundant foliage. "Stop!" He shrieked this command upward toward the dense, low-hanging purple clouds that perennially raced across the treetops, making one wonder how tall the trees really were. His eyes took quick inventory. It was exactly as he knew it would be: the milky-orange stream had become motionless and its minute whirlpools had stopped whirling; a nearby bee hung suspended over a paka plant, its transparent wings frozen in position for a downward stroke; and the heavy purple fluid overhead held fast in its manufacture of whorls and nimbi. With everything around him in a state of perfect tableau, Purnie hurried toward the ocean. If only the days weren't so short! he thought. There was so much to see and so little time. It seemed that everyone except him had seen the wonders of the beach country. The stories he had heard from his brothers and their friends had taunted him for as long as he could remember. So many times had he heard these thrilling tales that now, as he ran along, he could clearly picture the wonderland as though he were already there. There would be a rockslide of petrified logs to play on, the ocean itself with waves higher than a house, the comical three-legged tripons who never stopped munching on seaweed, and many kinds of other wonderful creatures found only at the ocean. He bounced through the forest as though the world was reserved this day just for him. And who could say it wasn't? he thought. Wasn't this his fifth birthday? He ran along feeling sorry for four-year-olds, and even for those who were only four and a half, for they were babies and wouldn't dare try slipping away to the ocean alone. But five! "I'll set you free, Mr. Bee—just wait and see!" As he passed one of the many motionless pollen-gathering insects he met on the way, he took care not to brush against it or disturb its interrupted task. When Purnie had stopped time, the bees—like all the other creatures he met—had been arrested in their native activities, and he knew that as soon as he resumed time, everything would pick up where it had left off. When he smelled an acid sweetness that told him the ocean was not far off, his pulse quickened in anticipation. Rather than spoil what was clearly going to be a perfect day, he chose to ignore the fact that he had been forbidden to use time-stopping as a convenience for journeying far from home. He chose to ignore the oft-repeated statement that an hour of time-stopping consumed more energy than a week of foot-racing. He chose to ignore the negative maxim that "small children who stop time without an adult being present, may not live to regret it." He chose, instead, to picture the beaming praise of family and friends when they learned of his brave journey. The journey was long, the clock stood still. He stopped long enough to gather some fruit that grew along the path. It would serve as his lunch during this day of promise. With it under his arm he bounded along a dozen more steps, then stopped abruptly in his tracks. He found himself atop a rocky knoll, overlooking the mighty sea! He was so overpowered by the vista before him that his "Hurrah!" came out as a weak squeak. The ocean lay at the ready, its stilled waves awaiting his command to resume their tidal sweep. The breakers along the shoreline hung in varying stages of disarray, some having already exploded into towering white spray while others were poised in smooth orange curls waiting to start that action. And there were new friends everywhere! Overhead, a flock of spora were frozen in a steep glide, preparatory to a beach landing. Purnie had heard of these playful creatures many times. Today, with his brothers in school, he would have the pets all to himself. Further down the beach was a pair of two-legged animals poised in mid-step, facing the spot where Purnie now stood. Some distance behind them were eight more, each of whom were motionless in a curious pose of interrupted animation. And down in the water, where the ocean ran itself into thin nothingness upon the sand, he saw standing here and there the comical tripons, those three-legged marine buffoons who made handsome careers of munching seaweed. "Hi there!" Purnie called. When he got no reaction, he remembered that he himself was "dead" to the living world: he was still in a zone of time-stopping, on the inside looking out. For him, the world would continue to be a tableau of mannikins until he resumed time. "Hi there!" he called again; but now his mental attitude was that he expected time to resume. It did! Immediately he was surrounded by activity. He heard the roar of the crashing orange breakers, he tasted the dew of acid that floated from the spray, and he saw his new friends continue the actions which he had stopped while back in the forest. He knew, too, that at this moment, in the forest, the little brook picked up its flow where it had left off, the purple clouds resumed their leeward journey up the valley, and the bees continued their pollen-gathering without having missed a single stroke of their delicate wings. The brook, the clouds, and the insects had not been interrupted in the least; their respective tasks had been performed with continuing sureness. It was time itself that Purnie had stopped, not the world around him. He scampered around the rockpile and down the sandy cliff to meet the tripons who, to him, had just come to life. "I can stand on my head!" He set down his lunch and balanced himself bottoms-up while his legs pawed the air in an effort to hold him in position. He knew it was probably the worst head-stand he had ever done, for he felt weak and dizzy. Already time-stopping had left its mark on his strength. But his spirits ran on unchecked. The tripon thought Purnie's feat was superb. It stopped munching long enough to give him a salutory wag of its rump before returning to its repast. Purnie ran from pillar to post, trying to see and do everything at once. He looked around to greet the flock of spora, but they had glided to a spot further along the shore. Then, bouncing up to the first of the two-legged animals, he started to burst forth with his habitual "Hi there!" when he heard them making sounds of their own. "... will be no limit to my operations now, Benson. This planet makes seventeen. Seventeen planets I can claim as my own!" "My, my. Seventeen planets. And tell me, Forbes, just what the hell are you going to do with them—mount them on the wall of your den back in San Diego?" "Hi there, wanna play?" Purnie's invitation got nothing more than startled glance from the animals who quickly returned to their chatter. He scampered up the beach, picked up his lunch, and ran back to them, tagging along at their heels. "I've got my lunch, want some?" "Benson, you'd better tell your men back there to stop gawking at the scenery and get to work. Time is money. I didn't pay for this expedition just to give your flunkies a vacation." The animals stopped so suddenly that Purnie nearly tangled himself in their heels. "All right, Forbes, just hold it a minute. Listen to me. Sure, it's your money that put us here; it's your expedition all the way. But you hired me to get you here with the best crew on earth, and that's just what I've done. My job isn't over yet. I'm responsible for the safety of the men while we're here, and for the safe trip home." "Precisely. And since you're responsible, get 'em working. Tell 'em to bring along the flag. Look at the damn fools back there, playing in the ocean with a three-legged ostrich!" "Good God, man, aren't you human? We've only been on this planet twenty minutes! Naturally they want to look around. They half expected to find wild animals or worse, and here we are surrounded by quaint little creatures that run up to us like we're long-lost brothers. Let the men look around a minute or two before we stake out your claim." "Bah! Bunch of damn children." As Purnie followed along, a leg shot out at him and missed. "Benson, will you get this bug-eyed kangaroo away from me!" Purnie shrieked with joy at this new frolic and promptly stood on his head. In this position he got an upside down view of them walking away. He gave up trying to stay with them. Why did they move so fast, anyway? What was the hurry? As he sat down and began eating his lunch, three more of the creatures came along making excited noises, apparently trying to catch up to the first two. As they passed him, he held out his lunch. "Want some?" No response. Playing held more promise than eating. He left his lunch half eaten and went down to where they had stopped further along the beach. "Captain Benson, sir! Miles has detected strong radiation in the vicinity. He's trying to locate it now." "There you are, Forbes. Your new piece of real estate is going to make you so rich that you can buy your next planet. That'll make eighteen, I believe." "Radiation, bah! We've found low-grade ore on every planet I've discovered so far, and this one'll be no different. Now how about that flag? Let's get it up, Benson. And the cornerstone, and the plaque." "All right, lads. The sooner we get Mr. Forbes's pennant raised and his claim staked out, the sooner we can take time to look around. Lively now!" When the three animals went back to join the rest of their group, the first two resumed walking. Purnie followed along. "Well, Benson, you won't have to look far for materials to use for the base of the flag pole. Look at that rockpile up there. "Can't use them. They're petrified logs. The ones on top are too high to carry down, and if we move those on the bottom, the whole works will slide down on top of us." "Well—that's your problem. Just remember, I want this flag pole to be solid. It's got to stand at least—" "Don't worry, Forbes, we'll get your monument erected. What's this with the flag? There must be more to staking a claim than just putting up a flag." "There is, there is. Much more. I've taken care of all requirements set down by law to make my claim. But the flag? Well, you might say it represents an empire, Benson. The Forbes Empire. On each of my flags is the word FORBES, a symbol of development and progress. Call it sentiment if you will." "Don't worry, I won't. I've seen real-estate flags before." "Damn it all, will you stop referring to this as a real-estate deal? What I'm doing is big, man. Big! This is pioneering." "Of course. And if I'm not mistaken, you've set up a neat little escrow system so that you not only own the planets, but you will virtually own the people who are foolish enough to buy land on them." "I could have your hide for talking to me like this. Damn you, man! It's people like me who pay your way. It's people like me who give your space ships some place to go. It's people like me who pour good money into a chancey job like this, so that people like you can get away from thirteen-story tenement houses. Did you ever think of that?" "I imagine you'll triple your money in six months." When they stopped, Purnie stopped. At first he had been interested in the strange sounds they were making, but as he grew used to them, and as they in turn ignored his presence, he hopped alongside chattering to himself, content to be in their company. He heard more of these sounds coming from behind, and he turned to see the remainder of the group running toward them. "Captain Benson! Here's the flag, sir. And here's Miles with the scintillometer. He says the radiation's getting stronger over this way!" "How about that, Miles?" "This thing's going wild, Captain. It's almost off scale." Purnie saw one of the animals hovering around him with a little box. Thankful for the attention, he stood on his head. "Can you do this?" He was overjoyed at the reaction. They all started making wonderful noises, and he felt most satisfied. "Stand back, Captain! Here's the source right here! This little chuck-walla's hotter than a plutonium pile!" "Let me see that, Miles. Well, I'll be damned! Now what do you suppose—" By now they had formed a widening circle around him, and he was hard put to think of an encore. He gambled on trying a brand new trick: he stood on one leg. "Benson, I must have that animal! Put him in a box." "Now wait a minute, Forbes. Universal Law forbids—" "This is my planet and I am the law. Put him in a box!" "With my crew as witness, I officially protest—" "Good God, what a specimen to take back. Radio-active animals! Why, they can reproduce themselves, of course! There must be thousands of these creatures around here someplace. And to think of those damn fools on Earth with their plutonium piles! Hah! Now I'll have investors flocking to me. How about it, Benson—does pioneering pay off or doesn't it?" "Not so fast. Since this little fellow is radioactive, there may be great danger to the crew—" "Now look here! You had planned to put mineral specimens in a lead box, so what's the difference? Put him in a box." "He'll die." "I have you under contract, Benson! You are responsible to me, and what's more, you are on my property. Put him in a box." Purnie was tired. First the time-stopping, then this. While this day had brought more fun and excitement than he could have hoped for, the strain was beginning to tell. He lay in the center of the circle happily exhausted, hoping that his friends would show him some of their own tricks. He didn't have to wait long. The animals forming the circle stepped back and made way for two others who came through carrying a box. Purnie sat up to watch the show. "Hell, Captain, why don't I just pick him up? Looks like he has no intention of running away." "Better not, Cabot. Even though you're shielded, no telling what powers the little fella has. Play it safe and use the rope." "I swear he knows what we're saying. Look at those eyes." "All right, careful now with that line." "Come on, baby. Here you go. That's a boy!" Purnie took in these sounds with perplexed concern. He sensed the imploring quality of the creature with the rope, but he didn't know what he was supposed to do. He cocked his head to one side as he wiggled in anticipation. He saw the noose spinning down toward his head, and, before he knew it, he had scooted out of the circle and up the sandy beach. He was surprised at himself for running away. Why had he done it? He wondered. Never before had he felt this fleeting twinge that made him want to protect himself. He watched the animals huddle around the box on the beach, their attention apparently diverted to something else. He wished now that he had not run away; he felt he had lost his chance to join in their fun. "Wait!" He ran over to his half-eaten lunch, picked it up, and ran back into the little crowd. "I've got my lunch, want some?" The party came to life once more. His friends ran this way and that, and at last Purnie knew that the idea was to get him into the box. He picked up the spirit of the tease, and deliberately ran within a few feet of the lead box, then, just as the nearest pursuer was about to push him in, he sidestepped onto safer ground. Then he heard a deafening roar and felt a warm, wet sting in one of his legs. "Forbes, you fool! Put away that gun!" "There you are, boys. It's all in knowing how. Just winged him, that's all. Now pick him up." The pang in his leg was nothing: Purnie's misery lay in his confusion. What had he done wrong? When he saw the noose spinning toward him again, he involuntarily stopped time. He knew better than to use this power carelessly, but his action now was reflex. In that split second following the sharp sting in his leg, his mind had grasped in all directions to find an acceptable course of action. Finding none, it had ordered the stoppage of time. The scene around him became a tableau once more. The noose hung motionless over his head while the rest of the rope snaked its way in transverse waves back to one of the two-legged animals. Purnie dragged himself through the congregation, whimpering from his inability to understand. As he worked his way past one creature after another, he tried at first to not look them in the eye, for he felt sure he had done something wrong. Then he thought that by sneaking a glance at them as he passed, he might see a sign pointing to their purpose. He limped by one who had in his hand a small shiny object that had been emitting smoke from one end; the smoke now billowed in lifeless curls about the animal's head. He hobbled by another who held a small box that had previously made a hissing sound whenever Purnie was near. These things told him nothing. Before starting his climb up the knoll, he passed a tripon which, true to its reputation, was comical even in fright. Startled by the loud explosion, it had jumped four feet into the air before Purnie had stopped time. Now it hung there, its beak stuffed with seaweed and its three legs drawn up into a squatting position. Leaving the assorted statues behind, he limped his way up the knoll, torn between leaving and staying. What an odd place, this ocean country! He wondered why he had not heard more detail about the beach animals. Reaching the top of the bluff, he looked down upon his silent friends with a feeling of deep sorrow. How he wished he were down there playing with them. But he knew at last that theirs was a game he didn't fit into. Now there was nothing left but to resume time and start the long walk home. Even though the short day was nearly over, he knew he didn't dare use time-stopping to get himself home in nothing flat. His fatigued body and clouded mind were strong signals that he had already abused this faculty. When Purnie started time again, the animal with the noose stood in open-mouthed disbelief as the rope fell harmlessly to the sand—on the spot where Purnie had been standing. "My God, he's—he's gone." Then another of the animals, the one with the smoking thing in his hand, ran a few steps toward the noose, stopped and gaped at the rope. "All right, you people, what's going on here? Get him in that box. What did you do with him?" The resumption of time meant nothing at all to those on the beach, for to them time had never stopped. The only thing they could be sure of was that at one moment there had been a fuzzy creature hopping around in front of them, and the next moment he was gone. "Is he invisible, Captain? Where is he?" "Up there, Captain! On those rocks. Isn't that him?" "Well, I'll be damned!" "Benson, I'm holding you personally responsible for this! Now that you've botched it up, I'll bring him down my own way." "Just a minute, Forbes, let me think. There's something about that fuzzy little devil that we should.... Forbes! I warned you about that gun!" Purnie moved across the top of the rockpile for a last look at his friends. His weight on the end of the first log started the slide. Slowly at first, the giant pencils began cascading down the short distance to the sand. Purnie fell back onto solid ground, horrified at the spectacle before him. The agonizing screams of the animals below filled him with hysteria. The boulders caught most of them as they stood ankle-deep in the surf. Others were pinned down on the sand. "I didn't mean it!" Purnie screamed. "I'm sorry! Can't you hear?" He hopped back and forth near the edge of the rise, torn with panic and shame. "Get up! Please get up!" He was horrified by the moans reaching his ears from the beach. "You're getting all wet! Did you hear me? Please get up." He was choked with rage and sorrow. How could he have done this? He wanted his friends to get up and shake themselves off, tell him it was all right. But it was beyond his power to bring it about. The lapping tide threatened to cover those in the orange surf. Purnie worked his way down the hill, imploring them to save themselves. The sounds they made carried a new tone, a desperate foreboding of death. "Rhodes! Cabot! Can you hear me?" "I—I can't move, Captain. My leg, it's.... My God, we're going to drown!" "Look around you, Cabot. Can you see anyone moving?" "The men on the beach are nearly buried, Captain. And the rest of us here in the water—" "Forbes. Can you see Forbes? Maybe he's—" His sounds were cut off by a wavelet gently rolling over his head. Purnie could wait no longer. The tides were all but covering one of the animals, and soon the others would be in the same plight. Disregarding the consequences, he ordered time to stop. Wading down into the surf, he worked a log off one victim, then he tugged the animal up to the sand. Through blinding tears, Purnie worked slowly and carefully. He knew there was no hurry—at least, not as far as his friends' safety was concerned. No matter what their condition of life or death was at this moment, it would stay the same way until he started time again. He made his way deeper into the orange liquid, where a raised hand signalled the location of a submerged body. The hand was clutching a large white banner that was tangled among the logs. Purnie worked the animal free and pulled it ashore. It was the one who had been carrying the shiny object that spit smoke. Scarcely noticing his own injured leg, he ferried one victim after another until there were no more in the surf. Up on the beach, he started unraveling the logs that pinned down the animals caught there. He removed a log from the lap of one, who then remained in a sitting position, his face contorted into a frozen mask of agony and shock. Another, with the weight removed, rolled over like an iron statue into a new position. Purnie whimpered in black misery as he surveyed the chaotic scene before him. At last he could do no more; he felt consciousness slipping away from him. He instinctively knew that if he lost his senses during a period of time-stopping, events would pick up where they had left off ... without him. For Purnie, this would be death. If he had to lose consciousness, he knew he must first resume time. Step by step he plodded up the little hill, pausing every now and then to consider if this were the moment to start time before it was too late. With his energy fast draining away, he reached the top of the knoll, and he turned to look down once more on the group below. Then he knew how much his mind and body had suffered: when he ordered time to resume, nothing happened. His heart sank. He wasn't afraid of death, and he knew that if he died the oceans would roll again and his friends would move about. But he wanted to see them safe. He tried to clear his mind for supreme effort. There was no urging time to start. He knew he couldn't persuade it by bits and pieces, first slowly then full ahead. Time either progressed or it didn't. He had to take one viewpoint or the other. Then, without knowing exactly when it happened, his mind took command.... His friends came to life. The first one he saw stir lay on his stomach and pounded his fists on the beach. A flood of relief settled over Purnie as sounds came from the animal. "What's the matter with me? Somebody tell me! Am I nuts? Miles! Schick! What's happening?" "I'm coming, Rhodes! Heaven help us, man—I saw it, too. We're either crazy or those damn logs are alive!" "It's not the logs. How about us? How'd we get out of the water? Miles, we're both cracking." "I'm telling you, man, it's the logs, or rocks or whatever they are. I was looking right at them. First they're on top of me, then they're piled up over there!" "Damnit, the logs didn't pick us up out of the ocean, did they? Captain Benson!" "Are you men all right?" "Yes sir, but—" "Who saw exactly what happened?" "I'm afraid we're not seeing right, Captain. Those logs—" "I know, I know. Now get hold of yourselves. We've got to round up the others and get out of here while time is on our side." "But what happened, Captain?" "Hell, Rhodes, don't you think I'd like to know? Those logs are so old they're petrified. The whole bunch of us couldn't lift one. It would take super-human energy to move one of those things." "I haven't seen anything super-human. Those ostriches down there are so busy eating seaweed—" "All right, let's bear a hand here with the others. Some of them can't walk. Where's Forbes?" "He's sitting down there in the water, Captain, crying like a baby. Or laughing. I can't tell which." "We'll have to get him. Miles, Schick, come along. Forbes! You all right?" "Ho-ho-ho! Seventeen! Seventeen! Seventeen planets, Benson, and they'll do anything I say! This one's got a mind of its own. Did you see that little trick with the rocks? Ho-ho!" "See if you can find his gun, Schick; he'll either kill himself or one of us. Tie his hands and take him back to the ship. We'll be along shortly." "Hah-hah-hah! Seventeen! Benson, I'm holding you personally responsible for this. Hee-hee!" Purnie opened his eyes as consciousness returned. Had his friends gone? He pulled himself along on his stomach to a position between two rocks, where he could see without being seen. By the light of the twin moons he saw that they were leaving, marching away in groups of two and three, the weak helping the weaker. As they disappeared around the curving shoreline, the voices of the last two, bringing up the rear far behind the others, fell faintly on his ears over the sound of the surf. "Is it possible that we're all crazy, Captain?" "It's possible, but we're not." "I wish I could be sure." "See Forbes up ahead there? What do you think of him?" "I still can't believe it." "He'll never be the same." "Tell me something. What was the most unusual thing you noticed back there?" "You must be kidding, sir. Why, the way those logs were off of us suddenly—" "Yes, of course. But I mean beside that." "Well, I guess I was kind of busy. You know, scared and mixed up." "But didn't you notice our little pop-eyed friend?" "Oh, him. I'm afraid not, Captain. I—I guess I was thinking mostly of myself." "Hmmm. If I could only be sure I saw him. If only someone else saw him too." "I'm afraid I don't follow you, sir." "Well, damn it all, you know that Forbes took a pot shot at him. Got him in the leg. That being the case, why would the fuzzy little devil come back to his tormentors—back to us—when we were trapped under those logs?" "Well, I guess as long as we were trapped, he figured we couldn't do him any more harm.... I'm sorry, that was a stupid answer. I guess I'm still a little shaky." "Forget it. Look, you go ahead to the ship and make ready for take-off. I'll join you in a few minutes. I think I'll go back and look around. You know. Make sure we haven't left anyone." "No need to do that. They're all ahead of us. I've checked." "That's my responsibility, Cabot, not yours. Now go on." As Purnie lay gathering strength for the long trek home, he saw through glazed eyes one of the animals coming back along the beach. When it was nearly directly below him, he could hear it making sounds that by now had become familiar. "Where are you?" Purnie paid little attention to the antics of his friend; he was beyond understanding. He wondered what they would say at home when he returned. "We've made a terrible mistake. We—" The sounds faded in and out on Purnie's ears as the creature turned slowly and called in different directions. He watched the animal walk over to the pile of scattered logs and peer around and under them. "If you're hurt I'd like to help!" The twin moons were high in the sky now, and where their light broke through the swirling clouds a double shadow was cast around the animal. With foggy awareness, Purnie watched the creature shake its head slowly, then walk away in the direction of the others. Purnie's eyes stared, without seeing, at the panorama before him. The beach was deserted now, and his gaze was transfixed on a shimmering white square floating on the ocean. Across it, the last thing Purnie ever saw, was emblazoned the word FORBES.
|
A. Self-interested and ignorant
|
What architecture has the neural network?
|
### Introduction
Following a turbulent election season, 2016's cyber world is awash with hate speech. Automatic detection of hate speech has become an urgent need since human supervision is unable to deal with large quantities of emerging texts. Context information, by our definition, is the text, symbols or any other kind of information related to the original text. While intuitively, context accompanying hate speech is useful for detecting hate speech, context information of hate speech has been overlooked in existing datasets and automatic detection models. Online hate speech tends to be subtle and creative, which makes context especially important for automatic hate speech detection. For instance, (1) barryswallows: Merkel would never say NO This comment is posted for the News titled by "German lawmakers approve 'no means no' rape law after Cologne assaults". With context, it becomes clear that this comment is a vicious insult towards female politician. However, almost all the publicly available hate speech annotated datasets do not contain context information. BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 . We have created a new dataset consisting of 1528 Fox News user comments, which were taken from 10 complete discussion threads for 10 widely read Fox News articles. It is different from previous datasets from the following two perspectives. First, it preserves rich context information for each comment, including its user screen name, all comments in the same thread and the news article the comment is written for. Second, there is no biased data selection and all comments in each news comment thread were annotated. In this paper, we explored two types of models, feature-based logistic regression models and neural network models, in order to incorporate context information in automatic hate speech detection. First, logistic regression models have been used in several prior hate speech detection studies BIBREF4 , BIBREF5 , BIBREF6 , BIBREF7 , BIBREF8 , BIBREF0 , BIBREF2 , BIBREF9 and various features have been tried including character-level and word-level n-gram features, syntactic features, linguistic features, and comment embedding features. However, all the features were derived from the to-be-classified text itself. In contrast, we experiment with logistic regression models using features extracted from context text as well. Second, neural network models BIBREF10 , BIBREF11 , BIBREF12 have the potential to capture compositional meanings of text, but they have not been well explored for online hate speech detection until recently BIBREF13 . We experiment with neural net models containing separate learning components that model compositional meanings of context information. Furthermore, recognizing unique strengths of each type of models, we build ensemble models of the two types of models. Evaluation shows that context-aware logistic regression models and neural net models outperform their counterparts that are blind with context information. Especially, the final ensemble models outperform a strong baseline system by around 10% in F1-score. ### Related Works
Recently, a few datasets with human labeled hate speech have been created, however, most of existing datasets do not contain context information. Due to the sparsity of hate speech in everyday posts, researchers tend to sample candidates from bootstrapping instead of random sampling, in order to increase the chance of seeing hate speech. Therefore, the collected data instances are likely to be from distinct contexts. For instance, in the Primary Data Set described in BIBREF14 and later used by BIBREF9 , 10% of the dataset is randomly selected while the remaining consists of comments tagged by users and editors. BIBREF15 built a balanced data set of 24.5k tweets by selecting from Twitter accounts that claimed to be racist or were deemed racist using their followed news sources. BIBREF5 collected hateful tweets related to the murder of Drummer Lee Rigby in 2013. BIBREF0 provided a corpus of 16k annotated tweets in which 3.3k are labeled as sexist and 1.9k are labeled as racist. They created this corpus by bootstrapping from certain key words ,specific hashtags and certain prolific users. BIBREF16 created a dataset of 9000 human labeled paragraphs that were collected using regular expression matching in order to find hate speech targeting Judaism and Israel. BIBREF7 extracted data instances from instagram that were associated with certain user accounts. BIBREF2 presented a very large corpus containing over 115k Wikipedia comments that include around 37k randomly sampled comments and the remaining 78k comments were selected from Wikipedia blocked comments. Most of existing hate speech detection models are feature-based and use features derived from the target text itself. BIBREF5 experimented with different classification methods including Bayesian Logistic Regression, Random Forest Decision Trees and SVMs, using features such as n-grams, reduced n-grams, dependency paths, and hateful terms. BIBREF0 proposed a logistic regression model using character n-gram features. BIBREF14 used the paragraph2vec for joint modeling of comments and words, then the generated embeddings were used as feature in a logistic regression model. BIBREF9 experimented with various syntactic, linguistic and distributional semantic features including word length, sentence length, part of speech tags, and embedding features, in order to improve performance of logistic regression classifiers. Recently, BIBREF17 surveyed current approaches for hate speech detection, which interestingly also called to attention on modeling context information for resolving difficult hate speech instances. ### Corpus Overview
The Fox News User Comments corpus consists of 1528 annotated comments (435 labeled as hateful) that were posted by 678 different users in 10 complete news discussion threads in the Fox News website. The 10 threads were manually selected and represent popular discussion threads during August 2016. All of the comments included in these 10 threads were annotated. The number of comments in each of the 10 threads is roughly equal. Rich context information was kept for each comment, including its user screen name, the comments and their nested structure and the original news article. The data corpus along with annotation guidelines is posted on github. ### Annotation Guidelines
Our annotation guidelines are similar to the guidelines used by BIBREF9 . We define hateful speech to be the language which explicitly or implicitly threatens or demeans a person or a group based upon a facet of their identity such as gender, ethnicity, or sexual orientation. The labeling of hateful speech in our corpus is binary. A comment will be labeled as hateful or non-hateful. ### Annotation Procedure
We identified two native English speakers for annotating online user comments. The two annotators first discussed and practices before they started annotation. They achieved a surprisingly high Kappa score BIBREF18 of 0.98 on 648 comments from 4 threads. We think that thorough discussions in the training stage is the key for achieving this high inter-agreement. For those comments which annotators disagreed on, we label them as hateful as long as one annotator labeled them as hateful. Then one annotator continued to annotate the remaining 880 comments from the remaining six discussion threads. ### Characteristics in Fox News User Comments corpus
Hateful comments in the Fox News User Comments Corpus is often subtle, creative and implicit. Therefore, context information is necessary in order to accurately identify such hate speech. The hatefulness of many comments depended on understanding their contexts. For instance, (3) mastersundholm: Just remember no trabjo no cervesa This comment is posted for the news "States moving to restore work requirements for food stamp recipients". This comment implies that Latino immigrants abuse the usage of food stamp policy, which is clearly a stereotyping. Many hateful comments use implicit and subtle language, which contain no clear hate indicating word or phrase. In order to recognize such hard cases, we hypothesize that neural net models are more suitable by capturing overall composite meanings of a comment. For instance, the following comment is a typical implicit stereotyping against women. (4) MarineAssassin: Hey Brianne - get in the kitchen and make me a samich. Chop Chop 11% of our annotated comments have more than 50 words each. In such long comments, the hateful indicators usually appear in a small region of a comment while the majority of the comment is neutral. For example, (5) TMmckay: I thought ...115 words... Too many blacks winning, must be racist and needs affirmative action to make whites equally win! Certain user screen names indicate hatefulness, which imply that comments posted by these users are likely to contain hate speech. In the following example, commie is a slur for communists. (6)nocommie11: Blah blah blah. Israel is the only civilized nation in the region to keep the unwashed masses at bay. ### Logistic Regression Models
In logistic regression models, we extract four types of features, word-level and character-level n-gram features as well as two types of lexicon derived features. We extract these four types of features from the target comment first. Then we extract these features from two sources of context texts, specifically the title of the news article that the comment was posted for and the screen name of the user who posted the comment. For logistic regression model implementation, we use l2 loss. We adopt the balanced class weight as described in Scikit learn. Logistic regression model with character-level n-gram features is presented as a strong baseline for comparison since it was shown very effective. BIBREF0 , BIBREF9 For character level n-grams, we extract character level bigrams, tri-grams and four-grams. For word level n-grams, we extract unigrams and bigrams. Linguistic Inquiry and Word Count, also called LIWC, has been proven useful for text analysis and classification BIBREF19 . In the LIWC dictionary, each word is labeled with several semantic labels. In our experiment, we use the LIWC 2015 dictionary which contain 125 semantic categories. Each word is converted into a 125 dimension LIWC vector, one dimension per semantic category. The LIWC feature vector for a comment or its context is a 125 dimension vector as well, which is the sum of all its words' LIWC vectors. NRC emotion lexicon contains a list of English words that were labeled with eight basic emotions (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust) and sentiment polarities (negative and positive) BIBREF20 . We use NRC emotion lexicon to capture emotion clues in text. Each word is converted into a 10 dimension emotion vector, corresponding to eight emotion types and two polarity labels. The emotion vector for a comment or its context is a 10 dimension vector as well, which is the sum of all its words' emotion vectors. As shown in table TABREF20 , given comment as the only input content, the combination of character n-grams, word n-grams, LIWC feature and NRC feature achieves the best performance. It shows that in addition to character level features, adding more features can improve hate speech detection performance. However, the improvement is limited. Compared with baseline model, the F1 score only improves 1.3%. In contrast, when context information was taken into account, the performance greatly improved. Specifically, after incorporating features extracted from the news title and username, the model performance was improved by around 4% in both F1 score and AUC score. This shows that using additional context based features in logistic regression models is useful for hate speech detection. ### Neural Network Models
Our neural network model mainly consists of three parallel LSTM BIBREF21 layers. It has three different inputs, including the target comment, its news title and its username. Comment and news title are encoded into a sequence of word embeddings. We use pre-trained word embeddings in word2vec. Username is encoded into a sequence of characters. We use one-hot encoding of characters. Comment is sent into a bi-directional LSTM with attention mechanism. BIBREF22 . News title and username are sent into a bi-directional LSTM. Note that we did not apply attention mechanism to the neural network models for username and news title because both types of context are relatively short and attention mechanism tends to be useful when text input is long. The three LSTM output layers are concatenated, then connected to a sigmoid layer, which outputs predictions. The number of hidden units in each LSTM used in our model is set to be 100. The recurrent dropout rate of LSTMs is set to 0.2. In addition, we use binary cross entropy as the loss function and a batch size of 128. The neural network models are trained for 30 epochs. As shown in table TABREF21 , given comment as the only input content, the bi-directional LSTM model with attention mechanism achieves the best performance. Note that the attention mechanism significantly improves the hate speech detection performance of the bi-directional LSTM model. We hypothesize that this is because hate indicator phrases are often concentrated in a small region of a comment, which is especially the case for long comments. ### Ensemble Models
To study the difference of logistic regression model and neural network model and potentially get performance improvement, we will build and evaluate ensemble models. As shown in table TABREF24 , both ensemble models significantly improved hate speech detection performance. Figure FIGREF28 shows the system prediction results of comments that were labeled as hateful in the dataset. It can be seen that the two models perform differently. We further examined predicted comments and find that both types of models have unique strengths in identifying certain types of hateful comments. The feature-based logistic regression models are capable of making good use of character-level n-gram features, which are powerful in identifying hateful comments that contains OOV words, capitalized words or misspelled words. We provide two examples from the hateful comments that were only labeled by the logistic regression model: (7)kmawhmf:FBLM. Here FBLM means fuck Black Lives Matter. This hateful comment contains only character information which can exactly be made use of by our logistic regression model. (8)SFgunrmn: what a efen loon, but most femanazis are. This comment deliberately misspelled feminazi for femanazis, which is a derogatory term for feminists. It shows that logistic regression model is capable in dealing with misspelling. The LSTM with attention mechanism are suitable for identifying specific small regions indicating hatefulness in long comments. In addition, the neural net models are powerful in capturing implicit hateful language as well. The following are two hateful comment examples that were only identified by the neural net model: (9)freedomscout: @LarJass Many religions are poisonous to logic and truth, that much is true...and human beings still remain fallen human beings even they are Redeemed by the Sacrifice of Jesus Christ. So there's that. But the fallacies of thinking cannot be limited or attributed to religion but to error inherent in human motivation, the motivation to utter self-centeredness as fallen sinful human beings. Nearly all of the world's many religions are expressions of that utter sinful nature...Christianity and Judaism being the sole exceptions. This comment is expressing the stereotyping against religions which are not Christian or Judaism. The hatefulness is concentrated within the two bolded segments. (10)mamahattheridge: blacks Love being victims. In this comment, the four words themselves are not hateful at all. But when combined together, it is clearly hateful against black people. ### Evaluation
We evaluate our model by 10 fold cross validation using our newly created Fox News User Comments Corpus. Both types of models use the exact same 10 folds of training data and test data. We report experimental results using multiple metrics, including accuracy, precision/recall/F1-score, and accuracy area under curve (AUC). ### Experimental Results
Table TABREF20 shows the performance of logistic regression models. The first section of table TABREF20 shows the performance of logistic regression models using features extracted from a target comment only. The result shows that the logistic regression model was improved in every metric after adding both word-level n-gram features and lexicon derived features. However, the improvements are moderate. The second section shows the performance of logistic regression models using the four types of features extracted from both a target comment and its contextsThe result shows that the logistic regression model using features extracted from a comment and both types of context achieved the best performance and obtained improvements of 2.8% and 2.5% in AUC score and F1-score respectively. Table TABREF21 shows the performance of neural network models. The first section of table TABREF21 shows the performance of several neural network models that use comments as the only input. The model names are self-explanatory. We can see that the attention mechanism coupled with the bi-directional LSTM neural net greatly improved the online hate speech detection, by 5.7% in AUC score. The second section of table TABREF21 shows performance of the best neural net model (bi-directional LSTM with attention) after adding additional learning components that take context as input. The results show that adding username and news title can both improve model performance. Using news title gives the best F1 score while using both news title and username gives the best AUC score. Table TABREF24 shows performance of ensemble models by combining prediction results of the best context-aware logistic regression model and the best context-aware neural network model. We used two strategies in combining prediction results of two types of models. Specifically, the Max Score Ensemble model made the final decisions based on the maximum of two scores assigned by the two separate models; instead, the Average Score Ensemble model used the average score to make final decisions. We can see that both ensemble models further improved hate speech detection performance compared with using one model only and achieved the best classification performance. Compared with the logistic regression baseline, the Max Score Ensemble model improved the recall by more than 20% with a comparable precision and improved the F1 score by around 10%, in addition, the Average Score Ensemble model improved the AUC score by around 7%. ### Conclusion
We demonstrated the importance of utilizing context information for online hate speech detection. We first presented a corpus of hateful speech consisting of full threads of online discussion posts. In addition, we presented two types of models, feature-based logistic regression models and neural network models, in order to incorporate context information for improving hate speech detection performance. Furthermore, we show that ensemble models leveraging strengths of both types of models achieve the best performance for automatic online hate speech detection. Table 1: Performance of Logistic Regression Models Table 2: Performance of Neural Network Models Table 3: Performance of Ensemble Models Figure 1: System Prediction Results of Comments that were Annotated as Hateful
|
three parallel LSTM BIBREF21 layers
|
What neural configurations are explored?
|
### Introduction
Students are exposed to simple arithmetic word problems starting in elementary school, and most become proficient in solving them at a young age. Automatic solvers of such problems could potentially help educators, as well as become an integral part of general question answering services. However, it has been challenging to write programs to solve even such elementary school level problems well. Solving a math word problem (MWP) starts with one or more sentences describing a transactional situation to be understood. The sentences are processed to produce an arithmetic expression, which is evaluated to provide an answer. Recent neural approaches to solving arithmetic word problems have used various flavors of recurrent neural networks (RNN) as well as reinforcement learning. Such methods have had difficulty achieving a high level of generalization. Often, systems extract the relevant numbers successfully but misplace them in the generated expressions. More problematic, they get the arithmetic operations wrong. The use of infix notation also requires pairs of parentheses to be placed and balanced correctly, bracketing the right numbers. There have been problems with parentheses placement as well. Correctly extracting the numbers in the problem is necessary. Figure FIGREF1 gives examples of some infix representations that a machine learning solver can potentially produce from a simple word problem using the correct numbers. Of the expressions shown, only the first one is correct. After carefully observing expressions that actual problem solvers have generated, we want to explore if the use of infix notation may itself be a part of the problem because it requires the generation of additional characters, the open and close parentheses, which must be balanced and placed correctly. The actual numbers appearing in MWPs vary widely from problem to problem. Real numbers take any conceivable value, making it almost impossible for a neural network to learn representations for them. As a result, trained programs sometimes generate expressions that have seemingly random numbers. For example, in some runs, a trained program could generate a potentially inexplicable expression such as $(25.01 - 4) * 9$ for the problem given in Figure FIGREF1, with one or more numbers not in the problem sentences. We hypothesize that replacing the numbers in the problem statement with generic tags like $\rm \langle n1 \rangle $, $\rm \langle n2 \rangle $, and $\rm \langle n3 \rangle $ and saving their values as a pre-processing step, does not take away from the generality of the solution, but suppresses the problem of fertility in number generation leading to the introduction of numbers not present in the question sentences. Another idea we want to test is whether a neural network which has been pre-trained to acquire language knowledge is better able to “understand" the problem sentences. Pre-training with a large amount of arithmetic-related text is likely to help develop such knowledge, but due to the lack of large such focused corpora, we want to test whether pre-training with a sufficient general corpus is beneficial. In this paper, we use the Transformer model BIBREF0 to solve arithmetic word problems as a particular case of machine translation from text to the language of arithmetic expressions. Transformers in various configurations have become a staple of NLP in the past two years. Past neural approaches did not treat this problem as pure translation like we do, and additionally, these approaches usually augmented the neural architectures with various external modules such as parse trees or used deep reinforcement learning, which we do not do. In this paper, we demonstrate that Transformers can be used to solve MWPs successfully with the simple adjustments we describe above. We compare performance on four individual datasets. In particular, we show that our translation-based approach outperforms state-of-the-art results reported by BIBREF1, BIBREF2, BIBREF3, BIBREF4, BIBREF5 by a large margin on three of four datasets tested. On average, our best neural architecture outperforms previous results by almost 10%, although our approach is conceptually more straightforward. We organize our paper as follows. The second section presents related work. Then, we discuss our approach. We follow by an analysis of experimental results and compare them to those of other recent approaches. We also discuss our successes and shortcomings. Finally, we share our concluding thoughts and end with our direction for future work. ### Related Work
Past strategies have used rules and templates to match sentences to arithmetic expressions. Some such approaches seemed to solve problems impressively within a narrow domain, but performed poorly when out of domain, lacking generality BIBREF6, BIBREF7, BIBREF8, BIBREF9. Kushman et al. BIBREF3 used feature extraction and template-based categorization by representing equations as expression forests and finding a near match. Such methods required human intervention in the form of feature engineering and development of templates and rules, which is not desirable for expandability and adaptability. Hosseini et al. BIBREF2 performed statistical similarity analysis to obtain acceptable results, but did not perform well with texts that were dissimilar to training examples. Existing approaches have used various forms of auxiliary information. Hosseini et al. BIBREF2 used verb categorization to identify important mathematical cues and contexts. Mitra and Baral BIBREF10 used predefined formulas to assist in matching. Koncel-Kedziorski et al. BIBREF11 parsed the input sentences, enumerated all parses, and learned to match, requiring expensive computations. Roy and Roth BIBREF12 performed searches for semantic trees over large spaces. Some recent approaches have transitioned to using neural networks. Semantic parsing takes advantage of RNN architectures to parse MWPs directly into equations or expressions in a math-specific language BIBREF9, BIBREF13. RNNs have shown promising results, but they have had difficulties balancing parenthesis, and also, sometimes incorrectly choose numbers when generating equations. Rehman et al. BIBREF14 used POS tagging and classification of equation templates to produce systems of equations from third-grade level MWPs. Most recently, Sun et al. BIBREF13 used a Bi-Directional LSTM architecture for math word problems. Huang et al. BIBREF15 used a deep reinforcement learning model to achieve character placement in both seen and novel equation templates. Wang et al. BIBREF1 also used deep reinforcement learning. ### Approach
We view math word problem solving as a sequence-to-sequence translation problem. RNNs have excelled in sequence-to-sequence problems such as translation and question answering. The recent introduction of attention mechanisms has improved the performance of RNN models. Vaswani et al. BIBREF0 introduced the Transformer network, which uses stacks of attention layers instead of recurrence. Applications of Transformers have achieved state-of-the-art performance in many NLP tasks. We use this architecture to produce character sequences that are arithmetic expressions. The models we experiment with are easy and efficient to train, allowing us to test several configurations for a comprehensive comparison. We use several configurations of Transformer networks to learn the prefix, postfix, and infix notations of MWP equations independently. Prefix and postfix representations of equations do not contain parentheses, which has been a source of confusion in some approaches. If the learned target sequences are simple, with fewer characters to generate, it is less likely to make mistakes during generation. Simple targets also may help the learning of the model to be more robust. Experimenting with all three representations for equivalent expressions may help us discover which one works best. We train on standard datasets, which are readily available and commonly used. Our method considers the translation of English text to simple algebraic expressions. After performing experiments by training directly on math word problem corpora, we perform a different set of experiments by pre-training on a general language corpus. The success of pre-trained models such as ELMo BIBREF16, GPT-2 BIBREF17, and BERT BIBREF18 for many natural language tasks, provides reasoning that pre-training is likely to produce better learning by our system. We use pre-training so that the system has some foundational knowledge of English before we train it on the domain-specific text of math word problems. However, the output is not natural language but algebraic expressions, which is likely to limit the effectiveness of such pre-training. ### Approach ::: Data
We work with four individual datasets. The datasets contain addition, subtraction, multiplication, and division word problems. AI2 BIBREF2. AI2 is a collection of 395 addition and subtraction problems, containing numeric values, where some may not be relevant to the question. CC BIBREF19. The Common Core dataset contains 600 2-step questions. The Cognitive Computation Group at the University of Pennsylvania gathered these questions. IL BIBREF4. The Illinois dataset contains 562 1-step algebra word questions. The Cognitive Computation Group compiled these questions also. MAWPS BIBREF20. MAWPS is a relatively large collection, primarily from other MWP datasets. We use 2,373 of 3,915 MWPs from this set. The problems not used were more complex problems that generate systems of equations. We exclude such problems because generating systems of equations is not our focus. We take a randomly sampled 95% of examples from each dataset for training. From each dataset, MWPs not included in training make up the testing data used when generating our results. Training and testing are repeated three times, and reported results are an average of the three outcomes. ### Approach ::: Representation Conversion
We take a simple approach to convert infix expressions found in the MWPs to the other two representations. Two stacks are filled by iterating through string characters, one with operators found in the equation and the other with the operands. From these stacks, we form a binary tree structure. Traversing an expression tree in pre-order results in a prefix conversion. Post-order traversal gives us a postfix expression. Three versions of our training and testing data are created to correspond to each type of expression. By training on different representations, we expect our test results to change. ### Approach ::: Pre-training
We pre-train half of our networks to endow them with a foundational knowledge of English. Pre-training models on significant-sized language corpora have been a common approach recently. We explore the pre-training approach using a general English corpus because the language of MWPs is regular English, interspersed with numerical values. Ideally, the corpus for pre-training should be a very general and comprehensive corpus like an English Wikipedia dump or many gigabytes of human-generated text scraped from the internet like GPT-2 BIBREF21 used. However, in this paper, we want to perform experiments to see if pre-training with a smaller corpus can help. In particular, for this task, we use the IMDb Movie Reviews dataset BIBREF22. This set contains 314,041 unique sentences. Since movie reviewers wrote this data, it is a reference to natural language not related to arithmetic. Training on a much bigger and general corpus may make the language model stronger, but we leave this for future work. We compare pre-trained models to non-pre-trained models to observe performance differences. Our pre-trained models are trained in an unsupervised fashion to improve the encodings of our fine-tuned solvers. In the pre-training process, we use sentences from the IMDb reviews with a target output of an empty string. We leave the input unlabelled, which focuses the network on adjusting encodings while providing unbiased decoding when we later change from IMDb English text to MWP-Data. ### Approach ::: Method: Training and Testing
The input sequence is a natural language specification of an arithmetic word problem. The MWP questions and equations have been encoded using the subword text encoder provided by the TensorFlow Datasets library. The output is an expression in prefix, infix, or postfix notation, which then can be manipulated further and solved to obtain a final answer. All examples in the datasets contain numbers, some of which are unique or rare in the corpus. Rare terms are adverse for generalization since the network is unlikely to form good representations for them. As a remedy to this issue, our networks do not consider any relevant numbers during training. Before the networks attempt any translation, we pre-process each question and expression by a number mapping algorithm. This algorithm replaces each numeric value with a corresponding identifier (e.g., $\langle n1 \rangle $, $\langle n2 \rangle $, etc.), and remembers the necessary mapping. We expect that this approach may significantly improve how networks interpret each question. When translating, the numbers in the original question are tagged and cached. From the encoded English and tags, a predicted sequence resembling an expression presents itself as output. Since each network's learned output resembles an arithmetic expression (e.g., $\langle n1 \rangle + \langle n2 \rangle * \langle n3 \rangle $), we use the cached tag mapping to replace the tags with the corresponding numbers and return a final mathematical expression. Three representation models are trained and tested separately: Prefix-Transformer, Postfix-Transformer, and Infix-Transformer. For each experiment, we use representation-specific Transformer architectures. Each model uses the Adam optimizer with $beta_1=0.95$ and $beta_2=0.99$ with a standard epsilon of $1 \times e^{-9}$. The learning rate is reduced automatically in each training session as the loss decreases. Throughout the training, each model respects a 10% dropout rate. We employ a batch size of 128 for all training. Each model is trained on MWP data for 300 iterations before testing. The networks are trained on a machine using 1 Nvidia 1080 Ti graphics processing unit (GPU). We compare medium-sized, small, and minimal networks to show if network size can be reduced to increase training and testing efficiency while retaining high accuracy. Networks over six layers have shown to be non-effective for this task. We tried many configurations of our network models, but report results with only three configurations of Transformers. Transformer Type 1: This network is a small to medium-sized network consisting of 4 Transformer layers. Each layer utilizes 8 attention heads with a depth of 512 and a feed-forward depth of 1024. Transformer Type 2: The second model is small in size, using 2 Transformer layers. The layers utilize 8 attention heads with a depth of 256 and a feed-forward depth of 1024. Transformer Type 3: The third type of model is minimal, using only 1 Transformer layer. This network utilizes 8 attention heads with a depth of 256 and a feed-forward depth of 512. ### Approach ::: Method: Training and Testing ::: Objective Function
We calculate the loss in training according to a mean of the sparse categorical cross-entropy formula. Sparse categorical cross-entropy BIBREF23 is used for identifying classes from a feature set, which assumes a large target classification set. Evaluation between the possible translation classes (all vocabulary subword tokens) and the produced class (predicted token) is the metric of performance here. During each evaluation, target terms are masked, predicted, and then compared to the masked (known) value. We adjust the model's loss according to the mean of the translation accuracy after predicting every determined subword in a translation. where $K = |Translation \; Classes|$, $J = |Translation|$, and $I$ is the number of examples. ### Approach ::: Method: Training and Testing ::: Experiment 1: Representation
Some of the problems encountered by prior approaches seem to be attributable to the use of infix notation. In this experiment, we compare translation BLEU-2 scores to spot the differences in representation interpretability. Traditionally, a BLEU score is a metric of translation quality BIBREF24. Our presented BLEU scores represent an average of scores a given model received over each of the target test sets. We use a standard bi-gram weight to show how accurate translations are within a window of two adjacent terms. After testing translations, we calculate an average BLEU-2 score per test set, which is related to the success over that data. An average of the scores for each dataset become the presented value. where $N$ is the number of test datasets, which is 4. ### Approach ::: Method: Training and Testing ::: Experiment 2: State-of-the-art
This experiment compares our networks to recent previous work. We count a given test score by a simple “correct versus incorrect" method. The answer to an expression directly ties to all of the translation terms being correct, which is why we do not consider partial precision. We compare average accuracies over 3 test trials on different randomly sampled test sets from each MWP dataset. This calculation more accurately depicts the generalization of our networks. ### Approach ::: Method: Training and Testing ::: Effect of Pre-training
We also explore the effect of language pre-training, as discussed earlier. This training occurs over 30 iterations, at the start of the two experiments, to introduce a good level of language understanding before training on the MWP data. The same Transformer architectures are also trained solely on the MWP data. We calculate the reported results as: where $R$ is the number of test repetitions, which is 3; $N$ is the number of test datasets, which is 4; $P$ is the number of MWPs, and $C$ is the number of correct equation translations. ### Results
We now present the results of our various experiments. We compare the three representations of target equations and three architectures of the Transformer model in each test. Results of Experiment 1 are given in Table TABREF21. For clarity, the number in parentheses in front of a row is the Transformer type. By using BLEU scores, we assess the translation capability of each network. This test displays how networks transform different math representations to a character summary level. We compare by average BLEU-2 accuracy among our tests in the Average column of Table TABREF21 to communicate these translation differences. To make it easier to understand the results, Table TABREF22 provides a summary of Table TABREF21. Looking at Tables TABREF21 and TABREF22, we note that both the prefix and postfix representations of our target language perform better than the generally used infix notation. The non-pre-trained models perform slightly better than the pre-trained models, and the small or Type 2 models perform slightly better than the minimal-sized and medium-sized Transformer models. The non-pre-trained type 2 prefix Transformer arrangement produced the most consistent translations. Table TABREF23 provides detailed results of Experiment 2. The numbers are absolute accuracies, i.e., they correspond to cases where the arithmetic expression generated is 100% correct, leading to the correct numeric answer. Results by BIBREF1, BIBREF2, BIBREF4, BIBREF5 are sparse but indicate the scale of success compared to recent past approaches. Prefix, postfix, and infix representations in Table TABREF23 show that network capabilities are changed by how teachable the target data is. The values in the last column of Table TABREF23 are summarized in Table TABREF24. How the models compare with respect to accuracy closely resembles the comparison of BLEU scores, presented earlier. Thus, BLEU scores seem to correlate well with accuracy values in our case. While our networks fell short of BIBREF1 AI2 testing accuracy, we present state-of-the-art results for the remaining three datasets. The AI2 dataset is tricky because it has numeric values in the word descriptions that are extraneous or irrelevant to the actual computation, whereas the other datasets have only relevant numeric values. The type 2 postfix Transformer received the highest testing average of 87.2%. Our attempt at language pre-training fell short of our expectations in all but one tested dataset. We had hoped that more stable language understanding would improve results in general. As previously mentioned, using more general and comprehensive corpora of language could help grow semantic ability. ### Results ::: Analysis
All of the network configurations used were very successful for our task. The prefix representation overall provides the most stable network performance. To display the capability of our most successful model (type 2 postfix Transformer), we present some outputs of the network in Figure FIGREF26. The models respect the syntax of math expressions, even when incorrect. For the majority of questions, our translators were able to determine operators based solely on the context of language. Our pre-training was unsuccessful in improving accuracy, even when applied to networks larger than those reported. We may need to use more inclusive language, or pre-train on very math specific texts to be successful. Our results support our thesis of infix limitation. ### Results ::: Analysis ::: Error Analysis
Our system, while performing above standard, could still benefit from some improvements. One issue originates from the algorithmic pre-processing of our questions and expressions. In Figure FIGREF27 we show an example of one such issue. The excerpt comes from a type 3 non-pre-trained Transformer test. The example shows an overlooked identifier, $\langle n1 \rangle $. The issue is attributed to the identifier algorithm only considering numbers in the problem. Observe in the question that the word “eight" is the number we expect to relate to $\langle n2 \rangle $. Our identifying algorithm could be improved by considering such number words and performing conversion to a numerical value. If our algorithm performed as expected, the identifier $\langle n1 \rangle $ relates with 4 (the first occurring number in the question) and $\langle n2 \rangle $ with 8 (the converted number word appearing second in the question). The overall translation was incorrect whether or not our algorithm was successful, but it is essential to analyze problems like these that may result in future improvements. Had all questions been tagged correctly, our performance would have likely improved. ### Conclusions and Future Work
In this paper, we have shown that the use of Transformer networks improves automatic math word problem-solving. We have also shown that the use of postfix target expressions performs better than the other two expression formats. Our improvements are well-motivated but straightforward and easy to use, demonstrating that the well-acclaimed Transformer architecture for language processing can handle MWPs well, obviating the need to build specialized neural architectures for this task. Extensive pre-training over much larger corpora of language has extended the capabilities of many neural approaches. For example, networks like BERT BIBREF18, trained extensively on data from Wikipedia, perform relatively better in many tasks. Pre-training on a much larger corpus remains an extension we would like to try. We want to work with more complex MWP datasets. Our datasets contain basic arithmetic expressions of +, -, * and /, and only up to 3 of them. For example, datasets such as Dolphin18k BIBREF25, consisting of web-answered questions from Yahoo! Answers, require a wider variety of arithmetic operators to be understood by the system. We have noticed that the presence of irrelevant numbers in the sentences for MWPs limits our performance. We can think of such numbers as a sort of adversarial threat to an MWP solver that stress-test it. It may be interesting to explore how to keep a network's performance high, even in such cases. With a hope to further advance this area of research and heighten interests, all of the code and data used is available on GitHub. ### Acknowledgement
The National Science Foundation supports the work reported in this paper under Grant No. 1659788. Any opinions, findings any conclusions or recommendations expressed in this work are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. TABLE I BLEU-2 COMPARISON FOR EXPERIMENT 1. TABLE II SUMMARY OF BLEU SCORES FROM TABLE I. TABLE III TEST RESULTS FOR EXPERIMENT 2 (* DENOTES AVERAGES ON PRESENT VALUES ONLY). TABLE IV SUMMARY OF ACCURACIES FROM TABLE III.
|
tried many configurations of our network models, but report results with only three configurations, Transformer Type 1, Transformer Type 2, Transformer Type 3
|
What is the true cause of Earth’s “plague” and what is its purpose?
A. The plague was facilitated by aliens, described as invisible, ovular beings. Their purpose is to clear Earth of all life and start their own colony.
B. The plague’s true cause is never revealed. Just as Charles suspects at the time of his death, the fall of the human race is completely unreasonable and meaningless.
C. The plague was facilitated by aliens, described as invisible, ovular beings. Their purpose is to exterminate all of Earth’s life in order to start their own planetary garden.
D. The plague was facilitated by aliens, described as invisible, ovular beings. Their purpose is to move from planet to planet exterminating living systems.
|
"Phone Me in Central Park" By JAMES McCONNELL There should be an epitaph for every man, big or little, but a really grand and special one for Loner Charlie. [Transcriber's Note: This etext was produced from Planet Stories Fall 1954. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] Charles turned over on his side to look at her. She lay quietly in the other bed, the most beautiful woman he had ever seen. She was blonde to perfection, exquisitely shaped, and the rich promise of her body was exposed to his view. "Why?" he thought as he looked at her. "Why did it have to happen like this?" The whole thing was still like a dream to him, and as yet he couldn't decide whether it was a good or a bad dream. A year ago she had been unattainable, a face to conjure with in erotic dreams, far beyond his ken. A year ago she had been a public idol, the most popular actress of the day. And he had been a nobody, full of a nobody's idle hopes and schemes. And now he was lying in the bed next to hers in her swank Manhattan apartment in the most exclusive hotel in town. The unrealness of the situation overwhelmed him. His mind was a picture of confused thoughts. Meanings and answers to his questions slithered out of his reach. "God," he said. It was not an exclamation, nor yet an expletive. It was a mere statement of fact. A thought teased at him. Charles looked at the woman again and decided that she still looked beautiful in spite of the harshness of the room's lighting. He touched buttons by the edge of the bed and the illumination quieted to a soft glow, wrapping her in a radiant halo. Charles smiled wanly and got up. He stood by the bed looking at her. "I could have fallen in love with you once. A year ago, perhaps, or longer. But not now. Not now." He turned away and walked to the window. "Now the world is dead. The whole world is dead." New York lay quietly below him. It was the hour of indecision when day has not quite made up its mind to leave and night has not yet attacked in force. The streetlights were already on, making geometric patterns through the dusk of Central Park. Some of the billboards were shining, their relays activated by darkness-sensitized solenoids. A reddish-orange pallor hung from the sky. It had been very pleasant that afternoon. She had given of herself freely, warmly, and Charles had accepted. But then he had known that she would. It was not him, it was the circumstances. Under the circumstances, she would have given herself to any man— "Why did it have to be her—or me? Why should it have to happen to anybody! Why!" She would have given herself to any man— His thoughts beat a rapid crescendo, activating emotions, stimulating sensations of angry rage. He wanted to cry, to weep angry tears of protest. To any man, WHO HAPPENED TO BE THE LAST MAN ON EARTH! Charles picked up a heavy book end off the table and crashed it through the thick pane of window glass. A gust of wind from the outside breezed through the shattered opening, attacking his olfactory patch with the retching smell of decaying flesh. Charles ignored it. Even smells had lost their customary meanings. He felt the rage build up inside again, tearing at his viscera. His stomach clenched up like an angry fist. "But I don't want to be the last man alive!" he shouted. "I don't know what to do! I don't know where to go, how to act! I just don't know—" A paroxysm of sobbing shook his body. Trembling, he dropped to his knees, his head against the cold firmness of the sill, his hands clutched tightly around the jagged edges of the window pane. In spite of the sharp pain that raced through his system, in spite of the bright, warm, red stream that trickled down his face, he knelt by the window for several minutes. " Maybe I'm not the last! " The thought struck him with suddenness, promisingly, edged with swelling comfort to fill his emptiness. Charles got up slowly, noticing for the first time that his fingers were badly cut. He wrapped a handkerchief around them and forgot them. He had to know—he had to find out. As he turned to leave, he noticed again the woman lying in radiant state upon the bed. He walked to her side and leaned over, kissing her gently on the forehead. As he straightened up, his leg caught against her arm, pushing it slightly. The woman's arm slipped from its position and dangled from the edge of the bed like a crazy pendulum. Charles picked it up and folded it across her now cold breasts. He started to pull the sheet over her nude form, then stopped, smiling at his conventionality. After all, it didn't make any difference now. The phonograph was near the door. On sudden impulse he switched it on, turned the volume up full, and in grim jest left it playing Rachmaninoff's Isle of the Dead on full automatic. The music haunted him down the hall to the elevator that he had to run himself. The lobby was littered with debris, human and otherwise. Charles ignored it. The street that led towards the Bureau of Vital Statistics was a mess of desolate carnage. Charles overlooked it. Shop fronts smashed, stores looted, gyro-cars wrecked, proud buildings defaced. "That was it," he said to himself. "Pride. We called this the 'Proud Era.' Everything was better and bigger and nicer to have. Buildings were taller, men were healthier, most of the problems of humanity seemed licked, or nearly so. It was a time of free power, each small unit of population, each section of town operating on perpetual, ever-lasting, automatic atomic piles. "We were free. We seemed, almost, to have accomplished something. The world was running well. No wonder we called it the 'Proud Era.' Life was fun, just a bowl of cherries, until...." Two years ago the animals had started dying. Strangely enough the rats had gone first, to anybody's notice. Sales of poison dropped, scientific laboratories chained to a perpetual rodent-cycle began to complain bitterly. Then the lovers who hunted out and haunted the lonely lanes through the countryside began to remark that the locusts were late that year. The Southern states joyously reported that mosquito control was working to an unprecedented degree. The largest cotton crop ever was forecast and rumors from Mexico had it that no one had died from scorpion bite in several weeks. A month later the meat animals, the birds and the household pets began dropping as rapidly as the flies which had dropped earlier. Congress was called into special session, as were all of the national governments around the world. The U.N. met at emergency sessions to cope with the situation. The president of the world-wide Society for the Prevention of Cruelty to Animals committed suicide. Within a year it was obvious to everyone that man was the only animal left on earth. The panic which had begun with the death of the animals was quieted somewhat by the fact that humans seemed immune to the pandemic. But the lakes full of dead fish caused a great stink and residents along the coasts began to move inland. Sales of perfumes and deodorants soared. Then just one year ago, the first human became infected with the strange malady. Within six months, half of the world's population was gone. Less than a month ago no more than a few thousand people remained in New York. And now.... "I've got to find out," Charles told himself. He meant it, of course, but in a sense he was afraid—afraid that his trip to the Bureau might give him an answer he didn't dare listen to. "But I've got to try." He walked on down the bloody street. Before the plague the Bureau of Vital Statistics had been one of man's crowning achievements. Housed as it was in a huge metallic globe of a building, it contained computers which kept exact account of every human on earth. Compulsory registration and the classification of each individual by means of the discrete patterns of his brain waves had accomplished for man what no ordinary census could have. The machine knew who was alive, who was dead, and where everybody was. Once a year the Bureau issued The Index, an exact accounting of Earth's four billion inhabitants. Four billion names and addresses, compressed into microprint, a tremendous achievement even for the "Proud Era." In all of his life, Charles had never once glanced at The Index. The average person had little necessity to do so since the Bureau information service would answer questions free of charge at any time. Reaching the gigantic building, Charles pushed aside the body of a young man and walked into the main foyer. Passing behind once-guarded doors, he entered the giant computer room and paused in admiration. Only once, before the plague, had he seen the interior of this room. But he still remembered it and he still recalled the powerful emotional experience it had been those many years ago. All children had to have a brain-wave recording made by the Bureau during the first month of their life. And again at the age of 10 each child returned to the Bureau for a recheck. It was for this latter recording that Charles had come to the Bureau some twenty-two years before and a friendly guard had let him peep briefly into the computer room. The impression of intense activity, of organized confusion, of mechanical wonder had remained with him the rest of his life. "So different now," he thought, surveying the room. "Now it's empty, so empty." The machine seemed to reflect the stillness, the very deadness of the world. The silence became unbearable. Charles walked to the master control panel. With newly acquired dexterity he switched the computer screens on and watched them glow to life. All around the world sensitive receiving stations pulsed to activity, sending out searching fingers, hunting for elusive patterns of neutral energy, mapping and tabulating the results. The main computer screen dominated one wall of the room. Other smaller screens clustered around it. On these screens could be graphed the population of any and every part of the globe. An illuminated counter immediately above it would give the numerical strength of the area being sampled while the screen would show population density by individual pinpoints of light that merged to form brightness patterns. "I'll try New York first," he said to himself, knowing that he was a coward, afraid to check the whole world from the start. "I'll start with New York and work up." Charles activated the switches that would flash a schematic map of New York on the screen. "There's bound to be somebody else left here. After all, there were at least twenty of us just a couple of days ago." And one of them, a beautiful woman, had invited him up to her apartment, not because she liked him, but because.... The main screen focused itself, the patterns shifting into a recognizable perceptual image. "Why, it was just yesterday (or was it the day before?) that ten of us, at least, met here to check the figures. There were lots of us alive then." Including the blond young woman who had died just this afternoon.... Charles stopped talking and forced his eyes upwards. Peripheral vision caught first the vague outlines of the lower part of the map. His eyes continued to move, slowly, reluctantly. They caught the over-all relief of Greater New York City—and then concentrated on the single, shining dot at the very heart of the map—and he understood. His eyes stabbed quickly for the counter above the screen. One. He gasped. The counter read one . Charles was by himself, the last person alive in all of New York City. He began to tremble violently. The silence of the room began to press quickly in on him. His frantic fingers searched for the computer controls. New York State. One. The entire United States. One. The western hemisphere, including islands. (Was that a point of light in Brazil? No. Just a ghost image). One. The Pacific area, Asia, Australia, Asia Minor, Russia and the Near East, Africa and then Europe. England! There was a light in England! Someone else still lived! The counter clicked forward. Two! His trembling stopped. He breathed again. "Of course. London was at least as populous as New York City before the plague. It's only logical that—" He stopped. For even as he spoke, the light winked out! The counter clicked again. One. Alone. Alone! Charles screamed. The bottom dropped out from under him! Why? Such a simple question, but in those three letters lay the essence of human nature. Why. The drive of curiosity. Stronger, in a way, than the so-called "basic" drives: hunger, thirst, sex, shelter, warmth, companionship, elimination. Certainly more decisive in the history of the race. Man began to think, to differentiate himself from the other animals, when he first asked the question: "Why?" But thinking about "why" didn't answer the question itself, Charles thought. He looked around him. He was sitting on a bench in Central Park, alone except for a few stray corpses. But the park was fairly free of bodies. "You've got about ten minutes warning," he said to himself. "I guess that most people wanted to die inside of something—inside of anything. Not out in the unprotected open." The silence was like a weight hanging around his neck. Not an insect noise, not the chirp of a bird, not the sound of a car nor the scream of a plane. Not even a breeze to whisper among the leaves, he thought. Civilization equals life equals noise. Silence equals.... Why. His mind kept returning to the question. Of all the people on earth, me. The last. Why me? Average, that's what he was. Height: 5'11". Weight: 165. Age: 32. Status: Married, once upon a time. The Norm, with no significant departures, all down the line. Church member, but not a good one. Could that be it? Could the most normal be the most perfect? Had he led the best of all possible lives? Was that it? Had God, in His infinite wisdom and mercy, spared his life, saved him, singled him out because he was most nearly a saint, most nearly Christ-like, most nearly.... Lies—His mind snapped back to reality. He half smiled. Saint? Christ? The Second Coming? He was no saint. Charles sighed. What about—? Chance. That was it! The laws of probability, the bell-shaped curve, normal distribution, rectilinear regression. More people per square foot in New York than elsewhere. The first person who died was from New York, so the last person who gave way to the disease should come from here too. Spin the wheel; throw the dice; toss the coin. So simple to explain by the laws of chance. No need for any underlying assumptions about good and evil, no need for teleological arguments concerning cause and effect. Simply explain it by chance. Somebody had to be the last to go and that was— "No," Charles said, standing up in the quiet of the spring evening. "No, chance won't do it. No man can reckon with chance. The mind rejects such things. There must be something beyond mere accident. There must be!" He sighed slowly. "So now I'm a hermit, whether or not I like it," he said in derision to the gravel path as he walked along it. "A hermit in the midst of a city of millions of—No, I forgot. There aren't any more people, are there?" It was hard to realize, even now. "A hermit, alone—and I haven't even got a cave...." Charles stopped walking suddenly. No cave, he thought. No place to sleep out the long one, no place to rest while time came to change things around and make them for the better. No place to hide. And suddenly it was the most important thing in life to him to find his "cave." It took him almost an hour to find the proper tools, and better than two hours more of hard, nighttime work to get the hole dug to his satisfaction. It took almost three hours to find the right sort of casket, durable but not too heavy for one man to handle. He carted it out to a grassy plot close to the center of the park where the grave was. He let the coffin down slowly into the depression, then piled up loose dirt on the sloping sides of the hole so that the rain would wash it down over him. "I can't very well bury myself," he said. "I guess it will rain after I'm gone." He looked carefully down at the metallic container. Wait now. There was something wrong, something missing. It was—oh, yes, he caught it. It was the stone. There wasn't any stone to go at the head of the grave. "I'll have to fix that." A sheet of metal, bent double, served for the monument proper. A nearby tool shed yielded up a can of paint and a brush. By the glow of one of the streetlights Charles worked out the inscription. "It ought to be something impressive," he thought out loud. "Something fitting the occasion." What did one say on these situations? There was so little chance to practice up for things like this. But it ought to be good, it ought to be proper. "'In this now hallowed corner of the planet Earth—' No. That sounds too ... too...." Make it simple, he thought. And he finally wrote: HERE LIES THE BODY OF THE LAST MAN ON EARTH Yes. That was it. Simple. Let whoever came afterwards figure out the rest. Let them decide. He smiled and finished the painting. Charles was hungry. He got up and started for one of the restaurants near the park. Later on, when there was more time, he'd find a piece of granite and move it to the plot. He could spend his free time carving on it, copying the inscription. He would make it into a real shrine; maybe he would practice up a bit and try to carve a statue to go with the stone. Somehow, though, since things were ready and it didn't make too much difference, it seemed to Charles that he'd probably have a long time to wait. "Maybe it's just a disease, and I'm immune. I was immune to smallpox. The vaccination never took. That's probably it." He smiled. Strange, but now he wanted very much to go on living, alone or not. There were things he could do, ways to keep occupied. He wouldn't mind it so much. But he wanted more and more desperately with each passing second to retain his foothold on the tenuous path of physical existence. The tantalizing thought of "why" puzzled its way back into his mind. But it seemed less pressing now that he had almost come to the conclusion that he would live for a long time. Later, in a few days perhaps, he would think about it. In a little while he'd have plenty of opportunity for hunting down the answer. This seemed good to him, for now he thought he almost had the answer, if there were an answer. He thought he had seen the solution peering out at him from the recesses of his mind, and he didn't like the expression on its face. Better to forget. Charles reached the broad boulevard. There was a large cafe just across from him, its front window caved in by a large truck. He stumbled and almost fell as he stepped from the curb. "Look at me, nervous as a cat." He was trembling noticeably as he started across the street. "I—" He started to say something, to think something. But some hidden part of his mind clamped down, obscuring the thought, rejecting the concept. The tremor turned to a shake before he reached the far curb, and the first burst of wild pain came as he laid his shoulder against the door to the restaurant. This was the way the plague began, but—His mind quickly repressed the idea. It couldn't be the plague. He was immune! Another burst of pulsating, shattering pain crashed through his body, tearing down the defenses of his mind, putting an end of his thoughts of immunity. Colors flared before his eyes, a persistent, irresistible susurrus flooded his ears. He wanted to protest, but there was no one to listen to him. He appealed to every divinity he knew, all the time knowing it would be useless. His body, out of his voluntary control, tried to run off in all directions at once. Charles struggled to end his body's disorganized responses, to channelize all his energy into one direction. His mind came back into action. He set up his goal; everything else seemed irrelevant: he had to get back to the park, to his hermit's cave, to his long, narrow home. He couldn't die until then. Ten minutes. He was allotted ten minutes before the end. It could have been ten years or ten seconds, for now objective time meant nothing to him. It was not a matter of measuring seconds and minutes. It was a matter of forgetting time and measuring space. He concentrated on the grave; he forced his body to become an unwilling machine. While he could, he walked, forcing himself on. When his legs gave way, he crawled. When his knees buckled, he rolled. When his stomach protested, he vomited. It made no difference. Charles refused to think. Machines, especially half-broken machines, do not think; they only work. Sweating, straining, bleeding, retching, he pushed himself towards his goal, trying to add one final touch of grace and custom to the rude irrationalness of it all. His eyes gave out a few feet from the pit. He felt his way towards it. Convulsions shook his body like a cat shakes a captive mouse. He humped his body forward between the seizures, hands outstretched, searching for the grave. And then he was upon it. One arm reached out for grass, and clutched bare space instead. He was home. He gathered energy from his final reservoirs of strength for one final movement that would throw him headlong into the shallow grave. He tensed his muscles, pulled his limbs up under him and started to roll into the hole. Instantly the thought struck him with paralyzing devastation. The answer to it all poked its face out from the recesses of his mind and sapped the last bit of his energy, corroding his nerves and dying muscles. Now he knew, and the knowing was the end of it. He collapsed at the edge of the pit. Only one arm hung loosely down into it, swinging senseless in the air, pointing accusingly at the empty coffin. The world will end, not with a bang, nor with a whimper, but with the last man's anguished cry at the unreasonableness of it all. Charles screamed. The large, invisible, ovular being that hung suspended over the Empire State Building rested from its exertion. Soon it was approached by another of its kind. "It is finished?" asked the second. "Yes. Just now. I am resting." "I can feel the emptiness of it." "It was very good. Where were you?" "On the next planet out. No beauty to it at all; no system. How was yours?" "Beautiful," said the first. "It went according to the strictest semantic relationship following the purest mathematical principles. They made it easy for me." "Good." "Well, where to now?" "There's another system about four thoughts away. We're due there soon." "All right. Let's go." "What's that you have there?" "Oh, this?" replied the first. "It's a higher neural order compendium the Things here made up. It's what I used." "You can't take it with you, you know. They don't allow souvenirs." "I know." "Well?" "All right, all right. You're so good, see if you can compute the scatter probability." The first being moved imperceptably and the heavy plastoid binding of the book disappeared. The thousands of pages dropped softly, caught at the wind like hungry sails, separated, and pulled by the fingers of gravity, went their disparate ways. Here a page scuttled into a broken window of the Chrysler Building (read the names: Aabat, Aabbs, Aabbt). Here a page landed upright on the head of one of the library lions and sloughed softly to the ground (read the names: Looman, Loomana, Loomanabsky). Here another page crept in between the cracks of a pier on the riverfront, dropping gently to the caressing eddies of the water (read the names: Smith, Smitha, Smitj). And here two pages danced down into Central Park, pirouetted, promenaded, and finally came to rest against a propped-up piece of metal (read the names: Whit, Whita, Whitacomb). It was not until the dusty morning sun stirred up the breezes that they fluttered down into the shallow hole beneath, unnoticed. The writing on the metal, until then partially obscured by the papers, became legible: HERE LIES THE BODY OF THE LAST MAN ON EARTH— CHARLES J. ZZYZST GO TO HELL!
|
D. The plague was facilitated by aliens, described as invisible, ovular beings. Their purpose is to move from planet to planet exterminating living systems.
|
How does Claney feel about Mikuta?
A. Claney likes Mikuta. He can trust Mikuta.
B. Claney doesn't like Mikuta. Mikuta makes too many mistakes.
C. Claney likes Mikuta, but Mikuta makes too many mistakes.
D. Claney doesn't like Mikuta. He can't trust Mikuta.
|
Brightside Crossing by Alan E. Nourse JAMES BARON was not pleased to hear that he had had a visitor when he reached the Red Lion that evening. He had no stomach for mysteries, vast or trifling, and there were pressing things to think about at this time. Yet the doorman had flagged him as he came in from the street: “A thousand pardons, Mr. Baron. The gentleman—he would leave no name. He said you’d want to see him. He will be back by eight.” Now Baron drummed his fingers on the table top, staring about the quiet lounge. Street trade was discouraged at the Red Lion, gently but persuasively; the patrons were few in number. Across to the right was a group that Baron knew vaguely—Andean climbers, or at least two of them were. Over near the door he recognized old Balmer, who had mapped the first passage to the core of Vulcan Crater on Venus. Baron returned his smile with a nod. Then he settled back and waited impatiently for the intruder who demanded his time without justifying it. Presently a small, grizzled man crossed the room and sat down at Baron’s table. He was short and wiry. His face held no key to his age—he might have been thirty or a thousand—but he looked weary and immensely ugly. His cheeks and forehead were twisted and brown, with scars that were still healing. The stranger said, “I’m glad you waited. I’ve heard you’re planning to attempt the Brightside.” Baron stared at the man for a moment. “I see you can read telecasts,” he said coldly. “The news was correct. We are going to make a Brightside Crossing.” “At perihelion?” “Of course. When else?” The grizzled man searched Baron’s face for a moment without expression. Then he said slowly, “No, I’m afraid you’re not going to make the Crossing.” “Say, who are you, if you don’t mind?” Baron demanded. “The name is Claney,” said the stranger. There was a silence. Then: “Claney? Peter Claney?” “That’s right.” Baron’s eyes were wide with excitement, all trace of anger gone. “Great balls of fire, man— where have you been hiding? We’ve been trying to contact you for months!” “I know. I was hoping you’d quit looking and chuck the whole idea.” “Quit looking!” Baron bent forward over the table. “My friend, we’d given up hope, but we’ve never quit looking. Here, have a drink. There’s so much you can tell us.” His fingers were trembling. Peter Claney shook his head. “I can’t tell you anything you want to hear.” “But you’ve got to. You’re the only man on Earth who’s attempted a Brightside Crossing and lived through it! And the story you cleared for the news—it was nothing. We need details . Where did your equipment fall down? Where did you miscalculate? What were the trouble spots?” Baron jabbed a finger at Claney’s face. “That, for instance—epithelioma? Why? What was wrong with your glass? Your filters? We’ve got to know those things. If you can tell us, we can make it across where your attempt failed—” “You want to know why we failed?” asked Claney. “Of course we want to know. We have to know.” “It’s simple. We failed because it can’t be done. We couldn’t do it and neither can you. No human beings will ever cross the Brightside alive, not if they try for centuries.” “Nonsense,” Baron declared. “We will.” Claney shrugged. “I was there. I know what I’m saying. You can blame the equipment or the men—there were flaws in both quarters—but we just didn’t know what we were fighting. It was the planet that whipped us, that and the Sun . They’ll whip you, too, if you try it.” “Never,” said Baron. “Let me tell you,” Peter Claney said. I’d been interested in the Brightside for almost as long as I can remember (Claney said). I guess I was about ten when Wyatt and Carpenter made the last attempt—that was in 2082, I think. I followed the news stories like a tri-V serial and then I was heartbroken when they just disappeared. I know now that they were a pair of idiots, starting off without proper equipment, with practically no knowledge of surface conditions, without any charts—they couldn’t have made a hundred miles—but I didn’t know that then and it was a terrible tragedy. After that, I followed Sanderson’s work in the Twilight Lab up there and began to get Brightside into my blood, sure as death. But it was Mikuta’s idea to attempt a Crossing. Did you ever know Tom Mikuta? I don’t suppose you did. No, not Japanese—Polish-American. He was a major in the Interplanetary Service for some years and hung onto the title after he gave up his commission. He was with Armstrong on Mars during his Service days, did a good deal of the original mapping and surveying for the Colony there. I first met him on Venus; we spent five years together up there doing some of the nastiest exploring since the Matto Grasso. Then he made the attempt on Vulcan Crater that paved the way for Balmer a few years later. I’d always liked the Major—he was big and quiet and cool, the sort of guy who always had things figured a little further ahead than anyone else and always knew what to do in a tight place. Too many men in this game are all nerve and luck, with no judgment. The Major had both. He also had the kind of personality that could take a crew of wild men and make them work like a well-oiled machine across a thousand miles of Venus jungle. I liked him and I trusted him. He contacted me in New York and he was very casual at first. We spent an evening here at the Red Lion, talking about old times; he told me about the Vulcan business, and how he’d been out to see Sanderson and the Twilight Lab on Mercury, and how he preferred a hot trek to a cold one any day of the year—and then he wanted to know what I’d been doing since Venus and what my plans were. “No particular plans,” I told him. “Why?” He looked me over. “How much do you weigh, Peter?” I told him one-thirty-five. “That much!” he said. “Well, there can’t be much fat on you, at any rate. How do you take heat?” “You should know,” I said. “Venus was no icebox.” “No, I mean real heat.” Then I began to get it. “You’re planning a trip.” “That’s right. A hot trip.” He grinned at me. “Might be dangerous, too.” “What trip?” “Brightside of Mercury,” the Major said. I whistled cautiously. “At aphelion?” He threw his head back. “Why try a Crossing at aphelion? What have you done then? Four thousand miles of butcherous heat, just to have some joker come along, use your data and drum you out of the glory by crossing at perihelion forty-four days later? No, thanks. I want the Brightside without any nonsense about it.” He leaned across me eagerly. “I want to make a Crossing at perihelion and I want to cross on the surface. If a man can do that, he’s got Mercury. Until then, nobody’s got Mercury. I want Mercury—but I’ll need help getting it.” I’d thought of it a thousand times and never dared consider it. Nobody had, since Wyatt and Carpenter disappeared. Mercury turns on its axis in the same time that it wheels around the Sun, which means that the Brightside is always facing in. That makes the Brightside of Mercury at perihelion the hottest place in the Solar System, with one single exception: the surface of the Sun itself. It would be a hellish trek. Only a few men had ever learned just how hellish and they never came back to tell about it. It was a real hell’s Crossing, but someday, I thought, somebody would cross it. I wanted to be along. The Twilight Lab, near the northern pole of Mercury, was the obvious jumping-off place. The setup there wasn’t very extensive—a rocket landing, the labs and quarters for Sanderson’s crew sunk deep into the crust, and the tower that housed the Solar ’scope that Sanderson had built up there ten years before. Twilight Lab wasn’t particularly interested in the Brightside, of course—the Sun was Sanderson’s baby and he’d picked Mercury as the closest chunk of rock to the Sun that could hold his observatory. He’d chosen a good location, too. On Mercury, the Brightside temperature hits 770° F. at perihelion and the Darkside runs pretty constant at -410° F. No permanent installation with a human crew could survive at either extreme. But with Mercury’s wobble, the twilight zone between Brightside and Darkside offers something closer to survival temperatures. Sanderson built the Lab up near the pole, where the zone is about five miles wide, so the temperature only varies 50 to 60 degrees with the libration. The Solar ’scope could take that much change and they’d get good clear observation of the Sun for about seventy out of the eighty-eight days it takes the planet to wheel around. The Major was counting on Sanderson knowing something about Mercury as well as the Sun when we camped at the Lab to make final preparations. Sanderson did. He thought we’d lost our minds and he said so, but he gave us all the help he could. He spent a week briefing Jack Stone, the third member of our party, who had arrived with the supplies and equipment a few days earlier. Poor Jack met us at the rocket landing almost bawling, Sanderson had given him such a gloomy picture of what Brightside was like. Stone was a youngster—hardly twenty-five, I’d say—but he’d been with the Major at Vulcan and had begged to join this trek. I had a funny feeling that Jack really didn’t care for exploring too much, but he thought Mikuta was God, followed him around like a puppy. It didn’t matter to me as long as he knew what he was getting in for. You don’t go asking people in this game why they do it—they’re liable to get awfully uneasy and none of them can ever give you an answer that makes sense. Anyway, Stone had borrowed three men from the Lab, and had the supplies and equipment all lined up when we got there, ready to check and test. We dug right in. With plenty of funds—tri-V money and some government cash the Major had talked his way around—our equipment was new and good. Mikuta had done the designing and testing himself, with a big assist from Sanderson. We had four Bugs, three of them the light pillow-tire models, with special lead-cooled cut-in engines when the heat set in, and one heavy-duty tractor model for pulling the sledges. The Major went over them like a kid at the circus. Then he said, “Have you heard anything from McIvers?” “Who’s he?” Stone wanted to know. “He’ll be joining us. He’s a good man—got quite a name for climbing, back home.” The Major turned to me. “You’ve probably heard of him.” I’d heard plenty of stories about Ted McIvers and I wasn’t too happy to hear that he was joining us. “Kind of a daredevil, isn’t he?” “Maybe. He’s lucky and skillful. Where do you draw the line? We’ll need plenty of both.” “Have you ever worked with him?” I asked. “No. Are you worried?” “Not exactly. But Brightside is no place to count on luck.” The Major laughed. “I don’t think we need to worry about McIvers. We understood each other when I talked up the trip to him and we’re going to need each other too much to do any fooling around.” He turned back to the supply list. “Meanwhile, let’s get this stuff listed and packed. We’ll need to cut weight sharply and our time is short. Sanderson says we should leave in three days.” Two days later, McIvers hadn’t arrived. The Major didn’t say much about it. Stone was getting edgy and so was I. We spent the second day studying charts of the Brightside, such as they were. The best available were pretty poor, taken from so far out that the detail dissolved into blurs on blow-up. They showed the biggest ranges of peaks and craters and faults, and that was all. Still, we could use them to plan a broad outline of our course. “This range here,” the Major said as we crowded around the board, “is largely inactive, according to Sanderson. But these to the south and west could be active. Seismograph tracings suggest a lot of activity in that region, getting worse down toward the equator—not only volcanic, but sub-surface shifting.” Stone nodded. “Sanderson told me there was probably constant surface activity.” The Major shrugged. “Well, it’s treacherous, there’s no doubt of it. But the only way to avoid it is to travel over the Pole, which would lose us days and offer us no guarantee of less activity to the west. Now we might avoid some if we could find a pass through this range and cut sharp east—” It seemed that the more we considered the problem, the further we got from a solution. We knew there were active volcanoes on the Brightside—even on the Darkside, though surface activity there was pretty much slowed down and localized. But there were problems of atmosphere on Brightside, as well. There was an atmosphere and a constant atmospheric flow from Brightside to Darkside. Not much—the lighter gases had reached escape velocity and disappeared from Brightside millennia ago—but there was CO 2 , and nitrogen, and traces of other heavier gases. There was also an abundance of sulfur vapor, as well as carbon disulfide and sulfur dioxide. The atmospheric tide moved toward the Darkside, where it condensed, carrying enough volcanic ash with it for Sanderson to estimate the depth and nature of the surface upheavals on Brightside from his samplings. The trick was to find a passage that avoided those upheavals as far as possible. But in the final analysis, we were barely scraping the surface. The only way we would find out what was happening where was to be there. Finally, on the third day, McIvers blew in on a freight rocket from Venus. He’d missed the ship that the Major and I had taken by a few hours, and had conned his way to Venus in hopes of getting a hop from there. He didn’t seem too upset about it, as though this were his usual way of doing things and he couldn’t see why everyone should get so excited. He was a tall, rangy man with long, wavy hair prematurely gray, and the sort of eyes that looked like a climber’s—half-closed, sleepy, almost indolent, but capable of abrupt alertness. And he never stood still; he was always moving, always doing something with his hands, or talking, or pacing about. Evidently the Major decided not to press the issue of his arrival. There was still work to do, and an hour later we were running the final tests on the pressure suits. That evening, Stone and McIvers were thick as thieves, and everything was set for an early departure after we got some rest. “And that,” said Baron, finishing his drink and signaling the waiter for another pair, “was your first big mistake.” Peter Claney raised his eyebrows. “McIvers?” “Of course.” Claney shrugged, glanced at the small quiet tables around them. “There are lots of bizarre personalities around a place like this, and some of the best wouldn’t seem to be the most reliable at first glance. Anyway, personality problems weren’t our big problem right then. Equipment worried us first and route next.” Baron nodded in agreement. “What kind of suits did you have?” “The best insulating suits ever made,” said Claney. “Each one had an inner lining of a fiberglass modification, to avoid the clumsiness of asbestos, and carried the refrigerating unit and oxygen storage which we recharged from the sledges every eight hours. Outer layer carried a monomolecular chrome reflecting surface that made us glitter like Christmas trees. And we had a half-inch dead-air space under positive pressure between the two layers. Warning thermocouples, of course—at 770 degrees, it wouldn’t take much time to fry us to cinders if the suits failed somewhere.” “How about the Bugs?” “They were insulated, too, but we weren’t counting on them too much for protection.” “You weren’t!” Baron exclaimed. “Why not?” “We’d be in and out of them too much. They gave us mobility and storage, but we knew we’d have to do a lot of forward work on foot.” Claney smiled bitterly. “Which meant that we had an inch of fiberglass and a half-inch of dead air between us and a surface temperature where lead flowed like water and zinc was almost at melting point and the pools of sulfur in the shadows were boiling like oatmeal over a campfire.” Baron licked his lips. His fingers stroked the cool, wet glass as he set it down on the tablecloth. “Go on,” he said tautly. “You started on schedule?” “Oh, yes,” said Claney, “we started on schedule, all right. We just didn’t quite end on schedule, that was all. But I’m getting to that.” He settled back in his chair and continued. We jumped off from Twilight on a course due southeast with thirty days to make it to the Center of Brightside. If we could cross an average of seventy miles a day, we could hit Center exactly at perihelion, the point of Mercury’s closest approach to the Sun—which made Center the hottest part of the planet at the hottest it ever gets. The Sun was already huge and yellow over the horizon when we started, twice the size it appears on Earth. Every day that Sun would grow bigger and whiter, and every day the surface would get hotter. But once we reached Center, the job was only half done—we would still have to travel another two thousand miles to the opposite twilight zone. Sanderson was to meet us on the other side in the Laboratory’s scout ship, approximately sixty days from the time we jumped off. That was the plan, in outline. It was up to us to cross those seventy miles a day, no matter how hot it became, no matter what terrain we had to cross. Detours would be dangerous and time-consuming. Delays could cost us our lives. We all knew that. The Major briefed us on details an hour before we left. “Peter, you’ll take the lead Bug, the small one we stripped down for you. Stone and I will flank you on either side, giving you a hundred-yard lead. McIvers, you’ll have the job of dragging the sledges, so we’ll have to direct your course pretty closely. Peter’s job is to pick the passage at any given point. If there’s any doubt of safe passage, we’ll all explore ahead on foot before we risk the Bugs. Got that?” McIvers and Stone exchanged glances. McIvers said: “Jack and I were planning to change around. We figured he could take the sledges. That would give me a little more mobility.” The Major looked up sharply at Stone. “Do you buy that, Jack?” Stone shrugged. “I don’t mind. Mac wanted—” McIvers made an impatient gesture with his hands. “It doesn’t matter. I just feel better when I’m on the move. Does it make any difference?” “I guess it doesn’t,” said the Major. “Then you’ll flank Peter along with me. Right?” “Sure, sure.” McIvers pulled at his lower lip. “Who’s going to do the advance scouting?” “It sounds like I am,” I cut in. “We want to keep the lead Bug light as possible.” Mikuta nodded. “That’s right. Peter’s Bug is stripped down to the frame and wheels.” McIvers shook his head. “No, I mean the advance work. You need somebody out ahead—four or five miles, at least—to pick up the big flaws and active surface changes, don’t you?” He stared at the Major. “I mean, how can we tell what sort of a hole we may be moving into, unless we have a scout up ahead?” “That’s what we have the charts for,” the Major said sharply. “Charts! I’m talking about detail work. We don’t need to worry about the major topography. It’s the little faults you can’t see on the pictures that can kill us.” He tossed the charts down excitedly. “Look, let me take a Bug out ahead and work reconnaissance, keep five, maybe ten miles ahead of the column. I can stay on good solid ground, of course, but scan the area closely and radio back to Peter where to avoid the flaws. Then—” “No dice,” the Major broke in. “But why not? We could save ourselves days!” “I don’t care what we could save. We stay together. When we get to the Center, I want live men along with me. That means we stay within easy sight of each other at all times. Any climber knows that everybody is safer in a party than one man alone—any time, any place.” McIvers stared at him, his cheeks an angry red. Finally he gave a sullen nod. “Okay. If you say so.” “Well, I say so and I mean it. I don’t want any fancy stuff. We’re going to hit Center together, and finish the Crossing together. Got that?” McIvers nodded. Mikuta then looked at Stone and me and we nodded, too. “All right,” he said slowly. “Now that we’ve got it straight, let’s go.” It was hot. If I forget everything else about that trek, I’ll never forget that huge yellow Sun glaring down, without a break, hotter and hotter with every mile. We knew that the first few days would be the easiest and we were rested and fresh when we started down the long ragged gorge southeast of the Twilight Lab. I moved out first; back over my shoulder, I could see the Major and McIvers crawling out behind me, their pillow tires taking the rugged floor of the gorge smoothly. Behind them, Stone dragged the sledges. Even at only 30 per cent Earth gravity they were a strain on the big tractor, until the ski-blades bit into the fluffy volcanic ash blanketing the valley. We even had a path to follow for the first twenty miles. I kept my eyes pasted to the big polaroid binocs, picking out the track the early research teams had made out into the edge of Brightside. But in a couple of hours we rumbled past Sanderson’s little outpost observatory and the tracks stopped. We were in virgin territory and already the Sun was beginning to bite. We didn’t feel the heat so much those first days out. We saw it. The refrig units kept our skins at a nice comfortable seventy-five degrees Fahrenheit inside our suits, but our eyes watched that glaring Sun and the baked yellow rocks going past, and some nerve pathways got twisted up, somehow. We poured sweat as if we were in a superheated furnace. We drove eight hours and slept five. When a sleep period came due, we pulled the Bugs together into a square, threw up a light aluminum sun-shield and lay out in the dust and rocks. The sun-shield cut the temperature down sixty or seventy degrees, for whatever help that was. And then we ate from the forward sledge—sucking through tubes—protein, carbohydrates, bulk gelatin, vitamins. The Major measured water out with an iron hand, because we’d have drunk ourselves into nephritis in a week otherwise. We were constantly, unceasingly thirsty. Ask the physiologists and psychiatrists why—they can give you have a dozen interesting reasons—but all we knew, or cared about, was that it happened to be so. We didn’t sleep the first few stops, as a consequence. Our eyes burned in spite of the filters and we had roaring headaches, but we couldn’t sleep them off. We sat around looking at each other. Then McIvers would say how good a beer would taste, and off we’d go. We’d have murdered our grandmothers for one ice-cold bottle of beer. After a few driving periods, I began to get my bearings at the wheel. We were moving down into desolation that made Earth’s old Death Valley look like a Japanese rose garden. Huge sun-baked cracks opened up in the floor of the gorge, with black cliffs jutting up on either side; the air was filled with a barely visible yellowish mist of sulfur and sulfurous gases. It was a hot, barren hole, no place for any man to go, but the challenge was so powerful you could almost feel it. No one had ever crossed this land before and escaped. Those who had tried it had been cruelly punished, but the land was still there, so it had to be crossed. Not the easy way. It had to be crossed the hardest way possible: overland, through anything the land could throw up to us, at the most difficult time possible. Yet we knew that even the land might have been conquered before, except for that Sun. We’d fought absolute cold before and won. We’d never fought heat like this and won. The only worse heat in the Solar System was the surface of the Sun itself. Brightside was worth trying for. We would get it or it would get us. That was the bargain. I learned a lot about Mercury those first few driving periods. The gorge petered out after a hundred miles and we moved onto the slope of a range of ragged craters that ran south and east. This range had shown no activity since the first landing on Mercury forty years before, but beyond it there were active cones. Yellow fumes rose from the craters constantly; their sides were shrouded with heavy ash. We couldn’t detect a wind, but we knew there was a hot, sulfurous breeze sweeping in great continental tides across the face of the planet. Not enough for erosion, though. The craters rose up out of jagged gorges, huge towering spears of rock and rubble. Below were the vast yellow flatlands, smoking and hissing from the gases beneath the crust. Over everything was gray dust—silicates and salts, pumice and limestone and granite ash, filling crevices and declivities—offering a soft, treacherous surface for the Bug’s pillow tires. I learned to read the ground, to tell a covered fault by the sag of the dust; I learned to spot a passable crack, and tell it from an impassable cut. Time after time the Bugs ground to a halt while we explored a passage on foot, tied together with light copper cable, digging, advancing, digging some more until we were sure the surface would carry the machines. It was cruel work; we slept in exhaustion. But it went smoothly, at first. Too smoothly, it seemed to me, and the others seemed to think so, too. McIvers’ restlessness was beginning to grate on our nerves. He talked too much, while we were resting or while we were driving; wisecracks, witticisms, unfunny jokes that wore thin with repetition. He took to making side trips from the route now and then, never far, but a little further each time. Jack Stone reacted quite the opposite; he grew quieter with each stop, more reserved and apprehensive. I didn’t like it, but I figured that it would pass off after a while. I was apprehensive enough myself; I just managed to hide it better. And every mile the Sun got bigger and whiter and higher in the sky and hotter. Without our ultra-violet screens and glare filters we would have been blinded; as it was our eyes ached constantly and the skin on our faces itched and tingled at the end of an eight-hour trek. But it took one of those side trips of McIvers’ to deliver the penultimate blow to our already fraying nerves. He had driven down a side-branch of a long canyon running off west of our route and was almost out of sight in a cloud of ash when we heard a sharp cry through our earphones. I wheeled my Bug around with my heart in my throat and spotted him through the binocs, waving frantically from the top of his machine. The Major and I took off, lumbering down the gulch after him as fast as the Bugs could go, with a thousand horrible pictures racing through our minds.... We found him standing stock-still, pointing down the gorge and, for once, he didn’t have anything to say. It was the wreck of a Bug; an old-fashioned half-track model of the sort that hadn’t been in use for years. It was wedged tight in a cut in the rock, an axle broken, its casing split wide open up the middle, half-buried in a rock slide. A dozen feet away were two insulated suits with white bones gleaming through the fiberglass helmets. This was as far as Wyatt and Carpenter had gotten on their Brightside Crossing. On the fifth driving period out, the terrain began to change. It looked the same, but every now and then it felt different. On two occasions I felt my wheels spin, with a howl of protest from my engine. Then, quite suddenly, the Bug gave a lurch; I gunned my motor and nothing happened. I could see the dull gray stuff seeping up around the hubs, thick and tenacious, splattering around in steaming gobs as the wheels spun. I knew what had happened the moment the wheels gave and, a few minutes later, they chained me to the tractor and dragged me back out of the mire. It looked for all the world like thick gray mud, but it was a pit of molten lead, steaming under a soft layer of concealing ash. I picked my way more cautiously then. We were getting into an area of recent surface activity; the surface was really treacherous. I caught myself wishing that the Major had okayed McIvers’ scheme for an advanced scout; more dangerous for the individual, maybe, but I was driving blind now and I didn’t like it. One error in judgment could sink us all, but I wasn’t thinking much about the others. I was worried about me , plenty worried. I kept thinking, better McIvers should go than me. It wasn’t healthy thinking and I knew it, but I couldn’t get the thought out of my mind. It was a grueling eight hours and we slept poorly. Back in the Bug again, we moved still more slowly—edging out on a broad flat plateau, dodging a network of gaping surface cracks—winding back and forth in an effort to keep the machines on solid rock. I couldn’t see far ahead, because of the yellow haze rising from the cracks, so I was almost on top of it when I saw a sharp cut ahead where the surface dropped six feet beyond a deep crack. I let out a shout to halt the others; then I edged my Bug forward, peering at the cleft. It was deep and wide. I moved fifty yards to the left, then back to the right. There was only one place that looked like a possible crossing; a long, narrow ledge of gray stuff that lay down across a section of the fault like a ramp. Even as I watched it, I could feel the surface crust under the Bug trembling and saw the ledge shift over a few feet.
|
A. Claney likes Mikuta. He can trust Mikuta.
|
What previous approaches did this method outperform?
|
### Introduction
Recently, a novel way of computing word embeddings has been proposed. Instead of computing one word embedding for each word which sums over all its occurrences, ignoring the appropriate word meaning in various contexts, the contextualized embeddings are computed for each word occurrence, taking into account the whole sentence. Three ways of computing such contextualized embeddings have been proposed: ELMo BIBREF0, BERT BIBREF1 and Flair BIBREF2, along with precomputed models. Peters et al. (2018) BIBREF0 obtain the proposed embeddings, called ELMo, from internal states of deep bidirectional language model, pretrained on a large corpus. Akbik et al. (2018) BIBREF2 introduced Flair, contextualized word embeddings obtained from internal states of a character-level bidirectional language model, thus significantly increasing state of the art of POS tagging, chunking and NER tasks. Last, but not least, Devlin et al. (2018) BIBREF1 employ a Transformer BIBREF3 to compute contextualized embeddings from preceeding and following context at the same time, at the cost of increased processing costs. The new BERT embeddings achieved state-of-the-art results in eleven natural language tasks. Using two of these methods, for which precomputed models for Czech are available, namely BERT and Flair, we present our models for four NLP tasks: part-of-speech (POS) tagging, lemmatization, dependency parsing and named entity recognition (NER). Adding the contextualized embeddings as optional inputs in strong artificial neural network baselines, we report state-of-the-art results in these four tasks. ### Related Work
As for the Prague Dependency Treebank (PDT) BIBREF4, most of the previous works are non-neural systems with one exception of BIBREF5 who hold the state of the art for Czech POS tagging and lemmatization, achieved with the recurrent neural network (RNN) using end-to-end trainable word embeddings and character-level word embeddings. Otherwise, Spoustová et al. (2009) BIBREF6 used an averaged perceptron for POS tagging. For parsing the PDT, Holan and Zabokrtský (2006) BIBREF7 and Novák and Žabokrtský (2007) BIBREF8 used a combination of non-neural parsing techniques . In the multilingual shared task CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies BIBREF9, raw text is processed and the POS tagging, lemmatization and dependency parsing are evaluated on the Universal Dependencies (UD) BIBREF10. Czech is one of the 57 evaluated languages. Interestingly, all 26 participant systems employed the artificial neural networks in some way. Of these, 3 participant systems used (a slightly modified variant of) the only newly presented contextualized embeddings called ELMo BIBREF0, most notably one of the shared task winners BIBREF11. BERT and Flair were not available at the time. For the Czech NER, Straková et al. (2016) BIBREF12 use an artificial neural network with word- and character-level word embeddings to perform NER on the Czech Named Entity Corpus (CNEC) BIBREF13, BIBREF14, BIBREF15. ### Datasets ::: Prague Dependency Treebank 3.5
The Prague Dependency Treebank 3.5 BIBREF4 is a 2018 edition of the core Prague Dependency Treebank. The Prague Dependency Treebank 3.5 contains the same texts as the previous versions since 2.0, and is divided into train, dtest, and etest subparts, where dtest is used as a development set and etest as a test set. The dataset consists of several layers – the morphological m-layer is the largest and contains morphological annotations (POS tags and lemmas), the analytical a-layer contains labeled dependency trees, and the t-layer is the smallest and contains tectogrammatical trees. The statistics of PDT 3.5 sizes is presented in Table TABREF7. A detailed description of the morphological system can be found in BIBREF16, a specification of the syntactic annotations has been presented in BIBREF17. We note that in PDT, lemmas with the same word form are disambiguated using a number suffix – for example, English lemmas for the word forms can (noun) and can (verb) would be annotated as can-1 and can-2. In evaluation, we compute: [noitemsep,topsep=0pt] POS tagging accuracy, lemmatization accuracy, unlabeled attachment score (UAS), labeled attachment score (LAS). ### Datasets ::: Universal Dependencies
The Universal Dependencies project BIBREF10 seeks to develop cross-linguistically consistent treebank annotation of morphology and syntax for many languages. We evaluate the Czech PDT treebank of UD 2.3 BIBREF18, which is an automated conversion of PDT 3.5 a-layer to Universal Dependencies annotation. The original POS tags are used to generate UPOS (universal POS tags), XPOS (language-specific POS tags, in this case the original PDT tags), and Feats (universal morphological features). The UD lemmas are the raw textual lemmas, so the discriminative numeric suffix of PDT is dropped. The dependency trees are converted according to the UD guidelines, adapting both the unlabeled trees and the dependency labels. To compute the evaluation scores, we use the official CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies BIBREF9 evaluation script, which produces the following metrics: [noitemsep,topsep=0pt] UPOS – universal POS tags accuracy, XPOS – language-specific POS tags accuracy, UFeats – universal subset of morphological features accuracy, Lemmas – lemmatization accuracy, UAS – unlabeled attachment score, LAS – labeled attachment score, MLAS – morphology-aware LAS, BLEX – bi-lexical dependency score. ### Datasets ::: Czech Named Entity Corpus
The Czech Named Entity Corpus 1.1 BIBREF13, BIBREF14 is a corpus of $5\,868$ Czech sentences with manually annotated $33\,662$ Czech named entities, classified according to a two-level hierarchy of 62 named entities. The Czech Named Entity Corpus 2.0 BIBREF15 contains $8\,993$ Czech sentences with manually annotated $35\,220$ Czech named entities, classified according to a two-level hierarchy of 46 named entities. We evaluate the NER task with the official CNEC evaluation script. Similarly to previous literature BIBREF13, BIBREF12 etc., the script only evaluates the first round annotation classes for the CNEC 1.1. For the CNEC 2.0, the script evaluates all annotated classes. ### Neural Architectures
All our neural architectures are recurrent neural networks (RNNs). The POS tagging, lemmatization and dependency parsing is performed with the UDPipe 2.0 (Section SECREF16) and NER is performed with our new sequence-to-sequence model (Section SECREF36). ### Neural Architectures ::: POS Tagging, Lemmatization, and Dependency Parsing
We perform POS tagging, lemmatization and dependency parsing using UDPipe 2.0 BIBREF19, one of the three winning systems of the CoNLL 2018 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies BIBREF9 and an overall winner of The 2018 Shared Task on Extrinsic Parser Evaluation BIBREF20. An overview of this architecture is presented in Figure FIGREF17 and the full details of the architecture and the training procedure are available in BIBREF19. ### Neural Architectures ::: POS Tagging, Lemmatization, and Dependency Parsing ::: POS Tagging and Lemmatization
The tagger employs a standard bi-LSTM architecture. After embedding input words, three bidirectional LSTM BIBREF21 layers are performed, followed by a softmax output layers for POS tags and lemmas. While a classification output layer is natural for POS tags, we also apply it to lemmatization and generate lemmas by classifying the input words into lemma generation rules, therefore considering lemmatization as another tagging task. We construct a lemma generation rule from a given form and lemma as follows: [noitemsep,topsep=0pt] We start by finding the longest continuous substring of the form and the lemma. If it is empty, we use the lemma itself as the class. If there is a common substring of the form and the lemma, we compute the shortest edit script converting the prefix of the form into the prefix of the lemma, and the shortest edit script converting the suffix of the form to the suffix of the lemma. The edit scripts permit the operations delete_current_char and insert_char(c). All above operations are performed case insensitively. To indicate correct casing of the lemma, we consider the lemma to be a concatenation of segments, where each segment is composed of either a sequence of lowercase characters, or a sequence of uppercase characters. We represent the lemma casing by encoding the beginning of every such segment, where the offsets in the first half of the lemma are computed relatively to the start of the lemma, and the offsets in the second half of the lemma are computed relatively to the end of the lemma. ### Neural Architectures ::: POS Tagging, Lemmatization, and Dependency Parsing ::: Dependency Parsing
The dependency parsing is again predicted using UDPipe 2.0 architecture. After embedding input words, three bidirectional LSTM BIBREF21 layers are again performed, followed by a biaffine attention layer BIBREF22 producing labeled dependency trees. In our evaluation we do not utilize gold POS tags and lemmas on the test set for dependency parsing. Instead, we consider three ways of employing them during parsing: [noitemsep,topsep=0pt] not using them at all; adding predicted POS tags and lemmas on input; perform joint training of POS tags, lemmatization, and dependency parsing. In this case, we share first two bidirectional LSTM layers between the tagger and the parser. ### Neural Architectures ::: POS Tagging, Lemmatization, and Dependency Parsing ::: Input Embeddings
In our baseline model, we use the end-to-end word embeddings and also character-level word embeddings (bidirectional GRUs, BIBREF23, BIBREF24, BIBREF25 of dimension 256) trained specifically for the task. Our architecture can optionally employ the following additional inputs [noitemsep,topsep=0pt] pretrained word embeddings (WE): For the PDT experiments, we generate the word embeddings with word2vec on a concatenation of large raw Czech corpora available from the LINDAT/CLARIN repository. For UD Czech, we use FastText word embeddings BIBREF27 of dimension 300, which we pretrain on Czech Wikipedia using segmentation and tokenization trained from the UD data. BERT BIBREF1: Pretrained contextual word embeddings of dimension 768 from the Base model. We average the last four layers of the BERT model to produce the embeddings. Because BERT utilizes word pieces, we decompose UD words into appropriate subwords and then average the generated embeddings over subwords belonging to the same word. Flair BIBREF2: Pretrained contextual word embeddings of dimension 4096. ### Neural Architectures ::: POS Tagging, Lemmatization, and Dependency Parsing ::: POS Tags and Lemmas Decoding
Optionally, we employ a morphological dictionary MorfFlex BIBREF28 during decoding. If the morphological dictionary is used, it may produce analyses for an input word as (POS tag, lemma) pairs. If any are generated, we choose the pair with maximum likelihood given by both the POS tag and lemmatization model. ### Neural Architectures ::: Named Entity Recognition
We use a novel approach BIBREF29 for nested named entity recognition (NER) to capture the nested entities in the Czech Named Entity Corpus. The nested entities are encoded in a sequence and the problem of nested NER is then viewed as a sequence-to-sequence (seq2seq) problem, in which the input sequence consists of the input tokens (forms) and the output sequence of the linearized entity labels. The system is a encoder-decoder architecture. The encoder is a bi-directional LSTM and the decoder is a LSTM. The encoded labels are predicted one by one by the decoder, until the decoder outputs the "<eow>" (end of word) label and moves to the next token. We use a hard attention on the word whose label(s) is being predicted. We train the network using the lazy variant of the Adam optimizer BIBREF30, which only updates accumulators for variables that appear in the current batch, with parameters $\beta _1=0.9$ and $\beta _2=0.98$. We use mini-batches of size 8. As a regularization, we apply dropout with rate $0.5$ and the word dropout replaces $20\%$ of words by the unknown token to force the network to rely more on context. We did not perform any complex hyperparameter search. In this model, we use the following word- and character-level word embeddings: [noitemsep,topsep=0pt] pretrained word embeddings: We use the FastText BIBREF27 word embeddings of dimension 300 from the publicly available Czech model. end-to-end word embeddings: We embed the input forms and lemmas (256 dimensions) and POS tags (one-hot). end-to-end character-level word embeddings: We use bidirectional GRUs BIBREF23, BIBREF24 of dimension 128 in line with BIBREF25: we represent every Unicode character with a vector of dimension 128, and concatenate GRU outputs for forward and reversed word characters. Optionally, we add the BERT BIBREF1 and the Flair BIBREF2 contextualized embeddings in the same way as in the UDPipe 2.0 (Section SECREF16). ### Results ::: POS Tagging and Lemmatization on PDT 3.5
The POS tagging and lemmatization results are presented in Table TABREF44. The word2vec word embeddings (WE) considerably increase performance compared to the baseline, especially in POS tagging. When only Flair embeddings are added to the baseline, we also observe an improvement, but not as high. We hypothesise that the lower performance (in contrast with the results reported in BIBREF2) is caused by the size of the training data, because we train the word2vec WE on considerably larger dataset than the Czech Flair model. However, when WE and Flair embeddings are combined, performance moderately increases, demonstrating that the two embedding methods produce at least partially complementary representations. The BERT embeddings alone bring highest improvement in performance. Furthermore, combination with WE or Flair again yields performance increase. The best results are achieved by exploiting all three embedding methods, substantially exceeding state-of-the-art results. Utilization of morphological dictionary improves prediction accuracy. However, as the performance of a model itself increases, the gains obtained by the morphological dictionary diminishes – for a model without any pretrained embeddings, morphological dictionary improves POS tagging by and lemmatization by $0.43\%$ and $0.45\%$, while the best performing model gains only $0.11\%$ and $0.23\%$. ### Results ::: Dependency Parsing on PDT 3.5
The evaluation of the contextualized embeddings methods as well as various ways of POS tag utilization is presented in Table TABREF44. Without POS tags and lemmas, the Flair embeddings bring only a slight improvement in dependency parsing when added to WE. In contrast, BERT embeddings employment results in substantial gains, increasing UAS and LAS by 1.6% and 2.1%. A combination of BERT and Flair embeddings does not result in any performance improvement, demonstrating that BERT syntactic representations encompass the Flair embeddings. When introducing POS tags and lemmas predicted by the best model from Section SECREF43 as inputs for dependency parsing, the performance increases only slightly. A better way of POS tags and lemmas exploitation is achieved in a joint model, which predicts POS tags, lemmas, and dependency trees simultaneously. Again, BERT embeddings bring significant improvements, but in contrast to syntax parsing only, adding Flair embeddings to BERT results in moderate gain – we hypothesise that the increase is due to the complementary morphological information present in Flair embeddings (cf. Section SECREF43). Note that the joint model achieves better parsing accuracy than the one given gold POS tags and lemmas on input. However, the POS tags and lemmas predicted by the joint model are of slightly lower quality compared to a standalone tagger of the best configuration from Section SECREF43. Table TABREF44 compares our best model with state-of-the-art results on PDT 2.0 (note that some of the related work used only a subset of PDT 2.0 and/or utilized gold morphological annotation). To our best knowledge, research on PDT parsing was performed mostly in the first decade of this century, therefore even our baseline model substantially surpasses previous works. Our best model with contextualized embeddings achieves nearly 50% error reduction both in UAS and LAS. ### Results ::: POS Tagging, Lemmatization and Dependency Parsing on Universal Dependencies
Table TABREF47 shows the performance of analyzed embedding methods in a joint model performing POS tagging, lemmatization, and dependency parsing on Czech PDT UD 2.3 treebank. This treebank is derived from PDT 3.5 a-layer, with original POS tags kept in XPOS, and the dependency trees and lemmas modified according to UD guidelines. We observe that the word2vec WEs perform similarly to Flair embeddings in this setting. Our hypothesis is that the word2vec WEs performance loss (compared to WEs in Section SECREF43) is caused by using a considerably smaller raw corpus to pretrain the WEs (Czech Wikipedia with 785M words, compared to 4G words used in Section SECREF43), due to licensing reasons. BERT embeddings once more deliver the highest improvement, especially in dependency parsing, and our best model employs all three embedding methods. In the previous ablation experiments, we used the gold segmentation and tokenization in the Czech PDT UD 2.3 treebank. For comparison with state of the art, Czech PDT UD 2.2 treebank without gold segmentation and tokenization is used in evaluation, according to the CoNLL 2018 shared task training and evaluation protocol. Our system reuses segmentation and tokenization produced by UDPipe 2.0 in the CoNLL 2018 shared task and surpasses previous works substantially in all metrics (bottom part of Table TABREF47). Comparing the results with a joint tagging and parsing PDT 3.5 model from Table TABREF7, we observe that the XPOS results are nearly identical as expected. Lemmatization on the UD treebank is performed without the discriminative numeric suffixes (see Section SECREF3) and therefore reaches better performance. Both UAS and LAS are also better on the UD treebank, which we assume is caused by the different annotation scheme. ### Results ::: Named Entity Recognition
Table TABREF47 shows NER results (F1 score) on CNEC 1.1 and CNEC 2.0. Our sequence-to-sequence (seq2seq) model which captures the nested entities, clearly surpasses the current Czech NER state of the art. Furthermore, significant improvement is gained when adding the contextualized word embeddings (BERT and Flair) as optional input to the LSTM encoder. The strongest model is a combination of the sequence-to-sequence architecture with both BERT and Flair contextual word embeddings. ### Conclusion
We have presented an evaluation of two contextualized embeddings methods, namely BERT and Flair. By utilizing these embeddings as input to deep neural networks, we have achieved state-of-the-art results in several Czech text processing tasks, namely in POS tagging, lemmatization, dependency parsing and named entity recognition. ### Acknowledgements
The work described herein has been supported by OP VVV VI LINDAT/CLARIN project (CZ.02.1.01/0.0/0.0/16_013/0001781) and it has been supported and has been using language resources developed by the LINDAT/CLARIN project (LM2015071) of the Ministry of Education, Youth and Sports of the Czech Republic. Table 1. Size of morphological and analytical annotations of PDT 3.5 train/development/test sets. Fig. 1. UDPipe 2.0 architecture overview. Table 2. POS tagging and lemmatization results (accuracy) on PDT 3.5. Bold indicates the best result, italics related work. †Reported on PDT 2.0, which has the same underlying corpus, with minor changes in morphological annotation (our model results differ at 0.1% on PDT 2.0). Table 3. Dependency tree parsing results on PDT 3.5 a-layer. Bold indicates the best result, italics POS tagging and lemmatization results. For comparison, we report results of a parser trained using gold POS tags and lemmas, and of a tagger trained on a-layer (both also in italics). Table 4. Dependency tree parsing results on PDT 2.0 a-layer. Bold indicates the best result, italics related work. †Possibly using gold POS tags. ‡Results as of 23 Mar 2019. Table 5. Czech PDT UD 2.3 results for POS tagging (UPOS: universal POS, XPOS: languagespecific POS, UFeats: universal morphological features), lemmatization and dependency parsing (UAS, LAS, MLAS, and BLEX scores). Bold indicates the best result, italics related work. Table 6. Named entity recognition results (F1) on the Czech Named Entity Corpus. Bold indicates the best result, italics related work.
|
Table TABREF44, Table TABREF44, Table TABREF47, Table TABREF47
|
What best describes how the overall tone changed from the beginning of the article?
A. From worrisome to frustrated.
B. From apathetic to solemn.
C. From lighthearted to tense.
D. From upbeat to sympathetic.
|
YOUNG READERS Atom Mystery 11 CHAPTER ONE It was only a dream. Eddie Taylor would like to have finished it, but the bar of morning sunlight poking in under the window shade pried his eyes open. The dream fled. Eddie kicked off the sheet, swung his feet to the floor, and groped under the bed for his tennis shoes. He heard his father’s heavy footsteps in the hallway. They stopped outside of his bedroom door. “You awake, Eddie?” “I’m awake, Dad,” Eddie answered. “Breakfast’s ready. Get washed and dressed.” 12 “Be right there,” Eddie said. Then, remembering the dream, he added, “Oh, Dad, is it all right if I use the Geiger counter today?” Mr. Taylor opened the door. He was a big man, broad-shouldered and still thin-waisted. Eddie found it easy to believe the stories he had heard about his father being an outstanding football player in his time. Even his glasses and the gray hair at his temples didn’t add much age, although Eddie knew it had been eighteen years since his father had played his last game of college football. “You may use the Geiger counter any time you want, Eddie,” Mr. Taylor said, “as long as you take good care of it. You figured out where you can find some uranium ore?” Eddie smiled sheepishly. “I—I had a dream,” he said. “Plain as day. It was out on Cedar Point. I was walking along over some rocks. Suddenly the Geiger counter began clicking like everything.” 13 “Cedar Point?” his father asked. “I’ve never been out there. But, from what I hear, there are plenty of rock formations. Might be worth a try, at that. You never can tell where you might strike some radioactivity.” “Do you believe in dreams, Dad?” “Well, now, that’s a tough question, son. I can’t say that I really do. Still, one clue is as good as another when it comes to hunting uranium ore, I guess. But right now we’d better get out to breakfast before your mother scalps us. Hurry it up.” His father turned and went back down the hallway toward the kitchen. Eddie pulled on his trousers and T shirt and went into the bathroom. He washed hurriedly, knowing that even if he missed a spot or two, he was fairly safe. During the summer months his freckles got so thick and dark that it would take a magnifying glass to detect any small smudges of dirt hiding among them. He plastered some water on his dark-red hair, pushed a comb through it, and shrugged as it snapped back almost to its original position. Oh, well, he had tried. 14 He grinned into the mirror, reached a finger into his mouth, and unhooked the small rubber bands from his tooth braces. He dropped them into the waste basket. He’d put fresh ones in after breakfast. He brushed his teeth carefully, taking particular pains around the metal braces. The tooth-straightening orthodontist had warned him about letting food gather around the metal clamps. It could start cavities. Finished, Eddie went out to breakfast. “Good morning, dear,” his mother greeted him, handing him a plate of eggs. “Hi, Mom,” Eddie said. “Gotta hurry. Big day today.” “So your father says. But I’m afraid your big day will have to start with sorting out and tying up those newspapers and magazines that have been collecting in the garage.” “Aw, Mom—” “Eddie, I asked you to do it three days ago. Remember? And the Goodwill truck comes around today.” “But, Mom—” 15 “No arguments, son,” his father put in calmly but firmly. “School vacation doesn’t mean that your chores around here are on vacation, too. Get at it right away, and you’ll still have time to hunt your uranium. “Well,” Mr. Taylor added, excusing himself from the table, “I’d better be getting over to school. I’m expecting to receive shipment of a new radioisotope today.” The very word excited Eddie. In fact, anything having to do with atomic science excited him. He knew something about isotopes—pronounced eye-suh-tope . You couldn’t have a father who was head of the atomic-science department at Oceanview College without picking up a little knowledge along the way. Eddie knew that a radioisotope was a material which had been “cooked” in an atomic reactor until it was “hot” with radioactivity. When carefully controlled, the radiation stored up in such isotopes was used in many beneficial ways. 16 “Why don’t college professors get summer vacations, too?” Eddie asked. One reason for asking that particular question was to keep from prying deeper into the subject of the radioisotope. Much of his father’s work at Oceanview College was of a secret nature. Eddie had learned not to ask questions about it. His father usually volunteered any information he wanted known, so Eddie stuck to questions which could and would be answered. “We get vacations,” his father said. “But—well, my work is a little different, you know. At the speed atomic science is moving today, we simply can’t afford to waste time. But don’t worry. We’ll take a week or so off before school starts in the fall. Maybe head for the mountains with our tent and sleeping bags.” “And Geiger counter?” Eddie asked eagerly. “Wouldn’t think of leaving it home,” his father said, smiling. “By the way, I put new batteries in it the other day. Take it easy on them. Remember to switch it off when you’re not actually using it.” “I will,” Eddie promised. He had forgotten several times before, weakening the batteries. 17 It took Eddie over an hour to sort out the newspapers and magazines in the garage, tie them in neat bundles, and place them out on the front curb for the Goodwill pickup. By that time the sun was high overhead. It had driven off the coolness which the ocean air had provided during the earlier hours. “Anything else, Mom?” he asked, returning to the house and getting the Geiger counter out of the closet. He edged toward the back door before his mother had much time to think of something more for him to do. “I guess not, dear,” Mrs. Taylor said, smiling over his hasty retreat. “What are you going to do?” “Think I’ll do a little prospecting,” Eddie said. “Where?” “Probably in the hills beyond the college,” Eddie said. The more he thought about it, the more he realized it was a little late in the day to go to Cedar Point. The best way to get there was by rowboat across Moon Bay, and that was too long a row to be starting now. Besides, there were plenty of other places around the outskirts of Oceanview where likely looking rock formations invited search with a Geiger counter. 18 “Are you going alone?” his mother asked. “Oh, guess I’ll stop by and see if Teena wants to go,” Eddie answered casually. He tried to make it sound as though he would be doing Teena Ross a big favor. After all, she was only a girl. Eddie didn’t figure a girl would make a very good uranium prospecting partner, but most of the fellows he knew were away at camp, or vacationing with their folks, or something like that. “She’ll enjoy it, I’m sure,” his mother said. “I’ll take Sandy, too,” Eddie said. “He needs the exercise.” “That’s a good idea, dear. Be back in time for an early dinner.” Eddie let Sandy off his chain. The taffy-colored cocker spaniel yipped wildly over his freedom, racing back and forth as Eddie started down the street. 19 Christina Ross—whom everybody called Teena—lived at the far end of the block. Eddie went around to the side door of the light-green stucco house and knocked. “Oh, hi, Eddie,” Teena greeted him, appearing at the screen door. “I was hoping you’d come over.” “Well, I—I just happened to be going by,” Eddie said. “Thought you might want to watch me do a little prospecting with the Geiger counter. But maybe you’re too busy.” That’s how to handle it, Eddie thought. Don’t act anxious. Let Teena be anxious. Then maybe she’ll even offer to bring along a couple of sandwiches or some fruit. “Oh, I’d love to go,” Teena said eagerly, “but I’m just finishing the dishes. Come on in.” “I’m in kind of a hurry.” “I’ll only be a minute.” She pushed the screen door open for him. “I’ll make us some sandwiches.” “Stay here, Sandy,” Eddie said. “Sit.” The dog minded, although he looked a bit rebellious. 20 Eddie went inside and followed Teena to the kitchen. He felt triumphant about the sandwiches. Teena tossed him a dish towel. “You dry them,” she said. “Who, me?” “Why not? You’re in a hurry, aren’t you? I can make the sandwiches while you dry the silverware.” She smiled, putting tiny crinkles in her small, slightly upturned nose. She wore her hair in a pony tail. Even though her hair was blond all year long, it seemed even lighter in the summer. Eddie couldn’t tell whether the sun had faded it, or whether her deep summer tan simply made her hair look lighter by contrast. Maybe both. “Hello, Eddie,” Mrs. Ross said, coming into the kitchen. “Looks like Teena put you to work.” “She always does, Mrs. Ross,” Eddie said, pretending great injury. “Don’t know why I keep coming over here.” “I know,” Teena spoke up quickly. “It’s because we’re friends, that’s why.” 21 Eddie knew she was right. They were friends—good friends. They had been ever since Eddie’s family had moved to Oceanview and his father had become head of the college’s atomic-science department. In fact, their parents were close friends, also. Teena’s father was chief engineer for the Acme Aviation Company, one of the coast town’s largest manufacturing concerns. “Well, I’ll be glad to finish them, Eddie,” Mrs. Ross offered. “I know how boys detest doing dishes.” “Oh, I don’t really mind, Mrs. Ross,” Eddie said. “Besides, Teena’s making sandwiches to take with us.” “Another prospecting trip?” Teena’s mother glanced at the Geiger counter which Eddie had set carefully on the dinette table. “I still think there must be some uranium around here,” Eddie insisted. “And we can find it if anyone can.” “I agree,” Mrs. Ross said. “But even if you don’t find it, you both seem to enjoy your hikes.” 22 “Oh, yes, it’s fun, Mother,” Teena replied, wrapping wax paper around a sandwich. “Guess I’m ready. I’ve got a bone for Sandy, too.” “Don’t go too far out from town,” Mrs. Ross cautioned, as Eddie picked up the Geiger counter. “And stick near the main roads. You know the rules.” “We sure do, Mrs. Ross,” Eddie assured her. “And we’ll be back early.” They walked past the college campus, and toward the rocky foothills beyond. At various rock mounds and outcroppings, Eddie switched on the Geiger counter. The needle of the dial on the black box wavered slightly. A slow clicking came through the earphones, but Eddie knew these indicated no more than a normal background count. There were slight traces of radioactivity in almost all earth or rocks. It was in the air itself, caused by mysterious and ever-present cosmic rays, so there was always a mild background count when the Geiger counter was turned on; but to mean anything, the needle had to jump far ahead on the gauge, and the clicking through the earphones had to speed up until it sounded almost like bacon frying in a hot skillet. 23 There was none of that today. After they had hiked and searched most of the forenoon, Eddie said, “We might as well call it a day, Teena. Doesn’t seem to be anything out here.” “It’s all right with me,” Teena agreed, plucking foxtails from Sandy’s ears. “Pretty hot, anyway. Let’s eat our sandwiches and go back home.” “All right,” Eddie said. “You know, one of these days I’d like to go out to Cedar Point and scout around. Maybe we’ll find something there.” Then he told Teena about his dream. Teena smiled. “A dream sure isn’t much to go on,” she said, “but they say it’s pretty out on Cedar Point. I’ll go any time you want to, Eddie.” She handed him one of the sandwiches. It was midafternoon by the time they arrived back at Teena’s house. They worked a while on a new jigsaw puzzle Teena had received on a recent birthday. Then Eddie said good-by and went on down the street toward his own home. 24 After putting Sandy on his long chain and filling his water dish, Eddie went in the back door. He put the Geiger counter in the closet and went into the kitchen. “What’s for dinner, Mom?” he asked. Mrs. Taylor turned from the sink. Eddie knew at once, just seeing the expression on his mother’s face, that something was wrong. “Dinner?” his mother said absently. “It’s not quite four o’clock yet, Eddie. Besides, dinner may be a little late today.” “But this morning you said it would be early,” Eddie reminded her, puzzled. “This morning I didn’t know what might happen.” 25 Then Eddie heard the sound of his father’s voice coming from the den. There was a strange urgent tone in it. The door to the den was open. Eddie went through the dining room and glanced into the den. His father sat stiffly behind his homemade desk, talking rapidly into the telephone. Eddie caught only the last few sketchy words. Then his father placed the telephone in its cradle, glanced up, and saw Eddie. If there had been even the slightest doubt in Eddie’s mind about something being wrong, it vanished now. Mr. Taylor looked years older than he had that very morning. Worry lay deep in his eyes. He fumbled thoughtfully with a pencil, turning it end over end on his desk. “Hello, son,” he said. He didn’t even ask whether Eddie had discovered any uranium ore that day. Always before, he had shown genuine interest in Eddie’s prospecting trips. “Dad,” Eddie said anxiously, “what—what’s the matter?” “It shows that much, does it, son?” his father said tiredly. “What’s wrong, Dad?” Eddie prompted. “Or can’t you tell me?” Mr. Taylor leaned back. “Quite a bit’s wrong, Eddie,” he said, “and I guess there’s no reason why I shouldn’t tell you. It’ll be in the evening papers, anyway.” 26 “Evening papers?” “Eddie, you remember me mentioning this morning about that radioisotope shipment I was expecting today?” “I remember,” Eddie said. “Did it come?” “It did—and it didn’t,” his father said. “What does that mean, Dad?” Eddie asked, puzzled. “The delivery truck arrived at the school with it,” his father explained, “but while the driver was inquiring where to put it, the container disappeared.” “Disappeared?” “The radioisotope was stolen, Eddie,” his father said slowly. “Stolen right out from under our noses!” 27 CHAPTER TWO At the moment, Eddie didn’t pry for further information on the theft of the valuable radioactive isotope. His father had plenty on his mind, as it was. The main information was in the evening Globe , which Eddie rushed out to get as soon as he heard it plop onto the front porch. He took the newspaper to his father to read first. After having finished, Mr. Taylor handed the paper to Eddie and leaned back thoughtfully in his chair. 28 “They’ve got it pretty straight, at that,” Mr. Taylor said, “but I’m afraid this is going to stir up quite a bit of trouble.” “It wasn’t your fault, was it, Dad?” Eddie defended. “It was as much mine as anybody’s, son,” his father said. “Probably more so. After all, I am head of the department. I knew about the shipment. That should make it my responsibility to see that it was properly received and placed in our atomic-materials storage vault. But there is little point in trying to place the blame on anyone. I’m willing to accept that part of it. The important thing is that we recover that radioisotope. Not only is it of a secret nature, but it is also dangerously radioactive if improperly handled.” “But—but wasn’t it in a safe container?” Eddie asked. 29 “Of course,” his father said. “There were only two ounces of it in a fifty-pound lead capsule. As long as it remains in that capsule it’s safe. As you know, the lead prevents any radiation from escaping. Out of that capsule, however, those two ounces of radioisotope can be very dangerous.” “Fifty pounds,” Eddie said thoughtfully. “That’s a pretty big thing to steal, isn’t it?” “Not when it’s lead, son,” his father replied. “Not much bigger than a two-quart milk bottle, in fact.” “Even at that, no kid could have taken it,” Eddie said. “Kid?” His father smiled thinly. “We don’t think it was any kid, Eddie. Not by a long shot. The whole thing was carefully planned and carefully carried out. It was not the work of amateurs.” Eddie read the newspaper account. The small truck from Drake Ridge, where one of the country’s newest atomic reactors was located, had arrived earlier than expected at Oceanview College. It had backed up to the receiving dock where all of the college supplies were delivered. Since deliveries during vacation months were few, there was no one on the dock when the truck arrived. A half hour later, when the delivery was expected, there would have been. The truck’s early arrival had caught them unprepared. 30 The driver had left the truck and had gone around the building to the front office. It had taken him less than five minutes to locate the receiving-dock foreman. Together, they had returned through the small warehouse and opened the rear door onto the dock. During that short time someone had pried open the heavy padlock on the delivery truck’s rear door and had stolen the fifty-pound lead capsule containing the radioisotope. Dusty footprints on the pavement around the rear of the truck indicated that two men had carried out the theft. A heavy iron pry bar had been dropped at the rear of the truck after the lock was sprung. It was a common type used by carpenters. There were no fingerprints or other identifying marks on it. The footprints were barely visible and of no help other than to indicate that two men were involved in the crime. 31 “Dad,” Eddie asked, looking up from the paper, “how could anyone carry away something weighing fifty pounds without being noticed?” “Chances are they had their car parked nearby,” his father said. “As you know, there are no fences or gates around Oceanview College. People come and go as they please. As a matter of fact, there are always quite a few automobiles parked around the shipping and receiving building, and parking space is scarce even during summer sessions. Anyone could park and wait there unnoticed. Or they could walk around without attracting any undue attention.” “But, Dad,” Eddie continued, “how would the men know that the delivery truck would arrive a half hour early?” “They wouldn’t,” his father said. “They may have had another plan. The way things worked out, they didn’t need to use it. The early delivery and the business of leaving the truck unguarded for a few minutes probably gave them a better opportunity than they had expected. At least, they took quick advantage of it.” 32 “I don’t see what anyone would want with a radioisotope,” Eddie said. “Maybe they figured there was something else inside of that lead capsule.” “That’s unlikely, son,” Mr. Taylor said. “Believe me, it was no common theft. Nor were the thieves ordinary thieves. That isotope was a new one. A very secret one. Our job at the college was to conduct various tests with it in order to find out exactly how it could best be put to use as a cure for disease, or for sterilizing food, or even as a source of power.” “Power?” Eddie said. “Boy, it must have been a strong isotope.” He knew that the strength of radioisotopes could be controlled largely by the length of time they were allowed to “cook” in an atomic reactor and soak up radioactivity. 33 “We weren’t planning to run a submarine with it,” his father said. “It wasn’t that strong. Still, it doesn’t take so very much radioactivity to make two ounces of an isotope quite powerful—and quite deadly. I only hope whoever stole it knows what he’s doing. However, I’m sure he does.” “You mean he must have been an atomic scientist himself?” Eddie asked. “Let’s just say he—or both of them—have enough training in the subject to know how to handle that isotope safely,” Mr. Taylor said. “But, Dad,” Eddie wondered, “what could they do with it?” “They could study it,” his father explained. “At least, they could send it somewhere to be broken down and studied. Being a new isotope, the formula is of great value.” “What do you mean, send it somewhere?” Eddie asked. “Perhaps to some other country.” “Then—then you mean whoever stole it were spies!” Eddie exclaimed breathlessly. “That’s entirely possible,” his father said. “In fact, it’s the only logical explanation I can think of. People simply don’t go around stealing radioactive isotopes without a mighty important reason.” 34 “Dinner’s ready,” Eddie’s mother called from the kitchen. During dinner Eddie wasn’t sure just what he was eating. The idea of spies stealing atomic materials kept building up in his mind. By the time dessert was finished, he was anxious to talk with someone, yet he knew he shouldn’t bother his father with any more questions. He asked if he could go over and visit with Teena for a while. “Well, you were together most of the day,” his mother said, “but I guess it’s all right. Be back in about an hour, though.” It was a balmy evening. On such evenings, he and Teena sometimes walked along the beach barefoot, collecting sea shells. Today Eddie had no desire to do that. He ran down the block. Teena answered his knock. “Come on in, Eddie,” she invited, seeming surprised to see him. “Mother and I are just finishing dinner.” “Oh, I figured you’d be through by now,” Eddie apologized, following her inside. 35 “Hello, Eddie,” Mrs. Ross said, but she didn’t seem as cheerful as usual. “Good evening, Mrs. Ross,” Eddie said. “I—I hope I’m not making a pest of myself.” He looked around for Mr. Ross, but Teena’s father apparently hadn’t arrived home from Acme Aircraft yet. There wasn’t a place set for him at the table, either. “You’re never a pest, Eddie,” Mrs. Ross assured him. “I was going to call your mother in a little while about that newspaper write-up.” “Oh, you read it?” Eddie said. “How could anyone miss it?” Teena said. “Right on the front page.” “I suppose your father is quite concerned over it,” Teena’s mother said. “Oh, yes,” Eddie affirmed. “He was the one who ordered the isotope.” “What’s an isotope?” Teena asked. “I’m not sure I know, either,” Mrs. Ross said. “Maybe we could understand more of what it’s all about if you could explain what a radioisotope is, Eddie.” 36 “Well,” Eddie said slowly, “it’s not easy to explain, but I’ll try. You know how rare uranium is. There’s not nearly enough of it to fill all the needs for radioactive materials. Besides, pure uranium is so powerful and expensive and dangerous to handle that it’s not a very good idea to try using it in its true form. So they build an atomic reactor like the one at Drake Ridge.” “We’ve driven by it,” Mrs. Ross said. “My, it’s a big place.” “I’ll say,” Eddie agreed. “Of course, only one building holds the reactor itself. It’s the biggest building near the center.” “I remember it,” Teena said. “Well, the reactor is about four stories high,” Eddie went on. “They call it a uranium ‘pile.’ It’s made up of hundreds and hundreds of graphite bricks. That’s where they get the name ‘pile’—from brick pile. Anyway, scattered around in between the bricks are small bits of uranium. Uranium atoms are radioactive. That is, they keep splitting up and sending out rays.” “Why do they do that?” Teena asked. 37 “It’s just the way nature made uranium, I guess,” Eddie said. “Most atoms stay in one piece, although they move around lickety-split all of the time. Uranium atoms not only move around, but they break apart. They shoot out little particles called neutrons. These neutrons hit other atoms and split them apart, sending out more neutrons. It’s a regular chain reaction.” “I’ve heard of chain reactions,” Mrs. Ross said. “Well, with all of the splitting up and moving around of the uranium atoms,” Eddie went on, “an awful lot of heat builds up. If they don’t control it—well, you’ve seen pictures of atomic-bomb explosions. That’s a chain reaction out of control.” “Out of control is right,” Teena said. 38 “But the atomic piles control the reaction,” Eddie said. “The graphite bricks keep the splitting-up atoms apart so one neutron won’t go smashing into other atoms unless they want it to. They have ways of controlling it so that only as much radiation builds up as they want. You can even hear the reactor hum as the radioactive rays go tearing through it. But by careful tending, the scientists keep the atomic collisions far enough apart so the thing doesn’t blow up.” “Boy, that sounds dangerous,” Teena said. “Well, they know just how to do it,” Eddie replied. “Aren’t the rays dangerous?” Mrs. Ross asked. “I’ll say they’re dangerous,” Eddie said. “But the whole pile is covered by a shield of concrete about eight feet thick. That keeps the rays from getting out and injuring the workmen.” “Goodness. Eight feet is a lot of cement.” “It takes a lot to stop radioactive atomic particles,” Eddie explained. “Especially the gamma rays. They’re the fastest and most dangerous, and the hardest to stop. Alpha and beta rays are fairly easy to stop. But the gamma rays are regular high-velocity invisible bullets. They’ll go right through a stone wall unless it’s plenty thick. Of course, you can’t see them. Not with even the most powerful microscope in the world.” 39 “I wouldn’t want to work around a place where I might get shot at by—by dangerous rays you can’t even see,” Teena said. “I would,” Eddie said. “Everyone is carefully protected. They see to that. Well, anyway, if all of those uranium atoms were shooting radioactive rays around inside of that pile and doing nothing, there would be an awful lot of energy going to waste. So the atomic scientists take certain elements which aren’t radioactive, but can be made radioactive, and shove small pieces of them into holes drilled in the pile.” “Isn’t that dangerous?” Teena asked. “They don’t shove them in with their bare hands,” Eddie said, trying not to show exasperation. “They use long holders to push the small chunks of material into the holes in the reactor. Then, as those uranium atoms keep splitting up and shooting particles around inside of the pile, some of them smack into the chunks of material, and stick there. Most elements will soak up radiation, just like a sponge soaks up water.” 40 “My, that’s interesting, Eddie,” Mrs. Ross said. “I’ve seen them do it,” Eddie said proudly, then added, “from behind a protective shield, of course. When the material has soaked up enough radiation, they pull it back out. They say it’s ‘cooked.’” “You mean it’s hot?” Teena asked. “It’s hot,” Eddie said, “but not like if it came out of a stove. By hot, they mean it’s radioactive. If you touched it, or even got near it, you would get burned, but you probably wouldn’t even know it for a while. It would be a radiation burn. That’s a kind of burn you don’t feel, but it destroys your blood cells and tissues, and—well, you’ve had it.” “So that’s what a radioisotope is,” Mrs. Ross said. “It’s like a sponge. Only instead of soaking up water, it soaks up radiation.” 41 “That’s about it,” Eddie said. “My dad says that as more is learned about the ways to use isotopes, the whole world is going to be improved. You’ve heard of radiocobalt for curing cancer. Well, that’s an isotope. They make it by cooking cobalt in an atomic reactor. Oh, there are hundreds of different isotopes. Like I said, isotopes can be made of most of the elements. And there are over a hundred elements. Some soak up a lot of radioactivity, and are strong and dangerous. Others absorb only a little and are pretty safe to use. Depends, too, on how long they let them cook in the reactor.” “What kind was the one stolen from the college today?” Teena asked. “Dad didn’t say exactly,” Eddie answered, “except he did say that if whoever took it didn’t know what he was doing and opened up the lead capsule, it could kill him. Of course, even the mild isotopes are deadly if they’re not handled right.” “My goodness, it is a serious matter, isn’t it?” Mrs. Ross said. 42 Eddie nodded. It was even more serious than its threat of danger to anyone who handled it carelessly. It was a new isotope—a secret isotope. His father hadn’t said whether it had been developed for curing things or for destroying things. But many radioisotopes could do either; it depended on how they were used. Eddie assumed that anyone who would stoop to stealing isotopes more than likely would be interested in their ability to destroy rather than their ability to benefit mankind. “Well, I certainly do hope everything works out all right,” Teena’s mother said. “So do I,” Teena agreed. Eddie glanced at the kitchen clock. “Oh, boy,” he said, “I’d better be heading back home. I didn’t mean to come over here and talk so long.” “Oh, we’re glad you did, Eddie,” Mrs. Ross said. “I’m afraid too few of us know anything about this atom business.” 43 “That’s right, Mrs. Ross,” Eddie agreed. “People should talk more and read more about it. After all, this is an atomic age. We might as well face it. My father says that in horse-and-buggy days everyone knew how to feed a horse and grease a wagon wheel. They knew what was needed to get the work done. But now that atoms are being harnessed to do the work, not many people even bother to find out what an atom is.” Mrs. Ross smiled. “I guess you’re right, Eddie,” she said, “but I wouldn’t quite know how to go about feeding an atom.” “Or greasing one,” Teena added. Eddie laughed. “I sure wouldn’t want the job of trying to feed a herd of them the size of a period,” he said. “Did you know that there are about three million billion atoms of carbon in a single period printed at the end of a sentence. That’s how small atoms are.” “Three million billion is a lot of something,” a man’s voice spoke behind him. “What are we talking about, Eddie?” “Oh, hello, Mr. Ross,” Eddie said, turning around and standing up. “I didn’t hear you come in.” 44 Teena’s father was a medium-sized man with light-brown hair which was getting somewhat thin on top. He was usually quite cheerful and full of fun, but tonight his face seemed unusually drawn and sober. He stepped to the table, leaned over, and gave both Teena and Mrs. Ross a kiss on the cheek. “Eddie was telling us about atoms,” Teena’s mother said. “Did you know there were three million billion of them in a period?” “How many in a comma?” Mr. Ross said to Eddie, then added quickly, “forget it, Eddie. It wasn’t very funny. I—I’m afraid I don’t feel very funny tonight.” “Sit down, dear,” Mrs. Ross said. “I’ll warm your dinner. You didn’t sound very cheerful when you called to say you would be late. How did everything go at the plant today?” “Not so good,” Teena’s father said tiredly. “In fact, not good at all.” Problems. It seemed that everyone had problems, Eddie thought, as he started to leave.
|
C. From lighthearted to tense.
|
Which of the follow characteristics does NOT give Moran the impression that the planet that the Nadine is approaching may be habitable?
A. shape of the ice cap
B. composition of the ice cap
C. location of the ice cap
D. size of the ice cap
|
PLANET of DREAD By MURRAY LEINSTER Illustrator ADKINS [Transcriber's Note: This etext was produced from Fantastic Stories of Imagination May 1962. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] I. Moran cut apart the yard-long monstrosity with a slash of flame. The thing presumably died, but it continued to writhe senselessly. He turned to see other horrors crawling toward him. Then he knew he was being marooned on a planet of endless terrors. Moran, naturally, did not mean to help in the carrying out of the plans which would mean his destruction one way or another. The plans were thrashed out very painstakingly, in formal conference on the space-yacht Nadine , with Moran present and allowed to take part in the discussion. From the viewpoint of the Nadine's ship's company, it was simply necessary to get rid of Moran. In their predicament he might have come to the same conclusion; but he was not at all enthusiastic about their decision. He would die of it. The Nadine was out of overdrive and all the uncountable suns of the galaxy shone steadily, remotely, as infinitesimal specks of light of every color of the rainbow. Two hours since, the sun of this solar system had been a vast glaring disk off to port, with streamers and prominences erupting about its edges. Now it lay astern, and Moran could see the planet that had been chosen for his marooning. It was a cloudy world. There were some dim markings near one lighted limb, but nowhere else. There was an ice-cap in view. The rest was—clouds. The ice-cap, by its existence and circular shape, proved that the planet rotated at a not unreasonable rate. The fact that it was water-ice told much. A water-ice ice-cap said that there were no poisonous gases in the planet's atmosphere. Sulfur dioxide or chlorine, for example, would not allow the formation of water-ice. It would have to be sulphuric-acid or hydrochloric-acid ice. But the ice-cap was simple snow. Its size, too, told about temperature-distribution on the planet. A large cap would have meant a large area with arctic and sub-arctic temperatures, with small temperate and tropical climate-belts. A small one like this meant wide tropical and sub-tropical zones. The fact was verified by the thick, dense cloud-masses which covered most of the surface,—all the surface, in fact, outside the ice-cap. But since there were ice-caps there would be temperate regions. In short, the ice-cap proved that a man could endure the air and temperature conditions he would find. Moran observed these things from the control-room of the Nadine , then approaching the world on planetary drive. He was to be left here, with no reason ever to expect rescue. Two of the Nadine's four-man crew watched out the same ports as the planet seemed to approach. Burleigh said encouragingly; "It doesn't look too bad, Moran!" Moran disagreed, but he did not answer. He cocked an ear instead. He heard something. It was a thin, wabbling, keening whine. No natural radiation sounds like that. Moran nodded toward the all-band speaker. "Do you hear what I do?" he asked sardonically. Burleigh listened. A distinctly artificial signal came out of the speaker. It wasn't a voice-signal. It wasn't an identification beacon, such as are placed on certain worlds for the convenience of interstellar skippers who need to check their courses on extremely long runs. This was something else. Burleigh said: "Hm ... Call the others, Harper." Harper, prudently with him in the control-room, put his head into the passage leading away. He called. But Moran observed with grudging respect that he didn't give him a chance to do anything drastic. These people on the Nadine were capable. They'd managed to recapture the Nadine from him, but they were matter-of-fact about it. They didn't seem to resent what he'd tried to do, or that he'd brought them an indefinite distance in an indefinite direction from their last landing-point, and they had still to re-locate themselves. They'd been on Coryus Three and they'd gotten departure clearance from its space-port. With clearance-papers in order, they could land unquestioned at any other space-port and take off again—provided the other space-port was one they had clearance for. Without rigid control of space-travel, any criminal anywhere could escape the consequences of any crime simply by buying a ticket to another world. Moran couldn't have bought a ticket, but he'd tried to get off the planet Coryus on the Nadine . The trouble was that the Nadine had clearance papers covering five persons aboard—four men and a girl Carol. Moran made six. Wherever the yacht landed, such a disparity between its documents and its crew would spark an investigation. A lengthy, incredibly minute investigation. Moran, at least, would be picked out as a fugitive from Coryus Three. The others were fugitives too, from some unnamed world Moran did not know. They might be sent back where they came from. In effect, with six people on board instead of five, the Nadine could not land anywhere for supplies. With five on board, as her papers declared, she could. And Moran was the extra man whose presence would rouse space-port officials' suspicion of the rest. So he had to be dumped. He couldn't blame them. He'd made another difficulty, too. Blaster in hand, he'd made the Nadine take off from Coryus III with a trip-tape picked at random for guidance. But the trip-tape had been computed for another starting-point, and when the yacht came out of overdrive it was because the drive had been dismantled in the engine-room. So the ship's location was in doubt. It could have travelled at almost any speed in practically any direction for a length of time that was at least indefinite. A liner could re-locate itself without trouble. It had elaborate observational equipment and tri-di star-charts. But smaller craft had to depend on the Galactic Directory. The process would be to find a planet and check its climate and relationship to other planets, and its flora and fauna against descriptions in the Directory. That was the way to find out where one was, when one's position became doubtful. The Nadine needed to make a planet-fall for this. The rest of the ship's company came into the control-room. Burleigh waved his hand at the speaker. "Listen!" They heard it. All of them. It was a trilling, whining sound among the innumerable random noises to be heard in supposedly empty space. "That's a marker," Carol announced. "I saw a costume-story tape once that had that sound in it. It marked a first-landing spot on some planet or other, so the people could find that spot again. It was supposed to be a long time ago, though." "It's weak," observed Burleigh. "We'll try answering it." Moran stirred, and he knew that every one of the others was conscious of the movement. But they didn't watch him suspiciously. They were alert by long habit. Burleigh said they'd been Underground people, fighting the government of their native world, and they'd gotten away to make it seem the revolt had collapsed. They'd go back later when they weren't expected, and start it up again. Moran considered the story probable. Only people accustomed to desperate actions would have remained so calm when Moran had used desperate measures against them. Burleigh picked up the transmitter-microphone. "Calling ground," he said briskly. "Calling ground! We pick up your signal. Please reply." He repeated the call, over and over and over. There was no answer. Cracklings and hissings came out of the speaker as before, and the thin and reedy wabbling whine continued. The Nadine went on toward the enlarging cloudy mass ahead. Burleigh said; "Well?" "I think," said Carol, "that we should land. People have been here. If they left a beacon, they may have left an identification of the planet. Then we'd know where we are and how to get to Loris." Burleigh nodded. The Nadine had cleared for Loris. That was where it should make its next landing. The little yacht went on. All five of its proper company watched as the planet's surface enlarged. The ice-cap went out of sight around the bulge of the globe, but no markings appeared. There were cloud-banks everywhere, probably low down in the atmosphere. The darker vague areas previously seen might have been highlands. "I think," said Carol, to Moran, "that if it's too tropical where this signal's coming from, we'll take you somewhere near enough to the ice-cap to have an endurable climate. I've been figuring on food, too. That will depend on where we are from Loris because we have to keep enough for ourselves. But we can spare some. We'll give you the emergency-kit, anyhow." The emergency-kit contained antiseptics, seeds, and a weapon or two, with elaborate advice to castaways. If somebody were wrecked on an even possibly habitable planet, the especially developed seed-strains would provide food in a minimum of time. It was not an encouraging thought, though, and Moran grimaced. She hadn't said anything about being sorry that he had to be marooned. Maybe she was, but rebels learn to be practical or they don't live long. Moran wondered, momentarily, what sort of world they came from and why they had revolted, and what sort of set-back to the revolt had sent the five off in what they considered a strategic retreat but their government would think defeat. Moran's own situation was perfectly clear. He'd killed a man on Coryus III. His victim would not be mourned by anybody, and somebody formerly in very great danger would now be safe, which was the reason for what Moran had done. But the dead man had been very important, and the fact that Moran had forced him to fight and killed him in fair combat made no difference. Moran had needed to get off-planet, and fast. But space-travel regulations are especially designed to prevent such escapes. He'd made a pretty good try, at that. One of the controls on space-traffic required a ship on landing to deposit its fuel-block in the space-port's vaults. The fuel-block was not returned until clearance for departure had been granted. But Moran had waylaid the messenger carrying the Nadine's fuel-block back to that space-yacht. He'd knocked the messenger cold and presented himself at the yacht with the fuel. He was admitted. He put the block in the engine's gate. He duly took the plastic receipt-token the engine only then released, and he drew a blaster. He'd locked two of the Nadine's crew in the engine-room, rushed to the control-room without encountering the others, dogged the door shut, and threaded in the first trip-tape to come to hand. He punched the take-off button and only seconds later the overdrive. Then the yacht—and Moran—was away. But his present companions got the drive dismantled two days later and once the yacht was out of overdrive they efficiently gave him his choice of surrendering or else. He surrendered, stipulating that he wouldn't be landed back on Coryus; he still clung to hope of avoiding return—which was almost certain anyhow. Because nobody would want to go back to a planet from which they'd carried away a criminal, even though they'd done it unwillingly. Investigation of such a matter might last for months. Now the space-yacht moved toward a vast mass of fleecy whiteness without any visible features. Harper stayed with the direction-finder. From time to time he gave readings requiring minute changes of course. The wabbling, whining signal was louder now. It became louder than all the rest of the space-noises together. The yacht touched atmosphere and Burleigh said; "Watch our height, Carol." She stood by the echometer. Sixty miles. Fifty. Thirty. A correction of course. Fifteen miles to surface below. Ten. Five. At twenty-five thousand feet there were clouds, which would be particles of ice so small that they floated even so high. Then clear air, then lower clouds, and lower ones still. It was not until six thousand feet above the surface that the planet-wide cloud-level seemed to begin. From there on down it was pure opacity. Anything could exist in that dense, almost palpable grayness. There could be jagged peaks. The Nadine went down and down. At fifteen hundred feet above the unseen surface, the clouds ended. Below, there was only haze. One could see the ground, at least, but there was no horizon. There was only an end to visibility. The yacht descended as if in the center of a sphere in which one could see clearly nearby, less clearly at a little distance, and not at all beyond a quarter-mile or so. There was a shaded, shadowless twilight under the cloud-bank. The ground looked like no ground ever seen before by anyone. Off to the right a rivulet ran between improbable-seeming banks. There were a few very small hills of most unlikely appearance. It was the ground, the matter on which one would walk, which was strangest. It had color, but the color was not green. Much of it was a pallid, dirty-yellowish white. But there were patches of blue, and curious veinings of black, and here and there were other colors, all of them unlike the normal color of vegetation on a planet with a sol-type sun. Harper spoke from the direction-finder; "The signal's coming from that mound, yonder." There was a hillock of elongated shape directly in line with the Nadine's course in descent. Except for the patches of color, it was the only considerable landmark within the half-mile circle in which anything could be seen at all. The Nadine checked her downward motion. Interplanetary drive is rugged and sure, but it does not respond to fine adjustment. Burleigh used rockets, issuing great bellowings of flame, to make actual contact. The yacht hovered, and as the rocket-flames diminished slowly she sat down with practically no impact at all. But around her there was a monstrous tumult of smoke and steam. When the rockets went off, she lay in a burned-out hollow some three or four feet deep with a bottom of solid stone. The walls of the hollow were black and scorched. It seemed that at some places they quivered persistently. There was silence in the control-room save for the whining noise which now was almost deafening. Harper snapped off the switch. Then there was true silence. The space-yacht had come to rest possibly a hundred yards from the mound which was the source of the space-signal. That mound shared the peculiarity of the ground as far as they could see through the haze. It was not vegetation in any ordinary sense. Certainly it was no mineral surface! The landing-pockets had burned away three or four feet of it, and the edge of the burned area smoked noisesomely, and somehow it looked as if it would reek. And there were places where it stirred. Burleigh blinked and stared. Then he reached up and flicked on the outside microphones. Instantly there was bedlam. If the landscape was strange, here, the sounds that came from it were unbelievable. There were grunting noises. There were clickings, uncountable clickings that made a background for all the rest. There were discordant howls and honkings. From time to time some thing unknown made a cry that sounded very much like a small boy trailing a stick against a picket fence, only much louder. Something hooted, maintaining the noise for an impossibly long time. And persistently, sounding as if they came from far away, there were booming noises, unspeakably deep-bass, made by something alive. And something shrieked in lunatic fashion and something else still moaned from time to time with the volume of a steam-whistle.... "This sounds and looks like a nice place to live," said Moran with fine irony. Burleigh did not answer. He turned down the outside sound. "What's that stuff there, the ground?" he demanded. "We burned it away in landing. I've seen something like it somewhere, but never taking the place of grass!" "That," said Moran as if brightly, "that's what I'm to make a garden in. Of evenings I'll stroll among my thrifty plantings and listen to the delightful sounds of nature." Burleigh scowled. Harper flicked off the direction-finder. "The signal still comes from that hillock yonder," he said with finality. Moran said bitingly; "That ain't no hillock, that's my home!" Then, instantly he'd said it, he recognized that it could be true. The mound was not a fold in the ground. It was not an up-cropping of the ash-covered stone on which the Nadine rested. The enigmatic, dirty-yellow-dirty-red-dirty-blue-and-dirty-black ground-cover hid something. It blurred the shape it covered, very much as enormous cobwebs made solid and opaque would have done. But when one looked carefully at the mound, there was a landing-fin sticking up toward the leaden skies. It was attached to a large cylindrical object of which the fore part was crushed in. The other landing-fins could be traced. "It's a ship," said Moran curtly. "It crash-landed and its crew set up a signal to call for help. None came, or they'd have turned the beacon off. Maybe they got the lifeboats to work and got away. Maybe they lived as I'm expected to live until they died as I'm expected to die." Burleigh said angrily; "You'd do what we are doing if you were in our shoes!" "Sure," said Moran, "but a man can gripe, can't he?" "You won't have to live here," said Burleigh. "We'll take you somewhere up by the ice-cap. As Carol said, we'll give you everything we can spare. And meanwhile we'll take a look at that wreck yonder. There might be an indication in it of what solar system this is. There could be something in it of use to you, too. You'd better come along when we explore." "Aye, aye, sir," said Moran with irony. "Very kind of you, sir. You'll go armed, sir?" Burleigh growled; "Naturally!" "Then since I can't be trusted with a weapon," said Moran, "I suggest that I take a torch. We may have to burn through that loathesome stuff to get in the ship." "Right," growled Burleigh again. "Brawn and Carol, you'll keep ship. The rest of us wear suits. We don't know what that stuff is outside." Moran silently went to the space-suit rack and began to get into a suit. Modern space-suits weren't like the ancient crudities with bulging metal casings and enormous globular helmets. Non-stretch fabrics took the place of metal, and constant-volume joints were really practical nowadays. A man could move about in a late-model space-suit almost as easily as in ship-clothing. The others of the landing-party donned their special garments with the brisk absence of fumbling that these people displayed in every action. "If there's a lifeboat left," said Carol suddenly, "Moran might be able to do something with it." "Ah, yes!" said Moran. "It's very likely that the ship hit hard enough to kill everybody aboard, but not smash the boats!" "Somebody survived the crash," said Burleigh, "because they set up a beacon. I wouldn't count on a boat, Moran." "I don't!" snapped Moran. He flipped the fastener of his suit. He felt all the openings catch. He saw the others complete their equipment. They took arms. So far they had seen no moving thing outside, but arms were simple sanity on an unknown world. Moran, though, would not be permitted a weapon. He picked up a torch. They filed into the airlock. The inner door closed. The outer door opened. It was not necessary to check the air specifically. The suits would take care of that. Anyhow the ice-cap said there were no water-soluble gases in the atmosphere, and a gas can't be an active poison if it can't dissolve. They filed out of the airlock. They stood on ash-covered stone, only slightly eroded by the processes which made life possible on this planet. They looked dubiously at the scorched, indefinite substance which had been ground before the Nadine landed. Moran moved scornfully forward. He kicked at the burnt stuff. His foot went through the char. The hole exposed a cheesy mass of soft matter which seemed riddled with small holes. Something black came squirming frantically out of one of the openings. It was eight or ten inches long. It had a head, a thorax, and an abdomen. It had wing-cases. It had six legs. It toppled down to the stone on which the Nadine rested. Agitatedly, it spread its wing-covers and flew away, droning loudly. The four men heard the sound above even the monstrous cacophony of cries and boomings and grunts and squeaks which seemed to fill the air. "What the devil—." Moran kicked again. More holes. More openings. More small tunnels in the cheese-like, curd-like stuff. More black things squirming to view in obvious panic. They popped out everywhere. It was suddenly apparent that the top of the soil, here, was a thick and blanket-like sheet over the whitish stuff. The black creatures lived and thrived in tunnels under it. Carol's voice came over the helmet-phones. " They're—bugs! " she said incredulously. " They're beetles! They're twenty times the size of the beetles we humans have been carrying around the galaxy, but that's what they are! " Moran grunted. Distastefully, he saw his predicament made worse. He knew what had happened here. He could begin to guess at other things to be discovered. It had not been practical for men to move onto new planets and subsist upon the flora and fauna they found there. On some new planets life had never gotten started. On such worlds a highly complex operation was necessary before humanity could move in. A complete ecological complex had to be built up; microbes to break down the rock for soil, bacteria to fix nitrogen to make the soil fertile; plants to grow in the new-made dirt and insects to fertilize the plants so they would multiply, and animals and birds to carry the seeds planet-wide. On most planets, to be sure, there were local, aboriginal plants and animals. But still terrestrial creatures had to be introduced if a colony was to feed itself. Alien plants did not supply satisfactory food. So an elaborate adaptation job had to be done on every planet before native and terrestrial living things settled down together. It wasn't impossible that the scuttling things were truly beetles, grown large and monstrous under the conditions of a new planet. And the ground.... "This ground stuff," said Moran distastefully, "is yeast or some sort of toadstool growth. This is a seedling world. It didn't have any life on it, so somebody dumped germs and spores and bugs to make it ready for plants and animals eventually. But nobody's come back to finish up the job." Burleigh grunted a somehow surprised assent. But it wasn't surprising; not wholly so. Once one mentioned yeasts and toadstools and fungi generally, the weird landscape became less than incredible. But it remained actively unpleasant to think of being marooned on it. "Suppose we go look at the ship?" said Moran unpleasantly. "Maybe you can find out where you are, and I can find out what's ahead of me." He climbed up on the unscorched surface. It was elastic. The parchment-like top skin yielded. It was like walking on a mass of springs. "We'd better spread out," added Moran, "or else we'll break through that skin and be floundering in this mess." "I'm giving the orders, Moran!" said Burleigh shortly. "But what you say does make sense." He and the others joined Moran on the yielding surface. Their footing was uncertain, as on a trampoline. They staggered. They moved toward the hillock which was a covered-over wrecked ship. The ground was not as level as it appeared from the Nadine's control-room. There were undulations. But they could not see more than a quarter-mile in any direction. Beyond that was mist. But Burleigh, at one end of the uneven line of advancing men, suddenly halted and stood staring down at something he had not seen before. The others halted. Something moved. It came out from behind a very minor spire of whitish stuff that looked like a dirty sheet stretched over a tall stone. The thing that appeared was very peculiar indeed. It was a—worm. But it was a foot thick and ten feet long, and it had a group of stumpy legs at its fore end—where there were eyes hidden behind bristling hair-like growths—and another set of feet at its tail end. It progressed sedately by reaching forward with its fore-part, securing a foothold, and then arching its middle portion like a cat arching its back, to bring its hind part forward. Then it reached forward again. It was of a dark olive color from one end to the other. Its manner of walking was insane but somehow sedate. Moran heard muffled noises in his helmet-phone as the others tried to speak. Carol's voice came anxiously; " What's the matter? What do you see? " Moran said with savage precision; "We're looking at an inch-worm, grown up like the beetles only more so. It's not an inch-worm any longer. It's a yard-worm." Then he said harshly to the men with him; "It's not a hunting creature on worlds where it's smaller. It's not likely to have turned deadly here. Come on!" He went forward over the singularly bouncy ground. The others followed. It was to be noted that Hallet the engineer, avoided the huge harmless creature more widely than most. They reached the mound which was the ship. Moran unlimbered his torch. He said sardonically; "This ship won't do anybody any good. It's old-style. That thick belt around its middle was dropped a hundred years ago, and more." There was an abrupt thickening of the cylindrical hull at the middle. There was an equally abrupt thinning, again, toward the landing-fins. The sharpness of the change was blurred over by the revolting ground-stuff growing everywhere. "We're going to find that this wreck has been here a century at least!" Without orders, he turned on the torch. A four-foot flame of pure blue-white leaped out. He touched its tip to the fungoid soil. Steam leaped up. He used the flame like a gigantic scalpel, cutting a square a yard deep in the whitish stuff, and then cutting it across and across to destroy it. Thick fumes arose, and quiverings and shakings began. Black creatures in their labyrinths of tunnels began to panic. Off to the right the blanket-like surface ripped and they poured out. They scuttled crazily here and there. Some took to wing. By instinct the other men—the armed ones—moved back from the smoke. They wore space-helmets but they felt that there should be an intolerable smell. Moran slashed and slashed angrily with the big flame, cutting a way to the metal hull that had fallen here before his grandfather was born. Sometimes the flame cut across things that writhed, and he was sickened. But above all he raged because he was to be marooned here. He could not altogether blame the others. They couldn't land at any colonized world with him on board without his being detected as an extra member of the crew. His fate would then be sealed. But they also would be investigated. Official queries would go across this whole sector of the galaxy, naming five persons of such-and-such description and such-and-such fingerprints, voyaging in a space-yacht of such-and-such size and registration. The world they came from would claim them as fugitives. They would be returned to it. They'd be executed. Then Carol's voice came in his helmet-phone. She cried out; " Look out! It's coming! Kill it! Kill it—. " He heard blast-rifles firing. He heard Burleigh pant commands. He was on his way out of the hollow he'd carved when he heard Harper cry out horribly. He got clear of the newly burned-away stuff. There was still much smoke and stream. But he saw Harper. More, he saw the thing that had Harper. It occurred to him instantly that if Harper died, there would not be too many people on the Nadine . They need not maroon him. In fact, they wouldn't dare. A ship that came in to port with two few on board would be investigated as thoroughly as one that had too many. Perhaps more thoroughly. So if Harper were killed, Moran would be needed to take his place. He'd go on from here in the Nadine , necessarily accepted as a member of her crew. Then he rushed, the flame-torch making a roaring sound. II. They went back to the Nadine for weapons more adequate for encountering the local fauna when it was over. Blast-rifles were not effective against such creatures as these. Torches were contact weapons but they killed. Blast-rifles did not. And Harper needed to pull himself together again, too. Also, neither Moran nor any of the others wanted to go back to the still un-entered wreck while the skinny, somehow disgusting legs of the thing still kicked spasmodically—quite separate—on the whitish ground-stuff. Moran had disliked such creatures in miniature form on other worlds. Enlarged like this. It seemed insane that such creatures, even in miniature, should painstakingly be brought across light-years of space to the new worlds men settled on. But it had been found to be necessary. The ecological system in which human beings belonged had turned out to be infinitely complicated. It had turned out, in fact, to be the ecological system of Earth, and unless all parts of the complex were present, the total was subtly or glaringly wrong. So mankind distastefully ferried pests as well as useful creatures to its new worlds as they were made ready for settlement. Mosquitos throve on the inhabited globes of the Rim Stars. Roaches twitched nervous antennae on the settled planets of the Coal-sack. Dogs on Antares had fleas, and scratched their bites, and humanity spread through the galaxy with an attendant train of insects and annoyances. If they left their pests behind, the total system of checks and balances which make life practical would get lopsided. It would not maintain itself. The vagaries that could result were admirably illustrated in and on the landscape outside the Nadine . Something had been left out of the seeding of this planet. The element—which might be a bacterium or a virus or almost anything at all—the element that kept creatures at the size called "normal" was either missing or inoperable here. The results were not desirable.
|
C. location of the ice cap
|
which tasks are used in BLUE benchmark?
|
### Introduction
With the growing amount of biomedical information available in textual form, there have been significant advances in the development of pre-training language representations that can be applied to a range of different tasks in the biomedical domain, such as pre-trained word embeddings, sentence embeddings, and contextual representations BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 , BIBREF4 . In the general domain, we have recently observed that the General Language Understanding Evaluation (GLUE) benchmark BIBREF5 has been successfully promoting the development of language representations of general purpose BIBREF2 , BIBREF6 , BIBREF7 . To the best of our knowledge, however, there is no publicly available benchmarking in the biomedicine domain. To facilitate research on language representations in the biomedicine domain, we present the Biomedical Language Understanding Evaluation (BLUE) benchmark, which consists of five different biomedicine text-mining tasks with ten corpora. Here, we rely on preexisting datasets because they have been widely used by the BioNLP community as shared tasks BIBREF8 . These tasks cover a diverse range of text genres (biomedical literature and clinical notes), dataset sizes, and degrees of difficulty and, more importantly, highlight common biomedicine text-mining challenges. We expect that the models that perform better on all or most tasks in BLUE will address other biomedicine tasks more robustly. To better understand the challenge posed by BLUE, we conduct experiments with two baselines: One makes use of the BERT model BIBREF7 and one makes use of ELMo BIBREF2 . Both are state-of-the-art language representation models and demonstrate promising results in NLP tasks of general purpose. We find that the BERT model pre-trained on PubMed abstracts BIBREF9 and MIMIC-III clinical notes BIBREF10 achieves the best results, and is significantly superior to other models in the clinical domain. This demonstrates the importance of pre-training among different text genres. In summary, we offer: (i) five tasks with ten biomedical and clinical text-mining corpora with different sizes and levels of difficulty, (ii) codes for data construction and model evaluation for fair comparisons, (iii) pretrained BERT models on PubMed abstracts and MIMIC-III, and (iv) baseline results. ### Related work
There is a long history of using shared language representations to capture text semantics in biomedical text and data mining research. Such research utilizes a technique, termed transfer learning, whereby the language representations are pre-trained on large corpora and fine-tuned in a variety of downstream tasks, such as named entity recognition and relation extraction. One established trend is a form of word embeddings that represent the semantic, using high dimensional vectors BIBREF0 , BIBREF11 , BIBREF12 . Similar methods also have been derived to improve embeddings of word sequences by introducing sentence embeddings BIBREF1 . They always, however, require complicated neural networks to be effectively used in downstream applications. Another popular trend, especially in recent years, is the context-dependent representation. Different from word embeddings, it allows the meaning of a word to change according to the context in which it is used BIBREF13 , BIBREF2 , BIBREF7 , BIBREF14 . In the scientific domain, BIBREF15 released SciBERT which is trained on scientific text. In the biomedical domain, BioBERT BIBREF3 and BioELMo BIBREF16 were pre-trained and applied to several specific tasks. In the clinical domain, BIBREF17 released a clinical BERT base model trained on the MIMIC-III database. Most of these works, however, were evaluated on either different datasets or the same dataset with slightly different sizes of examples. This makes it challenging to fairly compare various language models. Based on these reasons, a standard benchmarking is urgently required. Parallel to our work, BIBREF3 introduced three tasks: named entity recognition, relation extraction, and QA, while BIBREF16 introduced NLI in addition to named entity recognition. To this end, we deem that BLUE is different in three ways. First, BLUE is selected to cover a diverse range of text genres, including both biomedical and clinical domains. Second, BLUE goes beyond sentence or sentence pairs by including document classification tasks. Third, BLUE provides a comprehensive suite of codes to reconstruct dataset from scratch without removing any instances. ### Tasks
BLUE contains five tasks with ten corpora that cover a broad range of data quantities and difficulties (Table 1 ). Here, we rely on preexisting datasets because they have been widely used by the BioNLP community as shared tasks. ### Sentence similarity
The sentence similarity task is to predict similarity scores based on sentence pairs. Following common practice, we evaluate similarity by using Pearson correlation coefficients. BIOSSES is a corpus of sentence pairs selected from the Biomedical Summarization Track Training Dataset in the biomedical domain BIBREF18 . To develop BIOSSES, five curators judged their similarity, using scores that ranged from 0 (no relation) to 4 (equivalent). Here, we randomly select 80% for training and 20% for testing because there is no standard splits in the released data. MedSTS is a corpus of sentence pairs selected from Mayo Clinic’s clinical data warehouse BIBREF19 . To develop MedSTS, two medical experts graded the sentence's semantic similarity scores from 0 to 5 (low to high similarity). We use the standard training and testing sets in the shared task. ### Named entity recognition
The aim of the named entity recognition task is to predict mention spans given in the text BIBREF20 . The results are evaluated through a comparison of the set of mention spans annotated within the document with the set of mention spans predicted by the model. We evaluate the results by using the strict version of precision, recall, and F1-score. For disjoint mentions, all spans also must be strictly correct. To construct the dataset, we used spaCy to split the text into a sequence of tokens when the original datasets do not provide such information. BC5CDR is a collection of 1,500 PubMed titles and abstracts selected from the CTD-Pfizer corpus and was used in the BioCreative V chemical-disease relation task BIBREF21 . The diseases and chemicals mentioned in the articles were annotated independently by two human experts with medical training and curation experience. We use the standard training and test set in the BC5CDR shared task BIBREF22 . ShARe/CLEF eHealth Task 1 Corpus is a collection of 299 deidentified clinical free-text notes from the MIMIC II database BIBREF23 . The disorders mentioned in the clinical notes were annotated by two professionally trained annotators, followed by an adjudication step, resulting in high inter-annotator agreement. We use the standard training and test set in the ShARe/CLEF eHealth Tasks 1. ### Relation extraction
The aim of the relation extraction task is to predict relations and their types between the two entities mentioned in the sentences. The relations with types were compared to annotated data. We use the standard micro-average precision, recall, and F1-score metrics. DDI extraction 2013 corpus is a collection of 792 texts selected from the DrugBank database and other 233 Medline abstracts BIBREF24 . The drug-drug interactions, including both pharmacokinetic and pharmacodynamic interactions, were annotated by two expert pharmacists with a substantial background in pharmacovigilance. In our benchmark, we use 624 train files and 191 test files to evaluate the performance and report the micro-average F1-score of the four DDI types. ChemProt consists of 1,820 PubMed abstracts with chemical-protein interactions annotated by domain experts and was used in the BioCreative VI text mining chemical-protein interactions shared task BIBREF25 . We use the standard training and test sets in the ChemProt shared task and evaluate the same five classes: CPR:3, CPR:4, CPR:5, CPR:6, and CPR:9. i2b2 2010 shared task collection consists of 170 documents for training and 256 documents for testing, which is the subset of the original dataset BIBREF26 . The dataset was collected from three different hospitals and was annotated by medical practitioners for eight types of relations between problems and treatments. ### Document multilabel classification
The multilabel classification task predicts multiple labels from the texts. HoC (the Hallmarks of Cancers corpus) consists of 1,580 PubMed abstracts annotated with ten currently known hallmarks of cancer BIBREF27 . Annotation was performed at sentence level by an expert with 15+ years of experience in cancer research. We use 315 ( $\sim $ 20%) abstracts for testing and the remaining abstracts for training. For the HoC task, we followed the common practice and reported the example-based F1-score on the abstract level BIBREF28 , BIBREF29 . ### Inference task
The aim of the inference task is to predict whether the premise sentence entails or contradicts the hypothesis sentence. We use the standard overall accuracy to evaluate the performance. MedNLI is a collection of sentence pairs selected from MIMIC-III BIBREF30 . Given a premise sentence and a hypothesis sentence, two board-certified radiologists graded whether the task predicted whether the premise entails the hypothesis (entailment), contradicts the hypothesis (contradiction), or neither (neutral). We use the same training, development, and test sets in Romanov and Shivade BIBREF30 . ### Total score
Following the practice in BIBREF5 and BIBREF3 , we use a macro-average of F1-scores and Pearson scores to determine a system's position. ### Baselines
For baselines, we evaluate several pre-training models as described below. The original code for the baselines is available at https://github.com/ncbi-nlp/NCBI_BERT. ### BERT
BERT BIBREF7 is a contextualized word representation model that is pre-trained based on a masked language model, using bidirectional Transformers BIBREF31 . In this paper, we pre-trained our own model BERT on PubMed abstracts and clinical notes (MIMIC-III). The statistics of the text corpora on which BERT was pre-trained are shown in Table 2 . We initialized BERT with pre-trained BERT provided by BIBREF7 . We then continue to pre-train the model, using the listed corpora. We released our BERT-Base and BERT-Large models, using the same vocabulary, sequence length, and other configurations provided by BIBREF7 . Both models were trained with 5M steps on the PubMed corpus and 0.2M steps on the MIMIC-III corpus. BERT is applied to various downstream text-mining tasks while requiring only minimal architecture modification. For sentence similarity tasks, we packed the sentence pairs together into a single sequence, as suggested in BIBREF7 . For named entity recognition, we used the BIO tags for each token in the sentence. We considered the tasks similar to machine translation, as predicting the sequence of BIO tags from the input sentence. We treated the relation extraction task as a sentence classification by replacing two named entity mentions of interest in the sentence with pre-defined tags (e.g., @GENE$, @DRUG$) BIBREF3 . For example, we used “@CHEMICAL$ protected against the RTI-76-induced inhibition of @GENE$ binding.” to replace the original sentence “Citalopram protected against the RTI-76-induced inhibition of SERT binding.” in which “citalopram” and “SERT” has a chemical-gene relation. For multi-label tasks, we fine-tuned the model to predict multi-labels for each sentence in the document. We then combine the labels in one document and compare them with the gold-standard. Like BERT, we provided sources code for fine-tuning, prediction, and evaluation to make it straightforward to follow those examples to use our BERT pre-trained models for all tasks. Table 1: BLUE tasks Table 2: Corpora Table 3: Baseline performance on the BLUE task test sets.
|
Inference task
The aim of the inference task is to predict whether the premise sentence entails or contradicts the hypothesis sentence, Document multilabel classification
The multilabel classification task predicts multiple labels from the texts., Relation extraction
The aim of the relation extraction task is to predict relations and their types between the two entities mentioned in the sentences., Named entity recognition
The aim of the named entity recognition task is to predict mention spans given in the text , Sentence similarity
The sentence similarity task is to predict similarity scores based on sentence pairs
|
Why does Groz not want to go into the mountain?
A. He does not want to get separated from his team
B. He is scared of the wildlife that might try to attack
C. He is nervous about the technology left behind
D. He knows it will be hard to see the people he is chasing
|
WHEN THE MOUNTAIN SHOOK By Robert Abernathy Illustrated by Kelly Freas [Transcriber Note: This etext was produced from IF Worlds of Science Fiction March 1954. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] Dark was the Ryzga mountain and forbidding; steep were its cliffs and sheer its crevasses. But its outward perils could not compare with the Ryzgas themselves, who slept within, ready to wake and conquer.... At sunset they were in sight of the Ryzga mountain. Strangely it towered among the cliffs and snow-slopes of the surrounding ranges: an immense and repellently geometric cone, black, its sides blood-tinted by the dying sun. Neena shivered, even though the surrounding cold could not reach her. The ice-wind blew from the glacier, but Var's love was round her as a warming cloak, a cloak that glowed softly golden in the deepening twilight, even as her love was about him. Var said, "The Watcher's cave should be three miles beyond this pass." He stood rigid, trying to catch an echo of the Watcher's thoughts, but there was nothing. Perhaps the old man was resting. From the other direction, the long way that they two had come, it was not difficult to sense the thought of Groz. That thought was powerful, and heavy with vengeance. "Hurry," said Neena. "They're closer than they were an hour ago." She was beautiful and defiant, facing the red sunset and the black mountain. Var sensed her fear, and the love that had conquered it. He felt a wave of tenderness and bitterness. For him she had come to this. For the flame that had sprung between them at the Truce of New Grass, she had challenged the feud of their peoples and had left her home, to follow him. Now, if her father and his kinsmen overtook them, it would be death for Var, and for Neena living shame. Which of the two was worse was no longer a simple problem to Var, who had grown much older in the last days. "Wait," he commanded. While she waited he spun a dream, attaching it to the crags that loomed over the pass, and to the frozen ground underfoot. It was black night, as it would really be when Groz and his henchmen reached this place; lurid fire spewed from the Ryzga mountain, and strange lights dipped above it; and for good measure there was an avalanche in the dream, and hideous beasts rushed snapping and ravening from the crevices of the rock. "Oh!" cried Neena in involuntary alarm. Var sighed, shaking his head. "It won't hold them for long, but it's the best I can do now. Come on." There was no path. Now they were descending the steeper face of the sierra, and the way led over bottomless crevasses, sheer drops and sheer ascents, sheets of traitorous glare ice. Place after place had to be crossed on the air, and both grew weary with the effort such crossings cost. They hoarded their strength, helping one another; one alone might never have won through. It was starry night already when they saw the light from the Watcher's cave. The light shone watery and dim from beneath the hoary back of the glacier, and as they came nearer they saw why: the cave entrance was sealed by a sheet of ice, a frozen waterfall that fell motionless from the rocks above. They heard no sound. The two young people stared for a long minute, intrigued and fearful. Both had heard of this place, and the ancient who lived there to keep watch on the Ryzga mountain, as a part of the oldest legends of their childhood; but neither had been here before. But this was no time for shyness. Var eyed the ice-curtain closely to make sure that it was real, not dream-stuff; then he struck it boldly with his fist. It shattered and fell in a rain of splinters, sparkling in the light that poured from within. They felt the Watcher rouse, heard his footsteps, and finally saw him—a shrunken old man, white-haired, with a lined beardless face. The sight of him, more marred by age than anyone they had ever seen before, was disappointing. They had expected something more—an ancient giant, a tower of wisdom and strength. The Watcher was four hundred years old; beside him even Groz, who had always seemed so ancient, was like a boy. The Watcher peered at them in turn. "Welcome," he said in a cracked voice. He did not speak again; the rest of his conversation was in thought only. "Welcome indeed. I am too much alone here." "You were asleep!" said Var. Shock made his thought accusing, though he had not meant to be. The old man grinned toothlessly. "Never fear. Asleep or awake, I watch. Come in! You're letting in the wind." Inside the cave it was warm as summer. Var saw with some surprise that all the walls were sheathed in ice—warm to the touch, bound fast against melting by the Watcher's will. Light blazed in reflections from the ice walls, till there was no shadow in the place. Behind them began a tinkling of falling water, thawed from the glacial ridges above to descend sheet-wise over the cave mouth, freezing as it fell into lengthening icicles. The old man gazed at his work for a moment, then turned questioningly to the young pair. "We need a little rest out of the cold," said Var. "And food, if you can spare it. We're pursued." "Yes, yes. You shall have what I can give you. Make yourselves comfortable, and in one minute.... Pursued, eh? A pity. I see the world is as bad as it was when I was last in it." Hot food and drink were before them almost at once. The Watcher regarded them with compassion as their eyes brightened and some of the shadow of weariness lifted from them. "You have stolen your enemy's daughter, no doubt, young man? Such things happened when I was young." Warming to the old man now, Var sketched his and Neena's history briefly. "We should have been safe among my people by now. And before very long, I'm sure, I would have performed some deed which Groz would recognize as a worthy exploit, and would thus have healed the feud between our families. But our flight was found out too soon. They cut us off and forced us into the mountains, and now they are only a few hours behind us." "A pity, indeed. I would like to help you—but, you understand, I am the Mountain Watcher. I must be above feuds and families." Var nodded somberly, thinking that an old recluse would in any case be able to do little for them against Groz and his violent kinsfolk. "And what will you do now?" Var grinned mirthlessly. "We haven't much choice, since they're overtaking us. I have only one idea left: we can go where Groz may fear to follow us." "To the mountain, you mean." "And into it, if need be." The Watcher was broodingly silent; his eyes shifted to Neena, where she nestled by Var's side. He asked, "And you—are you willing to follow your lover in this?" Neena returned his gaze without flinching; then she looked sidelong at Var, and her lips curled with a proud and tender mockery. "Follow? Why, I will lead, if his courage should fail him." The old man said, "It is no part of my duty to dissuade you from this thing. You are free persons. But I must be sure that you know what you are doing. That is the second part of the law the First Watcher made: to guard lest the unwary and the ignorant should bring harm on themselves and on all men." "We know the stories," Var said brusquely. "In the hollow heart of their mountain the Ryzgas sleep, as they chose to do when their world crumbled. But if they are wakened, the mountain will tremble, and the Ryzgas will come forth." "Do you believe that?" "As one believes stories." "It is true," said the Watcher heavily. "In my youth I penetrated farther into the mountain than anyone before, farther even than did the First Watcher. I did not see the sleepers, nor will any man until they come again, but I met their sentries, the sentinel machines that guard them now as they have for two thousand years. When I had gone that far, the mountain began to shake, the force that is in the Earth rumbled below, and I returned in time." Now for the first time Var sensed the power in the old man's look, the power of four hundred years' wisdom. Var stared down at his hands. "The Ryzgas also were men," said the Watcher. "But they were such a race as the world has not seen before or since. There were tyrannies before the Ryzgas, there was lust for power, and atrocious cruelty; but such tyranny, power, and cruelty as theirs, had never been known. They ruled the Earth for four generations, and the Earth was too little for them. They laid the world waste, stripped it of metals and fuels and bored to its heart for energy, poisoned its seas and its air with the fume of their works, wrung its peoples dry for their labor ... and in each of those four generations they launched a ship of space. They were great and evil as no other people has been, because they wanted the stars. "Because of them we must build with dreams instead of iron, and our only fire is that of the Sun, and even now, two thousand years later, the Earth is still slowly recovering from the pangs and poison of that age. If you turn up the sod in the plain where the wild herds graze, you will find numberless fragments of rusted or corroded metal, bits of glass and strange plastic substances, debris of artifacts still showing the marks of their shaping—the scattered wreckage of the things they made. And we—we too are a remnant, the descendants of the few out of all humanity that survived when the Ryzgas' world went down in flame and thunder. "In the last generation of their power the Ryzgas knew by their science that the race of man would endure them no longer. They made ready their weapons, they mined the cities and the factories for destruction, making sure that their works and their knowledge would perish with them. Meanwhile they redoubled the yoke and the punishments, hastening the completion of the last of the starships. "From the memories that the old Watchers have left here, and from the memories of dead men that still echo in the air, I have gathered a picture of that world's end. I will show it to you...." Var and Neena stared, unstirring, with wide vacant eyes, while the old man wove a dream around them, and the bright ice-cave faded from their vision, and they saw— Black starless night, a sky of rolling smoke above the greatest city that was ever built. Only the angry light of fires relieved the city's darkness—that, and the blue-white lightning flashes that silhouetted the naked skeletons of buildings and were followed by thunder and a shaking of the earth. Along lightless streets, half choked with rubble and with the dead, poured a mad, hating horde. The recurrent flashes lit scarred faces, naked bodies blackened and maimed from the hell of the workshops where the Ryzgas' might had been forged, eyes that stared white and half sightless from the glare of the furnaces, gnarled hands that now at long last clutched the weapons of the last rebellion—a rebellion without hope of new life on a world gutted and smoldering from the fulfilment of the Ryzgas' dream, without slogans other than a cry for blood. Before them death waited around the citadel where the masters still fought. All round, from the lowest and most poisonous levels of the shattered city, the slaves swarmed up in their millions. And the lightning blazed, and the city howled and screamed and burned. Then, unbelievably, the thunder fell silent, and the silence swept outward like a wave, from ruined street to street. The mouths that had shouted their wrath were speechless, and the rage-blinded eyes were lifted in sudden awe. From the center, over the citadel, an immense white globe soared upward, rising swiftly without sound. They had never seen its like, but they knew. It was the last starship, and it was leaving. It poised motionless. For an instant the burning city lay mute; then the millions found voice. Some roared ferocious threats and curses; others cried desolately— wait! Then the whole city, the dark tumuli of its buildings and its leaping fires and tormented faces, and the black sky over it, seemed to twist and swim, like a scene under water when a great fish sweeps past, and the ship was gone. The stunned paralysis fell apart in fury. Flame towered over the citadel. The hordes ran and shrieked again toward the central inferno, and the city burned and burned.... Var blinked dazedly in the shadowless glow of the ice-cave. His arm tightened about Neena till she gasped. He was momentarily uncertain that he and she were real and here, such had been the force of the dream, a vision of such scope and reality as Var had never seen—no, lived through—before. With deep respect now he gazed upon the bent old man who was the Mountain Watcher. "Some of the Ryzgas took flight to the stars, and some perished on Earth. But there was a group of them who believed that their time to rule would come again. These raised a black mountain from the Earth's heart, and in hollows within it cast themselves into deathless sleep, their deathless and lifeless sentinels round them, to wait till someone dare arouse them, or until their chosen time—no one knows surely. "I have told you the story you know, and have shown you a glimpse of the old time, because I must make sure that you do not approach the mountain in ignorance. Our world is unwise and sometimes evil, full of arrogance, folly, and passion that are in the nature of man. Yet it is a happy world, compared to that the Ryzgas made and will make again." The Watcher eyed them speculatively. "Before all," he said finally, "this is a world where you are free to risk wakening the old tyrants, if in your own judgment your great need renders the chance worth taking." Neena pressed her face against Var's shoulder, hiding her eyes. In her mind as it groped for his there was a confusion of horror and pity. Var looked grimly at the Watcher, and would have spoken; but the Watcher seemed suddenly a very long way off, and Var could no longer feel his own limbs, his face was a numb mask. Dully he heard the old man say, "You are tired. Best sleep until morning." Var strove to cry out that there was no time, that Groz was near and that sleep was for infants and the aged, but his intention sank and drowned under wave upon wave of unconquerable languor. The bright cave swam and dissolved; his eyelids closed. Var woke. Daylight glimmered through the ice of the cave mouth. He had been unconscious, helpless, for hours! At the thought of that, panic gripped him. He had not slept since childhood, and he had forgotten how it was. He came to his feet in one quick movement, realizing in that action that sleep had refreshed his mind and body—realizing also that a footstep had wakened him. Across the cave he faced a young man who watched him coolly with dark piercing eyes that were familiar though he did not know the face. Neena sat up and stifled a cry of fright. Var growled, "Who are you? Where's the Watcher?" The other flashed white teeth in a smile. "I'm the Watcher," he answered. "Often I become a youth at morning, and relax into age as the day passes. A foolish amusement, no doubt, but amusements are few here." "You made us fall asleep. Groz will be on us—" "Groz and his people could not detect your thoughts as you slept. They were all night chasing elusive dreams on the high ridges, miles away." Var passed a hand across bewildered eyes. Neena said softly, "Thank you, Watcher." "Don't thank me. I take no sides in your valley feuds. But now you are rested, your minds are clear. Do you still mean to go on to the Ryzga mountain?" Not looking at the Watcher, Var muttered unsteadily, "We have no alternative." There was a liquid tinkling as the ice-curtain collapsed; the fresh breeze of morning swept into the cave. The youth beckoned to them, and they followed him outside. The glacial slope on which the cavern opened faced toward the mountain. It rose black and forbidding in the dawn as it had by sunset. To right and left of it, the grand cliffs, ocher and red, were lit splendidly by the morning sun, but the mountain of the Ryzgas drank in the light and gave nothing back. Below their feet the slope fell away into an opaque sea of fog, filling a mile-wide gorge. There was a sound of turbulent water, of a river dashed from rock to rock in its struggle toward the plain, but the curling fog hid everything. "You have an alternative," said the Watcher crisply. The two took their eyes from the black mountain and gazed at him in sudden hope, but his face was unsmiling. "It is this. You, Var, can flee up the canyon to the north, by a way I will show you, disguising your thoughts and masking your presence as well as you are able, while the girl goes in the other direction, southward, without seeking to conceal herself. Your pursuers will be deceived and follow her, and by the time they catch her it will be too late for them to overtake Var." That possibility had not occurred to them at all. Var and Neena looked at one another. Then by common consent they blended their minds into one. They thought, in the warm intimacy of unreserved understanding: " It would work: I-you would make the sacrifice of shame and mockery—yet these can be borne—that I-you might be saved from death—which is alone irreparable.... But to become I and you again—that cannot be borne. " They said in unison, "No. Not that." The Watcher's face did not change. He said gravely, "Very well. I will give you what knowledge I have that may help you when you enter the Ryzga mountain." Quickly, he impressed on them what he had learned of the structure of the mountain and of its guardian machines. Var closed his eyes, a little dizzied by the rapid flood of detail. "You are ready to go," said the Watcher. He spoke aloud, and his voice was cracked and harsh. Var opened his eyes in surprise, and saw that the Watcher had become again the hoary ancient of last night. Var felt a twinge of unfamiliar emotion; only by its echo in Neena's mind did he recognize it as a sense of guilt. He said stiffly, "You don't blame us?" "You have taken life in your own hands," rasped the Watcher. "Who does that needs no blessing and feels no curse. Go!" They groped through the fog above blank abysses that hid the snarling river, crept hand in hand, sharing their strength, across unstable dream bridges from crag to crag. Groz and his pack, in their numbers, would cross the gorge more surely and swiftly. When Var and Neena set foot at last on the cindery slope of the great volcanic cone, they sensed that the pursuit already halved their lead. They stood high on the side of the Ryzga mountain, and gazed at the doorway. It was an opaque yet penetrable well of darkness, opening into the face of a lava cliff, closed only by an intangible curtain—so little had the Ryzgas feared those who might assail them in their sleep. Var sent his thoughts probing beyond the curtain, listened intently, head thrown back, to their echoes that returned. The tunnel beyond slanted steeply downward. Var's hands moved, molding a radiant globe from the feeble sunshine that straggled through the fog-bank. With an abrupt motion he hurled it. The sun-globe vanished, as if the darkness had drunk it up, but though sight did not serve they both sensed that it had passed through to light up the depths beyond. For within the mountain something snapped suddenly alert—something alive yet not living, seeing yet blind. They felt light-sensitive cells tingle in response, felt electric currents sting along buried, long-idle circuits.... The two stood shivering together. The morning wind stirred, freshening, the fog lifted a little, and they heard a great voice crying, "There they are!" Var and Neena turned. Far out in the sea of fog, on a dream bridge that they could not see, stood Groz. He shook the staff he carried. It was too far to discern the rage that must contort his features, but the thought he hurled at them was a soundless bellow: "Young fools! I've caught you now!" Behind Groz the figures of his followers loomed up as striding shadows. Neena's hand tightened on Var's. Var sent a thought of defiance: "Go back! Or you'll drive us to enter the mountain!" Groz seemed to hesitate. Then he swung his staff up like a weapon, and for the two on the mountainside the world turned upside down, the mountain's black shoulder hung inverted above them and the dizzy gulf of sky was beneath. Var fought for footing with his balance gone, feeling Neena reel against him until, summoning all his strength, he broke the grip of the illusion and the world seemed to right itself. The mist billowed again and Groz was out of sight, but they could hear him exhorting his men to haste. Neena's face was deadly pale and her lips trembled, but her urgent whisper said, "Come on!" Together they plunged into the curtain of darkness. At Var's thought command Neena froze instantly. "Feel that!" he muttered, and she, listening, sensed it too: the infinitesimal trickle of currents behind what appeared to be a blank tunnel wall, a rising potential that seemed to whisper Ready ... ready.... The sun-globe floated behind them, casting light before them down the featureless tunnel that sloped always toward the mountain's heart. Var summoned it, and it drifted ahead, a dozen feet, a little more— Between wall and wall a blinding spindle of flame sprang into being, pulsed briefly with radiant energy that pained the eyes, and went out. The immaterial globe of light danced on before them. "Forward, before the charge builds up again!" said Var. A few feet further on, they stumbled over a pile of charred bones. Someone else had made it only this far. It was farther than the Watcher had gone into these uncharted regions, and only the utmost alertness of mind and sense had saved them from death in traps like this. But as yet the way was not blocked.... Then they felt the mountain begin to tremble. A very faint and remote vibration at first, then an increasingly potent shuddering of the floor under their feet and the walls around them. Somewhere far below immense energies were stirring for the first time in centuries. The power that was in the Earth was rising; great wheels commenced to turn, the mechanical servitors of the Ryzgas woke one by one and began to make ready, while their masters yet slept, for the moment of rebirth that might be near at hand. From behind, up the tunnel, came a clear involuntary thought of dismay, then a directed thought, echoing and ghostly in the confinement of the dark burrow: " Stop! —before you go too far!" Var faced that way and thought coldly: "Only if you return and let us go free." In the black reaches of the shaft his will groped for and locked with that of Groz, like the grip of two strong wrestlers. In that grip each knew with finality that the other's stubbornness matched his own—that neither would yield, though the mountain above them and the world outside should crumble to ruin around them. "Follow us, then!" They plunged deeper into the mountain. And the shaking of the mountain increased with every step, its vibrations became sound, and its sound was like that of the terrible city which they had seen in the dream. Through the slow-rolling thunder of the hidden machines seemed to echo the death-cries of a billion slaves, the despair of all flesh and blood before their monstrous and inhuman power. Without warning, lights went on. Blinking in their glare, Var and Neena saw that fifty paces before them the way opened out into a great rounded room that was likewise ablaze with light. Cautiously they crept forward to the threshold of that chamber at the mountain's heart. Its roof was vaulted; its circular walls were lined with panels studded with gleaming control buttons, levers, colored lights. As they watched light flicked on and off in changing patterns, registering the progressive changes in the vast complex of mechanisms for which this must be the central control station. Behind those boards circuits opened and closed in bewildering confusion; the two invaders felt the rapid shifting of magnetic fields, the fury of electrons boiling in vacuum.... For long moments they forgot the pursuit, forgot everything in wonder at this place whose remotest like they had never seen in the simplicity of their machineless culture. In all the brilliant space there was no life. They looked at one another, the same thought coming to both at once: perhaps, after two thousand years, the masters were dead after all, and only the machines remained? As if irresistibly drawn, they stepped over the threshold. There was a clang of metal like a signal. Halfway up the wall opposite, above a narrow ramp that descended between the instrument panels, a massive doorway swung wide, and in its opening a figure stood. Var and Neena huddled frozenly, half expecting each instant to be their last. And the Ryzga too stood motionless, looking down at them. He was a man of middle height and stocky build, clad in a garment of changing colors, of fabric delicate as dream-stuff. In his right hand, with the care one uses with a weapon, he grasped a gleaming metal tube; his other hand rested as for support against the frame of the doorway. That, and his movements when he came slowly down the ramp toward them, conveyed a queer suggestion of weariness or weakness, as if he were yet not wholly roused from his two millenia of slumber. But the Ryzga's manner and his mind radiated a consciousness of power, a pride and assurance of self that smote them like a numbing blow. With a new shock, Var realized that the Ryzga's thoughts were quite open. They had a terse, disconnected quality that was strange and unsettling, and in part they were couched in alien and unintelligible symbols. But there was no block. Apparently the Ryzga felt no need to close his mind in the presence of inferior creatures.... He paused with his back to the central control panel, and studied the interlopers with the dispassionate gaze of a scientist examining a new, but not novel, species of insect. His thoughts seemed to click, like metal parts of a mechanism falling into places prepared for them. The image occurred oddly to Var, to whom such a comparison would ordinarily have been totally strange. "Culture: late barbarism. Handwork of high quality—good. Physically excellent stock...." There was a complicated and incomprehensible schemata of numbers and abstract forms. "The time: two thousand years—more progress might have been expected, if any survivors at all initially postulated; but this will do. The pessimists were mistaken. We can begin again." Then, startlingly super-imposed on the cool progression of logical thought, came a wave of raw emotion, devastating in its force. It was a lustful image of a world once more obedient, crawling, laboring to do the Ryzgas' will— toward the stars, the stars! The icy calculation resumed: "Immobilize these and the ones indicated in the passage above. Then wake the rest...." Var was staring in fascination at the Ryzga's face. It was a face formed by the custom of unquestioned command; yet it was lined by a deeply ingrained weariness, the signs of premature age—denied, overridden by the driving will they had sensed a moment earlier. It was a sick man's face. The Ryzga's final thought clicked into place: Decision! He turned toward the switchboard behind him, reaching with practised certainty for one spot upon it. Neena screamed. Between the Ryzga and the control panel a nightmare shape reared up seven feet tall, flapping black amorphous limbs and flashing red eyes and white fangs. The Ryzga recoiled, and the weapon in his hand came up. There was an instantaneous glare like heat lightning, and the monster crumpled in on itself, twitched briefly and vanished. But in that moment a light of inspiration had flashed upon Var, and it remained. As the Ryzga stretched out his hand again, Var acted. The Ryzga froze, teetering off balance and almost falling, as a numbing grip closed down on all his motor nerves. Holding that grip, Var strode across the floor and looked straight into the Ryzga's frantic eyes. They glared back at him with such hatred and such evil that for an instant he almost faltered. But the Ryzga's efforts, as he strove to free himself from the neural hold, were as misdirected and unavailing as those of a child who has not learned to wrestle with the mind. Var had guessed right. When Neena in her terror had flung a dream monster into the Ryzga's way—a mere child's bogey out of a fairy tale—the Ryzga had not recognized it as such, but had taken it for a real being. Var laughed aloud, and with great care, as one communicates with an infant, he projected his thoughts into the other's mind. "There will be no new beginning for you in our world, Ryzga! In two thousand years, we've learned some new things. Now at last I understand why you built so many machines, such complicated arrangements of matter and energy to do simple tasks—it was because you knew no other way." Behind the hate-filled eyes the cold brain tried to reason still. "Barbarians...? Our party was wrong after all. After us the machine civilization could never rise again, because it was a fire that consumed its fuel. After us man could not survive on the Earth, because the conditions that made him great were gone. The survivors must be something else—capacities undeveloped by our science—after us the end of man, the beginning.... But those of us who chose to die were right." The tide of hate and sick desire rose up to drown all coherence. The Ryzga made a savage, wholly futile effort to lift the weapon in his paralyzed hand. Then his eyes rolled upward, and abruptly he went limp and fell in a heap, like a mechanical doll whose motive power has failed. Var felt Neena beside him, and drew her close. As she sobbed her relief, he continued to look down absently at the dead man. When at last he raised his head, he saw that the drama's end had had a further audience. In the outer doorway, backed by his clansmen, stood Groz, gazing first in stupefaction at the fallen Ryzga, then with something like awe at Var. Var eyed him for a long moment; then he smiled, and asked, "Well, Groz? Is our feud finished, or does your ambition for a worthy son-in-law go beyond the conqueror of the Ryzgas?"
|
C. He is nervous about the technology left behind
|
How do Martians communicate with men from Earth?
A. Without using logical sense, only the imagination.
B. By sensing and without the need for talking.
C. Through manuscripts and unwritten books.
D. Via time travel.
|
Confidence Game By JIM HARMON Illustrated by EPSTEIN [Transcriber's Note: This etext was produced from Galaxy Science Fiction June 1957. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] I admit it: I didn't know if I was coming or going—but I know that if I stuck to the old man, I was a comer ... even if he was a goner! Doc had this solemn human by the throat when I caught up with him. "Tonight," Doc was saying in his old voice that was as crackled and important as parchment, "tonight Man will reach the Moon. The golden Moon and the silver ship, symbols of greed. Tonight is the night when this is to happen." "Sure," the man agreed severely, prying a little worriedly at Doc's arthritic fingers that were clamped on his collar. "No argument. Sure, up we go. But leave me go or, so help me, I'll fetch you one in the teeth!" I came alongside and carefully started to lever the old man loose, one finger at a time. It had to be done this way. I had learned that during all these weeks and months. His hands looked old and crippled, but I felt they were the strongest in the world. If a half dozen winos in Seattle hadn't helped me get them loose, Doc and I would have been wanted for the murder of a North American Mountie. It was easier this night and that made me afraid. Doc's thin frame, layered with lumpy fat, was beginning to muscle-dance against my side. One of his times was coming on him. Then at last he was free of the greasy collar of the human. "I hope you'll forgive him, sir," I said, not meeting the man's eyes. "He's my father and very old, as you can see." I laughed inside at the absurd, easy lie. "Old events seem recent to him." The human nodded, Adam's apple jerking in the angry neon twilight. "'Memory Jump,' you mean. All my great-grandfathers have it. But Great-great-grandmother Lupos, funny thing, is like a schoolgirl. Sharp, you know. I.... Say, the poor old guy looks sick. Want any help?" I told the human no, thanks, and walked Doc toward the flophouse three doors down. I hoped we would make it. I didn't know what would happen if we didn't. Doc was liable to say something that might nova Sol, for all I knew. Martians approaching the corner were sensing at Doc and me. They were just cheap tourists slumming down on Skid Row. I hated tourists and especially I hated Martian tourists because I especially hated Martians. They were aliens . They weren't men like Doc and me. Then I realized what was about to happen. It was foolish and awful and true. I was going to have one of mine at the same time Doc was having his. That was bad. It had happened a few times right after I first found him, but now it was worse. For some undefinable reason, I felt we kept getting closer each of the times. I tried not to think about it and helped Doc through the fly-specked flophouse doors. The tubercular clerk looked up from the gaudy comics sections of one of those little tabloids that have the funnies a week in advance. "Fifteen cents a bed," he said mechanically. "We'll use one bed," I told him. "I'll give you twenty cents." I felt the round hard quarter in my pocket, sweaty hand against sticky lining. "Fifteen cents a bed," he played it back for me. Doc was quivering against me, his legs boneless. "We can always make it over to the mission," I lied. The clerk turned his upper lip as if he were going to spit. "Awright, since we ain't full up. In ad vance." I placed the quarter on the desk. "Give me a nickel." The clerk's hand fell on the coin and slid it off into the unknown before I could move, what with holding up Doc. "You've got your nerve," he said at me with a fine mist of dew. "Had a quarter all along and yet you Martian me down to twenty cents." He saw the look on my face. "I'll give you a room for the two bits. That's better'n a bed for twenty." I knew I was going to need that nickel. Desperately. I reached across the desk with my free hand and hauled the scrawny human up against the register hard. I'm not as strong in my hands as Doc, but I managed. "Give me a nickel," I said. "What nickel?" His eyes were big, but they kept looking right at me. "You don't have any nickel. You don't have any quarter, not if I say so. Want I should call a cop and tell him you were flexing a muscle?" I let go of him. He didn't scare me, but Doc was beginning to mumble and that did scare me. I had to get him alone. "Where's the room?" I asked. The room was six feet in all directions and the walls were five feet high. The other foot was finished in chicken wire. There was a wino singing on the left, a wino praying on the right, and the door didn't have any lock on it. At last, Doc and I were alone. I laid Doc out on the gray-brown cot and put his forearm over his face to shield it some from the glare of the light bulb. I swept off all the bedbugs in sight and stepped on them heavily. Then I dropped down into the painted stool chair and let my burning eyes rest on the obscene wall drawings just to focus them. I was so dirty, I could feel the grime grinding together all over me. My shaggy scalp still smarted from the alcohol I had stolen from a convertible's gas tank to get rid of Doc's and my cooties. Lucky that I never needed to shave and that my face was so dirty, no one would even notice that I didn't need to. The cramp hit me and I folded out of the chair onto the littered, uncovered floor. It stopped hurting, but I knew it would begin if I moved. I stared at a jagged cut-out nude curled against a lump of dust and lint, giving it an unreal distortion. Doc began to mumble louder. I knew I had to move. I waited just a moment, savoring the painless peace. Then, finally, I moved. I was bent double, but I got from the floor to the chair and found my notebook and orb-point in my hands. I found I couldn't focus both my mind and my eyes through the electric flashes of agony, so I concentrated on Doc's voice and trusted my hands would follow their habit pattern and construct the symbols for his words. They were suddenly distinguishable. " Outsider ... Thoth ... Dyzan ... Seven ... Hsan ... Beyond Six, Seven, Eight ... Two boxes ... Ralston ... Richard Wentworth ... Jimmy Christopher ... Kent Allard ... Ayem ... Oh, are ... see ...." His voice rose to a meaningless wail that stretched into non-existence. The pen slid across the scribbled face of the notebook and both dropped from my numb hands. But I knew. Somehow, inside me, I knew that these words were what I had been waiting for. They told everything I needed to know to become the most powerful man in the Solar Federation. That wasn't just an addict's dream. I knew who Doc was. When I got to thinking it was just a dream and that I was dragging this old man around North America for nothing, I remembered who he was. I remembered that he was somebody very important whose name and work I had once known, even if now I knew him only as Doc. Pain was a pendulum within me, swinging from low throbbing bass to high screaming tenor. I had to get out and get some. But I didn't have a nickel. Still, I had to get some. I crawled to the door and raised myself by the knob, slick with greasy dirt. The door opened and shut—there was no lock. I shouldn't leave Doc alone, but I had to. He was starting to cry. He didn't always do that. I listened to him for a moment, then tested and tasted the craving that crawled through my veins. I got back inside somehow. Doc was twisting on the cot, tears washing white streaks across his face. I shoved Doc's face up against my chest. I held onto him and let him bellow. I soothed the lanks of soiled white hair back over his lumpy skull. He shut up at last and I laid him down again and put his arm back across his face. (You can't turn the light off and on in places like that. The old wiring will blow the bulb half the time.) I don't remember how I got out onto the street. She was pink and clean and her platinum hair was pulled straight back, drawing her cheek-bones tighter, straightening her wide, appealing mouth, drawing her lean, athletic, feminine body erect. She was wearing a powder-blue dress that covered all of her breasts and hips and the upper half of her legs. The most wonderful thing about her was her perfume. Then I realized it wasn't perfume, only the scent of soap. Finally, I knew it wasn't that. It was just healthy, fresh-scrubbed skin. I went to her at the bus stop, forcing my legs not to stagger. Nobody would help a drunk. I don't know why, but nobody will help you if they think you are blotto. "Ma'am, could you help a man who's not had work?" I kept my eyes down. I couldn't look a human in the eye and ask for help. "Just a dime for a cup of coffee." I knew where I could get it for three cents, maybe two and a half. I felt her looking at me. She spoke in an educated voice, one she used, perhaps, as a teacher or supervising telephone operator. "Do you want it for coffee, or to apply, or a glass or hypo of something else?" I cringed and whined. She would expect it of me. I suddenly realized that anybody as clean as she was had to be a tourist here. I hate tourists. "Just coffee, ma'am." She was younger than I was, so I didn't have to call her that. "A little more for food, if you could spare it." I hadn't eaten in a day and a half, but I didn't care much. "I'll buy you a dinner," she said carefully, "provided I can go with you and see for myself that you actually eat it." I felt my face flushing red. "You wouldn't want to be seen with a bum like me, ma'am." "I'll be seen with you if you really want to eat." It was certainly unfair and probably immoral. But I had no choice whatever. "Okay," I said, tasting bitterness over the craving. The coffee was in a thick white cup before me on the counter. It was pale, grayish brown and steaming faintly. I picked it up in both hands to feel its warmth. Out of the corner of my eye, I could see the woman sitting on the stool beside me. She had no right to intrude. This moment should be mine, but there she sat, marring it for me, a contemptible tourist . I gulped down the thick, dark liquid brutally. It was all I could do. The cramp flowed out of my diaphragm. I took another swallow and was able to think straight again. A third swallow and I felt—good. Not abnormally stimulated, but strong, alert, poised on the brink of exhilaration. That was what coffee did for me. I was a caffeine addict. Earth-norm humans sometimes have the addiction to a slight extent, but I knew that as a Centurian I had it infinitely worse. Caffeine affected my metabolism like a pure alkaloid. The immediate effects weren't the same, but the need ran as deep. I finished the cup. I didn't order another because I wasn't a pure sensualist. I just needed release. Sometimes, when I didn't have the price of a cup, I would look around in alleys and find cola bottles with a few drops left in them. They have a little caffeine in them—not enough, never enough, but better than nothing. "Now what do you want to eat?" the woman asked. I didn't look at her. She didn't know. She thought I was a human—an Earth human. I was a man , of course, not an alien like a Martian. Earthmen ran the whole Solar Federation, but I was just as good as an Earthman. With my suntan and short mane, I could pass, couldn't I? That proved it, didn't it? "Hamburger," I said. "Well done." I knew that would probably be all they had fit to eat at a place like this. It might be horse meat, but then I didn't have the local prejudices. I didn't look at the woman. I couldn't. But I kept remembering how clean she looked and I was aware of how clean she smelled. I was so dirty, so very dirty that I could never get clean if I bathed every hour for the rest of my life. The hamburger was engulfed by five black-crowned, broken fingernails and raised to two rows of yellow ivory. I surrounded it like an ameba, almost in a single movement of my jaws. Several other hamburgers followed the first. I lost count. I drank a glass of milk. I didn't want to black out on coffee with Doc waiting for me. "Could I have a few to take with me, miss?" I pleaded. She smiled. I caught that out of the edge of my vision, but mostly I just felt it. "That's the first time you've called me anything but 'ma'am'," she said. "I'm not an old-maid schoolteacher, you know." That probably meant she was a schoolteacher, though. "No, miss," I said. "It's Miss Casey—Vivian Casey," she corrected. She was a schoolteacher, all right. No other girl would introduce herself as Miss Last Name. Then there was something in her voice.... "What's your name?" she said to me. I choked a little on a bite of stale bun. I had a name, of course . Everybody has a name, and I knew if I went off somewhere quiet and thought about it, mine would come to me. Meanwhile, I would tell the girl that my name was ... Kevin O'Malley. Abruptly I realized that that was my name. "Kevin," I told her. "John Kevin." "Mister Kevin," she said, her words dancing with bright absurdity like waterhose mist on a summer afternoon, "I wonder if you could help me ." "Happy to, miss," I mumbled. She pushed a white rectangle in front of me on the painted maroon bar. "What do you think of this?" I looked at the piece of paper. It was a coupon from a magazine. Dear Acolyte R. I. S. : Please send me FREE of obligation, in sealed wrapper, "The Scarlet Book" revealing to me how I may gain Secret Mastery of the Universe. Name : ........................ Address : ..................... The world disoriented itself and I was on the floor of the somber diner and Miss Vivian Casey was out of sight and scent. There was a five dollar bill tight in my fist. The counterman was trying to pull it out. I looked up at his stubbled face. "I had half a dozen hamburgers, a cup of coffee and a glass of milk. I want four more 'burgers to go and a pint of coffee. By your prices, that will be one sixty-five—if the lady didn't pay you." "She didn't," he stammered. "Why do you think I was trying to get that bill out of your hand?" I didn't say anything, just got up off the floor. After the counterman put down my change, I spread out the five dollar bill on the vacant bar, smoothing it. I scooped up my change and walked out the door. There was no one on the sidewalk, only in the doorways. First I opened the door on an amber world, then an azure one. Neon light was coming from the chickenwire border of the room, from a window somewhere beyond. The wino on one side of the room was singing and the one on the other side was praying, same as before. Only they had changed around—prayer came from the left, song from the right. Doc sat on the floor in the half-darkness and he had made a thing . My heart hammered at my lungs. I knew this last time had been different. Whatever it was was getting closer. This was the first time Doc had ever made anything. It didn't look like much, but it was a start. He had broken the light bulb and used the filament and screw bottom. His strong hands had unraveled some of the bed "springs"—metal webbing—and fashioned them to his needs. My orb-point pen had dissolved under his touch. All of them, useless parts, were made into a meaningful whole. I knew the thing had meaning, but when I tried to follow its design, I became lost. I put the paper container of warm coffee and the greasy bag of hamburgers on the wooden chair, hoping the odor wouldn't bring any hungry rats out of the walls. I knelt beside Doc. "An order, my boy, an order," he whispered. I didn't know what he meant. Was he suddenly trying to give me orders? He held something out to me. It was my notebook. He had used my pen, before dismantling it, to write something. I tilted the notebook against the neon light, now red wine, now fresh grape. I read it. "Concentrate," Doc said hoarsely. "Concentrate...." I wondered what the words meant. Wondering takes a kind of concentration. The words "First Edition" were what I was thinking about most. The heavy-set man in the ornate armchair was saying, "The bullet struck me as I was pulling on my boot...." I was kneeling on the floor of a Victorian living room. I'm quite familiar with Earth history and I recognized the period immediately. Then I realized what I had been trying to get from Doc all these months—time travel. A thin, sickly man was sprawled in the other chair in a rumpled dressing gown. My eyes held to his face, his pinpoint pupils and whitened nose. He was a condemned snowbird! If there was anything I hated or held in more contempt than tourists or Martians, it was a snowbird. "My clients have occasioned singular methods of entry into these rooms," the thin man remarked, "but never before have they used instantaneous materialization." The heavier man was half choking, half laughing. "I say—I say, I would like to see you explain this, my dear fellow." "I have no data," the thin man answered coolly. "In such instance, one begins to twist theories into fact, or facts into theories. I must ask this unemployed, former professional man who has gone through a serious illness and is suffering a more serious addiction to tell me the place and time from which he comes." The surprise stung. "How did you know?" I asked. He gestured with a pale hand. "To maintain a logical approach, I must reject the supernatural. Your arrival, unless hallucinatory—and despite my voluntary use of one drug and my involuntary experiences recently with another, I must accept the evidence of my senses or retire from my profession—your arrival was then super-normal. I might say super-scientific, of a science not of my or the good doctor's time, clearly. Time travel is a familiar folk legend and I have been reading an article by the entertaining Mr. Wells. Perhaps he will expand it into one of his novels of scientific romance." I knew who these two men were, with a tormenting doubt. "But the other—" "Your hands, though unclean, have never seen physical labor. Your cranial construction is of a superior type, or even if you reject my theories, concentration does set the facial features. I judge you have suffered an illness because of the inhibition of your beard growth. Your over-fondness for rum or opium, perhaps, is self-evident. You are at too resilient an age to be so sunk by even an amour. Why else then would you let yourself fall into such an underfed and unsanitary state?" He was so smug and so sure, this snowbird. I hated him. Because I couldn't trust to my own senses as he did. "You don't exist," I said slowly, painfully. "You are fictional creations." The doctor flushed darkly. "You give my literary agent too much credit for the addition of professional polish to my works." The other man was filling a large, curved pipe from something that looked vaguely like an ice-skate. "Interesting. Perhaps if our visitor would tell us something of his age with special reference to the theory and practice of temporal transference, Doctor, we would be better equipped to judge whether we exist." There was no theory or practice of time travel. I told them all I had ever heard theorized from Hindu yoga through Extra-sensory Perception to Relativity and the positron and negatron. "Interesting." He breathed out suffocating black clouds of smoke. "Presume that the people of your time by their 'Extra-sensory Perception' have altered the past to make it as they suppose it to be. The great historical figures are made the larger than life-size that we know them. The great literary creations assume reality." I thought of Cleopatra and Helen of Troy and wondered if they would be the goddesses of love that people imagined or the scrawny, big-nosed redhead and fading old woman of scholarship. Then I noticed the detective's hand that had been resting idly on a round brass weight of unknown sort to me. His tapered fingertips had indented the metal. His bright eyes followed mine and he smiled faintly. "Withdrawal symptoms." The admiration and affection for this man that had been slowly building up behind my hatred unbrinked. I remembered now that he had stopped. He was not really a snowbird. After a time, I asked the doctor a question. "Why, yes. I'm flattered. This is the first manuscript. Considering my professional handwriting, I recopied it more laboriously." Accepting the sheaf of papers and not looking back at these two great and good men, I concentrated on my own time and Doc. Nothing happened. My heart raced, but I saw something dancing before me like a dust mote in sunlight and stepped toward it.... ... into the effective range of Miss Casey's tiny gun. She inclined the lethal silver toy. "Let me see those papers, Kevin." I handed her the doctor's manuscript. Her breath escaped slowly and loudly. "It's all right. It's all right. It exists. It's real. Not even one of the unwritten ones. I've read this myself." Doc was lying on the cot, half his face twisted into horror. "Don't move, Kevin," she said. "I'll have to shoot you—maybe not to kill, but painfully." I watched her face flash blue, red, blue and knew she meant it. But I had known too much in too short a time. I had to help Doc, but there was something else. "I just want a drink of coffee from that container on the chair," I told her. She shook her head. "I don't know what you think it does to you." It was getting hard for me to think. "Who are you?" She showed me a card from her wrist purse. Vivian Casey, Constable, North American Mounted Police. I had to help Doc. I had to have some coffee. "What do you want?" "Listen, Kevin. Listen carefully to what I am saying. Doc found a method of time travel. It was almost a purely mathematical, topographical way divorced from modern physical sciences. He kept it secret and he wanted to make money with it. He was an idealist—he had his crusades. How can you make money with time travel?" I didn't know whether she was asking me, but I didn't know. All I knew was that I had to help Doc and get some coffee. "It takes money—money Doc didn't have—to make money," Miss Casey said, "even if you know what horse will come in and what stock will prosper. Besides, horse-racing and the stock market weren't a part of Doc's character. He was a scholar." Why did she keep using the past tense in reference to Doc? It scared me. He was lying so still with the left side of his face so twisted. I needed some coffee. "He became a book finder. He got rare editions of books and magazines for his clients in absolutely mint condition. That was all right—until he started obtaining books that did not exist ." I didn't know what all that was supposed to mean. I got to the chair, snatched up the coffee container, tore it open and gulped down the soothing liquid. I turned toward her and threw the rest of the coffee into her face. The coffee splashed out over her platinum hair and powder-blue dress that looked white when the neon was azure, purple when it was amber. The coffee stained and soiled and ruined, and I was fiercely glad, unreasonably happy. I tore the gun away from her by the short barrel, not letting my filthy hands touch her scrubbed pink ones. I pointed the gun generally at her and backed around the thing on the floor to the cot. Doc had a pulse, but it was irregular. I checked for a fever and there wasn't one. After that, I didn't know what to do. I looked up finally and saw a Martian in or about the doorway. "Call me Andre," the Martian said. "A common name but foreign. It should serve as a point of reference." I had always wondered how a thing like a Martian could talk. Sometimes I wondered if they really could. "You won't need the gun," Andre said conversationally. "I'll keep it, thanks. What do you want?" "I'll begin as Miss Casey did—by telling you things. Hundreds of people disappeared from North America a few months ago." "They always do," I told him. "They ceased to exist—as human beings—shortly after they received a book from Doc," the Martian said. Something seemed to strike me in the back of the neck. I staggered, but managed to hold onto the gun and stand up. "Use one of those sneaky Martian weapons again," I warned him, "and I'll kill the girl." Martians were supposed to be against the destruction of any life-form, I had read someplace. I doubted it, but it was worth a try. "Kevin," Andre said, "why don't you take a bath?" The Martian weapon staggered me again. I tried to say something. I tried to explain that I was so dirty that I could never get clean no matter how often I bathed. No words formed. "But, Kevin," Andre said, "you aren't that dirty." The blow shook the gun from my fingers. It almost fell into the thing on the floor, but at the last moment seemed to change direction and miss it. I knew something. "I don't wash because I drink coffee." "It's all right to drink coffee, isn't it?" he asked. "Of course," I said, and added absurdly, "That's why I don't wash." "You mean," Andre said slowly, ploddingly, "that if you bathed, you would be admitting that drinking coffee was in the same class as any other solitary vice that makes people wash frequently." I was knocked to my knees. "Kevin," the Martian said, "drinking coffee represents a major vice only in Centurian humanoids, not Earth-norm human beings. Which are you? " Nothing came out of my gabbling mouth. " What is Doc's full name? " I almost fell in, but at the last instant I caught myself and said, "Doctor Kevin O'Malley, Senior." From the bed, Doc said a word. "Son." Then he disappeared. I looked at that which he had made. I wondered where he had gone, in search of what. "He didn't use that," Andre said. So I was an Earthman, Doc's son. So my addiction to coffee was all in my mind. That didn't change anything. They say sex is all in your mind. I didn't want to be cured. I wouldn't be. Doc was gone. That was all I had now. That and the thing he left. "The rest is simple," Andre said. "Doc O'Malley bought up all the stock in a certain ancient metaphysical order and started supplying members with certain books. Can you imagine the effect of the Book of Dyzan or the Book of Thoth or the Seven Cryptical Books of Hsan or the Necronomican itself on human beings?" "But they don't exist," I said wearily. "Exactly, Kevin, exactly. They have never existed any more than your Victorian detective friend. But the unconscious racial mind has reached back into time and created them. And that unconscious mind, deeper than psychology terms the subconscious, has always known about the powers of ESP, telepathy, telekinesis, precognition. Through these books, the human race can tell itself how to achieve a state of pure logic, without food, without sex, without conflict—just as Doc has achieved such a state—a little late, true. He had a powerful guilt complex, even stronger than your withdrawal, over releasing this blessing on the inhabited universe, but reason finally prevailed. He had reached a state of pure thought." "The North American government has to have this secret, Kevin," the girl said. "You can't let it fall into the hands of the Martians." Andre did not deny that he wanted it to fall into his hands. I knew I could not let Doc's—Dad's—time travel thing fall into anyone's hands. I remembered that all the copies of the books had disappeared with their readers now. There must not be any more, I knew. Miss Casey did her duty and tried to stop me with a judo hold, but I don't think her heart was in it, because I reversed and broke it. I kicked the thing to pieces and stomped on the pieces. Maybe you can't stop the progress of science, but I knew it might be millenniums before Doc's genes and creative environment were recreated and time travel was rediscovered. Maybe we would be ready for it then. I knew we weren't now. Miss Casey leaned against my dirty chest and cried into it. I didn't mind her touching me. "I'm glad," she said. Andre flowed out of the doorway with a sigh. Of relief? I would never know. I supposed I had destroyed it because I didn't want the human race to become a thing of pure reason without purpose, direction or love, but I would never know for sure. I thought I could kick the habit—perhaps with Miss Casey's help—but I wasn't really confident. Maybe I had destroyed the time machine because a world without material needs would not grow and roast coffee.
|
B. By sensing and without the need for talking.
|
Which of these things contributed most to the lowered audience for UFC?
A. The better fighters were too expensive, so when they moved abroad the fanbase fell through
B. The scoring system defeated the purpose of the no-holds-barred sport which made it less exciting to watch
C. Other sports became more popular, and UFC ended up as another fad, leaving the fighters to return to their original combat sports
D. Misconceptions about the safety of the sport drove political spats that kicked UFC out of the spotlight
|
Fight Clubbed Fight Club , a movie about a fictional organization of men who strip down and beat each other to pulp, has provoked more than its share of media hand-wringing, particularly diatribes about Hollywood's infatuation with violence and Faludi-esque ruminations about the emasculated American male. Fight Club , however, has not sparked an iota of interest in a real organization of men who strip down and beat each other to pulp: the Ultimate Fighting Championship. UFC's flameout from national sensation to total irrelevance is a tragedy of American sports, a cautionary tale of prudishness, heavy-handed politics, and cultural myopia. UFC began in 1993 as a locker-room fantasy. What would happen if a kickboxer fought a wrestler? A karate champion fought a sumo champion? Promoters built an octagonal chain-link cage, invited eight top martial artists, and set them loose in no-holds-barred, bare-knuckles fights. "There are no rules!" bragged an early press release. Contestants would fight till "knockout, submission, doctor's intervention, or death." UFC allowed, even promoted, all notions of bad sportsmanship: kicking a man when he's down, hitting him in the groin, choking. Four-hundred-pound men were sent into the Octagon to maul guys half their size. Only biting and eye-gouging were forbidden. The gimmick entranced thousands of people (well, men). What happens when a 620-pound sumo champion fights a 200-pound kickboxer? Answer: The kickboxer knocks him silly in 35 seconds. They tuned in for bloodshed--"the damage," as fans like to call it. UFC fights could be horrifying. Tank Abbott, an ill-tempered, 270-pound street fighter, knocks out hapless opponent John Matua in 15 seconds. Then, before the ref can intervene, Abbott belts the unconscious Matua in the head, sending him into a fit, limbs quivering uncontrollably, blood spurting from his mouth. Abbott, naturally, became a cult hero and won a guest spot on Friends . (Matua walked out of the ring.) Soon, UFC was selling out huge arenas and drawing 300,000 pay-per-view subscribers for its quarterly competitions. But a subtle sport was emerging from the gimmicks and carnage. My passion for ultimate fighting (which is also called "extreme" or "no-holds-barred" fighting) began when I saw the finals of UFC IV. Royce Gracie, a 180-pound Brazilian jujitsu specialist, was matched against a 275-pound beast named Dan Severn, one of the top heavyweight wrestlers in the world and a national champion many times over. In 30 seconds, Severn had grabbed Gracie, flung him to the canvas, and mounted him. For the next 15 minutes, Severn pummeled and elbowed and head-butted the smaller man. Gracie's face grew drawn, and he squirmed wildly to avoid Severn's bombardment. Then, all of sudden, Gracie, still lying on his back, saw an opening, wrapped his arms and legs around Severn like a python and choked the giant into submission. UFC's caged matches revolutionized the idea of fighting. Nursed on boxing and Hollywood, Americans imagine fights as choreography, a dance of elegant combinations, roundhouse kicks, clean knockouts. The UFC punctured this. Boxers floundered. Experts in striking martial arts such as karate and tae kwon do, who fancied themselves the world's greatest fighters, found themselves pretzeled by jujitsu masters, who pulled them to the ground and slowly choked or leg-locked them. "UFC immediately debunked a lot of myths of fighting, of boxing, karate, kung fu. It showed the reality of what works in an actual fight," says Dave Meltzer, editor of Wrestling Observer . Instead of being carnivals of gore, UFC fights looked strangely like ... sex. Almost all fights ended on the ground, one man mounting the other in missionary position, the pair of them wiggling mysteriously along the canvas for five, 10, even 30 minutes. There were few spectacular knockouts. The referee--yes, there was always a referee--stopped many bouts, and in most others, fighters "tapped out," surrendering to mild-looking but agonizing chokes and joint locks. It was not barbarism. It was science. The UFC spawned a new breed of "mixed martial artists." World-class wrestlers learned to kickbox. Champion kickboxers learned to grapple. (The karate experts learned to stay home.) They became, without doubt, the best fighters in the world. (Click for more about the fighters.) Mike Tyson wouldn't last 30 seconds in an ultimate fighting match. When Olympic gold medal wrestler Kevin Jackson came to the UFC, a fighter named Frank Shamrock KO'd him with a submission hold in 16 seconds. Ultimate fighting schools began sprouting up all over the country, replacing the stylized gestures of the Eastern martial arts with techniques that actually work. UFC's promoters predicted that it would supplant boxing as America's martial art. Instead, it fell apart. The collapse began in 1996, when Sen. John McCain, R-Ariz., saw a UFC tape. McCain, a lifelong boxing fan, was horrified at the ground fighting, kicks, and head butts. It was "barbaric," he said. It was "not a sport." He sent letters to all 50 governors asking them to ban ultimate fighting. The outcry against "human cockfighting" became a crusade, and like many crusades, it was founded on misunderstanding. UFC fell victim to cultural determinism about what a fight is. In countries such as Brazil and Japan, where no-holds-barred fighting has a long history, it is popular and uncontroversial. But Americans adhere to the Marquis of Queensbury rules. A fight consists of an exchange of upper-body blows that halts when one fighter falls. Any blood sport can be barbaric, whether it's boxing or wrestling or ultimate fighting. It is impossible to draw a bright line between ultimate fighting and boxing. If anything, ultimate fighting is safer and less cruel than America's blood sport. For example, critics pilloried ultimate fighting because competitors fought with bare knuckles: To a nation accustomed to boxing gloves, this seemed revolting, an invitation to brain damage. But it's just the reverse: The purpose of boxing gloves is not to cushion the head but to shield the knuckles. Without gloves, a boxer would break his hands after a couple of punches to the skull. That's why ultimate fighters won't throw multiple skull punches. As a result, they avoid the concussive head wounds that kill boxers--and the long-term neurological damage that cripples them. Similarly, the chain-link fence surrounding the octagon looks grotesque. Critics have demanded that UFC install ropes instead. But ropes are a major cause of death and injury in boxing: Fighters hyperextend their necks when they are punched against the ropes, because nothing stops their heads from snapping back. The chain-link fence prevents hyperextension. When I tell people I'm an ultimate fighting fan, they invariably respond: "Don't people get killed all the time doing that?" But no one has ever been killed at the UFC--though boxers are killed every year. No one has even been seriously injured at the UFC. On the rare occasions when a bout has ended with a bloody knockout, the loser has always walked out of the ring. But this does not impress boxing fans, who are the most vigorous opponents of extreme fighting. McCain sat ringside at a boxing match where a fighter was killed. When I asked him to explain the moral distinction between boxing and ultimate fighting, he exploded at me, "If you can't see the moral distinction, then we have nothing to talk about!" Then he cut our interview short and stormed out of his office. But logic has not served the UFC well. Where McCain led, a prudish nation followed. George Will opined against UFC. The American Medical Association recommended a ban. New York state outlawed ultimate fighting, as did other states. The Nevada Athletic Commission refused to sanction UFC bouts, barring the UFC from the lucrative casino market. (One public TV station refused a UFC sponsorship ad. The only other organization the station ever rejected was the Ku Klux Klan.) Lawsuits blocked or delayed UFC events all over the country, forcing the promoters to spend millions in legal fees. The UFC was exiled from mega-arenas to ever-smaller venues in ever more out-of-the-way states: Louisiana, Iowa, and Alabama. The match I attended in October 1997 was held in the parking lot of a small Mississippi casino. The cable TV industry struck the fatal blow. In early 1997, McCain became chairman of the commerce committee, which oversees the cable industry. In April 1997, the president of the National Cable Television Association warned that UFC broadcasts could jeopardize the cable industry's influence in Washington. Time Warner, TCI, Request, Cablevision Systems, Viewer's Choice, and other major operators stopped airing UFC events, saying they were too violent for children. Never mind that 1) UFC only aired on pay-per-view, so children could not see it unless their parents paid for it; and 2) the same cable outfits carried boxing matches, R and NC-17 movies, and professional wrestling shows far more violent than UFC. The UFC's "addressable audience"--the potential number of PPV subscribers--shrank from 35 million at its peak to 7.5 million today. "It was a very cheap way for the cable companies to portray themselves as anti-violence. It did not cost them much and it made them look good in Washington," says Carol Klenfner, spokeswoman for UFC's parent company, SEG. The ultimate fighting industry did little to help its own cause. The UFC promoted itself less as a serious sport than as a circus of carnage. Its early ads emphasized extreme fighting's potential for death. UFC folks accused McCain, without any evidence, of opposing the sport as a favor to campaign contributors. Extreme fighting was tarnished when fighters from the other ultimate fighting operation, the now-defunct Battlecade, were arrested for violating Canadian prizefighting laws when they fought on an Indian reservation outside Montreal. In the past two years, an increasingly desperate UFC has been trying to assuage its critics. The competition, which had been gradually adding safety rules since the first fight, imposed even more. It institued rounds and a "10-point must" scoring system. It banned head butts and groin strikes. You can no longer kick a downed man or elbow someone in the back of the head. Fighters are required to wear thin martial arts gloves (a purely cosmetic change). The UFC imposed weight classes, ending the David-and-Goliath mismatches that made early fights so compelling. None of this soothed the cable operators, who have kept UFC off the air. The pay-per-view audience has plunged from 300,000 per show to 15,000. UFC can no longer afford its best fighters: Some are fighting overseas. Others, notably Ken Shamrock (Frank's brother), have become pro wrestlers. Fights have deteriorated. UFC is limping along, but it has been reduced to scheduling events in Japan and Brazil. "Sports fans want to grow with the sport," says former UFC fighter David Beneteau. "They want to recognize the athletes. They want to see the same fighters come back. When you compare UFC now to what it was, the fighters are not the same, the rules are not the same. The fans have no story to follow." Even as it disappears from public view, ultimate fighting is returning to its roots. Away from the scrutiny of the major media, state legislators, and McCain, kids are still learning mixed martial-arts techniques, and small-time promoters are quietly staging events. You can see Kage Kombat competitions at Dancing Waters nightclub in San Pedro, Calif. You can watch the Warrior's Challenge at a small Indian casino outside Sacramento. Texans compete in Houston's Dungal All Styles Fighting Championship. Tribal casinos in Northern Idaho are hosting small Pankration tournaments. The Extreme Fighting Challenge is popular in Iowa. The money is low; the crowds are small; and there's not a TV camera in sight. Ultimate fighting should have become boxing. Instead it has gone underground. It has become Fight Club.
|
D. Misconceptions about the safety of the sport drove political spats that kicked UFC out of the spotlight
|
What information is contained in the social graph of tweet authors?
|
### Introduction
Language identification is a crucial first step in textual data processing and is considered feasible over formal texts BIBREF0 . The task is harder for social media (e.g. Twitter) where text is less formal, noisier and can be written in wide range of languages. We focus on identifying similar languages, where surface-level content alone may not be sufficient. Our approach combines a content model with evidence propagated over the social network of the authors. For example, a user well-connected to users posting in a language is more likely to post in that language. Our system scores 76.63%, 1.4% higher than the top submission to the tweetLID workshop. ### Background
Traditional language identification compares a document with a language fingerprint built from n-gram bag-of-words (character or word level). Tweets carry additional metadata useful for identifying language, such as geolocation BIBREF1 , username BIBREF2 , BIBREF1 and urls mentioned in the tweet BIBREF2 . Other methods expand beyond the tweet itself to use a histogram of previously predicted languages, those of users @-mentioned and lexical content of other tweets in a discussion BIBREF1 . Discriminating between similar languages was the focus of the VarDial workshop BIBREF3 , and most submissions used content analysis. These methods make limited use of the social context in which the authors are tweeting – our research question is “Can we identify the language of a tweet using the social graph of the tweeter?”. Label propagation approaches BIBREF4 are powerful techniques for semi-supervised learning where the domain can naturally be described using an undirected graph. Each node contains a probability distribution over labels, which may be empty for unlabelled nodes, and these labels are propagated over the graph in an iterative fashion. Modified Adsorption (mad) BIBREF5 , is an extension that allows more control of the random walk through the graph. Applications of lp and mad are varied, including video recommendation BIBREF6 and sentiment analysis over Twitter BIBREF7 . ### Method
Our method predicts the language INLINEFORM0 for a tweet INLINEFORM1 by combining scores from a content model and a graph model that takes social context into account, as per Equation EQREF2 : DISPLAYFORM0 Where INLINEFORM0 are the content model parameters, INLINEFORM1 the social model parameters. ### Content model
Our content model is a 1 vs. all INLINEFORM0 regularised logistic regression model with character 2- to 5-grams features, not spanning over word boundaries. The scores for a tweet are normalised to obtain a probability distribution. ### Social model
We use a graph to model the social media context, relating tweets to one another, authors to tweets and other authors. Figure FIGREF7 shows the graph, composed of three types of nodes: tweets (T), users (U) and the “world” (W). Edges are created between nodes and weighted as follows: T-T the unigram cosine similarity between tweets, T-U weighted 100 between a tweet and its author, U-U weighted 1 between two users in a “follows” relationship and U-W weighted 0.001 to ensure a connected graph for the mad algorithm. We create the graph using all data, and training set tweets have an initial language label distribution. A naïve approach to building the tweet-tweet subgraph requires O( INLINEFORM0 ) comparisons, measuring the similarity of each tweet with all others. Instead, we performed INLINEFORM1 -nearest-neighbour classification on all tweets, represented as a bag of unigrams, and compared each tweet and the top- INLINEFORM2 neighbours. We use Junto (mad) BIBREF5 to propagate labels from labelled to unlabelled nodes. Upon convergence, we renormalise label scores for initially unlabelled nodes to find the value of INLINEFORM4 . ### Evaluation
The tweetLID workshop shared task requires systems to identify the language of tweets written in Spanish (es), Portuguese (pt), Catalan (ca), English (en), Galician (gl) and Basque (eu). Some language pairs are similar (es and ca; pt and gl) and this poses a challenge to systems that rely on content features alone. We use the supplied evaluation corpus, which has been manually labelled with six languages and evenly split into training and test collections. We use the official evaluation script and report precision, recall and F-score, macro-averaged across languages. This handles ambiguous tweets by permitting systems to return any of the annotated languages. Table TABREF10 shows that using the content model alone is more effective for languages that are distinct in our set of languages (i.e. English and Basque). For similar languages, adding the social model helps discriminate them (i.e. Spanish, Portuguese, Catalan and Galician), particularly those where a less-resourced language is similar to a more popular one. Using the social graph almost doubles the F-score for undecided (und) languages, either not in the set above or hard-to-identify, from 18.85% to 34.95%. Macro-averaged, our system scores 76.63%, higher than the best score in the competition: 75.2%. ### Conclusion
Our approach uses social information to help identify the language of tweets. This shows state-of-the-art performance, especially when discriminating between similar languages. A by-product of our approach is that users are assigned a language distribution, which may be useful for other tasks. Table 1: Experimental results. ♦/♠ are similar pairs. Figure 1: Graph topology. Rectangular nodes are tweets, circular nodes are users and the diamond represents the world. Some tweet nodes are labelled with an initial distribution over language labels and others are unlabelled.
|
the graph, composed of three types of nodes: tweets (T), users (U) and the “world” (W). Edges are created between nodes and weighted as follows: T-T the unigram cosine similarity between tweets, T-U weighted 100 between a tweet and its author, U-U weighted 1 between two users in a “follows” relationship and U-W weighted 0.001 to ensure a connected graph for the mad algorithm.
|
What datasets are used in this paper?
|
### Introduction
Semantic parsing aims to map natural language questions to the logical forms of their underlying meanings, which can be regarded as programs and executed to yield answers, aka denotations BIBREF0 . In the past few years, neural network based semantic parsers have achieved promising performances BIBREF1 , however, their success is limited to the setting with rich supervision, which is costly to obtain. There have been recent attempts at low-resource semantic parsing, including data augmentation methods which are learned from a small number of annotated examples BIBREF2 , and methods for adapting to unseen domains while only being trained on annotated examples in other domains. This work investigates neural semantic parsing in a low-resource setting, in which case we only have our prior knowledge about a limited number of simple mapping rules, including a small amount of domain-independent word-level matching tables if necessary, but have no access to either annotated programs or execution results. Our key idea is to use these rules to collect modest question-programs pairs as the starting point, and then leverage automatically generated examples to improve the accuracy and generality of the model. This presents three challenges including how to generate examples in an efficient way, how to measure the quality of generated examples which might contain errors and noise, and how to train a semantic parser that makes robust predictions for examples covered by rules and generalizes well to uncovered examples. We address the aforementioned challenges with a framework consisting of three key components. The first component is a data generator. It includes a neural semantic parsing model, which maps a natural language question to a program, and a neural question generation model, which maps a program to a natural language question. We learn these two models in a back-translation paradigm using pseudo parallel examples, inspired by its big success on unsupervised neural machine translation BIBREF3 , BIBREF4 . The second component is a quality controller, which is used for filtering out noise and errors contained in the pseudo data. We construct a phrase table with frequent mapping patterns, therefore noise and errors with low frequency can be filtered out. A similar idea has been worked as posterior regularization in neural machine translation BIBREF5 , BIBREF6 . The third component is a meta learner. Instead of transferring a model pretrained with examples covered by rules to the generated examples, we leverage model-agnostic meta-learning BIBREF7 , an elegant meta-learning algorithm which has been successfully applied to a wide range of tasks including few-shot learning and adaptive control. We regard different data sources as different tasks, and use outputs of the quality controller for stable training. We test our approach on three tasks with different programs, including SQL (and SQL-like) queries for both single-turn and multi-turn questions over web tables BIBREF8 , BIBREF9 , and subject-predicate pairs over a large-scale knowledge graph BIBREF10 . The program for SQL queries for single-turn questions and subject-predicate pairs over knowledge graph is simple while the program for SQL queries for multi-turn questions have top-tier complexity among currently proposed tasks. Results show that our approach yields large improvements over rule-based systems, and incorporating different strategies incrementally improves the overall performance. On WikiSQL, our best performing system achieves execution accuracy of 72.7%, comparable to a strong system learned from denotations BIBREF11 with an accuracy of 74.8%. ### Problem Statement
We focus on the task of executive semantic parsing. The goal is to map a natural language question/utterance INLINEFORM0 to a logical form/program INLINEFORM1 , which can be executed over a world INLINEFORM2 to obtain the correct answer INLINEFORM3 . We consider three tasks. The first task is single-turn table-based semantic parsing, in which case INLINEFORM0 is a self-contained question, INLINEFORM1 is a SQL query in the form of “SELECT agg col INLINEFORM2 WHERE col INLINEFORM3 = val INLINEFORM4 AND ...”, and INLINEFORM5 is a web table consisting of multiple rows and columns. We use WikiSQL BIBREF8 as the testbed for this task. The second task is multi-turn table-based semantic parsing. Compared to the first task, INLINEFORM6 could be a follow-up question, the meaning of which depends on the conversation history. Accordingly, INLINEFORM7 in this task supports additional operations that copy previous turn INLINEFORM8 to the current turn. We use SequentialQA BIBREF9 for evaluation. In the third task, we change INLINEFORM9 to a large-scale knowledge-graph (i.e. Freebase) and consider knowledge-based question answering for single-turn questions. We use SimpleQuestions BIBREF10 as the testbed, where the INLINEFORM10 is in the form of a simple INLINEFORM11 -calculus like INLINEFORM12 , and the generation of INLINEFORM13 is equivalent to the prediction of the predicate and the subject entity. We study the problem in a low-resource setting. In the training process, we don't have annotated logical forms INLINEFORM0 or execution results INLINEFORM1 . Instead, we have a collection of natural language questions for the task, a limited number of simple mapping rules based on our prior knowledge about the task, and may also have a small amount of domain-independent word-level matching tables if necessary. These rules are not perfect, with low coverage, and can even be incorrect for some situations. For instance, when predicting a SQL command in the first task, we have a prior knowledge that (1) WHERE values potentially have co-occurring words with table cells; (2) the words “more” and “greater” tend to be mapped to WHERE operator “ INLINEFORM2 ”; (3) within a WHERE clause, header and cell should be in the same column; and (4) the word “average” tends to be mapped to aggregator “avg”. Similarly, when predicting a INLINEFORM3 -calculus in the third task, the entity name might be present in the question, and among all the predicates connected to the entity, the predicate with maximum number of co-occurred words might be correct. We would like to study to what extent our model can achieve if we use rules as the starting point. ### Learning Algorithm
We describe our approach for low-resource neural semantic parsing in this section. We propose to train a neural semantic parser using back-translation and meta-learning. The learning process is summarized in Algorithm FIGREF1 . We describe the three components in this section, namely back-translation, quality control, and meta-learning. ### Back-Translation
Following the back-translation paradigm BIBREF3 , BIBREF4 , we have a semantic parser, which maps a natural language question INLINEFORM0 to a logical form INLINEFORM1 , and a question generator, which maps INLINEFORM2 to INLINEFORM3 . The semantic parser works for the primary task, and the question generator mainly works for generating pseudo datapoints. We start the training process by applying the rule INLINEFORM4 to a set of natural language questions INLINEFORM5 . The resulting dataset is considered as the training data to initialize both the semantic parser and the question generator. Afterwards, both models are improved following the back-translation protocol that target sequences should follow the real data distribution, yet source sequences can be generated with noises. This is based on the consideration that in an encoder-decoder model, the decoder is more sensitive to the data distribution than the encoder. We use datapoints from both models to train the semantic parser because a logical form is structural which follows a grammar, whose distribution is similar to the ground truth. ### Quality Controller
Directly using generated datapoints as supervised training data is not desirable because those generated datapoints contain noises or errors. To address this, we follow the application of posterior regularization in neural machine translation BIBREF5 , and implement a dictionary-based discriminator which is used to measure the quality of a pseudo data. The basic idea is that although these generated datapoints are not perfect, the frequent patterns of the mapping between a phrase in INLINEFORM0 to a token in INLINEFORM1 are helpful in filtering out noise in the generated data with low frequency BIBREF6 . There are multiple ways to collect the phrase table information, such as using statistical phrase-level alignment algorithms like Giza++ or directly counting the co-occurrence of any question word and logical form token. We use the latter one in this work. Further details are described in the appendix. ### Meta-Learning
A simple way to update the semantic parser is to merge the datapoints in hand and train a one-size-fits-all model BIBREF2 . However, this will hurt model's stability on examples covered by rules, and examples of the same task may vary widely BIBREF12 . Dealing with different types of examples requires the model to possess different abilities. For example, tackling examples uncovered by rules in WikiSQL requires the model to have the additional ability to map a column name to a totally different utterance, such as “country” to “nation”. Another simple solution is self-training BIBREF13 . One can train a model with examples covered by rules, and use the model as a teacher to make predictions on examples uncovered by rules and update the model on these predictions. However, self-training is somewhat tautological because the model is learned to make predictions which it already can produce. We learn the semantic parser with meta-learning, regarding learning from examples covered by rules or uncovered by rules as two (pseudo) tasks. Compared to the aforementioned strategies, the advantage of exploring meta-learning here is two-fold. First, we learn a specific model for each task, which provides guarantees about its stability on examples covered by rules. In the test phase, we can use the rule to detect which task an example belongs to, and use the corresponding task-specific model to make predictions. When dealing with examples covered by rules, we can either directly use rules to make predictions or use the updated model, depending on the accuracy of the learned model on the examples covered by rules on development set. Second, latent patterns of examples may vary widely in terms of whether or not they are covered by rules. Meta-learning is more desirable in this situation because it learns the model's ability to learn, improving model's versatility rather than mapping the latent patterns learned from datapoints in one distribution to datapoints in another distribution by force. Figure FIGREF1 is an illustration of data combination, self-training, and meta-learning. Meta-learning includes two optimizations: the learner that learns new tasks, and the meta-learner that trains the learner. In this work, the meta-learner is optimized by finding a good initialization that is highly adaptable. Specifically, we use model-agnostic meta-learning, MAML BIBREF7 , a powerful meta-learning algorithm with desirable properties including introducing no additional parameters and making no assumptions of the form of the model. In MAML, task-specific parameter INLINEFORM0 is initialized by INLINEFORM1 , and updated using gradient decent based on the loss function INLINEFORM2 of task INLINEFORM3 . In this work, the loss functions of two tasks are the same. The updated parameter INLINEFORM4 is then used to calculate the model's performance across tasks to update the parameter INLINEFORM5 . In this work, following the practical suggestions given by BIBREF17 , we update INLINEFORM6 in the inner-loop and regard the outputs of the quality controller as the input of both tasks. If we only have examples covered by rules, such as those used in the initialization phase, meta-learning learns to learn a good initial parameter that is evaluated by its usefulness on the examples from the same distribution. In the training phase, datapoints from both tasks are generated, and meta-learning learns to learn an initialization parameter which can be quickly and efficiently adapted to examples from both tasks. ### Experiment
We conduct experiments on three tasks to test our approach, including generating SQL (or SQL-like) queries for both single-turn and multi-turn questions over web tables BIBREF8 , BIBREF9 , and predicting subject-predicate pairs over a knowledge graph BIBREF10 . We describe task definition, base models, experiments settings and empirical results for each task, respectively. ### Table-Based Semantic Parsing
Given a natural language INLINEFORM0 and a table INLINEFORM1 with INLINEFORM2 columns and INLINEFORM3 rows as the input, the task is to output a SQL query INLINEFORM4 , which could be executed on table INLINEFORM5 to yield the correct answer of INLINEFORM6 . We conduct experiments on WikiSQL BIBREF8 , which provides 87,726 annotated question-SQL pairs over 26,375 web tables. In this work, we do not use either SQL queries or answers in the training process. We use execution accuracy as the evaluation metric, which measures the percentage of generated SQL queries that result in the correct answer. We describe our rules for WikiSQL here. We first detect WHERE values, which exactly match to table cells. After that, if a cell appears at more than one column, we choose the column name with more overlapped words with the question, with a constraint that the number of co-occurred words is larger than 1. By default, a WHERE operator is INLINEFORM0 , except for the case that surrounding words of a value contain keywords for INLINEFORM1 and INLINEFORM2 . Then, we deal with the SELECT column, which has the largest number of co-occurred words and cannot be same with any WHERE column. By default, the SELECT AGG is NONE, except for matching to any keywords in Table TABREF8 . The coverage of our rule on training set is 78.4%, with execution accuracy of 77.9%. We implement a neural network modular approach as the base model, which includes different modules to predict different SQL constituents. This approach is based on the understanding of the SQL grammar in WikiSQL, namely “SELECT $agg $column WHERE $column $op $value (AND $column $op $value)*”, where tokens starting with “$” are the slots to be predicted BIBREF18 . In practice, modular approaches typically achieve higher accuracy than end-to-end learning approach. Specifically, at the first step we implement a sequential labeling module to detect WHERE values and link them to table cells. Advantages of starting from WHERE values include that WHERE values are less ambiguous compared to other slots, and that the number of WHERE clauses can be naturally detected. After that, for each WHERE value, we use the preceding and following contexts in the question to predict its WHERE column and the WHERE operator through two unidirectional LSTM. Column attention BIBREF18 is used for predicting a particular column. Similar LSTM-based classifiers are used to predict SELECT column and SELECT aggregator. According to whether the training data can be processed by our rules, we divide it into two parts: rule covered part and rule uncovered part. For the rule covered part we could get rule covered training data using our rules. For the rule uncovered part we could also get training data using the trained Base model we have, we refer to these data as self-inference training data. Furthermore, we could get more training data by back translation, we refer to these data as question-generation training data. For all the settings, the Base Model is initialized with rule covered training data. In Base + Self Training Method, we finetune the Base model with self-inference training data. In Base + Question Generation Method, we use question-generation training data to finetune our model. In Base + BT Method, we use both self-inference and question-generation data to finetune our model. In Base + BT + QC, we add our quality controller. In Base + BT + QC + MAML, we further add meta-learning. Results are given in Table TABREF5 . We can see that back-translation, quality control and MAML incrementally improves the accuracy. Question generation is better than self-training here because the logical form in WikiSQL is relatively simple, so the distribution of the sampled logical forms is similar to the original one. In the back-translation setting, generated examples come from both self-training and the question generation model. The model performs better than rules on rule-covered examples, and improves the accuracy on uncovered examples. Figure FIGREF12 shows the learning curves of the COLUMN prediction model with or without using MAML. The model using MAML has a better starting point during training, which reflects the effectiveness of the pre-trained parameter. ### Knowledge-Based Question Answering
We test our approach on question answering over another genre of environment: knowledge graph consisting of subject-relation-object triples. Given a natural language question and a knowledge graph, the task aims to correctly answer the question with evidences from the knowledge graph. We do our study on SimpleQuestions BIBREF10 , which includes 108,442 simple questions, each of which is accompanied by a subject-relation-object triple. Questions are constructed in a way that subject and relation are mentioned in the question, and that object is the answer. The task requires predicting the entityId and the relation involved in the question. Our rule for KBQA is simple without using a curated mapping dictionary. First, we detect an entity from the question using strict string matching, with the constraint that only one entity from the KB has the same surface string and that the question contains only one entity. After that, we get the connected relations of the detected entity, and assign the relation as the one with maximum number of co-occurred words. The coverage of our rule on training set is 16.0%, with an accuracy of 97.3% for relation prediction. We follow BIBREF22 , and implement a KBQA pipeline consisting of three modules in this work. At the first step, we use a sequence labeling model, i.e. LSTM-CRF, to detect entity mention words in the question. After that, we use an entity linking model with BM25 built on Elasticsearch. Top-K ranked similar entities are retrieved as candidate list. Then, we get all the relations connected to entities in the candidate list as candidate relations, and use a relation prediction model, which is based on Match-LSTM BIBREF23 , to predict the relation. Finally, from all the entities connected to the predicted relation, we choose the one with highest BM25 score as the predicted entity. We use FB2M as the KB, which includes about 2 million triples. The settings are the same as those described in table-based semantic parsing. Results are given in Table TABREF10 , which are consistent with the numbers in WikiSQL. Using back-translation, quality control and MAML incrementally improves the accuracy, and our approach generalizes well to rule-uncovered examples. ### Conversational Table-Based Semantic Parsing
We consider the task of conversational table-based semantic parsing in this part. Compared to single-turn table-based semantic parsing as described in subsection SECREF6 , the meaning of a natural language may also depends on questions of past turns, which is the common ellipsis and co-reference phenomena in conversational agents. Given a natural language question at the current turn, a web table, and previous turn questions in a conversation as the input, the task aims to generate a program (i.e. logical form), which can be executed on the table to obtain the correct answer of the current turn question. We conduct experiments on SequentialQA BIBREF9 which is derived from the WikiTableQuestions dataset BIBREF19 . It contains 6,066 question sequences covering 17,553 question-answer pairs. Each sequence includes 2.9 natural language questions on average. Different from WikiSQL which provides the correct logical form for each question, SequentialQA only annotates the correct answer. This dataset is also harder than the previous two, since it requires complex, highly compositional logical forms to get the answer. Existing approaches are evaluated by question answering accuracy, which measures whether the predicted answer is correct or not. The pipeline of rules in SequentialQA is similar to that of WikiSQL. Compared to the grammar of WikiSQL, the grammar of SequentialQA has additional actions including copying the previous-turn logical form, no greater than, no more than, and negation. Table TABREF23 shows the additional word-level mapping table used in SequentialQA. The coverage of our rule on training set is 75.5%, with an accuracy of 38.5%. We implement a modular approach on top of a grammar of derivation rules (actions) as the base model. Similar to BIBREF9 , our grammar consists of predefined actions used for predicting SELECT column, WHERE column, WHERE operator, WHERE value, and determining whether it is required to copy the entire action sequence of the previous turn questions. After encoding a question and previous turn questions into vectors, we first use a controller module to predict an action sequence consisting of slots, and then use specific modules to predict the argument of each slot. Similar to BIBREF9 , we use a recurrent structure as the backbone of each module and use the softmax layer for making prediction. The settings are the same as those described in table-based semantic parsing. From Table TABREF20 , we can see that question generation does not work well on this task. This is because the difficulty in generating sequential questions and complex target logical forms. Applying MAML to examples not coming from question generation performs best. We leave contextual question generation as a future work. ### Conclusion and Future Directions
We present an approach to learn neural semantic parser from simple domain-independent rules, instead of annotated logical forms or denotations. Our approach starts from examples covered by rules, which are used to initialize a semantic parser and a question generator in a back-translation paradigm. Generated examples are measured and filtered based on statistic analysis, and then used with model-agnostic meta-learning, which guarantees model's accuracy and stability on rule-covered examples, and acquires the versatility to generalize well on rule-uncovered examples. We conduct experiments on three datasets for table-based and knowledge-based question answering tasks. Results show that incorporating different strategies incrementally improves the performance. Our best model on WikiSQL achieves comparable accuracy to the system learned from denotation. In the future, we plan to focus on more complex logical forms. Figure 1: An illustration of the difference between (a) data combination which learns a monolithic, one-size-fits-all model, (b) self-training which learns from predictions which the model produce and (c) meta-learning that reuse the acquired ability to learn. Table 1: Results on WikiSQL testset. BT stands for back-translation. QC stands for quality control. Table 2: Token-level dictionary for aggregators (upper group) and operators (lower group) in WikiSQL. Table 3: Results on SimpleQuestions testset. BT stands for back-translation. QC stands for quality control. Figure 2: Learning curve of the WHERE column prediction model on WikiSQL devset. Table 4: Results on SequentialQA testset. BT stands for back-translation. QC stands for quality control. Table 5: Token-level dictionary used for additional actions in SequentialQA.
|
WikiSQL, SimpleQuestions, SequentialQA
|
Were there any board member nominees who had substantially more votes against joining than the other nominees?
|
Evidence 0:
Proposal 1. With respect to the proposal to elect ten nominees to the Board of Directors (the Board), each for a one-year term expiring at the
annual meeting of shareholders to be held in 2023, the votes were cast for the proposal as set forth below:
Name
Votes For
Votes Against
Abstentions
Broker Non-Votes
Virginia C. Drosos
59,657,810
294,935
10,714,238
6,884,223
Alan D. Feldman
54,760,830
5,184,437
10,721,716
6,884,223
Richard A. Johnson
54,484,293
16,105,005
77,685
6,884,223
Guillermo G. Marmol
54,193,921
5,753,395
10,719,667
6,884,223
Darlene Nicosia
55,123,930
4,827,808
10,715,245
6,884,223
Steven Oakland
55,421,657
4,524,393
10,720,933
6,884,223
Ulice Payne, Jr.
54,993,396
4,950,917
10,722,670
6,884,223
Kimberly Underhill
55,046,260
4,906,500
10,714,223
6,884,223
Tristan Walker
55,528,794
4,419,340
10,718,849
6,884,223
Dona D. Young
53,876,257
6,074,467
10,716,259
6,884,223
Based on the votes set forth above, each of the ten nominees to the Board was duly elected.
|
Yes, his name is Richard A. Johnson
|
What task-specific features are used?
|
### Introduction
The increasing use of social media and microblogging services has broken new ground in the field of Information Extraction (IE) from user-generated content (UGC). Understanding the information contained in users' content has become one of the main goal for many applications, due to the uniqueness and the variety of this data BIBREF0 . However, the highly informal and noisy status of these sources makes it difficult to apply techniques proposed by the NLP community for dealing with formal and structured content BIBREF1 . In this work, we analyze a set of tweets related to a specific classical music radio channel, BBC Radio 3, interested in detecting two types of musical named entities, Contributor and Musical Work. The method proposed makes use of the information extracted from the radio schedule for creating links between users' tweets and tracks broadcasted. Thanks to this linking, we aim to detect when users refer to entities included into the schedule. Apart from that, we consider a series of linguistic features, partly taken from the NLP literature and partly specifically designed for this task, for building statistical models able to recognize the musical entities. To that aim, we perform several experiments with a supervised learning model, Support Vector Machine (SVM), and a recurrent neural network architecture, a bidirectional LSTM with a CRF layer (biLSTM-CRF). The contributions in this work are summarized as follows: The paper is structured as follows. In Section 2, we present a review of the previous works related to Named Entity Recognition, focusing on its application on UGC and MIR. Afterwards, in Section 3 it is presented the methodology of this work, describing the dataset and the method proposed. In Section 4, the results obtained are shown. Finally, in Section 5 conclusions are discussed. ### Related Work
Named Entity Recognition (NER), or alternatively Named Entity Recognition and Classification (NERC), is the task of detecting entities in an input text and to assign them to a specific class. It starts to be defined in the early '80, and over the years several approaches have been proposed BIBREF2 . Early systems were based on handcrafted rule-based algorithms, while recently several contributions by Machine Learning scientists have helped in integrating probabilistic models into NER systems. In particular, new developments in neural architectures have become an important resource for this task. Their main advantages are that they do not need language-specific knowledge resources BIBREF3 , and they are robust to the noisy and short nature of social media messages BIBREF4 . Indeed, according to a performance analysis of several Named Entity Recognition and Linking systems presented in BIBREF5 , it has been found that poor capitalization is one of the main issues when dealing with microblog content. Apart from that, typographic errors and the ubiquitous occurrence of out-of-vocabulary (OOV) words also cause drops in NER recall and precision, together with shortenings and slang, particularly pronounced in tweets. Music Information Retrieval (MIR) is an interdisciplinary field which borrows tools of several disciplines, such as signal processing, musicology, machine learning, psychology and many others, for extracting knowledge from musical objects (be them audio, texts, etc.) BIBREF6 . In the last decade, several MIR tasks have benefited from NLP, such as sound and music recommendation BIBREF7 , automatic summary of song review BIBREF8 , artist similarity BIBREF9 and genre classification BIBREF10 . In the field of IE, a first approach for detecting musical named entities from raw text, based on Hidden Markov Models, has been proposed in BIBREF11 . In BIBREF12 , the authors combine state-of-the-art Entity Linking (EL) systems to tackle the problem of detecting musical entities from raw texts. The method proposed relies on the argumentum ad populum intuition, so if two or more different EL systems perform the same prediction in linking a named entity mention, the more likely this prediction is to be correct. In detail, the off-the-shelf systems used are: DBpedia Spotlight BIBREF13 , TagMe BIBREF14 , Babelfy BIBREF15 . Moreover, a first Musical Entity Linking, MEL has been presented in BIBREF16 which combines different state-of-the-art NLP libraries and SimpleBrainz, an RDF knowledge base created from MusicBrainz after a simplification process. Furthermore, Twitter has also been at the center of many studies done by the MIR community. As example, for building a music recommender system BIBREF17 analyzes tweets containing keywords like nowplaying or listeningto. In BIBREF9 , a similar dataset it is used for discovering cultural listening patterns. Publicly available Twitter corpora built for MIR investigations have been created, among others the Million Musical Tweets dataset BIBREF18 and the #nowplaying dataset BIBREF19 . ### Methodology
We propose a hybrid method which recognizes musical entities in UGC using both contextual and linguistic information. We focus on detecting two types of entities: Contributor: person who is related to a musical work (composer, performer, conductor, etc). Musical Work: musical composition or recording (symphony, concerto, overture, etc). As case study, we have chosen to analyze tweets extracted from the channel of a classical music radio, BBC Radio 3. The choice to focus on classical music has been mostly motivated by the particular discrepancy between the informal language used in the social platform and the formal nomenclature of contributors and musical works. Indeed, users when referring to a musician or to a classical piece in a tweet, rarely use the full name of the person or of the work, as shown in Table 2. We extract information from the radio schedule for recreating the musical context to analyze user-generated tweets, detecting when they are referring to a specific work or contributor recently played. We manage to associate to every track broadcasted a list of entities, thanks to the tweets automatically posted by the BBC Radio3 Music Bot, where it is described the track actually on air in the radio. In Table 3, examples of bot-generated tweets are shown. Afterwards, we detect the entities on the user-generated content by means of two methods: on one side, we use the entities extracted from the radio schedule for generating candidates entities in the user-generated tweets, thanks to a matching algorithm based on time proximity and string similarity. On the other side, we create a statistical model capable of detecting entities directly from the UGC, aimed to model the informal language of the raw texts. In Figure 1, an overview of the system proposed is presented. ### Dataset
In May 2018, we crawled Twitter using the Python library Tweepy, creating two datasets on which Contributor and Musical Work entities have been manually annotated, using IOB tags. The first set contains user-generated tweets related to the BBC Radio 3 channel. It represents the source of user-generated content on which we aim to predict the named entities. We create it filtering the messages containing hashtags related to BBC Radio 3, such as #BBCRadio3 or #BBCR3. We obtain a set of 2,225 unique user-generated tweets. The second set consists of the messages automatically generated by the BBC Radio 3 Music Bot. This set contains 5,093 automatically generated tweets, thanks to which we have recreated the schedule. In Table 4, the amount of tokens and relative entities annotated are reported for the two datasets. For evaluation purposes, both sets are split in a training part (80%) and two test sets (10% each one) randomly chosen. Within the user-generated corpora, entities annotated are only about 5% of the whole amount of tokens. In the case of the automatically generated tweets, the percentage is significantly greater and entities represent about the 50%. ### NER system
According to the literature reviewed, state-of-the-art NER systems proposed by the NLP community are not tailored to detect musical entities in user-generated content. Consequently, our first objective has been to understand how to adapt existing systems for achieving significant results in this task. In the following sections, we describe separately the features, the word embeddings and the models considered. All the resources used are publicy available. We define a set of features for characterizing the text at the token level. We mix standard linguistic features, such as Part-Of-Speech (POS) and chunk tag, together with several gazetteers specifically built for classical music, and a series of features representing tokens' left and right context. For extracting the POS and the chunk tag we use the Python library twitter_nlp, presented in BIBREF1 . In total, we define 26 features for describing each token: 1)POS tag; 2)Chunk tag; 3)Position of the token within the text, normalized between 0 and 1; 4)If the token starts with a capital letter; 5)If the token is a digit. Gazetteers: 6)Contributor first names; 7)Contributor last names; 8)Contributor types ("soprano", "violinist", etc.); 9)Classical work types ("symphony", "overture", etc.); 10)Musical instruments; 11)Opus forms ("op", "opus"); 12)Work number forms ("no", "number"); 13)Work keys ("C", "D", "E", "F" , "G" , "A", "B", "flat", "sharp"); 14)Work Modes ("major", "minor", "m"). Finally, we complete the tokens' description including as token's features the surface form, the POS and the chunk tag of the previous and the following two tokens (12 features). We consider two sets of GloVe word embeddings BIBREF20 for training the neural architecture, one pre-trained with 2B of tweets, publicy downloadable, one trained with a corpora of 300K tweets collected during the 2014-2017 BBC Proms Festivals and disjoint from the data used in our experiments. The first model considered for this task has been the John Platt's sequential minimal optimization algorithm for training a support vector classifier BIBREF21 , implemented in WEKA BIBREF22 . Indeed, in BIBREF23 results shown that SVM outperforms other machine learning models, such as Decision Trees and Naive Bayes, obtaining the best accuracy when detecting named entities from the user-generated tweets. However, recent advances in Deep Learning techniques have shown that the NER task can benefit from the use of neural architectures, such as biLSTM-networks BIBREF3 , BIBREF4 . We use the implementation proposed in BIBREF24 for conducting three different experiments. In the first, we train the model using only the word embeddings as feature. In the second, together with the word embeddings we use the POS and chunk tag. In the third, all the features previously defined are included, in addition to the word embeddings. For every experiment, we use both the pre-trained embeddings and the ones that we created with our Twitter corpora. In section 4, results obtained from the several experiments are reported. ### Schedule matching
The bot-generated tweets present a predefined structure and a formal language, which facilitates the entities detection. In this dataset, our goal is to assign to each track played on the radio, represented by a tweet, a list of entities extracted from the tweet raw text. For achieving that, we experiment with the algorithms and features presented previously, obtaining an high level of accuracy, as presented in section 4. The hypothesis considered is that when a radio listener posts a tweet, it is possible that she is referring to a track which has been played a relatively short time before. In this cases, we want to show that knowing the radio schedule can help improving the results when detecting entities. Once assigned a list of entities to each track, we perform two types of matching. Firstly, within the tracks we identify the ones which have been played in a fixed range of time (t) before and after the generation of the user's tweet. Using the resulting tracks, we create a list of candidates entities on which performing string similarity. The score of the matching based on string similarity is computed as the ratio of the number of tokens in common between an entity and the input tweet, and the total number of token of the entity: DISPLAYFORM0 In order to exclude trivial matches, tokens within a list of stop words are not considered while performing string matching. The final score is a weighted combination of the string matching score and the time proximity of the track, aimed to enhance matches from tracks played closer to the time when the user is posting the tweet. The performance of the algorithm depends, apart from the time proximity threshold t, also on other two thresholds related to the string matching, one for the Musical Work (w) and one for the Contributor (c) entities. It has been necessary for avoiding to include candidate entities matched against the schedule with a low score, often source of false positives or negatives. Consequently, as last step Contributor and Musical Work candidates entities with respectively a string matching score lower than c and w, are filtered out. In Figure 2, an example of Musical Work entity recognized in an user-generated tweet using the schedule information is presented. The entities recognized from the schedule matching are joined with the ones obtained directly from the statistical models. In the joined results, the criteria is to give priority to the entities recognized from the machine learning techniques. If they do not return any entities, the entities predicted by the schedule matching are considered. Our strategy is justified by the poorer results obtained by the NER based only on the schedule matching, compared to the other models used in the experiments, to be presented in the next section. ### Results
The performances of the NER experiments are reported separately for three different parts of the system proposed. Table 6 presents the comparison of the various methods while performing NER on the bot-generated corpora and the user-generated corpora. Results shown that, in the first case, in the training set the F1 score is always greater than 97%, with a maximum of 99.65%. With both test sets performances decrease, varying between 94-97%. In the case of UGC, comparing the F1 score we can observe how performances significantly decrease. It can be considered a natural consequence of the complex nature of the users' informal language in comparison to the structured message created by the bot. In Table 7, results of the schedule matching are reported. We can observe how the quality of the linking performed by the algorithm is correlated to the choice of the three thresholds. Indeed, the Precision score increase when the time threshold decrease, admitting less candidates as entities during the matching, and when the string similarity thresholds increase, accepting only candidates with an higher degree of similarity. The behaviour of the Recall score is inverted. Finally, we test the impact of using the schedule matching together with a biLSTM-CRF network. In this experiment, we consider the network trained using all the features proposed, and the embeddings not pre-trained. Table 8 reports the results obtained. We can observe how generally the system benefits from the use of the schedule information. Especially in the testing part, where the neural network recognizes with less accuracy, the explicit information contained in the schedule can be exploited for identifying the entities at which users are referring while listening to the radio and posting the tweets. ### Conclusion
We have presented in this work a novel method for detecting musical entities from user-generated content, modelling linguistic features with statistical models and extracting contextual information from a radio schedule. We analyzed tweets related to a classical music radio station, integrating its schedule to connect users' messages to tracks broadcasted. We focus on the recognition of two kinds of entities related to the music field, Contributor and Musical Work. According to the results obtained, we have seen a pronounced difference between the system performances when dealing with the Contributor instead of the Musical Work entities. Indeed, the former type of entity has been shown to be more easily detected in comparison to the latter, and we identify several reasons behind this fact. Firstly, Contributor entities are less prone to be shorten or modified, while due to their longness, Musical Work entities often represent only a part of the complete title of a musical piece. Furthermore, Musical Work titles are typically composed by more tokens, including common words which can be easily misclassified. The low performances obtained in the case of Musical Work entities can be a consequences of these observations. On the other hand, when referring to a Contributor users often use only the surname, but in most of the cases it is enough for the system to recognizing the entities. From the experiments we have seen that generally the biLSTM-CRF architecture outperforms the SVM model. The benefit of using the whole set of features is evident in the training part, but while testing the inclusion of the features not always leads to better results. In addition, some of the features designed in our experiments are tailored to the case of classical music, hence they might not be representative if applied to other fields. We do not exclude that our method can be adapted for detecting other kinds of entity, but it might be needed to redefine the features according to the case considered. Similarly, it has not been found a particular advantage of using the pre-trained embeddings instead of the one trained with our corpora. Furthermore, we verified the statistical significance of our experiment by using Wilcoxon Rank-Sum Test, obtaining that there have been not significant difference between the various model considered while testing. The information extracted from the schedule also present several limitations. In fact, the hypothesis that a tweet is referring to a track broadcasted is not always verified. Even if it is common that radios listeners do comments about tracks played, or give suggestion to the radio host about what they would like to listen, it is also true that they might refer to a Contributor or Musical Work unrelated to the radio schedule. Table 2. Example of entities annotated and corresponding formal forms, from the user-generated tweet (1) in Table 1. Table 3. Examples of bot-generated tweets. Table 4. Tokens’ distributions within the two datasets: user-generated tweets (top) and bot-generated tweets (bottom) Fig. 2. Example of the workflow for recognizing entities in UGC using the information from the radio schedule Table 6. F1 score for Contributor(C) and Musical Work(MW) entities recognized from bot-generated tweets (top) and user-generated tweets (bottom) Table 7. Precision (P), Recall (R) and F1 score for Contributor (C) and Musical Work (MW) of the schedule matching algorithm. w indicates the Musical Work string similarity threshold, c indicates the Contributor string similarity threshold and t indicates the time proximity threshold in seconds Table 8. Precision (P), Recall (R) and F1 score for Contributor (C) and Musical Work (MW) entities recognized from user-generated tweets using the biLSTM-CRF network together with the schedule matching. The thresholds used for the matching are t=1200, w=0.5, c=0.5.
|
6)Contributor first names, 7)Contributor last names, 8)Contributor types ("soprano", "violinist", etc.), 9)Classical work types ("symphony", "overture", etc.), 10)Musical instruments, 11)Opus forms ("op", "opus"), 12)Work number forms ("no", "number"), 13)Work keys ("C", "D", "E", "F" , "G" , "A", "B", "flat", "sharp"), 14)Work Modes ("major", "minor", "m")
|
How did Pembroke get to Puerto Pacifico?
A. Pembroke traveled to Puerto Pacifico on a ship called the Elena Mia.
B. Pembroke traveled to Puerto Pacifico on a Colombian ship called The Valparaiso.
C. Pembroke arrived in Puerto Pacifico on the lifeboat he used to escape the sinking ship.
D. The aliens placed Pembroke in Puerto Pacifico after destroying his ship.
|
THE PERFECTIONISTS By ARNOLD CASTLE ILLUSTRATED by SUMMERS Is there something wrong with you? Do you fail to fit in with your group? Nervous, anxious, ill-at-ease? Happy about it? Lucky you! Frank Pembroke sat behind the desk of his shabby little office over Lemark's Liquors in downtown Los Angeles and waited for his first customer. He had been in business for a week and as yet had had no callers. Therefore, it was with a mingled sense of excitement and satisfaction that he greeted the tall, dark, smooth-faced figure that came up the stairs and into the office shortly before noon. "Good day, sir," said Pembroke with an amiable smile. "I see my advertisement has interested you. Please stand in that corner for just a moment." Opening the desk drawer, which was almost empty, Pembroke removed an automatic pistol fitted with a silencer. Pointing it at the amazed customer, he fired four .22 caliber longs into the narrow chest. Then he made a telephone call and sat down to wait. He wondered how long it would be before his next client would arrive. The series of events leading up to Pembroke's present occupation had commenced on a dismal, overcast evening in the South Pacific a year earlier. Bound for Sydney, two days out of Valparaiso, the Colombian tramp steamer Elena Mia had encountered a dense greenish fog which seemed vaguely redolent of citrus trees. Standing on the forward deck, Pembroke was one of the first to perceive the peculiar odor and to spot the immense gray hulk wallowing in the murky distance. Then the explosion had come, from far below the waterline, and the decks were awash with frantic crewmen, officers, and the handful of passengers. Only two lifeboats were launched before the Elena Mia went down. Pembroke was in the second. The roar of the sinking ship was the last thing he heard for some time. Pembroke came as close to being a professional adventurer as one can in these days of regimented travel, organized peril, and political restriction. He had made for himself a substantial fortune through speculation in a great variety of properties, real and otherwise. Life had given him much and demanded little, which was perhaps the reason for his restiveness. Loyalty to person or to people was a trait Pembroke had never recognized in himself, nor had it ever been expected of him. And yet he greatly envied those staunch patriots and lovers who could find it in themselves to elevate the glory and safety of others above that of themselves. Lacking such loyalties, Pembroke adapted quickly to the situation in which he found himself when he regained consciousness. He awoke in a small room in what appeared to be a typical modern American hotel. The wallet in his pocket contained exactly what it should, approximately three hundred dollars. His next thought was of food. He left the room and descended via the elevator to the restaurant. Here he observed that it was early afternoon. Ordering a full dinner, for he was unusually hungry, he began to study the others in the restaurant. Many of the faces seemed familiar; the crew of the ship, probably. He also recognized several of the passengers. However, he made no attempt to speak to them. After his meal, he bought a good corona and went for a walk. His situation could have been any small western American seacoast city. He heard the hiss of the ocean in the direction the afternoon sun was taking. In his full-gaited walk, he was soon approaching the beach. On the sand he saw a number of sun bathers. One in particular, an attractive woman of about thirty, tossed back her long, chestnut locks and gazed up intently at Pembroke as he passed. Seldom had he enjoyed so ingenuous an invitation. He halted and stared down at her for a few moments. "You are looking for someone?" she inquired. "Much of the time," said the man. "Could it be me?" "It could be." "Yet you seem unsure," she said. Pembroke smiled, uneasily. There was something not entirely normal about her conversation. Though the rest of her compensated for that. "Tell me what's wrong with me," she went on urgently. "I'm not good enough, am I? I mean, there's something wrong with the way I look or act. Isn't there? Please help me, please!" "You're not casual enough, for one thing," said Pembroke, deciding to play along with her for the moment. "You're too tense. Also you're a bit knock-kneed, not that it matters. Is that what you wanted to hear?" "Yes, yes—I mean, I suppose so. I can try to be more casual. But I don't know what to do about my knees," she said wistfully, staring across at the smooth, tan limbs. "Do you think I'm okay otherwise? I mean, as a whole I'm not so bad, am I? Oh, please tell me." "How about talking it over at supper tonight?" Pembroke proposed. "Maybe with less distraction I'll have a better picture of you—as a whole." "Oh, that's very generous of you," the woman told him. She scribbled a name and an address on a small piece of paper and handed it to him. "Any time after six," she said. Pembroke left the beach and walked through several small specialty shops. He tried to get the woman off his mind, but the oddness of her conversation continued to bother him. She was right about being different, but it was her concern about being different that made her so. How to explain that to her? Then he saw the weird little glass statuette among the usual bric-a-brac. It rather resembled a ground hog, had seven fingers on each of its six limbs, and smiled up at him as he stared. "Can I help you, sir?" a middle-aged saleswoman inquired. "Oh, good heavens, whatever is that thing doing here?" Pembroke watched with lifted eyebrows as the clerk whisked the bizarre statuette underneath the counter. "What the hell was that?" Pembroke demanded. "Oh, you know—or don't you? Oh, my," she concluded, "are you one of the—strangers?" "And if I were?" "Well, I'd certainly appreciate it if you'd tell me how I walk." She came around in front of the counter and strutted back and forth a few times. "They tell me I lean too far forward," she confided. "But I should think you'd fall down if you didn't." "Don't try to go so fast and you won't fall down," suggested Pembroke. "You're in too much of a hurry. Also those fake flowers on your blouse make you look frumpy." "Well, I'm supposed to look frumpy," the woman retorted. "That's the type of person I am. But you can look frumpy and still walk natural, can't you? Everyone says you can." "Well, they've got a point," said Pembroke. "Incidentally, just where are we, anyway? What city is this?" "Puerto Pacifico," she told him. "Isn't that a lovely name? It means peaceful port. In Spanish." That was fine. At least he now knew where he was. But as he left the shop he began checking off every west coast state, city, town, and inlet. None, to the best of his knowledge, was called Puerto Pacifico. He headed for the nearest service station and asked for a map. The attendant gave him one which showed the city, but nothing beyond. "Which way is it to San Francisco?" asked Pembroke. "That all depends on where you are," the boy returned. "Okay, then where am I?" "Pardon me, there's a customer," the boy said. "This is Puerto Pacifico." Pembroke watched him hurry off to service a car with a sense of having been given the runaround. To his surprise, the boy came back a few minutes later after servicing the automobile. "Say, I've just figured out who you are," the youngster told him. "I'd sure appreciate it if you'd give me a little help on my lingo. Also, you gas up the car first, then try to sell 'em the oil—right?" "Right," said Pembroke wearily. "What's wrong with your lingo? Other than the fact that it's not colloquial enough." "Not enough slang, huh? Well, I guess I'll have to concentrate on that. How about the smile?" "Perfect," Pembroke told him. "Yeah?" said the boy delightedly. "Say, come back again, huh? I sure appreciate the help. Keep the map." "Thanks. One more thing," Pembroke said. "What's over that way—outside the city?" "Sand." "How about that way?" he asked, pointing north. "And that way?" pointing south. "More of the same." "Any railroads?" "That we ain't got." "Buses? Airlines?" The kid shook his head. "Some city." "Yeah, it's kinda isolated. A lot of ships dock here, though." "All cargo ships, I'll bet. No passengers," said Pembroke. "Right," said the attendant, giving with his perfect smile. "No getting out of here, is there?" "That's for sure," the boy said, walking away to wait on another customer. "If you don't like the place, you've had it." Pembroke returned to the hotel. Going to the bar, he recognized one of the Elena Mia's paying passengers. He was a short, rectangular little man in his fifties named Spencer. He sat in a booth with three young women, all lovely, all effusive. The topic of the conversation turned out to be precisely what Pembroke had predicted. "Well, Louisa, I'd say your only fault is the way you keep wigglin' your shoulders up 'n' down. Why'n'sha try holdin' 'em straight?" "I thought it made me look sexy," the redhead said petulantly. "Just be yourself, gal," Spencer drawled, jabbing her intimately with a fat elbow, "and you'll qualify." "Me, me," the blonde with a feather cut was insisting. "What is wrong with me?" "You're perfect, sweetheart," he told her, taking her hand. "Ah, come on," she pleaded. "Everyone tells me I chew gum with my mouth open. Don't you hate that?" "Naw, that's part of your charm," Spencer assured her. "How 'bout me, sugar," asked the girl with the coal black hair. "Ah, you're perfect, too. You are all perfect. I've never seen such a collection of dolls as parade around this here city. C'mon, kids—how 'bout another round?" But the dolls had apparently lost interest in him. They got up one by one and walked out of the bar. Pembroke took his rum and tonic and moved over to Spencer's booth. "Okay if I join you?" "Sure," said the fat man. "Wonder what the hell got into those babes?" "You said they were perfect. They know they're not. You've got to be rough with them in this town," said Pembroke. "That's all they want from us." "Mister, you've been doing some thinkin', I can see," said Spencer, peering at him suspiciously. "Maybe you've figured out where we are." "Your bet's as good as mine," said Pembroke. "It's not Wellington, and it's not Brisbane, and it's not Long Beach, and it's not Tahiti. There are a lot of places it's not. But where the hell it is, you tell me. "And, by the way," he added, "I hope you like it in Puerto Pacifico. Because there isn't any place to go from here and there isn't any way to get there if there were." "Pardon me, gentlemen, but I'm Joe Valencia, manager of the hotel. I would be very grateful if you would give me a few minutes of honest criticism." "Ah, no, not you, too," groaned Spencer. "Look, Joe, what's the gag?" "You are newcomers, Mr. Spencer," Valencia explained. "You are therefore in an excellent position to point out our faults as you see them." "Well, so what?" demanded Spencer. "I've got more important things to do than to worry about your troubles. You look okay to me." "Mr. Valencia," said Pembroke. "I've noticed that you walk with a very slight limp. If you have a bad leg, I should think you would do better to develop a more pronounced limp. Otherwise, you may appear to be self-conscious about it." Spencer opened his mouth to protest, but saw with amazement that it was exactly this that Valencia was seeking. Pembroke was amused at his companion's reaction but observed that Spencer still failed to see the point. "Also, there is a certain effeminateness in the way in which you speak," said Pembroke. "Try to be a little more direct, a little more brusque. Speak in a monotone. It will make you more acceptable." "Thank you so much," said the manager. "There is much food for thought in what you have said, Mr. Pembroke. However, Mr. Spencer, your value has failed to prove itself. You have only yourself to blame. Cooperation is all we require of you." Valencia left. Spencer ordered another martini. Neither he nor Pembroke spoke for several minutes. "Somebody's crazy around here," the fat man muttered after a few moments. "Is it me, Frank?" "No. You just don't belong here, in this particular place," said Pembroke thoughtfully. "You're the wrong type. But they couldn't know that ahead of time. The way they operate it's a pretty hit-or-miss operation. But they don't care one bit about us, Spencer. Consider the men who went down with the ship. That was just part of the game." "What the hell are you sayin'?" asked Spencer in disbelief. "You figure they sunk the ship? Valencia and the waitress and the three babes? Ah, come on." "It's what you think that will determine what you do, Spencer. I suggest you change your attitude; play along with them for a few days till the picture becomes a little clearer to you. We'll talk about it again then." Pembroke rose and started out of the bar. A policeman entered and walked directly to Spencer's table. Loitering at the juke box, Pembroke overheard the conversation. "You Spencer?" "That's right," said the fat man sullenly. "What don't you like about me? The truth , buddy." "Ah, hell! Nothin' wrong with you at all, and nothin'll make me say there is," said Spencer. "You're the guy, all right. Too bad, Mac," said the cop. Pembroke heard the shots as he strolled casually out into the brightness of the hotel lobby. While he waited for the elevator, he saw them carrying the body into the street. How many others, he wondered, had gone out on their backs during their first day in Puerto Pacifico? Pembroke shaved, showered, and put on the new suit and shirt he had bought. Then he took Mary Ann, the woman he had met on the beach, out to dinner. She would look magnificent even when fully clothed, he decided, and the pale chartreuse gown she wore hardly placed her in that category. Her conversation seemed considerably more normal after the other denizens of Puerto Pacifico Pembroke had listened to that afternoon. After eating they danced for an hour, had a few more drinks, then went to Pembroke's room. He still knew nothing about her and had almost exhausted his critical capabilities, but not once had she become annoyed with him. She seemed to devour every factual point of imperfection about herself that Pembroke brought to her attention. And, fantastically enough, she actually appeared to have overcome every little imperfection he had been able to communicate to her. It was in the privacy of his room that Pembroke became aware of just how perfect, physically, Mary Ann was. Too perfect. No freckles or moles anywhere on the visible surface of her brown skin, which was more than a mere sampling. Furthermore, her face and body were meticulously symmetrical. And she seemed to be wholly ambidextrous. "With so many beautiful women in Puerto Pacifico," said Pembroke probingly, "I find it hard to understand why there are so few children." "Yes, children are decorative, aren't they," said Mary Ann. "I do wish there were more of them." "Why not have a couple of your own?" he asked. "Oh, they're only given to maternal types. I'd never get one. Anyway, I won't ever marry," she said. "I'm the paramour type." It was obvious that the liquor had been having some effect. Either that, or she had a basic flaw of loquacity that no one else had discovered. Pembroke decided he would have to cover his tracks carefully. "What type am I?" he asked. "Silly, you're real. You're not a type at all." "Mary Ann, I love you very much," Pembroke murmured, gambling everything on this one throw. "When you go to Earth I'll miss you terribly." "Oh, but you'll be dead by then," she pouted. "So I mustn't fall in love with you. I don't want to be miserable." "If I pretended I was one of you, if I left on the boat with you, they'd let me go to Earth with you. Wouldn't they?" "Oh, yes, I'm sure they would." "Mary Ann, you have two other flaws I feel I should mention." "Yes? Please tell me." "In the first place," said Pembroke, "you should be willing to fall in love with me even if it will eventually make you unhappy. How can you be the paramour type if you refuse to fall in love foolishly? And when you have fallen in love, you should be very loyal." "I'll try," she said unsurely. "What else?" "The other thing is that, as my mistress, you must never mention me to anyone. It would place me in great danger." "I'll never tell anyone anything about you," she promised. "Now try to love me," Pembroke said, drawing her into his arms and kissing with little pleasure the smooth, warm perfection of her tanned cheeks. "Love me my sweet, beautiful, affectionate Mary Ann. My paramour." Making love to Mary Ann was something short of ecstasy. Not for any obvious reason, but because of subtle little factors that make a woman a woman. Mary Ann had no pulse. Mary Ann did not perspire. Mary Ann did not fatigue gradually but all at once. Mary Ann breathed regularly under all circumstances. Mary Ann talked and talked and talked. But then, Mary Ann was not a human being. When she left the hotel at midnight, Pembroke was quite sure that she understood his plan and that she was irrevocably in love with him. Tomorrow might bring his death, but it might also ensure his escape. After forty-two years of searching for a passion, for a cause, for a loyalty, Frank Pembroke had at last found his. Earth and the human race that peopled it. And Mary Ann would help him to save it. The next morning Pembroke talked to Valencia about hunting. He said that he planned to go shooting out on the desert which surrounded the city. Valencia told him that there were no living creatures anywhere but in the city. Pembroke said he was going out anyway. He picked up Mary Ann at her apartment and together they went to a sporting goods store. As he guessed there was a goodly selection of firearms, despite the fact that there was nothing to hunt and only a single target range within the city. Everything, of course, had to be just like Earth. That, after all, was the purpose of Puerto Pacifico. By noon they had rented a jeep and were well away from the city. Pembroke and Mary Ann took turns firing at the paper targets they had purchased. At twilight they headed back to the city. On the outskirts, where the sand and soil were mixed and no footprints would be left, Pembroke hopped off. Mary Ann would go straight to the police and report that Pembroke had attacked her and that she had shot him. If necessary, she would conduct the authorities to the place where they had been target shooting, but would be unable to locate the spot where she had buried the body. Why had she buried it? Because at first she was not going to report the incident. She was frightened. It was not airtight, but there would probably be no further investigation. And they certainly would not prosecute Mary Ann for killing an Earthman. Now Pembroke had himself to worry about. The first step was to enter smoothly into the new life he had planned. It wouldn't be so comfortable as the previous one, but should be considerably safer. He headed slowly for the "old" part of town, aging his clothes against buildings and fences as he walked. He had already torn the collar of the shirt and discarded his belt. By morning his beard would grow to blacken his face. And he would look weary and hungry and aimless. Only the last would be a deception. Two weeks later Pembroke phoned Mary Ann. The police had accepted her story without even checking. And when, when would she be seeing him again? He had aroused her passion and no amount of long-distance love could requite it. Soon, he assured her, soon. "Because, after all, you do owe me something," she added. And that was bad because it sounded as if she had been giving some womanly thought to the situation. A little more of that and she might go to the police again, this time for vengeance. Twice during his wanderings Pembroke had seen the corpses of Earthmen being carted out of buildings. They had to be Earthmen because they bled. Mary Ann had admitted that she did not. There would be very few Earthmen left in Puerto Pacifico, and it would be simple enough to locate him if he were reported as being on the loose. There was no out but to do away with Mary Ann. Pembroke headed for the beach. He knew she invariably went there in the afternoon. He loitered around the stalls where hot dogs and soft drinks were sold, leaning against a post in the hot sun, hat pulled down over his forehead. Then he noticed that people all about him were talking excitedly. They were discussing a ship. It was leaving that afternoon. Anyone who could pass the interview would be sent to Earth. Pembroke had visited the docks every day, without being able to learn when the great exodus would take place. Yet he was certain the first lap would be by water rather than by spaceship, since no one he had talked to in the city had ever heard of spaceships. In fact, they knew very little about their masters. Now the ship had arrived and was to leave shortly. If there was any but the most superficial examination, Pembroke would no doubt be discovered and exterminated. But since no one seemed concerned about anything but his own speech and behavior, he assumed that they had all qualified in every other respect. The reason for transporting Earth People to this planet was, of course, to apply a corrective to any of the Pacificos' aberrant mannerisms or articulation. This was the polishing up phase. Pembroke began hobbling toward the docks. Almost at once he found himself face to face with Mary Ann. She smiled happily when she recognized him. That was a good thing. "It is a sign of poor breeding to smile at tramps," Pembroke admonished her in a whisper. "Walk on ahead." She obeyed. He followed. The crowd grew thicker. They neared the docks and Pembroke saw that there were now set up on the roped-off wharves small interviewing booths. When it was their turn, he and Mary Ann each went into separate ones. Pembroke found himself alone in the little room. Then he saw that there was another entity in his presence confined beneath a glass dome. It looked rather like a groundhog and had seven fingers on each of its six limbs. But it was larger and hairier than the glass one he had seen at the gift store. With four of its limbs it tapped on an intricate keyboard in front of it. "What is your name?" queried a metallic voice from a speaker on the wall. "I'm Jerry Newton. Got no middle initial," Pembroke said in a surly voice. "Occupation?" "I work a lot o' trades. Fisherman, fruit picker, fightin' range fires, vineyards, car washer. Anything. You name it. Been out of work for a long time now, though. Goin' on five months. These here are hard times, no matter what they say." "What do you think of the Chinese situation?" the voice inquired. "Which situation's 'at?" "Where's Seattle?" "Seattle? State o' Washington." And so it went for about five minutes. Then he was told he had qualified as a satisfactory surrogate for a mid-twentieth century American male, itinerant type. "You understand your mission, Newton?" the voice asked. "You are to establish yourself on Earth. In time you will receive instructions. Then you will attack. You will not see us, your masters, again until the atmosphere has been sufficiently chlorinated. In the meantime, serve us well." He stumbled out toward the docks, then looked about for Mary Ann. He saw her at last behind the ropes, her lovely face in tears. Then she saw him. Waving frantically, she called his name several times. Pembroke mingled with the crowd moving toward the ship, ignoring her. But still the woman persisted in her shouting. Sidling up to a well-dressed man-about-town type, Pembroke winked at him and snickered. "You Frank?" he asked. "Hell, no. But some poor punk's sure red in the face, I'll bet," the man-about-town said with a chuckle. "Those high-strung paramour types always raising a ruckus. They never do pass the interview. Don't know why they even make 'em." Suddenly Mary Ann was quiet. "Ambulance squad," Pembroke's companion explained. "They'll take her off to the buggy house for a few days and bring her out fresh and ignorant as the day she was assembled. Don't know why they keep making 'em, as I say. But I guess there's a call for that type up there on Earth." "Yeah, I reckon there is at that," said Pembroke, snickering again as he moved away from the other. "And why not? Hey? Why not?" Pembroke went right on hating himself, however, till the night he was deposited in a field outside of Ensenada, broke but happy, with two other itinerant types. They separated in San Diego, and it was not long before Pembroke was explaining to the police how he had drifted far from the scene of the sinking of the Elena Mia on a piece of wreckage, and had been picked up by a Chilean trawler. How he had then made his way, with much suffering, up the coast to California. Two days later, his identity established and his circumstances again solvent, he was headed for Los Angeles to begin his save-Earth campaign. Now, seated at his battered desk in the shabby rented office over Lemark's Liquors, Pembroke gazed without emotion at the two demolished Pacificos that lay sprawled one atop the other in the corner. His watch said one-fifteen. The man from the FBI should arrive soon. There were footsteps on the stairs for the third time that day. Not the brisk, efficient steps of a federal official, but the hesitant, self-conscious steps of a junior clerk type. Pembroke rose as the young man appeared at the door. His face was smooth, unpimpled, clean-shaven, without sweat on a warm summer afternoon. "Are you Dr. Von Schubert?" the newcomer asked, peering into the room. "You see, I've got a problem—" The four shots from Pembroke's pistol solved his problem effectively. Pembroke tossed his third victim onto the pile, then opened a can of lager, quaffing it appreciatively. Seating himself once more, he leaned back in the chair, both feet upon the desk. He would be out of business soon, once the FBI agent had got there. Pembroke was only in it to get the proof he would need to convince people of the truth of his tale. But in the meantime he allowed himself to admire the clipping of the newspaper ad he had run in all the Los Angeles papers for the past week. The little ad that had saved mankind from God-knew-what insidious menace. It read: ARE YOU IMPERFECT? LET DR. VON SCHUBERT POINT OUT YOUR FLAWS IT IS HIS GOAL TO MAKE YOU THE AVERAGE FOR YOUR TYPE FEE—$3.75 MONEY BACK IF NOT SATISFIED! THE END Transcriber's Note: This etext was produced from Amazing Science Fiction Stories January 1960. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
D. The aliens placed Pembroke in Puerto Pacifico after destroying his ship.
|
Where will Arvid 6 and Tendal 13 go after the end of the text?
A. To go back 6,000 years to re-attempt a Kanad recovery mission
B. To return to the Laughton's home in order to alter the crime scene
C. To travel to the Ultroom for Arvid 6 to face his consequences
D. To steal Phullam from his parents and get closer to recovering Kanad.
|
Transcriber's Note: This etext was produced from Space Science Fiction May 1952. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. THE ULTROOM ERROR by JERRY SOHL Smith admitted he had made an error involving a few murders—and a few thousand years. He was entitled to a sense of humor, though, even in the Ultroom! HB73782. Ultroom error. Tendal 13. Arvid 6. Kanad transfer out of 1609 complete, intact, but too near limit of 1,000 days. Next Kanad transfer ready. 1951. Reginald, son of Mr. and Mrs. Martin Laughton, 3495 Orland Drive, Marionville, Illinois, U. S. A. Arrive his 378th day. TB73782. Nancy Laughton sat on the blanket she had spread on the lawn in her front yard, knitting a pair of booties for the PTA bazaar. Occasionally she glanced at her son in the play pen, who was getting his daily dose of sunshine. He was gurgling happily, examining a ball, a cheese grater and a linen baby book, all with perfunctory interest. When she looked up again she noticed a man walking by—except he turned up the walk and crossed the lawn to her. He was a little taller than her husband, had piercing blue eyes and a rather amused set to his lips. "Hello, Nancy," he said. "Hello, Joe," she answered. It was her brother who lived in Kankakee. "I'm going to take the baby for a while," he said. "All right, Joe." He reached into the pen, picked up the baby. As he did so the baby's knees hit the side of the play pen and young Laughton let out a scream—half from hurt and half from sudden lack of confidence in his new handler. But this did not deter Joe. He started off with the child. Around the corner and after the man came a snarling mongrel dog, eyes bright, teeth glinting in the sunlight. The man did not turn as the dog threw himself at him, burying his teeth in his leg. Surprised, the man dropped the screaming child on the lawn and turned to the dog. Joe seemed off balance and he backed up confusedly in the face of the snapping jaws. Then he suddenly turned and walked away, the dog at his heels. "I tell you, the man said he was my brother and he made me think he was," Nancy told her husband for the tenth time. "I don't even have a brother." Martin Laughton sighed. "I can't understand why you believed him. It's just—just plain nuts, Nancy!" "Don't you think I know it?" Nancy said tearfully. "I feel like I'm going crazy. I can't say I dreamt it because there was Reggie with his bleeding knees, squalling for all he was worth on the grass—Oh, I don't even want to think about it." "We haven't lost Reggie, Nancy, remember that. Now why don't you try to get some rest?" "You—you don't believe me at all, do you, Martin?" When her husband did not answer, her head sank to her arms on the table and she sobbed. "Nancy, for heaven's sake, of course I believe you. I'm trying to think it out, that's all. We should have called the police." Nancy shook her head in her arms. "They'd—never—believe me either," she moaned. "I'd better go and make sure Reggie's all right." Martin got up out of his chair and went to the stairs. "I'm going with you," Nancy said, hurriedly rising and coming over to him. "We'll go up and look at him together." They found Reggie peacefully asleep in his crib in his room upstairs. They checked the windows and tucked in the blankets. They paused in the room for a moment and then Martin stole his arm around his wife and led her to the door. "As I've said, sergeant, this fellow hypnotized my wife. He made her think he was her brother. She doesn't even have a brother. Then he tried to get away with the baby." Martin leaned down and patted the dog. "It was Tiger here who scared him off." The police sergeant looked at the father, at Nancy and then at the dog. He scribbled notes in his book. "Are you a rich man, Mr. Laughton?" he asked. "Not at all. The bank still owns most of the house. I have a few hundred dollars, that's all." "What do you do?" "Office work, mostly. I'm a junior executive in an insurance company." "Any enemies?" "No ... Oh, I suppose I have a few people I don't get along with, like anybody else. Nobody who'd do anything like this, though." The sergeant flipped his notebook closed. "You'd better keep your dog inside and around the kid as much as possible. Keep your doors and windows locked. I'll see that the prowl car keeps an eye on the house. Call us if anything seems unusual or out of the way." Nancy had taken a sedative and was asleep by the time Martin finished cleaning the .30-.30 rifle he used for deer hunting. He put it by the stairs, ready for use, fully loaded, leaning it against the wall next to the telephone stand. The front door bell rang. He answered it. It was Dr. Stuart and another man. "I came as soon as I could, Martin," the young doctor said, stepping inside with the other man. "This is my new assistant, Dr. Tompkins." Martin and Tompkins shook hands. "The baby—?" Dr. Stuart asked. "Upstairs," Martin said. "You'd better get him, Dr. Tompkins, if we're to take him to the hospital. I'll stay here with Mr. Laughton. How've you been, Martin?" "Fine." "How's everything at the office?" "Fine." "And your wife?" "She's fine, too." "Glad to hear it, Martin. Mighty glad. Say, by the way, there's that bill you owe me. I think it's $32, isn't that right?" "Yes, I'd almost forgotten about it." "Why don't you be a good fellow and write a check for it? It's been over a year, you know." "That's right. I'll get right at it." Martin went over to his desk, opened it and started looking for his checkbook. Dr. Stuart stood by him, making idle comment until Dr. Tompkins came down the stairs with the sleeping baby cuddled against his shoulder. "Never mind the check, now, Martin. I see we're ready to go." He went over to his assistant and took the baby. Together they walked out the front door. "Good-bye," Martin said, going to the door. Then he was nearly bowled over by the discharge of the .30-.30. Dr. Stuart crumpled to the ground, the baby falling to the lawn. Dr. Tompkins whirled and there was a second shot. Dr. Tompkins pitched forward on his face. The figure of a woman ran from the house, retrieved the now squalling infant and ran back into the house. Once inside, Nancy slammed the door, gave the baby to the stunned Martin and headed for the telephone. "One of them was the same man!" she cried. Martin gasped, sinking into a chair with the baby. "I believed them," he said slowly and uncomprehendingly. "They made me believe them!" "Those bodies," the sergeant said. "Would you mind pointing them out to me, please?" "Aren't they—aren't they on the walk?" Mrs. Laughton asked. "There is nothing on the walk, Mrs. Laughton." "But there must be! I tell you I shot these men who posed as doctors. One of them was the same man who tried to take the baby this afternoon. They hypnotized my husband—" "Yes, I know, Mrs. Laughton. We've been through that." The sergeant went to the door and opened it. "Say, Homer, take another look around the walk and the bushes. There's supposed to be two of them. Shot with a .30-.30." He turned and picked up the gun and examined it again. "Ever shoot a gun before, Mrs. Laughton?" "Many times. Martin and I used to go hunting together before we had Reggie." The sergeant nodded. "You were taking an awful chance, shooting at a guy carrying your baby, don't you think?" "I shot him in the legs. The other—the other turned and I shot him in the chest. I could even see his eyes when he turned around. If I hadn't pulled the trigger then ... I don't want to remember it." The patrolman pushed the door open. "There's no bodies out here but there's some blood. Quite a lot of blood. A little to one side of the walk." The policemen went out. "Thank God you woke up, Nancy," Martin said. "I'd have let them have the baby." He reached over and smoothed the sleeping Reggie's hair. Nancy, who was rocking the boy, narrowed her eyes. "I wonder why they want our baby? He's just like any other baby. We don't have any money. We couldn't pay a ransom." "Reggie's pretty cute, though," Martin said. "You will have to admit that." Nancy smiled. Then she suddenly stopped rocking. "Martin!" He sat up quickly. "Where's Tiger?" Together they rose and walked around the room. They found him in a corner, eyes open, tongue protruding. He was dead. If we keep Reggie in the house much longer he'll turn out to be a hermit," Martin said at breakfast a month later. "He needs fresh air and sunshine." "I'm not going to sit on the lawn alone with him, Martin. I just can't, that's all. I'd be able to think of nothing but that day." "Still thinking about it? I think we'd have heard from them again if they were coming back. They probably got somebody else's baby by this time." Martin finished his coffee and rose to kiss her good-bye. "But for safety's sake I guess you'd better keep that gun handy." The morning turned into a brilliant, sunshiny day. Puffs of clouds moved slowly across the summer sky and a warm breeze rustled the trees. It would be a crime to keep Reggie inside on a day like this, Nancy thought. So she called Mrs. MacDougal, the next door neighbor. Mrs. MacDougal was familiar with what had happened to the Laughtons and she agreed to keep an eye on Nancy and Reggie and to call the police at the first sign of trouble. With a fearful but determined heart Nancy moved the play pen and set it up in the front yard. She spread a blanket for herself and put Reggie in the pen. Her heart pounded all the while and she watched the street for any strangers, ready to flee inside if need be. Reggie just gurgled with delight at the change in environment. This peaceful scene was disturbed by a speeding car in which two men were riding. The car roared up the street, swerved toward the parkway, tires screaming, bounced over the curb and sidewalk, straight toward the child and mother. Reggie, attracted by the sudden noise, looked up to see the approaching vehicle. His mother stood up, set her palms against her cheeks and shrieked. The car came on, crunched over the play pen, killing the child. The mother was hit and instantly killed, force of the blow snapping her spine and tossing her against the house. The car plunged on into a tree, hitting it a terrible blow, crumbling the car's forward end so it looked like an accordion. The men were thrown from the machine. "We'll never be able to prosecute in this case," the states attorney said. "At least not on a drunken driving basis." "I can't get over it," the chief of police said. "I've got at least six men who will swear the man was drunk. He staggered, reeled and gave the usual drunk talk. He reeked of whiskey." The prosecutor handed the report over the desk. "Here's the analysis. Not a trace of alcohol. He couldn't have even had a smell of near beer. Here's another report. This is his physical exam made not long afterwards. The man was in perfect health. Only variations are he had a scar on his leg where something, probably a dog, bit him once. And then a scar on his chest. It looked like an old gunshot wound, they said. Must have happened years ago." "That's odd. The man who accosted Mrs. Laughton in the afternoon was bitten by their dog. Later that night she said she shot the same man in the chest. Since the scars are healed it obviously couldn't be the same man. But there's a real coincidence for you. And speaking of the dogbite, the Laughton dog died that night. His menu evidently didn't agree with him. Never did figure what killed him, actually." "Any record of treatment on the man she shot?" "The men . You'll remember, there were two. No, we never found a trace of either. No doctor ever made a report of a gunshot wound that night. No hospital had a case either—at least not within several hundred miles—that night or several nights afterwards. Ever been shot with .30-.30?" The state attorney shook his head. "I wouldn't be here if I had." "I'll say you wouldn't. The pair must have crawled away to die God knows where." "Getting back to the man who ran over the child and killed Mrs. Laughton. Why did he pretend to be drunk?" It was the chief's turn to shake his head. "Your guess is as good as mine. There are a lot of angles to this case none of us understand. It looks deliberate, but where's the motive?" "What does the man have to say?" "I was afraid you'd get to him," the chief said, his neck reddening. "It's all been rather embarrassing to the department." He coughed self-consciously. "He's proved a strange one, all right. He says his name is John Smith and he's got cards to prove it, too—for example, a social security card. It looks authentic, yet there's no such number on file in Washington, so we've discovered. We've had him in jail for a week and we've all taken turns questioning him. He laughs and admits his guilt—in fact, he seems amused by most everything. Sometimes all alone in his cell he'll start laughing for no apparent reason. It gives you the creeps." The states attorney leaned back in his chair. "Maybe it's a case for an alienist." "One jump ahead of you. Dr. Stone thinks he's normal, but won't put down any I.Q. Actually, he can't figure him out himself. Smith seems to take delight in answering questions—sort of anticipates them and has the answer ready before you're half through asking." "Well, if Dr. Stone says he's normal, that's enough for me." The prosecutor was silent for a moment. Then, "How about the husband?" "Laughton? We're afraid to let him see him. All broken up. No telling what kind of a rumpus he'd start—especially if Smith started his funny business." "Guess you're right. Well, Mr. Smith won't think it's so funny when we hang criminal negligence or manslaughter on him. By the way, you've checked possible family connections?" "Nobody ever saw John Smith before. Even at the address on his driver's license. And there's no duplicate of that in Springfield, in case you're interested." The man who had laughingly told police his name was John Smith lay on his cot in the county jail, his eyes closed, his arms folded across his chest. This gave him the appearance of being alert despite reclining. Even as he lay, his mouth held a hint of a smile. Arvid 6—for John Smith was Arvid 6—had lain in that position for more than four hours, when suddenly he snapped his eyes open and appeared to be listening. For a moment a look of concern crossed his face and he swung his legs to the floor and sat there expectantly. Arvid 6 knew Tendal 13 had materialized and was somewhere in the building. Eventually there were some sounds from beyond the steel cell and doorway. There was a clang when the outer doorway was opened and Arvid 6 rose from his cot. "Your lawyer's here to see you," the jailer said, indicating the man with the brief case. "Ring the buzzer when you're through." The jailer let the man in, locked the cell door and walked away. The man threw the brief case on the jail cot and stood glaring. "Your damned foolishness has gone far enough. I'm sick and tired of it," he declared. "If you carry on any more we'll never get back to the Ultroom!" "I'm sorry, Tendal," the man on the cot said. "I didn't think—" "You're absolutely right. You didn't think. Crashing that car into that tree and killing that woman—that was the last straw. You don't even deserve to get back to our era. You ought to be made to rot here." "I'm really sorry about that," Arvid 6 said. You know the instructions. Just because you work in the Ultroom don't get to thinking human life doesn't have any value. We wouldn't be here if it hadn't. But to unnecessarily kill—" The older man shook his head. "You could have killed yourself as well and we'd never get the job done. As it is, you almost totally obliterated me." Tendal 13 paced the length of the cell and back again, gesturing as he talked. "It was only with the greatest effort I pulled myself back together again. I doubt that you could have done it. And then all the while you've been sitting here, probably enjoying yourself with your special brand of humor I have grown to despise." "You didn't have to come along at all, you know," Arvid 6 said. "How well I know! How sorry I am that I ever did! It was only because I was sorry for you, because someone older and more experienced than you was needed. I volunteered. Imagine that! I volunteered! Tendal 13 reaches the height of stupidity and volunteers to help Arvid 6 go back 6,000 years to bring Kanad back, to correct a mistake Arvid 6 made!" He snorted. "I still can't believe I was ever that stupid. I only prove it when I pinch myself and here I am. "Oh, you've been a joy to be with! First it was that hunt in ancient Mycenae when you let the lion escape the hunters' quaint spears and we were partly eaten by the lion in the bargain, although you dazzled the hunters, deflecting their spears. And then your zest for drink when we were with Octavian in Alexandria that led to everybody's amusement but ours when we were ambushed by Anthony's men. And worst of all, that English barmaid you became engrossed with at our last stop in 1609, when her husband mistook me for you and you let him take me apart piece by piece—" "All right, all right," Arvid 6 said. "I'll admit I've made some mistakes. You're just not adventurous, that's all." "Shut up! For once you're going to listen to me. Our instructions specifically stated we were to have as little as possible to do with these people. But at every turn you've got us more and more enmeshed with them. If that's adventure, you can have it." Tendal 13 sat down wearily and sank his head in his hands. "It was you who conceived the idea of taking Reggie right out of his play pen. 'Watch me take that child right out from under its mother's nose' were your exact words. And before I could stop you, you did. Only you forgot an important factor in the equation—the dog, Tiger. And you nursed a dogbite most of the afternoon before it healed. And then you took your spite out on the poor thing by suggesting suffocation to it that night. "And speaking of that night, you remember we agreed I was to do the talking. But no, you pulled a switch and captured Martin Laughton's attention. 'I came as soon as I could, Martin,' you said. And suddenly I played a very minor role. 'This is my new assistant, Dr. Tompkins,' you said. And then what happened? I get shot in the legs and you get a hole in your back. We were both nearly obliterated that time and we didn't even come close to getting the child. "Still you wanted to run the whole show. 'I'm younger than you,' you said. 'I'll take the wheel.' And the next thing I know I'm floating in space halfway to nowhere with two broken legs, a spinal injury, concussion and some of the finest bruises you ever saw." These twentieth century machines aren't what they ought to be," Arvid 6 said. "You never run out of excuses, do you, Arvid? Remember what you said in the Ultroom when you pushed the lever clear over and transferred Kanad back 6,000 years? 'My hand slipped.' As simple as that. 'My hand slipped.' It was so simple everyone believed you. You were given no real punishment. In a way it was a reward—at least to you—getting to go back and rescue the life germ of Kanad out of each era he'd be born in." Tendal 13 turned and looked steadily and directly at Arvid 6. "Do you know what I think? I think you deliberately pushed the lever over as far as it would go just to see what would happen . That's how simple I think it was." Arvid 6 flushed, turned away and looked at the floor. "What crazy things have you been doing since I've been gone?" Tendal 13 asked. Arvid 6 sighed. "After what you just said I guess it wouldn't amuse you, although it has me. They got to me right after the accident before I had a chance to collect my wits, dematerialize or anything—you said we shouldn't dematerialize in front of anybody." "That's right." "Well, I didn't know what to do. I could see they thought I was drunk, so I was. But they had a blood sample before I could manufacture any alcohol in my blood, although I implanted a memory in them that I reeked of it." He laughed. "I fancy they're thoroughly confused." "And you're thoroughly amused, no doubt. Have they questioned you?" "At great length. They had a psychiatrist in to see me. He was a queer fellow with the most stupid set of questions and tests I ever saw." "And you amused yourself with him." "I suppose you'd think so." "Who do you tell them you are?" "John Smith. A rather prevalent name here, I understand. I manufactured a pasteboard called a social security card and a driver's license—" "Never mind. It's easy to see you've been your own inimitable self. Believe me, if I ever get back to the Ultroom I hope I never see you again. And I hope I'll never leave there again though I'm rejuvenated through a million years." "Was Kanad's life germ transferred all right this time?" Tendal 13 shook his head. "I haven't heard. The transfers are getting more difficult all the time. In 1609, you'll remember, it was a case of pneumonia for the two-year-old. A simple procedure. It wouldn't work here. Medicine's too far along." He produced a notebook. "The last jump was 342 years, a little more than average. The next ought to be around 2250. Things will be more difficult than ever there, probably." "Do you think Kanad will be angry about all this?" "How would you like to have to go through all those birth processes, to have your life germ knocked from one era to the next?" "Frankly, I didn't think he'd go back so far." "If it had been anybody but Kanad nobody'd ever have thought of going back after it. The life germ of the head of the whole galactic system who came to the Ultroom to be transplanted to a younger body—and then sending him back beyond his original birth date—" Tendal 13 got up and commenced his pacing again. "Oh, I suppose Kanad's partly to blame, wanting rejuvenating at only 300 years. Some have waited a thousand or more or until their bones are like paper." "I just wonder how angry Kanad will be," Arvid muttered. HB92167. Ultroom Error. Tendal 13. Arvid 6. Kanad transfer out of 1951 complete. Next Kanad transfer ready. 2267. Phullam 19, son of Orla 39 and Rhoda R, 22H Level M, Hemisphere B, Quadrant 3, Sector I. Arrive his 329th Day. TB92167 Arvid 6 rose from the cot and the two men faced each other. "Before we leave, Arvid," Tendal 13 started to say. "I know, I know. You want me to let you handle everything." "Exactly. Is that too much to ask after all you've done?" "I guess I have made mistakes. From now on you be the boss. I'll do whatever you say." "I hope I can count on that." Tendal 13 rang the jail buzzer. The jailer unlocked the cell door. "You remember the chief said it's all right to take him with me, Matthews," Tendal 13 told the jailer. "Yes, I remember," the jailer said mechanically, letting them both out of the cell. They walked together down the jail corridor. When they came to another barred door the jailer fumbled with the keys and clumsily tried several with no luck. Arvid 6, an amused set to his mouth and devilment in his eyes, watched the jailer's expression as he walked through the bars of the door. He laughed as he saw the jailer's eyes bulge. "Arvid!" Tendal 13 walked briskly through the door, snatched Arvid 6 by the shoulders and shook him. The jailer watched stupified as the two men vanished in the middle of a violent argument.
|
D. To steal Phullam from his parents and get closer to recovering Kanad.
|
All of the following is 'recycled' to create extra 'food' EXCEPT for:
A. urine
B. hair
C. algae
D. bones
|
GOURMET By ALLEN KIM LANG [Transcriber's Note: This etext was produced from Galaxy Magazine April 1962. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] This was the endless problem of all spaceship cooks: He had to feed the men tomorrow on what they had eaten today! Unable to get out to the ballgame and a long way off from the girls, men on ships think about, talk about, bitch about their food. It's true that Woman remains a topic of thoughtful study, but discussion can never replace practice in an art. Food, on the other hand, is a challenge shipmen face three times a day, so central to their thoughts that a history of sea-faring can be read from a commissary list. In the days when salt-sea sailors were charting islands and spearing seals, for example, the fo'c's'le hands called themselves Lobscousers, celebrating the liquid hash then prominent in the marine menu. The Limey sailor got the name of the anti-scorbutic citrus squeezed into his diet, a fruit known to us mariners of a more sophisticated age only as garnish for our groundside gin-and-tonic. And today we Marsmen are called Slimeheads, honoring in our title the Chlorella and Scenedesmus algae that, by filling up the spaces within, open the road to the larger Space without. Should any groundsman dispute the importance of belly-furniture in history—whether it be exterminating whales, or introducing syphilis to the Fiji Islanders, or settling the Australian littoral with cross-coves from Middlesex and Hampshire—he is referred to the hundred-and-first chapter of Moby Dick , a book spooled in the amusement tanks of all but the smallest spacers. I trust, however, that no Marsman will undertake to review this inventory of refreshment more than a week from groundfall. A catalogue of sides of beef and heads of Leyden cheese and ankers of good Geneva would prove heavy reading for a man condemned to snack on the Chlorella-spawn of cis-Martian space. The Pequod's crew ate wormy biscuit and salt beef. Nimitz's men won their war on canned pork and beans. The Triton made her underwater periplus of Earth with a galley stocked with frozen pizza and concentrated apple-juice. But then, when sailors left the seas for the skies, a decline set in. The first amenity of groundside existence to be abandoned was decent food. The earliest men into the vacuum swallowed protein squeezings from aluminum tubes, and were glad enough to drop back to the groundsman's diet of steak and fried potatoes. Long before I was a boy in Med School, itching to look at black sky through a view-port, galley science had fulfilled the disgusting exordium of Isaiah 36:12, to feed the Slimeheads for breakfast today what was day-before-yesterday's table-scraps and jakes-water. The Ship's Cook, the man who accomplishes the daily miracle of turning offal into eatables, is in many ways the most vital man aboard a spacer. He can make morale or foment a mutiny. His power is paramount. Slimeheads remember the H. M. S. Ajax fiasco, for example, in which a galleyman leveled his Chlorella tanks with heavy water from the ship's shielding. Four officers and twenty-one Other Ranks were rescued from the Ajax in deep space, half dead from deuterium poisoning. We think of the Benjo Maru incident, too, caused by a Ship's Cook who allowed his algaeal staff-of-life to become contaminated with a fast-growing Saccharomycodes yeast. The Japanese vessel staggered to her pad at Piano West after a twenty-week drunk: the alien yeast had got into the stomach of every man aboard, where it fermented each subsequent bite he ate to a superior grade of sake . And for a third footnote to the ancient observation, "God sends food, and the Devil sends cooks," Marsmen will recall what happened aboard my ship the Charles Partlow Sale . The Sale blasted off from Brady Station in the middle of August, due in at Piano West in early May. In no special hurry, we were taking the low-energy route to Mars, a pathway about as long in time as the human period of gestation. Our cargo consisted mostly of Tien-Shen fir seedlings and some tons of an arctic grass-seed—these to be planted in the maria to squeeze out the native blue bugberry vines. We had aboard the Registry minimum of six men and three officers. Ship's Surgeon was myself, Paul Vilanova. Our Captain was Willy Winkelmann, the hardest man in space and very likely the fattest. Ship's Cook was Robert Bailey. Cooking aboard a spacer is a job combining the more frustrating tensions of biochemistry, applied mycology, high-speed farming, dietetics and sewage engineering. It's the Cook's responsibility to see that each man aboard gets each day no less than five pounds of water, two pounds of oxygen, and one-and-a-half pounds of dry food. This isn't just a paragraph from the Spacer Union Contract. It's a statement of the least fuel a man can run on. Twelve tons of water, oxygen, and food would have filled the cargo compartments to bursting, and left a small ship like the C. P. Sale no reason to reach for Mars. By allowing a colony of Chlorella algae to work over our used air, water and other effluvia, though, three tons of metabolites would see us through from Brady Station to Piano West and back. Recycling was the answer. The molecule of carbohydrate, fat, protein or mineral that didn't feed the crew fed the algae. And the algae fed us. All waste was used to fertilize our liquid fields. Even the stubble from our 2,680 shaves and the clippings from our 666 haircuts en route and back would be fed into the Chlorella tanks. Human hair is rich in essential amino acids. The algae—dried by the Cook, bleached with methyl alcohol to kill the smell and make the residue more digestible, disguised and seasoned in a hundred ways—served as a sort of meat-and-potatoes that never quite wore out. Our air and water were equally immortal. Each molecule of oxygen would be conversant with the alveoli of every man aboard by the end of our trip. Every drop of water would have been intimate with the glomeruli of each kidney on the ship before we grounded in. Groundling politicians are right enough when they say that we spacers are a breed apart. We're the one race of men who can't afford the luxury of squeamishness. Though I'm signed aboard as Ship's Surgeon, I seldom lift a knife in space. My employment is more in the nature of TS-card-puncher extraordinary. My duties are to serve as wailing-wall, morale officer, guardian of the medicinal whiskey and frustrator of mutual murder. Generally the man aboard who'd serve as the most popular murder-victim is the Cook. This trip, the-man-you-love-to-hate was our Captain. If the Cook hadn't problems enough with the chemical and psychic duties of his office, Winkelmann supplied the want. Captain Willy Winkelmann was the sort of man who, if he had to go into space at all, had best do so alone. If the Prussians had a Marine Corps, Winkelmann would have done splendidly as Drill Instructor for their boot camp. His heart was a chip of helium ice, his voice dripped sarcastic acid. The planet Earth was hardly large enough to accommodate a wart as annoying as Willy Winkelmann. Cheek-by-jowl every day in a nacelle the size of a Pullman car, our Captain quickly established himself as a major social hemorrhoid. The Captain's particular patsy was, of course, young Bailey the Cook. It was Winkelmann who saw humorous possibilities in the entry, "Bailey, Robert," on Ship's Articles. He at once renamed our unfortunate shipmate "Belly-Robber." It was Winkelmann who discussed haut cuisine and the properties of the nobler wines while we munched our algaeburgers and sipped coffee that tasted of utility water. And it was Captain Willy Winkelmann who never referred to the ship's head by any other name than The Kitchen Cabinet. Bailey tried to feed us by groundside standards. He hid the taste of synthetic methionine—an essential amino acid not synthesized by Chlorella—by seasoning our algaeal repasts with pinches of oregano and thyme. He tinted the pale-green dollops of pressed Chlorella pink, textured the mass to the consistency of hamburger and toasted the slabs to a delicate brown in a forlorn attempt to make mock-meat. For dessert, he served a fudge compounded from the dextrose-paste of the carbohydrate recycler. The crew thanked him. The Captain did not. "Belly-Robber," he said, his tone icy as winter wind off the North Sea, "you had best cycle this mess through the tanks again. There is a pun in my home country: Mensch ist was er isst. It means, you are what you eat. I think you are impertinent to suggest I should become this Schweinerei you are feeding me." Captain Winkelmann blotted his chin with his napkin, heaved his bulk up from the table, and climbed up the ladder from the dining-cubby. "Doc, do you like Winkelmann?" the Cook asked me. "Not much," I said. "I suspect that the finest gift our Captain can give his mother is to be absent from her on Mother's Day. But we've got to live with him. He's a good man at driving a ship." "I wish he'd leave off driving this Cook," Bailey said. "The fat swine!" "His plumpness is an unwitting tribute to your cooking, Bailey," I said. "He eats well. We all do. I've dined aboard a lot of spacers in my time, and I'll testify that you set a table second to none." Bailey took a handful of dried Chlorella from a bin and fingered it. It was green, smelled of swamp, and looked appetizing as a bedsore. "This is what I have to work with," he said. He tossed the stuff back into its bin. "In Ohio, which is my home country, in the presence of ladies, we'd call such garbage Horse-Leavings." "You'll never make Winkelmann happy," I said. "Even the simultaneous death of all other human beings could hardly make him smile. Keep up the good work, though, and you'll keep our Captain fat." Bailey nodded from his one-man cloud of gloom. I got a bottle of rye from Medical Stores and offered him a therapeutic draught. The Cook waved my gift aside. "Not now, Doc," he said. "I'm thinking about tomorrow's menu." The product of Bailey's cerebrations was on the mess table at noon the next day. We were each served an individual head of lettuce, dressed with something very like vinegar and oil, spiced with tiny leaves of burnet. How Bailey had constructed those synthetic lettuces I can only guess: the hours spent preparing a green Chlorella paste, rolling and drying and shaping each artificial leaf, the fitting together of nine heads like crisp, three-dimensional jigsaw puzzles. The pièce de résistance was again a "hamburger steak;" but this time the algaeal mass that made it up was buried in a rich, meaty gravy that was only faintly green. The essence-of-steak used in these Chlorella cutlets had been sprinkled with a lavish hand. Garlic was richly in evidence. "It's so tender," the radioman joked, "that I can hardly believe it's really steak." Bailey stared across the dining-cubby toward Winkelmann, silently imploring the Captain's ratification of his masterpiece. The big man's pink cheeks bulged and jumped with his chewing. He swallowed. "Belly-Robber," Winkelmann said, "I had almost rather you served me this pond-scum raw than have it all mucked-up with synthetic onions and cycler-salt." "You seem able enough to choke down Bailey's chow, Captain," I said. I gazed at Winkelmann's form, bulbous from a lifetime of surfeit feeding. "Yes, I eat it," the Captain said, taking and talking through another bite. "But I eat only as a man in the desert will eat worms and grasshoppers, to stay alive." "Sir, what in heaven's name do you expect from me?" Bailey pleaded. "Only good food," Winkelmann mumbled through his mouthful of disguised algae. He tapped his head with a finger. "This—the brain that guides the ship—cannot be coaxed to work on hog-slop. You understand me, Belly-Robber?" Bailey, his hands fisted at his sides, nodded. "Yes, sir. But I really don't know what I can do to please you." "You are a spacer and a Ship's Cook, not a suburban Hausfrau with the vapors," Winkelmann said. "I do not expect from you hysterics, tantrums or weeping. Only—can you understand this, so simple?—food that will keep my belly content and my brain alive." "Yes, sir," Bailey said, his face a picture of that offense the British term Dumb Insolence. Winkelmann got up and climbed the ladder to the pilot-cubicle. I followed him. "Captain," I said, "you're driving Bailey too hard. You're asking him to make bricks without straw." Winkelmann regarded me with his pale-blue stare. "You think, Doctor, that my cruelty to the Belly-Robber is the biliousness of a middle-aged man?" "Frankly, I can't understand your attitude at all," I said. "You accuse me of driving a man to make bricks without straw," Winkelmann said. "Very well, Doctor. It is my belief that if the Pharaoh's taskmaster had had my firmness of purpose, the Children of Israel would have made bricks with stubble. Necessity, Doctor, is the mother of invention. I am Bailey's necessity. My unkindnesses make him uncomfortable, I doubt that not. But I am forcing him to experiment, to improvise, to widen the horizons of his ingenuity. He will learn somehow to bring good food from Chlorella tanks." "You're driving him too hard, Sir," I said. "He'll crack." "Bailey will have some fifty thousand dollars' salary waiting when we ground at Brady Station," Captain Winkelmann said. "So much money buys many discomforts. That will be all, Doctor Vilanova." "Crew morale on the ship...." I began. "That will be all, Doctor Vilanova," Captain Winkelmann repeated. Bailey grew more silent as we threaded our way along the elliptical path to Mars. Each meal he prepared was a fresh attempt to propitiate the appetite of our splenetic Captain. Each such offering was condemned by that heartless man. Bailey began to try avoiding the Captain at mealtimes, but was frustrated by Winkelmann's orders. "Convey my compliments to the Chef, please," the Captain would instruct one of the crew, "and ask him to step down here a moment." And the Cook would cheerlessly appear in the dining-cubby, to have his culinary genius acidly called in question again. I myself do not doubt that Bailey was the finest Cook ever to go into Hohmann orbit. His every meal established a higher benchmark in brilliant galleymanship. We were served, for instance, an ersatz hot turkey supreme. The cheese-sauce was almost believable, the Chlorella turkey-flesh was white and tender. Bailey served with this delicacy a grainy and delicious "cornbread," and had extracted from his algae a lipid butter-substitute that soaked into the hot "bread" with a genuinely dairy smell. "Splendid, Bailey," I said. "We are not amused," said Captain Winkelmann, accepting a second helping of the pseudo-turkey. "You are improving, Belly-Robber, but only arithmetically. Your first efforts were so hideous as to require a geometric progression of improving excellence to raise them to mere edibility. By the time we are halfway 'round the Sun, I trust you will have learned to cook with the competence of a freshman Home Economics student. That will be all, Bailey." The crew and my fellow-officers were amused by Winkelmann's riding of Bailey; they were in addition gratified that the battle between their Captain and their Cook served to feed them so well. Most spacers embark on an outward voyage somewhat plump, having eaten enough on their last few days aground to smuggle several hundred calories of fat and many memories of good food aboard with them. This trip, none of the men had lost weight during the first four months in space. Winkelmann, indeed, seemed to have gained. His uniform was taut over his plump backside, and he puffed a bit up the ladders. I was considering suggesting to our Captain that he curtail his diet for reasons of health, a bit of advice that would have stood unique in the annals of space medicine, when Winkelmann produced his supreme insult to our Cook. Each man aboard a spacer is allowed ten kilograms of personal effects besides his uniforms, these being considered Ship's Furnishing. As his rank and responsibility merit, the Captain is allowed double this ration. He may thus bring aboard with him some forty-five pounds of books, playing-cards, knitting-wool, whiskey or what have you to help him while away the hours between the planets. Bailey, I knew for a fact, had used up his weight-allowance in bringing aboard a case of spices: marjoram and mint, costmary, file powder, basil and allspice, and a dozen others. Captain Winkelmann was not a reader, and had brought no books. Cards interested him not at all, as card-playing implies a sociability alien to his nature. He never drank aboard ship. I had supposed that he'd exercised his option of returning his personal-effects weight allowance to the owners for the consideration of one hundred dollars a kilogram. To collect the maximum allowance, spacers have been known to come aboard their ship mother-naked. But this was not the case with Winkelmann. His personal-effects baggage, an unlabeled cardboard box, appeared under the table at noon mess some hundred days out from Piano West. Winkelmann rested his feet on the mysterious box as he sat to eat. "What disgusting form does the ship's garbage appear in today, Belly-Robber?" he asked the Cook. Bailey frowned, but kept his temper, an asceticism in which by now he'd had much practice. "I've been working on the problem of steak, Sir," he said. "I think I've whipped the taste; what was left was to get the texture steak-like. Do you understand, Sir?" "I understand," Winkelmann growled. "You intend that your latest mess should feel like steak to the mouth, and not like baby-food. Right?" "Yes, Sir," Bailey said. "Well, I squeezed the steak-substrate—Chlorella, of course, with all sorts of special seasonings—through a sieve, and blanched the strands in hot algaeal oil. Then I chopped those strands to bits and rolled them out. Voila! I had something very close in texture to the muscle-fibers of genuine meat." "Remarkable, Bailey," I said. "It rather throws me off my appetite to hear how you muddle about with our food," the Captain said, his jowls settling into an expression of distaste. "It's quite all right to eat lobster, for example, but I never cared to see the ugly beast boiled before my eyes. Detail spoils the meal." Bailey lifted the cover off the electric warming-pan at the center of the table and tenderly lifted a small "steak" onto each of our plates. "Try it," he urged the Captain. Captain Winkelmann sliced off a corner of his algaeal steak. The color was an excellent medium-rare, the odor was the rich smell of fresh-broiled beef. Winkelmann bit down, chewed, swallowed. "Not too bad, Belly-Robber," he said, nodding. Bailey grinned and bobbed his head, his hands folded before him in an ecstasy of pleasure. A kind word from the Captain bettered the ruffles-and-flourishes of a more reasonable man. "But it still needs something ... something," Winkelmann went on, slicing off another portion of the tasty Chlorella. "Aha! I have it!" "Yes, Sir?" Bailey asked. "This, Belly-Robber!" Winkelmann reached beneath the mess-table and ripped open his cardboard carton. He brought out a bottle and unscrewed the cap. "Ketchup," he said, splattering the red juice over Bailey's masterpiece. "The scarlet burial-shroud for the failures of Cooks." Lifting a hunk of the "steak," streaming ketchup, to his mouth, Winkelmann chewed. "Just the thing," he smiled. "Damn you!" Bailey shouted. Winkelmann's smile flicked off, and his blue eyes pierced the Cook. "... Sir," Bailey added. "That's better," Winkelmann said, and took another bite. He said meditatively, "Used with caution, and only by myself, I believe I have sufficient ketchup here to see me through to Mars. Please keep a bottle on the table for all my future meals, Belly-Robber." "But, Sir...." Bailey began. "You must realize, Belly-Robber, that a dyspeptic Captain is a threat to the welfare of his ship. Were I to continue eating your surrealistic slops for another hundred days, without the small consolation of this sauce I had the foresight to bring with me, I'd likely be in no condition to jet us safely down to the Piano West pad. Do you understand, Belly-Robber?" he demanded. "I understand that you're an ungrateful, impossible, square-headed, slave-driving...." "Watch your noun," Winkelmann cautioned the Cook. "Your adjectives are insubordinate; your noun might prove mutinous." "Captain, you've gone too far," I said. Bailey, his fists knotted, was scarlet, his chest heaving with emotion. "Doctor, I must point out to you that it ill behooves the Ship's Surgeon to side with the Cook against the Captain," Winkelmann said. "Sir, Bailey has tried hard to please you," I said. "The other officers and the men have been more than satisfied with his work." "That only suggests atrophy of their taste buds," Winkelmann said. "Doctor, you are excused. As are you, Belly-Robber," he added. Bailey and I climbed from the mess compartment together. I steered him to my quarters, where the medical supplies were stored. He sat on my bunk and exploded into weeping, banging his fists against the metal bulkhead. "You'll have that drink now," I said. "No, dammit!" he shouted. "Orders," I said. I poured us each some fifty cc's of rye. "This is therapy, Bailey," I told him. He poured the fiery stuff down his throat like water and silently held out his glass for a second. I provided it. After a few minutes Bailey's sobbing ceased. "Sorry, Doc," he said. "You've taken more pressure than most men would," I said. "Nothing to be ashamed of." "He's crazy. What sane man would expect me to dip Wiener schnitzel and sauerkraut and Backhahndl nach suddeutscher Art out of an algae tank? I've got nothing but microscopic weeds to cook for him! Worn-out molecules reclaimed from the head; packaged amino acid additives. And he expects meals that would take the blue ribbon at the annual banquet of the Friends of Escoffier!" "Yours is an ancient plaint, Bailey," I said. "You've worked your fingers to the bone, slaving over a hot stove, and you're not appreciated. But you're not married to Winkelmann, remember. A year from now you'll be home in Ohio, fifty grand richer, set to start that restaurant of yours and forget about our fat Flying Dutchman." "I hate him," Bailey said with the simplicity of true emotion. He reached for the bottle. I let him have it. Sometimes alcohol can be an apt confederate of vis medicatrix naturae , the healing power of nature. Half an hour later I strapped Bailey into his bunk to sleep it off. That therapeutic drunk seemed to be just what he'd needed. For morning mess the next day we had a broth remarkable in horribleness, a pottage or boiled Chlorella vulgaris that looked and tasted like the vomit of some bottom-feeding sea-beast. Bailey, red-eyed and a-tremble, made no apology, and stared at Winkelmann as though daring him to comment. The Captain lifted a spoonful of the disgusting stuff to his lips, smacked and said, "Belly-Robber, you're improving a little at last." Bailey nodded and smiled. "Thank you, Sir," he said. I smiled, too. Bailey had conquered himself. His psychic defenses were now strong enough to withstand the Captain's fiercest assaults of irony. Our food would likely be bad the rest of this trip, but that was a price I was willing to pay for seeing destroyed the Willy Winkelmann theory of forcing a Cook to make bricks without straw. The Captain had pushed too hard. He'd need that ketchup for the meals to come, I thought. Noon mess was nearly as awful as breakfast had been. The coffee tasted of salt, and went largely undrunk. The men in the mess compartment were vehement in their protests, blaming the Captain, in his absence, for the decline in culinary standards. Bailey seemed not to care. He served the algaeburgers with half a mind, and hurried back into his galley oblivious of the taunts of his crewmates. There being only three seats in the Sale's mess compartment, we ate our meals in three shifts. That evening, going down the ladder to supper, my nose was met with a spine-tingling barbecue tang, a smell to make a man think of gray charcoal glowing in a picnic brazier, of cicadas chirping and green grass underfoot, of the pop and hiss of canned beer being church-keyed. "He's done it, Doc!" one of the first-shift diners said. "It actually tastes of food!" "Then he's beat the Captain at his game," I said. "The Dutchman won't want to mess ketchup on these steaks," the crewman said. I sat, unfolded my napkin, and looked with hope to the electric warming-pan at the center of the table. Bailey served the three of us with the small "steaks." Each contained about a pound of dried Chlorella, I judged, teasing mine with my fork. But they were drenched in a gravy rich as the stuff grandma used to make in her black iron skillet, peppery and seasoned with courageous bits of garlic. I cut a bit from my steak and chewed it. Too tender, of course; there are limits to art. But the pond-scum taste was gone. Bailey appeared in the galley door. I gestured for him to join me. "You've done it, Bailey," I said. "Every Slimehead in orbit will thank you for this. This is actually good ." "Thanks, Doc," Bailey said. I smiled and took another bite. "You may not realize it, Bailey; but this is a victory for the Captain, too. He drove you to this triumph; you couldn't have done it without him." "You mean he was just whipping me on, trying to make me do better?" Bailey asked. "He was driving you to do the impossible," I said; "and you did it. Our Captain may be a hard man, Bailey; but he did know how to coax maximum performance out of his Ship's Cook." Bailey stood up. "Do you like Captain Winkelmann, Doctor?" he asked. I thought about his question a moment. Winkelmann was good at his job. He persuaded his men by foul means, true; but it was all for the good of the ship and his crew. "Do I like Captain Winkelmann?" I asked, spearing another piece of my artificial steak. "Bailey, I'm afraid I'll have to admit that I do." Bailey smiled and lifted a second steak from the warming-pan onto my plate. "Then have another piece," he said.
|
D. bones
|
How good were test subjects at labeling the beers in round two?
A. Few of them got anything correct
B. None of them could guess any of them
C. Most of them got most things correct
D. Most of them got them perfect
|
More Booze You Can Use When we last heard from them, the members of the Slate beer-testing team were coping with lagers and trying to see if they could taste the 3-to-1 price difference between the most- and least-expensive brands. (Click for a wrap-up of the first round of beer tasting.) The answer was: They found one beer they really liked, Samuel Adams Boston Lager , and one they really hated, imported Grolsch from Holland. Both were expensive beers--Grolsch was the most expensive in the test--and otherwise the testers had a hard time telling beers apart. The members of the team, as noted in the original article, all hold day jobs at Microsoft, mainly as designers, managers, and coders for Microsoft Word. The point of the second test was not to find the difference between cheap and expensive beers but instead to compare a variety of top-of-the-line beers. Was there one kind the tasters preferred consistently? Could they detect any of the subtleties of brewing style and provenance that microbrew customers pay such attention to when choosing some Doppelbock over a cream ale? Since the tasting panel had left the first round grumbling that cheap lagers were not a fair test of their abilities, this second round of testing was advertised to the panel as a reward. Every beer in Round 2 would be a fancy beer. A microbrew. A "craft beer." A prestigious import. These were the kinds of beer the panel members said they liked--and the ones they said they were most familiar with. One aspect of the reward was that they would presumably enjoy the actual testing more--fewer rueful beer descriptions along the lines of "urine" or "get it away!" were expected than in the first round. The other aspect of anticipated reward was the panelists' unspoken but obvious assumption that this time they would "do better" on the test. Intellectual vanity being what it is, people who had fought for and won jobs at Microsoft and who still must fight every six months for primacy on the employee-ranking scale (which determines--gasp!--how many new stock options they receive) would assume that their skill as tasters was on trial, just as much as the beer was. Of course they were right, which is what made this round as amusing to administer as the first one had been. Here is what happened and what it meant: 1. Procedure. This was similar in most ways to the experimental approach of Round 1. The nine testers who showed up were a subset of the original 12. The missing three dropped out with excuses of "my wife is sick" (one person) and "meeting is running long" (two). As before, each tester found before him on a table 10 red plastic cups, labeled A through J. Each cup held 3 ounces of one of the beers. The A-to-J labeling scheme was the same for all testers. Instead of saltines for palate-cleansing, this time we had popcorn and nuts. As they began, the tasters were given these and only these clues: that the flight included one "holdover" beer from the previous round (Sam Adams); that it included at least one import (Bass); that it included at least one macrobrew , specifically, a member of the vast Anheuser-Busch family (Michelob Hefeweizen). After sampling all beers, the tasters rated them as follows: Overall quality points, from zero to 100, reflecting their personal, subjective fondness for the beer. Descriptions of and comments about each beer's taste--"smooth and nutty," "too strong," etc. If the first ranking was a measure of how good each beer was, this was an attempt to explain what made it good. Best and Worst , one of each from the group. Name that beer! The tasters were told that some of the drinks were Hefeweizens, some might be IPAs (India pale ales), some might be bitters, and so on. They were asked to put each beer in its proper category--and to name a specific brewery and brand if they could. The idea here was to test the veteran beer drinkers' claim to recognize the distinctive tastes of famous brands. (To see all the grids for all the beers, click .) 2. Philosophy. The first round of testing was All Lager. This second round was All Fancy, and Mainly Not Lager. As several correspondents (for instance, the of Best American Beers ) have helpfully pointed out, the definition of lager provided last time was not exactly "accurate." If you want to stay within the realm of textbook definitions, a lager is a beer brewed a particular way--slowly, at cool temperatures, with yeast that settles on the bottom of the vat. This is in contrast with an ale, which is brewed faster, warmer, and with the yeast on top. By this same reasoning, lagers don't have to be light-colored, weak-flavored, and watery, as mainstream American lagers are. In principle, lagers can be dark, fierce, manly. Therefore, the correspondents suggest, it was wrong to impugn Sam Adams or Pete's Wicked for deceptive labeling, in presenting their tawnier, more flavorful beers as lagers too. To this the beer scientist must say: Book-learning is fine in its place. But let's be realistic. Actual drinking experience teaches the American beer consumer that a) all cheap beers are lagers; and b) most lagers are light-colored and weak. The first test was designed to evaluate low-end beers and therefore had to be lager-centric. This one is designed to test fancy beers--but in the spirit of open-mindedness and technical accuracy, it includes a few "strong" lagers too. 3. Materials. The 10 test beers were chosen with several goals in mind: To cover at least a modest range of fancy beer types--extra special bitter, India pale ale, Hefeweizen, and so on. To include both imported and domestic beers. Among the domestic microbrews, there's an obvious skew toward beers from the Pacific Northwest. But as Microsoft would put it, that's a feature not a bug. These beers all came from the Safeway nearest the Redmond, Wash., "main campus" of Microsoft, and microbrews are supposed to be local. To include one holdover from the previous test, as a scientific control on our tasters' preferences. This was Sam Adams , runaway winner of Round 1. To include one fancy product from a monster-scale U.S. mass brewery, to see if the tasters liked it better or worse than the cute little microbrews. This was Michelob Hefeweizen , from the pride of St. Louis, Anheuser-Busch. Click for pricing information and pre-quaffing evaluations. The beers tasted were: 4. Data Analysis. a) Best and Worst. Compared to the lager test, we would expect the range of "best" choices to be more varied, since all the tested beers were supposed to be good. This expectation was most dramatically borne out in the "Best and Worst" rankings. The nine tasters cast a total of nine Worst votes and 11.5 Best votes. (Tester No. 1 turned in a sheet with three Best selections, or two more than his theoretical quota. Tester No. 4 listed a Best and a Best-minus, which counted as half a vote.) The results were clearest at the bottom: three Worsts for Pyramid Hefeweizen , even though most comments about the beer were more or less respectful. ("Bitter, drinkable.") But at the top and middle the situation was muddier: There were three Bests for Full Sail ESB , which most of the tasters later said they weren't familiar with, and 2.5 for Redhook IPA , which all the tasters knew. But each of these also got a Worst vote, and most of the other beers had a mixed reading. So far, the tasters are meeting expectations, finding something to like in nearly all these fancy beers. b) Overall preference points. Here the complications increase. The loser was again apparent: Pyramid Hefeweizen came in last on rating points, as it had in the Best/Worst derby. But the amazing dark horse winner was Michelob Hefeweizen . The three elements of surprise here, in ascending order of unexpectedness, are: This best-liked beer belonged to the same category, Hefeweizen, as the least-liked product, from Pyramid. This was also the only outright Anheuser-Busch product in the contest (the Redhooks are 75 percent A-B free). It is safe to say that all tasters would have said beforehand that they would rank an American macrobrew last, and Anheuser-Busch last of all. Although it clearly won on overall preference points, Michelob Hefeweizen was the only beer not to have received a single "Best" vote. The first two anomalies can be written off as testament to the power of a blind taste test. The third suggests an important difference in concepts of "bestness." Sometimes a product seems to be the best of a group simply because it's the most unusual or distinctive. This is why very high Wine Spectator ratings often go to wines that mainly taste odd. But another kind of bestness involves an unobtrusive, day-in day-out acceptability. That seems to be Michelob Hefe 's achievement here: no one's first choice, but high on everyone's list. Let's go to the charts: This table shows how the beers performed on "raw score"--that is, without the advanced statistical adjustment of throwing out the highest and lowest score each beer received. Next, we have "corrected average preference points," throwing out the high and low marks for each beer. The result is basically the same: It is worth noting the fate of Sam Adams on these charts. Here it ends up with a score of less than 61. These were the numbers awarded by the very same tasters who gave it a corrected preference rating of 83.33 the last time around--and 10 "Best" votes, vs. one Best (and one Worst) this time. The shift in Bests is understandable and demonstrates the importance of picking your competition. The severe drop in preference points illustrates more acutely the ancient principle of being a big fish in a small pond. These same tasters thought that Sam Adams was objectively much better when it was surrounded by Busch and Schmidt's. c) Value rankings. Last time this calculation led to what the colorful French would call a bouleversement. One of the cheapest beers, Busch, which had been in the lower ranks on overall preference points, came out at the top on value-for-money ratings, because it was so cheap. The big surprise now is that the highest-rated beer was also the cheapest one, Michelob Hefe , so the value calculation turned into a rout: Pyramid Hefeweizen was expensive on top of being unpopular, so its position at the bottom was hammered home--but not as painfully as that of Bass Ale . Bass had been in the respectable lower middle class of the preference rankings, so its disappointing Val-u-meter showing mainly reflects the fact that it was the only beer not on "sale" and therefore by far the costliest entry in the experiment. d) Taster skill. As members of the tasting panel began to suspect, they themselves were being judged while they judged the beer. One of the tasters, No. 7, decided to live dangerously and give specific brands and breweries for Samples A through J. This man was the only panel member whose job does not involve designing Microsoft Word--and the only one to identify two or more of the beers accurately and specifically. (He spotted Redhook IPA and Redhook ESB.) The fact that the beers correctly identified were the two most popular microbrews in the Seattle area suggests that familiarity is the main ingredient in knowing your beer. Many others were simply lost. Barely half the tasters, five of nine, recognized that Michelob Hefeweizen was a Hefeweizen. Before the test, nine of nine would have said that picking out a Hefe was easy, because of its cloudy look and wheaty flavor. Three tasters thought Sam Adams was an IPA ; two thought Redhook's IPA was a Hefeweizen. In fairness, six of nine testers identified Pyramid Hefeweizen as a Hefe, and six recognized Full Sail ESB as a bitter. Much in the fashion of blind men describing an elephant, here is a how the testers handled Sam Adams Boston Lager : 5. Implications and Directions for Future Research. Science does not always answer questions; often, it raises many new ones. This excursion into beer science mainly raises the question: What kind of people are we? If we are Gradgrind-like empiricists, living our life for "welfare maximization" as described in introductory econ. courses, the conclusion is obvious. We learned from the first experiment to buy either Sam Adams (when we wanted maximum lager enjoyment per bottle) or Busch (for maximum taste and snob appeal per dollar). From this second round we see an even more efficient possibility: Buy Michelob Hefeweizen and nothing else, since on the basis of this test it's the best liked and the cheapest beer. By the way, if there is a single company whose achievements the testing panel honored, it would be Anheuser-Busch . From its brewing tanks came two of the double-crown winners of the taste tests: plain old Busch , the Taste-o-meter and Snob-o-meter victor of Round 1, and Michelob Hefeweizen , the preference-point and Val-u-meter winner this time. But, of course, there is another possibility: that what is excluded in a blind taste test is in fact what we want, and are happy to pay for, when we sit down with a beer. The complicated label, the fancy bottle, the exotic concept that this beer has traveled from some far-off corner of Bohemia or even the Yakima Valley--all this may be cheap at the $1.25-per-pint cost difference between the cheapest and the most expensive beers. In elementary school, we all endured a standard science experiment: If you shut your eyes and pinch your nose closed, can you tell any difference in the taste of a slice of apple, of carrot, of pear? You can't--but that doesn't mean that from then on you should close your eyes, hold your nose, and chew a cheap carrot when you feel like having some fruit. There is a time and place for carrots, but also for juicy pears. There is a time for Busch, but also for Full Sail "Equinox." For scientists who want to continue this work at home, here are a few suggestions for further research: Tell the testers ahead of time what beers they will be drinking. Ask them to rank the beers, 1 through 10, based on how well they like them. Then compare the list with the "revealed preferences" that come from the blind test. As a variation, show them the list ahead of time and ask them to pick out the beer they know they love and the one they know they hate. Then compare this with the "after" list. If you're going to test imported lagers, try Foster's or Corona rather than Grolsch. Remember to stay strictly in the scientist's role. Don't take the test yourself.
|
A. Few of them got anything correct
|
What system is used as baseline?
|
### Introduction
The recent years have seen unprecedented forward steps for Natural Language Processing (NLP) over almost every NLP subtask, relying on the advent of large data collections that can be leveraged to train deep neural networks. However, this progress has solely been observed in languages with significant data resources, while low-resource languages are left behind. The situation for endangered languages is usually even worse, as the focus of the scientific community mostly relies in language documentation. The typical endangered language documentation process typically includes the creation of language resources in the form of word lists, audio and video recordings, notes, or grammar fragments, with the created resources then stored into large online linguistics archives. This process is often hindered by the so-called Transcription Bottleneck, but recent advances BIBREF0, BIBREF1 provide promising directions towards a solution for this issue. However, language documentation and linguistic description, although extremely important itself, does not meaningfully contribute to language conservation, which aims to ensure that the language stays in use. We believe that a major avenue towards continual language use is by further creating language technologies for endangered languages, essentially elevating them to the same level as high-resource, economically or politically stronger languages. The majority of the world's languages are categorized as synthetic, meaning that they have rich morphology, be it fusional, agglutinative, polysynthetic, or a mixture thereof. As Natural Language Processing (NLP) keeps expanding its frontiers to encompass more and more languages, modeling of the grammatical functions that guide language generation is of utmost importance. It follows, then, that the next crucial step for expanding NLP research on endangered languages is creating benchmarks for standard NLP tasks in such languages. With this work we take a small first step towards this direction. We present a resource that allows for benchmarking two NLP tasks in San Juan Quiahije Chatino, an endangered language spoken in southern Mexico: morphological analysis and morphological inflection, with a focus on the verb morphology of the language. We first briefly discuss the Chatino language and the intricacies of its verb morphology (§SECREF2), then describe the resource (§SECREF3), and finally present baseline results on both the morphological analysis and the inflection tasks using state-of-the-art neural models (§SECREF4). We make our resource publicly available online. ### The Chatino Language
Chatino is a group of languages spoken in Oaxaca, Mexico. Together with the Zapotec language group, the Chatino languages form the Zapotecan branch of the Otomanguean language family. There are three main Chatino languages: Zenzontepec Chatino (ZEN, ISO 639-2 code czn), Tataltepec Chatino (TAT, cta), and Eastern Chatino (ISO 639-2 ctp, cya, ctz, and cly) (E.Cruz 2011 and Campbell 2011). San Juan Quiahije Chatino (SJQ), the language of the focus of this study, belongs to Eastern Chatino, and is used by about 3000 speakers. ### The Chatino Language ::: Typology and Writing System
Eastern Chatino languages , including SJQ Chatino, are intensively tonal BIBREF2, BIBREF3. Tones mark both lexical and grammatical distinctions in Eastern Chatino languages. In SJQ Chatino, there are eleven tones. Three different systems for representing tone distinctions are employed in the literature: the S-H-M-L system of BIBREF2; the numeral system of BIBREF4; and the alphabetic system of BIBREF3. The correspondences among these three systems are given in Table . For present purposes, we will use numeral representations of the second sort. The number 1 represents a high pitch, 4 represents a low pitch, and double digits represent contour tones. ### The Chatino Language ::: Verb Morphology
SJQ Chatino verb inflection distinguishes four aspect/mood categories: completive (`I did'), progressive (`I am doing'), habitual (`I habitually do') and potential (`I might do'). In each of these categories, verbs inflect for three persons (first, second, third) and two numbers (singular, plural) and distinguish inclusive and exclusive categories of the first person plural (`we including you' vs `we excluding you'). Verbs can be classified into dozens of different conjugation classes. Each conjugation class involves its own tone pattern; each tone pattern is based on a series of three person/number (PN) triplets. A PN triplet [X, Y, Z] consists of three tones: tone X is employed in the third person singular as well as in all plural forms; tone Y is employed in the second person singular, and tone Z, in the third person singular. Thus, a verb's membership in a particular conjugation class entails the assignment of one tone triplet to completive forms, another to progressive forms, and a third to habitual and potential forms. The paradigm of the verb lyu1 `fall' in Table illustrates: the conjugation class to which this verb belongs entails the assignment of the triplet [1, 42, 20] to the completive, [1, 42, 32] to the progressive, and [20, 42, 32] to the habitual and potential. Verbs in other conjugation classes exhibit other triplet series. ### The Resource
We provide a hand-curated collection of complete inflection tables for 198 lemmata. The morphological tags follow the guidelines of the UniMorph schema BIBREF6, BIBREF7, in order to allow for the potential of cross-lingual transfer learning, and they are tagged with respect to: Person: first (1), second (2), and third (3) Number: singular (SG) ad plural (PL) Inclusivity (only applicable to first person plural verbs: inclusive (INCL) and exclusive (EXCL) Aspect/mood: completive (CPL), progressive (PROG), potential (POT), and habitual (HAB). Two examples of complete inflection tables for the verbs ndyu2 `fell from above' and lyu1 `fall' are shown in Table . Note how the first verb has the same PN triplet for all four aspect/mood categories, while the second paradigm is more representative in that it involves three different triplets (one for the completive, another for the progressive, and another for the habitual/potential). This variety is at the core of why the SJQ verb morphology is particularly interesting, and a challenging testcase for modern NLP systems. In total, we end up with 4716 groupings (triplets) of a lemma, a tag-set, and a form; we split these groupings randomly into a training set (3774 groupings), a development set (471 groupings), and test set (471 groupings). Basic statistics of the corpus are outlined in Table 1 . Compared to all the other languages from the Unimorph project, this puts SJQ Chatino in the low- to mid-resource category, but nonetheless it is more than enough for benchmarking purposes. ### Baseline Results ::: Inflectional realization
Inflectional realization defines the inflected forms of a lexeme/lemma. As a computational task, often referred to as simply “morphological inflection," inflectional realization is framed as a mapping from the pairing of a lemma with a set of morphological tags to the corresponding word form. For example, the inflectional realization of SJQ Chatino verb forms entails a mapping of the pairing of the lemma lyu1 `fall' with the tag-set 1;SG;PROG to the word form nlyon32. Morphological inflection has been thoroughly studied in monolingual high resource settings, especially through the recent SIGMORPHON challenges BIBREF8, BIBREF9, BIBREF10, with the latest iteration focusing more on low-resource settings, utilizing cross-lingual transfer BIBREF11. We use the guidelines of the state-of-the-art approach of BIBREF12 that achieved the highest inflection accuracy in the latest SIGMORPHON 2019 morphological inflection shared task. Our models are implemented in DyNet BIBREF13. Inflection results are outlined in Table . In the `standard' setting we simply train on the pre-defined training set, achieving an exact-match accuracy of 60% over the test set. Interestingly, the data augmentation approach of BIBREF12 that hallucinates new training paradigms based on character level alignments does not heed significant improvements in accuracy (only 2 percentage points increase, cf. with more than 15 percentage points increases in other languages). These results indicate that automatic morphological inflection for low-resource tonal languages like SJQ Chatino poses a particularly challenging setting, which perhaps requires explicit handling of tone information by the model. ### Baseline Results ::: Morphological Analysis
Morphological analysis is the task of creating a morphosyntactic description for a given word. It can be framed in a context-agnostic manner (as in our case) or within a given context, as for instance for the SIGMORPHON 2019 second shared task BIBREF11. We trained an encoder-decoder model that receives the form as character-level input, encodes it with a BiLSTM encoder, and then an attention enhanced decoder BIBREF14 outputs the corresponding sequence of morphological tags, implemented in DyNet. The baseline results are shown in Table . The exact-match accuracy of 67% is lower than the average accuracy that context-aware systems can achieve, and it highlights the challenge that the complexity of the tonal system of SJQ Chatino can pose. ### Baseline Results ::: Lemmatization
Lemmatization is the task of retrieving the underlying lemma from which an inflected form was derived. Although in some languages the lemma is distinct from all forms, in SJQ Chatino the lemma is defined as the completive third-person singular form. As a computational task, lemmatization entails producing the lemma given an inflected form (and possibly, given a set of morphological tags describing the input form). Popular approaches tackle it as a character-level edit sequence generation task BIBREF15, or as a character-level sequence-to-sequence task BIBREF16. For our baseline lemmatization systems we follow the latter approach. We trained a character level encoder-decoder model, similar to the above-mentioned inflection system, implemented in DyNet. The baseline results, with and without providing gold morphological tags along with the inflected form as input, are outlined in Table . We find that automatic lemmatization in SJQ Chatino achieves fairly high accuracy even with our simple baseline models (89% accuracy, $0.27$ average Levenshtein distance) and that providing the gold morphological tags provides a performance boost indicated by small improvements on both metrics. It it worth noting, though, that these results are also well below the $94--95\%$ average accuracy and $0.13$ average Levenshtein distance that lemmatization models achieved over 107 treebanks in 66 languages for the SIGMORPHON 2019 shared task BIBREF11. ### Related Work
Our work builds and expands upon previous work on Indigenous languages of the Americas specifically focusing on the complexity of their morphology. Among other works similar to ours, BIBREF17 focused on the morphology of Dene verbs, BIBREF18 on Arapaho verbs, BIBREF19 on Shipibo-Konibo, and BIBREF20 on Saint Lawrence Island and Central Siberian Yupik. BIBREF21 describe an approach for elicit complete inflection paradigms, with experiments in languages like Nahuatl. Our resource is the first one for SJQ Chatino, but it also provides an exciting new data point in the computational study of morphological analysis, lemmatization, and inflection, as it is the first one in a tonal language with explicit tonal markings in the writing system. In a similar vein, the Oto-Manguean Inflectional Class Database BIBREF22 provides a valuable resource for studying the verbal morphology of Oto-Manguean languages (including a couple of other Chatino variants: Yaitepec and Zenzotepec Chatino) but not in a format suitable for computational experiments. ### Conclusion
We presented a resource of 198 complete inflectional paradigms in San Juan Quiahije Chatino, which will facilitate research in computational morphological analysis and inflection for low-resource tonal languages and languages of Mesoamerica. We also provide strong baseline results on computational morphological analysis, lemmatization, and inflection realization, using character-level neural encoder-decoder systems. For future work, while we will keep expanding our resource to include more paradigms, we will also follow the community guidelines in extending our resource to include morphological analysis and inflection examples in context. ### Acknowledgements
Part of this work was done during the Workshop on Language Technology for Language Documentation and Revitalization. This material is based upon work generously supported by the National Science Foundation under grant 1761548. Table 1: Basic Statistics of our resource. Table 2: Three alternative systems for representing the SJQ Chatino tones. Table 3: Complete inflection paradigms for two lemmata: one with a single PN triple across all aspects (top), and one with three different PN triples (bottom). Table 4: Morphological Inflection Results Table 6: Lemmatization Results. Table 5: Morphological Analysis Results
|
DyNet
|
On which date was a CRT-ICD system successfully implanted in Mr. Romero?
Choose the correct answer from the following options:
A. 04/07/2016
B. 2016-05-07 00:00:00
C. 10/08/2022
D. 09/23/2022
E. 05/20/2023
|
### Patient Report 0
**Dear colleague, **
We would like to report to you about our patient, Mr. David Romero, born
on 02/16/1942, who was under our inpatient care from 03/25/2016 to
03/30/2016.
**Diagnoses:**
- Suspected myocarditis
- Uncomplicated biopsy, pending results
- LifeVest has been adjusted
- Left ventricular ejection fraction of 28%
- Chronic hepatitis C
- Status post hepatitis A
- Post-antiviral therapy
- Exclusion of relevant coronary artery disease
**Medical History:** The patient was admitted with suspected myocarditis
due to a significantly impaired pump function noticed during outpatient
visits. Anamnestically, the patient reported experiencing fatigue and
exertional dyspnea since mid-December, with no recollection of a
preceding infection. Antiviral therapy with Interferon/Ribavirin for
chronic Hepatitis C had been ongoing since November. An outpatient
evaluation had excluded relevant coronary artery disease.
**Current Presentation:** Suspected inflammatory/dilated cardiomyopathy,
Indication for biopsy
**Physical Examination:** The patient is fully oriented with good
general condition and normal mental state. Dry skin and mucous
membranes, normal breathing, no cyanosis. Cranial nerves are grossly
intact, no focal neurological deficits, good muscle strength and
sensitivity all around. Clear, rhythmic heart sounds, with a 2/6
systolic murmur at the apex. Lungs are evenly ventilated without rales.
Resonant percussion. Soft and supple abdomen without guarding, spleen
and liver not palpable. Normal bowel sounds.
**Coronary Angiography**: Globally significantly impaired left
ventricular function (EF: 28%)
[Myocardial biopsy:]{.underline} Uncomplicated retrieval of LV
endomyocardial biopsies
[Recommendation]{.underline}: A conservative medical approach is
recommended, and further therapeutic decisions will depend on the
histological, immunohistological, and molecular biological examination
results of the now-retrieved myocardial biopsies.
[Procedure]{.underline}: Femoral closure system is applied, 6 hours of
bed rest, administration of 100 mg/day of Aspirin for 4 weeks following
left ventricular heart biopsy.
**Echocardiography before Heart Catheterization**:
Performed in sinus rhythm. Satisfactory ultrasound condition.
[Findings]{.underline}: Moderately dilated left ventricle (LVDd 64mm).
Markedly reduced systolic LV function (EF 28%). Global longitudinal
strain (2D speckle tracking): -8.6%.
Regional wall motion abnormalities: despite global hypokinesia, the
posterolateral wall (basal) contracts best. Diastolic dysfunction Grade
1 (LV relaxation disorder) (E/A 0.7) (E/E\' mean 13.8). No LV
hypertrophy. Morphologically age-appropriate heart valves. Moderately
dilated left atrium (LA Vol. 71ml). Mild mitral valve insufficiency
(Grade 1 on a 3-grade scale). Normal-sized right ventricle. Moderately
reduced RV function Normal-sized right atrium. Minimal tricuspid valve
insufficiency (Grade 0-1 on a 3-grade scale). Systolic pulmonary artery
pressure in the normal range (systolic PAP 27mmHg).
No thrombus detected. Minimal pericardial effusion, circular, maximum
2mm, no hemodynamic relevance.
**Echocardiography after Heart Catheterization:**
[Indication]{.underline}: Follow-up on pericardial effusion.
[Examination]{.underline}: TTE at rest, including duplex and
quantitative determination of parameters. [Echocardiographic
Finding:]{.underline} Regarding pericardial effusion, the status is the
same. Circular effusion, maximum 2mm.
**ECG after Heart Catheterization:**
76/min, sinus rhythm, complete left bundle branch block.
**Summary:** On 03/26/2016, biopsy and left heart catheterization were
successfully performed without complications. Here, too, the patient
exhibited a significantly impaired pump function, currently at 28%.
**Therapy and Progression:**
Throughout the inpatient stay, the patient remained cardiorespiratorily
stable at all times. Malignant arrhythmias were ruled out via telemetry.
After the intervention, echocardiography showed no pericardial effusion.
The results of the endomyocardial biopsies are still pending. An
appointment for results discussion and evaluation of further procedures
at our facility should be scheduled in 3 weeks. Following the biopsy,
Aspirin 100 as specified should be given for 4 weeks. We intensified the
ongoing heart failure therapy and added Spironolactone to the
medication, recommending further escalation based on hemodynamic
tolerability.
**Current Recommendations:** Close cardiological follow-up examinations,
electrolyte monitoring, and echocardiography are advised. Depending on
the left ventricular ejection fraction\'s course, the implantation of an
ICD or ICD/CRT system should be considered after 3 months. On the day of
discharge, we initiated the adjustment of a Life Vest, allowing the
patient to return home in good general condition.
**Medication upon Discharge: **
**Medication ** **Dosage ** **Frequency**
---------------------------- ------------- ---------------
Aspirin 100 mg 1-0-0
Ramipril (Altace) 2.5 mg 1-0-1
Carvedilol (Coreg) 12.5 mg 1-0-1
Torasemide (Torem) 5 mg 1-0-0
Spironolactone (Aldactone) 25 mg 1-0-0
L-Thyroxine (Synthroid) 50 µg 1-0-0
**Lab results upon Discharge:**
**Parameter** **Results** **Reference Range**
------------------------ ------------- ---------------------
Absolute Erythroblasts 0.01/nL \< 0.01/nL
Sodium 134 mEq/L 136-145 mEq/L
Potassium 4.5 mEq/L 3.5-4.5 mEq/L
Creatinine (Jaffé) 1.25 mg/dL 0.70-1.20 mg/dL
Urea 50 mg/dL 17-48 mg/dL
Total Bilirubin 1.9 mg/dL \< 1.20 mg/dL
CRP 4.1 mg/L \< 5.0 mg/L
Troponin-T 78 ng/L \< 14 ng/L
ALT 67 U/L \< 41 U/L
AST 78 U/L \< 50 U/L
Alkaline Phosphatase 151 U/L 40-130 U/L
gamma-GT 200 U/L 8-61 U/L
Free Triiodothyronine 2.3 ng/L 2.00-4.40 ng/L
Free Thyroxine 14.2 ng/L 9.30-17.00 ng/L
TSH 4.1 mU/L 0.27-4.20 mU/L
Hemoglobin 11.6 g/dL 13.5-17.0 g/dL
Hematocrit 34.5% 39.5-50.5%
Erythrocytes 3.7 /pL 4.3-5.8/pL
Leukocytes 9.56/nL 3.90-10.50/nL
MCV 92.5 fL 80.0-99.0 fL
MCH 31.1 pg 27.0-33.5 pg
MCHC 33.6 g/dL 31.5-36.0 g/dL
MPV 8.9 fL 7.0-12.0 fL
RDW-CV 14.0% 11.5-15.0%
Quick 89% 78-123%
INR 1.09 0.90-1.25
PTT Actin-FS 25.3 sec. 22.0-29.0 sec.
### Patient Report 1
**Dear colleague, **
We are reporting on the pending findings of the myocardial biopsies
taken from Mr. David Romero, born on 02/16/1942 on 03/26/2016 due to the
deterioration of LV function from 40% to 28% after interferon therapy
for HCV infection.
**Diagnoses:**
- Suspected myocarditis
- LifeVest
- Left ventricular ejection fraction of 28%
- Chronic hepatitis C
- Status post hepatitis A
- Post-antiviral therapy
- Exclusion of relevant coronary artery disease
**Current Medication:**
**Medication ** **Dosage ** **Frequency**
---------------------------- ------------- ---------------
Aspirin 100 mg 1-0-0
Ramipril (Altace) 2.5 mg 1-0-1
Carvedilol (Coreg) 12.5 mg 1-0-1
Torasemide (Torem) 5 mg 1-0-0
Spironolactone (Aldactone) 25 mg 1-0-0
L-Thyroxine (Synthroid) 50 µg 1-0-0
**Myocardial Biopsy on 01/27/2014:**
[Molecular Biology:]{.underline}
PCR examinations performed under the question of myocardial infection
with cardiotropic pathogens yielded a positive detection of HCV-specific
RNA in myocardial tissue without quantification possibility
(methodically determined). Otherwise, there was no evidence of
myocardial infection with enteroviruses, adenoviruses, Epstein-Barr
virus, Human Herpes Virus Type 6 A/B, or Erythrovirus genotypes 1/2 in
the myocardium.
[Assessment]{.underline}: Positive HCV-mRNA detection in myocardial
tissue. This positive test result does not unequivocally prove an
infection of myocardial cells, as contamination of the tissue sample
with HCV-infected peripheral blood cells cannot be ruled out in chronic
hepatitis.
**Histology and Immunohistochemistry**:
Unremarkable endocardium, normal cell content of the interstitium with
only isolated lymphocytes and histiocytes in the histologically examined
samples. Quantitatively, immunohistochemically examined native
preparations showed borderline high CD3-positive lymphocytes with a
diffuse distribution pattern at 10.2 cells/mm2. No increased
perforin-positive cytotoxic T cells. The expression of cell adhesion
molecules is discreetly elevated. Otherwise, only slight perivascular
but no interstitial fibrosis. Cardiomyocytes are properly arranged and
slightly hypertrophied (average diameter around 23 µm), the surrounding
capillaries are unremarkable. No evidence of acute
inflammation-associated myocardial cell necrosis (no active myocarditis)
and no interstitial scars from previous myocyte loss. No lipomatosis.
[Assessment:]{.underline} Based on the myocardial biopsy findings, there
is positive detection of HCV-RNA in the myocardial tissue samples, with
the possibility of tissue contamination with HCV-infected peripheral
blood cells. Significant myocardial inflammatory reaction cannot be
documented histologically and immunohistochemically. In the endocardial
samples, apart from mild hypertrophy of properly arranged
cardiomyocytes, there are no significant signs of myocardial damage
(interstitial fibrosis or scars from previous myocyte loss). Therefore,
the present findings do not indicate the need for specific further
antiviral or anti-inflammatory therapy, and the existing heart failure
medication can be continued unchanged. If LV function impairment
persists for an extended period, there is an indication for
antiarrhythmic protection of the patient using an ICD.
### Patient Report 2
**Dear colleague, **
We thank you for referring your patient Mr. David Romero, born on
02/16/1942, to us for echocardiographic follow-up on 05/04/2016.
**Diagnoses:**
- Dilatated cardiomyopathy
- LifeVest
- Left ventricular ejection fraction of 28%
- Chronic Hepatitis C
- Status post Hepatitis A
- Post-antiviral therapy
- Exclusion of relevant coronary artery disease
- Type 2 diabetes mellitus
- Hypothyroidism
**Current Medication:**
**Medication ** **Dosage ** **Frequency**
---------------------------- ------------- ---------------
Aspirin 100 mg 1-0-0
Ramipril (Altace) 2.5 mg 1-0-1
Carvedilol (Coreg) 12.5 mg 1-0-1
Torem (Torasemide) 5 mg 1-0-0
Spironolactone (Aldactone) 25 mg 1-0-0
L-Thyroxine (Synthroid) 50 µg 1-0-0
**Physical Examination:** The patient is fully oriented with good
general condition and normal mental state. Dry skin and mucous
membranes, normal breathing, no cyanosis. Cranial nerves are grossly
intact, no focal neurological deficits, good muscle strength and
sensitivity all around. Clear, rhythmic heart sounds, with a 2/6
systolic murmur at the apex. Lungs are evenly ventilated without rales.
Resonant percussion. Soft and supple abdomen without pressure pain,
spleen and liver not palpable. Normal bowel sounds.
**Echocardiography: M-mode and 2-dimensional.**
The left ventricle measures approximately 65/56 mm (normal up to 56 mm).
The right atrium and right ventricle are of normal dimensions.
Global progressive reduction in contractility, morphologically
unremarkable.
In Doppler echocardiography, normal heart valves are observed.
Mitral valve insufficiency Grade I.
[Assessment]{.underline}: Dilated cardiomyopathy with slightly reduced
left ventricular function. MI I TII °, PAP 23 mm Hg + CVP. No more
pulmonary embolism detectable.
**Summary:**
Currently, the cardiac situation is stable, LVEDD slightly decreasing.
### Patient Report 3
**Dear colleague, **
We thank you for referring your patient, Mr. David Romero, born on
02/16/1942 to us for echocardiographic follow-up on 06/15/2016.
**Diagnoses:**
- Dilatated cardiomyopathy
- LifeVest
- Left ventricular ejection fraction of 28%
- Chronic Hepatitis C
- Status post Hepatitis A
- Post-antiviral therapy
- Exclusion of relevant coronary artery disease
- Type 2 diabetes mellitus
- Hypothyroidism
**Medication upon Admission:**
**Medication ** **Dosage ** **Frequency**
---------------------------- ------------- ---------------
Aspirin 100 mg 1-0-0
Ramipril (Altace) 2.5 mg 1-0-1
Carvedilol (Coreg) 12.5 mg 1-0-1
Torasemide (Torem) 5 mg 1-0-0
Spironolactone (Aldactone) 25 mg 1-0-0
L-Thyroxine (Synthroid) 50 µg 1-0-0
**Physical Examination:** The patient is fully oriented with good
general condition and normal mental state. Dry skin and mucous
membranes, normal breathing, no cyanosis. Cranial nerves are grossly
intact, no focal neurological deficits, good muscle strength and
sensitivity all around. Clear, rhythmic heart sounds, with a 2/6
systolic murmur at the apex. Lungs are evenly ventilated without rales.
Resonant percussion. Soft and supple abdomen without guarding, spleen
and liver not palpable. Normal bowel sounds.
**Echocardiography from 06/15/2016**: Good ultrasound conditions.
The left ventricle is dilated to approximately 65/57 mm (normal up to 56
mm). The left atrium is dilated to 48 mm. Normal thickness of the left
ventricular myocardium. Ejection fraction is around 28%. Heart valves
show normal flow velocities.
**Summary:**
Currently, the cardiac situation is stable, LVEDD slightly decreasing,
potassium and creatinine levels were obtained. If EF remains this low,
an ICD may be indicated.
**Lab results from 06/15/2016:**
**Parameter** **Result** **Reference Range**
----------------------------------- ------------ ---------------------
Reticulocytes 0.01/nL \< 0.01/nL
Sodium 135 mEq/L 136-145 mEq/L
Potassium 4.8 mEq/L 3.5-4.5 mEq/L
Creatinine 1.34 mg/dL 0.70-1.20 mg/dL
BUN 49 mg/dL 17-48 mg/dL
Total Bilirubin 1.9 mg/dL \< 1.20 mg/dL
C-reactive Protein 4.1 mg/L \< 5.0 mg/L
Troponin-T 78 ng/L \< 14 ng/L
ALT 67 U/L \< 41 U/L
AST 78 U/L \< 50 U/L
Alkaline Phosphatase 151 U/L 40-130 U/L
gamma-GT 200 U/L 8-61 U/L
Free Triiodothyronine (T3) 2.3 ng/L 2.00-4.40 ng/L
Free Thyroxine (T4) 14.2 ng/L 9.30-17.00 ng/L
Thyroid Stimulating Hormone (TSH) 4.1 mU/L 0.27-4.20 mU/L
Hemoglobin 11.6 g/dL 13.5-17.0 g/dL
Hematocrit 34.5% 39.5-50.5%
Red Blood Cell Count 3.7 M/µL 4.3-5.8 M/µL
White Blood Cell Count 9.56 K/µL 3.90-10.50 K/µL
Platelet Count 280 K/µL 150-370 K/µL
MC 92.5 fL 80.0-99.0 fL
MCH 31.1 pg 27.0-33.5 pg
MCHC 33.6 g/dL 31.5-36.0 g/dL
MPV 8.9 fL 7.0-12.0 fL
RDW-CV 14.0% 11.5-15.0%
Quick 89% 78-123%
INR 1.09 0.90-1.25
Partial Thromboplastin Time 25.3 sec. 22.0-29.0 sec.
### Patient Report 4
**Dear colleague, **
We are reporting to you about Mr. David Romero, born on 02/16/1942, who
presented himself at our Cardiology University Outpatient Clinic on
06/30/2016.
**Diagnoses:**
- Dilated cardiomyopathy
- Exclusion of coronary heart diseases
- Myocardial biopsy showed no inflammation
- Left bundle branch block
- Severely impaired left ventricular (LV) function (ejection fraction
around 30%)
- LifeVest
- Planned CRT-D implantation
- Chronic Hepatitis C
- Type 2 diabetes
**Current Medication:**
**Medication ** **Dosage ** **Frequency**
---------------------------- ---------------- ---------------
Aspirin 100 mg/tablet 1-0-0
Ramipril (Altace) 2.5 mg/tablet 1-0-1
Carvedilol (Coreg) 12.5 mg/tablet 1-0-1
Torasemide (Torem) 5 mg/tablet 1-0-0
Spironolactone (Aldactone) 25 mg/tablet 1-0-0
L-Thyroxine (Synthroid) 50 µg/tablet 1-0-0
**Echocardiography on 06/30/2016:** In sinus rhythm. Adequate ultrasound
window.
Moderately dilated left ventricle (LVDd 63mm). Significantly reduced
systolic LV function (EF biplane 29%). No LV hypertrophy.
**ECG on 06/30/2016:** Sinus rhythm, regular tracing, heart rate 69/min,
complete left bundle branch block, QRS 135 ms, ERBS with left bundle
branch block.
**Assessment**: Mr. Romero presents himself for the follow-up assessment
of known dilated cardiomyopathy. He currently reports minimal dyspnea.
Coronary heart disease has been ruled out. No virus was detected
bioptically. However, the recent echocardiography still shows severely
impaired LV function.
**Current Recommendations:** Given the presence of left bundle branch
block, there is an indication for CRT-D implantation. For this purpose,
we have scheduled a pre-admission appointment, with the implantation
planned for 07/04/2016. We kindly request a referral letter. The
LifeVest should continue to be worn until the implantation, despite the
pressure sores on the thorax.
### Patient Report 5
**Dear colleague, **
We would like to report to you about our patient, Mr. David Romero, born
on 02/16/1942, who was in our inpatient care from 07/04/2016 to
07/06/2016.
**Diagnoses:**
- Dilated cardiomyopathy
- Exclusion of coronary heart diseases
- Myocardial biopsy showed no inflammation
- Left bundle branch block
- Severely impaired left ventricular (LV) function (ejection fraction
around 30%)
- LifeVest
- Planned CRT-D implantation
- Chronic Hepatitis C
- Type 2 diabetes
**Medication upon Admission:**
**Medication ** **Dosage ** **Frequency**
---------------------------- ------------- ---------------
Aspirin 100 mg 1-0-0
Ramipril (Altace) 2.5 mg 1-0-1
Carvedilol (Coreg) 12.5 mg 1-0-1
Torem (Torasemide) 5 mg 1-0-0
Spironolactone (Aldactone) 25 mg 1-0-0
L-Thyroxine (Synthroid) 50 µg 1-0-0
Sitagliptin (Januvia) 100 mg 1-0-0
Insulin glargine (Lantus) 0-0-20IE
**Current Presentation:** The current admission was elective for CRT-D
implantation in dilated cardiomyopathy with severely impaired LV
function despite full heart failure medication and complete left bundle
branch block. Please refer to previous medical records for a detailed
history. On 07/05/2016, a CRT-ICD system was successfully implanted. The
peri- and post-interventional course was uncomplicated. Pneumothorax was
ruled out post-interventionally. The wound conditions are
irritation-free. The ICD card was given to the patient. We request
outpatient follow-up on the above-mentioned date for wound inspection
and CRT follow-up. Please adjust the known cardiovascular risk factors.
**Findings:**
**ECG upon Admission:** Sinus rhythm 66/min, PQ 176ms, QRS 126ms, QTc
432ms, Complete left bundle branch block with corresponding excitation
regression disorder.
**Procedure**: Implantation of a CRT-D with left ventricular multipoint
pacing left pectoral. Smooth triple puncture of the lateral left
subclavian vein and implantation of an active single-coil electrode in
the RV apex with very good electrical values. Trouble-free probing of
the CS and direct venography using a balloon occlusion catheter.
Identification of a suitable lateral vein and implantation of a
quadripolar electrode (Quartet, St. Jude Medical) with very good
electrical values. No phrenic stimulation up to 10 volts in all
polarities. Finally, implantation of an active P/S electrode in the
right atrial roof with equally very good electrical values. Connection
to the device and submuscular implantation. Wound irrigation and layered
wound closure with absorbable suture material. Finally, extensive
testing of all polarities of the LV electrode and activation of
multipoint pacing. Final setting of the ICD.
**Chest X-ray on 07/05/2016:**
[Clinical status, question, justifying indication:]{.underline} History
of CRT-D implantation. Question about lead position, pneumothorax?
**Findings**: New CRT-D unit left pectoral with leads projected onto the
right ventricle, the right atrium, and the sinus coronarius. No
pneumothorax.
Normal heart size. No pulmonary congestion. No diffuse infiltrates. No
pleural effusions.
**ECG at Discharge:** Continuous ventricular PM stimulation, HR: 66/min.
**Current Recommendations:**
- We request a follow-up appointment in our Pacemaker Clinic. Please
provide a referral slip.
- We ask for the protection of the left arm and avoidance of
elevations \> 90 degrees. Self-absorbing sutures have been used.
- We request regular wound checks.
### Patient Report 6
**Dear colleague, **
We thank you for referring your patient, Mr. David Romero, born on
02/16/1942, who presented to our Cardiological University Outpatient
Clinic on 08/26/2016.
**Diagnoses:**
- Dilated cardiomyopathy
- Exclusion of coronary heart diseases
- Myocardial biopsy showed no inflammation
- Left bundle branch block
- Severely impaired left ventricular (LV) function
- LifeVest
- CRT-D implantation
- Chronic Hepatitis C
- Type 2 diabetes
**Current Medication:**
**Medication ** **Dosage ** **Frequency**
---------------------------- ------------- ---------------
Aspirin 100 mg 1-0-0
Ramipril (Altace) 2.5 mg 1-0-1
Carvedilol (Coreg) 12.5 mg 1-0-1
Torem (Torasemide) 5 mg 1-0-0
Spironolactone (Aldactone) 25 mg 1-0-0
L-Thyroxine (Synthroid) 50 µg 1-0-0
Sitagliptin (Januvia) 100 mg 1-0-0
Insulin glargine (Lantus) 0-0-20IE
**Current Presentation**: Slightly increasing exertional dyspnea, no
coronary heart disease.
**Cardiovascular Risk Factors:**
- Family history: No
- Smoking: No
- Hypertension: No
- Diabetes: Yes
- Dyslipidemia: Yes
**Physical Examination:** The patient is fully oriented with good
general condition and normal mental state. Dry skin and mucous
membranes, normal breathing, no cyanosis. Cranial nerves are grossly
intact, no focal neurological deficits, good muscle strength and
sensitivity all around. Clear, rhythmic heart sounds, with a 2/6
systolic murmur at the apex. Lungs are evenly ventilated without rales.
Resonant percussion. Soft and supple abdomen without pressure pain,
spleen and liver not palpable. Normal bowel sounds.
**Findings**:
**Resting ECG:** Sinus rhythm, 83 bpm. Blood pressure: 120/70 mmHg.
**Echocardiography: M-mode and 2-dimensional**
Left ventricle dimensions: Approximately 57/45 mm (normal up to 56 mm),
moderately dilated
- Right atrium and right ventricle: Normal dimensions
- Normal thickness of left ventricular muscle
- Globally, mild reduction in contractility
- Heart valves: Morphologically normal
- Doppler-Echocardiography: No significant valve regurgitation
**Assessment**: Mildly dilated cardiomyopathy with slightly reduced left
ventricular function. Ejection fraction at 45 - 50%. Mild diastolic
dysfunction. Mild tricuspid regurgitation, pulmonary artery pressure 22
mm Hg, and left ventricular filling pressure slightly increased.
**Stress Echocardiography: Stress echocardiography with exercise test**
- Stress test protocol: Treadmill exercise test
- Reason for stress test: Exertional dyspnea
- Quality of the ultrasound: Good
- Initial workload: 50 watts
- Maximum workload achieved: 150 Watt
- Blood pressure response: Systolic BP increased from 112/80 mmHg to
175/90 mmHg
- Heart rate response: Increased from 71bpm to 124bpm
- Exercise terminated due to leg pain
**Resting ECG:** Sinus rhythm**.** No significant changes during
exercise
**Echocardiography at rest:** Normokinesis of all left ventricular
segments EF: 45 - 50%
**Echocardiography during exercise:** Increased contractility and wall
thickening of all segments
[Summary]{.underline}: No dynamic wall motion abnormalities. No evidence
of exercise-induced myocardial ischemia
**Carotid Doppler Ultrasound:** Both common carotid arteries are
smooth-walled**.** Intima-media thickness: 0.8 mm**.** Small plaque in
the carotid bulb on both sides**.** Normal flow in the internal and
external carotid arteries**.** Normal dimensions and flow in the
vertebral arteries
**Summary:** Non-obstructive carotid plaques**.** Indicated to lower LDL
to below 1.8 mmol/L
**Summary:**
- Stress echocardiography shows no evidence of ischemia, EF \>45-50%
- Carotid duplex shows minimal non-obstructive plaques
- Increase Simvastatin to 20 mg, target LDL-C \< 1.8 mmol/L
### Patient Report 7
**Dear colleague, **
We would like to inform you about the results of the cardiac
catheterization of Mr. David Romero, born on 02/16/1942 performed by us
on 08/10/2022.
**Diagnoses:**
- Dilated cardiomyopathy
- Exclusion of coronary heart diseases
- Myocardial biopsy showed no inflammation
- Left bundle branch block
- Severely impaired left ventricular (LV) function
- LifeVest
- CRT-D implantation
- Chronic Hepatitis C
- Type 2 diabetes
**Procedure:** Right femoral artery puncture. Left ventriculography with
a 5F pigtail catheter in the right anterior oblique projection. Coronary
angiography with 5F JL4.0 and 5F JR 4.0 catheters. End-diastolic
pressure in the left ventricle within the normal range, measured in
mmHg. No pathological pressure gradient across the aortic valve.
**Coronary angiography:**
- Unremarkable left main stem.
- The left anterior descending (LAD) artery shows mild wall changes,
with a maximum stenosis of 20-\<30%.
- The robust right coronary artery (RCA) is stenosed proximally by
30-40%, subsequently ectatic and then stenosed to 40-\<50% distally.
Slow contrast clearance. The right coronary artery is also stenosed
up to 30%.
- Left-dominant coronary circulation.
**Assessment**: Diffuse coronary atherosclerosis with less than 50%
stenosis in the RCA and evidence of endothelial dysfunction.
**Current Recommendations:**
- Initiation of Ranolazine
- Additional stress myocardial perfusion scintigraphy
### Patient Report 8
**Dear colleague, **
We would like to inform you about the results of the Myocardial
Perfusion Scintigraphy performed on our patient, Mr. David Romero, born
on 02/16/1942, on 09/23/2022.
**Diagnoses:**
- Dilated cardiomyopathy
- Exclusion of coronary heart diseases
- Myocardial biopsy showed no inflammation
- Left bundle branch block
- Severely impaired left ventricular (LV) function (ejection fraction
around 30%)
- LifeVest
- CRT-D implantation
- Chronic Hepatitis C
- Type 2 diabetes
**Physical Examination:** The patient is fully oriented with good
general condition and normal mental state. Dry skin and mucous
membranes, normal breathing, no cyanosis. Cranial nerves are grossly
intact, no focal neurological deficits, good muscle strength and
sensitivity all around. Clear, rhythmic heart sounds, with a 2/6
systolic murmur at the apex. Lungs are evenly ventilated without rales.
Resonant percussion. Soft and supple abdomen without guarding, spleen
and liver not palpable. Normal bowel sounds.
**Myocardial Perfusion Scintigraphy:**
The myocardial perfusion scintigraphy was conducted using 365 MBq of
99m-Technetium MIBI during pharmacological stress and 383 MBq of
99m-Technetium MIBI at rest.
[Technique]{.underline}: Initially, the patient was pharmacologically
stressed with the intravenous administration of 400 µg of Regadenoson
over 20 seconds, accompanied by ergometer exercise at 50 W.
Subsequently, the intravenous injection of the radiopharmaceutical was
performed. The maximum blood pressure achieved during the stress phase
was 143/84 mm Hg, and the maximum heart rate reached was 102 beats per
minute.
Approximately 60 minutes later, ECG-triggered acquisition of a
360-degree SPECT study was conducted with reconstructions of short and
long-axis slices.
Due to inhomogeneities in the myocardial wall segments during stress,
rest images were acquired on another examination day. Following the
intravenous injection of the radiopharmaceutical, ECG-triggered
acquisition of a 360-degree SPECT study was performed, including
short-axis and long-axis slices, approximately 60 minutes later.
[Clinical Information:]{.underline} Known coronary heart disease (RCA
50%). ICD/CRT pacemaker.
[Findings]{.underline}: No clear perfusion defects are seen in the
scintigraphic images acquired after pharmacologic exposure to
Regadenoson. This finding remains unchanged in the scintigraphic images
acquired at rest.
Quantitative analysis shows a normal-sized ventricle with a normal left
ventricular ejection fraction (LVEF) of 53% under exercise conditions
and 47% at rest (EDV 81 mL). There are no clear wall motion
abnormalities. In the gated SPECT analysis, there are no definite wall
motion abnormalities observed in both stress and rest conditions.
**Quantitative Scoring:**
- SSS (Summed Stress Score): 3 (4.4%)
- SRS (Summed Rest Score): 0 (0.0%)
- SDS (Summed Difference Score): 3 (4.4%)
**Assessment**: No evidence of myocardial perfusion defects with
Regadenoson stress or at rest. Normal ventricular size and function with
no significant wall motion abnormalities.
### Patient Report 9
**Dear colleague, **
We would like to report on our patient, Mr. David Romero, born on
02/16/1942, who was under our inpatient care from 05/20/2023 to
05/21/2023.
**Diagnoses:**
- Dilated cardiomyopathy
- Exclusion of coronary heart diseases
- Myocardial biopsy showed no inflammation
- Left bundle branch block
- Severely impaired left ventricular (LV) function
- LifeVest
- CRT-D implantation
- Chronic Hepatitis C
- Type 2 diabetes
**Medical History:** The patient was admitted for device replacement due
and upgrading to a CRT-P pacemaker. At admission, the patient reported
no complaints of fever, cough, dyspnea, chest pain, or melena.
**Physical Examination:** The patient is fully oriented with good
general condition and normal mental state. Dry skin and mucous
membranes, normal breathing, no cyanosis. Cranial nerves are grossly
intact, no focal neurological deficits, good muscle strength and
sensitivity all around. Clear, rhythmic heart sounds, with a 2/6
systolic murmur at the apex. Lungs are evenly ventilated without rales.
Resonant percussion. Soft and supple abdomen without guarding, spleen
and liver not palpable. Normal bowel sounds.
**Medication upon Admission**
**Medication** **Dosage** **Frequency**
--------------------------- -------------- ----------------------
Insulin glargine (Lantus) 450 E/1.5 ml 0-0-0-6-8 IU
Insulin lispro (Humalog) 300 E/3 ml 5-8 IU-5-8 IU-5-8 IU
Levothyroxine (Synthroid) 100 mcg 1-0-0-0
Colecalciferol 12.5 mcg 2-0-0-0
Atorvastatin (Lipitor) 21.7 mg 0-0-1-0
Amlodipine (Norvasc) 6.94 mg 1-0-0-0
Ramipril (Altace) 5 mg 1-0-0-0
Torasemide (Torem) 5 mg 0-0-0.5-0
Carvedilol (Coreg) 25 mg 0.5-0-0.5-0
Simvastatin (Zocor) 40 mg 0-0-0.5-0
Aspirin 100 mg 1-0-0-0
**Therapy and Progression:** The patient\'s current admission was
elective for the implantation of a 3-chamber CRT-D device due to device
depletion. The procedure was performed without complications on
05/20/2023. The post-interventional course was uneventful. The
implantation site showed no irritation or significant hematoma at the
time of discharge, and no pneumothorax was detected on X-ray.
To protect the surgical wound, we request dry wound dressing for the
next 10 days and clinical wound checks. Suture removal is not necessary
with absorbable suture material. We advise against arm elevation for the
next 4 weeks, avoiding heavy lifting on the side of the device pocket
and gradual, pain-adapted full range of motion after 4 weeks.
**Current Recommendations:** We kindly request an outpatient follow-up
appointment in our Pacemaker Clinic.
**Medication upon Discharge:**
**Medication ** **Dosage ** **Frequency**
----------------------------- --------------- -----------------------
Insulin glargine (Lantus) 450 E./1.5 ml 0-0-0-/6-8 IU
Insulin lispro (Humalog) 300 E./3 ml 5-8 IU/-5-8 IU/5-8 IU
Levothyroxine (Synthroid) 100 µg 1-0-0-0
Colecalciferol (Vitamin D3) 12.5 µg 2-0-0-0
Atorvastatin (Lipitor) 21.7 mg 0-0-1-0
Amlodipine (Norvasc) 6.94 mg 1-0-0-0
Ramipril (Altace) 5 mg 1-0-0-0
Torasemide (Torem) 5 mg 0-0-0.5-0
Carvedilol (Coreg) 25 mg 0.5-0-0.5-0
Simvastatin (Zocor) 40 mg 0-0-0.5-0
Aspirin 100 mg 1-0-0-0
Colecalciferol 12.5 µg 2-0-0-0
**Addition: Findings:**
**ECG at Discharge:** Sinus rhythm, ventricular pacing, QRS 122ms, QTc
472ms
**Rhythm Examination on 05/20/2023:**
[Results:]{.underline} Replacement of a 3-chamber CRT-D device (new:
SJM/Abbott Quadra Assura) due to impending battery depletion:
Uncomplicated replacement. Tedious freeing of the submuscular device and
proximal lead portions using a plasma blade. Extraction of the old
device. Connection to the new device. Avoidance of device fixation in
the submuscular position. Hemostasis by electrocauterization. Layered
wound closure. Skin closure with absorbable intracutaneous sutures. End
adjustment of the CRT-D device is complete. [Procedure]{.underline}:
Compression of the wound with a sandbag and local cooling. First
outpatient follow-up in 8 weeks through our pacemaker clinic (please
schedule an appointment before discharge). Postoperative chest X-ray is
not necessary. Cefuroxime 1.5 mg again tonight.
**Transthoracic Echocardiography on 05/18/2023**
**Results:** Globally mildly impaired systolic LV function. Diastolic
dysfunction Grade 1 (LV relaxation disorder).
- Right Ventricle: Normal-sized right ventricle. Normal RV function.
Pulmonary arterial pressure is normal.
- Left Atrium: Slightly dilated left atrium.
- Right Atrium: Normal-sized right atrium.
- Mitral Valve: Morphologically unremarkable. Minimal mitral valve
regurgitation.
- Aortic Valve: Mildly sclerotic aortic valve cusps. No aortic valve
insufficiency. No aortic valve stenosis (AV PGmax 7 mmHg).
- Tricuspid Valve: Delicate tricuspid valve leaflets. Minimal
tricuspid valve regurgitation (TR Pmax 26 mmHg).
- Pulmonary Valve: No pulmonary valve insufficiency. Pericardium: No
pericardial effusion.
**Assessment**: Examination in sinus rhythm with bundle branch block.
Moderate ultrasound windows. Normal-sized left ventricle (LVED 54 mm)
with mildly reduced systolic LV function (EF biplan 55%) with mildly
reduced contractility without regional emphasis. Mild LV hypertrophy,
predominantly septal, without obstruction. Diastolic dysfunction Grade 1
(E/A 0.47) with a normal LV filling index (E/E\' mean 3.5). Slightly
sclerotic aortic valve without stenosis, no AI. Slightly dilated left
atrium (LAVI 31 ml/m²). Minimal MI. Normal-sized right ventricle with
normal function. Normal-sized right atrium (RAVI 21 ml/m²). Minimal TI.
As far as assessable, systolic PA pressure is within the normal range.
The IVC cannot be viewed from the subcostal angle. No thrombi are
visible. As far as assessable, no pericardial effusion is visible.
**Chest X-ray in two planes on 05/20/2023: **
[Clinical Information, Question, Justification:]{.underline} Post CRT
device replacement. Inquiry about position, pneumothorax.
[Findings]{.underline}: No pneumothorax following CRT device
replacement.
### Patient Report 0
**Dear colleague, **
We are writing to provide an update on Mr. David Romero, born on
02/16/1942, who presented at our Rhythm Clinic on 09/29/2023.
**Diagnoses:**
- Dilated cardiomyopathy
- Exclusion of coronary heart diseases
- Myocardial biopsy showed no inflammation
- Left bundle branch block
- Severely impaired left ventricular (LV) function
- LifeVest
- CRT-D implantation
- Chronic Hepatitis C
- Type 2 diabetes
**Current Medication:**
**Medication** **Dosage** **Frequency**
----------------------------- ------------------ ---------------
Lantus (Insulin glargine) 450 Units/1.5 mL 0-0-0-/6-8
Humalog (Insulin lispro) 300 Units/3 mL 5-8/0/5-8/5-8
Levothyroxine (Synthroid) 100 mcg 1-0-0-0
Vitamin D3 (Colecalciferol) 12.5 mcg 2-0-0-0
Lipitor (Atorvastatin) 21.7 mg 0-0-1-0
Norvasc (Amlodipine) 6.94 mg 1-0-0-0
Altace (Ramipril) 5 mg 1-0-0-0
Demadex (Torasemide) 5 mg 0-0-0.5-0
Coreg (Carvedilol) 25 mg 0.5-0-0.5-0
Zocor (Simvastatin) 40 mg 0-0-0.5-0
Aspirin 100 mg 1-0-0-0
Vitamin D3 (Colecalciferol) 12.5 mcg 2-0-0-0
**Measurement Results:**
Battery/Capacitor: Status: OK, Voltage: 8.4V
- Right Atrial: 375 Ohms 3.80 mV 0.375 V 0.50 ms
- Right Ventricular: 388 Ohms 11.80 mV 0.750 V 0.50 ms
- Left Ventricular: 350 Ohms 0.625 V 0.50 ms
- Defibrillation Impedance: Right Ventricular: 48 Ohms
**Implant Settings:**
- Bradycardia Setting: Mode: DDD
- Tachycardia Settings: Zone Detection Interval (ms) Detection Beats
ATP Shocks Details Status
- VFVF 260 ms 30 /
- VTVT1 330 ms 55 /
<!-- -->
- Probe Settings: Lead Sensitivity Sensing Polarity/Vector
Amplification/Pulse Width Stimulation Polarity/Vector Auto Amplitude
Control
- Right Atrial: 0.30 mV Bipolar/ 1.375 V/0.50 ms Bipolar/
- Right Ventricular: Bipolar/ 2.000 V/0.50 ms Bipolar/
- Left Ventricular: 2.000 V/0.50 ms tip 1 - RV Coil
**Assessment:**
- Routine visit with normal device function.
- Normal sinus rhythm with a heart rate of 65/min.
- Balanced heart rate histogram with a plateau at 60-70 bpm.
- Wound conditions are unremarkable.
- Battery status: OK.
- Atrial probe: Intact
- Right ventricular probe: Intact
- Left ventricular probe: Intact
- A follow-up appointment for the patient is requested in 6 months.
**Lab results:**
**Parameter** **Result** **Reference Range**
----------------------------------- ------------ ---------------------
Reticulocytes 0.01/nL \< 0.01/nL
Sodium 137 mEq/L 136-145 mEq/L
Potassium 4.2 mEq/L 3.5-4.5 mEq/L
Creatinine 1.34 mg/dL 0.70-1.20 mg/dL
BUN 49 mg/dL 17-48 mg/dL
Total Bilirubin 1.8 mg/dL \< 1.20 mg/dL
C-reactive Protein 5.9 mg/L \< 5.0 mg/L
ALT 67 U/L \< 41 U/L
AST 78 U/L \< 50 U/L
Alkaline Phosphatase 151 U/L 40-130 U/L
Gamma-Glutamyl Transferase 200 U/L 8-61 U/L
Free Triiodothyronine (T3) 2.3 ng/L 2.00-4.40 ng/L
Free Thyroxine (T4) 14.2 ng/L 9.30-17.00 ng/L
Thyroid Stimulating Hormone (TSH) 4.1 mU/L 0.27-4.20 mU/L
Hemoglobin 11.6 g/dL 13.5-17.0 g/dL
Hematocrit 34.5% 39.5-50.5%
Red Blood Cell Count 3.7 M/µL 4.3-5.8 M/µL
White Blood Cell Count 9.56 K/µL 3.90-10.50 K/µL
MCV 92.7 fL 80.0-99.0 fL
MCH 31.8 pg 27.0-33.5 pg
MCHC 33.9 g/dL 31.5-36.0 g/dL
MPV 8.9 fL 7.0-12.0 fL
RDW-CV 14.2% 11.5-15.0%
Quick 89% 78-123%
INR 1.09 0.90-1.25
Partial Thromboplastin Time 25.3 sec. 22.0-29.0 sec.
|
2016-05-07 00:00:00
|
By how much does their model outperform the baseline in the cross-domain evaluation?
|
### Introduction
Sentiment Analysis (SA) is an active field of research in Natural Language Processing and deals with opinions in text. A typical application of classical SA in an industrial setting would be to classify a document like a product review into positve, negative or neutral sentiment polarity. In constrast to SA, the more fine-grained task of Aspect Based Sentiment Analysis (ABSA) BIBREF0, BIBREF1 aims at finding both the aspect of an entity like a restaurant and the sentiment associated with this aspect. It is important to note that ABSA comes in two variants. We will use the sentence “I love their dumplings” to explain these variants in detail. Both variants are implemented as a two-step procedure. The first variant is comprised of Aspect-Category Detection (ACD) followed by Aspect-Category Sentiment Classification (ACSC). ACD is a multilabel classification task, where a sentence can be associated with a set of predefined aspect categories like "food" and "service" in the restaurants domain. In the second step, ACSC, the sentiment polarity associated to the aspect is classified. For our example-sentence the correct result is (“food”, “positive”). The second variant consists of Aspect-Target Extraction (ATE) followed by Aspect-Target Sentiment Classification (ATSC). ATE is a sequence labeling task, where terms like “dumplings” are detected. In the second step, ATSC, the sentiment polarity associated to the aspect-target is determined. In our example the correct result is the tuple ("dumplings", "positive"). In this work, we focus on ATSC. In the last years, specialized neural architectures BIBREF2, BIBREF3 have been developed that substantially improved modeling of this target-context relationship. More recently, the Natural Language Processing community experienced a substantial shift towards using pre-trained language models BIBREF4, BIBREF5, BIBREF6, BIBREF7 as a base for many down-stream tasks, including ABSA BIBREF8, BIBREF9, BIBREF10. We still see huge potential that comes with this trend, this is why we approach the ATSC task using the BERT architecture. As shown by BIBREF9, for the ATSC task the performance of models that were pre-trained on general text corpora is improved substantially by finetuning the model on domain-specific corpora — in their case review corpora — that have not been used for pre-training BERT, or other language models. We extend the work by Xu et al. by further investigating the behavior of finetuning the BERT language model in relation to ATSC performance. In particular, our contributions are: The analysis of the influence of the amount of training-steps used for BERT language model finetuning on the Aspect-Target Sentiment Classification performance. The findings on how to exploit BERT language model finetuning enables us to achieve new state-of-the-art performance on the SemEval 2014 restaurants dataset. The analysis of cross-domain adaptation between the laptops and restaurants domain. Adaptation is tested by finetuning the BERT language model self-supervised on the target-domain and then supervised training on the ATSC task in the source-domain. In addition, the performance of training on the combination of both datasets is measured. ### Related Works
We separate our discussion of related work into two areas: First, neural methods applied to ATSC that have improved performance solely by model architecture improvements. Secondly, methods that additionally aim to transfer knowledge from semantically related tasks or domains. ### Related Works ::: Architecture Improvements for Aspect-Target Sentiment Classification
The datasets typically used for Aspect-Target Sentiment Classification are the SemEval 2014 Task 4 datasets BIBREF1 for the restaurants and laptops domain. Unfortunately, both datasets only have a small number of training examples. One common approach to compensate for insufficient training examples is to invent neural architectures that better model ATSC. For example, in the past a big leap in classification performance was achieved with the use of the Memory Network architecture BIBREF3, which uses memory to remember context words and explicitly models attention over both the target word and context. It was found that making full use of context words improves their model compared to previous models BIBREF2 that make use of left- and right-sided context independently. BIBREF8 proposed Attention Encoder Networks (AEN), a modification to the transformer architecture. The authors split the Multi-Head Attention (MHA) layers into Intra-MHA and Inter-MHA layers in order to model target words and context differently, which results in a more lightweight model compared to the transformer architecture. Another recent performance leap was achieved by BIBREF11, who model dependencies between sentiment words explicitly in sentences with more than one aspect-target by using a graph convolutional neural network. They show that their architecture performs particularly well if multiple aspects are present in a sentence. ### Related Works ::: Knowledge Transfer for Aspect-Target Sentiment Classification Analysis
Another approach to compensate for insufficient training examples is to transfer knowledge across domains or across similar tasks. BIBREF12 proposed Multi-Granularity Alignment Networks (MGAN). They use this architecture to transfer knowledge from both an aspect-category classification task and also across different domains. They built a large scale aspect-category dataset specifically for this. BIBREF13 transfer knowledge from a document-level sentiment classification task trained on the amazon review dataset BIBREF14. They successfully apply pre-training by reusing the weights of a Long Short Term Memory (LSTM) network BIBREF15 that has been trained on the document-level sentiment task. In addition, they apply multi-task learning where aspect and document-level tasks are learned simultaneously by minimizing a joint loss function. Similarly, BIBREF9 introduce a multi-task loss function to simultaneously optimize the BERT model's BIBREF7 pre-training objectives as well as a question answering task. In contrast to the methods described above that aim to transfer knowledge from a different source task like question answering or document-level sentiment classification, this paper aims at transferring knowledge across different domains by finetuning the BERT language model. ### Methodology
We approach the Aspect-Target Sentiment Classification task using a two-step procedure. We use the pre-trained BERT architecture as a basis. In the first step we finetune the pre-trained weights of the language model further in a self-supervised way on a domain-specific corpus. In the second step we train the finetuned language model in a supervised way on the ATSC end-task. In the following subsections, we discuss the BERT architecture, how we finetune the language model, and how we transform the ATSC task into a BERT sequence-pair classification task BIBREF10. Finally, we discuss the different end-task training and domain-specific finetuning combinations we employ to evaluate our model's generalization performance not only in-domain but also cross-domain. ### Methodology ::: BERT
The BERT model builds on many previous innovations: contextualized word representations BIBREF4, the transformer architecture BIBREF16, and pre-training on a language modeling task with subsequent end-to-end finetuning on a downstream task BIBREF5, BIBREF6. Due to being deeply bidirectional, the BERT architecture creates very powerful sequence representations that perform extremely well on many downstream tasks BIBREF7. The main innovation of BERT is that instead of using the objective of next-word prediction a different objective is used to train the language model. This objective consists of 2 parts. The first part is the masked language model objective, where the model learns to predict tokens, which have been randomly masked, from the context. The second part is the next-sequence prediction objective, where the model needs to predict if a sequence $B$ would naturally follow the previous sequence $A$. This objective enables the model to capture long-term dependencies better. Both objectives are discussed in more detail in the next section. As a base for our experiments we use the BERTBASE model, which has been pre-trained by the Google research team. It has the following parameters: 12 layers, 768 hidden dimensions per token and 12 attention heads. It has 110 Mio. parameters in total. For finetuning the BERT language model on a specific domain we use the weights of BERTBASE as a starting point. ### Methodology ::: BERT Language Model Finetuning
As the first step of our procedure we perform language model finetuning of the BERT model using domain-specific corpora. Algorithmically, this is equivalent to pre-training. The domain-specific language model finetuning as an intermediate step to ATSC has been shown by BIBREF9. As an extension to their paper we investigate the limits of language model finetuning in terms of how end-task performance is dependent on the amount of training steps. The training input representation for language model finetuning consists of two sequences $s_A$ and $s_B$ in the format of $"\textrm {[CLS]} \ s_{A} \ \textrm {[SEP]} \ s_{B} \ \textrm {[SEP]}"$, where [CLS] is a dummy token used for downstream classification and [SEP] are separator tokens. ### Methodology ::: BERT Language Model Finetuning ::: Masked Language Model Objective
The sequences $A$ and $B$ have tokens randomly masked out in order for the model to learn to predict them. The following example shows why domain-specific finetuning can alleviate the bias from pre-training on a Wikipedia corpus: "The touchscreen is an [MASK] device". In the fact-based context of Wikipedia the [MASK] could be "input" and in the review domain a typical guess could be the general opinion word "amazing". ### Methodology ::: BERT Language Model Finetuning ::: Next-Sentence Prediction
In order to train BERT to capture long-term dependencies better, the model is trained to predict if sequence $B$ follows sequence $A$. If this is the case, sequence A and sequence B are jointly sampled from the same document in the order they are occuring naturally. Otherwise the sequences are sampled randomly from the training corpus. ### Methodology ::: Aspect-Target Sentiment Classification
The ATSC task aims at classifying sentiment polarity into the three classes positive, negative, neutral with respect to an aspect-target. The input to the classifier is a tokenized sentence $s=s_{1:n}$ and a target $t=s_{j:j+m}$ contained in the sentence, where $j < j+m \le n$. Similar to previous work by BIBREF10, we transform the input into a format compatible with BERT sequence-pair classification tasks: $"\textrm {[CLS]} \ s \ \textrm {[SEP]} \ t \ \textrm {[SEP]}"$. In the BERT architecture the position of the token embeddings is structurally maintained after each Multi-Head Attention layer. Therefore, we refer to the last hidden representation of the [CLS] token as $h_{[CLS]} \in \mathbf {R}^{768 \times 1}$. The number of sentiment polarity classes is three. A distribution $p \in [0,1]^3$ over these classes is predicted using a fully-connected layer with 3 output neurons on top of $h_{[CLS]}$, followed by a softmax activation function where $b \in \mathbf {R}^3$ and $W \in \mathbf {R}^{3 \times 768}$. Cross-entropy is used as the training loss. The way we use BERT for classifying the sentiment polaritites is equivalent to how BERT is used for sequence-pair classification tasks in the original paper BIBREF7. ### Methodology ::: Domain Adaptation through Language Model Finetuning
In academia, it is common that the performance of a machine learning model is evaluated in-domain. This means that the model is evaluated on a test set that comes from the same distribution as the training set. In real-world applications this setting is not always valid, as the trained model is used to predict previously unseen data. In order to evaluate the performance of a machine learning model more robustly, its generalization error can be evaluated across different domains, i.e. cross-domain. Additionally, the model itself can be adapted towards a target domain. This is known as Domain Adaptation, which is a special case of Transductive Transfer Learning in the taxonomy of BIBREF17. Here, it is typically assumed that supervised data for a specific task is only available for a source domain $S$, whereas only unsupervised data is available in the target domain $T$. The goal is to optimize performance of the task in the target domain while transferring task-specific knowledge from the source domain. If we map this framework to our challenge, we define Aspect-Target Sentiment Classification as the transfer-task and BERT language model finetuning is used for domain adaptation. In terms of on which domain is finetuned on, the full transfer-procedure can be expressed in the following way: Here, $D_{LM}$ stands for the domain on which the language model is finetuned and can take on the values of Restaurants, Laptops or (Restaurants $\cup $ Laptops). The domain for training $D_{Train}$ can take on the same values, for the joint case case the training datasets for laptops and restaurants are simply combined. The domain for testing $D_{Test}$ can only be take on the values Restaurants or Laptops. Combining finetuning and training steps gives us nine different evaluation scenarios, which we group into the following four categories: ### Methodology ::: In-Domain Training
ATSC is trained on a domain-specific dataset and evaluated on the test set from the same domain. This can be expressed as $D_{LM} \rightarrow T \rightarrow T,$ where $T$ is our target domain and can be either Laptops or Restaurants. It is expected that the performance of the model is best if $D_{LM} = T$. ### Methodology ::: Cross-Domain Training
ATSC is trained on a domain-specific dataset and evaluated on the test set from the other domain. This can be expressed as $D_{LM} \rightarrow S \rightarrow T,$ where $S\ne T$ are source and target domain and can be either Laptops or Restaurants. ### Methodology ::: Cross-Domain Adaptation
As a special case of cross-domain Training we expect performance to be optimal if $D_{LM} = T$. This is the variant of Domain Adaptation and is written as $T \rightarrow S \rightarrow T.$ ### Methodology ::: Joint-Domain Training
ATSC is trained on both domain-specific datasets jointly and evaluated on both test sets independently. This can be expressed as $D_{LM} \rightarrow (S \cup T) \rightarrow T,$ where $S\ne T$ are source- and target domain and can either be Laptops or Restaurants. ### Experiments
In our experiments we aim to answer the following research questions (RQs): RQ1: How does the number of training iterations in the BERT language model finetuning stage influence the ATSC end-task performance? At what point does performance start to improve, when does it converge? RQ2: When trained in-domain, what ATSC end-task performance can be reached through fully exploitet finetuning of the BERT language model? RQ3: When trained cross-domain in the special case of domain adaptation, what ATSC end-task performance can be reached if BERT language model finetuning is fully exploitet? ### Experiments ::: Datasets for Classification and Language Model Finetuning
We conduct experiments using the two SemEval 2014 Task 4 Subtask 2 datasets BIBREF1 for the laptops and the restaurants domain. The two datasets contain sentences with multiple marked aspect terms that each have a 3-level sentiment polarity (positive, neutral or negative) associated. In the original dataset the conflict label is also present. Here, conflicting labels are dropped for reasons of comparability with BIBREF9. Both datasets are small, detailed statistics are shown in tab:datasets. For BERT language model finetuning we prepare three corpora for the two domains of laptops and restaurants. For the restaurants domain we use Yelp Dataset Challenge reviews and for the laptops domain we use Amazon Laptop reviews BIBREF14. For the laptop domain we filtered out reviews that appear in the SemEval 2014 laptops dataset to avoid training bias for the test data. To be compatible with the next-sentence prediction task used during fine tuning, we removed reviews containing less than two sentences. For the laptop corpus, $1,007,209$ sentences are left after pre-processing. For the restaurants domain more reviews are available, we sampled $10,000,000$ sentences to have a sufficient amount of data for fully exploitet language model finetuning. In order to compensate for the smaller amount of finetuning data in the laptops domain, we finetune for more epochs, 30 epochs in the case of the laptops domain compared to 3 epochs for the restaurants domain, so that the BERT model trains on about 30 million sentences in both cases. This means that 1 sentence can be seen multiple times with a different language model masking. We also create a mixed corpus to jointly finetune both domains. Here, we sample 1 Mio. restaurant reviews and combine them with the laptop reviews. This results in about 2 Mio. reviews that are finetuned for 15 epochs. The exact statistics for the three finetuning corpora are shown in the top of tab:datasets. To be able to reproduce our finetuning corpora, we make the code that is used to generate them available online. ### Experiments ::: Hyperparameters
We use BERTBASE (uncased) as the base for all of our experiments, with the exception of XLNetBASE (cased), which is used as one of the baseline models. For the BERT language model finetuning we use 32 bit floating point computations using the Adam optimizer BIBREF18. The batchsize is set to 32 while the learning rate is set to $3\cdot 10^{-5}$. The maximum input sequence length is set to 256 tokens, which amounts to about 4 sentences per sequence on average. As shown in tab:datasets, we finetune the language models on each domain so that the model trains a total of about 30 Mio. sentences (7.5 Mio. sequences). For training the BERT and XLNet models on the down-stream task of ATSC we use mixed 16 bit and 32 bit floating point computations, the Adam optimizer, and a learning rate of $3\cdot 10^{-5}$ and a batchsize of 32. We train the model for a total of 7 epochs. The validation accuracy converges after about 3 epochs of training on all datasets, but training loss still improves after that. It is important to note that all our results reported are the average of 9 runs with different random initializations. This is needed to measure significance of improvements, as the standard deviation in accuray amounts to roughly $1\%$ for all experiments, see fig:acc-dep-lmiterations. ### Experiments ::: Compared Methods
We compare in-domain results to current state of the art methods, which we will now describe briefly. SDGCN-BERT BIBREF11 explicitly models sentiment dependencies for sentences with multiple aspects with a graph convolutional network. This method is current state-of-the-art on the SemEval 2014 laptops dataset. AEN-BERT BIBREF8 is an attentional encoder network. When used on top of BERT embeddings this method performs especially well on the laptops dataset. BERT-SPC BIBREF8 is BERT used in sentence-pair classification mode. This is exactly the same method as our BERT-base baseline and therefore, we can cross-check the authors results. BERT-PT BIBREF9 uses multi-task fine-tuning prior to downstream classification, where the BERT language model is finetuned jointly with a question answering task. It performs state-of-the-art on the restaurants dataset prior to this paper. To our knowledge, cross- and joint-domain training on the SemEval 2014 Task 4 datasets has not been analyzed so far. Thus, we compare our method to two very strong baselines: BERT and XLNet. BERT-base BIBREF7 is using the pre-trained BERTBASE embeddings directly on the down-stream task without any domain specific language model finetuning. XLNet-base BIBREF19 is a method also based on general language model pre-training similar to BERT. Instead of randomly masking tokens for pre-training like in BERT a more general permutation objective is used, where all possible variants of masking are fully exploitet. Our models are BERT models whose language model has been finetuned on different domain corpora. BERT-ADA Lapt is the BERT language model finetuned on the laptops domain corpus. BERT-ADA Rest is the BERT language model finetuned on the restaurant domain corpus. BERT-ADA Joint is the BERT language model finetuned on the corpus containing an equal amount of laptops and restaurants reviews. ### Experiments ::: Results Analysis
The results of our experiments are shown in fig:acc-dep-lmiterations and tab:results respectively. To answer RQ1, which is concerned with details on domain-specific language model finetuning, we can see in fig:acc-dep-lmiterations that first of all, language model finetuning has a substantial effect on ATSC end-task performance. Secondly, we see that in the laptops domain the performance starts to increase at about 10 Mio. finetuned sentences. This is an interesting insight as one would expect a relation closer to a logarithmic curve. One reason might be that it takes many steps to train knowledge into the BERT language model due to its vast amount of parameters. The model already converges at around 17 Mio. sentences. More finetuning does not improve performance significantly. In addition, we find that different runs have a high variance, the standard deviation amounts to about $1\%$ in accuracy, which justifies averaging over 9 runs to measure differences in model performance reliably. To answer RQ2, which is concerned with in-domain ATSC performance, we see in tab:results that for the in-domain training case, our models BERT-ADA Lapt and BERT-ADA Rest achieve performance close to state-of-the-art on the laptops dataset and new state-of-the-art on the restaurants dataset with accuracies of $79.19\%$ and $87.14\%$, respectively. On the restaurants dataset, this corresponds to an absolute improvement of $2.2\%$ compared to the previous state-of-the-art method BERT-PT. Language model finetuning produces a larger improvement on the restaurants dataset. We think that one reason for that might be that the restaurants domain is underrepresented in the pre-training corpora of BERTBASE. Generally, we find that language model finetuning helps even if the finetuning domain does not match the evaluation domain. We think the reason for this might be that the BERT-base model is pre-trained more on knowledge-based corpora like Wikipedia than on text containing opinions. Another finding is that BERT-ADA Joint performs better on the laptops dataset than BERT-ADA Rest, although the unique amount of laptop reviews are the same in laptops- and joint-corpora. We think that confusion can be created when mixing the domains, but this needs to be investigated further. We also find that the XLNet-base baseline performs generally stronger than BERT-base and even outperforms BERT-ADA Lapt with an accuracy of $79.89\%$ on the laptops dataset. To answer RQ3, which is concerned with domain adaptation, we can see in the grayed out cells in tab:results, which correspond to the cross-domain adaption case where the BERT language model is trained on the target domain, that domain adaptation works well with $2.2\%$ absolute accuracy improvement on the laptops test set and even $3.6\%$ accuracy improvement on the restaurants test set compared to BERT-base. In general, the ATSC task generalizes well cross-domain, with about 2-$3\%$ drop in accuracy compared to in-domain training. We think the reason for this might be that syntactical relationships between the aspect-target and the phrase expressing sentiment polarity as well as knowing the sentiment-polarity itself are sufficient to solve the ATSC task in many cases. For the joint-training case, we find that combining both training datasets improves performance on both test sets. This result is intuitive, as more training data leads to better performance if the domains do not confuse each other. Interesting for the joint-training case is that the BERT-ADA Joint model performs especially strong when measured by the Macro-F1 metric. A reason for this might be that the SemEval 2014 datasets are imbalanced due to dominance of positive label. It seems like through finetuning the language model on both domains the model learns to classify the neutral class much better, especially in the laptops domain. ### Conclusion
We performed experiments on the task of Aspect-Target Sentiment Classification by first finetuning a pre-trained BERT model on a domain specific corpus with subsequent training on the down-stream classification task. We analyzed the behavior of the number of domain-specific BERT language model finetuning steps in relation to the end-task performance. With the findings on how to best exploit BERT language model finetuning we were able to train high performing models, which one of even performs as new state-of-the-art on SemEval 2014 Task 4 restaurants dataset. We further evaluated our models cross-domain to explore the robustness of Aspect-Target Sentiment Classification. We found that in general, this task transfers well between the laptops and the restaurants domain. As a special case we ran a cross-domain adaptation experiments, where the BERT language model is specifically finetuned on the target domain. We achieve significant improvement over unadapted models, a cross-domain adapted model performs even better than a BERT-base model that is trained in-domain. Overall, our findings reveal promising directions for follow-up work. The XLNet-base model performs strongly on the ATSC task. Here, domain-specific finetuning could probably bring significant performance improvements. Another interesting direction for future work would be to investigate cross-domain behavior for an additional domain like hotels, which is more similar to the restaurants domain. Here, it could be interesting to find out if the shared nature of these domain would results in more confusion or if they would behave synergetically. Table 1: Top: Detailed statistics of the corpora for BERT language model finetuning. Bottom: Number of labels for each category of the SemEval 2014 Task 4 Subtask 2 laptop and restaurant datasets for AspectTarget Sentiment Classification. Figure 1: Accuracy of Aspect-Target Sentiment Classification as a function of the number of sentences the BERT language model has been finetuned on. Marked dots (•) connected through the line are the averages (µ) over 9 runs, a single run is marked as a cross (×). The standard deviation (σ) curves are also drawn (µ ± σ). The model is trained on the SemEval 2014 Task 4 laptops dataset. The language model is finetuned on our laptops domain corpus. Table 2: Summary of results for Aspect-Target Sentiment Classification for in-domain, cross-domain, and jointdomain training on SemEval 2014 Task 4 Subtask 2 datasets. The cells with gray background correspond to the cross-domain adaptation case, where the language model is finetuned on the target domain. As evaluation metrics accuracy (Acc) and Macro-F1 (MF1) are used.
|
$2.2\%$ absolute accuracy improvement on the laptops test set, $3.6\%$ accuracy improvement on the restaurants test set
|
What would have likely happened if the bank robbers' car tires had not melted?
A. The car would have wrecked regardless and the robbers would have been caught.
B. The police would have stopped them in a chase.
C. The robbers would have gotten away from the scene.
D. The robbers would have later returned to rob the bank again and get caught.
|
CALL HIM NEMESIS By DONALD E. WESTLAKE Criminals, beware; the Scorpion is on your trail! Hoodlums fear his fury—and, for that matter, so do the cops! [Transcriber's Note: This etext was produced from Worlds of If Science Fiction, September 1961. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The man with the handkerchief mask said, "All right, everybody, keep tight. This is a holdup." There were twelve people in the bank. There was Mr. Featherhall at his desk, refusing to okay a personal check from a perfect stranger. There was the perfect stranger, an itinerant garage mechanic named Rodney (Rod) Strom, like the check said. There were Miss English and Miss Philicoff, the girls in the gilded teller cages. There was Mister Anderson, the guard, dozing by the door in his brown uniform. There was Mrs. Elizabeth Clayhorn, depositing her husband's pay check in their joint checking account, and with her was her ten-year-old son Edward (Eddie) Clayhorn, Junior. There was Charlie Casale, getting ten dollars dimes, six dollars nickels and four dollars pennies for his father in the grocery store down the street. There was Mrs. Dolly Daniels, withdrawing money from her savings account again. And there were three bank robbers. The three bank robbers looked like triplets. From the ground up, they all wore scuffy black shoes, baggy-kneed and unpressed khaki trousers, brown cracked-leather jackets over flannel shirts, white handkerchiefs over the lower half of their faces and gray-and-white check caps pulled low over their eyes. The eyes themselves looked dangerous. The man who had spoken withdrew a small but mean-looking thirty-two calibre pistol from his jacket pocket. He waved it menacingly. One of the others took the pistol away from Mister Anderson, the guard, and said to him in a low voice, "Think about retirement, my friend." The third one, who carried a black satchel like a doctor's bag, walked quickly around behind the teller's counter and started filling it with money. It was just like the movies. The man who had first spoken herded the tellers, Mr. Featherhall and the customers all over against the back wall, while the second man stayed next to Mr. Anderson and the door. The third man stuffed money into the black satchel. The man by the door said, "Hurry up." The man with the satchel said, "One more drawer." The man with the gun turned to say to the man at the door, "Keep your shirt on." That was all Miss English needed. She kicked off her shoes and ran pelting in her stocking feet for the door. The man by the door spread his arms out and shouted, "Hey!" The man with the gun swung violently back, cursing, and fired the gun. But he'd been moving too fast, and so had Miss English, and all he hit was the brass plate on Mr. Featherhall's desk. The man by the door caught Miss English in a bear hug. She promptly did her best to scratch his eyes out. Meanwhile, Mr. Anderson went scooting out the front door and running down the street toward the police station in the next block, shouting, "Help! Help! Robbery!" The man with the gun cursed some more. The man with the satchel came running around from behind the counter, and the man by the door tried to keep Miss English from scratching his eyes out. Then the man with the gun hit Miss English on the head. She fell unconscious to the floor, and all three of them ran out of the bank to the car out front, in which sat a very nervous-looking fourth man, gunning the engine. Everyone except Miss English ran out after the bandits, to watch. Things got very fast and very confused then. Two police cars came driving down the block and a half from the precinct house to the bank, and the car with the four robbers in it lurched away from the curb and drove straight down the street toward the police station. The police cars and the getaway car passed one another, with everybody shooting like the ships in pirate movies. There was so much confusion that it looked as though the bank robbers were going to get away after all. The police cars were aiming the wrong way and, as they'd come down with sirens wailing, there was a clear path behind them. Then, after the getaway car had gone more than two blocks, it suddenly started jouncing around. It smacked into a parked car and stopped. And all the police went running down there to clap handcuffs on the robbers when they crawled dazedly out of their car. "Hey," said Eddie Clayhorn, ten years old. "Hey, that was something, huh, Mom?" "Come along home," said his mother, grabbing his hand. "We don't want to be involved." "It was the nuttiest thing," said Detective-Sergeant Stevenson. "An operation planned that well, you'd think they'd pay attention to their getaway car, you know what I mean?" Detective-Sergeant Pauling shrugged. "They always slip up," he said. "Sooner or later, on some minor detail, they always slip up." "Yes, but their tires ." "Well," said Pauling, "it was a stolen car. I suppose they just grabbed whatever was handiest." "What I can't figure out," said Stevenson, "is exactly what made those tires do that. I mean, it was a hot day and all, but it wasn't that hot. And they weren't going that fast. I don't think you could go fast enough to melt your tires down." Pauling shrugged again. "We got them. That's the important thing." "Still and all, it's nutty. They're free and clear, barrelling out Rockaway toward the Belt, and all at once their tires melt, the tubes blow out and there they are." Stevenson shook his head. "I can't figure it." "Don't look a gift horse in the mouth," suggested Pauling. "They picked the wrong car to steal." "And that doesn't make sense, either," said Stevenson. "Why steal a car that could be identified as easily as that one?" "Why? What was it, a foreign make?" "No, it was a Chevvy, two-tone, three years old, looked just like half the cars on the streets. Except that in the trunk lid the owner had burned in 'The Scorpion' in big black letters you could see half a block away." "Maybe they didn't notice it when they stole the car," said Pauling. "For a well-planned operation like this one," said Stevenson, "they made a couple of really idiotic boners. It doesn't make any sense." "What do they have to say about it?" Pauling demanded. "Nothing, what do you expect? They'll make no statement at all." The squad-room door opened, and a uniformed patrolman stuck his head in. "The owner of that Chevvy's here," he said. "Right," said Stevenson. He followed the patrolman down the hall to the front desk. The owner of the Chevvy was an angry-looking man of middle age, tall and paunchy. "John Hastings," he said. "They say you have my car here." "I believe so, yes," said Stevenson. "I'm afraid it's in pretty bad shape." "So I was told over the phone," said Hastings grimly. "I've contacted my insurance company." "Good. The car's in the police garage, around the corner. If you'd come with me?" On the way around, Stevenson said, "I believe you reported the car stolen almost immediately after it happened." "That's right," said Hastings. "I stepped into a bar on my route. I'm a wine and liquor salesman. When I came out five minutes later, my car was gone." "You left the keys in it?" "Well, why not?" demanded Hastings belligerently. "If I'm making just a quick stop—I never spend more than five minutes with any one customer—I always leave the keys in the car. Why not?" "The car was stolen," Stevenson reminded him. Hastings grumbled and glared. "It's always been perfectly safe up till now." "Yes, sir. In here." Hastings took one look at his car and hit the ceiling. "It's ruined!" he cried. "What did you do to the tires?" "Not a thing, sir. That happened to them in the holdup." Hastings leaned down over one of the front tires. "Look at that! There's melted rubber all over the rims. Those rims are ruined! What did you use, incendiary bullets?" Stevenson shook his head. "No, sir. When that happened they were two blocks away from the nearest policeman." "Hmph." Hastings moved on around the car, stopping short to exclaim, "What in the name of God is that? You didn't tell me a bunch of kids had stolen the car." "It wasn't a bunch of kids," Stevenson told him. "It was four professional criminals, I thought you knew that. They were using it in a bank holdup." "Then why did they do that ?" Stevenson followed Hastings' pointing finger, and saw again the crudely-lettered words, "The Scorpion" burned black into the paint of the trunk lid. "I really don't know," he said. "It wasn't there before the car was stolen?" "Of course not!" Stevenson frowned. "Now, why in the world did they do that?" "I suggest," said Hastings with heavy sarcasm, "you ask them that." Stevenson shook his head. "It wouldn't do any good. They aren't talking about anything. I don't suppose they'll ever tell us." He looked at the trunk lid again. "It's the nuttiest thing," he said thoughtfully.... That was on Wednesday. The Friday afternoon mail delivery to the Daily News brought a crank letter. It was in the crank letter's most obvious form; that is, the address had been clipped, a letter or a word at a time, from a newspaper and glued to the envelope. There was no return address. The letter itself was in the same format. It was brief and to the point: Dear Mr. Editor: The Scorpion has struck. The bank robbers were captured. The Scorpion fights crime. Crooks and robbers are not safe from the avenging Scorpion. WARN YOUR READERS! Sincerely yours, THE SCORPION The warning was duly noted, and the letter filed in the wastebasket. It didn't rate a line in the paper. II The bank robbery occurred in late June. Early in August, a Brooklyn man went berserk. It happened in Canarsie, a section in southeast Brooklyn near Jamaica Bay. This particular area of Canarsie was a residential neighborhood, composed of one and two family houses. The man who went berserk was a Motor Vehicle Bureau clerk named Jerome Higgins. Two days before, he had flunked a Civil Service examination for the third time. He reported himself sick and spent the two days at home, brooding, a bottle of blended whiskey at all times in his hand. As the police reconstructed it later, Mrs. Higgins had attempted to awaken him on the third morning at seven-thirty, suggesting that he really ought to stop being so foolish, and go back to work. He then allegedly poked her in the eye, and locked her out of the bedroom. Mrs. Higgins then apparently called her sister-in-law, a Mrs. Thelma Stodbetter, who was Mr. Higgins' sister. Mrs. Stodbetter arrived at the house at nine o'clock, and spent some time tapping at the still-locked bedroom door, apparently requesting Mr. Higgins to unlock the door and "stop acting like a child." Neighbors reported to the police that they heard Mr. Higgins shout a number of times, "Go away! Can't you let a man sleep?" At about ten-fifteen, neighbors heard shots from the Higgins residence, a two-story one-family pink stucco affair in the middle of a block of similar homes. Mr. Higgins, it was learned later, had suddenly erupted from his bedroom, brandishing a .30-.30 hunting rifle and, being annoyed at the shrieks of his wife and sister, had fired seven shells at them, killing his wife on the spot and wounding his sister in the hand and shoulder. Mrs. Stodbetter, wounded and scared out of her wits, raced screaming out the front door of the house, crying for the police and shouting, "Murder! Murder!" At this point, neighbors called the police. One neighbor additionally phoned three newspapers and two television stations, thereby earning forty dollars in "news-tips" rewards. By chance, a mobile television unit was at that moment on the Belt Parkway, returning from having seen off a prime minister at Idlewild Airport. This unit was at once diverted to Canarsie, where it took up a position across the street from the scene of carnage and went to work with a Zoomar lens. In the meantime, Mister Higgins had barricaded himself in his house, firing at anything that moved. The two cameramen in the mobile unit worked their hearts out. One concentrated on the movements of the police and firemen and neighbors and ambulance attendants, while the other used the Zoomar lens to search for Mr. Higgins. He found him occasionally, offering the at-home audience brief glimpses of a stocky balding man in brown trousers and undershirt, stalking from window to window on the second floor of the house. The show lasted for nearly an hour. There were policemen everywhere, and firemen everywhere, and neighbors milling around down at the corner, where the police had roped the block off, and occasionally Mr. Higgins would stick his rifle out a window and shoot at somebody. The police used loudspeakers to tell Higgins he might as well give up, they had the place surrounded and could eventually starve him out anyway. Higgins used his own good lungs to shout obscenities back and challenge anyone present to hand-to-hand combat. The police fired tear gas shells at the house, but it was a windy day and all the windows in the Higgins house were either open or broken. Higgins was able to throw all the shells back out of the house again. The show lasted for nearly an hour. Then it ended, suddenly and dramatically. Higgins had showed himself to the Zoomar lens again, for the purpose of shooting either the camera or its operator. All at once he yelped and threw the rifle away. The rifle bounced onto the porch roof, slithered down to the edge, hung for a second against the drain, and finally fell barrel first onto the lawn. Meanwhile, Higgins was running through the house, shouting like a wounded bull. He thundered down the stairs and out, hollering, to fall into the arms of the waiting police. They had trouble holding him. At first they thought he was actually trying to get away, but then one of them heard what it was he was shouting: "My hands! My hands!" They looked at his hands. The palms and the palm-side of the fingers were red and blistering, from what looked like severe burns. There was another burn on his right cheek and another one on his right shoulder. Higgins, thoroughly chastened and bewildered, was led away for burn ointment and jail. The television crew went on back to Manhattan. The neighbors went home and telephoned their friends. On-duty policemen had been called in from practically all of the precincts in Brooklyn. Among them was Detective-Sergeant William Stevenson. Stevenson frowned thoughtfully at Higgins as that unhappy individual was led away, and then strolled over to look at the rifle. He touched the stock, and it was somewhat warm but that was all. He picked it up and turned it around. There, on the other side of the stock, burned into the wood, were the crudely-shaped letters, "The Scorpion." You don't get to be Precinct Captain on nothing but political connections. Those help, of course, but you need more than that. As Captain Hanks was fond of pointing out, you needed as well to be both more imaginative than most—"You gotta be able to second-guess the smart boys"—and to be a complete realist—"You gotta have both feet on the ground." If these were somewhat contradictory qualities, it was best not to mention the fact to Captain Hanks. The realist side of the captain's nature was currently at the fore. "Just what are you trying to say, Stevenson?" he demanded. "I'm not sure," admitted Stevenson. "But we've got these two things. First, there's the getaway car from that bank job. The wheels melt for no reason at all, and somebody burns 'The Scorpion' onto the trunk. Then, yesterday, this guy Higgins out in Canarsie. He says the rifle all of a sudden got too hot to hold, and he's got the burn marks to prove it. And there on the rifle stock it is again. 'The Scorpion'." "He says he put that on there himself," said the captain. Stevenson shook his head. "His lawyer says he put it on there. Higgins says he doesn't remember doing it. That's half the lawyer's case. He's trying to build up an insanity defense." "He put it on there himself, Stevenson," said the captain with weary patience. "What are you trying to prove?" "I don't know. All I know is it's the nuttiest thing I ever saw. And what about the getaway car? What about those tires melting?" "They were defective," said Hanks promptly. "All four of them at once? And what about the thing written on the trunk?" "How do I know?" demanded the captain. "Kids put it on before the car was stolen, maybe. Or maybe the hoods did it themselves, who knows? What do they say?" "They say they didn't do it," said Stevenson. "And they say they never saw it before the robbery and they would have noticed it if it'd been there." The captain shook his head. "I don't get it," he admitted. "What are you trying to prove?" "I guess," said Stevenson slowly, thinking it out as he went along, "I guess I'm trying to prove that somebody melted those tires, and made that rifle too hot, and left his signature behind." "What? You mean like in the comic books? Come on, Stevenson! What are you trying to hand me?" "All I know," insisted Stevenson, "is what I see." "And all I know," the captain told him, "is Higgins put that name on his rifle himself. He says so." "And what made it so hot?" "Hell, man, he'd been firing that thing at people for an hour! What do you think made it hot?" "All of a sudden?" "He noticed it all of a sudden, when it started to burn him." "How come the same name showed up each time, then?" Stevenson asked desperately. "How should I know? And why not, anyway? You know as well as I do these things happen. A bunch of teen-agers burgle a liquor store and they write 'The Golden Avengers' on the plate glass in lipstick. It happens all the time. Why not 'The Scorpion'? It couldn't occur to two people?" "But there's no explanation—" started Stevenson. "What do you mean, there's no explanation? I just gave you the explanation. Look, Stevenson, I'm a busy man. You got a nutty idea—like Wilcox a few years ago, remember him? Got the idea there was a fiend around loose, stuffing all those kids into abandoned refrigerators to starve. He went around trying to prove it, and getting all upset, and pretty soon they had to put him away in the nut hatch. Remember?" "I remember," said Stevenson. "Forget this silly stuff, Stevenson," the captain advised him. "Yes, sir," said Stevenson.... The day after Jerome Higgins went berserk, the afternoon mail brought a crank letter to the Daily News : Dear Mr. Editor, You did not warn your readers. The man who shot all those people could not escape the Scorpion. The Scorpion fights crime. No criminal is safe from the Scorpion. WARN YOUR READERS. Sincerely yours, THE SCORPION Unfortunately, this letter was not read by the same individual who had seen the first one, two months before. At any rate, it was filed in the same place, and forgotten. III Hallowe'en is a good time for a rumble. There's too many kids around for the cops to keep track of all of them, and if you're picked up carrying a knife or a length of tire chain or something, why, you're on your way to a Hallowe'en party and you're in costume. You're going as a JD. The problem was this schoolyard. It was a block wide, with entrances on two streets. The street on the north was Challenger territory, and the street on the south was Scarlet Raider territory, and both sides claimed the schoolyard. There had been a few skirmishes, a few guys from both gangs had been jumped and knocked around a little, but that had been all. Finally, the War Lords from the two gangs had met, and determined that the matter could only be settled in a war. The time was chosen: Hallowe'en. The place was chosen: the schoolyard. The weapons were chosen: pocket knives and tire chains okay, but no pistols or zip-guns. The time was fixed: eleven P.M. And the winner would have undisputed territorial rights to the schoolyard, both entrances. The night of the rumble, the gangs assembled in their separate clubrooms for last-minute instructions. Debs were sent out to play chicken at the intersections nearest the schoolyard, both to warn of the approach of cops and to keep out any non-combatant kids who might come wandering through. Judy Canzanetti was a Deb with the Scarlet Raiders. She was fifteen years old, short and black-haired and pretty in a movie-magazine, gum-chewing sort of way. She was proud of being in the Auxiliary of the Scarlet Raiders, and proud also of the job that had been assigned to her. She was to stand chicken on the southwest corner of the street. Judy took up her position at five minutes to eleven. The streets were dark and quiet. Few people cared to walk this neighborhood after dark, particularly on Hallowe'en. Judy leaned her back against the telephone pole on the corner, stuck her hands in the pockets of her Scarlet Raider jacket and waited. At eleven o'clock, she heard indistinct noises begin behind her. The rumble had started. At five after eleven, a bunch of little kids came wandering down the street. They were all about ten or eleven years old, and most of them carried trick-or-treat shopping bags. Some of them had Hallowe'en masks on. They started to make the turn toward the schoolyard. Judy said, "Hey, you kids. Take off." One of them, wearing a red mask, turned to look at her. "Who, us?" "Yes, you! Stay out of that street. Go on down that way." "The subway's this way," objected the kid in the red mask. "Who cares? You go around the other way." "Listen, lady," said the kid in the red mask, aggrieved, "we got a long way to go to get home." "Yeah," said another kid, in a black mask, "and we're late as it is." "I couldn't care less," Judy told them callously. "You can't go down that street." "Why not?" demanded yet another kid. This one was in the most complete and elaborate costume of them all, black leotards and a yellow shirt and a flowing: black cape. He wore a black and gold mask and had a black knit cap jammed down tight onto his head. "Why can't we go down there?" this apparition demanded. "Because I said so," Judy told him. "Now, you kids get away from here. Take off." "Hey!" cried the kid in the black-and-yellow costume. "Hey, they're fighting down there!" "It's a rumble," said Judy proudly. "You twerps don't want to be involved." "Hey!" cried the kid in the black-and-yellow costume again. And he went running around Judy and dashing off down the street. "Hey, Eddie!" shouted one of the other kids. "Eddie, come back!" Judy wasn't sure what to do next. If she abandoned her post to chase the one kid who'd gotten through, then maybe all the rest of them would come running along after her. She didn't know what to do. A sudden siren and a distant flashing red light solved her problems. "Cheez," said one of the kids. "The cops!" "Fuzz!" screamed Judy. She turned and raced down the block toward the schoolyard, shouting, "Fuzz! Fuzz! Clear out, it's the fuzz!" But then she stopped, wide-eyed, when she saw what was going on in the schoolyard. The guys from both gangs were dancing. They were jumping around, waving their arms, throwing their weapons away. Then they all started pulling off their gang jackets and throwing them away, whooping and hollering. They were making such a racket themselves that they never heard Judy's warning. They didn't even hear the police sirens. And all at once both schoolyard entrances were full of cops, a cop had tight hold of Judy and the rumble was over. Judy was so baffled and terrified that everything was just one great big blur. But in the middle of it all, she did see the little kid in the yellow-and-black costume go scooting away down the street. And she had the craziest idea that it was all his fault. Captain Hanks was still in his realistic cycle this morning, and he was impatient as well. "All right, Stevenson," he said. "Make it fast, I've got a lot to do this morning. And I hope it isn't this comic-book thing of yours again." "I'm afraid it is, Captain," said Stevenson. "Did you see the morning paper?" "So what?" "Did you see that thing about the gang fight up in Manhattan?" Captain Hanks sighed. "Stevenson," he said wearily, "are you going to try to connect every single time the word 'scorpion' comes up? What's the problem with this one? These kid gangs have names, so what?" "Neither one of them was called 'The Scorpions,'" Stevenson told him. "One of them was the Scarlet Raiders and the other gang was the Challengers." "So they changed their name," said Hanks. "Both gangs? Simultaneously? To the same name?" "Why not? Maybe that's what they were fighting over." "It was a territorial war," Stevenson reminded him. "They've admitted that much. It says so in the paper. And it also says they all deny ever seeing that word on their jackets until after the fight." "A bunch of juvenile delinquents," said Hanks in disgust. "You take their word?" "Captain, did you read the article in the paper?" "I glanced through it." "All right. Here's what they say happened: They say they started fighting at eleven o'clock. And they just got going when all at once all the metal they were carrying—knives and tire chains and coins and belt buckles and everything else—got freezing cold, too cold to touch. And then their leather jackets got freezing cold, so cold they had to pull them off and throw them away. And when the jackets were later collected, across the name of the gang on the back of each one had been branded 'The Scorpion.'" "Now, let me tell you something," said Hanks severely. "They heard the police sirens, and they threw all their weapons away. Then they threw their jackets away, to try to make believe they hadn't been part of the gang that had been fighting. But they were caught before they could get out of the schoolyard. If the squad cars had showed up a minute later, the schoolyard wouldn't have had anything in it but weapons and jackets, and the kids would have been all over the neighborhood, nice as you please, minding their own business and not bothering anybody. That's what happened. And all this talk about freezing cold and branding names into jackets is just some smart-alec punk's idea of a way to razz the police. Now, you just go back to worrying about what's happening in this precinct and forget about kid gangs up in Manhattan and comic book things like the Scorpion, or you're going to wind up like Wilcox, with that refrigerator business. Now, I don't want to hear any more about this nonsense, Stevenson." "Yes, sir," said Stevenson.
|
C. The robbers would have gotten away from the scene.
|
What classification task was used to evaluate the cross-lingual adaptation method described in this work?
|
### Introduction
Universal Dependencies (UD) BIBREF0, BIBREF1, BIBREF2 is an ongoing project aiming to develop cross-lingually consistent treebanks for different languages. UD provided a framework for consistent annotation of grammar (parts of speech, morphological features, and syntactic dependencies) across different human languages. The annotation schema relies on Universal Stanford Dependencies BIBREF3 and Google Universal POS tags BIBREF4. The general principle is to provide universal annotation ; meanwhile, each language can add language-specific relations to the universal pool when necessary. The main goal of UD project is to facilitate multi-lingual parser production and cross-lingual learningFOOTREF1. Cross-lingual learning is the task of gaining advantages from high-resource languages in terms of annotated data to build a model for low-resource languages. This paradigm of learning is now an invaluable tool for improving the performance of natural language processing in low-resource languages. Based on the universal annotations of the UD project, there are several works on cross-lingual tasks. Most of them focus on grammar-related tasks such as POS tagging BIBREF5 and dependency parsing BIBREF6, BIBREF7, BIBREF8. In this paper, we are going to study the effectiveness of UD in making cross-lingual models for more complex tasks such as semantic relation extraction and paraphrase identification. To the best of our knowledge, no work was done on the application of UD annotations in the mentioned tasks. Universal dependencies approach for cross-lingual learning is based on the fact that UD captures similarities as well as idiosyncrasies among typologically different languages. The important characteristic of UD annotations is that although the UD parse trees of parallel sentences in different languages may not be completely equivalent, they have many similar sub-trees, in the sense that at least core parts of trees are equal BIBREF9. In this paper, we study two cross-lingual tasks : semantic relation extraction and paraphrase identification. The former is the task of identifying semantic connections between entities in a sentence ; while the training and test data are in different languages. The latter is defined to determine whether two sentences are paraphrase or not ; while the training' pairs of sentences are in a different language from the test data. To employ similarities of UD trees of different languages to train cross-lingual models, we propose to use syntactic based methods which ideally can deal with parsing information of data. We found that tree kernels allow to estimate the similarities among texts directly from their parse trees. They are known to operate on dependency parse trees and automatically generate robust prediction models based on the similarities of them. We have made parallel dataset for each task and presented the cross-lingual variant of kernel functions for them. Evaluation by the parallel test data reveals that the accuracy of models trained by a language and tested on the other languages get close to mono-lingual when the syntactic parsers are trained with UD corpora. This suggests that syntactic patterns trained on the UD trees can be invariant with respect to very different languages. To compare the proposed approach with the cross-lingual variant of neural models, we employed several state-of-the-art deep networks and equipped them with pre-trained bi-lingual word embeddings. English training data are fed into the networks, which create a mapping between the input and output values. Then test set is given to the trained network. Results show that the tree-based models outperform end-to-end neural models in cross-lingual experiments. Moreover, we employed Tree-LSTM network BIBREF10 with UD parse trees, which is capable to produce semantic representation from tree-ordered input data. Tree-LSTM doesn't directly deal with syntactic features of the input sentence, rather it processes the input tokens in order of placing in a tree, e.g. from bottom to up or vice versa. Experiments show superiority of Tree-LSTM trained by UD trees over sequential models like LSTM in cross-lingual evaluations. This paper is organized as follows : Section SECREF2 describes how UD approach allows to capture similarities and differences across diverse languages. Section SECREF3 presents tree-based models for cross-lingual learning of PI and RE tasks. Section SECREF4 presents an empirical study on cross-lingual learning using UD. Finally Section SECREF5 gives the analysis and conclusion remarks. ### Transfer Learning via Universal Dependencies
The Universal Dependencies project aims to produce consistent dependency treebanks and parsers for many languages BIBREF0, BIBREF1, BIBREF2. The most important achievements of the project are the cross-lingual annotation guidelines and sets of universal POS and the grammatical relation tags. Consequentially many treebanks have been developed for different languages. The general rule of UD project is to provide a universal tag set ; however each language can add language-specific relations to the universal pool or omit some tags. To capture similarities and differences across languages, UD uses a representation consisting of three components : (i) dependency relations between lexical words ; (ii) function words modifying lexical words ; and (iii) morphological features associated with words BIBREF9. The underlying principle of the syntactic annotation schema of the UD project is that dependencies hold between content words, while function words attach to the content word that they further specify BIBREF3. There is an important difference between UD schema and Stanford Typed Dependencies (STD) BIBREF11 as the STD schema chooses function words as heads : prepositions in prepositional phrases, and copula verbs that have a prepositional phrase as their complement. Although the UD parse graphs of a sentence in different languages may not be completely equal, they have similar core parts. Figure FIGREF5 shows the UD graph of English sentence “The memo presents details about the lineup management" and its translation into French and Farsi. Both the similarities and differences of UD graphs are demonstrated in that figure. Most of the nodes and edges are similar. Farsi has the language-specific relation “compound :lvc", which relates the noun part of the compound verb to the verbal part as depicted in Figure FIGREF5. So far, UD treebanks have been developed for over 70 languages and all of them are freely available for download. UD project released a pipeline, called UDPipe, which is used to train models for UD parsing using the UD treebanks BIBREF12. UD parsing and similarity of UD structures in different languages provide facilities to train multi-lingual models. In what follows, we focus on two tasks, paraphrase identification and semantic relation extraction, and present cross-learning models for them. ### Cross-Lingual Tree-based Models
To employ UD parsing in cross-lingual learning, there should be a training algorithm that is capable of utilizing similarities of UD parse trees in different languages. Kernel methods such as SVM use a similarity function, which is called kernel function, to assign a similarity score to pairs of data samples. A kernel function $K$ over an object space $X$ is symmetric, positive semi-definite function $K: X \times X \rightarrow [0,\infty )$ that assigns a similarity score to two instances of $X$, where $K(x,y)=\phi (x)\cdot \phi (y)=\sum {\phi _{i}(x)\phi _{i}(y)}$. Here, $\phi (x)$ is a mapping function from the data object in $X$ to the high-dimensional feature space. Using the kernel function, it is not necessary to extract all features one by one and then multiply the feature vectors. Instead, kernel functions compute the final value directly based on the similarity of data examples. Tree kernels are the most popular kernels for many natural language processing tasks BIBREF13, BIBREF14. Tree Kernels compute the number of common substructures between two trees $T_1$ and $T_2$ without explicitly considering the whole fragment space BIBREF15. Suppose the set $\mathcal {F}=\lbrace f_1,f_2, \dots , f_{|\mathcal {F}|} \rbrace $ be the tree fragment space and $\mathcal {X}_i(n)$ be an indicator function that is 1 if the $f_i$ rooted at node $n$ and equals to 0, otherwise. Now, tree kernel over $T_1$ and $T_2$ is defined as below BIBREF15 : where $N_{T_1}$ and $N_{T_2}$ are the set of nodes of $T_1$ and $T_2$, respectively and which shows the number of common fragments rooted in $n_1$ and $n_2$ nodes. Different tree kernels vary in their definition of $\Delta $ function and fragment type. There are three important characterizations of fragment type BIBREF16 : SubTree, SubSet Tree and Partial Tree. A SubTree is defined by taking a node of a tree along with all its descendants. SubSet Tree is more general and does not necessarily contain all of the descendants. Instead, it must be generated by utilizing the same grammatical rule set of the original trees. A Partial Tree is more general and relaxes SubSet Tree's constraints. Some popular tree kernels are SubSet Tree Kernel (SST), Partial Tree Kernel (PTK) BIBREF17 and Smoothing Partial Tree Kernel (SPTK) BIBREF15. In the next section, we employ the tree kernels along with UD parse trees for solving cross-lingual tasks. ### Cross-Lingual Tree-based Models ::: Cross-Lingual Paraphrase Identification
Paraphrase Identification (PI) is the task of determining whether two sentences are paraphrase or not. It is considered a binary classification task. The best mono-lingual methods often achieve about 85% accuracy over this corpus BIBREF14, BIBREF18. Filice et al. BIBREF14 extended the tree kernels described in the previous section to operate on text pairs. The underlying idea is that this task is characterized by several syntactic/semantic patterns that a kernel machine can automatically capture from the training material. We can assess a text pair as a paraphrase if it shows a valid transformation rule that we observed in the training data. The following example can clarify this concept. A simple paraphrase rewriting rule is the active-passive transformation, such as in “Federer beat Nadal” and “Nadal was defeated by Federer”. The same transformation can be observed in other paraphrases, such as in “Mark studied biology” and “Biology was learned by Mark”. Although these two pairs of paraphrases have completely different topics, they have a very similar syntactic structure. Tree kernel combinations can capture this inter-pair similarity and allow a learning algorithm such as SVM to learn the syntactic-semantic patterns characterizing valid paraphrases. Given a tree kernel $TK$ and text pairs $p_i = (i_1, i_2)$, the best tree kernel combination for the paraphrase identification task described in BIBREF14 is the following : SMTK ( pa, pb ) = softmax ( TK(a1,b1)TK(a2, b2), TK(a1,b2)TK(a2,b1) ) where softmax$(x_1,x_2)= \frac{1}{m} \log \left(e^{m x_1} + e^{m x_2}\right)$ is a simple function approximating the max operator, which cannot be directly used in kernel formulations, as it can create non valid kernel functions. In this kernel combination the two different alignments between the trees of the two pairs are tried and the best alignment is chosen. This allows to exploit the inherent symmetry of the Paraphrase Identification task (i.e., if $a$ is a paraphrase of $b$, it also implies that $b$ is a paraphrase of $a$). When we adopt the universal dependencies, different languages have a common formalism to represent text syntax, and tree kernels, that mostly operate at a syntactical level, can still provide reliable similarity estimations, i.e., $SM_{TK}(p_a, p_b)$ can work even if $p_a$ and $p_b$ have different languages. This allows operating in a cross-lingual setting. For instance, we can use a model trained on a high-resource language for classifying textual data of a poor-resource language. In addition to the syntactic similarity evaluation, the PTK and SPTK which are used in the $SM_{TK}$ formulation also perform a lexical matching among the words of the trees to be compared. ### Cross-Lingual Tree-based Models ::: Cross-Lingual Semantic Relation Extraction
Relation Extraction (RE) is defined as the task of identifying semantic relations between entities in a text. The goal is to determine whether there is a semantic relation between two given entities in a text, and also to specify the type of relationship if present. RE is an important part of Information Extraction BIBREF19. Relation extraction methods often focus on the Shortest Dependency Path (SDP) between entities BIBREF20. However, there are some crucial differences between UD annotation principles and others parse formalisms that causes us to reconsider SDP of UD trees. Considering the sentence : “The most common $[$audits$]_{e1}$ were about $[$waste$]_{e2}$ and recycling", there is a Message-Topic relation between $e1$ and $e2$. The most informative words of the sentence for the relation are “were" and “about" ; while the other words of the sentence can be ignored and the same relation is still realized. It is a crucial challenge of relation extraction methods that important information may appear at any part of the sentence. Most previous works assume that the words lying in the window surrounding entities are enough to extract the relation governing entities BIBREF21, BIBREF22. However, words of a sentence are often reordered when the sentence is translated into other languages. Therefore, using words in the window surrounding entities may result in an accurate model for mono-lingual experiments, but not necessarily for cross-lingual ones. Regarding UD parsing, there are several significant differences between universal annotation schema and other schemas for dependency parsing. Two main differences are related to prepositions and copula verbs. According to the UD annotation guidelines, prepositions are attached to the head of a nominal, and copula verbs are attached to the head of a clause. However in other schemas, prepositions are often the root of the nominal, and the clause is attached to the copula. Figure FIGREF12 shows the parse tree of the example : “The most common $[$audits$]_{e1}$ were about $[$waste$]_{e2}$ and recycling". The tree is produced by the ARK parser, which does not follow universal schema. As mentioned before, “were" and “about" lie on the SDP between $e1$ and $e2$. However, considering the UD parse tree depicted in Figure FIGREF12, there is no word in the SDP ; while both “were" and “about" are attached to $e2$. As a result, we propose that the words which are dependent on the entities be considered to be the informative words in addition to the SDP's words. We use these words for making a cross-lingual model. Kernel functions have several interesting characteristics. The combination of kernel functions in a linear or polynomial way results in a valid kernel function BIBREF23. Composite kernel functions are built on individual kernels ; each of them captures part of the features of a data object. Tree kernels capture the data's syntactic structure, while a word sequence kernel considers the words of a sequence in a particular order. To define a cross-lingual kernel, we have adopted the composite kernel used by the Nguyen et al. BIBREF16 : where $K_{P-e}$ is a polynomial kernel. Its base kernel is an entity kernel ($K_E$), which is applied to an entity-related feature vector consisting of (named) entity type, mention type, headword, and POS tag. $K_{SST}$ is the Sub-Set Tree (SST) kernel, which is applied to the Path-Enclosed Tree (PET) of the constituency tree structure. PET is the smallest common subtree including the two entities BIBREF24, BIBREF25. $K_{PT}$ is the Partial Tree kernel BIBREF17, which is applied to the dependency-based tree structures. Parameter $\alpha $ weighs the kernels. To incorporate the most informative words of the sentence into the model, the feature vector $V_o$ is defined similarly to the work of Hashimoto et al. BIBREF21. They proposed concatenating these vectors to make the $V_o$ : the vector representing $e1$, the vector representing $e2$, the average of vectors representing words between two entities, the average of vectors representing words in a window before $e1$, and the average of vectors representing words in a window after $e2$. Since $V_o$ is defined based on the position of words in the sentence and thus is not necessary a cross-lingual consistent feature vector, we propose to define feature vector $V_{ud}$ by concatenating these vectors : the vector representing $e1$, the vector representing $e2$, the average of vectors representing words in the shortest path between two entities (instead of words between $e1$ and $e2$), the average of vectors representing words dependent to $e1$ (instead of words before $e1$), and the average of vectors representing words dependent to $e2$ (instead of words after $e2$). $V_{ud}$ is cross-lingually consistent provided that the words are picked up from UD parse trees and represented by multi-lingual embeddings. Based on the $CK$ defined in formula DISPLAY_FORM13 and the feature vectors $V_o$ and $V_{ud}$, the following composite kernels are proposed : where $K_{P-o}$ is polynomial kernel applied on a feature vector $V_o$. where $K_{P-ud}$ is polynomial kernel applied on a feature vector $V_{ud}$. Constituency parsing of a sentence in a language depends on the syntactic rules governing the position of words. In general, constituency parse trees of a sentence in different languages are different. So, the constituency tree should not be involved in the cross-lingual model. Here, $CK_2$ is our proposed kernel, which is used for CL-RE. However, $CK_1$ and $CK_3$ can also be used for cross-lingual experiments subject to the similarity of syntactic parsing of the source and target languages. SST kernel works only on the constituency trees and not on the dependency trees BIBREF17. Therefore, for evaluating the similarity of dependency trees, PT kernel is used. The PT kernel cannot process labels on the edges ; so dependency trees are converted to the Lexical Centered Tree (LCT) format BIBREF15 and then PT kernel is applied on the transformed trees. In LCT format, the lexical is kept at the center and the other information related to that lexical, such as POS tag and grammatical relation, is then added as its children. MultiWord Expression (MWE) is a lexeme made up a sequence of two or more lexemes as each lexeme has its own meaning, but the meaning of the whole expression cannot (or at least can only partially) be computed from the meaning of its parts. MWE displays lexical, syntactic, semantic, pragmatic and/or statistical idiosyncrasies BIBREF26. The nature of MWE leads us to deal with the whole lexemes as a word. Fortunately, MWE can be identified from the parse tree. There are three types of dependency relations for MWE in UD parsing : flat, fixed, and compound. According to UD guidelines, the flat relation is used for exocentric (headless) semi-fixed MWEs like names (Walter Burley Griffin) and dates (20 November). The fixed relation applies to completely fixed grammaticized (function word-like) MWE (like instead of, such as), whereas compound applies to endocentric (headed) MWE (like apple pie). To produce feature vector $V_{ud}$, it is better to treat MWE as single words, especially MWE with fixed relations between parts because considering each part of the MWE separately and averaging their embedding vectors may result in a meaningless vector. This point matters when the words of low-resource languages are first translated into other languages and then presented by an embedding of that language. Therefore, the procedure of producing feature vector $V_{ud}$ should be modified with a simple heuristic : every node of the UD tree within the shortest path between two entities or dependent to $e1$ or $e2$ which have a child node with fixed dependency type is considered with its child as one word. If the child has also a child with a fixed dependency, all of them are considered as one word. For example, Figure FIGREF17 shows the UD tree of a Farsi sentence which is the translation of the English sentence in Figure FIGREF12. Entities are distinguished from other nodes by putting a circle around them. The 5th and 6th nodes from the left make a multiword expression that means “about". Applying the above heuristic results in them both being considered as a single word and so the correct translation to another language is found. Some other examples of Farsi MWEs are “قبل از آن که/before", “در حالی که/while", “به درون/into", “به جز/except", and “بر روی/on". In French language there are also MWEs, such as “bien que/although", “en tant que/as", “tant de/so many", “afin de/in order to", “prés de/near". Apart from fixed, flat, and compound, there are grammatical relations which are language-specific and show MWE structures BIBREF27. If the target language has language-specific relations, the above heuristic should be applied to them. For example, compound :lvc relation, which is defined for several languages including Farsi, represents the dependence from the noun part to the light verb part of compound verbs. An example of this relation was shown in Figure FIGREF5. The words “ارائه/presentation" and “میدهد/give" together mean “present". ### Experiments
In this section, the experimental analysis of the proposed models is presented. We have implemented the cross-lingual variant of kernel functions for PI and RE tasks as described in section SECREF3 and measured the accuracy of models by testing them on the parallel data set. The main advantage of the proposed method is that it needs no data of the test language, in the sense that the model trained using the training data of a language, e.g. English, is directly used in the other languages, e.g. Farsi, Arabic, etc. From this point of view, the proposed method can only be compared with those methods that use no data (neither labeled nor un-labeled) of the test language or parallel corpus or machine translators between the training and test languages. One solution for cross-lingual tasks is to equip the high accurate neural networks proposed for each task with pre-trained multi-lingual word embeddings, without any change in the architecture of the network. Therefore, we re-implemented some deep methods and compared the proposed approach with them for both PI and RE tasks. ### Experiments ::: Paraphrase Identification
For this task, we made a parallel test dataset and implemented PT and SPT kernels and compared the results with two-channel CNN of Wang et al. BIBREF18. ### Experiments ::: Paraphrase Identification ::: Construction of Parallel Dataset
To prepare a multi-language corpus for PI, we employed an existing English corpus with its Arabic translation and made Farsi correspondence. Microsoft Research Paraphrase Corpus (MSRC) BIBREF28 mostly used by the researches for English PI task. It contains 4,076 and 1,725 pairs of sentences for the training and test, respectively. This data has been extracted from news sources on the web, and has been annotated by humans whether each pair captures a paraphrase equivalence relationship. PI relates to the task of Semantic Textual Similarity (STS), in which the goal is to capture the degree of equivalence of meaning rather than making a binary decision. SemEval-2017 task 1 put the emphasis on multi-lingual STS BIBREF29. They selected 510 pairs from the test part of the MSRC corpus, and translated them into Arabic by Arabic native speakers. All data have been manually tagged with a number from 0 to 5 to show the degree of similarity. The Arabic part of the STS dataset of SemEval-2017 is parallel to some parts of the MSRC test corpus. So there is a parallel English-Arabic dataset. Because of the similarity between PI and STS tasks, the dataset of STS can also be used in the PI task, just by converting the scores to 0 or 1. So, the original binary scores of the STS dataset have been retrieved from the MSRC corpus. As a result, a corpus with 510 pairs of English sentences and Arabic translation for PI task is ready. In addition to Arabic translation, we produced correspondence Farsi data by translation of parallel English-Arabic dataset into Farsi by a Farsi native speaker. In the experiments, MSRC corpus was divided as follows : 1) the training part of MSRC corpus for training ; 2) those data from test part of MSRC, which we don't have their Arabic or Farsi counterpart, for tuning hyper-parameters as development set ; and 3) 510 parallel English-Arabic-Farsi from the test part of MSRC for the test. Therefore, our training and test data have 4076 and 510 samples, respectively. Table TABREF21 shows the statistics of our data. ### Experiments ::: Paraphrase Identification ::: Tools and Setup
The classifiers were trained with the C-SVM learning algorithm within KeLP BIBREF30, which is a kernel-based machine learning framework and implemented tree kernels. We employed PT and SPT kernel functions. For evaluating node similarity in SPTK function, we used the same method described in BIBREF14 : if $n_1$ and $n_2$ are two identical syntactic nodes, $\sigma (n_1,n_2)$ denoted the similarity of $n_1$ and $n_2$ and is equal to 1. If $n_1$ and $n_2$ are two lexical nodes with the same POS tag, their similarity is computed as the cosine similarity of the corresponding vectors in a wordspace. In all other cases $\sigma = 0$. English wordspace was generated by using word2vec tool. In the cross-lingual setup, we need a vocabulary to find the translation of lexical nodes and then compute their similarity in a wordspace. For English-Arabic experiments, we used Almaany dictionary to find the translation of Arabic words into English. For English-Farsi experiments, we used the Aryanpour dictionary to extract the English equivalent of Farsi words. To evaluate the performance of the classifiers we used Accuracy and F$_1$ as the previous works BIBREF31, BIBREF32, BIBREF18. For dependency parsing, UDPipe was used, which is a trainable pipeline for tokenization, tagging, lemmatization, and dependency parsing. We used version 2.4 of the UD pre-trained models of English, Arabic, and Farsi. To implement the CNN network of Wang et al. BIBREF18, we used the same word embedding they used. They set the size of the word vector dimension as d =300, and pre-trained the vectors with the word2vec toolkit on the English Gigaword (LDC2011T07). Hyper-parameters of the network are the same as their work. ### Experiments ::: Paraphrase Identification ::: Results
We first examine the tree kernels in the mono-lingual and then in the cross-lingual learning. ### Experiments ::: Paraphrase Identification ::: Results ::: Evaluation of tree-based models in mono-lingual learning
In the first experiment, we benchmark the UD-based models on the monolingual dataset. So, we employed the original split of MSRC corpus and trained models using PT and SPT kernels. These models essentially work based on the lexico-syntactic patterns observed in training sentences. Filice et al. BIBREF14 proposed several kernels including linear, graph and SPT kernels. They showed the best accuracy is obtained using the combination of them. However, we use only tree kernels in cross-lingual experiments, to measure how much we can rely on the similarities of UD parse trees in different languages. As Table TABREF29 shows, tree kernels including PTK and SPTK show comparable results according to the accuracy and F$_1$ measures. This means that PT and SPT kernels, which are trained by UD parse trees, make accurate models that can be used in solving the PI task. In the next experiment, we use these models to evaluate Arabic and Farsi test data. ### Experiments ::: Paraphrase Identification ::: Results ::: Evaluation of tree-based models with UD in cross-lingual learning
Now, we employ the parallel dataset for cross-lingual evaluation of the UD-based model trained by English data. A baseline for this task is the majority voting in that what we get if we always predict the most frequent label of the training data. A better baseline for cross-lingual PI is to use some neural models and couple them with pre-induced multilingual embeddings. So, we re-run the two-channel CNN model of Wang et al. BIBREF18 by our test data. Upper bound for the cross-lingual experiment is considered the accuracy of the model when it is evaluated by the data of the same language of the training data, e.g. English. Table TABREF30 shows that using PTK 61.6% of accuracy is obtained for English test data. It is 57.7% and 57.3% for Arabic and Farsi, respectively ; while the accuracy of the majority baseline is 50.6%. CNN model obtained similar accuracy but much lower F$_1$ scores. Comparing the results of Tables TABREF29 and TABREF30 reveals that the accuracy of both kernels drops significantly when they are tested by our small test data. The reason is that the distribution of MSRC training data over positive and negative classes is significantly different from our test data. Specifically, 67.5% of MSRC's training data are positive ; while 50.5% of our test data are positive. ### Experiments ::: Paraphrase Identification ::: Results ::: Evaluation of tree-based models with parse formalisms rather than UD
In this experiment, we produced dependency parse trees of Farsi data employing Hazm parser which is trained on non-UD tree-bank. Table TABREF30 shows that in this case accuracy of the models significantly drops. Taking a deeper look at the tree kernels, PTK doesn't use the similarity of words and works based on exact matching of them. So, in cross-lingual experiments, it considers only the similarity of trees. In this case, accuracy on Farsi test data is 50.6% which is the same as the majority baseline. This experiment reveals that the trees of parallel sentences that are produced by UD parsers are significantly more similar than the trees generated by other formalisms. ### Experiments ::: Relation Extraction
In this section, we explain the experiments of cross-lingual RE and present the results. Specifically, we compared tree-based methods including combination of tree kernels and TreeLSTM with deep methods of CNN BIBREF33, Bi-LSTM BIBREF34 and RCNN BIBREF35. ### Experiments ::: Relation Extraction ::: Construction of Parallel Dataset
SemEval 2010 released a dataset for relation extraction in task 8 BIBREF36, which is used by many researchers. This dataset contains 8000 samples for the training and 2717 samples for the test. It was annotated with 19 types of relations : 9 semantically different relationships (with two directions) and an undirected Other class. A brief description of these relation types is given in Table TABREF34. The SemEval-2010 dataset is in English. For cross-lingual experiments, the first 1000 samples of the test part were translated into Farsi and French. Two native Farsi and French speakers with high expertise in English were asked to translate the data. ### Experiments ::: Relation Extraction ::: Tools and Setup
Similar to PI's experiments, KeLP was used to implement the kernel combination. The strategy for dealing with multiple classes is “one versus others”. For constituency parsing, Stanford CoreNLP was used that contains pre-trained models for English and French within the Stanford package. For parsing Farsi data, the University of Tehran’s constituency parser BIBREF37 was used. Parameter $\alpha $ of the formula DISPLAY_FORM14-DISPLAY_FORM16 is 0.23 as the previous works BIBREF16. To obtain bi-lingual word embeddings, the multiCluster method of Ammar et al. BIBREF38 was used and 512-dimensional vectors were trained for English, French, and Farsi. ### Experiments ::: Relation Extraction ::: Result
We first examine the tree kernels in the mono-lingual and then in the cross-lingual learning. ### Experiments ::: Relation Extraction ::: Result ::: Evaluation of tree-based models in mono-lingual learning
There is a huge amount of works on RE, which mainly utilizes neural networks. These methods use different features including lexical, grammatical, and semantic features such as POS, WordNet, and dependency parsing. Table TABREF39 shows the state-of-the-art neural models evaluated by SemEval 2010-task 8 test set (2717 samples). The best proposed method, $CK_1$, obtained 84.0% of F$_1$ which is comparable with the others. ### Experiments ::: Relation Extraction ::: Result ::: Evaluation of tree-based models with UD in cross-lingual learning
Table TABREF40 shows accuracy of 84.2% F$_1$ score for $CK_1$ when tested on the first 1000 samples of English test data. The accuracy of this model for its Farsi and French counterparts is 53.4% and 61.2% respectively. This kernel employs sentence context, and so it didn't show exciting results in the cross-lingual experiment ; especially for Farsi data. This is because Farsi is one of the SOV languages, in contrast to English and French, which are SVO. This means verbs are usually at the end of the sentence in Farsi. When the sentence's verb is highly informative for the relation between two entities, it places outside the window surrounding two entities and so it doesn't contribute to the feature vector $V_o$. Table TABREF40 show the F$_1$ score of the models trained by $CK_2$ and $CK_3$. These kernels utilize the context words of the UD trees. Comparing three kernels, F$_1$ increased from 53.4% to 65.2% for Farsi, and to 67.5% for the French test data. The best result for Farsi came from kernel $CK_2$ ; whereas $CK_3$ performed better with the French data. Thus, it can be concluded that the constituency-based parse trees of English and French data have more similar sub-trees than English and Farsi. The reason partially relates to the common tool for English and French ; because Stanford CoreNLP has pre-trained models for both of these languages. Therefore, English and French models followed the same schema, while Farsi adopted different schema for constituency parsing. In addition to the composite kernels, we trained a Tree-LSTM model over the UD parse trees. Tree-LSTM doesn't process the syntactic features of the input sentence, rather it takes the tokens in order of the tree's node. However, to contribute the grammatical features, for each token its word embedding was concatenated to its dependency type embedding and its POS tag embedding. The resulting network obtained 80.0% of F$_1$ when tested by English test data. F$_1$ of this model is 52.0% for Farsi and 55.6% for French. Although the Tree-LSTM model obtained lower F$_1$ in comparison with the tree kernels, it still does better than deep baselines : we re-implemented the CNN model of Qin et al. BIBREF33, Att-BiLSTM of Zhou et al. BIBREF34, and RCNN of Lai et al. BIBREF35. All networks use bilingual word embeddings in the embedding layer. As Table TABREF40 shows the best F$_1$ scores were obtained by RCNN which utilizes CNN over the LSTM layer. However, the results are significantly lower than the UD-based models, specifically in Farsi. Because word order of Farsi and English sentences are very different ; as Farsi is SOV and English is SVO. ### Experiments ::: Relation Extraction ::: Result ::: Effect of Multi-Word Expressions
Last two rows of Table TABREF40 show the F$_1$ score of the model trained on the English training data using the $CK2$ and $CK3$, in which MWEs were considered to be a single node within the dependency tree, as described at the end of Section SECREF10. The accuracy of $CK_2$ mainly increased for the Farsi data, because Farsi has many multi-word expressions such as compound verbs. Farsi has only about 250 simple verbs and all the other verbs are compound BIBREF43. Considering MWE as a single node causes all the tokens which compose a verb to be treated as a single word, and so the true translation will be found when searching for that word in dictionaries. Figure FIGREF46 shows the F$_1$ scores of best models for different semantic classes. ### Discussion and Conclusion
Taking a deeper look at the proposed method, most of the mis-classifications of the cross-lingual tree models are related to the following issues : Structural Difference : The main reason for the error of classifiers is structural differences. Although UD tries to produce as most similar trees as it can for parallel sentences, there are many language-specific dependency patterns that could not be neglected. Lexical Gap : Words mainly convey the meaning of the sentence. A lexical gap between source and target languages usually ruins the accuracy of cross-lingual models. Confusion of different senses on a surface : Words of different languages usually have multiple senses. Confusion of different senses of words causes incorrect translation of words, because dictionaries translate word to word, but not word-sense to word-sense. On the other hand, Word Sense Disambiguation (WSD) is a difficult task and needs additional resources such as high-quality multi-lingual wordnets BIBREF44. Incorrect translation of prepositions : Prepositions are very informative for the RE task. Hashimoto et al. presented the five most informative unigrams and three-grams for three types of relations of the SemEval 2010-task 8 dataset BIBREF21, which are shown in Table TABREF47. Wang et al. BIBREF42 also presented the most representative trigrams for different relations on the same data set. Also, Lahbib et al. BIBREF45 presented the most common Arabic prepositions and showed that each one reflects some specific kinds of semantic relations. Confusion of senses for prepositions is a very common issue in word-to-word translation. Phrasal verbs : Phrasal verbs, which have a metaphorical meaning, often cannot be translated word for word. For example, the Farsi verb “از دست دادن / to give from hand”, means “lose". When the most informative chunk of the sentence is the phrasal verb, the proposed method does not capture the true meaning. In general, more lexical and structural similarities between the source and target languages increase the accuracy of UD-based transfer learning. As future works, it is proposed that the UD-based approach is studied for other cross-lingual learning tasks and other languages along with different learning algorithms that are capable of dealing with parse trees. Figure 1: UD annotations for equivalent sentences in English, Farsi, and French. English translation of each word is written below it. Figure 2: Two different parse trees for the sentence “The most common [audits]e1 were about [waste]e2 and recycling". Figure 3: Example of MWE in a Farsi sentence. The nodes 5th and 6th from the left make an MWE, while only the 6th node is dependent on the entity. Table 1: MSRC corpus split Table 3: Results of cross-lingual PI over our split of MSRC dataset. Table 2: Results of the mono-lingual English PI using tree kernels compared with previous methods on the original split of the MSRC dataset. Table 4: Relation types of SemEval 2010 dataset [15]. Table 5: F1 scores and features used by mono-lingual RE methods evaluated on the SemEval 2010-task 8 dataset. Table 6: F1 scores of different tree-based models of RE compared to the neural models (our reimplementation) on the SemEval 2010 dataset. Figure 4: F1 scores of different classes on the Farsi and French test sets using CK2 and CK3 kernels respectively (the best model for each language). Table 7: Top five most informative words for some relations of the SemEval 2010-task 8 dataset [14].
|
Paraphrase Identification
|
Which clinically validated survey tools are used?
|
### Introduction
Combat veterans diagnosed with PTSD are substantially more likely to engage in a number of high risk activities including engaging in interpersonal violence, attempting suicide, committing suicide, binge drinking, and drug abuse BIBREF0. Despite improved diagnostic screening, outpatient mental health and inpatient treatment for PTSD, the syndrome remains treatment resistant, is typically chronic, and is associated with numerous negative health effects and higher treatment costs BIBREF1. As a result, the Veteran Administration's National Center for PTSD (NCPTSD) suggests to reconceptualize PTSD not just in terms of a psychiatric symptom cluster, but focusing instead on the specific high risk behaviors associated with it, as these may be directly addressed though behavioral change efforts BIBREF0. Consensus prevalence estimates suggest that PTSD impacts between 15-20% of the veteran population which is typically chronic and treatment resistant BIBREF0. The PTSD patients support programs organized by different veterans peer support organization use a set of surveys for local weekly assessment to detect the intensity of PTSD among the returning veterans. However, recent advanced evidence-based care for PTSD sufferers surveys have showed that veterans, suffered with chronic PTSD are reluctant in participating assessments to the professionals which is another significant symptom of war returning veterans with PTSD. Several existing researches showed that, twitter posts of war veterans could be a significant indicator of their mental health and could be utilized to predict PTSD sufferers in time before going out of control BIBREF2, BIBREF3, BIBREF4, BIBREF5, BIBREF6, BIBREF7, BIBREF8. However, all of the proposed methods relied on either blackbox machine learning methods or language models based sentiments extraction of posted texts which failed to obtain acceptability and trust of clinicians due to the lack of their explainability. In the context of the above research problem, we aim to answer the following research questions Given clinicians have trust on clinically validated PTSD assessment surveys, can we fill out PTSD assessment surveys using twitter posts analysis of war-veterans? If possible, what sort of analysis and approach are needed to develop such XAI model to detect the prevalence and intensity of PTSD among war-veterans only using the social media (twitter) analysis where users are free to share their everyday mental and social conditions? How much quantitative improvement do we observe in our model's ability to explain both detection and intensity estimation of PTSD? In this paper, we propose LAXARY, an explainable and trustworthy representation of PTSD classification and its intensity for clinicians. The key contributions of our work are summarized below, The novelty of LAXARY lies on the proposed clinical surveys-based PTSD Linguistic dictionary creation with words/aspects which represents the instantaneous perturbation of twitter-based sentiments as a specific pattern and help calculate the possible scores of each survey question. LAXARY includes a modified LIWC model to calculate the possible scores of each survey question using PTSD Linguistic Dictionary to fill out the PTSD assessment surveys which provides a practical way not only to determine fine-grained discrimination of physiological and psychological health markers of PTSD without incurring the expensive and laborious in-situ laboratory testing or surveys, but also obtain trusts of clinicians who are expected to see traditional survey results of the PTSD assessment. Finally, we evaluate the accuracy of LAXARY model performance and reliability-validity of generated PTSD Linguistic Dictionary using real twitter users' posts. Our results show that, given normal weekly messages posted in twitter, LAXARY can provide very high accuracy in filling up surveys towards identifying PTSD ($\approx 96\%$) and its intensity ($\approx 1.2$ mean squared error). ### Overview
Fig. FIGREF7 shows a schematic representation of our proposed model. It consists of the following logical steps: (i) Develop PTSD Detection System using twitter posts of war-veterans(ii) design real surveys from the popular symptoms based mental disease assessment surveys; (iii) define single category and create PTSD Linguistic Dictionary for each survey question and multiple aspect/words for each question; (iv) calculate $\alpha $-scores for each category and dimension based on linguistic inquiry and word count as well as the aspects/words based dictionary; (v) calculate scaling scores ($s$-scores) for each dimension based on the $\alpha $-scores and $s$-scores of each category based on the $s$-scores of its dimensions; (vi) rank features according to the contributions of achieving separation among categories associated with different $\alpha $-scores and $s$-scores; and select feature sets that minimize the overlap among categories as associated with the target classifier (SGD); and finally (vii) estimate the quality of selected features-based classification for filling up surveys based on classified categories i.e. PTSD assessment which is trustworthy among the psychiatry community. ### Related Works
Twitter activity based mental health assessment has been utmost importance to the Natural Language Processing (NLP) researchers and social media analysts for decades. Several studies have turned to social media data to study mental health, since it provides an unbiased collection of a person's language and behavior, which has been shown to be useful in diagnosing conditions. BIBREF9 used n-gram language model (CLM) based s-score measure setting up some user centric emotional word sets. BIBREF10 used positive and negative PTSD data to train three classifiers: (i) one unigram language model (ULM); (ii) one character n-gram language model (CLM); and 3) one from the LIWC categories $\alpha $-scores and found that last one gives more accuracy than other ones. BIBREF11 used two types of $s$-scores taking the ratio of negative and positive language models. Differences in language use have been observed in the personal writing of students who score highly on depression scales BIBREF2, forum posts for depression BIBREF3, self narratives for PTSD (BIBREF4, BIBREF5), and chat rooms for bipolar BIBREF6. Specifically in social media, differences have previously been observed between depressed and control groups (as assessed by internet-administered batteries) via LIWC: depressed users more frequently use first person pronouns (BIBREF7) and more frequently use negative emotion words and anger words on Twitter, but show no differences in positive emotion word usage (BIBREF8). Similarly, an increase in negative emotion and first person pronouns, and a decrease in third person pronouns, (via LIWC) is observed, as well as many manifestations of literature findings in the pattern of life of depressed users (e.g., social engagement, demographics) (BIBREF12). Differences in language use in social media via LIWC have also been observed between PTSD and control groups (BIBREF13). All of the prior works used some random dictionary related to the human sentiment (positive/negative) word sets as category words to estimate the mental health but very few of them addressed the problem of explainability of their solution to obtain trust of clinicians. Islam et. al proposed an explainable topic modeling framework to rank different mental health features using Local Interpretable Model-Agnostic Explanations and visualize them to understand the features involved in mental health status classification using the BIBREF14 which fails to provide trust of clinicians due to its lack of interpretability in clinical terms. In this paper, we develop LAXARY model where first we start investigating clinically validated survey tools which are trustworthy methods of PTSD assessment among clinicians, build our category sets based on the survey questions and use these as dictionary words in terms of first person singular number pronouns aspect for next level LIWC algorithm. Finally, we develop a modified LIWC algorithm to estimate survey scores (similar to sentiment category scores of naive LIWC) which is both explainable and trustworthy to clinicians. ### Demographics of Clinically Validated PTSD Assessment Tools
There are many clinically validated PTSD assessment tools that are being used both to detect the prevalence of PTSD and its intensity among sufferers. Among all of the tools, the most popular and well accepted one is Domain-Specific Risk-Taking (DOSPERT) Scale BIBREF15. This is a psychometric scale that assesses risk taking in five content domains: financial decisions (separately for investing versus gambling), health/safety, recreational, ethical, and social decisions. Respondents rate the likelihood that they would engage in domain-specific risky activities (Part I). An optional Part II assesses respondents' perceptions of the magnitude of the risks and expected benefits of the activities judged in Part I. There are more scales that are used in risky behavior analysis of individual's daily activities such as, The Berlin Social Support Scales (BSSS) BIBREF16 and Values In Action Scale (VIAS) BIBREF17. Dryhootch America BIBREF18, BIBREF19, a veteran peer support community organization, chooses 5, 6 and 5 questions respectively from the above mentioned survey systems to assess the PTSD among war veterans and consider rest of them as irrelevant to PTSD. The details of dryhootch chosen survey scale are stated in Table TABREF13. Table!TABREF14 shows a sample DOSPERT scale demographic chosen by dryhootch. The threshold (in Table TABREF13) is used to calculate the risky behavior limits. For example, if one individual's weekly DOSPERT score goes over 28, he is in critical situation in terms of risk taking symptoms of PTSD. Dryhootch defines the intensity of PTSD into four categories based on the weekly survey results of all three clinical survey tools (DOSPERT, BSSS and VIAS ) High risk PTSD: If one individual veteran's weekly PTSD assessment scores go above the threshold for all three PTSD assessment tools i.e. DOSPERT, BSSS and VIAS, then he/she is in high risk situation which needs immediate mental support to avoid catastrophic effect of individual's health or surrounding people's life. Moderate risk PTSD: If one individual veteran's weekly PTSD assessment scores go above the threshold for any two of the three PTSD assessment tools, then he/she is in moderate risk situation which needs close observation and peer mentoring to avoid their risk progression. Low risk PTSD: If one individual veteran's weekly PTSD assessment scores go above the threshold for any one of the three PTSD assessment tools, then he/she has light symptoms of PTSD. No PTSD: If one individual veteran's weekly PTSD assessment scores go below the threshold for all three PTSD assessment tools, then he/she has no PTSD. ### Twitter-based PTSD Detection
To develop an explainable model, we first need to develop twitter-based PTSD detection algorithm. In this section, we describe the data collection and the development of our core LAXARY model. ### Twitter-based PTSD Detection ::: Data Collection
We use an automated regular expression based searching to find potential veterans with PTSD in twitter, and then refine the list manually. First, we select different keywords to search twitter users of different categories. For example, to search self-claimed diagnosed PTSD sufferers, we select keywords related to PTSD for example, post trauma, post traumatic disorder, PTSD etc. We use a regular expression to search for statements where the user self-identifies as being diagnosed with PTSD. For example, Table TABREF27 shows a self-identified tweet posts. To search veterans, we mostly visit to different twitter accounts of veterans organizations such as "MA Women Veterans @WomenVeterans", "Illinois Veterans @ILVetsAffairs", "Veterans Benefits @VAVetBenefits" etc. We define an inclusion criteria as follows: one twitter user will be part of this study if he/she describes himself/herself as a veteran in the introduction and have at least 25 tweets in last week. After choosing the initial twitter users, we search for self-identified PTSD sufferers who claim to be diagnosed with PTSD in their twitter posts. We find 685 matching tweets which are manually reviewed to determine if they indicate a genuine statement of a diagnosis for PTSD. Next, we select the username that authored each of these tweets and retrieve last week's tweets via the Twitter API. We then filtered out users with less than 25 tweets and those whose tweets were not at least 75% in English (measured using an automated language ID system.) This filtering left us with 305 users as positive examples. We repeated this process for a group of randomly selected users. We randomly selected 3,000 twitter users who are veterans as per their introduction and have at least 25 tweets in last one week. After filtering (as above) in total 2,423 users remain, whose tweets are used as negative examples developing a 2,728 user's entire weeks' twitter posts where 305 users are self-claimed PTSD sufferers. We distributed Dryhootch chosen surveys among 1,200 users (305 users are self claimed PTSD sufferers and rest of them are randomly chosen from previous 2,423 users) and received 210 successful responses. Among these responses, 92 users were diagnosed as PTSD by any of the three surveys and rest of the 118 users are diagnosed with NO PTSD. Among the clinically diagnosed PTSD sufferers, 17 of them were not self-identified before. However, 7 of the self-identified PTSD sufferers are assessed with no PTSD by PTSD assessment tools. The response rates of PTSD and NO PTSD users are 27% and 12%. In summary, we have collected one week of tweets from 2,728 veterans where 305 users claimed to have diagnosed with PTSD. After distributing Dryhootch surveys, we have a dataset of 210 veteran twitter users among them 92 users are assessed with PTSD and 118 users are diagnosed with no PTSD using clinically validated surveys. The severity of the PTSD are estimated as Non-existent, light, moderate and high PTSD based on how many surveys support the existence of PTSD among the participants according to dryhootch manual BIBREF18, BIBREF19. ### Twitter-based PTSD Detection ::: Pre-processing
We download 210 users' all twitter posts who are war veterans and clinically diagnosed with PTSD sufferers as well which resulted a total 12,385 tweets. Fig FIGREF16 shows each of the 210 veteran twitter users' monthly average tweets. We categorize these Tweets into two groups: Tweets related to work and Tweets not related to work. That is, only the Tweets that use a form of the word “work*” (e.g. work,worked, working, worker, etc.) or “job*” (e.g. job, jobs, jobless, etc.) are identified as work-related Tweets, with the remaining categorized as non-work-related Tweets. This categorization method increases the likelihood that most Tweets in the work group are indeed talking about work or job; for instance, “Back to work. Projects are firing back up and moving ahead now that baseball is done.” This categorization results in 456 work-related Tweets, about 5.4% of all Tweets written in English (and 75 unique Twitter users). To conduct weekly-level analysis, we consider three categorizations of Tweets (i.e. overall Tweets, work-related Tweets, and non work-related Tweets) on a daily basis, and create a text file for each week for each group. ### Twitter-based PTSD Detection ::: PTSD Detection Baseline Model
We use Coppersmith proposed PTSD classification algorithm to develop our baseline blackbox model BIBREF11. We utilize our positive and negative PTSD data (+92,-118) to train three classifiers: (i) unigram language model (ULM) examining individual whole words, (ii) character n-gram language model (CLM), and (iii) LIWC based categorical models above all of the prior ones. The LMs have been shown effective for Twitter classification tasks BIBREF9 and LIWC has been previously used for analysis of mental health in Twitter BIBREF10. The language models measure the probability that a word (ULM) or a string of characters (CLM) was generated by the same underlying process as the training data. We first train one of each language model ($clm^{+}$ and $ulm^{+}$) from the tweets of PTSD users, and another model ($clm^{-}$ and $ulm^{-}$) from the tweets from No PTSD users. Each test tweet $t$ is scored by comparing probabilities from each LM called $s-score$ A threshold of 1 for $s-score$ divides scores into positive and negative classes. In a multi-class setting, the algorithm minimizes the cross entropy, selecting the model with the highest probability. For each user, we calculate the proportion of tweets scored positively by each LIWC category. These proportions are used as a feature vector in a loglinear regression model BIBREF20. Prior to training, we preprocess the text of each tweet: we replace all usernames with a single token (USER), lowercase all text, and remove extraneous whitespace. We also exclude any tweet that contained a URL, as these often pertain to events external to the user. We conduct a LIWC analysis of the PTSD and non-PTSD tweets to determine if there are differences in the language usage of PTSD users. We applied the LIWC battery and examined the distribution of words in their language. Each tweet was tokenized by separating on whitespace. For each user, for a subset of the LIWC categories, we measured the proportion of tweets that contained at least one word from that category. Specifically, we examined the following nine categories: first, second and third person pronouns, swear, anger, positive emotion, negative emotion, death, and anxiety words. Second person pronouns were used significantly less often by PTSD users, while third person pronouns and words about anxiety were used significantly more often. ### LAXARY: Explainable PTSD Detection Model
The heart of LAXARY framework is the construction of PTSD Linguistic Dictionary. Prior works show that linguistic dictionary based text analysis has been much effective in twitter based sentiment analysis BIBREF21, BIBREF22. Our work is the first of its kind that develops its own linguistic dictionary to explain automatic PTSD assessment to confirm trustworthiness to clinicians. ### LAXARY: Explainable PTSD Detection Model ::: PTSD Linguistic Dictionary Creation
We use LIWC developed WordStat dictionary format for our text analysis BIBREF23. The LIWC application relies on an internal default dictionary that defines which words should be counted in the target text files. To avoid confusion in the subsequent discussion, text words that are read and analyzed by WordStat are referred to as target words. Words in the WordStat dictionary file will be referred to as dictionary words. Groups of dictionary words that tap a particular domain (e.g., negative emotion words) are variously referred to as subdictionaries or word categories. Fig FIGREF8 is a sample WordStat dictionary. There are several steps to use this dictionary which are stated as follows: Pronoun selection: At first we have to define the pronouns of the target sentiment. Here we used first person singular number pronouns (i.e., I, me, mine etc.) that means we only count those sentences or segments which are only related to first person singular number i.e., related to the person himself. Category selection: We have to define the categories of each word set thus we can analyze the categories as well as dimensions' text analysis scores. We chose three categories based on the three different surveys: 1) DOSPERT scale; 2) BSSS scale; and 3) VIAS scale. Dimension selection: We have to define the word sets (also called dimension) for each category. We chose one dimension for each of the questions under each category to reflect real survey system evaluation. Our chosen categories are state in Fig FIGREF20. Score calculation $\alpha $-score: $\alpha $-scores refer to the Cronbach's alphas for the internal reliability of the specific words within each category. The binary alphas are computed on the ratio of occurrence and non-occurrence of each dictionary word whereas the raw or uncorrected alphas are based on the percentage of use of each of the category words within texts. ### LAXARY: Explainable PTSD Detection Model ::: Psychometric Validation of PTSD Linguistic Dictionary
After the PTSD Linguistic Dictionary has been created, we empirically evaluate its psychometric properties such as reliability and validity as per American Standards for educational and psychological testing guideline BIBREF24. In psychometrics, reliability is most commonly evaluated by Cronbach's alpha, which assesses internal consistency based on inter-correlations and the number of measured items. In the text analysis scenario, each word in our PTSD Linguistic dictionary is considered an item, and reliability is calculated based on each text file's response to each word item, which forms an $N$(number of text files) $\times $ $J$(number of words or stems in a dictionary) data matrix. There are two ways to quantify such responses: using percentage data (uncorrected method), or using "present or not" data (binary method) BIBREF23. For the uncorrected method, the data matrix comprises percentage values of each word/stem are calculated from each text file. For the binary method, the data matrix quantifies whether or not a word was used in a text file where "1" represents yes and "0" represents no. Once the data matrix is created, it is used to calculate Cronbach's alpha based on its inter-correlation matrix among the word percentages. We assess reliability based on our selected 210 users' Tweets which further generated a 23,562 response matrix after running the PTSD Linguistic Dictionary for each user. The response matrix yields reliability of .89 based on the uncorrected method, and .96 based on the binary method, which confirm the high reliability of our PTSD Dictionary created PTSD survey based categories. After assessing the reliability of the PTSD Linguistic dictionary, we focus on the two most common forms of construct validity: convergent validity and discriminant validity BIBREF25. Convergent validity provides evidence that two measures designed to assess the same construct are indeed related; discriminate validity involves evidence that two measures designed to assess different constructs are not too strongly related. In theory, we expect that the PTSD Linguistic dictionary should be positively correlated with other negative PTSD constructs to show convergent validity, and not strongly correlated with positive PTSD constructs to show discriminant validity. To test these two types of validity, we use the same 210 users' tweets used for the reliability assessment. The results revealed that the PTSD Linguistic dictionary is indeed positively correlated with negative construct dictionaries, including the overall negative PTSD dictionary (r=3.664,p$<$.001). Table TABREF25 shows all 16 categorical dictionaries. These results provide strong support for the measurement validity for our newly created PTSD Linguistic dictionary. ### LAXARY: Explainable PTSD Detection Model ::: Feature Extraction and Survey Score Estimation
We use the exact similar method of LIWC to extract $\alpha $-scores for each dimension and categories except we use our generated PTSD Linguistic Dictionary for the task BIBREF23. Thus we have total 16 $\alpha $-scores in total. Meanwhile, we propose a new type of feature in this regard, which we called scaling-score ($s$-score). $s$-score is calculated from $\alpha $-scores. The purpose of using $s$-score is to put exact scores of each of the dimension and category thus we can apply the same method used in real weekly survey system. The idea is, we divide each category into their corresponding scale factor (i.e., for DOSPERT scale, BSSS scale and VIAS scales) and divide them into 8, 3 and 5 scaling factors which are used in real survey system. Then we set the $s$-score from the scaling factors from the $\alpha $-scores of the corresponding dimension of the questions. The algorithm is stated in Figure FIGREF23. Following Fig FIGREF23, we calculate the $s$-score for each dimension. Then we add up all the $s$-score of the dimensions to calculate cumulative $s$-score of particular categories which is displayed in Fig FIGREF22. Finally, we have total 32 features among them 16 are $\alpha $-scores and 16 are $s$-scores for each category (i.e. each question). We add both of $\alpha $ and $s$ scores together and scale according to their corresponding survey score scales using min-max standardization. Then, the final output is a 16 valued matrix which represent the score for each questions from three different Dryhootch surveys. We use the output to fill up each survey, estimate the prevalence of PTSD and its intensity based on each tool's respective evaluation metric. ### Experimental Evaluation
To validate the performance of LAXARY framework, we first divide the entire 210 users' twitter posts into training and test dataset. Then, we first developed PTSD Linguistic Dictionary from the twitter posts from training dataset and apply LAXARY framework on test dataset. ### Experimental Evaluation ::: Results
To provide an initial results, we take 50% of users' last week's (the week they responded of having PTSD) data to develop PTSD Linguistic dictionary and apply LAXARY framework to fill up surveys on rest of 50% dataset. The distribution of this training-test dataset segmentation followed a 50% distribution of PTSD and No PTSD from the original dataset. Our final survey based classification results showed an accuracy of 96% in detecting PTSD and mean squared error of 1.2 in estimating its intensity given we have four intensity, No PTSD, Low Risk PTSD, Moderate Risk PTSD and High Risk PTSD with a score of 0, 1, 2 and 3 respectively. Table TABREF29 shows the classification details of our experiment which provide the very good accuracy of our classification. To compare the outperformance of our method, we also implemented Coppersmith et. al. proposed method and achieved an 86% overall accuracy of detecting PTSD users BIBREF11 following the same training-test dataset distribution. Fig FIGREF28 illustrates the comparisons between LAXARY and Coppersmith et. al. proposed method. Here we can see, the outperformance of our proposed method as well as the importance of $s-score$ estimation. We also illustrates the importance of $\alpha -score$ and $S-score$ in Fig FIGREF30. Fig FIGREF30 illustrates that if we change the number of training samples (%), LAXARY models outperforms Coppersmith et. al. proposed model under any condition. In terms of intensity, Coppersmith et. al. totally fails to provide any idea however LAXARY provides extremely accurate measures of intensity estimation for PTSD sufferers (as shown in Fig FIGREF31) which can be explained simply providing LAXARY model filled out survey details. Table TABREF29 shows the details of accuracies of both PTSD detection and intensity estimation. Fig FIGREF32 shows the classification accuracy changes over the training sample sizes for each survey which shows that DOSPERT scale outperform other surveys. Fig FIGREF33 shows that if we take previous weeks (instead of only the week diagnosis of PTSD was taken), there are no significant patterns of PTSD detection. ### Challenges and Future Work
LAXARY is a highly ambitious model that targets to fill up clinically validated survey tools using only twitter posts. Unlike the previous twitter based mental health assessment tools, LAXARY provides a clinically interpretable model which can provide better classification accuracy and intensity of PTSD assessment and can easily obtain the trust of clinicians. The central challenge of LAXARY is to search twitter users from twitter search engine and manually label them for analysis. While developing PTSD Linguistic Dictionary, although we followed exactly same development idea of LIWC WordStat dictionary and tested reliability and validity, our dictionary was not still validated by domain experts as PTSD detection is highly sensitive issue than stress/depression detection. Moreover, given the extreme challenges of searching veterans in twitter using our selection and inclusion criteria, it was extremely difficult to manually find the evidence of the self-claimed PTSD sufferers. Although, we have shown extremely promising initial findings about the representation of a blackbox model into clinically trusted tools, using only 210 users' data is not enough to come up with a trustworthy model. Moreover, more clinical validation must be done in future with real clinicians to firmly validate LAXARY model provided PTSD assessment outcomes. In future, we aim to collect more data and run not only nationwide but also international-wide data collection to establish our innovation into a real tool. Apart from that, as we achieved promising results in detecting PTSD and its intensity using only twitter data, we aim to develop Linguistic Dictionary for other mental health issues too. Moreover, we will apply our proposed method in other types of mental illness such as depression, bipolar disorder, suicidal ideation and seasonal affective disorder (SAD) etc. As we know, accuracy of particular social media analysis depends on the dataset mostly. We aim to collect more data engaging more researchers to establish a set of mental illness specific Linguistic Database and evaluation technique to solidify the genralizability of our proposed method. ### Conclusion
To promote better comfort to the trauma patients, it is really important to detect Post Traumatic Stress Disorder (PTSD) sufferers in time before going out of control that may result catastrophic impacts on society, people around or even sufferers themselves. Although, psychiatrists invented several clinical diagnosis tools (i.e., surveys) by assessing symptoms, signs and impairment associated with PTSD, most of the times, the process of diagnosis happens at the severe stage of illness which may have already caused some irreversible damages of mental health of the sufferers. On the other hand, due to lack of explainability, existing twitter based methods are not trusted by the clinicians. In this paper, we proposed, LAXARY, a novel method of filling up PTSD assessment surveys using weekly twitter posts. As the clinical surveys are trusted and understandable method, we believe that this method will be able to gain trust of clinicians towards early detection of PTSD. Moreover, our proposed LAXARY model, which is first of its kind, can be used to develop any type of mental disorder Linguistic Dictionary providing a generalized and trustworthy mental health assessment framework of any kind. Fig. 1. Overview of our framework Fig. 2. WordStat dictionary sample TABLE I DRYHOOTCH CHOSEN PTSD ASSESSMENT SURVEYS (D: DOSPERT, B: BSSS AND V: VIAS) DEMOGRAPHICS TABLE II SAMPLE DRYHOOTCH CHOSEN QUESTIONS FROM DOSPERT Fig. 3. Each 210 users’ average tweets per month Fig. 4. Category Details Fig. 5. S-score table details Fig. 6. Comparisons between Coppersmith et. al. and our method TABLE V LAXARY MODEL BASED CLASSIFICATION DETAILS Fig. 7. Percentages of Training dataset and their PTSD detection accuracy results comparisons. Rest of the dataset has been used for testing Fig. 9. Percentages of Training dataset and their Accuracies for each Survey Tool. Rest of the dataset has been used for testing Fig. 8. Percentages of Training dataset and their Mean Squared Error (MSE) of PTSD Intensity. Rest of the dataset has been used for testing Fig. 10. Weekly PTSD detection accuracy change comparisons with baseline model
|
DOSPERT, BSSS and VIAS
|
Why did Moynihan shoot Stanley?
A. It was an accident.
B. Stanley tried to poison him.
C. Stanley was protecting Albert Peet.
D. He was hired to shoot Stanley.
|
Red Witch of Mercury By EMMETT McDOWELL Death was Jaro Moynahan's stock in trade, and every planet had known his touch. But now, on Mercury, he was selling his guns into the weirdest of all his exploits—gambling his life against the soft touch of a woman's lips. [Transcriber's Note: This etext was produced from Planet Stories Summer 1945. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] On the stage of Mercury Sam's Garden , a tight-frocked, limber-hipped, red-head was singing " The Lady from Mars ." The song was a rollicking, ribald ditty, a favorite of the planters and miners, the space pilots and army officers who frequented the garden. The girl rendered it with such gusto that the audience burst into a roar of applause. She bent her head in acknowledgment so that her bronze red hair fell down about her face. There was perspiration on her upper lip and temples. Her crimson mouth wore a fixed smile. Her eyes were frightened. The man, who had accompanied the singer on the piano, sat at the foot of the stage, his back to the crowded tables. He did not look up at the singer but kept his pale, immature face bent over the keys, while his fingers lightly, automatically picked out the tune. Sweat trickled down the back of his neck, plastered his white coat to his back. Without looking up, he said: "Have you spotted him?" His voice was pitched to reach the singer alone. The girl, with an almost imperceptible gesture, shook her head. The night was very hot; but then it is always hot on Mercury, the newest, the wildest, the hottest of Earth's frontiers. Fans spaced about the garden's walls sluggishly stirred the night air, while the men and women sitting at the tables drank heavily of Latonka, the pale green wine of Mercury. Only the native waiters, the enigmatic, yellow-eyed Mercurians, seemed unaffected by the heat. They didn't sweat at all. Up on the stage the singer was about to begin another number when she stiffened. "Here he is," she said to the pianist without moving her lips. The pianist swung around on his stool, lifted his black eyes to the gate leading to the street. Just within the entrance, a tall, thin man was standing. He looked like a gaunt gray wolf loitering in the doorway. His white duraloes suit hung faultlessly. His black hair was close-cropped, his nose thin and aquiline. For a moment he studied the crowded garden before making his way to a vacant table. "Go on," said the pianist in a flat voice. The red-head shivered. Stepping from the stage she picked her way through the tables until she came to the one occupied by the newcomer. "May I join you?" she asked in a low voice. The man arose. "Of course. I was expecting you. Here, sit down." He pulled out a chair, motioned for the waiter. The Mercurian, his yellow incurious eyes like two round topazes, sidled up. "Bring us a bottle of Latonka from the Veederman region, well iced." The waiter slipped away. "So," said the red-head; "you have come. I did not think you would be in time." Her hands were clenched in her lap. The knuckles were white. The man said nothing. "I did not want to call you in, Jaro Moynahan." It was the first time she had used his name. "You have the reputation of being unpredictable. I don't trust you, but since...." She stopped as the waiter placed glasses on the table and deftly poured the pale green wine. The man, Jaro Moynahan, raised his glass. "Here's to the revolution," he said. His low voice carried an odd, compelling note. His eyes, light blue and amused, were pale against his brown face. The girl drew in her breath. "No! Mercury is not ready for freedom. Only a handful of fanatics are engineering the revolution. The real Mercurian patriots are against it, but they are afraid to protest. You've got to believe me. The revolution is scheduled to break during the Festival of the Rains. If it does, the Terrestrials here will be massacred. The Mercurians hate them. We haven't but a handful of troops." Jaro Moynahan wiped the sweat from his forehead with a fine duraweb handkerchief. "I had forgotten how abominably hot it can be here." The girl ignored the interruption. "There is one man; he is the leader, the very soul of the revolution. The Mercurians worship him. They will do whatever he says. Without him they would be lost. He is the rebel, Karfial Hodes. I am to offer you ten thousand Earth notes to kill Karfial Hodes." Jaro Moynahan refilled their empty glasses. He was a big man, handsome in a gaunt fashion. Only his eyes were different. They were flat and a trifle oblique with straight brows. The pupils were a pale and penetrating blue that could probe like a surgeon's knife. Now he caught the girl's eyes and held them with his own as a man spears a fish. "Why call me all the way from Mars for that? Why not have that gunman at the piano rub Hodes out?" The girl started, glanced at the pianist, said with a shiver: "We can't locate Karfial Hodes. Don't look at me that way, Jaro. You frighten me. I'm telling the truth. We can't find him. That's why we called you. You've got to find him, Jaro. He's stirring up all Mercury." "Who's putting up the money?" "I can't tell you." "Ah," said Jaro Moynahan; "so that's the way it is." "That's the way it is." "There isn't much time," he said after a moment. "The Rains are due any day now." "No," the girl replied. "But we think he's here in the city." "Why? What makes you think that?" "He was seen," she began, then stopped with a gasp. The lights had gone out. It was as unexpected as a shot in the back. One moment the garden was glowing in light, the next the hot black night swooped down on the revelers, pressing against their eyes like dark wool. The fans about the walls slowed audibly and stopped. It grew hotter, closer. Jaro Moynahan slipped sideways from the table. He felt something brush his sleeve. Somewhere a girl giggled. "What's coming off here?" growled a petulant male voice. Other voices took up the plaint. Across the table from Jaro there was the feel of movement; he could sense it. An exclamation was suddenly choked off as if a hand had been clamped over the girl's mouth. "Red!" said Jaro in a low voice. There was no answer. "Red!" he repeated, louder. Unexpectedly, the deep, ringing voice of Mercury Sam boomed out from the stage. "It's all right. The master fuse blew out. The lights will be on in a moment." On the heels of his speech the lights flashed on, driving the night upward. The fans recommenced their monotonous whirring. Jaro Moynahan glanced at the table. The red-headed singer was gone. So was the pianist. Jaro Moynahan sat quietly back down and poured himself another glass of Latonka. The pale green wine had a delicate yet exhilarating taste. It made him think of cool green grapes beaded with dew. On the hot, teeming planet of Mercury it was as refreshing as a cold plunge. He wondered who was putting up the ten thousand Earth notes? Who stood to lose most in case of a revolution? The answer seemed obvious enough. Who, but Albert Peet. Peet controlled the Latonka trade for which there was a tremendous demand throughout the Universe. And what had happened to the girl. Had the rebels abducted her. If so, he suspected that they had caught a tartar. The Red Witch had the reputation of being able to take care of herself. He beckoned a waiter, paid his bill. As the Mercurian started to leave, a thought struck Jaro. These yellow-eyed Mercurians could see as well in the dark as any alley-prowling cat. For centuries they had lived most their lives beneath ground to escape the terrible rays of the sun. Only at night did they emerge to work their fields and ply their trades. He peeled off a bill, put it in the waiter's hands. "What became of the red-headed singer?" The Mercurian glanced at the bill, then back at the Earthman. There was no expression in his yellow eyes. "She and the man, the queer white one who plays the piano, slipped out the gate to the street." Jaro shrugged, dismissed the waiter. He had not expected to get much information from the waiter, but he was not a man to overlook any possibility. If the girl had been abducted, only Mercurians could have engineered it in the dark; and the Mercurians were a clannish lot. Back on the narrow alley-like street Jaro Moynahan headed for his hostelry. By stretching out his arms he could touch the buildings on either side: buildings with walls four feet thick to keep out the heat of the sun. Beneath his feet, he knew, stretched a labyrinth of rooms and passages. Somewhere in those rat-runs was Karfial Hodes, the revolutionist, and the girl. At infrequent intervals green globes cut a hole in the night, casting a faint illumination. He had just passed one of these futile street lamps when he thought he detected a footfall behind him. It was only the whisper of a sound, but as he passed beyond the circle of radiation, he flattened himself in a doorway. Nothing stirred. There was no further sound. Again he started forward, but now he was conscious of shadows following him. They were never visible, but to his trained ears there came stealthy, revealing noises: the brush of cloth against the baked earth walls, the sly shuffle of a step. He ducked down a bisecting alley, faded into a doorway. Immediately all sounds of pursuit stopped. But as soon as he emerged he was conscious again of the followers. In the dense, humid night, he was like a blind man trying to elude the cat-eyed Mercurians. Jaro Moynahan In the East a sullen red glow stained the heavens like the reflection of a fire. The Mercurian dawn was about to break. With an oath, he set out again for his hostelry. He made no further effort to elude the followers. Once back in his room, Jaro Moynahan stripped off his clothes, unbuckled a shoulder holster containing a compressed air slug gun, stepped under the shower. His body was lean and brown as his face and marked with innumerable scars. There were small round puckered scars and long thin ones, and his left shoulder bore the unmistakable brownish patch of a ray burn. Stepping out of the shower, he dried, rebuckled on the shoulder holster, slipped into pajamas. The pajamas were blue with wide gaudy stripes. Next he lit a cigarette and stretching out on the bed began to contemplate his toes with singular interest. He had, he supposed, killed rather a lot of men. He had fought in the deadly little wars of the Moons of Jupiter for years, then the Universal Debacle of 3368, after that the Martian Revolution as well as dozens of skirmishes between the Federated Venusian States. No, there was little doubt but that he had killed quite a number of men. But this business of hunting a man through the rat-runs beneath the city was out of his line. Furthermore, there was something phony about the entire set up. The Mercurians, he knew, had been agitating for freedom for years. Why, at this time when the Earth Congress was about to grant them self-government, should they stage a revolution? A loud, authoritative rapping at the door interrupted further speculation. He swung his bare feet over the edge of the bed, stood up and ground out his cigarette. Before he could reach the door the rapping came again. Throwing off the latch, he stepped back, balancing on the balls of his feet. "Come in," he called. The door swung open. A heavy set man entered, shut and locked the door, then glanced around casually. His eyes fastened on Jaro. He licked his lips. "Mr. Moynahan, the—ah—professional soldier, I believe." His voice was high, almost feminine. "I'm Albert Peet." He held out a fat pink hand. Jaro said nothing. He ignored the hand, waited, poised like a cat. Mr. Peet licked his lips again. "I have come, Mr. Moynahan, on a matter of business, urgent business. I had not intended to appear in this matter. I preferred to remain behind the scenes, but the disappearance of Miss Mikail has—ah—forced my hand." He paused. Jaro still said nothing. Miss Mikail must be the red-headed singer, whom at different times he had known under a dozen different aliases. He doubted that even she remembered her right name. "Miss Mikail made you a proposition?" Albert Peet's voice was tight. "Yes," said Jaro. "You accepted?" "Why, no. As it happened she was abducted before I had the chance." Mr. Peet licked his lips. "But you will, surely you will. Unless Karfial Hodes is stopped immediately there will be a bloody uprising all over the planet during the Festival of the Rains. Earth doesn't realize the seriousness of the situation." "Then I was right; it is you who are putting up the ten thousand Earth notes." "Not entirely," said Peet uncomfortably. "There are many of us here, Mercurians as well as Earthmen, who recognize the danger. We have—ah—pooled our resources." "But you stand to lose most in case of a successful revolution?" "Perhaps. I have a large interest in the Latonka trade. It is—ah—lucrative." Jaro Moynahan lit a cigarette, sat down on the edge of the bed. "Why beat about the bush," he asked with a sudden grin. "Mr. Peet, you've gained control of the Latonka trade. Other Earthmen are in control of the mines and the northern plantations. Together you form perhaps the strongest combine the Universe has ever seen. You actually run Mercury, and you've squeezed out every possible penny. Every time self-government has come before the Earth Congress you've succeeded in blocking it. You are, perhaps, the most cordially-hated group anywhere. I don't wonder that you are afraid of a revolution." Mr. Peet took out a handkerchief and mopped his forehead. "Fifteen thousand Earth notes I can offer you. But no more. That is as high as I can go." Jaro laughed. "How did you know Red had been kidnapped?" "We have a very efficient information system. I had the report of Miss Mikail's abduction fifteen minutes after the fact." Jaro raised his eyebrows. "Perhaps then you know where she is?" Mr. Peet shook his head. "No. Karfial Hodes' men abducted her." A second rapping at the door caused them to exchange glances. Jaro went to the door, opened it. The pianist at the gardens was framed in the entrance. His black eyes burned holes in his pale boyish face. His white suit was blotched with sweat and dirt. "They told me Mr. Peet was here," he said. "It's for you," said Jaro over his shoulder. Mr. Peet came to the door. "Hello, Stanley. I thought Hodes had you? Where's Miss Mikail?" "I got away. Look, Mr. Peet, I got to see you alone." Albert Peet said, "Would you excuse me, Mr. Moynahan?" He licked his lips. "I'll just step out into the hall a moment." He went out, drawing the door shut after him. Jaro lit a cigarette. He padded nervously back and forth across the room, his bare feet making no noise. He sat down on the edge of the bed. He got up and ground out the cigarette. He went to the door, but did not open it. Instead, he took another turn about the room. Again he came to a halt before the door, pressed his ear against the panel. For a long time he listened but could distinguish no murmur of voices. With an oath he threw open the door. The hall was empty. II Jaro returned to his room, stripped off his pajamas, climbed back into his suit. He tested the slug gun. It was a flat, ugly weapon which hurled a slug the size of a quarter. He preferred it because, though he seldom shot to kill, it stopped a man like a well placed mule's hoof. He adjusted the gun lightly in its holster in order that it wouldn't stick if he were called upon to use it in a hurry. Then he went out into the hall. At the desk he inquired if any messages had come for him. There were none, but the clerk had seen Mr. Peet with a young fellow take the incline to the underground. Above the clerk's head a newsograph was reeling off the current events almost as soon as they happened. Jaro read: " Earth Congress suspends negotiations on Mercurian freedom pending investigation of rumored rebellion. Terrestrials advised to return to Earth. Karfial Hodes, Mercurian patriot, being sought. " Jaro descended the incline to the network of burrows which served as streets during the flaming days. Here in the basements and sub-basements were located the shops and dram houses where the Mercurians sat around little tables drinking silently of the pale green Latonka. The burrows were but poorly lit, the natives preferring the cool gloom, and Jaro had to feel his way, rubbing shoulders with the strange, silent populace. But when he reached the Terrestrial quarter of the city, bright radoxide lights took the place of the green globes, and there was a sprinkling of Colonial guards among the throng. Jaro halted before a door bearing a placard which read: "LATONKA TRUST" He pushed through the door into a rich carpeted reception room. At the far end was a second door beside which sat a desk, door and desk being railed off from the rest of the office. The door into Albert Peet's inner sanctum was ajar. Jaro could distinguish voices; then quite clearly he heard Albert Peet say in a high girlish tone: "Stanley, I thought I left you in the native quarter. Why did you follow me? How many times have I told you never to come here?" The reply was unintelligible. Then the pale-faced young man came through the door shutting it after himself. At the sight of Jaro Moynahan he froze. "What're you sneaking around here for?" Jaro settled himself warily, his light blue eyes flicking over the youth. "Let's get this straight," he said mildly. "I've known your kind before. Frankly, ever since I saw you I've had to repress a desire to step on you as I might a spider." The youth's black eyes were hot as coals, his fingers twitching. His hands began to creep upward. "You dirty ..." he began, but he got no further. Jaro Moynahan shot him in the shoulder. The compressed air slug gun had seemed to leap into Jaro's hand. The big slug, smacked the gunman's shoulder with a resounding thwack, hurled him against the wall. Jaro vaulted the rail, deftly relieved him of two poisoned needle guns. "I'll get you for this," said Stanley, his mouth twisted in pain. "You've broken my shoulder. I'll kill you." The door to the inner sanctum swung open. "What's happened?" cried Albert Peet in distress. "What's wrong with you, Stanley?" "This dirty slob shot me in the shoulder." "But how badly?" Peet was wringing his hands. "Nothing serious," said Jaro. "He'll have his arm in a sling for a while. That's all." "Stanley," said Mr. Peet. "You're bleeding all over my carpet. Why can't you go in the washroom. There's a tile floor in there. If you hadn't disobeyed this wouldn't have happened. You and your fights. Has anyone called a doctor? Where's Miss Webb? Miss Webb! Oh, Miss Webb! That girl. Miss Webb!" Stanley climbed to his feet, swayed a moment drunkenly, then wobbled out a door on the left just as a tall brunette hurried in from the right. She had straight black hair which hung not quite to her shoulders, and dark brown eyes, and enough of everything else to absorb Jaro's attention. "Oh!" exclaimed Miss Webb as she caught sight of the blood staining the carpet. Joan Webb "There's been an—ah—accident," said Mr. Peet, and he licked his lips. "Call a doctor, Miss Webb." Miss Webb raised an eyebrow, went to the visoscreen. In a moment she had tuned in the prim starched figure of a nurse seated at a desk. "Could Dr. Baer rush right over here? There's been an accident." "Rush over where?" said the girl in the visoscreen. "These gadgets aren't telepathic, honey." "Oh," said Miss Webb, "the offices of the Latonka Trust." The girl in the visoscreen thawed like ice cream in the sun. "I'm sure Dr. Baer can come. He'll be there in a moment." "Thank you," said Miss Webb. She flicked the machine off, then added: "You trollop." Mr. Peet regarded Jaro Moynahan with distress. "Really, Mr. Moynahan, was it necessary to shoot Stanley? Isn't that—ah—a little extreme? I'm afraid it might incapacitate him, and I had a job for him." "Oh," cried Miss Webb, her brown eyes crackling. "Did you shoot that poor boy? Aren't you the big brave man?" "Poor boy?" said Jaro mildly. "Venomous little rattlesnake. I took these toys away from him." He held out the poisoned dart guns. "You take them, Mr. Peet. Frankly, they give me the creeps. They might go off. A scratch from one of those needles would be enough." Mr. Peet accepted the guns gingerly. He held them as if they might explode any minute. He started to put them in his pocket, thought better of it, glanced around helplessly. "Here, Miss Webb," he said, "do something with these. Put them in my desk." Miss Webb's eyes grew round as marbles. "I wouldn't touch one of those nasty little contraptions for all the Latonka on Mercury." "Here, I'll take them," said Stanley coming back into the room. He had staunched the flow of blood. His face was even whiter, if possible. Jaro eyed him coldly as with his good hand the youth dropped the dart guns back into their holsters. "Act like you want to use those and I'll put a slug in your head next time." "Now, Mr. Moynahan." Mr. Peet licked his lips nervously. "Stanley, go into my office. The doctor will be here in a moment. Miss Webb, you may go home. I'll have no more work for you today." Albert Peet led Stanley through the door. Jaro and Miss Webb were alone. With his eye on the door, Jaro said: "When you go out, turn left toward the native quarter. Wait for me in the first grog shop you come to." Miss Webb raised her eyebrows. "What's this? A new technique?" "Look," began Jaro annoyed. "My eyes are practically popping out of my head now," she interrupted. "Another morning like this and I take the first space liner back to Earth." She jammed her hat on backward, snatched her bag from the desk drawer. "I'm not trying to pick you up. This is...." "How disappointing." Jaro began again patiently. "Wait for me in the first grog shop. There's something I must know. It's important." He cleared his throat. "Don't you find the heat rather uncomfortable, Miss Webb. But perhaps you've become accustomed to it." Mr. Peet came back into the room. "Why, no, I mean yes," replied Miss Webb, a blank expression in her eyes. "Goodbye, Miss Webb," said Mr. Peet firmly. Jaro grinned and winked at her. Miss Webb tottered out of the room. As the door closed behind the girl, Albert Peet licked his lips, said: "Mr. Moynahan, I suppose my disappearance back at your room requires some explanation. But the fact is that Stanley brought an important bit of news." He paused. Jaro said nothing. "You might be interested to know that Miss Mikail is quite safe. Karfial Hodes has her, but Stanley assures me she will be quite safe." Again he paused. As Jaro remained silent, his neck mottled up pinkly. "The fact is, Mr. Moynahan, that we won't need you after all. I realize that we've put you to considerable trouble and we're prepared to pay you whatever you believe your time is worth. Say five hundred Earth notes?" "That's fair enough," replied Jaro. Albert Peet sighed. "I have the check made out." "Only," continued Jaro coldly, "I'm not ready to be bought off. I think I'll deal myself a hand in this game." Mr. Peet's face fell. "You won't reconsider?" "Sorry," said Jaro; "but I've got a date. I'm late now." He started to leave. "Stanley!" called Albert Peet. The pale-faced young man appeared in the doorway, the dart gun in his good hand. Jaro Moynahan dropped on his face, jerking out his slug gun as he fell. There was a tiny plop like a cap exploding. He heard the whisper of the poisoned dart as it passed overhead. Then he fired from the floor. The pale-faced young man crumpled like an empty sack. Jaro got up, keeping an eye on Albert Peet, brushed off his knees. "You've killed him," said Peet. "If I were you, Mr. Moynahan, I would be on the next liner back to Earth." Without answering, Jaro backed watchfully from the room. Once Jaro Moynahan had regained the street, he mopped his forehead with his handkerchief. Whatever was going on, these boys played for keeps. Warily he started down the passage toward the native quarter. At the first basement grog shop he turned in. His eyes swept the chamber, then he grinned. At a corner table, a tall glass of Latonka before her, sat Miss Webb. Her hat was still on backwards, and she was perched on the edge of her chair as if ready to spring up and away like a startled faun. " Bang! " said Jaro coming up behind her and poking a long brown finger in the small of her back. Miss Webb uttered a shriek, jerked so violently that her hat tilted over one eye. She regarded him balefully from beneath the brim. "Never a dull moment," she gritted. Still grinning, Jaro sat down. "I'm Jaro Moynahan, Miss Webb. I think Albert Peet forgot to introduce us. There's some skullduggery going on here that I'm particularly anxious to get to the bottom of. I thought you might be able to help me." "Yes," replied Miss Webb sweetly. A native waiter, attracted no doubt by her scream, came over and took Jaro's order. "All right," Jaro smiled, but his pale blue eyes probed the girl thoughtfully. "I'll have to confide certain facts which might be dangerous for you to know. Are you game, Miss Webb?" "Since we're going to be so chummy," she replied; "you might begin by calling me Joan. You make me feel downright ancient." "Well then," he said. "In the first place, I just killed that baby-faced gunman your boss had in his office." " Awk! " said Joan, choking on the Latonka. "It was self-defense," he hastened to assure her. "He took a pot shot at me with that poisoned dart gun." "But the police!" she cried, as she caught her breath. "There'll never be an investigation. Albert Peet will see to that. I was called here on what I supposed was a legitimate revolution. Instead I was offered ten thousand Earth notes to assassinate the leader of the revolution." "What revolution? I'm going around in circles." "The Mercurians, of course." "I don't believe it," said the girl. "The Mercurians are the most peaceable people in the Universe. They've been agitating for freedom, yes. But they believe in passive resistance. I don't believe you could induce a Mercurian to kill, even in self-protection. That's why Albert Peet and the rest of the combine had such an easy time gaining control of the Latonka trade." "Score one," breathed Jaro, "I begin to see light. Miss Webb—ah, Joan—I've a notion that we're going to be a great team. How do you happen to be Albert Peet's private secretary?" "A gal's gotta eat. But the truth is, I was quitting. The Latonka Trust is almost on the rocks. Their stock has been dropping like a meteor." Jaro Moynahan raised his oblique brows but did not interrupt. "Albert Peet," she continued, "has been trying to sell out but nobody will touch the stock, not since it looks as if the Earth Congress is going to grant the Mercurians their freedom. Everybody knows that the first thing the Mercurians will do, will be to boot out the Latonka Trust." "What about this Karfial Hodes?" said Jaro. "I've heard that he's inciting the Mercurians to rebellion. The newscaster had a line about the revolution too. The government has advised all Terrestrials to return to Earth." "It's not true," Joan flared. "It's all a pack of lies invented by the Latonka Trust. I know." "But I should think rumors like that would run down the Latonka stock."
|
B. Stanley tried to poison him.
|
What is the future of the Pandora?
A. It will stay on the planet forever.
B. It will return to Earth to report back on what they found.
C. It will rescue Hennessy’s crew and the exploring party.
D. It will remain in space.
|
Spawning Ground By LESTER DEL REY They weren't human. They were something more—and something less—they were, in short, humanity's hopes for survival! [Transcriber's Note: This etext was produced from Worlds of If Science Fiction, September 1961. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The Starship Pandora creaked and groaned as her landing pads settled unevenly in the mucky surface of the ugly world outside. She seemed to be restless to end her fool's errand here, two hundred light years from the waiting hordes on Earth. Straining metal plates twanged and echoed through her hallways. Captain Gwayne cursed and rolled over, reaching for his boots. He was a big, rawboned man, barely forty; but ten years of responsibility had pressed down his shoulders and put age-feigning hollows under his reddened eyes. The starlanes between Earth and her potential colonies were rough on the men who traveled them now. He shuffled toward the control room, grumbling at the heavy gravity. Lieutenant Jane Corey looked up, nodding a blonde head at him as he moved toward the ever-waiting pot of murky coffee. "Morning, Bob. You need a shave." "Yeah." He swallowed the hot coffee without tasting it, then ran a hand across the dark stubble on his chin. It could wait. "Anything new during the night?" "About a dozen blobs held something like a convention a little ways north of us. They broke up about an hour ago and streaked off into the clouds." The blobs were a peculiarity of this planet about which nobody knew anything. They looked like overgrown fireballs, but seemed to have an almost sentient curiosity about anything moving on the ground. "And our two cadets sneaked out again. Barker followed them, but lost them in the murk. I've kept a signal going to guide them back." Gwayne swore softly to himself. Earth couldn't turn out enough starmen in the schools, so promising kids were being shipped out for training as cadets on their twelfth birthday. The two he'd drawn, Kaufman and Pinelli, seemed to be totally devoid of any sense of caution. Of course there was no obvious need for caution here. The blobs hadn't seemed dangerous, and the local animals were apparently all herbivorous and harmless. They were ugly enough, looking like insects in spite of their internal skeletons, with anywhere from four to twelve legs each on their segmented bodies. None acted like dangerous beasts. But something had happened to the exploration party fifteen years back, and to the more recent ship under Hennessy that was sent to check up. He turned to the port to stare out at the planet. The Sol-type sun must be rising, since there was a dim light. But the thick clouds that wrapped the entire world diffused its rays into a haze. For a change, it wasn't raining, though the ground was covered by thick swirls of fog. In the distance, the tops of shrubs that made a scrub forest glowed yellow-green. Motions around them suggested a herd of feeding animals. Details were impossible to see through the haze. Even the deep gorge where they'd found Hennessy's carefully buried ship was completely hidden by the fog. There were three of the blobs dancing about over the grazing animals now, as they often seemed to do. Gwayne stared at them for a minute, trying to read sense into the things. If he had time to study them.... But there was no time. Earth had ordered him to detour here, after leaving his load of deep-sleep stored colonists on Official World 71, to check on any sign of Hennessy. He'd been here a week longer than he should have stayed already. If there was no sign in another day or so of what had happened to the men who'd deserted their ship and its equipment, he'd have to report back. He would have left before, if a recent landslip hadn't exposed enough of the buried ship for his metal locators to spot from the air by luck. It had obviously been hidden deep enough to foil the detectors originally. "Bob!" Jane Corey's voice cut through his pondering. "Bob, there are the kids!" Before he could swing to follow her pointing finger, movement caught his eye. The blobs had left the herd. Now the three were streaking at fantastic speed to a spot near the ship, to hover excitedly above something that moved there. He saw the two cadets then, heading back to the waiting ship, just beyond the movement he'd seen through the mist. Whatever was making the fog swirl must have reached higher ground. Something began to heave upwards. It was too far to see clearly, but Gwayne grabbed the microphone, yelling into the radio toward the cadets. They must have seen whatever it was just as the call reached them. Young Kaufman grabbed at Pinelli, and they swung around together. Then the mists cleared. Under the dancing blobs, a horde of things was heading for the cadets. Shaggy heads, brute bodies vaguely man-like! One seemed to be almost eight feet tall, leading the others directly toward the spacesuited cadets. Some of the horde were carrying spears or sticks. There was a momentary halt, and then the leader lifted one arm, as if motioning the others forward. "Get the jeeps out!" Gwayne yelled at Jane. He yanked the door of the little officers' lift open and jabbed the down button. It was agonizingly slow, but faster than climbing down. He ripped the door back at the exit deck. Men were dashing in, stumbling around in confusion. But someone was taking over now—one of the crew women. The jeeps were lining up. One, at the front, was stuttering into life, and Gwayne dashed for it as the exit port slid back. There was no time for suits or helmets. The air on the planet was irritating and vile smelling, but it could be breathed. He leaped to the seat, to see that the driver was Doctor Barker. At a gesture, the jeep rolled down the ramp, grinding its gears into second as it picked up speed. The other two followed. There was no sign of the cadets at first. Then Gwayne spotted them; surrounded by the menacing horde. Seen from here, the things looked horrible in a travesty of manhood. The huge leader suddenly waved and pointed toward the jeeps that were racing toward him. He made a fantastic leap backwards. Others swung about, two of them grabbing up the cadets. The jeep was doing twenty miles an hour now, but the horde began to increase the distance, in spite of the load of the two struggling boys! The creatures dived downward into lower ground, beginning to disappear into the mists. "Follow the blobs," Gwayne yelled. He realized now he'd been a fool to leave his suit; the radio would have let him keep in contact with the kids. But it was too late to go back. The blobs danced after the horde. Barker bounced the jeep downward into a gorge. Somewhere the man had learned to drive superlatively; but he had to slow as the fog thickened lower down. Then it cleared to show the mob of creatures doubling back on their own trail to confuse the pursuers. There was no time to stop. The jeep plowed through them. Gwayne had a glimpse of five-foot bodies tumbling out of the way. Monstrously coarse faces were half hidden by thick hair. A spear crunched against the windshield from behind, and Gwayne caught it before it could foul the steering wheel. It had a wickedly beautiful point of stone. The creatures vanished as Barker fought to turn to follow them. The other jeeps were coming up, by the sound of their motors, but too late to help. They'd have to get to the group with the cadets in a hurry or the horde would all vanish in the uneven ground, hidden by the fog. A blob dropped down, almost touching Gwayne. He threw up an instinctive hand. There was a tingling as the creature seemed to pass around it. It lifted a few inches and drifted off. Abruptly, Barker's foot ground at the brake. Gwayne jolted forward against the windshield, just as he made out the form of the eight-foot leader. The thing was standing directly ahead of him, a cadet on each shoulder. The wheels locked and the jeep slid protestingly forward. The creature leaped back. But Gwayne was out of the jeep before it stopped, diving for the figure. It dropped the boys with a surprised grunt. The arms were thin and grotesque below the massively distorted shoulders, but amazingly strong. Gwayne felt them wrench at him as his hands locked on the thick throat. A stench of alien flesh was in his nose as the thing fell backwards. Doc Barker had hit it seconds after the captain's attack. Its head hit rocky ground with a dull, heavy sound, and it collapsed. Gwayne eased back slowly, but it made no further move, though it was still breathing. Another jeep had drawn up, and men were examining the cadets. Pinelli was either laughing or crying, and Kaufman was trying to break free to kick at the monster. But neither had been harmed. The two were loaded onto a jeep while men helped Barker and Gwayne stow the bound monster on another before heading back. "No sign of skull fracture. My God, what a tough brute!" Barker shook his own head, as if feeling the shock of the monster's landing. "I hope so," Gwayne told him. "I want that thing to live—and you're detailed to save it and revive it. Find out if it can make sign language or draw pictures. I want to know what happened to Hennessy and why that ship was buried against detection. This thing may be the answer." Barker nodded grimly. "I'll try, though I can't risk drugs on an alien metabolism." He sucked in on the cigarette he'd dug out, then spat sickly. Smoke and this air made a foul combination. "Bob, it still makes no sense. We've scoured this planet by infra-red, and there was no sign of native villages or culture. We should have found some." "Troglodytes, maybe," Gwayne guessed. "Anyhow, send for me when you get anything. I've got to get this ship back to Earth. We're overstaying our time here already." The reports from the cadets were satisfactory enough. They'd been picked up and carried, but no harm had been done them. Now they were busy being little heroes. Gwayne sentenced them to quarters as soon as he could, knowing their stories would only get wilder and less informative with retelling. If they could get any story from the captured creature, they might save time and be better off than trying to dig through Hennessy's ship. That was almost certainly spoorless by now. The only possible answer seemed to be that the exploring expedition and Hennessy's rescue group had been overcome by the aliens. It was an answer, but it left a lot of questions. How could the primitives have gotten to the men inside Hennessy's ship? Why was its fuel dumped? Only men would have known how to do that. And who told these creatures that a space ship's metal finders could be fooled by a little more than a hundred feet of solid rock? They'd buried the ship cunningly, and only the accidental slippage had undone their work. Maybe there would never be a full answer, but he had to find something—and find it fast. Earth needed every world she could make remotely habitable, or mankind was probably doomed to extinction. The race had blundered safely through its discovery of atomic weapons into a peace that had lasted two hundred years. It had managed to prevent an interplanetary war with the Venus colonists. It had found a drive that led to the stars, and hadn't even found intelligent life there to be dangerous on the few worlds that had cultures of their own. But forty years ago, observations from beyond the Solar System had finally proved that the sun was going to go nova. It wouldn't be much of an explosion, as such things go—but it would render the whole Solar System uninhabitable for millenia. To survive, man had to colonize. And there were no worlds perfect for him, as Earth had been. The explorers went out in desperation to find what they could; the terraforming teams did what they could. And then the big starships began filling worlds with colonists, carried in deep sleep to conserve space. Almost eighty worlds. The nearest a four month journey from Earth and four more months back. In another ten years, the sun would explode, leaving man only on the footholds he was trying to dig among other solar systems. Maybe some of the strange worlds would let men spread his seed again. Maybe none would be spawning grounds for mankind in spite of the efforts. Each was precious as a haven for the race. If this world could be used, it would be nearer than most. If not, as it now seemed, no more time could be wasted here. Primitives could be overcome, maybe. It would be ruthless and unfair to strip them of their world, but the first law was survival. But how could primitives do what these must have done? He studied the spear he had salvaged. It was on a staff made of cemented bits of smaller wood from the scrub growth, skillfully laminated. The point was of delicately chipped flint, done as no human hand had been able to do for centuries. "Beautiful primitive work," he muttered. Jane pulled the coffee cup away from her lips and snorted. "You can see a lot more of it out there," she suggested. He went to the port and glanced out. About sixty of the things were squatting in the clearing fog, holding lances and staring at the ship. They were perhaps a thousand yards away, waiting patiently. For what? For the return of their leader—or for something that would give the ship to them? Gwayne grabbed the phone and called Barker. "How's the captive coming?" Barker's voice sounded odd. "Physically fine. You can see him. But—" Gwayne dropped the phone and headed for the little sick bay. He swore at Doc for not calling him at once, and then at himself for not checking up sooner. Then he stopped at the sound of voices. There was the end of a question from Barker and a thick, harsh growling sound that lifted the hair along the nape of Gwayne's neck. Barker seemed to understand, and was making a comment as the captain dashed in. The captive was sitting on the bunk, unbound and oddly unmenacing. The thick features were relaxed and yet somehow intent. He seemed to make some kind of a salute as he saw Gwayne enter, and his eyes burned up unerringly toward the device on the officer's cap. "Haarroo, Cabbaan!" the thing said. "Captain Gwayne, may I present your former friend, Captain Hennessy?" Barker said. There was a grin on the doctor's lips, but his face was taut with strain. The creature nodded slowly and drew something from the thick hair on its head. It was the golden comet of a captain. "He never meant to hurt the kids—just to talk to them," Barker cut in quickly. "I've got some of the story. He's changed. He can't talk very well. Says they've had to change the language around to make the sounds fit, and he's forgotten how to use what normal English he can. But it gets easier as you listen. It's Hennessy, all right. I'm certain." Gwayne had his own ideas on that. It was easy for an alien to seize on the gold ornament of a captive earthman, even to learn a little English, maybe. But Hennessy had been his friend. "How many barmaids in the Cheshire Cat? How many pups did your oldest kid's dog have? How many were brown?" The lips contorted into something vaguely like a smile, and the curiously shaped fingers that could handle no human-designed equipment spread out. Three. Seven. Zero. The answers were right. By the time the session was over, Gwayne had begun to understand the twisted speech from inhuman vocal cords better. But the story took a long time telling. When it was finished, Gwayne and Barker sat for long minutes in silence. Finally Gwayne drew a shuddering breath and stood up. "Is it possible, Doc?" "No," Barker said flatly. He spread his hands and grimaced. "No. Not by what I know. But it happened. I've looked at a few tissues under the microscope. The changes are there. It's hard to believe about their kids. Adults in eight years, but they stay shorter. It can't be a hereditary change—the things that affect the body don't change the germ plasm. But in this case, what changed Hennessy is real, so maybe the fact that the change is passed on is as real as he claims." Gwayne led the former Hennessy to the exit. The waiting blobs dropped down to touch the monstrous man, then leaped up again. The crowd of monsters began moving forward toward their leader. A few were almost as tall as Hennessy, but most were not more than five feet high. The kids of the exploring party.... Back in the control room, Gwayne found the emergency release levers, set the combinations and pressed the studs. There was a hiss and gurgle as the great tanks of fuel discharged their contents out onto the ground where no ingenuity could ever recover it to bring life to the ship again. He'd have to tell the men and women of the crew later, after he'd had time to organize things and present it all in a way they could accept, however much they might hate it at first. But there was no putting off giving the gist of it to Jane. "It was the blobs," he summarized it. "They seem to be amused by men. They don't require anything from us, but they like us around. Hennessy doesn't know why. They can change our cells, adapt us. Before men came, all life here had twelve legs. Now they're changing that, as we've seen. "And they don't have to be close to do it. We've all been outside the hull. It doesn't show yet—but we're changed. In another month, Earth food would kill us. We've got to stay here. We'll bury the ships deeper this time, and Earth won't find us. They can't risk trying a colony where three ships vanish, so we'll just disappear. And they'll never know." Nobody would know. Their children—odd children who matured in eight years—would be primitive savages in three generations. The Earth tools would be useless, impossible for the hands so radically changed. Nothing from the ship would last. Books could never be read by the new eyes. And in time, Earth wouldn't even be a memory to this world. She was silent a long time, staring out of the port toward what must now be her home. Then she sighed. "You'll need practice, but the others don't know you as well as I do, Bob. I guess we can fix it so they'll believe it all. And it's too late now. But we haven't really been changed yet, have we?" "No," he admitted. Damn his voice! He'd never been good at lying. "No. They have to touch us. I've been touched, but the rest could go back." She nodded. He waited for the condemnation, but there was only puzzlement in her face. "Why?" And then, before he could answer, her own intelligence gave her the same answer he had found for himself. "The spawning ground!" It was the only thing they could do. Earth needed a place to plant her seed, but no world other than Earth could ever be trusted to preserve that seed for generation after generation. Some worlds already were becoming uncertain. Here, though, the blobs had adapted men to the alien world instead of men having to adapt the whole planet to their needs. Here, the strange children of man's race could grow, develop and begin the long trek back to civilization. The gadgets would be lost for a time. But perhaps some of the attitudes of civilized man would remain to make the next rise to culture a better one. "We're needed here," he told her, his voice pleading for the understanding he couldn't yet fully give himself. "These people need as rich a set of bloodlines as possible to give the new race strength. The fifty men and women on this ship will be needed to start them with a decent chance. We can't go to Earth, where nobody would believe or accept the idea—or even let us come back. We have to stay here." She smiled then and moved toward him, groping for his strength. "Be fruitful," she whispered. "Be fruitful and spawn and replenish an earth." "No," he told her. "Replenish the stars." But she was no longer listening, and that part of his idea could wait. Some day, though, their children would find a way to the starlanes again, looking for other worlds. With the blobs to help them, they could adapt to most worlds. The unchanged spirit would lead them through all space, and the changing bodies would claim worlds beyond numbering. Some day, the whole universe would be a spawning ground for the children of men!
|
A. It will stay on the planet forever.
|
How does the author feel about Carolyn?
A. Carolyn is confident, composed, and in control.
B. Carolyn is a primal force.
C. Carolyn tries very hard to appear perfect and in control. It's hard to feel animosity toward her.
D. Carolyn is shrill.
|
A Good Year for the Roses? Early in American Beauty , Lester Burnham (Kevin Spacey), a weary reporter for a media magazine, masturbates in the shower while informing us in voice-over that we're witnessing the highlight of his day. He peers through tired eyes out the window at his manicured suburban tract-house lawn, where his wife, Carolyn (Annette Bening)--whose gardening clogs, he points out, are color-coordinated with the handles of her shears--snips roses (American beauties) and twitters about Miracle-Gro to a gay yuppie (Scott Bakula) on the other side of a white picket fence. "I have lost something," says Lester. "I'm not exactly sure what it is but I know I didn't always feel this ... sedated." Apparently, Lester doesn't realize that snipped roses are garden-variety symbols of castration, or he'd know what he has lost. But the makers of American Beauty are about to give Lester his roses back. At a high-school basketball game, Lester is transfixed by a blonde cheerleader named Angela (Mena Suvari), who is twirling alongside his daughter, Jane (Thora Burch). Ambient noise falls away, the crowd disappears, and there she is, Lester's angel, writhing in slow motion--just for him. She opens her jacket (she's naked underneath) and red rose petals drift out. Later, Lester envisions her on a bed of red petals, then immersed in a bath of red petals. Back in the roses for the first time in years, he's soon pumping iron, smoking pot, and telling off his frigid wife and faceless bosses, convinced that whatever he has lost he's getting back, baby. The movie is convinced, too--which is odd, since the fantasy of an underage cheerleader making a middle-aged man's wilted roses bloom is a tad ... primitive. But American Beauty doesn't feel primitive. It feels lustrously hip and aware, and a lot of critics are making big claims for it. The script, by Alan Ball, a playwright and former sitcom writer, carries an invigorating blast of counterculture righteousness, along with the kind of pithily vicious marital bickering that makes some viewers (especially male) say, "Yeah! Tell that bitch off!" More important, it has a vein of metaphysical yearning, which the director, Sam Mendes, mines brilliantly. A hotshot English theater director (his Cabaret revival is still on the boards in New York), Mendes gives the film a patina of New Age lyricism and layer upon layer of visual irony. The movie's surface is velvety and immaculate--until the action is abruptly viewed through the video camera of the teen-age voyeur next door (Wes Bentley), and the graininess of the video image (along with the plangent music) suggests how unstable the molecules that constitute our "reality" really are. Mendes can distend the real into the surreal with imperceptible puffs. Aided by his cinematographer, Conrad Hall, and editors, Tariq Anwar and Chris Greenbury, he creates an entrancing vision of the American nuclear family on the verge of a meltdown. A merican Beauty is so wittily written and gorgeously directed that you might think you're seeing something archetypal--maybe even the Great American Movie. But when you stop and smell the roses ... Well, that scent isn't Miracle-Gro. The hairpin turns from farce to melodrama, from satire to bathos, are fresh and deftly navigated, but almost every one of the underlying attitudes is smug and easy: from the corporate flunky named "Brad" to the interchangeable gay neighbors (they're both called "Jim") to the brutally homophobic patriarch next door, an ex-Marine colonel (Chris Cooper) who has reduced his wife (the normally exuberant Allison Janney) to a catatonic mummy and his son, Ricky (Bentley), to a life of subterranean deception. (The colonel's idea of bliss is watching an old Ronald Reagan military picture on television: How's that for subtle?) Lester's wife, Carolyn, is even more stridently caricatured. A real-estate broker who fails to sell a big house (her only potential customers are blank-faced African-Americans, Indian-Americans, and surly lesbians), she wears a mask of perky efficiency and insists on listening to Muzak while she and her husband and daughter eat her "nutritious yet savory" dinners. It's amazing that Mendes and Ball get away with recycling so many stale and reactionary ideas under the all-purpose rubric of "black comedy." But it's also possible that those ideas have rarely been presented so seductively. Several months ago, Daniel Menaker in Slate in contemporary film in which the protagonist attempts to break through our cultural and technological anesthetization into "the real." That's the theme here, too, and it's extraordinarily potent, at times even heartbreaking. The symbols, however, have been cunningly reversed. In movies like sex, lies, and videotape (1989), the protagonist has to put away the video camera to "get real"; in American Beauty , it's Ricky Fitts, the damaged stoner videomaker next door, who sees beauty where nonartists see only horror or nothingness. In the film's most self-consciously poetic set piece, Ricky shows Lester's dour daughter Jane--in whom he recognizes a kindred spirit--a video of a plastic bag fluttering up, down, and around on invisible currents of wind. Ricky speaks of glimpsing in the bag's trajectory an "entire life behind things"--a "benevolent force" that holds the universe together. The teen-ager, who likes to train his lenses on dead bodies of animals and people, sells wildly expensive marijuana to Lester and somehow passes on this notion of "beauty." By the end, Lester is mouthing the same sentiments and has acquired the same deadpan radiance. That must be some really good shit they're smoking. It's not the druggy philosophizing, however, that makes American Beauty an emotional workout. It's that the caricatures are grounded in sympathy instead of derision. Everyone on screen is in serious pain. The manipulative sexpot Angela, who taunts her friend Jane with the idea of seducing her dad, acts chiefly out of a terror of appearing ordinary. As the military martinet, Cooper goes against the grain, turning Col. Fitts into a sour bulldog whose capaciously baggy eyes are moist with sadness over his inability to reach out. (When he stands helplessly in the rain at the end, the deluge completes him.) The character of Carolyn is so shrill as to constitute a libel on the female sex, but there isn't a second when Bening sends the woman up. She doesn't transcend the part, she fills it to the brim, anatomizes it. You can't hate Carolyn because the woman is trying so hard--to appear confident, composed, in control. When she fails to sell that house, she closes the shades and lets go with a naked wail--it's the sound of a vacuum crying to be filled--then furiously slaps herself while sputtering, "Shut up--you're weak--shut up. " Then she breathes, regains her go-get-'em poise, replaces her mask. Carolyn isn't a complicated dramatic construction, but Bening gives her a primal force. An actress who packs more psychological detail into a single gesture than others get into whole scenes, Bening was barreling down the road to greatness before she hit a speed bump called Warren. It's a joy to observe her--both here and in Neil Jordan's In Dreams (1999)--back at full throttle. American Beauty is Spacey's movie, though. He gives it--how weird to write this about Spacey, who made his name playing flamboyantly self-involved psychopaths--a heart. Early on, he lets his face and posture go slack and his eyes blurry. He mugs like crazy, telegraphing Lester's "loserness." But Spacey's genius is for mugging in character. He makes us believe that it's Lester who's caricaturing himself , and that bitter edge paves the way for the character's later, more comfortably Spacey-like scenes of insult and mockery. He even makes us take Lester's final, improbably rhapsodic moments straight. But do the filmmakers take them straight? If I read it correctly, the movie is saying that American society is unjust and absurd and loveless--full of people so afraid of seeming ordinary that they lose their capacity to see. It's saying that our only hope is to cultivate a kind of stoned aesthetic detachment whereby even a man with his brains blown out becomes an object of beauty and a signpost to a Higher Power. But to scrutinize a freshly dead body and not ask how it got that way--or if there's anyone nearby with a gun who might want to add to the body count--strikes me as either moronic or insane or both. The kind of detachment the movie is peddling isn't artistic, it isn't life--it's nihilism at its most fatuous. In the end, American Beauty is New Age Nihilism. Kevin Costner is 11 years older than he was as Crash Davis, the over-the-hill minor-league catcher in Bull Durham (1988), but he can still get away with playing a professional ballplayer. He moves and acts like a celebrity jock, and he can make his narcissistic self-containment look as if he's keeping something in reserve--to protect his "instrument," as it were. In For Love of the Game , he's a 40ish Detroit Tigers pitcher having his last hurrah: The team has been sold and the new owners don't necessarily want him back. For about half an hour, it's a great sports movie. Costner stands on the mound shaking off the signals of his longtime catcher (John C. Reilly); he forces himself to tune out the huge Yankee Stadium crowd (the background blurs before our eyes and the sound drops out); and he mutters darkly at a succession of batters, some old nemeses, some old buddies. He also thinks about his Manhattan-based ex-girlfriend (Kelly Preston), who tearfully told him that morning that things were absolutely over and she was moving to London. There's an appealing flashback to how they met (he stopped to fix her car while on the way to Yankee Stadium), then it's back to the game for more nail-biting at bats. But pretty soon the relationship flashbacks start coming thick and fast, and the balance of the movie shifts to whether Kevin can commit to Kelly and Kelly can commit to Kevin or whether his only commitment could ever be to the ball and the diamond and the game. Maybe it's because I'm a baseball nut that I hated to leave the mound. But maybe it's also because the relationships scenes are soft-focus, generic, and woozily drawn-out, whereas the stuff in the stadium is sharply edited and full of texture. The rhythms of the game feel right; the rhythms of the romance feel embarrassingly Harlequin, and the picture drags on for over two hours. I can't believe that the director, Sam Raimi ( The Evil Dead , 1983; last year's A Simple Plan ) thought that all those scenes of Costner and Preston staring into space while the piano plinks would end up in the final cut, but Raimi apparently gave up control of the final cut for the sake of making his first, real mainstream picture. He might as well have stuck his head over the plate and said, "Bean me."
|
C. Carolyn tries very hard to appear perfect and in control. It's hard to feel animosity toward her.
|
How big are this dataset and catalogue?
|
### Introduction
Abusive online content, such as hate speech and harassment, has received substantial attention over the past few years for its malign social effects. Left unchallenged, abusive content risks harminng those who are targeted, toxifying public discourse, exacerbating social tensions and could lead to the exclusion of some groups from public spaces. As such, systems which can accurately detect and classify online abuse at scale, in real-time and without bias are of central interest to tech companies, policymakers and academics. Most detection systems rely on having the right training dataset, reflecting one of the most widely accepted mantras in computer science: Garbage In, Garbage Out. Put simply: to have systems which can detect and classify abusive online content effectively, one needs appropriate datasets with which to train them. However, creating training datasets is often a laborious and non-trivial task – and creating datasets which are non-biased, large and theoretically-informed is even more difficult (BIBREF0 p. 189). We address this issue by examining and reviewing publicly available datasets for abusive content detection, which we provide access to on a new dedicated website, hatespeechdata.com. In the first section, we examine previous reviews and present the four research aims which guide this paper. In the second section, we conduct a critical and in-depth analysis of the available datasets, discussing first what their aim is, how tasks have been described and what taxonomies have been constructed and then, second, what they contain and how they were annotated. In the third section, we discuss the challenges of open science in this research area and elaborates different ways of sharing training datasets, including the website hatespeechdata.com In the final section, we draw on our findings to establish best practices when creating datasets for abusive content detection. ### Background
The volume of research examining the social and computational aspects of abusive content detection has expanded prodigiously in the past five years. This has been driven by growing awareness of the importance of the Internet more broadly BIBREF1, greater recognition of the harms caused by online abuse BIBREF2, and policy and regulatory developments, such as the EU's Code of Conduct on Hate, the UK Government's `Online Harms' white paper BIBREF3, Germany's NetzDG laws, the Public Pledge on Self-Discipline for the Chinese Internet Industry, and France's anti-hate regulation BIBREF2. In 2020 alone, three computer science venues will host workshops on online hate (TRAC and STOC at LREC, and WOAH at EMNLP), and a shared task at 2019's SemEval on online abuse detection reports that 800 teams downloaded the training data and 115 submitted detection systems BIBREF4. At the same time, social scientific interventions have also appeared, deepening our understanding of how online abuse spreads BIBREF5 and how its harmful impact can be mitigated and challenged BIBREF6. All analyses of online abuse ultimately rely on a way of measuring it, which increasingly means having a method which can handle the sheer volume of content produced, shared and engaged with online. Traditional qualitative methods cannot scale to handle the hundreds of millions of posts which appear on each major social media platform every day, and can also introduce inconsistencies and biase BIBREF7. Computational tools have emerged as the most promising way of classifying and detecting online abuse, drawing on work in machine learning, Natural Language Processing (NLP) and statistical modelling. Increasingly sophisticated architectures, features and processes have been used to detect and classify online abuse, leveraging technically sophisticated methods, such as contextual word embeddings, graph embeddings and dependency parsing. Despite their many differences BIBREF8, nearly all methods of online abuse detection rely on a training dataset, which is used to teach the system what is and is not abuse. However, there is a lacuna of research on this crucial aspect of the machine learning process. Indeed, although several general reviews of the field have been conducted, no previous research has reviewed training datasets for abusive content detection in sufficient breadth or depth. This is surprising given (i) their fundamental importance in the detection of online abuse and (ii) growing awareness that several existing datasets suffer from many flaws BIBREF9, BIBREF10. Close relevant work includes: Schmidt and Wiegand conduct a comprehensive review of research into the detection and classification of abusive online content. They discuss training datasets, stating that `to perform experiments on hate speech detection, access to labelled corpora is essential' (BIBREF8, p. 7), and briefly discuss the sources and size of the most prominent existing training datasets, as well as how datasets are sampled and annotated. Schmidt and Wiegand identify two key challenges with existing datasets. First, `data sparsity': many training datasets are small and lack linguistic variety. Second, metadata (such as how data was sampled) is crucial as it lets future researchers understand unintended biases, but is often not adequately reported (BIBREF8, p. 6). Waseem et al.BIBREF11 outline a typology of detection tasks, based on a two-by-two matrix of (i) identity- versus person- directed abuse and (ii) explicit versus implicit abuse. They emphasise the importance of high-quality datasets, particularly for more nuanced expressions of abuse: `Without high quality labelled data to learn these representations, it may be difficult for researchers to come up with models of syntactic structure that can help to identify implicit abuse.' (BIBREF11, p. 81) Jurgens et al.BIBREF12 discuss also conduct a critical review of hate speech detection, and note that `labelled ground truth data for building and evaluating classifiers is hard to obtain because platforms typically do not share moderated content due to privacy, ethical and public relations concerns.' (BIBREF12, p. 3661) They argue that the field needs to `address the data scarcity faced by abuse detection research' in order to better address more complex rsearch issues and pressing social challenges, such as `develop[ing] proactive technologies that counter or inhibit abuse before it harms' (BIBREF12, pp. 3658, 3661). Vidgen et al. describe several limitations with existing training datasets for abusive content, most noticeably how `they contain systematic biases towards certain types and targets of abuse.' BIBREF13[p.2]. They describe three issues in the quality of datasets: degradation (whereby datasets decline in quality over time), annotation (whereby annotators often have low agreement, indicating considerable uncertainty in class assignments) and variety (whereby `The quality, size and class balance of datasets varies considerably.' [p. 6]). Chetty and AlathurBIBREF14 review the use of Internet-based technologies and online social networks to study the spread of hateful, offensive and extremist content BIBREF14. Their review covers both computational and legal/social scientific aspects of hate speech detection, and outlines the importance of distinguishing between different types of group-directed prejudice. However, they do not consider training datasets in any depth. Fortuna and NunesBIBREF15 provide an end-to-end review of hate speech research, including the motivations for studying online hate, definitional challenges, dataset creation/sharing, and technical advances, both in terms of feature selection and algorithmic architecture (BIBREF15, 2018). They delineate between different types of online abuse, including hate, cyberbullying, discrimination and flaming, and add much needed clarity to the field. They show that (1) dataset size varies considerably but they are generally small (mostly containing fewer than 10,000 entries), (2) Twitter is the most widely-studied platform, and (3) most papers research hate speech per se (i.e. without specifying a target). Of those which do specify a target, racism and sexism are the most researched. However, their review focuses on publications rather than datasets: the same dataset might be used in multiple studies, limiting the relevance of their review for understanding the intrinsic role of training datasets. They also only engage with datasets fairly briefly, as part of a much broader review. Several classification papers also discuss the most widely used datasets, including Davidson et al. BIBREF16 who describe five datasets, and Salminen et al. who review 17 datasets and describe four in detail BIBREF17. This paper addresses this lacuna in existing research, providing a systematic review of available training datasets for online abuse. To provide structure to this review, we adopt the `data statements' framework put forward by Bender and Friedman BIBREF18, as well as other work providing frameworks, schema and processes for analysing NLP artefacts BIBREF19, BIBREF20, BIBREF21. Data statements are a way of documenting the decisions which underpin the creation of datasets used for Natural Language Processing (NLP). They formalise how decisions should be documented, not only ensuring scientific integrity but also addressing `the open and urgent question of how we integrate ethical considerations in the everyday practice of our field' (BIBREF18, p. 587). In many cases, we find that it is not possible to fully recreate the level of detail recorded in an original data statement from how datasets are described in publications. This reinforces the importance of proper documentation at the point of dataset creation. As the field of online abusive content detection matures, it has started to tackle more complex research challenges, such as multi-platform, multi-lingual and multi-target abuse detection, and systems are increasingly being deployed in `the wild' for social scientific analyses and for content moderation BIBREF5. Such research heightens the focus on training datasets as exactly what is being detected comes under greater scrutiny. To enhance our understanding of this domain, our review paper has four research aims. Research Aim One: to provide an in-depth and critical analysis of the available training datasets for abusive online content detection. Research Aim Two: to map and discuss ways of addressing the lack of dataset sharing, and as such the lack of `open science', in the field of online abuse research. Research Aim Three: to introduce the website hatespeechdata.com, as a way of enabling more dataset sharing. Research Aim Four: to identify best practices for creating an abusive content training dataset. ### Analysis of training datasets
Relevant publications have been identified from four sources to identify training datasets for abusive content detection: The Scopus database of academic publications, identified using keyword searches. The ACL Anthology database of NLP research papers, identified using keyword searches. The ArXiv database of preprints, identified using keyword searches. Proceedings of the 1st, 2nd and 3rd workshops on abusive language online (ACL). Most publications report on the creation of one abusive content training dataset. However, some describe several new datasets simultaneously or provide one dataset with several distinct subsets of data BIBREF22, BIBREF23, BIBREF24, BIBREF25. For consistency, we separate out each subset of data where they are in different languages or the data is collected from different platforms. As such, the number of datasets is greater than the number publications. All of the datasets were released between 2016 and 2019, as shown in Figure FIGREF17. ### Analysis of training datasets ::: The purpose of training datasets ::: Problems addressed by datasets
Creating a training dataset for online abuse detection is typically motivated by the desire to address a particular social problem. These motivations can inform how a taxonomy of abusive language is designed, how data is collected and what instructions are given to annotators. We identify the following motivating reasons, which were explicitly referenced by dataset creators. Reducing harm: Aggressive, derogatory and demeaning online interactions can inflict harm on individuals who are targeted by such content and those who are not targeted but still observe it. This has been shown to have profound long-term consequences on individuals' well-being, with some vulnerable individuals expressing concerns about leaving their homes following experiences of abuse BIBREF26. Accordingly, many dataset creators state that aggressive language and online harassment is a social problem which they want to help address Removing illegal content: Many countries legislate against certain forms of speech, e.g. direct threats of violence. For instance, the EU's Code of Conduct requires that all content that is flagged for being illegal online hate speech is reviewed within 24 hours, and removed if necessary BIBREF27. Many large social media platforms and tech companies adhere to this code of conduct (including Facebook, Google and Twitter) and, as of September 2019, 89% of such content is reviewed in 24 hours BIBREF28. However, we note that in most cases the abuse that is marked up in training datasets falls short of the requirements of illegal online hate – indeed, as most datasets are taken from public API access points, the data has usually already been moderated by the platforms and most illegal content removed. Improving health of online conversations: The health of online communities can be severely affected by abusive language. It can fracture communities, exacerbate tensions and even repel users. This is not only bad for the community and for civic discourse in general, it also negatively impacts engagement and thus the revenue of the host platforms. Therefore, there is a growing impetus to improve user experience and ensure online dialogues are healthy, inclusive and respectful where possible. There is ample scope for improvement: a study showed that 82% of personal attacks on Wikipedia against other editors are not addressed BIBREF29. Taking steps to improve the health of exchanges in online communities will also benefit commercial and voluntary content moderators. They are routinely exposed to such content, often with insufficient safeugards, and sometimes display symptoms similar to those of PTSD BIBREF30. Automatic tools could help to lessen this exposure, reducing the burden on moderators. ### Analysis of training datasets ::: The purpose of training datasets ::: Uses of datasets: How detection tasks are defined
Myriad tasks have been addressed in the field of abusive online content detection, reflecting the different disciplines, motivations and assumptions behind research. This has led to considerable variation in what is actually detected under the rubric of `abusive content', and establishing a degree of order over the diverse categorisations and subcategorisations is both difficult and somewhat arbitrary. Key dimensions which dataset creators have used to categorise detection tasks include who/what is targeted (e.g. groups vs. individuals), the strength of content (e.g. covert vs. overt), the nature of the abuse (e.g. benevolent vs. hostile sexism BIBREF31), how the abuse manifests (e.g. threats vs. derogatory statements), the tone (e.g. aggressive vs. non-aggressive), the specific target (e.g. ethnic minorities vs. women),and the subjective perception of the reader (e.g. disrespectful vs. respectful). Other important dimensions include the theme used to express abuse (e.g. Islamophobia which relies on tropes about terrorism vs. tropes about sexism) and the use of particular linguistic devices, such as appeals to authority, sincerity and irony. All of these dimensions can be combined in different ways, producing a large number of intersecting tasks. Consistency in how tasks are described will not necessarily ensure that datasets can be used interchangeably. From the description of a task, an annotation framework must be developed which converts the conceptualisation of abuse into a set of standards. This formalised representation of the `abuse' inevitably involves shortcuts, imperfect rules and simplifications. If annotation frameworks are developed and applied differently, then even datasets aimed at the same task can still vary considerably. Nonetheless, how detection tasks for online abuse are described is crucial for how the datasets – and in turn the systems trained on them – can subsequently be used. For example, a dataset annotated for hate speech can be used to examine bigoted biases, but the reverse is not true. How datasets are framed also impacts whether, and how, datasets can be combined to form large `mega-datasets' – a potentially promising avenue for overcoming data sparsity BIBREF17. In the remainder of this section, we provide a framework for splitting out detection tasks along the two most salient dimensions: (1) the nature of abuse and (2) the granularity of the taxonomy. ### Analysis of training datasets ::: The purpose of training datasets ::: Uses of datasets: How detection tasks are defined ::: Detection tasks: the nature of abuse
This refers to what is targeted/attacked by the content and, subsequently, how the taxonomy has been designed/framed by the dataset creators. The most well-established taxonomic distinction in this regard is the difference between (i) the detection of interpersonal abuse, and (ii) the detection of group-directed abuse BIBREF11). Other authors have sought to deductively theorise additional categories, such as `concept-directed' abuse, although these have not been widely adopted BIBREF13. Through an inductive investigation of existing training datasets, we extend this binary distinction to four primary categories of abuse which have been studied in previous work, as well as a fifth `Mixed' category. Person-directed abuse. Content which directs negativity against individuals, typically through aggression, insults, intimidation, hostility and trolling, amongst other tactics. Most research falls under the auspices of `cyber bullying', `harassment' and `trolling' BIBREF23, BIBREF32, BIBREF33. One major dataset of English Wikipedia editor comments BIBREF29 focuses on the `personal attack' element of harassment, drawing on prior investigations that mapped out harassment in that community. Another widely used dataset focuses on trolls' intent to intimidate, distinguishing between direct harassment and other behaviours BIBREF34. An important consideration in studies of person-directed abuse is (a) interpersonal relations, such as whether individuals engage in patterns of abuse or one-off acts and whether they are known to each other in the `real' world (both of which are a key concern in studies of cyberbullying) and (b) standpoint, such as whether individuals directly engage in abuse themselves or encourage others to do so. For example, the theoretically sophisticated synthetic dataset provided by BIBREF33 identifies not only harassment but also encouragement to harassment. BIBREF22 mark up posts from computer game forums (World of Warcraft and League of Legends) for cyberbullying and annotate these as $\langle $offender, victim, message$\rangle $ tuples. Group-directed abuse. Content which directs negativity against a social identity, which is defined in relation to a particular attribute (e.g. ethnic, racial, religious groups)BIBREF35. Such abuse is often directed against marginalised or under-represented groups in society. Group-directed abuse is typically described as `hate speech' and includes use of dehumanising language, making derogatory, demonising or hostile statements, making threats, and inciting others to engage in violence, amongst other dangerous communications. Common examples of group-directed abuse include sexism, which is included in datasets provided by BIBREF36, BIBREF37, BIBREF38, BIBREF39, BIBREF33 and racism, which is directly targeted in BIBREF36, BIBREF40. In some cases, specific types of group-directed abuse are subsumed within a broader category of identity-directed abuse, as in BIBREF41, BIBREF42, BIBREF4. Determining the limits of any group-directed abuse category requires careful theoretical reflection, as with the decision to include ethnic, caste-based and certain religious prejudices under `racism'. There is no `right' answer to such questions as they engage with ontological concerns about identification and `being' and the politics of categorization. Flagged content. Content which is reported by community members or assessed by community and professional content moderators. This covers a broad range of focuses as moderators may also remove spam, sexually inappropriate content and other undesirable contributions. In this regard, `flagged' content is akin to the concept of `trolling', which covers a wide range of behaviours, from jokes and playful interventions through to sinister personal attacks such as doxxing BIBREF43. Some forms of trolling can be measured with tools such as the Global Assessment of Internet Trolling (GAIT) BIBREF43. Incivility. Content which is considered to be incivil, rude, inappropriate, offensive or disrespectful BIBREF24, BIBREF25, BIBREF44. Such categories are usually defined with reference to the tone that the author adopts rather than the substantive content of what they express, which is the basis of person- and group- directed categories. Such content usually contains obscene, profane or otherwise `dirty' words. This can be easier to detect as closed-class lists are effective at identifying single objectionable words (e.g. BIBREF45). However, one concern with this type of research is that the presence of `dirty' words does not necessarily signal malicious intent or abuse; they may equally be used as intensifiers or colloquialisms BIBREF46. At the same time, detecting incivility can be more difficult as it requires annotators to infer the subjective intent of the speaker or to understand (or guess) the social norms of a setting and thus whether disrespect has been expressed BIBREF42. Content can be incivil without directing hate against a group or person, and can be inappropriate in one setting but not another: as such it tends to be more subjective and contextual than other types of abusive language. Mixed. Content which contains multiple types of abuse, usually a combination of the four categories discussed above. The intersecting nature of online language means that this is common but can also manifest in unexpected ways. For instance, female politicians may receive more interpersonal abuse than other politicians. This might not appear as misogyny because their identity as women is not referenced – but it might have motivated the abuse they were subjected to. Mixed forms of abuse require further research, and have thus far been most fully explored in the OLID dataset provided by BIBREF4, who explore several facets of abuse under one taxonomy. ### Analysis of training datasets ::: The purpose of training datasets ::: Uses of datasets: How detection tasks are defined ::: Detection tasks: Granularity of taxonomies
This refers to how much detail a taxonomy contains, reflected in the number of unique classes. The most important and widespread distinction is whether a binary class is used (e.g. Hate / Not) or a multi-level class, such as a tripartite split (typically, Overt, Covert and Non-abusive). In some cases, a large number of complex classes are created, such as by combining whether the abuse is targeted or not along with its theme and strength. In general, Social scientific analyses encourage creating a detailed taxonomy with a large number of fine-grained categories. However, this is only useful for machine learning if there are enough data points in each category and if annotators are capable of consistently distinguishing between them. Complex annotation schemas may not result in better training datasets if they are not implemented in a robust way. As such, it is unsurprising that binary classification schemas are the most prevalent, even though they are arguably the least useful given the variety of ways in which abuse can be articulated. This can range from the explicit and overt (e.g. directing threats against a group) to more subtle behaviours, such as micro-aggressions and dismissing marginalised groups' experiences of prejudice. Subsuming both types of behaviour within one category not only risks making detection difficult (due to considerable in-class variation) but also leads to a detection system which cannot make important distinctions between qualitatively different types of content. This has severe implications for whether detection systems trained on such datasets can actually be used for downstream tasks, such as content moderation and social scientific analysis. Drawing together the nature and granularity of abuse, our analyses identify a hierarchy of taxonomic granularity from least to most granular: Binary classification of a single `meta' category, such as hate/not or abuse/not. This can lead to very general and vague research, which is difficult to apply in practice. Binary classification of a single type of abuse, such as person-directed or group-directed. This can be problematic given that abuse is nearly always directed against a group rather than `groups' per se. Binary classification of abuse against a single well-defined group, such as racism/not or Islamophobia/not, or interpersonal abuse against a well-defined cohort, such as MPs and young people. Multi-class (or multi-label) classification of different types of abuse, such as: Multiple targets (e.g. racist, sexist and non-hateful content) or Multiple strengths (e.g. none, implicit and explicit content). Multiple types (e.g. threats versus derogatory statements or benevolent versus hostile statements). Multi-class classification of different types of abuse which is integrated with other dimensions of abuse. ### Analysis of training datasets ::: The content of training datasets ::: The `Level' of content
49 of the training datasets are annotated at the level of the post, one dataset is annotated at the level of the user BIBREF47, and none of them are annotated at the level of the comment thread. Only two publications indicate that the entire conversational thread was presented to annotators when marking up individual entries, meaning that in most cases this important contextual information is not used. 49 of the training datasets contain only text. This is a considerable limitation of existing research BIBREF13, especially given the multimodal nature of online communication and the increasing ubiquity of digital-specific image-based forms of communication such as Memes, Gifs, Filters and Snaps BIBREF48. Although some work has addressed the task of detecting hateful images BIBREF49, BIBREF50, this lead to the creation of a publically available labelled training dataset in only one case BIBREF51. To our knowledge, no research has tackled the problem of detecting hateful audio content. This is a distinct challenge; alongside the semantic content audio also contains important vocal cues which provide more opportunities to investigate (but also potentially misinterpret) tone and intention. ### Analysis of training datasets ::: The content of training datasets ::: Language
The most common language in the training datasets is English, which appears in 20 datasets, followed by Arabic and Italian (5 datasets each), Hindi-English (4 datasets) and then German, Indonesian and Spanish (3 datasets). Noticeably, several major languages, both globally and in Europe, do not appear, which suggests considerable unevenness in the linguistic and cultural focuses of abusive language detection. For instance, there are major gaps in the coverage of European languages, including Danish and Dutch. Surprisingly, French only appears once. The dominance of English may be due to how we sampled publications (for which we used English terms), but may also reflect different publishing practices in different countries and how well-developed abusive content research is. ### Analysis of training datasets ::: The content of training datasets ::: Source of data
Training datasets use data collected from a range of online spaces, including from mainstream platforms, such as Twitter, Wikipedia and Facebook, to more niche forums, such as World of Warcraft and Stormfront. In most cases, data is collected from public sources and then manually annotated but in others data is sourced through proprietary data sharing agreements with host platforms. Unsurprisingly, Twitter is the most widely used source of data, accounting for 27 of the datasets. This reflects wider concerns in computational social research that Twitter is over-used, primarily because it has a very accessible API for data collection BIBREF52, BIBREF53. Facebook and Wikipedia are the second most used sources of data, accounting for three datasets each – although we note that all three Wikipedia datasets are reported in the same publication. Many of the most widely used online platforms are not represented at all, or only in one dataset, such as Reddit, Weibo, VK and YouTube. The lack of diversity in where data is collected from limits the development of detection systems. Three main issues emerge: Linguistic practices vary across platforms. Twitter only allows 280 characters (previously only 140), provoking stylistic changes BIBREF54, and abusive content detection systems trained on this data are unlikely to work as well with longer pieces of text. Dealing with longer pieces of text could necessitate different classification systems, potentially affecting the choice of algorithmic architecture. Additionally, the technical affordances of platforms may affect the style, tone and topic of the content they host. The demographics of users on different platforms vary considerably. Social science research indicates that `digital divides' exist, whereby online users are not representative of wider populations and differ across different online spaces BIBREF53, BIBREF55, BIBREF56. Blank draws attention to how Twitter users are usually younger and wealthier than offline populations; over reliance on data from Twitter means, in effect, that we are over-sampling data from this privileged section of society. Blank also shows that there are also important cross-national differences: British Twitters are better-educated than the offline British population but the same is not true for American Twitter users compared with the offline American population BIBREF56. These demographic differences are likely to affect the types of content that users produce. Platforms have different norms and so host different types and amounts of abuse. Mainstream platforms have made efforts in recent times to `clean up' content and so the most overt and aggressive forms of abuse, such as direct threats, are likely to be taken down BIBREF57. However, more niche platforms, such as Gab or 4chan, tolerate more offensive forms of speech and are more likely to contain explicit abuse, such as racism and very intrusive forms of harassment, such as `doxxing' BIBREF58, BIBREF59, BIBREF60. Over-reliance on a few sources of data could mean that datasets are biased towards only a subset of types of abuse. ### Analysis of training datasets ::: The content of training datasets ::: Size
The size of the training datasets varies considerably from 469 posts to 17 million; a difference of four orders of magnitude. Differences in size partly reflect different annotation approaches. The largest datasets are from proprietary data sharing agreements with platforms. Smaller datasets tend to be carefully collected and then manually annotated. There are no established guidelines for how large an abusive language training dataset needs to be. However, smaller datasets are problematic because they contain too little linguistic variation and increase the likelihood of overfitting. Rizoiu et al.BIBREF61 train detection models on only a proportion of the Davidson et al. and Waseem training datasets and show that this leads to worse performance, with a lower F1-Score, particularly for `data hungry' deep learning approaches BIBREF61. At the same time, `big' datasets alone are not a panacea for the challenges of abusive content classification. Large training datasets which have been poorly sampled, annotated with theoretically problematic categories or inexpertly and unthoughtfully annotated, could still lead to the development of poor classification systems. The challenges posed by small datasets could potentially be overcome through machine learning techniques such as `semi-supervised' and `active' learning BIBREF62, although these have only been limitedly applied to abusive content detection so far BIBREF63. Sharifirad et al. propose using text augmentation and new text generation as a way of overcoming small datasets, which is a promising avenue for future research BIBREF64. ### Analysis of training datasets ::: The content of training datasets ::: Class distribution and sampling
Class distribution is an important, although often under-considered, aspect of the design of training datasets. Datasets with little abusive content will lack linguistic variation in terms of what is abusive, thereby increasing the risk of overfitting. More concerningly, the class distribution directly affects the nature of the engineering task and how performance should be evaluated. For instance, if a dataset is 70% hate speech then a zero-rule classification system (i.e. where everything is categorised as hate speech) will achieve 70% precision and 100% recall. This should be used as a baseline for evaluating performance: 80% precision is less impressive compared with this baseline. However, 80% precision on an evenly balanced dataset would be impressive. This is particularly important when evaluating the performance of ternary classifiers, when classes can be considerably imbalanced. On average, 35% of the content in the training datasets is abusive. However, class distributions vary considerably, from those with just 1% abusive content up to 100%. These differences are largely a product of how data is sampled and which platform it is taken from. Bretschneider BIBREF22 created two datasets without using purposive sampling, and as such they contain very low levels of abuse ( 1%). Other studies filter data collection based on platforms, time periods, keywords/hashtags and individuals to increase the prevalence of abuse. Four datasets comprise only abusive content; three cases are synthetic datasets, reported on in one publication BIBREF65, and in the other case the dataset is an amendment to an existing dataset and only contains misogynistic content BIBREF37. Purposive sampling has been criticised for introducing various forms of bias into datasets BIBREF66, such as missing out on mis-spelled content BIBREF67 and only focusing on the linguistic patterns of an atypical subset of users. One pressing risk is that a lot of data is sampled from far right communities – which means that most hate speech classifiers implicitly pick up on right wing styles of discourse rather than hate speech per se. This could have profound consequences for our understanding of online political dialogue if the classifiers are applied uncritically to other groups. Nevertheless, purposive sampling is arguably a necessary step when creating a training dataset given the low prevalence of abuse on social media in general BIBREF68. ### Analysis of training datasets ::: The content of training datasets ::: Identity of the content creators
The identity of the users who originally created the content in training datasets is described in only two cases. In both cases the data is synthetic BIBREF65, BIBREF33. Chung et al. use `nichesourcing' to synthetically generate abuse, with experts in tackling hate speech creating hateful posts. Sprugnoli et al. ask children to adopt pre-defined roles in an experimental classroom setup, and ask them to engage in a cyberbullying scenario. In most of the non-synthetic training datasets, some information is given about the sampling criteria used to collect data, such as hashtags. However, this does not provide direct insight into who the content creators are, such as their identity, demographics, online behavioural patterns and affiliations. Providing more information about content creators may help address biases in existing datasets. For instance, Wiegand et al. show that 70% of the sexist tweets in the highly cited Waseem and Hovy dataset BIBREF36 come from two content creators and that 99% of the racist tweets come from just one BIBREF66. This is a serious constraint as it means that user-level metadata is artificially highly predictive of abuse. And, even when user-level metadata is not explicitly modelled, detection systems only need to pick up on the linguistic patterns of a few authors to nominally detect abuse. Overall, the complete lack of information about which users have created the content in most training datasets is a substantial limitation which may be driving as-yet-unrecognised biases. This can be remedied through the methodological rigour implicit in including a data statement with a corpus. ### Analysis of training datasets ::: Annotation of training datasets ::: Annotation process
How training datasets are annotated is one of the most important aspects of their creation. A range of annotation processes are used in training datasets, which we split into five high-level categories: Crowdsourcing (15 datasets). Crowdsourcing is widely used in NLP research because it is relatively cheap and easy to implement. The value of crowdsourcing lies in having annotations undertaken by `a large number of non-experts' (BIBREF69, p. 278) – any bit of content can be annotated by multiple annotators, effectively trading quality for quantity. Studies which use crowdsourcing with only a few annotators for each bit of content risk minimising quality without counterbalancing it with greater quantity. Furthermore, testing the work of many different annotators can be challenging BIBREF70, BIBREF71 and ensuring they are paid an ethical amount may make the cost comparable to using trained experts. Crowdsourcing has also been associated with `citizen science' initiatives to make academic research more accessible but this may not be fully realised in cases where annotation tasks are laborious and low-skilled BIBREF72, BIBREF20. Academic experts (22 datasets). Expert annotation is time-intensive but is considered to produce higher quality annotations. Waseem reports that `systems trained on expert annotations outperform systems trained on amateur annotations.' BIBREF73 and, similarly, D'Orazio et al. claim, `although expert coding is costly, it produces quality data.' BIBREF74. However, the notion of an `expert' remains somewhat fuzzy within abusive content detection research. In many cases, publications only report that `an expert' is used, without specifying the nature of their expertise – even though this can vary substantially. For example, an expert may refer to an NLP practitioner, an undergraduate student with only modest levels of training, a member of an attacked social group relevant to the dataset or a researcher with a doctorate in the study of prejudice. In general, we anticipate that experts in the social scientific study of prejudice/abuse would perform better at annotation tasks then NLP experts who may not have any direct expertise in the conceptual and theoretical issues of abusive content annotation. In particular, one risk of using NLP practitioners, whether students or professionals, is that they might `game' training datasets based on what they anticipate is technically feasible for existing detection systems. For instance, if existing systems perform poorly when presented with long range dependencies, humour or subtle forms of hate (which are nonetheless usually discernible to human readers) then NLP experts could unintentionally use this expectation to inform their annotations and not label such content as hateful. Professional moderators (3 datasets). Professional moderators offer a standardized approach to content annotated, implemented by experienced workers. This should, in principle, result in high quality annotations. However, one concern is that moderators are output-focused as their work involves determining whether content should be allowed or removed from platforms; they may not provide detailed labels about the nature of abuse and may also set the bar for content labelled `abusive' fairly high, missing out on more nuance and subtle varieties. In most cases, moderators will annotate for a range of unacceptable content, such as spam and sexual content, and this must be marked in datasets. A mix of crowdsourcing and experts (6 datasets). Synthetic data creation (4 datasets). Synthetic datasets are an interesting option as they are inherently non-authentic and therefore not necessarily representative of how abuse manifests in real-world situations. However, if they are created in realistic conditions by experts or relevant content creators then they can mimic real behaviour and have the added advantage that they may have broader coverage of different types of abuse. They are also usually easier to share. ### Analysis of training datasets ::: Annotation of training datasets ::: Identity of the annotators
The data statements framework given by Bender and Friedman emphasises the importance of understanding who has completed annotations. Knowing who the annotators are is important because `their own “social address" influences their experience with language and thus their perception of what they are annotating.' BIBREF18 In the context of online abuse, Binns et al. show that the gender of annotators systematically influences what annotations they provide BIBREF75. No annotator will be well-versed in all of the slang or coded meanings used to construct abusive language. Indeed, many of these coded meanings are deliberately covert and obfuscated BIBREF76. To help mitigate these challenges, annotators should be (a) well-qualified and (b) diverse. A homogeneous group of annotators will be poorly equipped to catch all instances of abuse in a corpus. Recruiting an intentionally mixed groups of annotators is likely to yield better recall of abuse and thus a more precise dataset BIBREF77. Information about annotators is unfortunately scarce. In 23 of the training datasets no information is given about the identity of annotators; in 17 datasets very limited information is given, such as whether the annotator is a native speaker of the language; and in just 10 cases is detailed information given. Interestingly, only 4 out of these 10 datasets are in the English language. Relevant information about annotators can be split into (i) Demographic information and (ii) annotators' expertise and experience. In none of the training sets is the full range of annotator information made available, which includes: Demographic information. The nature of the task affects what information should be provided, as well as the geographic and cultural context. For instance, research on Islamophobia should include, at the very least, information about annotators' religious affiliation. Relevant variables include: Age Ethnicity and race Religion Gender Sexual Orientation Expertise and experience. Relevant variables include: Field of research Years of experience Research status (e.g. research assistant or post-doc) Personal experiences of abuse. In our review, none of the datasets contained systematic information about whether annotators had been personally targeted by abuse or had viewed such abuse online, even though this can impact annotators' perceptions. Relevant variables include: Experiences of being targeted by online abuse. Experiences of viewing online abuse. ### Analysis of training datasets ::: Annotation of training datasets ::: Guidelines for annotation
A key source of variation across datasets is whether annotators were given detailed guidelines, very minimal guidelines or no guidelines at all. Analysing this issue is made difficult by the fact that many dataset creators do not share their annotation guidelines. 21 of the datasets we study do not provide the guidelines and 14 only provide them in a highly summarised form. In just 15 datasets is detailed information given (and these are reported on in just 9 publications). Requiring researchers to publish annotation guidelines not only helps future researchers to better understand what datasets contain but also to improve and extend them. This could be crucial for improving the quality of annotations; as Ross et al. recommend, `raters need more detailed instructions for annotation.' BIBREF78 The degree of detail given in guidelines is linked to how the notion of `abuse' is understood. Some dataset creators construct clear and explicit guidelines in an attempt to ensure that annotations are uniform and align closely with social scientific concepts. In other cases, dataset creators allow annotators to apply their own perception. For instance, in their Portuguese language dataset, Fortuna et al. ask annotators to `evaluate if according to your opinion, these tweets contain hate speech' BIBREF38. The risk here is that authors' perceptions may differ considerably; Salminen et al. show that online hate interpretation varies considerably across individuals BIBREF79. This is also reflected in inter-annotator agreement scores for abusive content, which is often very low, particularly for tasks which deploy more than just a binary taxonomy. However, it is unlikely that annotators could ever truly divorce themselves from their own social experience and background to decide on a single `objective' annotation. Abusive content annotation is better understood, epistemologically, as an intersubjective process in which agreement is constructed, rather than an objective process in which a `true' annotation is `found'. For this reason, some researchers have shifted the question of `how can we achieve the correct annotation?' to `who should decide what the correct annotation is?' BIBREF73. Ultimately, whether annotators should be allowed greater freedom in making annotations, and whether this results in higher quality datasets, needs further research and conceptual examination. Some aspects of abusive language present fundamental issues that are prone to unreliable annotation, such as Irony, Calumniation and Intent. They are intrinsically difficult to annotate given a third-person perspective on a piece of text as they involve making a judgement about indeterminate issues. However, they cannot be ignored given their prevalence in abusive content and their importance to how abuse is expressed. Thus, although they are fundamentally conceptual problems, these issues also present practical problems for annotators, and should be addressed explicitly in coding guidelines. Otherwise, as BIBREF80 note, these issues are likely to drive type II errors in classification, i.e. labelling non-hate-speech utterances as hate speech. ### Analysis of training datasets ::: Annotation of training datasets ::: Guidelines for annotation ::: Irony
This covers statements that have a meaning contrary to that one might glean at first reading. Lachenicht BIBREF81 notes that Irony goes against Grice's quality maxim, and as such Ironic content requires closer attention from the reader as it is prone to being misinterpreted. Irony is a particularly difficult issue as in some cases it is primarily intended to provide humour (and thus might legitimately be considered non-abusive) but in other cases is used as a way of veiling genuine abuse. Previous research suggests that the problem is widespread. Sanguinetti et al. BIBREF82 find irony in 11% of hateful tweets in Italian. BIBREF25 find that irony is one of the most common phenomena in self-deleted comments; and that the prevalence of irony is 33.9% amongst deleted comments in a Croatian comment dataset and 18.1% amongst deleted comments in a Slovene comment dataset. Furthermore, annotating irony (as well as related constructs, such as sarcasm and humour) is inherently difficult. BIBREF83 report that agreement on sarcasm amongst annotators working in English is low, something echoed by annotations of Danish content BIBREF84. Irony is also one of the most common reasons for content to be re-moderated on appeal, according to Pavlopoulos et al. BIBREF24. ### Analysis of training datasets ::: Annotation of training datasets ::: Guidelines for annotation ::: Calumniation
This covers false statements, slander, and libel. From the surveyed set, this is annotated in datasets for Greek BIBREF24 and for Croatian and Slovene BIBREF25. Its prevalence varies considerably across these two datasets and reliable estimations of the prevalence of false statements are not available. Calumniation is not only an empirical issue, it also raises conceptual problems: should false information be considered abusive if it slanders or demeans a person? However, if the information is then found out to be true does it make the content any less abusive? Given the contentiousness of `objectivity', and the lack of consensus about most issues in a `post-truth' age BIBREF85, who should decide what is considered true? And, finally, how do we determine whether the content creator knows whether something is true? These ontological, epistemological and social questions are fundamental to the issue of truth and falsity in abusive language. Understandably, most datasets do not taken any perspective on the truth and falsity of content. This is a practical solution: given error rates in abusive language detection as well as error rates in fact-checking, a system which combined both could be inapplicable in practice. ### Analysis of training datasets ::: Annotation of training datasets ::: Guidelines for annotation ::: Intent
This information about the utterer's state of mind is a core part of how many types of abusive language are defined. Intent is usually used to emphasize the wrongness of abusive behaviour, such as spreading, inciting, promoting or justifying hatred or violence towards a given target, or sending a message that aims at dehumanising, delegitimising, hurting or intimidating them BIBREF82. BIBREF81 postulate that “aggravation, invective and rudeness ... may be performed with varying degrees of intention to hurt", and cite five legal degrees of intent BIBREF86. However, it is difficult to discern the intent of another speaker in a verbal conversation between humans, and even more difficult to do so through written and computer-mediated communications BIBREF87. Nevertheless, intent is particularly important for some categories of abuse such as bullying, maliciousness and hostility BIBREF34, BIBREF32. Most of the guidelines for the datasets we have studied do not contain an explicit discussion of intent, although there are exceptions. BIBREF88 include intent as a core part of their annotation standard, noting that understanding context (such as by seeing a speakers' other online messages) is crucial to achieving quality annotations. However, this proposition poses conceptual challenges given that people's intent can shift over time. Deleted comments have been used to study potential expressions of regret by users and, as such, a change in their intent BIBREF89, BIBREF25; this has also been reported as a common motivator even in self-deletion of non-abusive language BIBREF90. Equally, engaging in a sequence of targeted abusive language is an indicator of aggressive intent, and appears in several definitions. BIBREF23 require an “intent to physically assert power over women" as a requirement for multiple categories of misogynistic behaviour. BIBREF34 find that messages that are “unapologetically or intentionally offensive" fit in the highest grade of trolling under their schema. Kenny et al. BIBREF86 note how sarcasm, irony, and humour complicate the picture of intent by introducing considerable difficulties in discerning the true intent of speakers (as discussed above). Part of the challenge is that many abusive terms, such as slurs and insults, are polysemic and may be co-opted by an ingroup into terms of entertainment and endearment BIBREF34. ### Dataset sharing ::: The challenges and opportunities of achieving Open Science
All of the training datasets we analyse are publicly accessible and as such can be used by researchers other than the authors of the original publication. Sharing data is an important aspect of open science but also poses ethical and legal risks, especially in light of recent regulatory changes, such as the introduction of GPDR in the UK BIBREF91, BIBREF92. This problem is particularly acute with abusive content, which can be deeply shocking, and some training datasets from highly cited publications have not been made publicly available BIBREF93, BIBREF94, BIBREF95. Open science initiatives can also raise concerns amongst the public, who may not be comfortable with researchers sharing their personal data BIBREF96, BIBREF97. The difficulty of sharing data in sensitive areas of research is reflected by the Islamist extremism research website, `Jihadology'. It chose to restrict public access in 2019, following efforts by Home Office counter-terrorism officials to shut it down completely. They were concerned that, whilst it aimed to support academic research into Islamist extremism, it may have inadvertently enabled individuals to radicalise by making otherwise banned extremist material available. By working with partners such as the not-for-profit Tech Against Terrorism, Jihadology created a secure area in the website, which can only be accessed by approved researchers. Some of the training datasets in our list have similar requirements, and can only be accessed following a registration process. Open sharing of datasets is not only a question of scientific integrity and a powerful way of advancing scientific knowledge. It is also, fundamentally, a question of fairness and power. Opening access to datasets will enable less-well funded researchers and organisations, which includes researchers in the Global South and those working for not-for-profit organisations, to steer and contribute to research. This is a particularly pressing issue in a field which is directly concerned with the experiences of often-marginalised communities and actors BIBREF36. For instance, one growing concern is the biases encoded in detection systems and the impact this could have when they are applied in real-world settings BIBREF9, BIBREF10. This research could be further advanced by making more datasets and detection systems more easily available. For instance, Binns et al. use the detailed metadata in the datasets provided by Wulczyn et al. to investigate how the demographics of annotators impacts the annotations they make BIBREF75, BIBREF29. The value of such insights is only clear after the dataset has been shared – and, equally, is only possible because of data sharing. More effective ways of sharing datasets would address the fact that datasets often deteriorate after they have been published BIBREF13. Several of the most widely used datasets provide only the annotations and IDs and must be `rehydrated' to collect the content. Both of the datasets provided by Waseem and Hovy and Founta et al. must be collected in this way BIBREF98, BIBREF36, and both have degraded considerably since they were first released as the tweets are no longer available on Twitter. Chung et al. also estimate that within 12 months the recently released dataset for counterspeech by Mathew et al. had lost more than 60% of its content BIBREF65, BIBREF58. Dataset degradation poses three main risks: First, if less data is available then there is a greater likelihood of overfitting. Second, the class distributions usually change as proportionally more of the abusive content is taken down than the non-abusive. Third, it is also likely that the more overt forms of abuse are taken down, rather than the covert instances, thereby changing the qualitative nature of the dataset. ### Dataset sharing ::: Research infrastructure: Solutions for sharing training datasets
The problem of data access and sharing remains unresolved in the field of abusive content detection, much like other areas of computational research BIBREF99. At present, an ethical, secure and easy way of sharing sensitive tools and resources has not been developed and adopted in the field. More effective dataset sharing would (1) greater collaboration amongst researchers, (2) enhance the reproducibility of research by encouraging greater scrutiny BIBREF100, BIBREF101, BIBREF102 and (3) substantively advance the field by enabling future researchers to better understand the biases and limitations of existing research and to identify new research directions. There are two main challenges which must be overcome to ensure that training datasets can be shared and used by future researchers. First, dataset quality: the size, class distribution and quality of their content must be maintained. Second, dataset access: access to datasets must be controlled so that researchers can use them, whilst respecting platforms' Terms of Service and not allowing potential extremists from having access. These problems are closely entwined and the solutions available, which follow, have implications for both of them. Synthetic datasets. Four of the datasets we have reviewed were developed synthetically. This resolves the dataset quality problem but introduces additional biases and limitations because the data is not real. Synthetic datasets still need to be shared in such a way as to limit access for potential extremists but face no challenges from Platforms' Terms of Services. Data `philanthropy' or `donations'. These are defined as `the act of an individual actively consenting to donate their personal data for research' BIBREF97. Donated data from many individuals could then be combined and shared – but it would still need to be annotated. A further challenge is that many individuals who share abusive content may be unwilling to `donate' their data as this is commonly associated with prosocial motivations, creating severe class imbalances BIBREF97. Data donations could also open new moral and ethical issues; individuals' privacy could be impacted if data is re-analysed to derive new unexpected insights BIBREF103. Informed consent is difficult given that the exact nature of analyses may not be known in advance. Finally, data donations alone do not solve how access can be responsibly protected and how platforms' Terms of Service can be met. For these reasons, data donations are unlikely to be a key part of future research infrastructure for abusive content detection. Platform-backed sharing. Platforms could share datasets and support researchers' access. There are no working examples of this in abusive content detection research, but it has been successfully used in other research areas. For instance, Twitter has made a large dataset of accounts linked to potential information operations, known as the “IRA" dataset (Internet Research Agency). This would require considerably more interfaces between academia and industry, which may be difficult given the challenges associated with existing initiatives, such as Social Science One. However, in the long term, we propose that this is the most effective solution for the problem of sharing training datasets. Not only because it removes Terms of Service limitations but also because platforms have large volumes of original content which has been annotated in a detailed way. This could take one of two forms: platforms either make content which has violated their Community Guidelines available directly or they provide special access post-hoc to datasets which researchers have collected publicly through their API - thereby making sure that datasets do not degrade over time. Data trusts. Data trusts have been described as a way of sharing data `in a fair, safe and equitable way' ( BIBREF104 p. 46). However, there is considerable disagreement as to what they entail and how they would operate in practice BIBREF105. The Open Data Institute identifies that data trusts aim to make data open and accessible by providing a framework for storing and accessing data, terms and mechanisms for resolving disputes and, in some cases, contracts to enforce them. For abusive content training datasets, this would provide a way of enabling datasets to be shared, although it would require considerable institutional, legal and financial commitments. Arguably, the easiest way of ensuring data can be shared is to maintain a very simple data trust, such as a database, which would contain all available abusive content training datasets. This repository would need to be permissioned and access controlled to address concerns relating to privacy and ethics. Such a repository could substantially reduce the burden on researchers; once they have been approved to the repository, they could access all datasets publicly available – different levels of permission could be implemented for different datasets, depending on commercial or research sensitivity. Furthermore, this repository could contain all of the metadata reported with datasets and such information could be included at the point of deposit, based on the `data statements' work of Bender and Friedman BIBREF18. A simple API could be developed for depositing and reading data, similar to that of the HateBase. The permissioning system could be maintained either through a single institution or, to avoid power concentrating amongst a small group of researchers, through a decentralised blockchain. ### Dataset sharing ::: A new repository of training datasets: Hatespeechdata.com
The resources and infrastructure to create a dedicated data trust and API for sharing abusive content training datasets is substantial and requires considerable further engagement with research teams in this field. In the interim, to encourage greater sharing of datasets, we have launched a dedicated website which contains all of the datasets analysed here: https://hatespeechdata.com. Based on the analysis in the previous sections, we have also provided partial data statements BIBREF18. The website also contains previously published abusive keyword dictionaries, which are not analysed here but some researchers may find useful. Note that the website only contains information/data which the original authors have already made publicly available elsewhere. It will be updated with new datasets in the future. ### Best Practices for training dataset creation
Much can be learned from existing efforts to create abusive language datasets. We identify best practices which emerge at four distinct points in the process of creating a training dataset: (1) task formation, (2) data selection, (3) annotation, and (4) documentation. ### Best Practices for training dataset creation ::: Task formation: Defining the task addressed by the dataset
Dataset creation should be `problem driven' BIBREF106 and should address a well-defined and specific task, with a clear motivation. This will directly inform the taxonomy design, which should be well-specified and engage with social scientific theory as needed. Defining a clear task which the dataset addresses is especially important given the maturation of the field, ongoing terminological disagreement and the complexity of online abuse. The diversity of phenomena that fits under the umbrella of abusive language means that `general purpose' datasets are unlikely to advance the field. New datasets are most valuable when they address a new target, generator, phenomenon, or domain. Creating datasets which repeat existing work is not nearly as valuable. ### Best Practices for training dataset creation ::: Selecting data for abusive language annotation
Once the task is established, dataset creators should select what language will be annotated, where data will be sampled from and how sampling will be completed. Any data selection exercise is bound to give bias, and so it is important to record what decisions are made (and why) in this step. Dataset builders should have a specific target size in mind and also have an idea of the minimum amount of data this is likely to be needed for the task. This is also where steps 1 and 2 intersect: the data selection should be driven by the problem that is addressed rather than what is easy to collect. Ensuring there are enough positive examples of abuse will always be challenging as the prevalence of abuse is so low. However, given that purposive sampling inevitably introduces biases, creators should explore a range of options before determining the best one – and consider using multiple sampling methods at once, such as including data from different times, different locations, different types of users and different platforms. Other options include using measures of linguistic diversity to maximize the variety of text included in datasets, or including words that cluster close to known abusive terms. ### Best Practices for training dataset creation ::: Annotating abusive language
Annotators must be hired, trained and given appropriate guidelines. Annotators work best with solid guidelines, that are easy to grasp and have clear examples BIBREF107. The best examples are both illustrative, in order to capture the concepts (such as `threatening language') and provide insight into `edge cases', which is content that only just crosses the line into abuse. Decisions should be made about how to handle intrinsically difficult aspects of abuse, such as irony, calumniation and intent (see above). Annotation guidelines should be developed iteratively by dataset creators; by working through the data, rules can be established for difficult or counter-intuitive coding decisions, and a set of shared practices developed. Annotators should be included in this iterative process. Discussions with annotators the language that they have seen “in the field" offers an opportunity to enhance and refine guidelines - and even taxonomies. Such discussions will lead to more consistent data and provide a knowledge base to draw on for future work. To achieve this, it is important to adopt an open culture where annotators are comfortable providing open feedback and also describing their uncertainties. Annotators should also be given emotional and practical support (as well as appropriate financial compensation), and the harmful and potentially triggering effects of annotating online abuse should be recognised at all times. For a set of guidelines to help protect the well-being of annotators, see BIBREF13. ### Best Practices for training dataset creation ::: Documenting methods, data, and annotators
The best training datasets provide as much information as possible and are well-documented. When the method behind them is unclear, they are hard to evaluate, use and build on. Providing as much information as possible can open new and unanticipated analyses and gives more agency to future researchers who use the dataset to create classifiers. For instance, if all annotators' codings are provided (rather than just the `final' decision) then a more nuanced and aware classifier could be developed as, in some cases, it can be better to maximise recall of annotations rather than maximise agreement BIBREF77. Our review found that most datasets have poor methodological descriptions and few (if any) provide enough information to construct an adequate data statement. It is crucial that dataset creators are up front about their biases and limitations: every dataset is biased, and this is only problematic when the biases are unknown. One strategy for doing this is to maintain a document of decisions made when designing and creating the dataset and to then use it to describe to readers the rationale behind decisions. Details about the end-to-end dataset creation process are welcomed. For instance, if the task is crowdsourced then a screenshot of the micro-task presented to workers should be included, and the top-level parameters should be described (e.g. number of workers, maximum number of tasks per worker, number of annotations per piece of text) BIBREF20. If a dedicated interface is used for the annotation, this should also be described and screenshotted as the interface design can influence the annotations. ### Best Practices for training dataset creation ::: Best practice summary
Unfortunately, as with any burgeoning field, there is confusion and overlap around many of the phenomena discussed in this paper; coupled with the high degree of variation in the quality of method descriptions, it has lead to many pieces of research that are hard to combine, compare, or re-use. Our reflections on best practices are driven by this review and the difficulties of creating high quality training datasets. For future researchers, we summarise our recommendations in the following seven points: Bear in mind the purpose of the dataset; design the dataset to help address questions and problems from previous research. Avoid using `easy to access' data, and instead explore new sources which may have greater diversity. Consider what biases may be created by your sampling method. Determine size based on data sparsity and having enough positive classes rather than `what is possible'. Establish a clear taxonomy to be used for the task, with meaningful and theoretically sound categories. Provide annotators with guidelines; develop them iteratively and publish them with your dataset. Consider using trained annotators given the complexities of abusive content. Involve people who have direct experience of the abuse which you are studying whenever possible (and provided that you can protect their well-being). Report on every step of the research through a Data Statement. ### Conclusion
This paper examined a large set of datasets for the creation of abusive content detection systems, providing insight into what they contain, how they are annotated, and how tasks have been framed. Based on an evidence-driven review, we provided an extended discussion of how to make training datasets more readily available and useful, including the challenges and opportunities of open science as well as the need for more research infrastructure. We reported on the development of hatespeechdata.com – a new repository for online abusive content training datasets. Finally, we outlined best practices for creation of training datasets for detection of online abuse. We have effectively met the four research aims elaborated at the start of the paper. Training detection systems for online abuse is a substantial challenge with real social consequences. If we want the systems we develop to be useable, scalable and with few biases then we need to train them on the right data: garbage in will only lead to garbage out. Fig. 1. Year in which training datasets were originally published Fig. 2. Primary language of dataset Fig. 3. Platform from which data is gathered Fig. 4. Distribution of dataset sizes Fig. 5. Relative size of “abusive" data class Table 1. Datasets surveyed. Synthetic data use indicated with asterisk.
|
from 469 posts to 17 million
|
What is the most likely reason that Secretary of State Blendwell appointed Malloy for the peace talks?
A. He had the most recent practice maintaining peace.
B. He was the closest diplomat available.
C. He was the most qualified to deal with the situation.
D. He has a reputation for not compromising.
|
IN CASE OF FIRE By RANDALL GARRETT There are times when a broken tool is better than a sound one, or a twisted personality more useful than a whole one. For instance, a whole beer bottle isn't half the weapon that half a beer bottle is ... Illustrated by Martinez In his office apartment, on the top floor of the Terran Embassy Building in Occeq City, Bertrand Malloy leafed casually through the dossiers of the four new men who had been assigned to him. They were typical of the kind of men who were sent to him, he thought. Which meant, as usual, that they were atypical. Every man in the Diplomatic Corps who developed a twitch or a quirk was shipped to Saarkkad IV to work under Bertrand Malloy, Permanent Terran Ambassador to His Utter Munificence, the Occeq of Saarkkad. Take this first one, for instance. Malloy ran his finger down the columns of complex symbolism that showed the complete psychological analysis of the man. Psychopathic paranoia. The man wasn't technically insane; he could be as lucid as the next man most of the time. But he was morbidly suspicious that every man's hand was turned against him. He trusted no one, and was perpetually on his guard against imaginary plots and persecutions. Number two suffered from some sort of emotional block that left him continually on the horns of one dilemma or another. He was psychologically incapable of making a decision if he were faced with two or more possible alternatives of any major importance. Number three ... Malloy sighed and pushed the dossiers away from him. No two men were alike, and yet there sometimes seemed to be an eternal sameness about all men. He considered himself an individual, for instance, but wasn't the basic similarity there, after all? He was—how old? He glanced at the Earth calendar dial that was automatically correlated with the Saarkkadic calendar just above it. Fifty-nine next week. Fifty-nine years old. And what did he have to show for it besides flabby muscles, sagging skin, a wrinkled face, and gray hair? Well, he had an excellent record in the Corps, if nothing else. One of the top men in his field. And he had his memories of Diane, dead these ten years, but still beautiful and alive in his recollections. And—he grinned softly to himself—he had Saarkkad. He glanced up at the ceiling, and mentally allowed his gaze to penetrate it to the blue sky beyond it. Out there was the terrible emptiness of interstellar space—a great, yawning, infinite chasm capable of swallowing men, ships, planets, suns, and whole galaxies without filling its insatiable void. Malloy closed his eyes. Somewhere out there, a war was raging. He didn't even like to think of that, but it was necessary to keep it in mind. Somewhere out there, the ships of Earth were ranged against the ships of the alien Karna in the most important war that Mankind had yet fought. And, Malloy knew, his own position was not unimportant in that war. He was not in the battle line, nor even in the major production line, but it was necessary to keep the drug supply lines flowing from Saarkkad, and that meant keeping on good terms with the Saarkkadic government. The Saarkkada themselves were humanoid in physical form—if one allowed the term to cover a wide range of differences—but their minds just didn't function along the same lines. For nine years, Bertrand Malloy had been Ambassador to Saarkkad, and for nine years, no Saarkkada had ever seen him. To have shown himself to one of them would have meant instant loss of prestige. To their way of thinking, an important official was aloof. The greater his importance, the greater must be his isolation. The Occeq of Saarkkad himself was never seen except by a handful of picked nobles, who, themselves, were never seen except by their underlings. It was a long, roundabout way of doing business, but it was the only way Saarkkad would do any business at all. To violate the rigid social setup of Saarkkad would mean the instant closing off of the supply of biochemical products that the Saarkkadic laboratories produced from native plants and animals—products that were vitally necessary to Earth's war, and which could be duplicated nowhere else in the known universe. It was Bertrand Malloy's job to keep the production output high and to keep the materiel flowing towards Earth and her allies and outposts. The job would have been a snap cinch in the right circumstances; the Saarkkada weren't difficult to get along with. A staff of top-grade men could have handled them without half trying. But Malloy didn't have top-grade men. They couldn't be spared from work that required their total capacity. It's inefficient to waste a man on a job that he can do without half trying where there are more important jobs that will tax his full output. So Malloy was stuck with the culls. Not the worst ones, of course; there were places in the galaxy that were less important than Saarkkad to the war effort. Malloy knew that, no matter what was wrong with a man, as long as he had the mental ability to dress himself and get himself to work, useful work could be found for him. Physical handicaps weren't at all difficult to deal with. A blind man can work very well in the total darkness of an infrared-film darkroom. Partial or total losses of limbs can be compensated for in one way or another. The mental disabilities were harder to deal with, but not totally impossible. On a world without liquor, a dipsomaniac could be channeled easily enough; and he'd better not try fermenting his own on Saarkkad unless he brought his own yeast—which was impossible, in view of the sterilization regulations. But Malloy didn't like to stop at merely thwarting mental quirks; he liked to find places where they were useful . The phone chimed. Malloy flipped it on with a practiced hand. "Malloy here." "Mr. Malloy?" said a careful voice. "A special communication for you has been teletyped in from Earth. Shall I bring it in?" "Bring it in, Miss Drayson." Miss Drayson was a case in point. She was uncommunicative. She liked to gather in information, but she found it difficult to give it up once it was in her possession. Malloy had made her his private secretary. Nothing—but nothing —got out of Malloy's office without his direct order. It had taken Malloy a long time to get it into Miss Drayson's head that it was perfectly all right—even desirable—for her to keep secrets from everyone except Malloy. She came in through the door, a rather handsome woman in her middle thirties, clutching a sheaf of papers in her right hand as though someone might at any instant snatch it from her before she could turn it over to Malloy. She laid them carefully on the desk. "If anything else comes in, I'll let you know immediately, sir," she said. "Will there be anything else?" Malloy let her stand there while he picked up the communique. She wanted to know what his reaction was going to be; it didn't matter because no one would ever find out from her what he had done unless she was ordered to tell someone. He read the first paragraph, and his eyes widened involuntarily. "Armistice," he said in a low whisper. "There's a chance that the war may be over." "Yes, sir," said Miss Drayson in a hushed voice. Malloy read the whole thing through, fighting to keep his emotions in check. Miss Drayson stood there calmly, her face a mask; her emotions were a secret. Finally, Malloy looked up. "I'll let you know as soon as I reach a decision, Miss Drayson. I think I hardly need say that no news of this is to leave this office." "Of course not, sir." Malloy watched her go out the door without actually seeing her. The war was over—at least for a while. He looked down at the papers again. The Karna, slowly being beaten back on every front, were suing for peace. They wanted an armistice conference—immediately. Earth was willing. Interstellar war is too costly to allow it to continue any longer than necessary, and this one had been going on for more than thirteen years now. Peace was necessary. But not peace at any price. The trouble was that the Karna had a reputation for losing wars and winning at the peace table. They were clever, persuasive talkers. They could twist a disadvantage to an advantage, and make their own strengths look like weaknesses. If they won the armistice, they'd be able to retrench and rearm, and the war would break out again within a few years. Now—at this point in time—they could be beaten. They could be forced to allow supervision of the production potential, forced to disarm, rendered impotent. But if the armistice went to their own advantage ... Already, they had taken the offensive in the matter of the peace talks. They had sent a full delegation to Saarkkad V, the next planet out from the Saarkkad sun, a chilly world inhabited only by low-intelligence animals. The Karna considered this to be fully neutral territory, and Earth couldn't argue the point very well. In addition, they demanded that the conference begin in three days, Terrestrial time. The trouble was that interstellar communication beams travel a devil of a lot faster than ships. It would take more than a week for the Earth government to get a vessel to Saarkkad V. Earth had been caught unprepared for an armistice. They objected. The Karna pointed out that the Saarkkad sun was just as far from Karn as it was from Earth, that it was only a few million miles from a planet which was allied with Earth, and that it was unfair for Earth to take so much time in preparing for an armistice. Why hadn't Earth been prepared? Did they intend to fight to the utter destruction of Karn? It wouldn't have been a problem at all if Earth and Karn had fostered the only two intelligent races in the galaxy. The sort of grandstanding the Karna were putting on had to be played to an audience. But there were other intelligent races throughout the galaxy, most of whom had remained as neutral as possible during the Earth-Karn war. They had no intention of sticking their figurative noses into a battle between the two most powerful races in the galaxy. But whoever won the armistice would find that some of the now-neutral races would come in on their side if war broke out again. If the Karna played their cards right, their side would be strong enough next time to win. So Earth had to get a delegation to meet with the Karna representatives within the three-day limit or lose what might be a vital point in the negotiations. And that was where Bertrand Malloy came in. He had been appointed Minister and Plenipotentiary Extraordinary to the Earth-Karn peace conference. He looked up at the ceiling again. "What can I do?" he said softly. On the second day after the arrival of the communique, Malloy made his decision. He flipped on his intercom and said: "Miss Drayson, get hold of James Nordon and Kylen Braynek. I want to see them both immediately. Send Nordon in first, and tell Braynek to wait." "Yes, sir." "And keep the recorder on. You can file the tape later." "Yes, sir." Malloy knew the woman would listen in on the intercom anyway, and it was better to give her permission to do so. James Nordon was tall, broad-shouldered, and thirty-eight. His hair was graying at the temples, and his handsome face looked cool and efficient. Malloy waved him to a seat. "Nordon, I have a job for you. It's probably one of the most important jobs you'll ever have in your life. It can mean big things for you—promotion and prestige if you do it well." Nordon nodded slowly. "Yes, sir." Malloy explained the problem of the Karna peace talks. "We need a man who can outthink them," Malloy finished, "and judging from your record, I think you're that man. It involves risk, of course. If you make the wrong decisions, your name will be mud back on Earth. But I don't think there's much chance of that, really. Do you want to handle small-time operations all your life? Of course not. "You'll be leaving within an hour for Saarkkad V." Nordon nodded again. "Yes, sir; certainly. Am I to go alone?" "No," said Malloy, "I'm sending an assistant with you—a man named Kylen Braynek. Ever heard of him?" Nordon shook his head. "Not that I recall, Mr. Malloy. Should I have?" "Not necessarily. He's a pretty shrewd operator, though. He knows a lot about interstellar law, and he's capable of spotting a trap a mile away. You'll be in charge, of course, but I want you to pay special attention to his advice." "I will, sir," Nordon said gratefully. "A man like that can be useful." "Right. Now, you go into the anteroom over there. I've prepared a summary of the situation, and you'll have to study it and get it into your head before the ship leaves. That isn't much time, but it's the Karna who are doing the pushing, not us." As soon as Nordon had left, Malloy said softly: "Send in Braynek, Miss Drayson." Kylen Braynek was a smallish man with mouse-brown hair that lay flat against his skull, and hard, penetrating, dark eyes that were shadowed by heavy, protruding brows. Malloy asked him to sit down. Again Malloy went through the explanation of the peace conference. "Naturally, they'll be trying to trick you every step of the way," Malloy went on. "They're shrewd and underhanded; we'll simply have to be more shrewd and more underhanded. Nordon's job is to sit quietly and evaluate the data; yours will be to find the loopholes they're laying out for themselves and plug them. Don't antagonize them, but don't baby them, either. If you see anything underhanded going on, let Nordon know immediately." "They won't get anything by me, Mr. Malloy." By the time the ship from Earth got there, the peace conference had been going on for four days. Bertrand Malloy had full reports on the whole parley, as relayed to him through the ship that had taken Nordon and Braynek to Saarkkad V. Secretary of State Blendwell stopped off at Saarkkad IV before going on to V to take charge of the conference. He was a tallish, lean man with a few strands of gray hair on the top of his otherwise bald scalp, and he wore a hearty, professional smile that didn't quite make it to his calculating eyes. He took Malloy's hand and shook it warmly. "How are you, Mr. Ambassador?" "Fine, Mr. Secretary. How's everything on Earth?" "Tense. They're waiting to see what is going to happen on Five. So am I, for that matter." His eyes were curious. "You decided not to go yourself, eh?" "I thought it better not to. I sent a good team, instead. Would you like to see the reports?" "I certainly would." Malloy handed them to the secretary, and as he read, Malloy watched him. Blendwell was a political appointee—a good man, Malloy had to admit, but he didn't know all the ins and outs of the Diplomatic Corps. When Blendwell looked up from the reports at last, he said: "Amazing! They've held off the Karna at every point! They've beaten them back! They've managed to cope with and outdo the finest team of negotiators the Karna could send." "I thought they would," said Malloy, trying to appear modest. The secretary's eyes narrowed. "I've heard of the work you've been doing here with ... ah ... sick men. Is this one of your ... ah ... successes?" Malloy nodded. "I think so. The Karna put us in a dilemma, so I threw a dilemma right back at them." "How do you mean?" "Nordon had a mental block against making decisions. If he took a girl out on a date, he'd have trouble making up his mind whether to kiss her or not until she made up his mind for him, one way or the other. He's that kind of guy. Until he's presented with one, single, clear decision which admits of no alternatives, he can't move at all. "As you can see, the Karna tried to give us several choices on each point, and they were all rigged. Until they backed down to a single point and proved that it wasn't rigged, Nordon couldn't possibly make up his mind. I drummed into him how important this was, and the more importance there is attached to his decisions, the more incapable he becomes of making them." The Secretary nodded slowly. "What about Braynek?" "Paranoid," said Malloy. "He thinks everyone is plotting against him. In this case, that's all to the good because the Karna are plotting against him. No matter what they put forth, Braynek is convinced that there's a trap in it somewhere, and he digs to find out what the trap is. Even if there isn't a trap, the Karna can't satisfy Braynek, because he's convinced that there has to be—somewhere. As a result, all his advice to Nordon, and all his questioning on the wildest possibilities, just serves to keep Nordon from getting unconfused. "These two men are honestly doing their best to win at the peace conference, and they've got the Karna reeling. The Karna can see that we're not trying to stall; our men are actually working at trying to reach a decision. But what the Karna don't see is that those men, as a team, are unbeatable because, in this situation, they're psychologically incapable of losing." Again the Secretary of State nodded his approval, but there was still a question in his mind. "Since you know all that, couldn't you have handled it yourself?" "Maybe, but I doubt it. They might have gotten around me someway by sneaking up on a blind spot. Nordon and Braynek have blind spots, but they're covered with armor. No, I'm glad I couldn't go; it's better this way." The Secretary of State raised an eyebrow. " Couldn't go, Mr. Ambassador?" Malloy looked at him. "Didn't you know? I wondered why you appointed me, in the first place. No, I couldn't go. The reason why I'm here, cooped up in this office, hiding from the Saarkkada the way a good Saarkkadic bigshot should, is because I like it that way. I suffer from agoraphobia and xenophobia. "I have to be drugged to be put on a spaceship because I can't take all that empty space, even if I'm protected from it by a steel shell." A look of revulsion came over his face. "And I can't stand aliens!" THE END Transcriber's Note: This etext was produced from Astounding Science Fiction March 1960. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
B. He was the closest diplomat available.
|
What is a significant irony in the successful colonization of the moon?
A. Earth needs materials from the moon to survive, while the moon needs materials from the Earth
B. The government is just as ineffective on the moon as it is on Earth
C. Moon inhabitants are less free on the moon than they used to be on Earth
D. The greed of humankind is destroying the newly colonized moon just as it is destroying Earth
|
ALL DAY SEPTEMBER By ROGER KUYKENDALL Illustrated by van Dongen [Transcriber's Note: This etext was produced from Astounding Science Fiction June 1959. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] Some men just haven't got good sense. They just can't seem to learn the most fundamental things. Like when there's no use trying—when it's time to give up because it's hopeless.... The meteor, a pebble, a little larger than a match head, traveled through space and time since it came into being. The light from the star that died when the meteor was created fell on Earth before the first lungfish ventured from the sea. In its last instant, the meteor fell on the Moon. It was impeded by Evans' tractor. It drilled a small, neat hole through the casing of the steam turbine, and volitized upon striking the blades. Portions of the turbine also volitized; idling at eight thousand RPM, it became unstable. The shaft tried to tie itself into a knot, and the blades, damaged and undamaged were spit through the casing. The turbine again reached a stable state, that is, stopped. Permanently stopped. It was two days to sunrise, where Evans stood. It was just before sunset on a spring evening in September in Sydney. The shadow line between day and night could be seen from the Moon to be drifting across Australia. Evans, who had no watch, thought of the time as a quarter after Australia. Evans was a prospector, and like all prospectors, a sort of jackknife geologist, selenologist, rather. His tractor and equipment cost two hundred and fifty thousand dollars. Fifty thousand was paid for. The rest was promissory notes and grubstake shares. When he was broke, which was usually, he used his tractor to haul uranium ore and metallic sodium from the mines at Potter's dike to Williamson Town, where the rockets landed. When he was flush, he would prospect for a couple of weeks. Once he followed a stampede to Yellow Crater, where he thought for a while that he had a fortune in chromium. The chromite petered out in a month and a half, and he was lucky to break even. Evans was about three hundred miles east of Williamson Town, the site of the first landing on the Moon. Evans was due back at Williamson Town at about sunset, that is, in about sixteen days. When he saw the wrecked turbine, he knew that he wouldn't make it. By careful rationing, he could probably stretch his food out to more than a month. His drinking water—kept separate from the water in the reactor—might conceivably last just as long. But his oxygen was too carefully measured; there was a four-day reserve. By diligent conservation, he might make it last an extra day. Four days reserve—plus one is five—plus sixteen days normal supply equals twenty-one days to live. In seventeen days he might be missed, but in seventeen days it would be dark again, and the search for him, if it ever began, could not begin for thirteen more days. At the earliest it would be eight days too late. "Well, man, 'tis a fine spot you're in now," he told himself. "Let's find out how bad it is indeed," he answered. He reached for the light switch and tried to turn it on. The switch was already in the "on" position. "Batteries must be dead," he told himself. "What batteries?" he asked. "There're no batteries in here, the power comes from the generator." "Why isn't the generator working, man?" he asked. He thought this one out carefully. The generator was not turned by the main turbine, but by a small reciprocating engine. The steam, however, came from the same boiler. And the boiler, of course, had emptied itself through the hole in the turbine. And the condenser, of course— "The condenser!" he shouted. He fumbled for a while, until he found a small flashlight. By the light of this, he reinspected the steam system, and found about three gallons of water frozen in the condenser. The condenser, like all condensers, was a device to convert steam into water, so that it could be reused in the boiler. This one had a tank and coils of tubing in the center of a curved reflector that was positioned to radiate the heat of the steam into the cold darkness of space. When the meteor pierced the turbine, the water in the condenser began to boil. This boiling lowered the temperature, and the condenser demonstrated its efficiency by quickly freezing the water in the tank. Evans sealed the turbine from the rest of the steam system by closing the shut-off valves. If there was any water in the boiler, it would operate the engine that drove the generator. The water would condense in the condenser, and with a little luck, melt the ice in there. Then, if the pump wasn't blocked by ice, it would return the water to the boiler. But there was no water in the boiler. Carefully he poured a cup of his drinking water into a pipe that led to the boiler, and resealed the pipe. He pulled on a knob marked "Nuclear Start/Safety Bypass." The water that he had poured into the boiler quickly turned into steam, and the steam turned the generator briefly. Evans watched the lights flicker and go out, and he guessed what the trouble was. "The water, man," he said, "there is not enough to melt the ice in the condenser." He opened the pipe again and poured nearly a half-gallon of water into the boiler. It was three days' supply of water, if it had been carefully used. It was one day's supply if used wastefully. It was ostentatious luxury for a man with a month's supply of water and twenty-one days to live. The generator started again, and the lights came on. They flickered as the boiler pressure began to fail, but the steam had melted some of the ice in the condenser, and the water pump began to function. "Well, man," he breathed, "there's a light to die by." The sun rose on Williamson Town at about the same time it rose on Evans. It was an incredibly brilliant disk in a black sky. The stars next to the sun shone as brightly as though there were no sun. They might have appeared to waver slightly, if they were behind outflung corona flares. If they did, no one noticed. No one looked toward the sun without dark filters. When Director McIlroy came into his office, he found it lighted by the rising sun. The light was a hot, brilliant white that seemed to pierce the darkest shadows of the room. He moved to the round window, screening his eyes from the light, and adjusted the polaroid shade to maximum density. The sun became an angry red brown, and the room was dark again. McIlroy decreased the density again until the room was comfortably lighted. The room felt stuffy, so he decided to leave the door to the inner office open. He felt a little guilty about this, because he had ordered that all doors in the survey building should remain closed except when someone was passing through them. This was to allow the air-conditioning system to function properly, and to prevent air loss in case of the highly improbable meteor damage. McIlroy thought that on the whole, he was disobeying his own orders no more flagrantly than anyone else in the survey. McIlroy had no illusions about his ability to lead men. Or rather, he did have one illusion; he thought that he was completely unfit as a leader. It was true that his strictest orders were disobeyed with cheerful contempt, but it was also true his mildest requests were complied with eagerly and smoothly. Everyone in the survey except McIlroy realized this, and even he accepted this without thinking about it. He had fallen into the habit of suggesting mildly anything that he wanted done, and writing orders he didn't particularly care to have obeyed. For example, because of an order of his stating that there would be no alcoholic beverages within the survey building, the entire survey was assured of a constant supply of home-made, but passably good liquor. Even McIlroy enjoyed the surreptitious drinking. "Good morning, Mr. McIlroy," said Mrs. Garth, his secretary. Morning to Mrs. Garth was simply the first four hours after waking. "Good morning indeed," answered McIlroy. Morning to him had no meaning at all, but he thought in the strictest sense that it would be morning on the Moon for another week. "Has the power crew set up the solar furnace?" he asked. The solar furnace was a rough parabola of mirrors used to focus the sun's heat on anything that it was desirable to heat. It was used mostly, from sun-up to sun-down, to supplement the nuclear power plant. "They went out about an hour ago," she answered, "I suppose that's what they were going to do." "Very good, what's first on the schedule?" "A Mr. Phelps to see you," she said. "How do you do, Mr. Phelps," McIlroy greeted him. "Good afternoon," Mr. Phelps replied. "I'm here representing the Merchants' Bank Association." "Fine," McIlroy said, "I suppose you're here to set up a bank." "That's right, I just got in from Muroc last night, and I've been going over the assets of the Survey Credit Association all morning." "I'll certainly be glad to get them off my hands," McIlroy said. "I hope they're in good order." "There doesn't seem to be any profit," Mr. Phelps said. "That's par for a nonprofit organization," said McIlroy. "But we're amateurs, and we're turning this operation over to professionals. I'm sure it will be to everyone's satisfaction." "I know this seems like a silly question. What day is this?" "Well," said McIlroy, "that's not so silly. I don't know either." "Mrs. Garth," he called, "what day is this?" "Why, September, I think," she answered. "I mean what day ." "I don't know, I'll call the observatory." There was a pause. "They say what day where?" she asked. "Greenwich, I guess, our official time is supposed to be Greenwich Mean Time." There was another pause. "They say it's September fourth, one thirty a.m. " "Well, there you are," laughed McIlroy, "it isn't that time doesn't mean anything here, it just doesn't mean the same thing." Mr. Phelps joined the laughter. "Bankers' hours don't mean much, at any rate," he said. The power crew was having trouble with the solar furnace. Three of the nine banks of mirrors would not respond to the electric controls, and one bank moved so jerkily that it could not be focused, and it threatened to tear several of the mirrors loose. "What happened here?" Spotty Cade, one of the electrical technicians asked his foreman, Cowalczk, over the intercommunications radio. "I've got about a hundred pinholes in the cables out here. It's no wonder they don't work." "Meteor shower," Cowalczk answered, "and that's not half of it. Walker says he's got a half dozen mirrors cracked or pitted, and Hoffman on bank three wants you to replace a servo motor. He says the bearing was hit." "When did it happen?" Cade wanted to know. "Must have been last night, at least two or three days ago. All of 'em too small for Radar to pick up, and not enough for Seismo to get a rumble." "Sounds pretty bad." "Could have been worse," said Cowalczk. "How's that?" "Wasn't anybody out in it." "Hey, Chuck," another technician, Lehman, broke in, "you could maybe get hurt that way." "I doubt it," Cowalczk answered, "most of these were pinhead size, and they wouldn't go through a suit." "It would take a pretty big one to damage a servo bearing," Cade commented. "That could hurt," Cowalczk admitted, "but there was only one of them." "You mean only one hit our gear," Lehman said. "How many missed?" Nobody answered. They could all see the Moon under their feet. Small craters overlapped and touched each other. There was—except in the places that men had obscured them with footprints—not a square foot that didn't contain a crater at least ten inches across, there was not a square inch without its half-inch crater. Nearly all of these had been made millions of years ago, but here and there, the rim of a crater covered part of a footprint, clear evidence that it was a recent one. After the sun rose, Evans returned to the lava cave that he had been exploring when the meteor hit. Inside, he lifted his filter visor, and found that the light reflected from the small ray that peered into the cave door lighted the cave adequately. He tapped loose some white crystals on the cave wall with his geologist's hammer, and put them into a collector's bag. "A few mineral specimens would give us something to think about, man. These crystals," he said, "look a little like zeolites, but that can't be, zeolites need water to form, and there's no water on the Moon." He chipped a number of other crystals loose and put them in bags. One of them he found in a dark crevice had a hexagonal shape that puzzled him. One at a time, back in the tractor, he took the crystals out of the bags and analyzed them as well as he could without using a flame which would waste oxygen. The ones that looked like zeolites were zeolites, all right, or something very much like it. One of the crystals that he thought was quartz turned out to be calcite, and one of the ones that he was sure could be nothing but calcite was actually potassium nitrate. "Well, now," he said, "it's probably the largest natural crystal of potassium nitrate that anyone has ever seen. Man, it's a full inch across." All of these needed water to form, and their existence on the Moon puzzled him for a while. Then he opened the bag that had contained the unusual hexagonal crystals, and the puzzle resolved itself. There was nothing in the bag but a few drops of water. What he had taken to be a type of rock was ice, frozen in a niche that had never been warmed by the sun. The sun rose to the meridian slowly. It was a week after sunrise. The stars shone coldly, and wheeled in their slow course with the sun. Only Earth remained in the same spot in the black sky. The shadow line crept around until Earth was nearly dark, and then the rim of light appeared on the opposite side. For a while Earth was a dark disk in a thin halo, and then the light came to be a crescent, and the line of dawn began to move around Earth. The continents drifted across the dark disk and into the crescent. The people on Earth saw the full moon set about the same time that the sun rose. Nickel Jones was the captain of a supply rocket. He made trips from and to the Moon about once a month, carrying supplies in and metal and ores out. At this time he was visiting with his old friend McIlroy. "I swear, Mac," said Jones, "another season like this, and I'm going back to mining." "I thought you were doing pretty well," said McIlroy, as he poured two drinks from a bottle of Scotch that Jones had brought him. "Oh, the money I like, but I will say that I'd have more if I didn't have to fight the union and the Lunar Trade Commission." McIlroy had heard all of this before. "How's that?" he asked politely. "You may think it's myself running the ship," Jones started on his tirade, "but it's not. The union it is that says who I can hire. The union it is that says how much I must pay, and how large a crew I need. And then the Commission ..." The word seemed to give Jones an unpleasant taste in his mouth, which he hurriedly rinsed with a sip of Scotch. "The Commission," he continued, making the word sound like an obscenity, "it is that tells me how much I can charge for freight." McIlroy noticed that his friend's glass was empty, and he quietly filled it again. "And then," continued Jones, "if I buy a cargo up here, the Commission it is that says what I'll sell it for. If I had my way, I'd charge only fifty cents a pound for freight instead of the dollar forty that the Commission insists on. That's from here to Earth, of course. There's no profit I could make by cutting rates the other way." "Why not?" asked McIlroy. He knew the answer, but he liked to listen to the slightly Welsh voice of Jones. "Near cost it is now at a dollar forty. But what sense is there in charging the same rate to go either way when it takes about a seventh of the fuel to get from here to Earth as it does to get from there to here?" "What good would it do to charge fifty cents a pound?" asked McIlroy. "The nickel, man, the tons of nickel worth a dollar and a half on Earth, and not worth mining here; the low-grade ores of uranium and vanadium, they need these things on Earth, but they can't get them as long as it isn't worth the carrying of them. And then, of course, there's the water we haven't got. We could afford to bring more water for more people, and set up more distilling plants if we had the money from the nickel. "Even though I say it who shouldn't, two-eighty a quart is too much to pay for water." Both men fell silent for a while. Then Jones spoke again: "Have you seen our friend Evans lately? The price of chromium has gone up, and I think he could ship some of his ore from Yellow Crater at a profit." "He's out prospecting again. I don't expect to see him until sun-down." "I'll likely see him then. I won't be loaded for another week and a half. Can't you get in touch with him by radio?" "He isn't carrying one. Most of the prospectors don't. They claim that a radio that won't carry beyond the horizon isn't any good, and one that will bounce messages from Earth takes up too much room." "Well, if I don't see him, you let him know about the chromium." "Anything to help another Welshman, is that the idea?" "Well, protection it is that a poor Welshman needs from all the English and Scots. Speaking of which—" "Oh, of course," McIlroy grinned as he refilled the glasses. " Slainte, McIlroy, bach. " [Health, McIlroy, man.] " Slainte mhor, bach. " [Great Health, man.] The sun was halfway to the horizon, and Earth was a crescent in the sky when Evans had quarried all the ice that was available in the cave. The thought grew on him as he worked that this couldn't be the only such cave in the area. There must be several more bubbles in the lava flow. Part of his reasoning proved correct. That is, he found that by chipping, he could locate small bubbles up to an inch in diameter, each one with its droplet of water. The average was about one per cent of the volume of each bubble filled with ice. A quarter of a mile from the tractor, Evans found a promising looking mound of lava. It was rounded on top, and it could easily be the dome of a bubble. Suddenly, Evans noticed that the gauge on the oxygen tank of his suit was reading dangerously near empty. He turned back to his tractor, moving as slowly as he felt safe in doing. Running would use up oxygen too fast. He was halfway there when the pressure warning light went on, and the signal sounded inside his helmet. He turned on his ten-minute reserve supply, and made it to the tractor with about five minutes left. The air purifying apparatus in the suit was not as efficient as the one in the tractor; it wasted oxygen. By using the suit so much, Evans had already shortened his life by several days. He resolved not to leave the tractor again, and reluctantly abandoned his plan to search for a large bubble. The sun stood at half its diameter above the horizon. The shadows of the mountains stretched out to touch the shadows of the other mountains. The dawning line of light covered half of Earth, and Earth turned beneath it. Cowalczk itched under his suit, and the sweat on his face prickled maddeningly because he couldn't reach it through his helmet. He pushed his forehead against the faceplate of his helmet and rubbed off some of the sweat. It didn't help much, and it left a blurred spot in his vision. That annoyed him. "Is everyone clear of the outlet?" he asked. "All clear," he heard Cade report through the intercom. "How come we have to blow the boilers now?" asked Lehman. "Because I say so," Cowalczk shouted, surprised at his outburst and ashamed of it. "Boiler scale," he continued, much calmer. "We've got to clean out the boilers once a year to make sure the tubes in the reactor don't clog up." He squinted through his dark visor at the reactor building, a gray concrete structure a quarter of a mile distant. "It would be pretty bad if they clogged up some night." "Pressure's ten and a half pounds," said Cade. "Right, let her go," said Cowalczk. Cade threw a switch. In the reactor building, a relay closed. A motor started turning, and the worm gear on the motor opened a valve on the boiler. A stream of muddy water gushed into a closed vat. When the vat was about half full, the water began to run nearly clear. An electric eye noted that fact and a light in front of Cade turned on. Cade threw the switch back the other way, and the relay in the reactor building opened. The motor turned and the gears started to close the valve. But a fragment of boiler scale held the valve open. "Valve's stuck," said Cade. "Open it and close it again," said Cowalczk. The sweat on his forehead started to run into his eyes. He banged his hand on his faceplate in an unconscious attempt to wipe it off. He cursed silently, and wiped it off on the inside of his helmet again. This time, two drops ran down the inside of his faceplate. "Still don't work," said Cade. "Keep trying," Cowalczk ordered. "Lehman, get a Geiger counter and come with me, we've got to fix this thing." Lehman and Cowalczk, who were already suited up started across to the reactor building. Cade, who was in the pressurized control room without a suit on, kept working the switch back and forth. There was light that indicated when the valve was open. It was on, and it stayed on, no matter what Cade did. "The vat pressure's too high," Cade said. "Let me know when it reaches six pounds," Cowalczk requested. "Because it'll probably blow at seven." The vat was a light plastic container used only to decant sludge out of the water. It neither needed nor had much strength. "Six now," said Cade. Cowalczk and Lehman stopped halfway to the reactor. The vat bulged and ruptured. A stream of mud gushed out and boiled dry on the face of the Moon. Cowalczk and Lehman rushed forward again. They could see the trickle of water from the discharge pipe. The motor turned the valve back and forth in response to Cade's signals. "What's going on out there?" demanded McIlroy on the intercom. "Scale stuck in the valve," Cowalczk answered. "Are the reactors off?" "Yes. Vat blew. Shut up! Let me work, Mac!" "Sorry," McIlroy said, realizing that this was no time for officials. "Let me know when it's fixed." "Geiger's off scale," Lehman said. "We're probably O.K. in these suits for an hour," Cowalczk answered. "Is there a manual shut-off?" "Not that I know of," Lehman answered. "What about it, Cade?" "I don't think so," Cade said. "I'll get on the blower and rouse out an engineer." "O.K., but keep working that switch." "I checked the line as far as it's safe," said Lehman. "No valve." "O.K.," Cowalczk said. "Listen, Cade, are the injectors still on?" "Yeah. There's still enough heat in these reactors to do some damage. I'll cut 'em in about fifteen minutes." "I've found the trouble," Lehman said. "The worm gear's loose on its shaft. It's slipping every time the valve closes. There's not enough power in it to crush the scale." "Right," Cowalczk said. "Cade, open the valve wide. Lehman, hand me that pipe wrench!" Cowalczk hit the shaft with the back of the pipe wrench, and it broke at the motor bearing. Cowalczk and Lehman fitted the pipe wrench to the gear on the valve, and turned it. "Is the light off?" Cowalczk asked. "No," Cade answered. "Water's stopped. Give us some pressure, we'll see if it holds." "Twenty pounds," Cade answered after a couple of minutes. "Take her up to ... no, wait, it's still leaking," Cowalczk said. "Hold it there, we'll open the valve again." "O.K.," said Cade. "An engineer here says there's no manual cutoff." "Like Hell," said Lehman. Cowalczk and Lehman opened the valve again. Water spurted out, and dwindled as they closed the valve. "What did you do?" asked Cade. "The light went out and came on again." "Check that circuit and see if it works," Cowalczk instructed. There was a pause. "It's O.K.," Cade said. Cowalczk and Lehman opened and closed the valve again. "Light is off now," Cade said. "Good," said Cowalczk, "take the pressure up all the way, and we'll see what happens." "Eight hundred pounds," Cade said, after a short wait. "Good enough," Cowalczk said. "Tell that engineer to hold up a while, he can fix this thing as soon as he gets parts. Come on, Lehman, let's get out of here." "Well, I'm glad that's over," said Cade. "You guys had me worried for a while." "Think we weren't worried?" Lehman asked. "And it's not over." "What?" Cade asked. "Oh, you mean the valve servo you two bashed up?" "No," said Lehman, "I mean the two thousand gallons of water that we lost." "Two thousand?" Cade asked. "We only had seven hundred gallons reserve. How come we can operate now?" "We picked up twelve hundred from the town sewage plant. What with using the solar furnace as a radiator, we can make do." "Oh, God, I suppose this means water rationing again." "You're probably right, at least until the next rocket lands in a couple of weeks." PROSPECTOR FEARED LOST ON MOON IPP Williamson Town, Moon, Sept. 21st. Scientific survey director McIlroy released a statement today that Howard Evans, a prospector is missing and presumed lost. Evans, who was apparently exploring the Moon in search of minerals was due two days ago, but it was presumed that he was merely temporarily delayed. Evans began his exploration on August 25th, and was known to be carrying several days reserve of oxygen and supplies. Director McIlroy has expressed a hope that Evans will be found before his oxygen runs out. Search parties have started from Williamson Town, but telescopic search from Palomar and the new satellite observatory are hindered by the fact that Evans is lost on the part of the Moon which is now dark. Little hope is held for radio contact with the missing man as it is believed he was carrying only short-range, intercommunications equipment. Nevertheless, receivers are ... Captain Nickel Jones was also expressing a hope: "Anyway, Mac," he was saying to McIlroy, "a Welshman knows when his luck's run out. And never a word did he say." "Like as not, you're right," McIlroy replied, "but if I know Evans, he'd never say a word about any forebodings." "Well, happen I might have a bit of Welsh second sight about me, and it tells me that Evans will be found." McIlroy chuckled for the first time in several days. "So that's the reason you didn't take off when you were scheduled," he said. "Well, yes," Jones answered. "I thought that it might happen that a rocket would be needed in the search." The light from Earth lighted the Moon as the Moon had never lighted Earth. The great blue globe of Earth, the only thing larger than the stars, wheeled silently in the sky. As it turned, the shadow of sunset crept across the face that could be seen from the Moon. From full Earth, as you might say, it moved toward last quarter. The rising sun shone into Director McIlroy's office. The hot light formed a circle on the wall opposite the window, and the light became more intense as the sun slowly pulled over the horizon. Mrs. Garth walked into the director's office, and saw the director sleeping with his head cradled in his arms on the desk. She walked softly to the window and adjusted the shade to darken the office. She stood looking at McIlroy for a moment, and when he moved slightly in his sleep, she walked softly out of the office. A few minutes later she was back with a cup of coffee. She placed it in front of the director, and shook his shoulder gently. "Wake up, Mr. McIlroy," she said, "you told me to wake you at sunrise, and there it is, and here's Mr. Phelps." McIlroy woke up slowly. He leaned back in his chair and stretched. His neck was stiff from sleeping in such an awkward position. "'Morning, Mr. Phelps," he said. "Good morning," Phelps answered, dropping tiredly into a chair. "Have some coffee, Mr. Phelps," said Mrs. Garth, handing him a cup. "Any news?" asked McIlroy. "About Evans?" Phelps shook his head slowly. "Palomar called in a few minutes back. Nothing to report and the sun was rising there. Australia will be in position pretty soon. Several observatories there. Then Capetown. There are lots of observatories in Europe, but most of them are clouded over. Anyway the satellite observatory will be in position by the time Europe is." McIlroy was fully awake. He glanced at Phelps and wondered how long it had been since he had slept last. More than that, McIlroy wondered why this banker, who had never met Evans, was losing so much sleep about finding him. It began to dawn on McIlroy that nearly the whole population of Williamson Town was involved, one way or another, in the search. The director turned to ask Phelps about this fact, but the banker was slumped in his chair, fast asleep with his coffee untouched. It was three hours later that McIlroy woke Phelps. "They've found the tractor," McIlroy said. "Good," Phelps mumbled, and then as comprehension came; "That's fine! That's just line! Is Evans—?" "Can't tell yet. They spotted the tractor from the satellite observatory. Captain Jones took off a few minutes ago, and he'll report back as soon as he lands. Hadn't you better get some sleep?" Evans was carrying a block of ice into the tractor when he saw the rocket coming in for a landing. He dropped the block and stood waiting. When the dust settled from around the tail of the rocket, he started to run forward. The air lock opened, and Evans recognized the vacuum suited figure of Nickel Jones. "Evans, man!" said Jones' voice in the intercom. "Alive you are!" "A Welshman takes a lot of killing," Evans answered. Later, in Evans' tractor, he was telling his story: "... And I don't know how long I sat there after I found the water." He looked at the Goldburgian device he had made out of wire and tubing. "Finally I built this thing. These caves were made of lava. They must have been formed by steam some time, because there's a floor of ice in all of 'em. "The idea didn't come all at once, it took a long time for me to remember that water is made out of oxygen and hydrogen. When I remembered that, of course, I remembered that it can be separated with electricity. So I built this thing. "It runs an electric current through water, lets the oxygen loose in the room, and pipes the hydrogen outside. It doesn't work automatically, of course, so I run it about an hour a day. My oxygen level gauge shows how long." "You're a genius, man!" Jones exclaimed. "No," Evans answered, "a Welshman, nothing more." "Well, then," said Jones, "are you ready to start back?" "Back?" "Well, it was to rescue you that I came." "I don't need rescuing, man," Evans said. Jones stared at him blankly. "You might let me have some food," Evans continued. "I'm getting short of that. And you might have someone send out a mechanic with parts to fix my tractor. Then maybe you'll let me use your radio to file my claim." "Claim?" "Sure, man, I've thousands of tons of water here. It's the richest mine on the Moon!" THE END
|
A. Earth needs materials from the moon to survive, while the moon needs materials from the Earth
|
How was reliability measured?
|
### Introduction
Social media are sometimes used to disseminate hateful messages. In Europe, the current surge in hate speech has been linked to the ongoing refugee crisis. Lawmakers and social media sites are increasingly aware of the problem and are developing approaches to deal with it, for example promising to remove illegal messages within 24 hours after they are reported BIBREF0 . This raises the question of how hate speech can be detected automatically. Such an automatic detection method could be used to scan the large amount of text generated on the internet for hateful content and report it to the relevant authorities. It would also make it easier for researchers to examine the diffusion of hateful content through social media on a large scale. From a natural language processing perspective, hate speech detection can be considered a classification task: given an utterance, determine whether or not it contains hate speech. Training a classifier requires a large amount of data that is unambiguously hate speech. This data is typically obtained by manually annotating a set of texts based on whether a certain element contains hate speech. The reliability of the human annotations is essential, both to ensure that the algorithm can accurately learn the characteristics of hate speech, and as an upper bound on the expected performance BIBREF1 , BIBREF2 . As a preliminary step, six annotators rated 469 tweets. We found that agreement was very low (see Section 3). We then carried out group discussions to find possible reasons. They revealed that there is considerable ambiguity in existing definitions. A given statement may be considered hate speech or not depending on someone's cultural background and personal sensibilities. The wording of the question may also play a role. We decided to investigate the issue of reliability further by conducting a more comprehensive study across a large number of annotators, which we present in this paper. Our contribution in this paper is threefold: ### Hate Speech
For the purpose of building a classifier, warner2012 define hate speech as “abusive speech targeting specific group characteristics, such as ethnic origin, religion, gender, or sexual orientation”. More recent approaches rely on lists of guidelines such as a tweet being hate speech if it “uses a sexist or racial slur” BIBREF2 . These approaches are similar in that they leave plenty of room for personal interpretation, since there may be differences in what is considered offensive. For instance, while the utterance “the refugees will live off our money” is clearly generalising and maybe unfair, it is unclear if this is already hate speech. More precise definitions from law are specific to certain jurisdictions and therefore do not capture all forms of offensive, hateful speech, see e.g. matsuda1993. In practice, social media services are using their own definitions which have been subject to adjustments over the years BIBREF3 . As of June 2016, Twitter bans hateful conduct. With the rise in popularity of social media, the presence of hate speech has grown on the internet. Posting a tweet takes little more than a working internet connection but may be seen by users all over the world. Along with the presence of hate speech, its real-life consequences are also growing. It can be a precursor and incentive for hate crimes, and it can be so severe that it can even be a health issue BIBREF4 . It is also known that hate speech does not only mirror existing opinions in the reader but can also induce new negative feelings towards its targets BIBREF5 . Hate speech has recently gained some interest as a research topic on the one hand – e.g. BIBREF6 , BIBREF4 , BIBREF7 – but also as a problem to deal with in politics such as the No Hate Speech Movement by the Council of Europe. The current refugee crisis has made it evident that governments, organisations and the public share an interest in controlling hate speech in social media. However, there seems to be little consensus on what hate speech actually is. ### Compiling A Hate Speech Corpus
As previously mentioned, there is no German hate speech corpus available for our needs, especially not for the very recent topic of the refugee crisis in Europe. We therefore had to compile our own corpus. We used Twitter as a source as it offers recent comments on current events. In our study we only considered the textual content of tweets that contain certain keywords, ignoring those that contain pictures or links. This section provides a detailed description of the approach we used to select the tweets and subsequently annotate them. To find a large amount of hate speech on the refugee crisis, we used 10 hashtags that can be used in an insulting or offensive way. Using these hashtags we gathered 13 766 tweets in total, roughly dating from February to March 2016. However, these tweets contained a lot of non-textual content which we filtered out automatically by removing tweets consisting solely of links or images. We also only considered original tweets, as retweets or replies to other tweets might only be clearly understandable when reading both tweets together. In addition, we removed duplicates and near-duplicates by discarding tweets that had a normalised Levenshtein edit distance smaller than .85 to an aforementioned tweet. A first inspection of the remaining tweets indicated that not all search terms were equally suited for our needs. The search term #Pack (vermin or lowlife) found a potentially large amount of hate speech not directly linked to the refugee crisis. It was therefore discarded. As a last step, the remaining tweets were manually read to eliminate those which were difficult to understand or incomprehensible. After these filtering steps, our corpus consists of 541 tweets, none of which are duplicates, contain links or pictures, or are retweets or replies. As a first measurement of the frequency of hate speech in our corpus, we personally annotated them based on our previous expertise. The 541 tweets were split into six parts and each part was annotated by two out of six annotators in order to determine if hate speech was present or not. The annotators were rotated so that each pair of annotators only evaluated one part. Additionally the offensiveness of a tweet was rated on a 6-point Likert scale, the same scale used later in the study. Even among researchers familiar with the definitions outlined above, there was still a low level of agreement (Krippendorff's INLINEFORM0 ). This supports our claim that a clearer definition is necessary in order to be able to train a reliable classifier. The low reliability could of course be explained by varying personal attitudes or backgrounds, but clearly needs more consideration. ### Methods
In order to assess the reliability of the hate speech definitions on social media more comprehensively, we developed two online surveys in a between-subjects design. They were completed by 56 participants in total (see Table TABREF7 ). The main goal was to examine the extent to which non-experts agree upon their understanding of hate speech given a diversity of social media content. We used the Twitter definition of hateful conduct in the first survey. This definition was presented at the beginning, and again above every tweet. The second survey did not contain any definition. Participants were randomly assigned one of the two surveys. The surveys consisted of 20 tweets presented in a random order. For each tweet, each participant was asked three questions. Depending on the survey, participants were asked (1) to answer (yes/no) if they considered the tweet hate speech, either based on the definition or based on their personal opinion. Afterwards they were asked (2) to answer (yes/no) if the tweet should be banned from Twitter. Participants were finally asked (3) to answer how offensive they thought the tweet was on a 6-point Likert scale from 1 (Not offensive at all) to 6 (Very offensive). If they answered 4 or higher, the participants had the option to state which particular words they found offensive. After the annotation of the 20 tweets, participants were asked to voluntarily answer an open question regarding the definition of hate speech. In the survey with the definition, they were asked if the definition of Twitter was sufficient. In the survey without the definition, the participants were asked to suggest a definition themselves. Finally, sociodemographic data were collected, including age, gender and more specific information regarding the participant's political orientation, migration background, and personal position regarding the refugee situation in Europe. The surveys were approved by the ethical committee of the Department of Computer Science and Applied Cognitive Science of the Faculty of Engineering at the University of Duisburg-Essen. ### Preliminary Results and Discussion
Since the surveys were completed by 56 participants, they resulted in 1120 annotations. Table TABREF7 shows some summary statistics. To assess whether the definition had any effect, we calculated, for each participant, the percentage of tweets they considered hate speech or suggested to ban and their mean offensiveness rating. This allowed us to compare the two samples for each of the three questions. Preliminary Shapiro-Wilk tests indicated that some of the data were not normally distributed. We therefore used the Wilcoxon-Mann-Whitney (WMW) test to compare the three pairs of series. The results are reported in Table TABREF7 . Participants who were shown the definition were more likely to suggest to ban the tweet. In fact, participants in group one very rarely gave different answers to questions one and two (18 of 500 instances or 3.6%). This suggests that participants in that group aligned their own opinion with the definition. We chose Krippendorff's INLINEFORM0 to assess reliability, a measure from content analysis, where human coders are required to be interchangeable. Therefore, it measures agreement instead of association, which leaves no room for the individual predilections of coders. It can be applied to any number of coders and to interval as well as nominal data. BIBREF8 This allowed us to compare agreement between both groups for all three questions. Figure FIGREF8 visualises the results. Overall, agreement was very low, ranging from INLINEFORM0 to INLINEFORM1 . In contrast, for the purpose of content analysis, Krippendorff recommends a minimum of INLINEFORM2 , or a minimum of INLINEFORM3 for applications where some uncertainty is unproblematic BIBREF8 . Reliability did not consistently increase when participants were shown a definition. To measure the extent to which the annotations using the Twitter definition (question one in group one) were in accordance with participants' opinions (question one in group two), we calculated, for each tweet, the percentage of participants in each group who considered it hate speech, and then calculated Pearson's correlation coefficient. The two series correlate strongly ( INLINEFORM0 ), indicating that they measure the same underlying construct. ### Conclusion and Future Work
This paper describes the creation of our hate speech corpus and offers first insights into the low agreement among users when it comes to identifying hateful messages. Our results imply that hate speech is a vague concept that requires significantly better definitions and guidelines in order to be annotated reliably. Based on the present findings, we are planning to develop a new coding scheme which includes clear-cut criteria that let people distinguish hate speech from other content. Researchers who are building a hate speech detection system might want to collect multiple labels for each tweet and average the results. Of course this approach does not make the original data any more reliable BIBREF8 . Yet, collecting the opinions of more users gives a more detailed picture of objective (or intersubjective) hatefulness. For the same reason, researchers might want to consider hate speech detection a regression problem, predicting, for example, the degree of hatefulness of a message, instead of a binary yes-or-no classification task. In the future, finding the characteristics that make users consider content hateful will be useful for building a model that automatically detects hate speech and users who spread hateful content, and for determining what makes users disseminate hateful content. ### Acknowledgments
This work was supported by the Deutsche Forschungsgemeinschaft (DFG) under grant No. GRK 2167, Research Training Group ”User-Centred Social Media”. Table 1: Summary statistics with p values and effect size estimates from WMW tests. Not all participants chose to report their age or gender. Figure 1: Reliability (Krippendorff’s a) for the different groups and questions
|
level of agreement (Krippendorff's INLINEFORM0 )
|
What incorrect assumption does Trella make about Blessing?
A. He would be thrilled to hear that Quest is alive and well
B. He murdered Dr. Mansard and got away with it
C. He turned Mansard's son into an android
D. He has no prior knowledge of the contents of Mansard's documents
|
Transcriber's Note: Every effort has been made to replicate this text as faithfully as possible; changes (corrections of spelling and punctuation) made to the original text are marked like this . The original text appears when hovering the cursor over the marked text. This e-text was produced from Amazing Science Fiction Stories March 1959. Extensive research did not uncover any evidence that the U. S. copyright on this publication was renewed. 50 THE JUPITER WEAPON By CHARLES L. FONTENAY He was a living weapon of destruction— immeasurably powerful, utterly invulnerable. There was only one question: Was he human? Trella feared she was in for trouble even before Motwick's head dropped forward on his arms in a drunken stupor. The two evil-looking men at the table nearby had been watching her surreptitiously, and now they shifted restlessly in their chairs. Trella had not wanted to come to the Golden Satellite. It was a squalid saloon in the rougher section of Jupiter's View, the terrestrial dome-colony on Ganymede. Motwick, already drunk, had insisted. A woman could not possibly make her way through these streets alone to the better section of town, especially one clad in a silvery evening dress. Her only hope was that this place had a telephone. Perhaps she could call one of Motwick's friends; she had no one on Ganymede she could call a real friend herself. Tentatively, she pushed her chair back from the table and arose. She had to brush close by the other table to get to the bar. As she did, the dark, slick-haired man reached out and grabbed her around the waist with a steely arm. Trella swung with her whole body, and slapped him so hard he nearly fell from his chair. As she walked swiftly toward the bar, he leaped up to follow her. There were only two other people in the Golden Satellite: the fat, mustached bartender and a short, square-built man at the bar. The latter swung around at the pistol-like report of her slap, and she saw that, though no more than four and a half feet tall, he was as heavily muscled as a lion. 51 His face was clean and open, with close-cropped blond hair and honest blue eyes. She ran to him. “Help me!” she cried. “Please help me!” He began to back away from her. “I can't,” he muttered in a deep voice. “I can't help you. I can't do anything.” The dark man was at her heels. In desperation, she dodged around the short man and took refuge behind him. Her protector was obviously unwilling, but the dark man, faced with his massiveness, took no chances. He stopped and shouted: “Kregg!” The other man at the table arose, ponderously, and lumbered toward them. He was immense, at least six and a half feet tall, with a brutal, vacant face. Evading her attempts to stay behind him, the squat man began to move down the bar away from the approaching Kregg. The dark man moved in on Trella again as Kregg overtook his quarry and swung a huge fist like a sledgehammer. Exactly what happened, Trella wasn't sure. She had the impression that Kregg's fist connected squarely with the short man's chin before he dodged to one side in a movement so fast it was a blur. But that couldn't have been, because the short man wasn't moved by that blow that would have felled a steer, and Kregg roared in pain, grabbing his injured fist. “The bar!” yelled Kregg. “I hit the damn bar!” At this juncture, the bartender took a hand. Leaning far over the bar, he swung a full bottle in a complete arc. It smashed on Kregg's head, splashing the floor with liquor, and Kregg sank stunned to his knees. The dark man, who had grabbed Trella's arm, released her and ran for the door. Moving agilely around the end of the bar, the bartender stood over Kregg, holding the jagged-edged bottleneck in his hand menacingly. “Get out!” rumbled the bartender. “I'll have no coppers raiding my place for the likes of you!” Kregg stumbled to his feet and staggered out. Trella ran to the unconscious Motwick's side. “That means you, too, lady,” said the bartender beside her. “You and your boy friend get out of here. You oughtn't to have come here in the first place.” “May I help you, Miss?” asked a deep, resonant voice behind her. She straightened from her anxious examination of Motwick. The squat man was standing there, an apologetic look on his face. She looked contemptuously at the massive muscles whose help had been denied her. Her arm ached where the dark man had grasped it. The broad face before 52 her was not unhandsome, and the blue eyes were disconcertingly direct, but she despised him for a coward. “I'm sorry I couldn't fight those men for you, Miss, but I just couldn't,” he said miserably, as though reading her thoughts. “But no one will bother you on the street if I'm with you.” “A lot of protection you'd be if they did!” she snapped. “But I'm desperate. You can carry him to the Stellar Hotel for me.” The gravity of Ganymede was hardly more than that of Earth's moon, but the way the man picked up the limp Motwick with one hand and tossed him over a shoulder was startling: as though he lifted a feather pillow. He followed Trella out the door of the Golden Satellite and fell in step beside her. Immediately she was grateful for his presence. The dimly lighted street was not crowded, but she didn't like the looks of the men she saw. The transparent dome of Jupiter's View was faintly visible in the reflected night lights of the colonial city, but the lights were overwhelmed by the giant, vari-colored disc of Jupiter itself, riding high in the sky. “I'm Quest Mansard, Miss,” said her companion. “I'm just in from Jupiter.” “I'm Trella Nuspar,” she said, favoring him with a green-eyed glance. “You mean Io, don't you—or Moon Five?” “No,” he said, grinning at her. He had an engaging grin, with even white teeth. “I meant Jupiter.” “You're lying,” she said flatly. “No one has ever landed on Jupiter. It would be impossible to blast off again.” “My parents landed on Jupiter, and I blasted off from it,” he said soberly. “I was born there. Have you ever heard of Dr. Eriklund Mansard?” “I certainly have,” she said, her interest taking a sudden upward turn. “He developed the surgiscope, didn't he? But his ship was drawn into Jupiter and lost.” “It was drawn into Jupiter, but he landed it successfully,” said Quest. “He and my mother lived on Jupiter until the oxygen equipment wore out at last. I was born and brought up there, and I was finally able to build a small rocket with a powerful enough drive to clear the planet.” She looked at him. He was short, half a head shorter than she, but broad and powerful as a man might be who had grown up in heavy gravity. He trod the street with a light, controlled step, seeming to deliberately hold himself down. “If Dr. Mansard succeeded in landing on Jupiter, why didn't anyone ever hear from him again?” she demanded. “Because,” said Quest, “his radio was sabotaged, just as his ship's drive was.” “Jupiter strength,” she murmured, looking him over coolly. 53 “You wear Motwick on your shoulder like a scarf. But you couldn't bring yourself to help a woman against two thugs.” He flushed. “I'm sorry,” he said. “That's something I couldn't help.” “Why not?” “I don't know. It's not that I'm afraid, but there's something in me that makes me back away from the prospect of fighting anyone.” Trella sighed. Cowardice was a state of mind. It was peculiarly inappropriate, but not unbelievable, that the strongest and most agile man on Ganymede should be a coward. Well, she thought with a rush of sympathy, he couldn't help being what he was. They had reached the more brightly lighted section of the city now. Trella could get a cab from here, but the Stellar Hotel wasn't far. They walked on. Trella had the desk clerk call a cab to deliver the unconscious Motwick to his home. She and Quest had a late sandwich in the coffee shop. “I landed here only a week ago,” he told her, his eyes frankly admiring her honey-colored hair and comely face. “I'm heading for Earth on the next spaceship.” “We'll be traveling companions, then,” she said. “I'm going back on that ship, too.” For some reason she decided against telling him that the assignment on which she had come to the Jupiter system was to gather his own father's notebooks and take them back to Earth. Motwick was an irresponsible playboy whom Trella had known briefly on Earth, and Trella was glad to dispense with his company for the remaining three weeks before the spaceship blasted off. She found herself enjoying the steadier companionship of Quest. As a matter of fact, she found herself enjoying his companionship more than she intended to. She found herself falling in love with him. Now this did not suit her at all. Trella had always liked her men tall and dark. She had determined that when she married it would be to a curly-haired six-footer. She was not at all happy about being so strongly attracted to a man several inches shorter than she. She was particularly unhappy about feeling drawn to a man who was a coward. The ship that they boarded on Moon Nine was one of the newer ships that could attain a hundred-mile-per-second velocity and take a hyperbolic path to Earth, but it would still require fifty-four days to make the trip. So Trella was delighted to find that the ship was the Cometfire and its skipper was her old friend, dark-eyed, curly-haired Jakdane Gille. “Jakdane,” she said, flirting with him with her eyes as in 54 days gone by, “I need a chaperon this trip, and you're ideal for the job.” “I never thought of myself in quite that light, but maybe I'm getting old,” he answered, laughing. “What's your trouble, Trella?” “I'm in love with that huge chunk of man who came aboard with me, and I'm not sure I ought to be,” she confessed. “I may need protection against myself till we get to Earth.” “If it's to keep you out of another fellow's clutches, I'm your man,” agreed Jakdane heartily. “I always had a mind to save you for myself. I'll guarantee you won't have a moment alone with him the whole trip.” “You don't have to be that thorough about it,” she protested hastily. “I want to get a little enjoyment out of being in love. But if I feel myself weakening too much, I'll holler for help.” The Cometfire swung around great Jupiter in an opening arc and plummeted ever more swiftly toward the tight circles of the inner planets. There were four crew members and three passengers aboard the ship's tiny personnel sphere, and Trella was thrown with Quest almost constantly. She enjoyed every minute of it. She told him only that she was a messenger, sent out to Ganymede to pick up some important papers and take them back to Earth. She was tempted to tell him what the papers were. Her employer had impressed upon her that her mission was confidential, but surely Dom Blessing could not object to Dr. Mansard's son knowing about it. All these things had happened before she was born, and she did not know what Dom Blessing's relation to Dr. Mansard had been, but it must have been very close. She knew that Dr. Mansard had invented the surgiscope. This was an instrument with a three-dimensional screen as its heart. The screen was a cubical frame in which an apparently solid image was built up of an object under an electron microscope. The actual cutting instrument of the surgiscope was an ion stream. By operating a tool in the three-dimensional screen, corresponding movements were made by the ion stream on the object under the microscope. The principle was the same as that used in operation of remote control “hands” in atomic laboratories to handle hot material, and with the surgiscope very delicate operations could be performed at the cellular level. Dr. Mansard and his wife had disappeared into the turbulent atmosphere of Jupiter just after his invention of the surgiscope, and it had been developed by Dom Blessing. Its success had built Spaceway Instruments, Incorporated, which Blessing headed. Through all these years since Dr. Mansard's disappearance, 55 Blessing had been searching the Jovian moons for a second, hidden laboratory of Dr. Mansard. When it was found at last, he sent Trella, his most trusted secretary, to Ganymede to bring back to him the notebooks found there. Blessing would, of course, be happy to learn that a son of Dr. Mansard lived, and would see that he received his rightful share of the inheritance. Because of this, Trella was tempted to tell Quest the good news herself; but she decided against it. It was Blessing's privilege to do this his own way, and he might not appreciate her meddling. At midtrip, Trella made a rueful confession to Jakdane. “It seems I was taking unnecessary precautions when I asked you to be a chaperon,” she said. “I kept waiting for Quest to do something, and when he didn't I told him I loved him.” “What did he say?” “It's very peculiar,” she said unhappily. “He said he can't love me. He said he wants to love me and he feels that he should, but there's something in him that refuses to permit it.” She expected Jakdane to salve her wounded feelings with a sympathetic pleasantry, but he did not. Instead, he just looked at her very thoughtfully and said no more about the matter. He explained his attitude after Asrange ran amuck. Asrange was the third passenger. He was a lean, saturnine individual who said little and kept to himself as much as possible. He was distantly polite in his relations with both crew and other passengers, and never showed the slightest spark of emotion … until the day Quest squirted coffee on him. It was one of those accidents that can occur easily in space. The passengers and the two crewmen on that particular waking shift (including Jakdane) were eating lunch on the center-deck. Quest picked up his bulb of coffee, but inadvertently pressed it before he got it to his lips. The coffee squirted all over the front of Asrange's clean white tunic. “I'm sorry!” exclaimed Quest in distress. The man's eyes went wide and he snarled. So quickly it seemed impossible, he had unbuckled himself from his seat and hurled himself backward from the table with an incoherent cry. He seized the first object his hand touched—it happened to be a heavy wooden cane leaning against Jakdane's bunk—propelled himself like a projectile at Quest. Quest rose from the table in a sudden uncoiling of movement. He did not unbuckle his safety belt—he rose and it snapped like a string. For a moment Trella thought he was going to meet Asrange's assault. But he fled in a long leap toward the companionway leading to the astrogation deck 56 above. Landing feet-first in the middle of the table and rebounding, Asrange pursued with the stick upraised. In his haste, Quest missed the companionway in his leap and was cornered against one of the bunks. Asrange descended on him like an avenging angel and, holding onto the bunk with one hand, rained savage blows on his head and shoulders with the heavy stick. Quest made no effort to retaliate. He cowered under the attack, holding his hands in front of him as if to ward it off. In a moment, Jakdane and the other crewman had reached Asrange and pulled him off. When they had Asrange in irons, Jakdane turned to Quest, who was now sitting unhappily at the table. “Take it easy,” he advised. “I'll wake the psychosurgeon and have him look you over. Just stay there.” Quest shook his head. “Don't bother him,” he said. “It's nothing but a few bruises.” “Bruises? Man, that club could have broken your skull! Or a couple of ribs, at the very least.” “I'm all right,” insisted Quest; and when the skeptical Jakdane insisted on examining him carefully, he had to admit it. There was hardly a mark on him from the blows. “If it didn't hurt you any more than that, why didn't you take that stick away from him?” demanded Jakdane. “You could have, easily.” “I couldn't,” said Quest miserably, and turned his face away. Later, alone with Trella on the control deck, Jakdane gave her some sober advice. “If you think you're in love with Quest, forget it,” he said. “Why? Because he's a coward? I know that ought to make me despise him, but it doesn't any more.” “Not because he's a coward. Because he's an android!” “What? Jakdane, you can't be serious!” “I am. I say he's an android, an artificial imitation of a man. It all figures. “Look, Trella, he said he was born on Jupiter. A human could stand the gravity of Jupiter, inside a dome or a ship, but what human could stand the rocket acceleration necessary to break free of Jupiter? Here's a man strong enough to break a spaceship safety belt just by getting up out of his chair against it, tough enough to take a beating with a heavy stick without being injured. How can you believe he's really human?” Trella remembered the thug Kregg striking Quest in the face and then crying that he had injured his hand on the bar. “But he said Dr. Mansard was his father,” protested Trella. “Robots and androids frequently look on their makers as their parents,” said Jakdane. “Quest may not even know he's 57 artificial. Do you know how Mansard died?” “The oxygen equipment failed, Quest said.” “Yes. Do you know when?” “No. Quest never did tell me, that I remember.” “He told me: a year before Quest made his rocket flight to Ganymede! If the oxygen equipment failed, how do you think Quest lived in the poisonous atmosphere of Jupiter, if he's human?” Trella was silent. “For the protection of humans, there are two psychological traits built into every robot and android,” said Jakdane gently. “The first is that they can never, under any circumstances, attack a human being, even in self defense. The second is that, while they may understand sexual desire objectively, they can never experience it themselves. “Those characteristics fit your man Quest to a T, Trella. There is no other explanation for him: he must be an android.” Trella did not want to believe Jakdane was right, but his reasoning was unassailable. Looking upon Quest as an android, many things were explained: his great strength, his short, broad build, his immunity to injury, his refusal to defend himself against a human, his inability to return Trella's love for him. It was not inconceivable that she should have unknowingly fallen in love with an android. Humans could love androids, with real affection, even knowing that they were artificial. There were instances of android nursemaids who were virtually members of the families owning them. She was glad now that she had not told Quest of her mission to Ganymede. He thought he was Dr. Mansard's son, but an android had no legal right of inheritance from his owner. She would leave it to Dom Blessing to decide what to do about Quest. Thus she did not, as she had intended originally, speak to Quest about seeing him again after she had completed her assignment. Even if Jakdane was wrong and Quest was human—as now seemed unlikely—Quest had told her he could not love her. Her best course was to try to forget him. Nor did Quest try to arrange with her for a later meeting. “It has been pleasant knowing you, Trella,” he said when they left the G-boat at White Sands. A faraway look came into his blue eyes, and he added: “I'm sorry things couldn't have been different, somehow.” “Let's don't be sorry for what we can't help,” she said gently, taking his hand in farewell. Trella took a fast plane from White Sands, and twenty-four hours later walked up the front steps of the familiar brownstone house on the outskirts of Washington. Dom Blessing himself met her at the door, a stooped, graying 58 man who peered at her over his spectacles. “You have the papers, eh?” he said, spying the brief case. “Good, good. Come in and we'll see what we have, eh?” She accompanied him through the bare, windowless anteroom which had always seemed to her such a strange feature of this luxurious house, and they entered the big living room. They sat before a fire in the old-fashioned fireplace and Blessing opened the brief case with trembling hands. “There are things here,” he said, his eyes sparkling as he glanced through the notebooks. “Yes, there are things here. We shall make something of these, Miss Trella, eh?” “I'm glad they're something you can use, Mr. Blessing,” she said. “There's something else I found on my trip, that I think I should tell you about.” She told him about Quest. “He thinks he's the son of Dr. Mansard,” she finished, “but apparently he is, without knowing it, an android Dr. Mansard built on Jupiter.” “He came back to Earth with you, eh?” asked Blessing intently. “Yes. I'm afraid it's your decision whether to let him go on living as a man or to tell him he's an android and claim ownership as Dr. Mansard's heir.” Trella planned to spend a few days resting in her employer's spacious home, and then to take a short vacation before resuming her duties as his confidential secretary. The next morning when she came down from her room, a change had been made. Two armed men were with Dom Blessing at breakfast and accompanied him wherever he went. She discovered that two more men with guns were stationed in the bare anteroom and a guard was stationed at every entrance to the house. “Why all the protection?” she asked Blessing. “A wealthy man must be careful,” said Blessing cheerfully. “When we don't understand all the implications of new circumstances, we must be prepared for anything, eh?” There was only one new circumstance Trella could think of. Without actually intending to, she exclaimed: “You aren't afraid of Quest? Why, an android can't hurt a human!” Blessing peered at her over his spectacles. “And what if he isn't an android, eh? And if he is—what if old Mansard didn't build in the prohibition against harming humans that's required by law? What about that, eh?” Trella was silent, shocked. There was something here she hadn't known about, hadn't even suspected. For some reason, Dom Blessing feared Dr. Eriklund Mansard … or his heir … or his mechanical servant. She was sure that Blessing was wrong, that Quest, whether man or android, intended no 59 harm to him. Surely, Quest would have said something of such bitterness during their long time together on Ganymede and aspace, since he did not know of Trella's connection with Blessing. But, since this was to be the atmosphere of Blessing's house, she was glad that he decided to assign her to take the Mansard papers to the New York laboratory. Quest came the day before she was scheduled to leave. Trella was in the living room with Blessing, discussing the instructions she was to give to the laboratory officials in New York. The two bodyguards were with them. The other guards were at their posts. Trella heard the doorbell ring. The heavy oaken front door was kept locked now, and the guards in the anteroom examined callers through a tiny window. Suddenly alarm bells rang all over the house. There was a terrific crash outside the room as the front door splintered. There were shouts and the sound of a shot. “The steel doors!” cried Blessing, turning white. “Let's get out of here.” He and his bodyguards ran through the back of the house out of the garage. Blessing, ahead of the rest, leaped into one of the cars and started the engine. The door from the house shattered and Quest burst through. The two guards turned and fired together. He could be hurt by bullets. He was staggered momentarily. Then, in a blur of motion, he sprang forward and swept the guards aside with one hand with such force that they skidded across the floor and lay in an unconscious heap against the rear of the garage. Trella had opened the door of the car, but it was wrenched from her hand as Blessing stepped on the accelerator and it leaped into the driveway with spinning wheels. Quest was after it, like a chunky deer, running faster than Trella had ever seen a man run before. Blessing slowed for the turn at the end of the driveway and glanced back over his shoulder. Seeing Quest almost upon him, he slammed down the accelerator and twisted the wheel hard. The car whipped into the street, careened, and rolled over and over, bringing up against a tree on the other side in a twisted tangle of wreckage. With a horrified gasp, Trella ran down the driveway toward the smoking heap of metal. Quest was already beside it, probing it. As she reached his side, he lifted the torn body of Dom Blessing. Blessing was dead. “I'm lucky,” said Quest soberly. “I would have murdered him.” “But why, Quest? I knew he was afraid of you, but he didn't tell me why.” “It was conditioned into me,” answered Quest “I didn't know 60 it until just now, when it ended, but my father conditioned me psychologically from my birth to the task of hunting down Dom Blessing and killing him. It was an unconscious drive in me that wouldn't release me until the task was finished. “You see, Blessing was my father's assistant on Ganymede. Right after my father completed development of the surgiscope, he and my mother blasted off for Io. Blessing wanted the valuable rights to the surgiscope, and he sabotaged the ship's drive so it would fall into Jupiter. “But my father was able to control it in the heavy atmosphere of Jupiter, and landed it successfully. I was born there, and he conditioned me to come to Earth and track down Blessing. I know now that it was part of the conditioning that I was unable to fight any other man until my task was finished: it might have gotten me in trouble and diverted me from that purpose.” More gently than Trella would have believed possible for his Jupiter-strong muscles, Quest took her in his arms. “Now I can say I love you,” he said. “That was part of the conditioning too: I couldn't love any woman until my job was done.” Trella disengaged herself. “I'm sorry,” she said. “Don't you know this, too, now: that you're not a man, but an android?” He looked at her in astonishment, stunned by her words. “What in space makes you think that?” he demanded. “Why, Quest, it's obvious,” she cried, tears in her eyes. “Everything about you … your build, suited for Jupiter's gravity … your strength … the fact that you were able to live in Jupiter's atmosphere after the oxygen equipment failed. I know you think Dr. Mansard was your father, but androids often believe that.” He grinned at her. “I'm no android,” he said confidently. “Do you forget my father was inventor of the surgiscope? He knew I'd have to grow up on Jupiter, and he operated on the genes before I was born. He altered my inherited characteristics to adapt me to the climate of Jupiter … even to being able to breathe a chlorine atmosphere as well as an oxygen atmosphere.” Trella looked at him. He was not badly hurt, any more than an elephant would have been, but his tunic was stained with red blood where the bullets had struck him. Normal android blood was green. “How can you be sure?” she asked doubtfully. “Androids are made,” he answered with a laugh. “They don't grow up. And I remember my boyhood on Jupiter very well.” He took her in his arms again, and this time she did not resist. His lips were very human. THE END
|
A. He would be thrilled to hear that Quest is alive and well
|
Why do the gangs pick Halloween night to fight?
A. The schoolyard would be empty as kids would be out.
B. They could be out past curfew without suspicion. No one would question why kids were going out on Halloween night.
C. The cops would be preoccupied with other matters, and it was easy to explain why you had a weapon on you.
D. The cops wouldn't be on lookout on a night like Halloween, so they can get away with doing what they want.
|
CALL HIM NEMESIS By DONALD E. WESTLAKE Criminals, beware; the Scorpion is on your trail! Hoodlums fear his fury—and, for that matter, so do the cops! [Transcriber's Note: This etext was produced from Worlds of If Science Fiction, September 1961. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The man with the handkerchief mask said, "All right, everybody, keep tight. This is a holdup." There were twelve people in the bank. There was Mr. Featherhall at his desk, refusing to okay a personal check from a perfect stranger. There was the perfect stranger, an itinerant garage mechanic named Rodney (Rod) Strom, like the check said. There were Miss English and Miss Philicoff, the girls in the gilded teller cages. There was Mister Anderson, the guard, dozing by the door in his brown uniform. There was Mrs. Elizabeth Clayhorn, depositing her husband's pay check in their joint checking account, and with her was her ten-year-old son Edward (Eddie) Clayhorn, Junior. There was Charlie Casale, getting ten dollars dimes, six dollars nickels and four dollars pennies for his father in the grocery store down the street. There was Mrs. Dolly Daniels, withdrawing money from her savings account again. And there were three bank robbers. The three bank robbers looked like triplets. From the ground up, they all wore scuffy black shoes, baggy-kneed and unpressed khaki trousers, brown cracked-leather jackets over flannel shirts, white handkerchiefs over the lower half of their faces and gray-and-white check caps pulled low over their eyes. The eyes themselves looked dangerous. The man who had spoken withdrew a small but mean-looking thirty-two calibre pistol from his jacket pocket. He waved it menacingly. One of the others took the pistol away from Mister Anderson, the guard, and said to him in a low voice, "Think about retirement, my friend." The third one, who carried a black satchel like a doctor's bag, walked quickly around behind the teller's counter and started filling it with money. It was just like the movies. The man who had first spoken herded the tellers, Mr. Featherhall and the customers all over against the back wall, while the second man stayed next to Mr. Anderson and the door. The third man stuffed money into the black satchel. The man by the door said, "Hurry up." The man with the satchel said, "One more drawer." The man with the gun turned to say to the man at the door, "Keep your shirt on." That was all Miss English needed. She kicked off her shoes and ran pelting in her stocking feet for the door. The man by the door spread his arms out and shouted, "Hey!" The man with the gun swung violently back, cursing, and fired the gun. But he'd been moving too fast, and so had Miss English, and all he hit was the brass plate on Mr. Featherhall's desk. The man by the door caught Miss English in a bear hug. She promptly did her best to scratch his eyes out. Meanwhile, Mr. Anderson went scooting out the front door and running down the street toward the police station in the next block, shouting, "Help! Help! Robbery!" The man with the gun cursed some more. The man with the satchel came running around from behind the counter, and the man by the door tried to keep Miss English from scratching his eyes out. Then the man with the gun hit Miss English on the head. She fell unconscious to the floor, and all three of them ran out of the bank to the car out front, in which sat a very nervous-looking fourth man, gunning the engine. Everyone except Miss English ran out after the bandits, to watch. Things got very fast and very confused then. Two police cars came driving down the block and a half from the precinct house to the bank, and the car with the four robbers in it lurched away from the curb and drove straight down the street toward the police station. The police cars and the getaway car passed one another, with everybody shooting like the ships in pirate movies. There was so much confusion that it looked as though the bank robbers were going to get away after all. The police cars were aiming the wrong way and, as they'd come down with sirens wailing, there was a clear path behind them. Then, after the getaway car had gone more than two blocks, it suddenly started jouncing around. It smacked into a parked car and stopped. And all the police went running down there to clap handcuffs on the robbers when they crawled dazedly out of their car. "Hey," said Eddie Clayhorn, ten years old. "Hey, that was something, huh, Mom?" "Come along home," said his mother, grabbing his hand. "We don't want to be involved." "It was the nuttiest thing," said Detective-Sergeant Stevenson. "An operation planned that well, you'd think they'd pay attention to their getaway car, you know what I mean?" Detective-Sergeant Pauling shrugged. "They always slip up," he said. "Sooner or later, on some minor detail, they always slip up." "Yes, but their tires ." "Well," said Pauling, "it was a stolen car. I suppose they just grabbed whatever was handiest." "What I can't figure out," said Stevenson, "is exactly what made those tires do that. I mean, it was a hot day and all, but it wasn't that hot. And they weren't going that fast. I don't think you could go fast enough to melt your tires down." Pauling shrugged again. "We got them. That's the important thing." "Still and all, it's nutty. They're free and clear, barrelling out Rockaway toward the Belt, and all at once their tires melt, the tubes blow out and there they are." Stevenson shook his head. "I can't figure it." "Don't look a gift horse in the mouth," suggested Pauling. "They picked the wrong car to steal." "And that doesn't make sense, either," said Stevenson. "Why steal a car that could be identified as easily as that one?" "Why? What was it, a foreign make?" "No, it was a Chevvy, two-tone, three years old, looked just like half the cars on the streets. Except that in the trunk lid the owner had burned in 'The Scorpion' in big black letters you could see half a block away." "Maybe they didn't notice it when they stole the car," said Pauling. "For a well-planned operation like this one," said Stevenson, "they made a couple of really idiotic boners. It doesn't make any sense." "What do they have to say about it?" Pauling demanded. "Nothing, what do you expect? They'll make no statement at all." The squad-room door opened, and a uniformed patrolman stuck his head in. "The owner of that Chevvy's here," he said. "Right," said Stevenson. He followed the patrolman down the hall to the front desk. The owner of the Chevvy was an angry-looking man of middle age, tall and paunchy. "John Hastings," he said. "They say you have my car here." "I believe so, yes," said Stevenson. "I'm afraid it's in pretty bad shape." "So I was told over the phone," said Hastings grimly. "I've contacted my insurance company." "Good. The car's in the police garage, around the corner. If you'd come with me?" On the way around, Stevenson said, "I believe you reported the car stolen almost immediately after it happened." "That's right," said Hastings. "I stepped into a bar on my route. I'm a wine and liquor salesman. When I came out five minutes later, my car was gone." "You left the keys in it?" "Well, why not?" demanded Hastings belligerently. "If I'm making just a quick stop—I never spend more than five minutes with any one customer—I always leave the keys in the car. Why not?" "The car was stolen," Stevenson reminded him. Hastings grumbled and glared. "It's always been perfectly safe up till now." "Yes, sir. In here." Hastings took one look at his car and hit the ceiling. "It's ruined!" he cried. "What did you do to the tires?" "Not a thing, sir. That happened to them in the holdup." Hastings leaned down over one of the front tires. "Look at that! There's melted rubber all over the rims. Those rims are ruined! What did you use, incendiary bullets?" Stevenson shook his head. "No, sir. When that happened they were two blocks away from the nearest policeman." "Hmph." Hastings moved on around the car, stopping short to exclaim, "What in the name of God is that? You didn't tell me a bunch of kids had stolen the car." "It wasn't a bunch of kids," Stevenson told him. "It was four professional criminals, I thought you knew that. They were using it in a bank holdup." "Then why did they do that ?" Stevenson followed Hastings' pointing finger, and saw again the crudely-lettered words, "The Scorpion" burned black into the paint of the trunk lid. "I really don't know," he said. "It wasn't there before the car was stolen?" "Of course not!" Stevenson frowned. "Now, why in the world did they do that?" "I suggest," said Hastings with heavy sarcasm, "you ask them that." Stevenson shook his head. "It wouldn't do any good. They aren't talking about anything. I don't suppose they'll ever tell us." He looked at the trunk lid again. "It's the nuttiest thing," he said thoughtfully.... That was on Wednesday. The Friday afternoon mail delivery to the Daily News brought a crank letter. It was in the crank letter's most obvious form; that is, the address had been clipped, a letter or a word at a time, from a newspaper and glued to the envelope. There was no return address. The letter itself was in the same format. It was brief and to the point: Dear Mr. Editor: The Scorpion has struck. The bank robbers were captured. The Scorpion fights crime. Crooks and robbers are not safe from the avenging Scorpion. WARN YOUR READERS! Sincerely yours, THE SCORPION The warning was duly noted, and the letter filed in the wastebasket. It didn't rate a line in the paper. II The bank robbery occurred in late June. Early in August, a Brooklyn man went berserk. It happened in Canarsie, a section in southeast Brooklyn near Jamaica Bay. This particular area of Canarsie was a residential neighborhood, composed of one and two family houses. The man who went berserk was a Motor Vehicle Bureau clerk named Jerome Higgins. Two days before, he had flunked a Civil Service examination for the third time. He reported himself sick and spent the two days at home, brooding, a bottle of blended whiskey at all times in his hand. As the police reconstructed it later, Mrs. Higgins had attempted to awaken him on the third morning at seven-thirty, suggesting that he really ought to stop being so foolish, and go back to work. He then allegedly poked her in the eye, and locked her out of the bedroom. Mrs. Higgins then apparently called her sister-in-law, a Mrs. Thelma Stodbetter, who was Mr. Higgins' sister. Mrs. Stodbetter arrived at the house at nine o'clock, and spent some time tapping at the still-locked bedroom door, apparently requesting Mr. Higgins to unlock the door and "stop acting like a child." Neighbors reported to the police that they heard Mr. Higgins shout a number of times, "Go away! Can't you let a man sleep?" At about ten-fifteen, neighbors heard shots from the Higgins residence, a two-story one-family pink stucco affair in the middle of a block of similar homes. Mr. Higgins, it was learned later, had suddenly erupted from his bedroom, brandishing a .30-.30 hunting rifle and, being annoyed at the shrieks of his wife and sister, had fired seven shells at them, killing his wife on the spot and wounding his sister in the hand and shoulder. Mrs. Stodbetter, wounded and scared out of her wits, raced screaming out the front door of the house, crying for the police and shouting, "Murder! Murder!" At this point, neighbors called the police. One neighbor additionally phoned three newspapers and two television stations, thereby earning forty dollars in "news-tips" rewards. By chance, a mobile television unit was at that moment on the Belt Parkway, returning from having seen off a prime minister at Idlewild Airport. This unit was at once diverted to Canarsie, where it took up a position across the street from the scene of carnage and went to work with a Zoomar lens. In the meantime, Mister Higgins had barricaded himself in his house, firing at anything that moved. The two cameramen in the mobile unit worked their hearts out. One concentrated on the movements of the police and firemen and neighbors and ambulance attendants, while the other used the Zoomar lens to search for Mr. Higgins. He found him occasionally, offering the at-home audience brief glimpses of a stocky balding man in brown trousers and undershirt, stalking from window to window on the second floor of the house. The show lasted for nearly an hour. There were policemen everywhere, and firemen everywhere, and neighbors milling around down at the corner, where the police had roped the block off, and occasionally Mr. Higgins would stick his rifle out a window and shoot at somebody. The police used loudspeakers to tell Higgins he might as well give up, they had the place surrounded and could eventually starve him out anyway. Higgins used his own good lungs to shout obscenities back and challenge anyone present to hand-to-hand combat. The police fired tear gas shells at the house, but it was a windy day and all the windows in the Higgins house were either open or broken. Higgins was able to throw all the shells back out of the house again. The show lasted for nearly an hour. Then it ended, suddenly and dramatically. Higgins had showed himself to the Zoomar lens again, for the purpose of shooting either the camera or its operator. All at once he yelped and threw the rifle away. The rifle bounced onto the porch roof, slithered down to the edge, hung for a second against the drain, and finally fell barrel first onto the lawn. Meanwhile, Higgins was running through the house, shouting like a wounded bull. He thundered down the stairs and out, hollering, to fall into the arms of the waiting police. They had trouble holding him. At first they thought he was actually trying to get away, but then one of them heard what it was he was shouting: "My hands! My hands!" They looked at his hands. The palms and the palm-side of the fingers were red and blistering, from what looked like severe burns. There was another burn on his right cheek and another one on his right shoulder. Higgins, thoroughly chastened and bewildered, was led away for burn ointment and jail. The television crew went on back to Manhattan. The neighbors went home and telephoned their friends. On-duty policemen had been called in from practically all of the precincts in Brooklyn. Among them was Detective-Sergeant William Stevenson. Stevenson frowned thoughtfully at Higgins as that unhappy individual was led away, and then strolled over to look at the rifle. He touched the stock, and it was somewhat warm but that was all. He picked it up and turned it around. There, on the other side of the stock, burned into the wood, were the crudely-shaped letters, "The Scorpion." You don't get to be Precinct Captain on nothing but political connections. Those help, of course, but you need more than that. As Captain Hanks was fond of pointing out, you needed as well to be both more imaginative than most—"You gotta be able to second-guess the smart boys"—and to be a complete realist—"You gotta have both feet on the ground." If these were somewhat contradictory qualities, it was best not to mention the fact to Captain Hanks. The realist side of the captain's nature was currently at the fore. "Just what are you trying to say, Stevenson?" he demanded. "I'm not sure," admitted Stevenson. "But we've got these two things. First, there's the getaway car from that bank job. The wheels melt for no reason at all, and somebody burns 'The Scorpion' onto the trunk. Then, yesterday, this guy Higgins out in Canarsie. He says the rifle all of a sudden got too hot to hold, and he's got the burn marks to prove it. And there on the rifle stock it is again. 'The Scorpion'." "He says he put that on there himself," said the captain. Stevenson shook his head. "His lawyer says he put it on there. Higgins says he doesn't remember doing it. That's half the lawyer's case. He's trying to build up an insanity defense." "He put it on there himself, Stevenson," said the captain with weary patience. "What are you trying to prove?" "I don't know. All I know is it's the nuttiest thing I ever saw. And what about the getaway car? What about those tires melting?" "They were defective," said Hanks promptly. "All four of them at once? And what about the thing written on the trunk?" "How do I know?" demanded the captain. "Kids put it on before the car was stolen, maybe. Or maybe the hoods did it themselves, who knows? What do they say?" "They say they didn't do it," said Stevenson. "And they say they never saw it before the robbery and they would have noticed it if it'd been there." The captain shook his head. "I don't get it," he admitted. "What are you trying to prove?" "I guess," said Stevenson slowly, thinking it out as he went along, "I guess I'm trying to prove that somebody melted those tires, and made that rifle too hot, and left his signature behind." "What? You mean like in the comic books? Come on, Stevenson! What are you trying to hand me?" "All I know," insisted Stevenson, "is what I see." "And all I know," the captain told him, "is Higgins put that name on his rifle himself. He says so." "And what made it so hot?" "Hell, man, he'd been firing that thing at people for an hour! What do you think made it hot?" "All of a sudden?" "He noticed it all of a sudden, when it started to burn him." "How come the same name showed up each time, then?" Stevenson asked desperately. "How should I know? And why not, anyway? You know as well as I do these things happen. A bunch of teen-agers burgle a liquor store and they write 'The Golden Avengers' on the plate glass in lipstick. It happens all the time. Why not 'The Scorpion'? It couldn't occur to two people?" "But there's no explanation—" started Stevenson. "What do you mean, there's no explanation? I just gave you the explanation. Look, Stevenson, I'm a busy man. You got a nutty idea—like Wilcox a few years ago, remember him? Got the idea there was a fiend around loose, stuffing all those kids into abandoned refrigerators to starve. He went around trying to prove it, and getting all upset, and pretty soon they had to put him away in the nut hatch. Remember?" "I remember," said Stevenson. "Forget this silly stuff, Stevenson," the captain advised him. "Yes, sir," said Stevenson.... The day after Jerome Higgins went berserk, the afternoon mail brought a crank letter to the Daily News : Dear Mr. Editor, You did not warn your readers. The man who shot all those people could not escape the Scorpion. The Scorpion fights crime. No criminal is safe from the Scorpion. WARN YOUR READERS. Sincerely yours, THE SCORPION Unfortunately, this letter was not read by the same individual who had seen the first one, two months before. At any rate, it was filed in the same place, and forgotten. III Hallowe'en is a good time for a rumble. There's too many kids around for the cops to keep track of all of them, and if you're picked up carrying a knife or a length of tire chain or something, why, you're on your way to a Hallowe'en party and you're in costume. You're going as a JD. The problem was this schoolyard. It was a block wide, with entrances on two streets. The street on the north was Challenger territory, and the street on the south was Scarlet Raider territory, and both sides claimed the schoolyard. There had been a few skirmishes, a few guys from both gangs had been jumped and knocked around a little, but that had been all. Finally, the War Lords from the two gangs had met, and determined that the matter could only be settled in a war. The time was chosen: Hallowe'en. The place was chosen: the schoolyard. The weapons were chosen: pocket knives and tire chains okay, but no pistols or zip-guns. The time was fixed: eleven P.M. And the winner would have undisputed territorial rights to the schoolyard, both entrances. The night of the rumble, the gangs assembled in their separate clubrooms for last-minute instructions. Debs were sent out to play chicken at the intersections nearest the schoolyard, both to warn of the approach of cops and to keep out any non-combatant kids who might come wandering through. Judy Canzanetti was a Deb with the Scarlet Raiders. She was fifteen years old, short and black-haired and pretty in a movie-magazine, gum-chewing sort of way. She was proud of being in the Auxiliary of the Scarlet Raiders, and proud also of the job that had been assigned to her. She was to stand chicken on the southwest corner of the street. Judy took up her position at five minutes to eleven. The streets were dark and quiet. Few people cared to walk this neighborhood after dark, particularly on Hallowe'en. Judy leaned her back against the telephone pole on the corner, stuck her hands in the pockets of her Scarlet Raider jacket and waited. At eleven o'clock, she heard indistinct noises begin behind her. The rumble had started. At five after eleven, a bunch of little kids came wandering down the street. They were all about ten or eleven years old, and most of them carried trick-or-treat shopping bags. Some of them had Hallowe'en masks on. They started to make the turn toward the schoolyard. Judy said, "Hey, you kids. Take off." One of them, wearing a red mask, turned to look at her. "Who, us?" "Yes, you! Stay out of that street. Go on down that way." "The subway's this way," objected the kid in the red mask. "Who cares? You go around the other way." "Listen, lady," said the kid in the red mask, aggrieved, "we got a long way to go to get home." "Yeah," said another kid, in a black mask, "and we're late as it is." "I couldn't care less," Judy told them callously. "You can't go down that street." "Why not?" demanded yet another kid. This one was in the most complete and elaborate costume of them all, black leotards and a yellow shirt and a flowing: black cape. He wore a black and gold mask and had a black knit cap jammed down tight onto his head. "Why can't we go down there?" this apparition demanded. "Because I said so," Judy told him. "Now, you kids get away from here. Take off." "Hey!" cried the kid in the black-and-yellow costume. "Hey, they're fighting down there!" "It's a rumble," said Judy proudly. "You twerps don't want to be involved." "Hey!" cried the kid in the black-and-yellow costume again. And he went running around Judy and dashing off down the street. "Hey, Eddie!" shouted one of the other kids. "Eddie, come back!" Judy wasn't sure what to do next. If she abandoned her post to chase the one kid who'd gotten through, then maybe all the rest of them would come running along after her. She didn't know what to do. A sudden siren and a distant flashing red light solved her problems. "Cheez," said one of the kids. "The cops!" "Fuzz!" screamed Judy. She turned and raced down the block toward the schoolyard, shouting, "Fuzz! Fuzz! Clear out, it's the fuzz!" But then she stopped, wide-eyed, when she saw what was going on in the schoolyard. The guys from both gangs were dancing. They were jumping around, waving their arms, throwing their weapons away. Then they all started pulling off their gang jackets and throwing them away, whooping and hollering. They were making such a racket themselves that they never heard Judy's warning. They didn't even hear the police sirens. And all at once both schoolyard entrances were full of cops, a cop had tight hold of Judy and the rumble was over. Judy was so baffled and terrified that everything was just one great big blur. But in the middle of it all, she did see the little kid in the yellow-and-black costume go scooting away down the street. And she had the craziest idea that it was all his fault. Captain Hanks was still in his realistic cycle this morning, and he was impatient as well. "All right, Stevenson," he said. "Make it fast, I've got a lot to do this morning. And I hope it isn't this comic-book thing of yours again." "I'm afraid it is, Captain," said Stevenson. "Did you see the morning paper?" "So what?" "Did you see that thing about the gang fight up in Manhattan?" Captain Hanks sighed. "Stevenson," he said wearily, "are you going to try to connect every single time the word 'scorpion' comes up? What's the problem with this one? These kid gangs have names, so what?" "Neither one of them was called 'The Scorpions,'" Stevenson told him. "One of them was the Scarlet Raiders and the other gang was the Challengers." "So they changed their name," said Hanks. "Both gangs? Simultaneously? To the same name?" "Why not? Maybe that's what they were fighting over." "It was a territorial war," Stevenson reminded him. "They've admitted that much. It says so in the paper. And it also says they all deny ever seeing that word on their jackets until after the fight." "A bunch of juvenile delinquents," said Hanks in disgust. "You take their word?" "Captain, did you read the article in the paper?" "I glanced through it." "All right. Here's what they say happened: They say they started fighting at eleven o'clock. And they just got going when all at once all the metal they were carrying—knives and tire chains and coins and belt buckles and everything else—got freezing cold, too cold to touch. And then their leather jackets got freezing cold, so cold they had to pull them off and throw them away. And when the jackets were later collected, across the name of the gang on the back of each one had been branded 'The Scorpion.'" "Now, let me tell you something," said Hanks severely. "They heard the police sirens, and they threw all their weapons away. Then they threw their jackets away, to try to make believe they hadn't been part of the gang that had been fighting. But they were caught before they could get out of the schoolyard. If the squad cars had showed up a minute later, the schoolyard wouldn't have had anything in it but weapons and jackets, and the kids would have been all over the neighborhood, nice as you please, minding their own business and not bothering anybody. That's what happened. And all this talk about freezing cold and branding names into jackets is just some smart-alec punk's idea of a way to razz the police. Now, you just go back to worrying about what's happening in this precinct and forget about kid gangs up in Manhattan and comic book things like the Scorpion, or you're going to wind up like Wilcox, with that refrigerator business. Now, I don't want to hear any more about this nonsense, Stevenson." "Yes, sir," said Stevenson.
|
C. The cops would be preoccupied with other matters, and it was easy to explain why you had a weapon on you.
|
How do they obtain language identities?
|
### Introduction
Code-switching (CS) speech is defined as the alternation of languages in an utterance, it is a pervasive communicative phenomenon in multilingual communities. Therefore, developing a CS speech recognition (CSSR) system is of great interest. However, the CS scenario presents challenges to recognition system BIBREF0. Some attempts based on DNN-HMM framework have been made to alleviate these problems BIBREF1, BIBREF2. The methods usually contain components including acoustic, language, and lexicon models that are trained with different object separately, which would lead to sub-optimal performance. And the design of complicated lexicon including different languages would consume lots of human efforts. Therefore, end-to-end framework for CSSR has received increasing attention recently BIBREF3, BIBREF4, BIBREF5. Examples of such models include connectionist temporal classification (CTC) BIBREF6, attention-based encoder-decoder models BIBREF7, BIBREF8, and the recurrent neural network transducer (RNN-T) BIBREF9, BIBREF10, BIBREF11, BIBREF12. These methods combine acoustic, language, and lexicon models into a single model with joint training. And the RNN-T and attention-based models trained with large speech corpus perform competitively compared to the state-of-art model in some tasks BIBREF13. However, the lack of CS training data poses serious problem to end-to-end methods. To address the problem, language identity information is utilized to improve the performance of recognition BIBREF3, BIBREF4, BIBREF5. They are usually based on CTC or attention-based encoder-decoder models or the combination of both. However, previous works use an additional language identification (LID) model as an auxiliary module, which causes the system complex. In this paper, we propose an improved RNN-T model with language bias to alleviate the problem. The model is trained to predict language IDs as well as the subwords. To ensure the model can learn CS information, we add language IDs in the CS point of transcription, as illustrated in Fig. 1. In the figure, we use the arrangements of different geometric icons to represent the CS distribution. Compared with normal text, the tagged data can bias the RNN-T to predict language IDs in CS points. So our method can model the CS distribution directly, no additional LID model is needed. Then we constrain the input word embedding with its corresponding language ID, which is beneficial for model to learn the language identity information from transcription. In the inference process, the predicted language IDs are used to adjust the output posteriors. The experiment results on CS corpus show that our proposed method outperforms the RNN-T baseline (without language bias) significantly. Overall, our best model achieves 16.2% and 12.9% relative error reduction on two test sets, respectively. To our best knowledge, this is the first attempt of using the RNN-T model with language bias as an end-to-end CSSR strategy. The rest of the paper is organized as follows. In Section 2, we review RNN-T model. In Section 3, we describe the intuition of the proposed model. In Section 4, we present the experimental setups, and in Section 5, we report and discuss the experiment results in detail. Finally, we conclude the paper in Section 6. ### review of RNN-T
Although CTC has been applied successfully in the context of speech recognition, it assumes that outputs at each step are independent of the previous predictions BIBREF6. RNN-T is an improved model based on CTC, it augments with a prediction network, which is explicitly conditioned on the previous outputs BIBREF10, as illustrated in Fig. 2(a). Let $\mathnormal { \mathbf {X} = (\mathbf {x}_{1}, \mathbf {x}_{2}, ... , \mathbf {x}_{T})}$ be the acoustic input sequence, where $T$ is the frame number of sequence. Let $\mathbf {Y} = (\mathnormal {y}_{1}, \mathnormal {y}_{2}, ... , \mathnormal {y}_{U})$ be the corresponding sequence of output targets (without language IDs) over the RNN-T output space $\mathcal {Y}$, and $\mathcal {Y}^{*}$ be the set of all possible sequence over $\mathcal {Y}$. In the context of ASR, the input sequence is much longer than output targets, i.e., $T>U$. Because the frame-level alignments of the target label are unknown, RNN-T augments the output set with an additional symbol, refer to as the $\mathit {blank}$ symbol, denoted as $\phi $, i.e., $\bar{\mathcal {Y}} \in \mathcal {Y} \cup \lbrace \phi \rbrace $. We denote $\hat{\mathbf {Y}} \in \bar{\mathcal {Y}}^{*}$ as an alignment, which are equivalent to $(\mathnormal {y}_{1}, \mathnormal {y}_{2},\mathnormal {y}_{3}) \in \mathcal {Y}^*$ after operation $\mathcal {B}$, such as $\hat{\mathbf {Y}} = (\mathnormal {y}_{1}, \phi , \mathnormal {y}_{2}, \phi , \phi , \mathnormal {y}_{3}) \in \bar{\mathcal {Y}}^{*}$. Given the input sequence $\mathbf {X}$, RNN-T models the conditional probability $P(\mathbf {Y} \in \mathcal {Y}^* | \mathbf {X})$ by marginalizing over all possible alignments: where $\mathcal {B}$ is the function that removes consecutive identical symbols and then removing any blank from a given alignment in $\bar{\mathcal {Y}}^{*}$. An RNN-T model consists of three different networks as illustrated in Fig. 2(a). (a) Encoder network (referred to as transcription network) maps the acoustic features into higher level representation $\mathbf {h}_{t}^{enc} = f^{enc}(\lbrace \mathbf {x}_{\tau }\rbrace _{1 \le \tau \le t})$. (b) Prediction network produces output vector $\mathbf {p}_{u} = f^{pred}(\lbrace \mathnormal {y}_{v}\rbrace _{1 \le v \le u-1})$ based on the previous non-blank input label. (c) Joint network computes logits by combining the outputs of the previous two networks $z_{t,u} = f^{joint} (\mathbf {h}_{t}^{enc}, \mathbf {p}_{u})$. These logits are then passed to a softmax layer to define a probability distribution. The model can be trained by maximizing the log-likelihood of $P(\mathbf {Y} \in \mathcal {Y}^* | \mathbf {X})$. ### RNN-T with language bias
In this paper, we aim to build a concise end-to-end CSSR model that can handle the speech recognition and LID simultaneously. For this task, we augment the output symbols set with language IDs $<chn>$ and $<eng>$ as shown in Fig. 1, i.e., $\hat{\mathcal {Y}} \in \bar{\mathcal {Y}} \cup \lbrace <chn>,<eng>\rbrace $. The intuition behind it is that the CS in the transcript may obey a certain probability distribution, and this distribution can be learned by neural network. The properties of RNN-T is key for the problem. It can predict rich set of target symbols such as speaker role and "end-of-word" symbol, which are not related to the input feature directly BIBREF14, BIBREF15. So the language IDs can also be treated as the output symbols. What's more, RNN-T can seamlessly integrate the acoustic and linguistic information. The prediction network of it can be viewed as an RNN language model which predict the current label given history labels BIBREF10. So it is effective in incorporating LID into the language model. In general, predicting language IDs only from text data is difficult. However, the joint training mechanism of RNN-T allows it to combine the language and acoustic information to model the CS distribution. Furthermore, the tagged text can bias the RNN-T to predict language IDs which indicates CS points, yet the model trained with normal text can not do this. That is why we choose RNN-T to build the end-to-end CSSR system. To promote the model to learn CS distribution more efficient, We concatenate a short vector to all the English word embedding and the English tag $<eng>$ embedding, another different vector for Mandarin, as shown in the bottom of Fig. 2(b). This enhances the dependence of word embedding to its corresponding language ID. In the training process, RNN-T model can learn the distinction information between the two languages easily. The experiment results show that the word embedding constraint is an effective technology. In the inference process, we use the predicted language ID to adjust the output posteriors, as shown in the head of Fig. 2(b). This can bias the model to predict a certain language words more likely in the next-step decode. Overall, our proposed method can handle the speech recognition and LID simultaneously in a simple way, and without increasing additional burden. This study provides new insights into the CS information of text data and its application in end-to-end CSSR system. As a final note, the training and inference algorithms of the proposed model are similar to the standard RNN-T model. ### Experiments setups ::: Dataset
We conduct experiments on SEAME (South East Asia Mandarin English), a spontaneous conversational bilingual speech corpus BIBREF16. Most of the utterances contain both Mandarin and English uttered by interviews and conversations. We use the standard data partitioning rule of previous works which consists of three parts: $train$, $test_{sge}$ and $test_{man}$ (see Table 1) BIBREF2. $test_{sge}$ is biased to Southeast Asian accent English speech and $test_{man}$ is biased to Mandarin speech. Building an end-to-end model requires lots of training data, we apply speech speed perturbation to augment speech data BIBREF17. By manipulation, we get 3 times the data, with the speed rate of 0.9, 1, and 1.1 of the original speech respectively. We use the augmented data to build our DNN-HMM system and RNN-T system. ### Experiments setups ::: DNN-HMM Baseline System
In addition to the RNN-T baseline system, we also build a conventional DNN-HMM baseline for comparison. The model is based on time delay neural network (TDNN) which trained with lattice-free maximum mutual information (LF-MMI) BIBREF18. The TDNN model has 7 hidden layers with 512 units and the input acoustic future is 13-dimensional Mel-frequency cepstrum coefficient (MFCC). For language modeling, we use SRI language modeling toolkit BIBREF19 to build 4-gram language model with the training transcription. And we construct the lexicon by combining CMU English lexicon and our Mandarin lexicon. ### Experiments setups ::: RNN-T System
We construct the RNN-T baseline system as described in Section 3.1. The encoder network of RNN-T model consists of 4 layers of 512 long short-term memory (LSTM). The prediction network is 2 layers with 512 LSTM units. And the joint network consists of single feed-forward layer of 512 units with tanh activate function. The input acoustic features of encoder network are 80-dimensional log Mel-filterbank with 25ms windowing and 10ms frame shift. Mean and normalization is applied to the futures. And the input words embedding of prediction network is in 512 dimensions continuous numerical vector space. During training, the ADAM algorithm is used as the optimization method, we set the initial learning rate as 0.001 and decrease it linearly when there is no improvement on the validation set. To reduce the over-fitting problem, the dropout rate is set to 0.2 throughout all the experiments. In the inference process, the beam-search algorithm BIBREF9 with beam size 35 is used to decode the model. All the RNN-T models are trained from scratch use PyTorch. ### Experiments setups ::: Wordpieces
For Mandarin-English CSSR task, it is a natural way to construct output units by using characters. However, there are several thousands of Chinese characters and 26 English letters. Meanwhile, the acoustic counterpart of Chinese character is much longer than English letter. So, the character modeling unit will result in significant discrepancy problem between the two languages. To balance the problem, we adopt BPE subword BIBREF20 as the English modeling units. The targets of our RNN-T baseline system contains 3090 English wordpieces and 3643 Chinese characters. The BPE subword units can not only increase the duration of English modeling units but also maintain a balance unit number of two languages. ### Experiments setups ::: Evaluation Metrics
In this paper, we use mixed error rate (MER) to evaluate the experiment results of our methods. The MER is defined as the combination of word error rate (WER) for English and character error rate (CER) for Mandarin. This metrics can balance the Mandarin and English error rates better compared to the WER or CER. ### Results and Analysis ::: Results of RNN-T Model
Table 2 reports our main experiment results of different setups with the standard decode in inference. It is obvious that the MER of end-to-end systems are not as competitive as the LF-MMI TDNN system. The result is consistent with some other reports BIBREF3. However, data augmentation is more effective for end-to-end system than TDNN system. It suggests that the gap between our RNN-T and TDNN may further reduce with increasing data. Furthermore, We can also observe that all the experiment results in $test_{sge}$ is much worse than $test_{man}$. This is probably that the accent English in data $test_{sge}$ is more difficult for the recognition system. Bilinguals usually have serious accent problem, which poses challenge to CSSR approaches. Because the data augmentation technology can significantly reduce the MER of end-to-end model, we conduct all the following experiments based on augmented training data. In order to fairly compare the results of proposed methods with baseline, we remove all the language IDs in the decoded transcription. We can find that The performance of RNN-T model trained (without word embedding constraint) with tagged transcription is much better than the RNN-T baseline. It achieves 9.3% and 7.6% relative MER reduction on two test sets respectively. This shows that the tagged text can improve the modeling ability of RNN-T for the CSSR problem. It is the main factor that causes the MER reduction in our experiments. Furthermore, word embedding constraint can also improve the performance of the system though not significant. Overall, our proposed methods yields improved results without increasing additional training or inference burden. ### Results and Analysis ::: Effect of Language IDs Re-weighted Decode
We then evaluate the system performance by adjusting the weights of next-step predictions in decode process. Table 3 shows the results of RNN-T model with different language IDs weights in inference. It is obvious that the re-weighted methods outperform the model with standard decode process. This suggests that the predicted language IDs can effectively guide the model decoding. Because the model assigns language IDs to the recognized words directly, the language IDs error rate is hard to compute. This result may imply that the prediction accuracy of our method is high enough to guide decoding. Meanwhile, We also find that the re-weighted method is more effective on the $test_{man}$ than $test_{sge}$. This could be caused by higher language IDs prediction accuracy in $test_{man}$. The results of the two different $\lambda $ have similarly MER, and we set $\lambda =0.2$ in the following experiments. ### Results and Analysis ::: Results of Language Model Re-score
Table 4 shows the MER results of the N-best (N=35) re-scoring with N-gram and neural language models. The language models are both trained with the tagged training transcription. We see that the language re-scoring can further improve the performance of models. It reveals that the prediction network of RNN-T still has room to be further optimization. Finally, compared to the RNN-T baseline without data augment, the best results of proposed method can achieve 25.9% and 21.3% relative MER reduction on two dev sets respectively. compared to the RNN-T baseline with data augment, the proposed method can achieve 16.2% and 12.9% relative MER reduction. For both scenarios, our RNN-T methods can achieve better performance than baselines. ### Conclusions and Future Work
In this work we develop an improved RNN-T model with language bias for end-to-end Mandarin-English CSSR task. Our method can handle the speech recognition and LID simultaneously, no additional LID system is needed. It yields consistent improved results of MER without increasing training or inference burden. Experiment results on SEAME show that proposed approaches significantly reduce the MER of two dev sets from 33.3% and 44.9% to 27.9% and 39.1% respectively. In the future, we plan to pre-train the prediction network of RNN-T model using large text corpus, and then finetune the RNN-T model with labeled speech data by frozen the prediction network. ### Acknowledgment
This work is supported by the National Key Research & Development Plan of China (No.2018YFB1005003) and the National Natural Science Foundation of China (NSFC) (No.61425017, No.61831022, No.61773379, No.61771472) Fig. 1. Code-switching distribution diagram. Fig. 2. Basic RNN-T model (a) and RNN-T with language bias (b). Table 1. Data Statistics of SEAME [3] Table 3. The MER language IDs re-weighted decode method in inference. Table 4. The MER of language model rescore. Table 2. The MER of different setups with standard decode method in inference.
|
model is trained to predict language IDs as well as the subwords, we add language IDs in the CS point of transcriptio
|
Before Peggy's parents reveal their decision, was it obvious that they would let her move?
A. Totally. Her parents sounded supportive in every possible way and they had the resources to get her multiple auditions in New York.
B. Not at all. Her parents had to argue about it for a while, and she was feeling nostalgic around her neighborhood so it looked like she was going to stay in town.
C. Not entirely. But, their conversation with Peggy along with Jean's conversation with Peggy supplied strong evidence that they would say yes.
D. Totally. Peggy had won so many awards and participated in so much theater that it would have been horrible parenting to make her stay.
|
PEGGY FINDS THE THEATER I Dramatic Dialogue “Of course, this is no surprise to us,” Thomas Lane said to his daughter Peggy, who perched tensely on the edge of a kitchen stool. “We could hardly have helped knowing that you’ve wanted to be an actress since you were out of your cradle. It’s just that decisions like this can’t be made quickly.” “But, Dad!” Peggy almost wailed. “You just finished saying yourself that I’ve been thinking about this and wanting it for years! You can’t follow that by calling it a quick decision!” She turned to her mother, her hazel eyes flashing under a mass of dark chestnut curls. “Mother, you understand, don’t you?” Mrs. Lane smiled gently and placed her soft white hand on her daughter’s lean brown one. “Of course I understand, Margaret, and so does your father. We both want to do what’s best for you, not to stand in your way. The only question is whether the time is right, or if you should wait longer.” 2 “Wait! Mother—Dad—I’m years behind already! The theater is full of beginners a year and even two years younger than I am, and girls of my age have lots of acting credits already. Besides, what is there to wait for?” Peggy’s father put down his coffee cup and leaned back in the kitchen chair until it tilted on two legs against the wall behind him. He took his time before answering. When he finally spoke, his voice was warm and slow. “Peg, I don’t want to hold up your career. I don’t have any objections to your wanting to act. I think—judging from the plays I’ve seen you in at high school and college—that you have a real talent. But I thought that if you would go on with college for three more years and get your degree, you would gain so much worth-while knowledge that you’d use and enjoy for the rest of your life—” “But not acting knowledge!” Peggy cried. “There’s more to life than that,” her father put in. “There’s history and literature and foreign languages and mathematics and sciences and music and art and philosophy and a lot more—all of them fascinating and all important.” “None of them is as fascinating as acting to me,” Peggy replied, “and none of them is nearly as important to my life.” 3 Mrs. Lane nodded. “Of course, dear. I know just how you feel about it,” she said. “I would have answered just the same way when I was your age, except that for me it was singing instead of acting. But—” and here her pleasant face betrayed a trace of sadness—“but I was never able to be a singer. I guess I wasn’t quite good enough or else I didn’t really want it hard enough—to go on with all the study and practice it needed.” She paused and looked thoughtfully at her daughter’s intense expression, then took a deep breath before going on. “What you must realize, Margaret, is that you may not quite make the grade. We think you’re wonderful, but the theater is full of young girls whose parents thought they were the most talented things alive; girls who won all kinds of applause in high-school and college plays; girls who have everything except luck. You may be one of these girls, and if you are, we want you to be prepared for it. We want you to have something to fall back on, just in case you ever need it.” Mr. Lane, seeing Peggy’s hurt look, was quick to step in with reassurance. “We don’t think you’re going to fail, Peg. We have every confidence in you and your talents. I don’t see how you could miss being the biggest success ever—but I’m your father, not a Broadway critic or a play producer, and I could be wrong. And if I am wrong, I don’t want you to be hurt. All I ask is that you finish college and get a teacher’s certificate so that you can always find useful work if you have to. Then you can try your luck in the theater. Doesn’t that make sense?” 4 Peggy stared at the faded linoleum on the floor for a few moments before answering. Then, looking first at her mother and then at her father, she replied firmly, “No, it doesn’t! It might make sense if we were talking about anything else but acting, but we’re not. If I’m ever going to try, I’ll have a better chance now than I will in three years. But I can see your point of view, Dad, and I’ll tell you what—I’ll make a bargain with you.” “What sort of bargain, Peg?” her father asked curiously. “If you let me go to New York now, and if I can get into a good drama school there, I’ll study and try to find acting jobs at the same time. That way I’ll still be going to school and I’ll be giving myself a chance. And if I’m not started in a career in one year, I’ll go back to college and get my teacher’s certificate before I try the theater again. How does that sound to you?” “It sounds fair enough,” Tom Lane admitted, “but are you so confident that you’ll see results in one year? After all, some of our top stars worked many times that long before getting any recognition.” “I don’t expect recognition in one year, Dad,” Peggy said. “I’m not that conceited or that silly. All I hope is that I’ll be able to get a part in that time, and maybe be able to make a living out of acting. And that’s probably asking too much. If I have to, I’ll make a living at something else, maybe working in an office or something, while I wait for parts. What I want to prove in this year is that I can act. If I can’t, I’ll come home.” 5 “It seems to me, Tom, that Margaret has a pretty good idea of what she’s doing,” Mrs. Lane said. “She sounds sensible and practical. If she were all starry-eyed and expected to see her name in lights in a few weeks, I’d vote against her going, but I’m beginning to think that maybe she’s right about this being the best time.” “Oh, Mother!” Peggy shouted, jumping down from the stool and throwing her arms about her mother’s neck. “I knew you’d understand! And you understand too, don’t you, Dad?” she appealed. Her father replied in little puffs as he drew on his pipe to get it started. “I ... never said ... I didn’t ... understand you ... did I?” His pipe satisfactorily sending up thick clouds of fragrant smoke, he took it out of his mouth before continuing more evenly. “Peg, your mother and I are cautious only because we love you so much and want what’s going to make you happy. At the same time, we want to spare you any unnecessary unhappiness along the way. Remember, I’m not a complete stranger to show business. Before I came out here to Rockport to edit the Eagle , I worked as a reporter on one of the best papers in New York. I saw a lot ... I met a lot of actors and actresses ... and I know how hard the city often was for them. But I don’t want to protect you from life. That’s no good either. Just let me think about it a little longer and let me talk to your mother some more.” 6 Mrs. Lane patted Peggy’s arm and said, “We won’t keep you in suspense long, dear. Why don’t you go out for a walk for a while and let us go over the situation quietly? We’ll decide before bedtime.” Peggy nodded silently and walked to the kitchen door, where she paused to say, “I’m just going out to the barn to see if Socks is all right for the night. Then maybe I’ll go down to Jean’s for a while.” As she stepped out into the soft summer dusk she turned to look back just in time to see her mother throw her a comically exaggerated wink of assurance. Feeling much better, Peggy shut the screen door behind her and started for the barn. Ever since she had been a little girl, the barn had been Peggy’s favorite place to go to be by herself and think. Its musty but clean scent of straw and horses and leather made her feel calm and alive. Breathing in its odor gratefully, she walked into the half-dark to Socks’s stall. As the little bay horse heard her coming, she stamped one foot and softly whinnied a greeting. Peggy stopped first at the bag that hung on the wall among the bridles and halters and took out a lump of sugar as a present. Then, after stroking Socks’s silky nose, she held out her palm with the sugar cube. Socks took it eagerly and pushed her nose against Peggy’s hand in appreciation. As Peggy mixed some oats and barley for her pet and checked to see that there was enough straw in the stall, she thought about her life in Rockport and the new life that she might soon be going to. 7 Rockport, Wisconsin, was a fine place, as pretty a small town as any girl could ask to grow up in. And not too small, either, Peggy thought. Its 16,500 people supported good schools, an excellent library, and two good movie houses. What’s more, the Rockport Community College attracted theater groups and concert artists, so that life in the town had always been stimulating. And of course, all of this was in addition to the usual growing-up pleasures of swimming and sailing, movie dates, and formal dances—everything that a girl could want. Peggy had lived all her life here, knew every tree-shaded street, every country road, field, lake, and stream. All of her friends were here, friends she had known since her earliest baby days. It would be hard to leave them, she knew, but there was no doubt in her mind that she was going to do so. If not now, then as soon as she possibly could. It was not any dissatisfaction with her life, her friends, or her home that made Peggy want to leave Rockport. She was not running away from anything, she reminded herself; she was running to something. To what? To the bright lights, speeding taxis, glittering towers of a make-believe movie-set New York? Would it really be like that? Or would it be something different, something like the dreary side-street world of failure and defeat that she had also seen in movies? 8 Seeing the image of herself hungry and tired, going from office to office looking for a part in a play, Peggy suddenly laughed aloud and brought herself back to reality, to the warm barn smell and the big, soft-eyed gaze of Socks. She threw her arm around the smooth bay neck and laid her face next to the horse’s cheek. “Socks,” she murmured, “I need some of your horse sense if I’m going to go out on my own! We’ll go for a fast run in the morning and see if some fresh air won’t clear my silly mind!” With a final pat, she left the stall and the barn behind, stepping out into the deepening dusk. It was still too early to go back to the house to see if her parents had reached a decision about her future. Fighting down an impulse to rush right into the kitchen to see how they were coming along, Peggy continued down the driveway and turned left on the slate sidewalk past the front porch of her family’s old farmhouse and down the street toward Jean Wilson’s house at the end of the block. As she walked by her own home, she noticed with a familiar tug at her heart how the lilac bushes on the front lawn broke up the light from the windows behind them into a pattern of leafy lace. For a moment, or maybe a little more, she wondered why she wanted to leave this. What for? What could ever be better? 9 II Dramatic Decision Upstairs at the Wilsons’, Peggy found Jean swathed in bath towels, washing her long, straight red hair, which was now white with lather and piled up in a high, soapy knot. “You just washed it yesterday!” Peggy said. “Are you doing it again—or still?” Jean grinned, her eyes shut tight against the soapsuds. “Again, I’m afraid,” she answered. “Maybe it’s a nervous habit!” “It’s a wonder you’re not bald, with all the rubbing you give your hair,” Peggy said with a laugh. “Well, if I do go bald, at least it will be with a clean scalp!” Jean answered with a humorous crinkle of her freckled nose. Taking a deep breath and puffing out her cheeks comically, she plunged her head into the basin and rinsed off the soap with a shampoo hose. When she came up at last, dripping-wet hair was tightly plastered to the back of her head. “There!” she announced. “Don’t I look beautiful?” 10 After a brisk rubdown with one towel, Jean rolled another dry towel around her head like an Indian turban. Then, having wrapped herself in an ancient, tattered, plaid bathrobe, she led Peggy out of the steamy room and into her cozy, if somewhat cluttered, bedroom. When they had made themselves comfortable on the pillow-strewn daybeds, Jean came straight to the point. “So the grand debate is still going on, is it? When do you think they’ll make up their minds?” she asked. “How do you know they haven’t decided anything yet?” Peggy said, in a puzzled tone. “Oh, that didn’t take much deduction, my dear Watson,” Jean laughed. “If they had decided against the New York trip, your face would be as long as Socks’s nose, and it’s not half that long. And if the answer was yes, I wouldn’t have to wait to hear about it! You would have been flying around the room and talking a mile a minute. So I figured that nothing was decided yet.” “You know, if I were as smart as you,” Peggy said thoughtfully, “I would have figured out a way to convince Mother and Dad by now.” “Oh, don’t feel bad about being dumb,” Jean said in mock tones of comfort. “If I were as pretty and talented as you are, I wouldn’t need brains, either!” With a hoot of laughter, she rolled quickly aside on the couch to avoid the pillow that Peggy threw at her. A short, breathless pillow fight followed, leaving the girls limp with laughter and with Jean having to retie her towel turban. From her new position, flat on the floor, Peggy looked up at her friend with a rueful smile. 11 “You know, I sometimes think that we haven’t grown up at all!” she said. “I can hardly blame my parents for thinking twice—and a lot more—before treating me like an adult.” “Nonsense!” Jean replied firmly. “Your parents know a lot better than to confuse being stuffy with being grown-up and responsible. And, besides, I know that they’re not the least bit worried about your being able to take care of yourself. I heard them talking with my folks last night, and they haven’t got a doubt in the world about you. But they know how hard it can be to get a start as an actress, and they want to be sure that you have a profession in case you don’t get a break in show business.” “I know,” Peggy answered. “We had a long talk about it this evening after dinner.” Then she told her friend about the conversation and her proposed “bargain” with her parents. “They both seemed to think it was fair,” she concluded, “and when I went out, they were talking it over. They promised me an answer by bedtime, and I’m over here waiting until the jury comes in with its decision. You know,” she said suddenly, sitting up on the floor and crossing her legs under her, “I bet they wouldn’t hesitate a minute if you would only change your mind and decide to come with me and try it too!” 12 After a moment’s thoughtful silence, Jean answered slowly, “No, Peg. I’ve thought this all out before, and I know it would be as wrong for me as it is right for you. I know we had a lot of fun in the dramatic groups, and I guess I was pretty good as a comedienne in a couple of the plays, but I know I haven’t got the real professional thing—and I know that you have. In fact, the only professional talent I think I do have for the theater is the ability to recognize talent when I see it—and to recognize that it’s not there when it isn’t!” “But, Jean,” Peggy protested, “you can handle comedy and character lines as well as anyone I know!” Jean nodded, accepting the compliment and seeming at the same time to brush it off. “That doesn’t matter. You know even better than I that there’s a lot more to being an actress—a successful one—than reading lines well. There’s the ability to make the audience sit up and notice you the minute you walk on, whether you have lines or not. And that’s something you can’t learn; you either have it, or you don’t. It’s like being double-jointed. I can make an audience laugh when I have good lines, but you can make them look at you and respond to you and be with you all the way, even with bad lines. That’s why you’re going to go to New York and be an actress. And that’s why I’m not.” “But, Jean—” Peggy began. 13 “No buts!” Jean cut in. “We’ve talked about this enough before, and I’m not going to change my mind. I’m as sure about what I want as you are about what you want. I’m going to finish college and get my certificate as an English teacher.” “And what about acting? Can you get it out of your mind as easily as all that?” Peggy asked. “That’s the dark and devious part of my plan,” Jean answered with a mysterious laugh that ended in a comic witch’s cackle and an unconvincing witch-look that was completely out of place on her round, freckled face. “Once I get into a high school as an English teacher, I’m going to try to teach a special course in the literature of the theater and maybe another one in stagecraft. I’m going to work with the high-school drama group and put on plays. That way, I’ll be in a spot where I can use my special talent of recognizing talent. And that way,” she added, becoming much more serious, “I have a chance really to do something for the theater. If I can help and encourage one or two people with real talent like yours, then I’ll feel that I’ve really done something worth while.” Peggy nodded silently, not trusting herself to speak for fear of saying something foolishly sentimental, or even of crying. Her friend’s earnestness about the importance of her work and her faith in Peggy’s talent had touched her more than she could say. 14 The silence lasted what seemed a terribly long time, until Jean broke it by suddenly jumping up and flinging a last pillow which she had been hiding behind her back. Running out of the bedroom, she called, “Come on! I’ll race you down to the kitchen for cocoa! By the time we’re finished, it’ll be about time for your big Hour of Decision scene!” It was nearly ten o’clock when Peggy finally felt that her parents had had enough time to talk things out. Leaving the Wilson house, she walked slowly despite her eagerness, trying in all fairness to give her mother and father every minute she could. Reaching her home, she cut across the lawn behind the lilac bushes, to the steps up to the broad porch that fronted the house. As she climbed the steps, she heard her father’s voice raised a little above its normal soft, deep tone, but she could not make out the words. Crossing the porch, she caught sight of him through the window. He was speaking on the telephone, and now she caught his words. “Fine. Yes.... Yes—I think we can. Very well, day after tomorrow, then. That’s right—all three of us. And, May—it’ll be good to see you again, after all these years! Good-by.” As Peggy entered the room, her father put down the phone and turned to Mrs. Lane. “Well, Betty,” he said, “it’s all set.” “What’s all set, Dad?” Peggy said, breaking into a run to her father’s side. 15 “Everything’s all set, Peg,” her father said with a grin. “And it’s set just the way you wanted it! There’s not a man in the world who can hold out against two determined women.” He leaned back against the fireplace mantel, waiting for the explosion he felt sure was to follow his announcement. But Peggy just stood, hardly moving a muscle. Then she walked carefully, as if she were on the deck of a rolling ship, to the big easy chair and slowly sat down. “Well, for goodness’ sake!” her mother cried. “Where’s the enthusiasm?” Peggy swallowed hard before answering. When her voice came, it sounded strange, about two tones higher than usual. “I ... I’m trying to be sedate ... and poised ... and very grown-up,” she said. “But it’s not easy. All I want to do is to—” and she jumped out of the chair—“to yell whoopee !” She yelled at the top of her lungs. After the kisses, the hugs, and the first excitement, Peggy and her parents adjourned to the kitchen, the favorite household conference room, for cookies and milk and more talk. “Now, tell me, Dad,” Peggy asked, her mouth full of oatmeal cookies, no longer “sedate” or “poised,” but her natural, bubbling self. “Who was that on the phone, and where are the three of us going, and what’s all set?” 16 “One thing at a time,” her father said. “To begin with, we decided almost as soon as you left that we were going to let you go to New York to try a year’s experience in the theater. But then we had to decide just where you would live, and where you should study, and how much money you would need, and a whole lot of other things. So I called New York to talk to an old friend of mine who I felt would be able to give us some help. Her name is May Berriman, and she’s spent all her life in the theater. In fact, she was a very successful actress. Now she’s been retired for some years, but I thought she might give us some good advice.” “And did she?” Peggy asked. “We were luckier than I would have thought possible,” Mrs. Lane put in. “It seems that May bought a big, old-fashioned town house and converted it into a rooming house especially for young actresses. She always wanted a house of her own with a garden in back, but felt it was foolish for a woman living alone. This way, she can afford to run a big place and at the same time not be alone. And best of all, she says she has a room that you can have!” “Oh, Mother! It sounds wonderful!” Peggy exulted. “I’ll be with other girls my own age who are actresses, and living with an experienced actress! I’ll bet she can teach me loads!” “I’m sure she can,” her father said. “And so can the New York Dramatic Academy.” “Dad!” Peggy shouted, almost choking on a cooky. “Don’t tell me you’ve managed to get me accepted there! That’s the best dramatic school in the country! How—?” 17 “Don’t get too excited, Peg,” Mr. Lane interrupted. “You’re not accepted anywhere yet, but May Berriman told me that the Academy is the best place to study acting, and she said she would set up an audition for you in two days. The term starts in a couple of weeks, so there isn’t much time to lose.” “Two days! Do you mean we’ll be going to New York day after tomorrow, just like that?” “Oh, no,” her mother answered calmly. “We’re going to New York tomorrow on the first plane that we can get seats on. Your father doesn’t believe in wasting time, once his mind is made up.” “Tomorrow?” Peggy repeated, almost unable to believe what she had heard. “What are we sitting here talking for, then? I’ve got a million things to do! I’ve got to get packed ... I’ve got to think of what to read for the audition! I can study on the plane, I guess, but ... oh! I’ll be terrible in a reading unless I can have more time! Oh, Mother, what parts will I do? Where’s the Shakespeare? Where’s—” “Whoa!” Mr. Lane said, catching Peggy’s arm to prevent her from rushing out of the kitchen. “Not now, young lady! We’ll pack in the morning, talk about what you should read, and take an afternoon plane to New York. But tonight, you’d better think of nothing more than getting to bed. This is going to be a busy time for all of us.” Reluctantly, Peggy agreed, recognizing the sense of what her father said. She finished her milk and cookies, kissed her parents good night and went upstairs to bed. But it was one thing to go to bed and another to go to sleep. 18 Peggy lay on her back, staring at the ceiling and the patterns of light and shade cast by the street lamp outside as it shone through the leaves of the big maple tree. As she watched the shifting shadows, she reviewed the roles she had played since her first time in a high-school play. Which should she refresh herself on? Which ones would she do best? And which ones were most suited to her now? She recognized that she had grown and developed past some of the roles which had once seemed perfectly suited to her talent and her appearance. But both had changed. She was certainly not a mature actress yet, from any point of view, but neither was she a schoolgirl. Her trim figure was well formed; her face had lost the undefined, simple cuteness of the early teens, and had gained character. She didn’t think she should read a young romantic part like Juliet. Not that she couldn’t do it, but perhaps something sharper was called for. Perhaps Viola in Twelfth Night ? Or perhaps not Shakespeare at all. Maybe the people at the Academy would think she was too arty or too pretentious? Maybe she should do something dramatic and full of stormy emotion, like Blanche in A Streetcar Named Desire ? Or, better for her development and age, a light, brittle, comedy role...? 19 Nothing seemed quite right. Peggy’s thoughts shifted with the shadows overhead. All the plays she had ever seen or read or acted in melted together in a blur, until the characters from one seemed to be talking with the characters from another and moving about in an enormous set made of pieces from two or three different plays. More actors kept coming on in a fantastic assortment of costumes until the stage was full. Then the stage lights dimmed, the actors joined hands across the stage to bow, the curtain slowly descended, the lights went out—and Peggy was fast asleep.
|
C. Not entirely. But, their conversation with Peggy along with Jean's conversation with Peggy supplied strong evidence that they would say yes.
|
By how much do they outperform state-of-the-art solutions on SWDA and MRDA?
|
### Introduction
Dialogue Act Recognition (DAR) is an essential problem in modeling and detecting discourse structure. The goal of DAR is to attach semantic labels to each utterance in a conversation and recognize the speaker's intention, which can be regarded as a sequence labeling task. Many applications have benefited from the use of automatic dialogue act recognition such as dialogue systems, machine translation, automatic speech recognition, topic identification and talking avatars BIBREF0 BIBREF1 BIBREF2 . One of the primary applications of DAR is to support task-oriented discourse agent system. Knowing the past utterances of DA can help ease the prediction of the current DA state, thus help to narrow the range of utterance generation topics for the current turn. For instance, the "Greeting" and "Farewell" acts are often followed with another same type utterances, the "Answer" act often responds to the former "Question" type utterance. Thus if we can correctly recognize the current dialogue act, we can easily predict the following utterance act and generate a corresponding response. Table 1 shows a snippet of the kind of discourse structure in which we are interested. The essential problem of DAR lies on predicting the utterance's act by referring to contextual utterances with act labels. Most of existing models adopt handcrafted features and formulate the DAR as a multi-classification problem. However, these methods which adopt feature engineering process and multi-classification algorithms reveal deadly weakness from two aspects: First, they are labor intensive and can not scale up well across different datasets. Furthermore, they abandon the useful correlation information among contextual utterances. Typical multi-classification algorithms like SVM, Naive Bayes BIBREF3 BIBREF4 BIBREF5 can not account for the contextual dependencies and classify the DA label in isolation. It is evident that during a conversation, the speaker's intent is influenced by the former utterance such as the previous "Greeting" and "Farewell" examples. To tackle these two problems, some works have turn to structured prediction algorithm along with deep learning tactics such as DRLM-Conditional BIBREF6 , LSTM-Softmax BIBREF0 and RCNN BIBREF7 . However, most of them failed to utilize the empirical effectiveness of attention in the graphical structured network and relies completely on the hidden layers of the network, which may cause the structural bias. A further limitation is that although these works claim they have considered the contextual correlations, in fact they view the whole conversation as a flat sequence and neglect the dual dependencies in the utterance level and act level BIBREF8 BIBREF9 BIBREF10 . Until now, the achieved performances in DAR field are still far behind human annotator's accuracy. In this paper, we present the problem of DAR from the viewpoint of extending richer CRF-attentive structural dependencies along with neural network without abandoning end-to-end training. For simplicity, we call the framework as CRF-ASN (CRF-Attentive Structured Network). Specifically, we propose the hierarchical semantic inference integrated with memory mechanism on the utterance modeling. The memory mechanism is adopted in order to enable the model to look beyond localized features and have access to the entire sequence. The hierarchical semantic modeling learns different levels of granularity including word level, utterance level and conversation level. We then develop internal structured attention network on the linear-chain conditional random field (CRF) to specify structural dependencies in a soft manner. This approach generalizes the soft-selection attention on the structural CRF dependencies and takes into account the contextual influence on the nearing utterances. It is notably that the whole process is differentiable thus can be trained in an end-to-end manner. The main contributions of this paper are as follows: The rest of this paper is organized as follows. In section 2, we introduce the problem of dialogue act recognition from the viewpoint of introducing CRF-structured attention, and propose the CRF-attentive structural network with hierarchical semantic inference and memory mechanism. A variety of experimental results are presented in Section 3. We have a comprehensive analysis on the experiment results and conduct the ablations to prove the availability of our model. We then provide a brief review of the related work about dialogue act recognition problem in Section 4. Finally, we provide some concluding remarks in Section 5. ### CRF-attentive Structured Network
In this section, we study the problem of dialogue act recognition from the viewpoint of extending rich CRF-attentive structural dependencies. We first present the hierarchical semantic inference with memory mechanism from three levels: word level, utterance level and conversation level. We then develop graphical structured attention to the linear chain conditional random field to fully utilize the contextual dependencies. ### The problem
Before presenting the problem, we first introduce some basic mathematical notions and terminologies for dialogue act recognition. Formally, we assume the input is in the form of sequence pairs: INLINEFORM0 with INLINEFORM1 . INLINEFORM2 is the input of the INLINEFORM3 -th conversation in dataset INLINEFORM4 and INLINEFORM5 is the INLINEFORM6 -th targeted dialogue act type. Each conversation INLINEFORM7 is composed of a sequence of utterances which denoted as INLINEFORM8 with aligned act types INLINEFORM9 . We have each dialogue act type assigned to utterance INLINEFORM10 and each associated INLINEFORM11 denoted the possible dialogue act belongs to INLINEFORM12 act types. Again each utterance consists of a sequence of diverse words INLINEFORM13 . Most of the previous models do not leverage the implicit and intrinsic dependencies among dialogue act and utterances. They just consider a conversation as a flat structure with an extremely long chain of words. However, such a construction suffers vanishing gradient problem as the extremely long words become impractical in the neural network back-propagation training process. To alleviate this problem, we consider the conversation to be a hierarchical structure composed of three level encoders: first encode each word in a fine grained manner, and the second encoder operates at the utterance level, the last encoder encode each utterance in the conversation level. Each encoder is based on the previous one thus can make sure the output of the previous one can capture the dependencies across the conversation. Here we take an example to illustrate the sequence structure in Figure 1. Apart from hierarchical neural encoders, we also integrate external memory to allow the model to have unrestricted access to the whole sequence rather than localized features as in RNNs. Naturally the dialogue act recognition problem can be regarded as a sequence labeling task which can be assigned dialogue act through multi-classification method or the structured prediction algorithms. In our formulation, we adopt the linear chain conditional random field (CRF) along with hierarchical attentive encoders for the structured prediction. Instead of labeling each utterance in isolation, structured prediction models such as HMM, CRF can better capture the contextual dependencies among utterances. In our model, we define the structured attention model as being an extended attention model which provides an alternative approach to incorporate the machinery of structural inference directly into our neural network. ### Hierarchical Semantic Network
Due to the hierarchical nature of conversations, our proposed model is constructed at multiple levels of granularity, e.g. word level, utterance level and conversation level. The representation of a conversation can be composed by each utterance INLINEFORM0 , and each INLINEFORM1 can be obtained by combining the representations of constituent words INLINEFORM2 . Taking inspiration from Memory Networks and incorporate so-called memory hops, we adopt the memory enhanced contextual representations in order to have unrestricted access to the whole sequence rather than localized features as former recurrent neural network. Here we include the memory enhanced hierarchical representation in Figure 2 to depict the conversation level representation. As illustrated in Figure 2, the hierarchical semantic network can be divided into two parts: (1) fine grained embedding layer (2) memory enhanced contextual representation layer. The second part can be further broken down into three main components: (a) the input memory INLINEFORM0 which takes in the output from the word embedding layer (b) the contextual attention which takes the consideration of the former utterance and the latter one. (c) the output memory INLINEFORM1 which is obtained from the input memory connected with the attention mechanism. The weights are determined by measuring the similarity between the input memory and the current utterance input. Fine Grained Embedding: For a given conversation, each utterance INLINEFORM0 is encoded by a fine grained embedding layer. We first try to utilize the rich lexical factors and linguistic properties to enhance the word representation. For each word token INLINEFORM1 in each utterance, we initialized the word embedding using pretrained embeddings such as Word2vec or Glove. Furthermore, in order to tackle the out-of-vocabulary (OOV) problem, we adopt the character-level word embedding via CNN to combine with pretrained word level embeddings. We also extend the lexical factors via POS tag and NER tag to enhance the utterance understanding. The obtained four factors are concatenated to form a rich lexical representation as: INLINEFORM2 Since we consider the bidirectional GRU to encode the representation of each utterance, we concatenate the outputs from the forward and backward GRU hidden representations at the time step. For each utterance INLINEFORM0 which consists a sequence of words INLINEFORM1 , the original semantic representation is as follows: INLINEFORM2 Here we utilize INLINEFORM0 and INLINEFORM1 to represent the word level embedding function and utterance level encoder in our hierarchical model. After obtained the original semantic representations on each utterance, we later apply the memory enhanced contextual layer to further explore the correlations between utterances. Memory Enhanced Contextual Representation: Every utterance in a conversation is encoded with INLINEFORM0 , where INLINEFORM1 is the encoding function via Bi-GRU to map the input words into a vector INLINEFORM2 . The original sequence utterances are denoted as INLINEFORM3 . While this original semantic representation can be the input component in the context of memory network. In order to tackle the drawback of insensitivity to temporal information between memory cells, we adopt the approach in injecting temporal signal into the memory using a contextual recurrent encoding: INLINEFORM4 where INLINEFORM0 , INLINEFORM1 , INLINEFORM2 are learnable parameters. It is a remarkable fact that the new sequence INLINEFORM0 can be seen as the contextual integrated representations which take consider of the former utterances and the latter ones. The injected temporal signal can further explore the contextual influence on the current input utterance. We thus can make use of this obtained INLINEFORM1 to represent another INLINEFORM2 which cares more about the context influence. For the current input utterance INLINEFORM0 , in memory networks, the input is required to be in the same space as the input memory. Here we adopt the popular attention mechanism in the memory by measuring the relevance between current input utterance INLINEFORM1 and the contextual new representation INLINEFORM2 . The relevance is measured with a softmax function: INLINEFORM3 Once the attention weights have been computed, the output memory can be used to generate the final output of the memory layer in the form of a weighted sum over the attention and the input utterance: INLINEFORM0 The output allows the model to have unrestricted access to elements in previous steps as opposed to a single hidden state INLINEFORM0 in recurrent neural networks. Thereby we can effectively detect the long range dependencies among utterances in a conversation. To further extend the complex reasoning over multiple supporting facts from memory, we adopt a stacking operation which stacks hops between the original utterance semantic representation INLINEFORM0 and the k-th output hop INLINEFORM1 to be the input to the INLINEFORM2 th hop: INLINEFORM3 where INLINEFORM0 encodes not only information at the current step ( INLINEFORM1 ), but also relevant knowledge from the contextual memory ( INLINEFORM2 ). Note that in the scope of this work, we limit the number of hops to 1 to ease the computational cost. ### Structured CRF-Attention Network
Traditional attention networks have proven to be an effective approach for embedding categorical inference within a deep neural network. However, In DAR problem, we need to further explore the structural dependencies among utterances and dialogue acts. As we see, utterances in a conversation are not exist independently. The latter utterance may be the responding answer to the former question, or that the chunk of utterances are in the same act type. Here we consider generalizing selection to types of chunks selecting attention, and propose the structured attention to model richer dependencies by incorporating structural distributions within networks. Such a structured attention can be interpreted as using soft-selection that considers all possible structures over the utterance input. In our paper, we formulate the DAR as a sequence labeling problem. It is a natural choice to assign a label to each element in the sequence via linear chain CRF, which enable us to model dependencies among labels. Here we do not directly apply the original linear chain CRF to the learned utterance. Although the dependencies among utterances have been captured by the former hierarchical semantic networks, we still need to further explore the dialogue act dependencies in the label level. For dialogue act sequence labeling problem, greedily predicting the dialogue act at each time-step might not optimal the solution. Instead, it is better to look into the correlations in both utterance level and the dialogue act level in order to jointly decode the best chain of dialogue acts. Formally, let INLINEFORM0 represent a sequence of utterance inputs, let INLINEFORM1 be the corresponding dialogue act sequence. Variable INLINEFORM2 are discrete latent act variables INLINEFORM3 with sample space INLINEFORM4 that encodes the desired selection among these inputs. The aim of the structured attention is to produce a sequence aware INLINEFORM5 INLINEFORM6 based on the utterances INLINEFORM7 and the dialogue act sequence INLINEFORM8 . We assume the attentive distribution INLINEFORM9 , where we condition INLINEFORM10 on the input utterances INLINEFORM11 and the dialogue act sequence INLINEFORM12 . Here we assume the utterances in the conversation as an undirected graph structure with INLINEFORM13 vertices. The CRF is parameterized with clique potentials INLINEFORM14 , indicating the subset of INLINEFORM15 give by clique INLINEFORM16 . Under this definition, the attention probability is defined as INLINEFORM17 . For symmetry, we use the softmax in a general sense, i.e. INLINEFORM18 , where INLINEFORM19 is the implied recognition function. Here INLINEFORM20 comes from the former memory enhanced deep model over utterances INLINEFORM21 and corresponding dialogue acts INLINEFORM22 . The INLINEFORM0 INLINEFORM1 over the utterances and dialogue acts is defined as expectation: INLINEFORM2 where we assume the annotation function INLINEFORM0 factors into INLINEFORM1 . The annotation function is defined to simply return the selected hidden state. The INLINEFORM2 INLINEFORM3 can be interpreted as an dialogue act aware attentive conversation as taking the expectation of the annotation function with respect to a latent variable INLINEFORM4 , where INLINEFORM5 is parameterized to be function of utterances INLINEFORM6 and dialogue acts INLINEFORM7 . The expectation is a linear combination of the input representation and represents how much attention will be focused on each utterance according to the dialogue act sequence. We can model the structural dependencies distribution over the latent INLINEFORM0 with a linear chain CRF with n states: INLINEFORM1 where INLINEFORM0 is the pairwise potential for INLINEFORM1 and INLINEFORM2 . Notice that the utterance INLINEFORM3 and the dialogue act sequence INLINEFORM4 are both obtained from downstream learned representation. The marginal distribution INLINEFORM5 can be calculated efficiently in linear time via the forward-backward algorithm. These marginals further allow us to implicitly sum over the linear chain conditional random field. We refer to this type of attention layer as a INLINEFORM6 INLINEFORM7 INLINEFORM8 , where we can explicitly look into the undirected graphical CRF structure to find which utterances are in the same chunk or in isolation. Here we define the node potentials with a unary CRF setting: INLINEFORM0 where for each utterance we summarize the possible dialogue act to perform sequential reasoning. Given the potential, we compute the structural marginals INLINEFORM0 using the forward-backward algorithm, which is then used to compute the final probability of predicting the sequence of dialogue acts as: INLINEFORM1 ### End-to-End Training
We adopt the maximum likelihood training estimation to learn the CRF-attentive structured parameters. Given the training set INLINEFORM0 with INLINEFORM1 conversation pairs, the log likelihood can be written as: INLINEFORM2 where we denote the INLINEFORM0 as the set of parameters within neural networks from hierarchical layers: word embedding layer, memory enhanced utterance modeling layer, CRF-attentive structured layer. We define the objective function in training process: DISPLAYFORM0 INLINEFORM0 is a hyper-parameter to trade-off the training loss and regularization. By using SGD optimization with the diagonal variant of AdaGrad, at time step t, the parameter INLINEFORM1 is updated as follows: DISPLAYFORM0 where INLINEFORM0 is the initial learning rate and INLINEFORM1 is the sub-gradient at time t. Notice that one of our contributions is to apply CRF structural attention as the final layer of deep models. The whole model can be trained in an end-to-end manner. Here we consider the standard Viterbi algorithm for computing the distribution INLINEFORM0 . The main procedure is summarized in Algorithm 1. For testing, we adopt Viterbi algorithm to obtain the optimal sequence by using dynamic programming techniques. The testing procedure can be written as: INLINEFORM0 [t] Viterbi algorithm for CRF-ASN [1] The observation space INLINEFORM0 The state space INLINEFORM0 The observation sequence INLINEFORM0 The probabilities INLINEFORM0 The most likely hidden state sequence INLINEFORM0 Construct transition matrix INLINEFORM0 , each element stores the transition probability of transiting from state INLINEFORM1 to state INLINEFORM2 Construct emission matrix INLINEFORM3 , each element stores the probability of observing INLINEFORM4 from state INLINEFORM5 each state INLINEFORM6 INLINEFORM7 INLINEFORM8 each observation INLINEFORM9 each state INLINEFORM10 INLINEFORM11 INLINEFORM12 INLINEFORM13 INLINEFORM14 INLINEFORM15 INLINEFORM16 INLINEFORM17 X ### Experiments
In this section, we conduct several experiments on two public DA datasets SwDA and MRDA, and show the effectiveness of our approach CRF-ASN for dialogue act recognition. ### Data Preparation
We evaluate the performance of our method on two benchmark DA datasets: Switchboard Dialogue Act Corpus (SwDA) and The ICSI Meeting Recorder Dialogue Act Corpus (MRDA). These two datasets have been widely used to conduct the dialogue act recognition or the dialogue act classification tasks by several prior studies. SwDA: Switchboard Dialogue Act Corpus is a large hand-labeled dataset of 1155 conversations from the Switchboard corpus of spontaneous human-to-human telephone speech. Each conversation involved two randomly selected strangers who had been charged with talking informally about one of several, self-selected general interest topics. For each utterance, together with a variety of automatic and semiautomatic tools, the tag set distinguishes 42 mutually exclusive utterance types via DAMSL taxonomy. The top five frequent DA types include STATEMENT, BACKCHANNEL / ACKNOWLEDGE, OPINION, ABANDONED / UNINTERPRETABLE, AGREEMENT / ACCEPT. We list the top five percentages of utterance type in the overall corpus in table2. MRDA: The ICSI Meeting Recorder Dialogue Act Corpus consists of hand-annotated dialog act, adjacency pair, and hotspot labels for the 75 meetings in the ICSI meeting corpus. The MRDA scheme provides several class-maps and corresponding scripts for grouping related tags together into smaller number of DAs. In this work we use the most widely used class-map that groups all tags into 5 DAs, i.e., Disruption (D) indicates the current Dialogue Act is interrupted. BackChannel (B) are utterances which are not made directly by a speaker as a response and do not function in a way that elicits a response either. FloorGrabber (F) are dialogue acts for grabbing or maintaining the floor. Question (Q) is for eliciting listener feedback. And finally, unless an utterance is completely indecipherable or else can be further described by a general tag, then its default status is Statement (S). We respectively list the percentage of the five general dialogue acts in table 3. From the table 2 and table 3, we can see the datasets are highly imbalanced in terms of label distributions. The dialogue act type STATEMENT occupies the largest proportion in both two datasets. Following the second place is the BACKCHANNEL act type which somewhat reflect the speaker's speech style. We present the detailed data preparation procedure for obtaining the clear dataset. For two datasets, we performed pre-processing steps in order to filter out the noise and some informal nature of utterances. We first strip the exclamations and commas, and then we convert the characters into lower-case. Notice that for SwDA, we only get the training and testing datasets. In order to smooth the training step and tune the parameters, we depart the original training dataset into two parts, one for training and the other small part used to be the validation set. We list the detailed statistics of the two datasets in table 4. ### Evaluation Criteria
We mainly evaluate the performance of our proposed CRF-ASN method based on the widely-used evaluation criteria for dialogue act recognition, Accuracy. The Accuracy is the normalized criteria of accessing the quality of the predicted dialogue acts based on the testing utterance set INLINEFORM0 . Given the testing conversation INLINEFORM1 with its ground-truth dialogue acts INLINEFORM2 , we denote the predicted dialogue acts from our CRF-ASN method by INLINEFORM3 . We now introduce the evaluation criteria below. INLINEFORM4 ### Implemental Details
We preprocess each utterance using the library of nltk BIBREF11 and exploit the popular pretrained word embedding Glove with 100 dimensional vectors BIBREF12 . The size of char-level embedding is also set as 100-dimensional and is obtained by CNN filters under the instruction of Kim BIBREF13 . The Gated Recurrent Unit BIBREF14 which is variant from LSTM BIBREF15 is employed throughout our model. We adopt the AdaDelta BIBREF16 optimizer for training with an initial learning rate of 0.005. We also apply dropout BIBREF17 between layers with a dropout rate of 0.2. For the memory network enhanced reasoning, we set the number of hops as 1 to preliminary learn the contextual dependencies among utterances. We do not set too many hops as increasing the number of GRU layers reduced the accuracy of the model. Early stopping is also used on the validation set with a patience of 5 epochs. Conversations with the same number of utterances were grouped together into mini-batches, and each utterance in a mini-batch was padded to the maximum length for that batch. The maximum batch-size allowed was 48. During training, we set the moving averages of all weights as the exponential decay rate of 0.999 BIBREF18 . The whole training process takes approximately 14 hours on a single 1080Ti GPU. All the hyper-parameters were selected by tuning one hyper-parameter at a time while keeping the others fixed. ### Performance Comparisons
We compare our propose method with other several state-of-the-art methods for the problem of dialogue act recognition as follows: Bi-LSTM-CRF BIBREF19 method builds a hierarchical bidirectional LSTM as a base unit and the conditional random field as the top layer to do the dialogue act recognition task. DRLM-Conditional BIBREF20 method combines postive aspects of neural network architectures with probabilistic graphical models. The model combines a recurrent neural network language model with a latent variable model over shallow discourse structure. LSTM-Softmax BIBREF0 method applies a deep LSTM structure to classify dialogue acts via softmax operation. The authors claim that the word embeddings, dropout, weight decay and number of LSTM layers all have large effect on the final performance. RCNN BIBREF8 method composes both sentence model and discourse model to extend beyond the single sentence. The authors propose hierarchical CNN on sentence model and RNN on the contextual discourses. CNN BIBREF21 method incorporates the preceding short texts to classify dialogue act. The authors demonstrate that adding sequential information improves the quality of the predictions. HMM BIBREF5 method treats the discourse structure of a conversation as a hidden Markov model and the individual dialogue acts as observations emanating from the model states. CRF Simple baseline which applies the text encoding and CRF-based structure prediction on the DAR problem. SVM Simple baseline which applies the text encoding and multi-classification algorithm on the DAR problem. Among them, The former five approaches eg. Bi-LSTM-CRF, DRLM-Conditional, LSTM-Softmax, RCNN, CNN all adopt the deep neural network model in order to better capture the utterances semantic representations. The latter three methods (HMM, CRF, SVM) just employ the simple feature selection on the text processing. About half of the baselines including Bi-LSTM-CRF, DRLM-Conditional, HMM, CRF consider the graphical structured prediction while the others eg. RCNN, CNN, LSTM-Softmax, SVM just adopt the traditional multi-classification algorithms. Table 5 and Table 6 respectively show the experimental Accuracy results of the methods on the SwDA and MRDA datasets. The hyper-parameters and parameters which achieve the best performance on the validation set are chosen to conduct the testing evaluation. The experiments reveal some interesting points: The results show that our proposed model CRF-ASN obviously outperforms the state-of-the-art baselines on both SwDA and MRDA datasets. Numerically, Our model improves the DAR accuracy over Bi-LSTM-CRF by 2.1% and 0.8% on SwDA and MRDA respectively. It is remarkable that our CRF-ASN method is nearly close to the human annotators' performance on SwDA, which is very convincing to prove the superiority of our model. The deep neural networks outperform the other feature-based models. We can see the last three non-deep models obtain worse performance than the top five deep-based methods. This suggests that the performance of dialogue act recognition can be improved significantly with discriminative deep neural networks, either in convolutional neural network or the recurrent neural network. Apart from deep learning tactics, the problem formulations are also critical to the DAR problem. We see structured prediction approaches eg. CRF-ASN, Bi-LSTM-CRF obtain better results than multi-classification eg. LSTM-Softmax. What's more, under the same text encoding situation, the CRF-based model achieves much better results than the SVM-based method. Which can fully prove the superiority of the structured prediction formulation. We also notice that CRF is better than HMM when adopted to the DAR task. The major differences between our proposed model CRF-ASN and the strong baseline BI-LSTM-CRF lie in two aspects: First we adopt a more fine grained manner to encode the utterances and utilize the memory enhanced mechanism to compute the contextual dependencies. Second we employ an adapted structured attention network on the CRF layer, rather than directly apply the original CRF on the utterances. These two modifications are essential and improve the performance significantly. ### Ablation Results
We respectively evaluate the individual contribution of the proposed module in our model. We conduct thorough ablation experiments on the SwDA dataset, which are recorded on the table 7. To make it fair, we only modify one module at a time and fix the other components to be in the same settings. We replace the proposed structured CRF-attention layer to simple CRF, the results show structured CRF-attention layer results in major improvement in the accuracy, approximately over 2.1% absolute points. We further replace the structure prediction formulation to multi-classification on SVM, the results drop dramatically, which illustrate the benefit of considering structural dependencies among utterances. We replace the fine-grained word INLINEFORM0 to the simple Glove vector. The results suggest that fine grained word embedding is useful to represent a text. We also adapt the context state INLINEFORM1 to only care its neighbor utterances. The result is not satisfying, which conveys us that the basic text understanding is critical in the semantic representations. We replace the memory network to directly apply CRF layer to the utterance layer. We also conduct a comparing experiment which plus the original utterance to memory enhanced output. The two results show the designed hierarchical memory-enhanced components are helpful in the utterance understanding and modeling the contextual influence. ### Visualization
In Figure 3, we visualize of the output edge marginals produced by the CRF-ASN model for a conversation. In this instance, the actual dialogue act recognition procedure is displayed as INLINEFORM0 . In the testing step, the model is uncertain and select the most attentive path to maximize the true dialogue act recognition. Here we can see from the marginal edges the path INLINEFORM1 occupies more attentive weights than the path INLINEFORM2 in predicting the dialogue act label. Thus we ultimately select the right way to recognize the dialogue act. Figure 4 shows the confusion heatmap of our proposed CRF-ASN model for the SwDA dataset. Each element in the heatmap denotes the rate that the predicted label is the same to the true label. We can see from the diagonal, the <sd,sd> <b,b> pairs achieve the most satisfying matching score while <qyd, qyd> is much worse than other pairs. This can be explained that the sd (statement) and b(acknowledge) have clearly self-identification while qyd(Declarative Yes-No-Question) is more easier to be mistakenly recognized. We can see that <qyd,qy> which represents (Declarative Yes-No-Questio,Yes-No-Question) is indeed hard to recognize since their dialogue type are too similar with each other. For another reason, we notice that due to the bias of the ground truth, there are some cases that we predict the dialogue act correctly while the ground truth is wrong. To some reason, classifying so many fine-grained dialogue act labels is not easy for human annotators, besides the human-subjectivity occupies an important role in recognizing the dialogue act. ### Related Work
In this section, we briefly review some related work on dialogue act recognition and attention network. ### Dialogue Act Recognition
The main task of dialogue act recognition is to assign an act label to each utterance in a conversation, which can be defined as a supervised problem due to the properties that each utterance has a corresponding act label. Most of the existing work for the problem of dialogue act recognition can be categorized as following two groups. Regarding the DAR as a multi-classification problem. Reithinger et al. BIBREF22 present deal with the dialogue act classification using a statistically based language model. Webb et al. BIBREF23 apply diverse intra-utterance features involving word n-gram cue phrases to understand the utterance and do the classification. Geertzen et al. BIBREF24 propose a multidimensional approach to distinguish and annotate units in dialogue act segmentation and classification. Grau et al. BIBREF3 focus on the dialogue act classification using a Bayesian approach. Serafin et al. BIBREF25 employ Latent Semantic Analysis (LSA) proper and augmented method to work for dialogue act classification. Chen et al. BIBREF26 had an empirical investigation of sparse log-linear models for improved dialogue act classification. Milajevs et al. BIBREF27 investigate a series of compositional distributional semantic models to dialogue act classification. Regarding the DAR as a sequence labeling problem. Stolcke et al. BIBREF5 treat the discourse structure of a conversation as a hidden Markov model and the individual dialogue acts as observations emanating from the model states. Tavafi et al. BIBREF28 study the effectiveness of supervised learning algorithms SVM-HMM for DA modeling across a comprehensive set of conversations. Similar to the SVM-HMM, Surendran et al. BIBREF29 also use a combination of linear support vector machines and hidden markov models for dialog act tagging in the HCRC MapTask corpus. Lendvai et al. BIBREF30 explore two sequence learners with a memory-based tagger and conditional random fields into turn-internal DA chunks. Boyer et al. BIBREF31 also applied HMM to discover internal dialogue strategies inherent in the structure of the sequenced dialogue acts. Galley et al. BIBREF32 use skip-chain conditional random field to model non-local pragmatic dependencies between paired utterances. Zimmermann et al. BIBREF33 investigate the use of conditional random fields for joint segmentation and classification of dialog acts exploiting both word and prosodic features. Recently, approaches based on deep learning methods improved many state-of-the-art techniques in NLP including DAR accuracy on open-domain conversations BIBREF7 BIBREF34 BIBREF6 BIBREF35 BIBREF21 . Kalchbrenner et al. BIBREF7 used a mixture of CNN and RNN. CNNs were used to extract local features from each utterance and RNNs were used to create a general view of the whole dialogue. Khanpour et al. BIBREF0 design a deep neural network model that benefits from pre-trained word embeddings combined with a variation of the RNN structure for the DA classification task. Ji et al. BIBREF6 also investigated the performance of using standard RNN and CNN on DA classification and got the cutting edge results on the MRDA corpus using CNN. Lee et al. BIBREF21 proposes a model based on CNNs and RNNs that incorporates preceding short texts as context to classify current DAs. Zhou et al. BIBREF34 combine heterogeneous information with conditional random fields for Chinese dialogue act recognition. Kumar et al. BIBREF35 build a hierarchical encoder with CRF to learn multiple levels of utterance and act dependencies. Unlike the previous studies, we formulate the problem from the viewpoint of integrating contextual dependencies in both utterance level and the act label level. We not only consider the fine grained multi-level semantic representations, but also integrate the structured attention network to further capture the structure designpendencies in the CRF layer. ### Attention Network
Attention mechanism has become an essential component in text understanding in recent years. Since the first work proposed by Bahdanau et al. BIBREF36 that adopt the attention mechanism in neural machine translation, attention mechanism based neural networks have become a major trend in diverse text researching field, such as in machine comprehension BIBREF37 BIBREF38 BIBREF39 BIBREF40 , machine translation BIBREF41 BIBREF42 , abstract summarization BIBREF43 BIBREF44 , text classification BIBREF45 BIBREF46 BIBREF47 and so on. The principle of attention mechanism is to select the most pertinent piece of information, rather than using all available information, a large part of it being irrelevant to compute the neural response. In our work, we propose the CRF-attentive structured network in order to encode the internal utterance inference with dialogue acts. The structured attention is a more general attention mechanism which take account of the graphical dependencies and allow for extending attention beyond the standard soft-selection approach. The most similar work to our model is proposed by Kim et al. BIBREF48 . Kim et al. also experiment with two different classes of structured attention networks: subsequence selection and syntactic selection. However, the objectives of these two networks aims to segment the structure dependencies, which are quite different from our DAR task. In DAR task we care more on the dialogue act influences on the overall conversation structure, thus the former structured attention may not be suitable for our problem. ### Conclusion
In this paper, we formulate the problem of dialogue act recognition from the viewpoint of capturing hierarchical rich utterance representations and generalize richer CRF attentive graphical structural dependencies without abandoning end-to-end training. We propose the CRF-Attentive Structured Network (CRF-ASN) for the problem. We implement the model in two steps. We first encode the rich semantic representation on the utterance level by incorporating hierarchical granularity and memory enhanced inference mechanism. The learned utterance representation can capture long term dependencies across the conversation. We next adopt the internal structured attention network to compute the dialogue act influence and specify structural dependencies in a soft manner. This approach enable the soft-selection attention on the structural CRF dependencies and take account of the contextual influence on the nearing utterances. We demonstrate the efficacy of our method using the well-known public datasets SwDA and MRDA. The extensive experiments demonstrate that our model can achieve better performance than several state-of-the-art solutions to the problem. Table 1: A snippet of a conversation sample. Each utterance has related dialogue act label. Figure 1: An illustration of the hierarchical conversation structure. The input of the model is a conversation which consist of n utterances u1,u2, ...,un with corresponding dialogue act labels a1,a2, ...,an . Each utterance is composed of diverse length of words in the character level Ec and the word level Ew . Notice that utterances are not exist independent, utterances have contextual relations with each other. Figure 2: The Overview of learning memory enhanced hierarchical conversation representation architecture. The momory hop is set to 1. First concatenate the rich word embedding and obtain the original utterance representation ut from the basic BiGRU. The hidden state ht represents the contextual encoding which cares the former and the latter utterance dependencies. After summarizing hierarchical memory enhanced output ot and the original utterance ut , we get the final representation ut denoted in a bold form. Table 2: Top five percentages of utterance type in the SWDA corpus Table 3: Top five percentages of utterance type in the MRDA corpus Table 4: |C | is the number of Dialogue Act classes, |V | is the vocabulary size. Training, Validation and Testing indicate the number of conversations (number of utterances) in the respective splits. Table 5: Comparing Accuracy of our method (CRF-ASN) with other methods in the literature on SwDA dataset. Table 6: Comparing Accuracy of our method (CRF-ASN) with other methods in the literature on the MRDA dataset. Table 7: Component ablations on SwDA dataset Figure 3: Visualization of the structured attention distribution over conditional random field sequence. For example, the edges represent the marginal probabilities p(z |u,Θ), and the nodes represent the utterances and corresponding dialogue acts. In this figure we can see For utterance u4, dialogue a3 is the most suitable predicting label as the edge (4→ 5→ 3) is the most attentive path. Figure 4: Confusion heatmap of CRF-ASN model for the SwDA dataset. There are totally 10 DA labels, where the row denotes the true label and the column denotes the predicted label.
|
improves the DAR accuracy over Bi-LSTM-CRF by 2.1% and 0.8% on SwDA and MRDA respectively
|
Whose dog was thrown a birthday party? What is the article doing with this detail?
A. Thomas Maier’s dog. The article uses this to demonstrate how Condé Nast has become a successful in group.
B. S.I. Newhouse Jr.’s dog. The article uses this to demonstrate the absurd expenditure of the Condé Nast magazines.
C. S.I. Newhouse Jr.’s dog. The article uses this to demonstrate that the absurd expenditure of the Condé Nast has a kind side
D. Thomas Maier’s dog. The article uses this to demonstrate the absurd expenditure of the Condé Nast magazines.
|
Let Si Get This During a typical lunch time at the Royalton Hotel restaurant in midtown Manhattan, The New Yorker 's Tina Brown might be installed at her usual table, and Vogue 's Anna Wintour might be at her usual table (chewing on her usual meal--a $25 hamburger). Vanity Fair 's Graydon Carter might be there too, although he has transferred his main allegiance to a place called Patroon. Filling out the room are other editors, publicists, and writers from these magazines and GQ and House & Garden and so on. And one man, who probably isn't there himself, picks up every tab. Some of the lesser fry may even utter the Condé Nast mantra--though it is hardly necessary at the Royalton--as they grab for the check: "Let Si get this." S.I. "Si" Newhouse Jr. and his younger brother, Donald, control Advance Publications, one of America's largest privately held companies. (Estimate of their combined wealth: $13 billion.) Donald tends to Advance's hugely profitable newspaper, radio, and TV holdings. Si runs the less profitable but more glamorous properties. These are the 15 Condé Nast magazines, including (in descending order of fabulousness) Vogue , Vanity Fair , GQ , Condé Nast Traveler , House & Garden , Allure , Details , Self , Mademoiselle , and Glamour ; ; and Random House. The expense-account lunch is a hallowed journalistic tradition. But consider a day in the life of an editor working for Si Newhouse. (Donald's editors are a different story, as they will be happy to tell you.) It's a closed economy where almost all human needs and desires can be gratified with a miraculous, unlimited currency called the Si. A Lincoln Town Car is waiting outside your door in the morning to take you to work. The car, which costs $50 an hour, is written into your contract. First stop, breakfast with a writer at the Four Seasons. The check may be as little as $40. When you reach the office, you realize you're out of cigarettes. No problem--you send your assistant to buy a pack for you. She gets reimbursed from petty cash ($3). (Could be worse for the assistant: She could be forced to pick up her boss's birth-control pills, or her boss's pet from the vet, or presents for her boss's children--regular duties for Condé Nast underlings.) You've forgotten to return the video your kids watched yesterday, so you have a messenger take it back to Blockbuster. Si spends $20; you save a $1.50 late fee. Then there's lunch. The magazines account for more than a quarter of daytime revenues at the Four Seasons and the Royalton. A modest lunch for two at the Royalton (no fancy wine or anything) might cost $80. But Si's generosity extends to even assistants and sub-sub-editors, dining on sushi at their desks. If you spend $10 or less on lunch, and claim you were working, Si pays. At Vogue and Vanity Fair , almost everyone has a "working lunch" every day . An editor at Allure says that "working lunches" there are limited to 10 a month. Back at the office, you hear that a friend at another Newhouse magazine has been promoted, so you send flowers. The tab: $100. Si pays. (One of my favorite Condé Nast stories is of an editor who had just been promoted to an extremely senior job. His office was jammed with congratulatory flowers and cards. All had been sent by fellow Condé Nast staffers. All had been billed to the company.) Four o'clock, and it's snack time. Your assistant joins the mob in the lobby newsstand. She bills your candy bar, juice, and cigarettes (as well as her own candy bar, juice, and cigarettes) to the magazine ($15). After all, it's a "working snack." Later, there's a birthday party for your assistant. You order champagne and a cake--on the company, of course, and present her with your gift--a Prada wallet ($200). Later, she submits the expense sheet for it. Finally, after a Random House book party at Le Cirque 2000 (estimated cost to Si: $35,000), your car ferries you home. Newhouse expense stories are a staple of New York literary-journalistic conversation. Stories about the $10,000 in expenses that a New Yorker editor billed for a single month. About the interior-decorating costs for the fashion-magazine editor who likes to have her office photographs rearranged every few months. About the hotel tab for the big-name New York writer who spent three weeks in Washington's Hay-Adams (basic room: $285 a night) researching a Vanity Fair story that will never run. About the Vogue editor who has furnished her summer house from items purchased for fashion shoots--beautiful furniture, designer pillows, coffee-table books. Vogue assistants have nicknamed the house "Petty Cash Junction." None of the 39 past and present Newhouse employees I spoke to for this story would talk on the record, for . And the nature of the subject makes it hard to separate apocrypha from the truth. Did Condé Nast pay, as sources insist it did, hundreds of thousands of dollars in back taxes on behalf of an editor who didn't bother to file tax forms? Did an editor really expense $20,000 in a weeklong trip to Paris? The people who pay the bills are not talking. But every example of excess cited here was told to me by at least one source (and usually more than one) in a position to know. Need a facial? Treat yourself and bill it to Si. This is what is called "scouting." It is also a great way to get free haircuts. To be fair, Si doesn't pay for all such treats. There is also a much-honored tradition of accepting tribute from companies that Condé Nast magazines cover. One magazine exec reportedly got so much loot last Christmas--Cuban cigars, "crates of wine," designer suits ("It was like a Spanish galleon")--that he needed three cars to cart it home. At yuletide, even midlevel fashion-mag writers and editors are inundated with "cashmere sweaters, Versace pillows, coats ..." recalls one ex- Vogue staffer wistfully. At the top of the masthead, the perks are perkier. His Si-ness (their joke, not mine) does not expect his editors in chief to actually live on their million-dollar salaries. He also gives them clothing allowances (up to $50,000 a year). He buys them cars of their choice and hires chauffeurs to drive them. He offers them low- or no-interest home loans. GQ editor Art Cooper reportedly received two $1-million loans, one for a Manhattan apartment, the other for a Connecticut farm. Tina Brown and her husband, Harold Evans, former president of Random House, reportedly just took a $2-million boost to buy a $3.7-million Manhattan house. Si's favorite courtiers lead lives of jaw-dropping privilege. When she was editor of British Vogue , Wintour commuted between London and New York--on the Concorde. Another Si confidant decided his office didn't feel right, so he hired one of the grandmasters of feng shui to rearrange it. Some editors prepare for trips by Federal Expressing their luggage to their destination. Why? "So you don't have to carry your bags. No one would be caught dead carrying a bag." Condé Nast has also created a class of mandarin journalists, writers who live much better than they ever could if they wrote only for normal magazines. One free-lancer tells of building much of a summer traveling with her husband in the West and Europe around a couple of Condé Nast assignments. Last summer, The New Yorker sent a staffer to Venice to cover the Venice Film Festival. The weeklong trip, which must have cost thousands, resulted in a short piece. Writers, of course, are nowhere near as profligate as photographers. Stories of wasteful shoots abound: the matching seaweed that had to be flown from California to the Caribbean for a fashion photo; the Annie Liebovitz Vanity Fair cover shot of Arnold Schwarzenegger that reportedly cost $100,000; the Vogue shoot in Africa in which, an ex- Vogue editor claims, the photographer and his huge entourage wined and dined to the tune of "hundreds of thousands of dollars." And then there are the parties. Last month The New Yorker spent--and this is not a joke--$500,000 on a two-day "Next Conference" at the Disney Institute in Florida, in connection with a special issue on the same theme. In order to get Vice President Gore, who was traveling in California at the time, The New Yorker paid for him and his entourage to fly Air Force Two from California to Florida and back. And vice presidents are not the only things that Condé Nast flies in for parties. The New Yorker once shipped silverware from New York to Chicago for a dinner. ("What, they don't have silverware in Chicago?" asks a New Yorker staffer.) Vanity Fair toted food from New York to Washington for this year's party on the night of the White House Correspondents Dinner. (What, they don't have food in Washington?) That annual Washington do has grown from an after-dinner gathering for drinks at a contributor's apartment to two huge blasts--before and after the dinner itself--at a rented embassy. VF 's annual Oscar-night party has become a similar institution in Hollywood. In addition to the parties themselves, Si also naturally pays to fly in VF staffers and to put them up at top hotels. (What, they don't have editors in Washington or L.A.?) Some Condé Nast parties are so ridiculous that even other Condé Nasties make fun of them. This week's New Yorker , for example, mocks a recent Vogue party in honor of food writer Jeffrey Steingarten. According to The New Yorker , Wintour so detested the carpet at Le Cirque 2000 that she ordered the florist to cover it with autumn leaves (handpicked, of course). The apogee of party absurdity is Vanity Fair 's sponsorship of an annual London dinner for the Serpentine Museum in Hyde Park. As one observer puts it, "Vanity Fair , an American magazine, pays more than $100,000 to a British art museum solely so that it can sponsor a dinner where Graydon Carter gets to sit next to Princess Diana." The princess was the museum's patron. Actually, paying $100,000 for face time with Princess Di may not have been a foolish investment for a magazine so dependent on peddling her image. And Condé Nast's excess has other plausible justifications as well. Some top editors may earn their perks. Vogue and GQ make millions, according to industry analysts. Vanity Fair is enjoying banner years, and while it probably hasn't made back the millions Newhouse lost in starting it up, it is certainly in the black. The New Yorker loses money--how much may even surpass perks as a topic of Newhouse gossip and speculation. On the other hand, The New Yorker is the most talked-about magazine in America, and Tina Brown is the most talked-about editor. That is worth something. Public media companies such as Time Warner (or, for that matter, Microsoft) can entice and hold journalists with stock options. Advance is private, so Newhouse uses other golden handcuffs. He runs a lifestyle prison. Top editors stay because they could never afford to live in a house as nice as the one Si's interest-free loan bought them, or to host parties as nice as the ones Si's party planners throw for them. Condé Nast's magazines are all about glamour, wealth, prestige. To uphold that image, magazine editors need to circulate at the top of New York society. But the top of New York society consists of people who make far more money than magazine editors do--investment bankers, corporate chieftains, and fashion designers. Million-dollar salaries aren't enough to mix as equals with the Trumps and Karans. Si's perks are equalizers. And they say it's not as good as it used to be. In 1992, according to Thomas Maier's biography of Newhouse, the editor of Self held a birthday party for Si Newhouse's dog . (Owners ate caviar; dogs drank Evian.) The lowliest assistants used to take car services home. But new Condé Nast CEO Steve Florio has restricted cars and catering. Editors who used to fly the Concorde now fly first-class; those who used to fly first-class now fly business. Expense accounts are scrutinized. Even so, today's Condé Nast is economical only by Condé Nast standards. The belt is tighter, but it's still hand-tooled, hand-tanned, and fashioned from the finest Italian leather.
|
B. S.I. Newhouse Jr.’s dog. The article uses this to demonstrate the absurd expenditure of the Condé Nast magazines.
|
What, according to the author, is the main flaw of The Slums of Beverly Hills?
A. The female characters are reduced to naive, sex-obsessed girls, when they are much more complex in reality
B. The director too obviously uses the film as an outlet for resolving her own childhood devastations
C. It is difficult for the audience to make sense of the director's absurd juxtapositions
D. The audience never gets to see the children interact within the context that motivates their father to uproot their lives
|
Dirty Laundry Now and then, a documentary film comes along that makes us re-examine the rules that unofficially govern the genre: Can there be a middle ground between fiction and fact? Can a documentary use scripted scenes and yet remain ontologically authentic? How much can you stylize material before you alter the reality that you're striving, at least in theory, to capture? Unmade Beds , Nicholas Barker's " 'real life' feature film," has proudly worn its mongrel status as a "directed" documentary of single life in the big city, employing, in the face of criticism, what amounts to a cackling-punk defiance. The movie tracks four aging New Yorkers--two men, two women--through their lonely dating rituals, in the process depicting a universe of lusty, coupled-up haves and downcast, excluded have-nots, all viewed Rear Window -style through rectangular openings in the massive apartment houses in which they reside. This is not cinema vérité , and nothing has been left to chance. The director selected his four subjects from many hundreds of potential candidates, followed them around for months, and then scripted their monologues and dialogues to reflect what he says he saw. Calling his own film "an exercise in mendacity," Barker goes on, "I'm quite happy to tell lies about my characters and even collude with their self-delusions if it enables me to communicate larger dramatic truths." Spurned by U.S. distributors, Unmade Beds opened two weeks ago in a small screening room in downtown Manhattan, where it proceeded to set box office records and generate lots of (largely favorable) press. In part due to smart publicity, which has bannered some of the bad reviews and commentary ("I have to tell you that this film upset me so much that I really don't want to have anything to do with it"--a New York publicist), it threatens to become a cause célèbre --and to be coming soon to a theater near you. It's always nice to see distributors proved wrong about the merits of "difficult" films, but in this case I think they did the decent thing. Unmade Beds isn't just bad--it's obnoxiously, noxiously bad, a freak show for the empathetically challenged. The outrage it has prompted isn't the Puritan kind; it's more like legitimate revulsion at watching a blowhard pervert people's lives in the name of "larger dramatic truths." Those truths are large, all right. Take Michael, the 40-year-old, 5 foot 4 inch lonely guy who has been looking for a wife for almost two decades. If you were to walk past him on the street, you might think that a man of his small stature might have some trouble getting dates and be rather bitter about it. The larger dramatic truth is that Michael has lots of trouble getting dates and is very bitter about it. Just in case you feel too sorry for him, however, Barker is careful to include a homophobic monologue in which Michael complains about young women who waste their lives hanging out with effeminate males. Michael turns out to be the film's most sympathetic subject--by a wide margin. At least he's not Mikey, a paunchy 54-year-old who writes but can't sell screenplays and who always flees blind dates, because the women he gets fixed up with are "mutts." Sounding like one of the low-level gangsters who posture like kingpins in Donnie Brasco , Mikey talks a lot about mutts. He also reminisces about that 24 hour period in the '70s when he managed to sleep with three different beautiful women, whose pictures he shows off. These days, all he meets are mutts. He comes off as a pathetic little loser--a mutt. Aimee, on the other hand, is a pathetic big loser, weighing in at 225 pounds. Determined to get married before she turns 30, she generally is filmed beside bags of groceries and assorted junk foods. She cries about her situation to her thin friend, Laurie, who, in one scene, gently mentions Aimee's weight. Clearly the scene is scripted, but Aimee does a good job acting taken aback. She has always been fat--and she's "OK with it," and a man just has to accept it. This is followed by more talk about how you attract men. Will they respect you if you call them back? If you express too much interest? "Or," the viewer thinks, "if you're 225 pounds?" The only natural performer here is Brenda, a garrulous exhibitionist who blossoms with the camera on her--she could have a career as a Penny Marshall-style character actress. Divorced and aging, Brenda needs money and is willing to charge for her sexual services. It shouldn't be too difficult, because men are always showing her their dicks ("I'm up to two dicks a day"). They meet her and, a few minutes later, they show her their dicks. Weird, huh? What Barker leaves out (it's in a New York Observer article) is that Brenda, a former lap dancer, works in marketing at a strip joint. Presumably, men standing next to her in line at McDonald's don't show her their dicks. Nor, presumably, does she show them her breasts--although she bares them for Barker's camera, jabbering about her body while she doffs her clothes and steps into the shower and soaps up. Barker might have crafted his subjects' monologues from their own words, but he has robbed them of their spontaneity--and, thus, of their essence. They aren't thinking or trying to come to grips with their situations in front of your eyes, because they already know what they're going to say: They've been fixed like butterflies on the ends of pins and held up for voyeuristic inspection. The scenes with friends and confidantes have a crude, programmatic purpose. You can imagine the director composing a shot (the shots are tightly composed and elaborately lighted) and reminding them, "In this scene she points out that you should lose weight and you get shocked and defensive. Ready ... Action." Call me square, but I find this antithetical to the documentary spirit. An Englishman who trained as an anthropologist before going to work for BBC Television, Barker clearly made up his mind about his material before his cameras began to roll--so it's no surprise that it feels prechewed and predigested. When reality interfered (Brenda apparently did not go through with a marriage to an immigrant in search of a green card for $10,000, as she does on-screen), Barker brushed the truth aside as immaterial, following her up the steps of City Hall in her wedding dress because it was "true to her character." But what separates documentary from fiction is that real people are often more complicated, and more conflicted, than finished characters--as Brenda proved to be more (or, at least, other) than the sum of her parts. That's the kind of truth that reveals itself to documentary filmmakers after the fact, when they go over footage and discover unexpected patterns, dissonances, glimmers of a universe that's richer and messier than the one they set out to portray. So what are Barker's "larger dramatic truths"? Single people in big cities can be desperate. Single people fear they're going to die alone--unloved and unloving. People are judged and, in turn, judge others by how they look. Big news. One could argue, charitably, that the movie is meant to be prescriptive, that Barker intends for us to regard the ways in which his subjects delude themselves and thereby learn to see through our own self-delusions. But Barker hasn't concocted a larger dramatic structure that would hold those larger dramatic truths together and help us comprehend where these people went wrong. He dramatizes right up to the point where a dramatist would be expected to provide some insight--and then, hey, he's a documentarian. Unmade Beds might make a good date movie. There's little to argue about in its subjects' personalities--both males and females will find them repulsive--and the picture the film paints of single life in the big city is so bleak that you'll probably want to jump into bed with whoever is sitting next to you. Anything to keep from turning into one of those people. The Slums of Beverly Hills also walks a line between two genres, in this case coming-of-age sex comedy and autobiographical monologue. Tamara Jenkins, the writer and first-time director, has an eye for absurd juxtapositions that was obviously sharpened by the pain of her nomadic upbringing. Her protagonist (Natasha Lyonne) spends her teen-age years being shuttled with her two brothers from one cheap dive to another in the 90210 ZIP code, all because her egregiously unsuccessful father (Alan Arkin) wants them to be educated in the best schools. ("Furniture's temporary; education is permanent.") It's a major omission, then, that we never see those schools or the kids' interaction with their stable, well-to-do Beverly Hills counterparts. We can't tell if the father is, on some weird level, justified in his fervor, or whether he's screwing up his children--subjecting them to humiliation and robbing them of a sense of permanence--for no reason. Jenkins hasn't quite figured out how to shape her narrative, which is full of episodes that are there because they actually happened but that don't have a payoff. I almost wish she'd included more voice-over narration, more commentary on the things that, as a filmmaker, she hasn't learned to bring out. The Slums of Beverly Hills never gels, but it has a likable spirit, and it's exceedingly easy on the eye, with lots of pretty girls and wry evocations of '70s fashions and decor. The father, to obtain financial support from his wealthy brother (Carl Reiner), volunteers to take in his vaguely schizzy, dipsomaniacal niece (Marisa Tomei). She and her cousin compare breasts, play with vibrators, and talk in pig Latinish gibberish, but Jenkins never lets the proceedings get too sentimental: The whimsy is always cut with an acidic awareness of the family's desperation. "Are we middle-class now?" ask the children, hopefully, before another crisis sends them back into their van, cruising past the movie stars' mansions, in the mean streets of Beverly Hills. Grading on the steep curve established by summer blockbuster seasons past, these have turned out to be a pretty good few months at the movies. Even the commercial swill ( Deep Impact , Armageddon , The Mask of Zorro , Small Soldiers , Snake Eyes , Halloween: H20 ) has been of a high grade, and Saving Private Ryan and Return to Paradise were Vitalis slaps in the kisser for people woozy from all the warm weather escapism. Out of Sight was tender and charming, as was, in its gross-out way, There's Something About Mary . And, on the indie front, The Opposite of Sex , Buffalo 66 , and Pi have proved that there's still commercial life after Sundance. Sure, we had stinkers, but even Godzilla was fun to jeer at. And there's something reassuring about the fact that The Avengers is so rotten: proof yet again that people with piles of money can hire wizard production designers but can't fake class. I don't know who the credited screenwriter, Don MacPherson, is, but it's unlikely that he has ever seen an episode of the old Avengers , let alone sussed out the source of its appeal. Opening with a slapstick sequence of agent John Steed (Ralph Fiennes) doing kung fu, the film shifts to a scene in which he meets Mrs. Peel (Uma Thurman) while sitting naked in a sauna with only a newspaper to cover his private parts. The series was erotic in a way only prim English humor can be: The Old Boy Steed was capable of throwing a punch and bonking someone with his bowler, but he left the karate kicking to his liberated, leather-suited distaff associate. Here their roles have been witlessly muddled, and MacPherson's idea of banter is to have the pair complete each other's clichés. Whereas the original Steed, Patrick Macnee, was to the English Men's Club born, Fiennes is an eternal caddie. The willowy Thurman looks great in her outfits, but it's ever more apparent that she isn't much of an actress--at least, not a trained one--and her attempts at insouciance are embarrassingly arch. As the eccentric master villain who controls the weather, even Sean Connery is flat-out terrible, acting high on the hog. To think Connery once found the Bond films so far beneath him! When he sputters lines like "Time to die!" one imagines Dr. No, Goldfinger, and Blofeld snickering in the wings.
|
D. The audience never gets to see the children interact within the context that motivates their father to uproot their lives
|
How are the substitution rules built?
|
### Introduction
In text mining and Natural Language Processing (NLP), a lemmatizer is a tool used to determine the basic form of a word (lemma). Lemmatization differs from stemming in the way this base form is determined. While stemmers chop off word endings to reach the common stem of words, lemmatizers take into account the morphology of the words in order to produce the common morphological base form, i.e., the form of the word found in a dictionary. This type of text normalization is an important step in pre-processing morphologically complex languages, like Icelandic, before conducting various tasks, such as machine translation, text mining and information retrieval. To give an example from the Icelandic language, lemmatization helps find all instances of the personal pronoun ég “I” in a text corpus, taking into account all inflectional forms (ég, mig, mér, mín, við, okkur, and okkar). These variations of each word can be up to 16 for nouns and over a hundred for adjectives and verbs. The value of being able to reduce the number of different surface forms that appear for each word is therefore evident, as otherwise it is hard or even impossible to correctly determine word frequency in a corpus, or to look up all instances of a particular term. In this paper, we describe and evaluate Nefnir BIBREF0 , a new open source lemmatizer for Icelandic. Nefnir uses suffix substitution rules derived (learned) from the Database of Modern Icelandic Inflection (DMII) BIBREF1 , which contains over 5.8 million inflectional forms. This new lemmatizer was used for large-scale lemmatization of the Icelandic Gigaword Corpus BIBREF2 with promising results, but a formal evaluation had not been carried out. Our evaluation of Nefnir indicates that, compared to previously published results, it obtains the highest lemmatization accuracy of Icelandic, with 99.55% accuracy given correct part-of-speech (PoS) tags, and 96.88% accuracy given text tagged with a PoS tagger. ### Related work
The most basic approach to lemmatization is a simple look-up in a lexicon. This method has the obvious drawback that words that are not in the lexicon cannot be processed. To solve this, word transformation rules have been used to analyze the surface form of the word (the token) in order to produce the base form. These rules can either be hand-crafted or learned automatically using machine learning. When hand-crafting the rules that are used to determine the lemmas, a thorough knowledge of the morphological features of the language is needed. This is a time-consuming task, further complicated in Icelandic by the extensive inflectional system BIBREF1 . An example of a hand-crafted lemmatizer is the morphological analyzer that is part of the Czech Dependency Treebank BIBREF3 . Machine learning methods emerged to make the rule-learning process more effective, and various algorithms have been developed. These methods rely on training data, which can be a corpus of words and their lemmas or a large morphological lexicon BIBREF4 . By analyzing the training data, transformation rules are formed, which can subsequently be used to find lemmas in new texts, given the word forms. In addition, maching learning lemmatizers based on deep neural networks (DNNs) have recently emerged (see for example finnlem BIBREF5 for Finnish and LemmaTag BIBREF6 for German, Czech and Arabic). Along with the best rule-derived machine learning methods, these are now the state-of-the-art approaches to lemmatizers for morphologically complex languages. The biggest problem in lemmatization is the issue of unknown words, i.e. words not found in the training corpus or the underlying lexicon of the lemmatizer. This has been handled in various ways, such as by only looking at the suffix of a word to determine the lemma, thereby lemmatizing unseen words that (hopefully) share the same morphological rules as a known word BIBREF7 . DNN-based lemmatizers may prove useful in solving this issue, as they have their own inherent ways of handling these out-of-vocabulary (OOV) words, such as by using character-level context BIBREF8 . Previous to Nefnir, two lemmatization tools had been developed for Icelandic. We will now briefly mention these lemmatizers, before describing Nefnir further. ### CST Lemmatizer
The CST Lemmatizer BIBREF4 is a rule-based lemmatizer that has been trained for Icelandic on the Icelandic Frequency Dictionary (IFD) corpus, consisting of about 590,000 tokens BIBREF9 . This is a language-independent lemmatizer that only looks at the suffix of the word as a way of lemmatizing OOV words, and can be used on both tagged and untagged input. The authors of Lemmald (see Section SECREF2 ) trained and evaluated the CST Lemmatizer on the IFD and observed a 98.99% accuracy on correctly tagged text and 93.15% accuracy on untagged text, in a 10-fold cross-validation, where each test set contained about 60,000 tokens. Another evaluation of this lemmatizer for Icelandic BIBREF10 reports around 90% accuracy on a random sample of 600 words from the IFD, when the input has been PoS tagged automatically (with a tagging accuracy of 91.5%). The PoS tagger used was IceTagger BIBREF11 , which is part of the IceNLP natural language processing toolkit BIBREF12 . These results indicate that the accuracy of this lemmatizer is very dependent upon the tags it is given. To our knowledge, the Icelandic CST Lemmatizer model is not openly available. ### Lemmald
The second tool is Lemmald BIBREF13 , which is part of the IceNLP toolkit. It uses a mixed method of data-driven machine learning (using the IFD as a training corpus) and linguistic rules, as well as providing the option of looking up word forms in the DMII. Given correct PoS tagging of the input, Lemmald's accuracy measures at 98.54%, in a 10-fold cross-validation. The authors note that the CST Lemmatizer performs better than Lemmald when trained on the same data, without the added DMII lookup. The DMII lookup for Lemmald delivers a statistically significant improvement on the accuracy (99.55%), but it is not provided with the IceNLP distribution, so this enhancement is not available for public use. When used for lemmatization of the Icelandic Tagged Corpus (MÍM) BIBREF14 , the lemmatization accuracy of Lemmald was roughly estimated at around 90%. ### System Description
The main difference between Nefnir and the two previously described lemmatizers for Icelandic, CST Lemmatizer and Lemmald, is that Nefnir derives its rules from a morphological database, the DMII, whereas the other two are trained on a corpus, the IFD. Note that the IFD only consists of about 590,000 tokens, while the DMII contains over 5.8 million inflectional forms. Nefnir uses suffix substitution rules, derived from the DMII to lemmatize tagged text. An example of such a rule is (ngar, nkfn, ar INLINEFORM0 ur), which can be applied to any word form with the suffix ngar that has the PoS tag nkfn (a masculine plural noun in the nominative case), transforming the suffix from ar to ur. This rule could, for example, be applied to the word form kettlingar “kittens” to obtain the corresponding lemma, kettlingur. Words are lemmatized using the rule with the longest shared suffix and the same tag. Each inflectional form in the DMII is annotated with a grammatical tag and lemma. As the DMII is limited to inflected words, the training data is supplemented with a hand-curated list of approximately 4,500 uninflected words (such as adverbs, conjunctions and prepositions) and abbreviations. To account for subtle differences between the tagsets used in the DMII and by the Icelandic PoS taggers, Nefnir translates all tags to an intermediate tagset which is a subset of both. Rules are successively generated and applied to the training set, with each new rule minimizing the number of remaining errors. Rules continue to be generated until the number of errors cannot be reduced. The process is as follows: Rules are only generated if they can correctly lemmatize at least two examples in the training set. A dictionary is created for words which are incorrectly lemmatized by the rules, for example because they require a unique transformation, such as from við “we” to ég “I”. Once trained, Nefnir lemmatizes words using the dictionary if they are present, or else with the most specific applicable rule. A rule is generated for every suffix in a word form, with some restrictions. For base words, Nefnir considers all suffixes, from the empty string to the full word. For skó “shoes”, an inflected form of the word skór “shoe”, rules are generated for the suffixes INLINEFORM0 , ó, kó and skó. However, Nefnir does not create rules for suffixes that are shorter than the transformation required to lemmatize the word. For example, for bækur “books”, which requires the transformation ækur INLINEFORM1 ók (the lemma for bækur is bók), only the suffixes ækur and bækur are considered. Compounding is highly productive in Icelandic and compound words comprise a very large portion of the vocabulary. This is reflected in the DMII, where over 88% of all words are compounds BIBREF15 . Any of the open word classes can be combined to form a compound, and there is no theoretical limit to how many words they can consist of. Due to the abundance of compounds in the training data, and the freedom with which they can be formed, Nefnir places additional restrictions on which suffixes to consider when generating rules for them. Suffixes for the final part of a compound are generated in the same manner as for base words, growing part by part thereafter. For example, the compound word fjall+göngu+skó “hiking boots” would yield rules for the suffixes INLINEFORM0 , ó, kó, skó, gönguskó and fjallgönguskó. Allowing suffixes to grow freely past the final part of the compound may result in overfitting as the rules adapt to incidental patterns in the training data. ### Evaluation
We have evaluated the output of Nefnir against a reference corpus of 21,093 tokens and their correct lemmas. Samples for the reference corpus were extracted from two larger corpora, in order to obtain a diverse vocabulary: Samples were extracted at random from these two corpora, roughly 10,000 tokens from each, and the lemmas manually reviewed, following the criteria laid out in the preface of the IFD BIBREF9 . The incentive when performing the evaluation was to create a diverse corpus of text samples containing foreign words, misspellings and other OOV words. Such words are likely to appear in real-world NLP tasks, and pose special problems for lemmatizers. In the proofread and literature-heavy IFD corpus, which was used for training and evaluating the previous two lemmatizers, these OOV words are less prevalent. Consequently, the test corpus used here is not directly comparable with the corpus used to evaluate Lemmald and the CST Lemmatizer for Icelandic. On the other hand, it is more diverse and offers more challenging problems for the lemmatizer. One of the motivations of this work was to determine how well Nefnir performs when lemmatizing text which has been PoS tagged automatically, without any manual review, as such manual labour is usually not feasible in large-scale NLP tasks. For this purpose, we created two versions of the test corpus, one with the correct PoS tags, and another tagged using IceTagger BIBREF11 . The accuracy of IceTagger is further enhanced using data from the DMII. Measured against the correct PoS tags, the accuracy of the PoS tags in the reference corpus is 95.47%. Accuracy of the lemmatizaton was measured by comparing the reference corpus lemmas with the obtained lemmas from Nefnir. This was done for both the correctly tagged corpus (gold tags) and the automatically tagged one (IceTagger tags). As seen in Table TABREF10 , the accuracy for the test file with the correct PoS tags is 99.55%, with 94 errors in 21,093 tokens. For the text tagged automatically with IceTagger, the accuracy is 96.88%, with 658 errors. These results indicate that given correct PoS tags, Nefnir obtains high accuracy, with under a hundred errors in the whole corpus sample. This is comparable to the score reported for Lemmald, when DMII lookup has been added (99.55%). In fact, it can be argued that a higher score is hard to come by, as natural language always contains some unforeseen issues that are hard to accommodate for, such as OOV words, misspellings, colloquialisms, etc. When Nefnir bases its lemmas on the automatically PoS tagged text, the accuracy decreases, from 99.55% to 96.88%, resulting in six times as many errors. We can classify the errors made by Nefnir into the following main categories: The most prevalent error categories when the PoS tags are correct are foreign words and proper names, such as foreign names of people, products and companies. A special issue that often came up is the cliticized definite article in Icelandic proper names. This is quite common in organization names (Síminn, Samfylkingin), titles of works of art (Svanurinn), names of ships (Vonin), buildings (Kringlan), etc. Ultimately, it depends on the aim of the lemmatization how these should be handled, but in this evaluation we assume as a general rule that they should be lemmatized with the definite article (Síminn, and not sími or Sími). The same applies to the plural, in names such as Hjálmar “helmets” (band) and Katlar (place name). In the automatically tagged data, tagging errors are the most common source of lemmatization errors, such as when læknum (referring to the plural dative of the masculine noun læknir “doctor”) is tagged as being in the singular, which leads to it being incorrectly lemmatized as lækur “brook”. This was to be expected, as the rules learned from the DMII rely on the correct tagging of the input. However, as the authors of Lemmald comment, as long as the word class is correct, the lemmatizer can usually still find the correct lemma BIBREF13 . The main reason for the high accuracy in our view lies in the richness of the DMII data. No lexicon can ever include all words of a particular language, as new words appear every day, but most often, new words in Icelandic are compounds, created from words already present in the DMII. This explains how rare or unknown words such as the adjective fuglglaður “bird-happy”, which appears in the corpus data, can be correctly lemmatized using the suffix rule for glaður “happy”. As mentioned above, Nefnir, the CST Lemmatizer for Icelandic, and Lemmald have not been evaluated using the same reference corpus. The accuracy of the three lemmatizers are, therefore, not directly comparable, but our results indicate that Nefnir obtains the highest accuracy. ### Conclusion
We described and evaluated Nefnir, a new open source lemmatizer for Icelandic. It uses suffix substitution rules, derived from a large morphological database, to lemmatize tagged text. Evaluation shows that Nefnir obtains high accuracy for both correctly and automatically PoS-tagged input. As taggers for Icelandic gradually get better, we can expect to see the lemmatization accuracy go up as well. Expanding the morphological database with more proper names may also help to achieve even higher accuracy. Table 1: Results of the evaluation, with the accuracy and the total number of errors found.
|
from the Database of Modern Icelandic Inflection (DMII) BIBREF1
|
What is the size of the labelled dataset?
|
### Background
In the light of increased vaccine hesitance in various countries, consistent monitoring of public beliefs and opinions about the national immunization program is important. Besides performing qualitative research and surveys, real-time monitoring of social media data about vaccination is a valuable tool to this end. The advantage is that one is able to detect and respond to possible vaccine concerns in a timely manner, that it generates continuous data and that it consists of unsolicited, voluntary user-generated content. Several studies that analyse tweets have already been conducted, providing insight in the content that was tweeted most during the 2009 H1N1 outbreak BIBREF0, the information flow between users with a certain sentiment during this outbreak BIBREF1, or trends in tweets that convey, for example, the worries on efficacy of HPV vaccines BIBREF2, BIBREF3. While human coders are best at deploying world knowledge and interpreting the intention behind a text, manual coding of tweets is laborious. The above-mentioned studies therefore aimed at developing and evaluating a system to code tweets automatically. There are several systems in place that make use of this automatic coding. The Vaccine Confidence Project BIBREF4 is a real-time worldwide internet monitor for vaccine concerns. The Europe Media Monitor (EMM) BIBREF5 was installed to support EU institutions and Member State organizations with, for example, the analysis real-time news for medical and health-related topics and with early warning alerts per category and country. MEDISYS, derived from the EMM and developed by the Joint Research Center of the European Commission BIBREF6, is a media monitoring system providing event-based surveillance to rapidly identify potential public health threats based on information from media reports. These systems cannot be used directly for the Netherlands because they do not contain search words in Dutch, are missing an opinion-detection functionality, or do not include categories of the proper specificity. Furthermore, opinions towards vaccination are contextualized by national debates rather than a multinational debate BIBREF7, which implies that a system for monitoring vaccination stance on Twitter should ideally be trained and applied to tweets with a similar language and nationality. Finally, by creating an automatic system for mining public opinions on vaccination concerns, one can continue training and adapting the system. We therefore believe it will be valuable to build our own system. Besides analysing the content of tweets, several other applications that use social media with regard to vaccination have been proposed. They, for example, use data about internet search activity and numbers of tweets as a proxy for (changes in) vaccination coverage or for estimating epidemiological patterns. Huang et al. BIBREF8 found a high positive correlation between reported influenza attitude and behavior on Twitter and influenza vaccination coverage in the US. In contrast, Aquino et al. BIBREF9 found an inverse correlation between Mumps, Measles, Rubella (MMR) vaccination coverage and tweets, Facebook posts and internet search activity about autism and MMR vaccine in Italy. This outcome was possibly due to a decision of the Court of Justice in one of the regions to award vaccine-injury compensation for a case of autism. Wagner, Lampos, Cox and Pebody BIBREF10 assessed the usefulness of geolocated Twitter posts and Google search as source data to model influenza rates, by measuring their fit to the traditional surveillance outcomes and analyzing the data quality. They find that Google search could be a useful alternative to the regular means of surveillance, while Twitter posts are not correlating well due to a lower volume and bias in demographics. Lampos, de Bie and Christianinni BIBREF11 also make use of geolocated Twitter posts to track academics, and present a monitoring tool with a daily flu-score based on weighted keywords. Various studies BIBREF12, BIBREF13, BIBREF14 show that estimates of influenza-like illness symptoms mentioned on Twitter can be exploited to track reported disease levels relatively accurately. However, other studies BIBREF15, BIBREF16 showed that this was only the case when looking at severe cases (e.g. hospitalizations, deaths) or only for the start of the epidemic when interest from journalists was still high. Other research focuses on detecting discussion communities on vaccination in Twitter BIBREF17 or analysing semantic networks BIBREF18 to identify the most relevant and influential users as well as to better understand complex drivers of vaccine hesitancy for public health communication. Tangherlini et al. BIBREF19 explore what can be learned about the vaccination discussion from the realm of “mommy blogs”: parents posting messages about children’s health care on forum websites. They aim to obtain insights in the underlying narrative frameworks, and analyse the topics of the messages using Latent Dirichlet Allocation (LDA) BIBREF20. They find that the most prominent frame is a focus on the exemption of one’s child from receiving a vaccination in school. The motivation against vaccination is most prominently based on personal belief about health, but could also be grounded in religion. Surian et al. BIBREF21 also apply topic modeling to distinguish dominant opinions in the discussion about vaccination, and focus on HPV vaccination as discussed on Twitter. They find a common distinction between tweets reporting on personal experience and tweets that they characterize as `evidence’ (statements of having had a vaccination) and `advocacy’ (statements that support vaccination). Most similar to our work is the study by Du, Xu, Song, Liu and Tao BIBREF2. With the ultimate aim to improve the vaccine uptake, they applied supervised machine learning to analyse the stance towards vaccination as conveyed on social media. Messages were labeled as either related to vaccination or unrelated, and, when related, as ‘positive’, ‘negative’ or ‘neutral’. The ‘negative’ category was further broken down into several considerations, such as ‘safety’ and ‘cost’. After having annotated 6,000 tweets, they trained a classifier on different combinations of features, obtaining the highest macro F1-score (the average of the separate F1-scores for each prediction category) of $0.50$ and micro F1-score (the F1-score over all predictions) of $0.73$. Tweets with a negative stance that point to safety risks could best be predicted, at an optimal F1 score of $0.75$, while the other five sub-categories with a negative stance were predicted at an F1 score below $0.5$ or even $0.0$. Like Du et al. BIBREF2, we focus on analysing sentiment about vaccination using Twitter as a data source and applying supervised machine learning approaches to extract public opinion from tweets automatically. In contrast, in our evaluation we focus on detecting messages with a negative stance in particular. Accurately monitoring such messages helps to recognize discord in an early stage and take appropriate action. We do train machine learning classifiers on modeling other categories than the negative stance, evaluating whether this is beneficial to detecting tweets with a negative stance. For example, we study whether it is beneficial to this task to model tweets with a positive and neutral stance as well. We also inquire whether a more fine-grained categorization of sentiment (e.g.: worry, relief, frustration and informing) offers an advantage. Apart from comparing performance in the context of different categorizations, we compare different machine learning algorithms and compare data with different levels of annotation reliability. Finally, the performance of the resulting systems is compared to regular sentiment analysis common to social media monitoring dashboards. At the public health institute in the Netherlands, we make use of social media monitoring tools offered by Coosto. For defining whether a message is positive, negative or neutral with regard to vaccination, this system makes use of the presence or absence of positive or negative words in the messages. We believe that we could increase the sensitivity and specificity of the sentiment analysis by using supervised machine learning approaches trained on a manually coded dataset. The performance of our machine learning approaches is therefore compared to the sentiment analysis that is currently applied in the Coosto tool. ### Implementation
We set out to curate a corpus of tweets annotated for their stance towards vaccination, and to employ this corpus to train a machine learning classifier to distinguish tweets with a negative stance towards vaccination from other tweets. In the following, we will describe the stages of data acquisition, from collection to labeling. ### Implementation ::: Data collection
We queried Twitter messages that refer to a vaccination-related key term from TwiNL , a database with IDs of Dutch Twitter messages from January 2012 onwards BIBREF22. In contrast to the open Twitter Search API, which only allows one to query tweets posted within the last seven days, TwiNL makes it possible to collect a much larger sample of Twitter posts, ranging several years. We queried TwiNL for different key terms that relate to the topic of vaccination in a five-year period, ranging from January 1, 2012 until February 8, 2017. Query terms that we used were the word `vaccinatie’ (Dutch for `vaccination’) and six other terms closely related to vaccination, with and without a hashtag (`#’). Among the six words is `rijksvaccinatieprogramma’, which refers to the vaccination programme in The Netherlands. An overview of all query terms along with the number of tweets that could be collected based on them is displayed in Table TABREF5. We collected a total of 96,566 tweets from TwiNL, which we filtered in a number of ways. First, retweets were removed, as we wanted to focus on unique messages. This led to a removal of 31% of the messages. Second, we filtered out messages that contain a URL. Such messages often share a news headline and include a URL to refer to the complete news message. As a news headline does not reflect the stance of the person who posted the tweet, we decided to apply this filtering step. It is likely that part of the messages with a URL do include a message composed by the sender itself, but this step helps to clean many unwanted messages. Third, we removed messages that include a word related to animals and traveling (`dier’, animal; `landbouw’, agriculture; and `teek’, tick), as we strictly focus on messages that refer to vaccination that is part of the governmental vaccination program. 27,534 messages were left after filtering. This is the data set that is used for experimentation. ### Implementation ::: Data annotation
The stance towards vaccination was categorized into `Negative’, `Neutral’, `Positive’ and `Not clear’. The latter category was essential, as some posts do not convey enough information about the stance of the writer. In addition to the four-valued stance classes we included separate classes grouped under relevance, subject and sentiment as annotation categories. With these additional categorizations we aimed to obtain a precise grasp of all possibly relevant tweet characteristics in relation to vaccination, which could help in a machine learning setting. The relevance categories were divided into `Relevant’, `Relevant abroad’ and `Irrelevant’. Despite our selection of vaccination-related keywords, tweets that mention these words might not refer to vaccination at all. A word like `vaccine’ might be used in a metaphorical sense, or the tweet might refer to vaccination of animals. The subject categorization was included to describe what the tweet is about primarily: `Vaccine’, `Disease’ or `Both’. We expected that a significant part of the tweets would focus on the severeness of a disease when discussing vaccination. Distinguishing these tweets could help the detection of the stance as well. Finally, the sentiment of tweets was categorized into `Informative’, `Angry/Frustration’, `Worried/Fear/Doubts’, `Relieved’ and `Other’, where the latter category lumps together occasional cases of humor, sarcasm, personal experience, and question raised. These categories were based on the article by BIBREF0, and emerged from analysing their H1N1-related tweets. The `Informative’ category refers to a typical type of message in which information is shared, potentially in support of a negative or positive stance towards vaccination. If the message contained more than one sentiment, the first sentiment identified was chosen. Table TABREF6 shows examples of tweets for the above-mentioned categories. We aimed at a sufficient number of annotated tweets to feed a machine learning classifier with. The majority of tweets were annotated twice. We built an annotation interface catered to the task. Upon being presented with the text of a Twitter post, the annotator was first asked whether the tweet was relevant. In case it was deemed relevant, the tweet could be annotated for the other categorizations. Otherwise, the user could click `OK’ after which he or she was directly presented with a new Twitter post. The annotator was presented with sampled messages that were either not annotated yet or annotated once. We ensured a fairly equal distribution of these two types, so that most tweets would be annotated twice. As annotators, we hired four student assistants and additionally made use of the Radboud Research Participation System. We asked participants to annotate for the duration of an hour, in exchange for a voucher valued ten Euros, or one course credit. Before starting the annotation, the participants were asked to read the annotation manual, with examples and an extensive description of the categories, and were presented with a short training round in which feedback on their annotations was given. The annotation period lasted for six weeks. We stopped when the number of applicants dropped. A total of 8,259 tweets were annotated, of which 6,472 were annotated twice (78%). 65 annotators joined in the study, with an average of $229.5$ annotated tweets per person. The number of annotations per person varied considerably, with $2,388$ tweets coded by the most active annotator. This variation is due to the different ways in which annotators were recruited: student-assistants were recruited for several days, while participants recruited through the Radboud Research Participation System could only join for the duration of an hour. We calculated inter-annotator agreement by Krippendorff's Alpha BIBREF23, which accounts for different annotator pairs and empty values. To also zoom in on the particular agreement by category, we calculated mutual F-scores for each of the categories. This metric is typically used to evaluate system performance by category on gold standard data, but could also be applied to annotation pairs by alternating the roles of the two annotators between classifier and ground truth. A summary of the agreement by categorization is given in Table TABREF10. While both the Relevance and Subject categorizations are annotated at a percent agreement of $0.71$ and $0.70$, their agreement scores are only fair, at $\alpha =0.27$ and $\alpha =0.29$. The percent agreement on Stance and Sentiment, which carry more categories than the former two, is $0.54$ for both. Their agreement scores are also fair, at $\alpha =0.35$ and $\alpha =0.34$. The mutual F-scores show marked differences in agreement by category, where the categories that were annotated most often typically yield a higher score. This holds for the Relevant category ($0.81$), the Vaccine category ($0.79$) and the Positive category ($0.64$). The Negative category yields a mutual F-score of $0.42$, which is higher than the more frequently annotated categories Neutral ($0.23$) and Not clear ($0.31$). We found that these categories are often confused. After combining the annotations of the two, the stance agreement would be increased to $\alpha =0.43$. The rather low agreement over the annotation categories indicates the difficulty of interpreting stance and sentiment in tweets that discuss the topic of vaccination. We therefore proceed with caution to categorize the data for training and testing our models. The agreed upon tweets will form the basis of our experimental data, as was proposed by Jakubiçek, Kovar and Rychly BIBREF24, while the other data is added as additional training material to see if the added quantity is beneficial to performance. We will also annotate a sample of the agreed upon tweets, to make sure that these data are reliable in spite of the low agreement rate. ### Implementation ::: Data categorization
The labeled data that we composed based on the annotated tweets are displayed in Table TABREF11. We combined the Relevant and Relevant abroad categories into one category (`Relevant’), as only a small part of the tweets was annotated as Relevant abroad. We did not make use of the subject annotations, as a small minority of the tweets that were relevant referred a disease only. For the most important categorization, stance, we included all annotated labels. Finally, we combined part of the more frequent sentiment categories with Positive. We distinguish three types of labeled tweets: `strict’, `lax’ and `one’. The strictly labeled tweets were labeled by both annotators with the same label. The lax labels describe tweets that were only annotated with a certain category by one of the coders. The categories were ordered by importance to decide on the lax labels. For instance, in case of the third categorization, Negative was preferred over Positive, followed by Neutral, Not clear and Irrelevant. If one of the annotators labeled a tweet as Positive and the other as Neutral, the lax label for this tweet is Positive. In table TABREF11, the categories are ordered by preference as imposed on the lax labeling. The `one' labeling applies to all tweets that were annotated by only one annotator. Note that the total counts can differ between label categorizations due to the lax labeling: the counts for Positive labels in the Polarity + sentiment labeling (Positive + Frustration, Positive + Information and Positive + other) do not add up to the count of the Positive label in the Polarity labeling. With the `strict’, `lax’ and `one’ labeling, we end up with four variants of data to experiment with: only strict, strict + lax, strict + one and strict + lax + one. The strict data, which are most reliable, are used in all variants. By comparing different combinations of training data, we test whether the addition of less reliably labeled data (lax and/or one) boosts performance. The four labelings have an increasing granularity, where the numbers of examples for the Negative category are stable across each labeling. In the first labeling, these examples are contrasted with any other tweet. It hence comprises a binary classification task. In the second labeling, irrelevant tweets are indicated in a separate category. The Other class here represents all relevant tweets that do not convey a negative stance towards vaccination. In the third labeling, this class is specified as the stance categories Positive, Neutral and Not clear. In the fourth labeling, the Positive category, which is the most frequent polarity class, is further split into `Positive + frustration’, `Positive + Information’ and `Positive + Other’. Positivity about vaccination combined with a frustration sentiment reflects tweets that convey frustration about the arguments of people who are negative about vaccination (e.g.: "I just read that a 17 year old girl died of the measles. Because she did not want an inoculation due to strict religious beliefs. -.- #ridiculous"). The Positive + Information category reflects tweets that provide information in favor of vaccination, or combined with a positive stance towards vaccination (e.g.: "#shingles is especially common with the elderly and chronically diseased. #vaccination can prevent much suffering. #prevention"). In line with Kovár, Rychlý and Jakubíček BIBREF25, we evaluate system performance only on the reliable part of the annotations - the instances labeled with the same label by two annotators. As the overall agreement is not sufficient, with Krippendorff's Alpha ranging between $0.27$ and $0.35$, the first author annotated 300 tweets sampled from the strict data (without knowledge of the annotations) to rule out the possibility that these agreed upon annotations are due to chance agreement. Comparing these new annotations to the original ones, the Negative category and the Positive category are agreed upon at mutual F-scores of $0.70$ and $0.81$. The percent agreement on the binary classification scheme (e.g.: Negative versus Other) is $0.92$, with $\alpha =0.67$, which decreases to $\alpha =0.55$ for the Relevance categorization, $\alpha =0.54$ for the Polarity categorization and $\alpha =0.43$ for the Polarity + Sentiment categorization. We find that instances of a negative and positive stance can be clearly identified by humans, while the labels Neutral and Not Clear are less clear cut. Since it is our focus to model tweets with a negative stance, the agreement on the binary decision between Negative and Other is just sufficient to use for experimentation based on Krippendorff’s BIBREF26 remark that “$\alpha \ge .667$ is the lowest conceivable limit” (p.241). In our experimental set-up we will therefore only evaluate our system performance on distinguishing the Negative category from any other category in the strict data. ### Implementation ::: Experimental Set-up
For each combination of labeling (four types of labeling) and training data (four combinations of training data) we train a machine learning classifier to best distinguish the given labels. Two different classifiers are compared: Multinomial Naive Bayes and Support Vector Machines (SVM). In total, this makes for 32 variants (4 labelings $\times $ 4 combinations of training data $\times $ 2 classifiers). All settings are tested through ten-fold cross-validation on the strict data and are compared against two rule-based sentiment analysis baselines and two random baselines. All components of the experimental set-up are described in more detail below. ### Implementation ::: Experimental Set-up ::: Preprocessing
To properly distinguish word tokens and punctuation we tokenized the tweets by means of Ucto, a rule-based tokenizer with good performance on the Dutch language, and with a configuration specific for Twitter. Tokens were lowercased in order to focus on the content. Punctuation was maintained, as well as emoji and emoticons. Such markers could be predictive in the context of a discussion such as vaccination. To account for sequences of words and characters that might carry useful information, we extracted word unigrams, bigrams, and trigrams as features. Features were coded binary, i.e. set to 1 if a feature is seen in a message and set to 0 otherwise. During training, all features apart from the top 15,000 most frequent ones were removed. ### Implementation ::: Experimental Set-up ::: Machine Learning
We applied two machine learning algorithms with a different perspective on the data: Multinomial Naive Bayes and SVM. The former algorithm is often used on textual data. It models the Bayesian probability of features to belong to a class and makes predictions based on a linear calculation. Features are naively seen as independent of one another BIBREF27. In their simplest form, SVMs are binary linear classifiers that make use of kernels. They search for the optimal hyperplane in the feature space that maximizes the geometric margin between any two classes. The advantage of SVMs is that they provide a solution to a global optimization problem, thereby reducing the generalization error of the classifier BIBREF28. We applied both algorithms by means of the scikit-learn toolkit, a python library that offers implementations of many machine learning algorithms BIBREF29. To cope with imbalance in the number of instances per label, for Multinomial Naive Bayes we set the Alpha parameter to $0.0$ and muted the fit prior. For SVM, we used a linear kernel with the $C$ parameter set to $1.0$ and a balanced class weight. ### Implementation ::: Experimental Set-up ::: Baselines
As baselines, we applied two rule-based sentiment analysis systems for Dutch as well as two random baselines. The first rule-based sentiment analysis system is Pattern, an off-the-shelf sentiment analysis system that makes use of a list of adjectives with a positive or negative weight, based on human annotations BIBREF30. Sentences are assigned a score between $-1.0$ and $1.0$ by multiplying the scores of their adjectives. Bigrams like `horribly good’ are seen as one adjective, where the adjective `horribly’ increases the positivity score of `good’. We translated the polarity score into the discrete labels `Negative’, `Positive’ and `Neutral’ by using the training data to infer which threshold leads to the best performance on the `Negative’ category. The second baseline is the sentiment analysis offered by the social media monitoring dashboard Coosto. As Coosto is a commercial product, there is no public documentation on their sentiment analysis tool. In addition to these two baselines, we applied two random baselines: predicting the negative class randomly for 50% of the messages and predicting the negative class randomly for 15% of the messages. The latter proportion relates to the proportion of vaccination-hesitant tweets in the strictly labeled data on which we test the systems. ### Implementation ::: Evaluation
We evaluate performance by means of ten-fold cross-validation on the strictly labeled data. In each of the folds, 90% of the strictly labeled data is used as training data, which are complemented with the laxly labeled data and/or the data labeled by one annotator, in three of the four training data variants. Performance is always tested on the strict data. As evaluation metrics we calculate the F1-score and the Area Under the ROC Curve (AUC) on predicting the negative stance towards vaccination in the test tweets. ### Results
We trained machine learning (ML) classifiers to distinguish Twitter messages with a negative stance towards vaccination, alternating three aspects of the system: the labels to train on, the composition of the training data and the ML algorithm. The results are presented in Table TABREF15, as the F1-score and AUC of any setting on correctly predicting tweets with a negative stance. Systems with specific combinations of the ML classifier and size of the training data are given in the rows of the table. The four types of labelings are listed in the columns. The results show a tendency for each of the three manipulations. Regarding the ML algorithm, SVM consistently outperforms Naive Bayes for this task. Furthermore, adding additional training data, albeit less reliable, generally improves performance. Training a model on all available data (strict + lax + one) leads to an improvement over using only the strict data, while adding only the laxly labeled data is generally better than using all data. Adding only the data labeled by one annotator often leads to a worse performance. With respect to the labeling, the Polarity-sentiment labeling generally leads to the best outcomes, although the overall best outcome is yielded by training an SVM on Polarity labeling with strict data appended by lax data, at an area under the curve score of $0.66$. The best reported performance is an F1-score of $0.36$ and an AUC of $0.66$. In comparison to the baselines (Table TABREF16), these scores are considerably higher. Nevertheless, there is room for improvement. The performance of the random baselines, with F1-scores of $0.18$ (50%) and $0.13$ (15%), indicates that the minimal performance on this task is rather low. The rule-based sentiment analyses yield better performances, at an F1-score of $0.20$ for Pattern and $0.25$ for Coosto. To analyse the behavior of the best ML system, we present a confusion table of its classifications in Table TABREF17. The Irrelevant category is most often classified with one of the other categories, while the Positive and Negative categories are the biggest confusables. The classifier is possibly identifying features that denote a stance, but struggles to distinguish positive from negative. To gain insight into the potential of increasing the amount of training data, we applied the best ML system (SVM trained on strict and lax data on the polarity labels) on 10% of the strictly labeled data, starting with a small sample of the data and increasing it to all available data (excluding the test data). The learning curve is presented in Figure FIGREF18. It shows an improved performance until the last training data is added, indicating that more training data would likely yield better performance. ### Results ::: Comparison machine learning and rule-based sentiment analysis
A confusion table of the predictions of the best of the two rule-based baselines, Pattern, and the best ML system is displayed in Table TABREF19. Only 192 tweets are labeled by both systems as Negative, while the best ML system accounts for almost double this amount and Pattern for three times as much. Comparing the predictions to the gold standard labeling, 99 of the tweets predicted only by the best ML system as Negative are correct (27%), opposed to 51 that are exclusive to Pattern (8%). Of the tweets that were classified by both as negative, 63 are correct (33%). This shows that the approaches have a rather complementary view on tweets with a negative stance. To gain more insight into the behavior of both approaches, we applied them to 15,577 unlabeled tweets. Table TABREF20 presents a confusion table with the numbers of tweets that were classified as Negative or another category by both approaches. Again, pattern accounts for the majority of negatively labeled messages, and the overlap is small. Two of the authors validated for a sample of 600 messages whether they actually manifested a negative attitude towards vaccination: 200 messages that were uniquely classified by the best ML system as Negative, 200 messages that were solely labeled as Negative by Pattern and 200 messages that were classified by both systems as Negative. This validation showed the same tendency as for the labeled data, with a higher precision of the best ML system in comparison to Pattern (33.5% versus 21% of the messages correctly predicted) and the highest precision when both systems predicted the negative class (36%). The complementary view on tweets with a negative stance between the best ML system and rule-based sentiment analysis becomes clear from their differing predictions. To make this difference concrete, we present a selection of the messages predicted as Negative by both systems in Table TABREF21. The first three are only predicted by the best ML system as Negative, and not by Pattern, while the fourth until the sixth examples are only seen as Negative by Pattern. Where the former give arguments (`can not be compared...’, `kids are dying from it’) or take stance (`I’m opposed to...’), the latter examples display more intensified words and exclamations (`that’s the message!!’, `Arrogant’, `horrific’) and aggression towards a person or organization. The last three tweets are seen by both systems as Negative. They are characterized by intensified words that linked strongly to a negative stance towards vaccination (`dangerous’, `suffering’, `get lost with your compulsory vaccination’). Table TABREF21 also features tweets that were predicted as Negative by neither the best ML-system nor Pattern, representing the most difficult instances of the task. The first two tweets include markers that explicitly point to a negative stance, such as `not been proven' and `vaccinating is nonsense'. The third tweet manifests a negative stance by means of the sarcastic phrase `way to go' (English translation). The use of sarcasm, where typically positive words are used to convey a negative valence, complicates this task of stance prediction. The last tweet advocates an alternative to vaccination, which implicitly can be explained as a negative stance towards vaccination. Such implicitly packaged viewpoints also hamper the prediction of negative stance. Both sarcasm and implicit stance could be addressed by specific modules. ### Results ::: Improving recall
For monitoring the number of Twitter messages over time that are negative towards vaccination, it is arguably more important to detect them at a high recall than at a high precision. False positives (messages incorrectly flagged as Negative) could be filtered manually by a human end user, while False Negatives (messages with a negative stance that are not detected) will be missed. We set out to improve recall, making use of classifier confidence scores and the complementary classifications of Pattern and the best ML system. A first recall-improving approach is to reset the prediction threshold for the Negative category. For any given instance, the SVM classifier estimates the probability of all categories it was trained on. It will predict the Negative category for an instance if its probability exceeds the probabilities of the other categories. This prediction can be altered by changing the threshold; setting the threshold higher will generally mean that fewer instances will be predicted as a Negative category (corresponding to a higher precision), whereas setting it lower will mean more instances will be predicted as such (corresponding to a higher recall). Thus, the balance between precision and recall can be set as desired, to favor one or another. However, in many cases, changing the threshold will not lead to a (strong) increase in overall performance. Figure FIGREF22 presents the balance between recall and precision as a result of predicting the Negative category with the best ML system, when the threshold for this category is altered from lowest to highest. Compared to the standard recall of $0.43$ at a precision of $0.29$, increasing the recall to $0.60$ would lead to a drop of precision to $0.21$. The F1-score would then decrease to $0.31$. A second means by which recall might be improved is to employ ensemble classification. The comparison in the previous section between the best ML method and rule-based sentiment analysis revealed that both systems have a rather disjoint perspective on negative stance: many more tweets are labeled as `Negative' by only one of the two systems than by both. We therefore built an ensemble system that follows both systems in their perspective on tweets with a negative stance: for each tweet, if either of the systems predicts the Negative category, the ensemble system makes this prediction. The performance of the ensemble system is presented in Table TABREF23. Of the 343 tweets in the test set that are labeled as Negative, 210 are retrieved by the ensemble system. The result is a recall of $0.61$. The system does overshoot in its categorization of tweets as Negative: this category is predicted for 1,168 tweets (about 40% of total test set of 2,886 tweets). The result is a precision of $0.18$. In comparison to lowering the prediction threshold of the ML system, the ensemble system thus yields a slightly worse trade-off between precision and recall. ### Discussion
With an F1-score of $0.36$, our system lags behind the $0.75$ F1-score reported by Du et al.BIBREF2. Several factors might have influenced this difference. A first factor is the low proportion of tweets with the label `Negative' in our dataset. In the strict labeling condition, only 343 cases are labeled as negative by two annotators, against 2,543 labeled as positive – the negative cases only comprise 13% of all instances. In the study of Du et al., the anti-vaccination category comprises 24% of all instances (1,445 tweets). More (reliable) examples might have helped in our study to train a better model of negative tweets. Secondly, Du et al. BIBREF2 focused on the English language domain, while we worked with Dutch Twitter messages. The Dutch Twitter realm harbors less data to study than the English one, and might bring forward different discussions when it comes to the topic of vaccination. It could be that the senders' stance towards vaccination is more difficult to pinpoint within these discussions. In line with this language difference, a third prominent factor that might have led to a higher performance in the study of Du et al.BIBREF2 is that they focus on a particular case of vaccination (e.g.: HPV vaccination) and split the anti-vaccination category into several more specific categories that describe the motivation of this stance. The diverse motivations for being against vaccination are indeed reflected in several other studies that focus on identifying discussion communities and viewpoints BIBREF17, BIBREF21, BIBREF19. While splitting the data into more specific categories will lead to less examples per category, it could boost performance on predicting certain categories due to a larger homogeneity. Indeed, the most dominant negative category in the study by Du et al.BIBREF2, dubbed `NegSafety' and occurring in 912 tweets (63% of all negative tweets), yielded the highest F1-score of $0.75$. While two less frequent categories were predicted at an F1-score of $0.0$, this outcome shows the benefit of breaking down the motivations behind a negative stance towards vaccination. A major limitation of our study is that the agreement rates for all categorizations are low. This is also the case in other studies, like BIBREF8, who report an agreement of $K = 0.40$ on polarity categorization. Foremost, this reflects the difficulty of the task. The way in which the stance towards vaccination is manifested in a tweet depends on the author, his or her specific viewpoint, the moment in time at which a tweet was posted, and the possible conversation thread that precedes it. Making a judgment solely based on the text could be difficult without this context. Agreement could possibly be improved by presenting the annotator with the preceding conversation as context to the text. Furthermore, tweets could be coded by more than two annotators. This would give insight into the subtleties of the data, with a graded scale of tweets that clearly manifest a negative stance towards vaccination to tweets that merely hint at such a stance. Such a procedure could likewise help to generate more reliable examples to train a machine learning classifier. The low agreement rates also indicate that measuring stance towards vaccination in tweets is a too difficult task to assign only to a machine. We believe that the human-in-the-loop could be an important asset in any monitoring dashboard that focuses on stance in particular discussions. The system will have an important role in filtering the bigger stream of messages, leaving the human ideally with a controllable set of messages to sift through to end up with reliable statistics on the stance that is seen in the discussion at any point in time. In the analysis section, we explored two approaches to increase recall of messages with a negative stance, which would be most useful in this scenario. Lowering the prediction threshold showed to be most effective to this end. Our primary aim in future work is to improve performance. We did not experiment with different types of features in our current study. Word embeddings might help to include more semantics in our classifier’s model. In addition, domain knowledge could be added by including word lists, and different components might be combined to address different features of the data (e.g.: sarcasm and implicit stance). We also aim to divide the negative category into the specific motivations behind a negative stance towards vaccination, like in the study of Du et al.BIBREF2, so as to obtain more homogeneous categories. Parallel to this new categorization of the data, adding more labeled data appears to be the most effective way to improve our model. The learning curve that we present in Figure FIGREF18 shows that there is no performance plateau reached with the current size of the data. An active learning setting BIBREF31, starting with the current system, could be applied to select additional tweets to annotate. Such a setting could be incorporated in the practical scenario where a human-in-the-loop judges the messages that were flagged as displaying a negative stance by the system. The messages that are judged as correctly and incorrectly predicted could be added as additional reliable training data to improve upon the model. We have installed a dashboard that is catered for such a procedure, starting with the machine learning system that yielded the best performance in our current study. ### Conclusions
We set out to train a classifier to distinguish Twitter messages that display a negative stance towards vaccination from other messages that discuss the topic of vaccination. Based on a set of 8,259 tweets that mention a vaccination-related keyword, annotated for their relevance, stance and sentiment, we tested a multitude of machine learning classifiers, alternating the algorithm, the reliability of training data and the labels to train on. The best performance, with a precision of $0.29$, a recall of $0.43$, an F1-score of $0.36$ and an AUC of $0.66$, was yielded by training an SVM classifier on strictly and laxly labeled data to distinguish irrelevant tweets and polarity categories. The baselines, with an optimal F1-score of $0.25$ (rule-based sentiment analysis), were considerably outperformed. The latter shows the benefit of machine-learned classifiers on domain-specific sentiment: despite being trained on a reasonably small amount of data, the machine-learning approach outperforms general-purpose sentiment analysis tools. ### Availability and requirements
Project name: Prikbord Project home page: http://prikbord.science.ru.nl/ Operating system: Linux Programming language: Python, javascript Other requirements: Django 1.5.11 or higher, MongoDB 2.6.10, pymongo 2.7.2 or higher, requests 2.13.0 or higher License: GNU GPL Any restrictions to use by non-academics: licence needed ### Abbreviations
EMM: Europe Media Monitor MMR: Mumps, Measles, Rubella LDA: Latent Dirichlet Allocation ML: Machine learning SVM: Support Vector Machines AUC: Area under the ROC Curve Clf: Classifier NB: Naive Bayes Pr: Precision Re: Recall ### Declarations ::: Ethics approval and consent to participate
Not applicable. ### Declarations ::: Consent for publication
Not applicable. ### Declarations ::: Availability of data and materials
http://cls.ru.nl/fkunneman/data_stance_vaccination.zip ### Declarations ::: Competing interests
The authors declare that they have no competing interests. ### Declarations ::: Funding
This study has been funded by the Rijksinstituut voor Volksgezondheid en Milieu. ### Declarations ::: Author's contributions
FK has set up the annotations procedure, performed the Machine Learning experiments and analysis, annotated tweets in the analysis and did a major part of the writing. ML has done part of the writing in the Introduction and Conclusion sections. AW has advised on the experimentation and analysis. AB has advised on the experimentation and has edited the complete text. LM has set up the annotations procedure, annotated tweets in the analysis and has done a major part of the writing. All authors read and approved the final manuscript. ### Declarations ::: Acknowledgements
We thank Erik Tjong Kim Sang for the development and support of the http://twiqs.nl service. We also thank the ones who have contributed with annotations. Table 1 Overview of the number of Twitter messages that were queried from TwiNL and filtered, from the period between January 2012 and February 2017 . ‘URLs’ refers to tweets with a URL. ‘blacklist’ refers to words related to animal vaccination and vaccination related to travelling: ‘dier’ (animal), ‘landbouw’ (agriculture), and ‘teek’ (tick). Table 2 Specification of the annotation categories Table 3 Agreement scores for all four categorizations; mutual F-score is reported by category. Table 4 Overview of data set (the cells indicate the number of examples per label and data type) Table 5 Machine Learning performance of correctly predicting the label of tweets with a negative stance (Clf = Classifier, NB = Naive Bayes, SVM = Support Vector Machines, AUC = Area under the curve). Table 6 Baseline performance of correctly predicting the label of tweets with a negative stance (for comparison, the best ML system is included; Pr = Precision, Re = Recall, AUC = Area under the Curve). Table 7 Confusion table of the classification of tweets in the best ML setting (SVM trained on Polarity labeling with strict data appended by lax data). The vertical axes give gold standard labels, the horizontal axes give the classifier decisions. Numbers given in bold are accurate classifications. Table 9 Confusion table of the unlabeled Twitter messages predicted as ‘Negative’ or another category between Pattern and the best ML system. Table 8 Confusion table of the labeled Twitter messages predicted as ‘Negative’ or another category between Pattern and the best ML system. Table 10 Examples of tweets that were classified by the best ML system and/or pattern as ‘Negative’ (for privacy reasons, user mentions are replaced with ‘@USER’).
|
27,534 messages
|
What type of syntactic analysis is performed?
|
### Introduction
A common way for marking information about gender, number, and case in language is morphology, or the structure of a given word in the language. However, different languages mark such information in different ways – for example, in some languages gender may be marked on the head word of a syntactic dependency relation, while in other languages it is marked on the dependent, on both, or on none of them BIBREF0 . This morphological diversity creates a challenge for machine translation, as there are ambiguous cases where more than one correct translation exists for the same source sentence. For example, while the English sentence “I love language” is ambiguous with respect to the gender of the speaker, Hebrew marks verbs for the gender of their subject and does not allow gender-neutral translation. This allows two possible Hebrew translations – one in a masculine and the other in a feminine form. As a consequence, a sentence-level translator (either human or machine) must commit to the gender of the speaker, adding information that is not present in the source. Without additional context, this choice must be done arbitrarily by relying on language conventions, world knowledge or statistical (stereotypical) knowledge. Indeed, the English sentence “I work as a doctor” is translated into Hebrew by Google Translate using the masculine verb form oved, indicating a male speaker, while “I work as a nurse” is translated with the feminine form ovedet, indicating a female speaker (verified on March 2019). While this is still an issue, there have been recent efforts to reduce it for specific language pairs. We present a simple black-box method to influence the interpretation chosen by an NMT system in these ambiguous cases. More concretely, we construct pre-defined textual hints about the gender and number of the speaker and the audience (the interlocutors), which we concatenate to a given input sentence that we would like to translate accordingly. We then show that a black-box NMT system makes the desired morphological decisions according to the given hint, even when no other evidence is available on the source side. While adding those hints results in additional text on the target side, we show that it is simple to remove, leaving only the desired translation. Our method is appealing as it only requires simple pre-and-post processing of the inputs and outputs, without considering the system internals, or requiring specific annotated data and training procedure as in previous work BIBREF1 . We show that in spite of its simplicity, it is effective in resolving many of the ambiguities and improves the translation quality in up to 2.3 BLEU when given the correct hints, which may be inferred from text metadata or other sources. Finally, we perform a fine-grained syntactic analysis of the translations generated using our method which shows its effectiveness. ### Morphological Ambiguity in Translation
Different languages use different morphological features marking different properties on different elements. For example, English marks for number, case, aspect, tense, person, and degree of comparison. However, English does not mark gender on nouns and verbs. Even when a certain property is marked, languages differ in the form and location of the marking BIBREF0 . For example, marking can occur on the head of a syntactic dependency construction, on its argument, on both (requiring agreement), or on none of them. Translation systems must generate correct target-language morphology as part of the translation process. This requires knowledge of both the source-side and target-side morphology. Current state-of-the-art translation systems do capture many aspects of natural language, including morphology, when a relevant context is available BIBREF2 , BIBREF3 , but resort to “guessing” based on the training-data statistics when it is not. Complications arise when different languages convey different kinds of information in their morphological systems. In such cases, a translation system may be required to remove information available in the source sentence, or to add information not available in it, where the latter can be especially tricky. ### Black-Box Knowledge Injection
Our goal is to supply an NMT system with knowledge regarding the speaker and interlocutor of first-person sentences, in order to produce the desired target-side morphology when the information is not available in the source sentence. The approach we take in the current work is that of black-box injection, in which we attempt to inject knowledge to the input in order to influence the output of a trained NMT system, without having access to its internals or its training procedure as proposed by vanmassenhove-hardmeier-way:2018:EMNLP. We are motivated by recent work by BIBREF4 who showed that NMT systems learn to track coreference chains when presented with sufficient discourse context. We conjecture that there are enough sentence-internal pronominal coreference chains appearing in the training data of large-scale NMT systems, such that state-of-the-art NMT systems can and do track sentence-internal coreference. We devise a wrapper method to make use of this coreference tracking ability by introducing artificial antecedents that unambiguously convey the desired gender and number properties of the speaker and audience. More concretely, a sentence such as “I love you” is ambiguous with respect to the gender of the speaker and the gender and number of the audience. However, sentences such as “I love you, she told him” are unambiguous given the coreference groups {I, she} and {you, him} which determine I to be feminine singular and you to be masculine singular. We can thus inject the desired information by prefixing a sentence with short generic sentence fragment such as “She told him:” or “She told them that”, relying on the NMT system's coreference tracking abilities to trigger the correctly marked translation, and then remove the redundant translated prefix from the generated target sentence. We observed that using a parataxis construction (i.e. “she said to him:”) almost exclusively results in target-side parataxis as well (in 99.8% of our examples), making it easy to identify and strip the translated version from the target side. Moreover, because the parataxis construction is grammatically isolated from the rest of the sentence, it can be stripped without requiring additional changes or modification to the rest of the sentence, ensuring grammaticality. ### Experiments & Results
To demonstrate our method in a black-box setting, we focus our experiments on Google's machine translation system (GMT), accessed through its Cloud API. To test the method on real-world sentences, we consider a monologue from the stand-up comedy show “Sarah Silverman: A Speck of Dust”. The monologue consists of 1,244 English sentences, all by a female speaker conveyed to a plural, gender-neutral audience. Our parallel corpora consists of the 1,244 English sentences from the transcript, and their corresponding Hebrew translations based on the Hebrew subtitles. We translate the monologue one sentence at a time through the Google Cloud API. Eyeballing the results suggest that most of the translations use the incorrect, but default, masculine and singular forms for the speaker and the audience, respectively. We expect that by adding the relevant condition of “female speaking to an audience” we will get better translations, affecting both the gender of the speaker and the number of the audience. To verify this, we experiment with translating the sentences with the following variations: No Prefix—The baseline translation as returned by the GMT system. “He said:”—Signaling a male speaker. We expect to further skew the system towards masculine forms. “She said:”—Signaling a female speaker and unknown audience. As this matches the actual speaker's gender, we expect an improvement in translation of first-person pronouns and verbs with first-person pronouns as subjects. “I said to them:”—Signaling an unknown speaker and plural audience. “He said to them:”—Masculine speaker and plural audience. “She said to them:”—Female speaker and plural audience—the complete, correct condition. We expect the best translation accuracy on this setup. “He/she said to him/her”—Here we set an (incorrect) singular gender-marked audience, to investigate our ability to control the audience morphology. ### Quantitative Results
We compare the different conditions by comparing BLEU BIBREF5 with respect to the reference Hebrew translations. We use the multi-bleu.perl script from the Moses toolkit BIBREF6 . Table shows BLEU scores for the different prefixes. The numbers match our expectations: Generally, providing an incorrect speaker and/or audience information decreases the BLEU scores, while providing the correct information substantially improves it - we see an increase of up to 2.3 BLEU over the baseline. We note the BLEU score improves in all cases, even when given the wrong gender of either the speaker or the audience. We hypothesise this improvement stems from the addition of the word “said” which hints the model to generate a more “spoken” language which matches the tested scenario. Providing correct information for both speaker and audience usually helps more than providing correct information to either one of them individually. The one outlier is providing “She” for the speaker and “her” for the audience. While this is not the correct scenario, we hypothesise it gives an improvement in BLEU as it further reinforces the female gender in the sentence. ### Qualitative Results
The BLEU score is an indication of how close the automated translation is to the reference translation, but does not tell us what exactly changed concerning the gender and number properties we attempt to control. We perform a finer-grained analysis focusing on the relation between the injected speaker and audience information, and the morphological realizations of the corresponding elements. We parse the translations and the references using a Hebrew dependency parser. In addition to the parse structure, the parser also performs morphological analysis and tagging of the individual tokens. We then perform the following analysis. Speaker's Gender Effects: We search for first-person singular pronouns with subject case (ani, unmarked for gender, corresponding to the English I), and consider the gender of its governing verb (or adjectives in copular constructions such as `I am nice'). The possible genders are `masculine', `feminine' and `both', where the latter indicates a case where the none-diacriticized written form admits both a masculine and a feminine reading. We expect the gender to match the ones requested in the prefix. Interlocutors' Gender and Number Effects: We search for second-person pronouns and consider their gender and number. For pronouns in subject position, we also consider the gender and number of their governing verbs (or adjectives in copular constructions). For a singular audience, we expect the gender and number to match the requested ones. For a plural audience, we expect the masculine-plural forms. Results: Speaker. Figure FIGREF3 shows the result for controlling the morphological properties of the speaker ({he, she, I} said). It shows the proportion of gender-inflected verbs for the various conditions and the reference. We see that the baseline system severely under-predicts the feminine form of verbs as compared to the reference. The “He said” conditions further decreases the number of feminine verbs, while the “I said” conditions bring it back to the baseline level. Finally, the “She said” prefixes substantially increase the number of feminine-marked verbs, bringing the proportion much closer to that of the reference (though still under-predicting some of the feminine cases). Results: Audience. The chart in Figure FIGREF3 shows the results for controlling the number of the audience (...to them vs nothing). It shows the proportion of singular vs. plural second-person pronouns on the various conditions. It shows a similar trend: the baseline system severely under-predicts the plural forms with respect to the reference translation, while adding the “to them” condition brings the proportion much closer to that of the reference. ### Comparison to vanmassenhove-hardmeier-way:2018:EMNLP
Closely related to our work, vanmassenhove-hardmeier-way:2018:EMNLP proposed a method and an English-French test set to evaluate gender-aware translation, based on the Europarl corpus BIBREF7 . We evaluate our method (using Google Translate and the given prefixes) on their test set to see whether it is applicable to another language pair and domain. Table shows the results of our approach vs. their published results and the Google Translate baseline. As may be expected, Google Translate outperforms their system as it is trained on a different corpus and may use more complex machine translation models. Using our method improves the BLEU score even further. ### Other Languages
To test our method’s outputs on multiple languages, we run our pre-and post-processing steps with Google Translate using examples we sourced from native speakers of different languages. For every example we have an English sentence and two translations in the corresponding language, one in masculine and one in feminine form. Not all examples are using the same source English sentence as different languages mark different information. Table shows that for these specific examples our method worked on INLINEFORM0 of the languages we had examples for, while for INLINEFORM1 languages both translations are masculine, and for 1 language both are feminine. ### Related Work
E17-1101 showed that given input with author traits like gender, it is possible to retain those traits in Statistical Machine Translation (SMT) models. W17-4727 showed that incorporating morphological analysis in the decoder improves NMT performance for morphologically rich languages. burlot:hal-01618387 presented a new protocol for evaluating the morphological competence of MT systems, indicating that current translation systems only manage to capture some morphological phenomena correctly. Regarding the application of constraints in NMT, N16-1005 presented a method for controlling the politeness level in the generated output. DBLP:journals/corr/FiclerG17aa showed how to guide a neural text generation system towards style and content parameters like the level of professionalism, subjective/objective, sentiment and others. W17-4811 showed that incorporating more context when translating subtitles can improve the coherence of the generated translations. Most closely to our work, vanmassenhove-hardmeier-way:2018:EMNLP also addressed the missing gender information by training proprietary models with a gender-indicating-prefix. We differ from this work by treating the problem in a black-box manner, and by addressing additional information like the number of the speaker and the gender and number of the audience. ### Conclusions
We highlight the problem of translating between languages with different morphological systems, in which the target translation must contain gender and number information that is not available in the source. We propose a method for injecting such information into a pre-trained NMT model in a black-box setting. We demonstrate the effectiveness of this method by showing an improvement of 2.3 BLEU in an English-to-Hebrew translation setting where the speaker and audience gender can be inferred. We also perform a fine-grained syntactic analysis that shows how our method enables to control the morphological realization of first and second-person pronouns, together with verbs and adjectives related to them. In future work we would like to explore automatic generation of the injected context, or the use of cross-sentence context to infer the injected information. Table 1: BLEU results on the Silverman dataset Figure 1: Gender inflection statistics for verbs governed by first-person pronouns. Table 2: Comparison of our approach (using Google Translate) to Vanmassenhove et al. (2018) on their English-French gender corpus. Table 3: Examples of languages where the speaker’s gender changes morphological markings in different languages, and translations using the prefix “He said:” or “She said:” accordingly
|
Speaker's Gender Effects, Interlocutors' Gender and Number Effects
|
Why did Miss Jervis think that Don worked for the government?
A. Because it seemed as though everyone in the area worked for the government.
B. His appearance made her think so.
C. Because he was familiar with Senator Bobby Thebold.
D. Because he was handcuffed to a briefcase.
|
And Then the Town Took Off by RICHARD WILSON ACE BOOKS, INC. 23 West 47th Street, New York 36, N.Y. AND THEN THE TOWN TOOK OFF Copyright ©, 1960, by Ace Books, Inc. All Rights Reserved For Felicitas K. Wilson THE SIOUX SPACEMAN Copyright ©, 1960, by Ace Books, Inc. Printed in U.S.A. THE CITY THAT RAN OFF THE MAP The town of Superior, Ohio, certainly was living up to its name! In what was undoubtedly the most spectacular feat of the century, it simply picked itself up one night and rose two full miles above Earth! Radio messages stated simply that Superior had seceded from Earth. But Don Cort, stranded on that rising town, was beginning to suspect that nothing was simple about Superior except its citizens. Calmly they accepted their rise in the world as being due to one of their local townspeople, a crackpot professor. But after a couple of weeks of floating around, it began to be obvious that the professor had no idea how to get them down. So then it was up to Cort: either find a way to anchor Superior, or spend the rest of his days on the smallest—and the nuttiest—planet in the galaxy! I The town of Superior, Ohio, disappeared on the night of October 31. A truck driver named Pierce Knaubloch was the first to report it. He had been highballing west along Route 202, making up for the time he'd spent over a second cup of coffee in a diner, when he screeched to a stop. If he'd gone another twenty-five feet he'd have gone into the pit where Superior had been. Knaubloch couldn't see the extent of the pit because it was too dark, but it looked big. Bigger than if a nitro truck had blown up, which was his first thought. He backed up two hundred feet, set out flares, then sped off to a telephone. The state police converged on the former site of Superior from several directions. Communicating by radiophone across the vast pit, they confirmed that the town undoubtedly was missing. They put in a call to the National Guard. The guard surrounded the area with troops—more than a thousand were needed—to keep people from falling into the pit. A pilot who flew over it reported that it looked as if a great ice-cream scoop had bitten into the Ohio countryside. The Pennsylvania Railroad complained that one of its passenger trains was missing. The train's schedule called for it to pass through but not stop at Superior at 11:58. That seemed to fix the time of the disappearance at midnight. The truck driver had made his discovery shortly after midnight. Someone pointed out that October 31 was Halloween and that midnight was the witching hour. Somebody else said nonsense, they'd better check for radiation. A civil defense official brought up a Geiger counter, but no matter how he shook it and rapped on it, it refused to click. A National Guard officer volunteered to take a jeep down into the pit, having found a spot that seemed navigable. He was gone a long time but when he came out the other side he reported that the pit was concave, relatively smooth, and did not smell of high explosives. He'd found no people, no houses—no sign of anything except the pit itself. The Governor of Ohio asked Washington whether any unidentified planes had been over the state. Washington said no. The Pentagon and the Atomic Energy Commission denied that they had been conducting secret experiments. Nor had there been any defense plants in Superior that might have blown up. The town's biggest factory made kitchen sinks and the next biggest made bubble gum. A United Airlines pilot found Superior early on the morning of November 1. The pilot, Captain Eric Studley, who had never seen a flying saucer and hoped never to see one, was afraid now that he had. The object loomed out of a cloudbank at twelve thousand feet and Studley changed course to avoid it. He noted with only minimum satisfaction that his co-pilot also saw the thing and wondered why it wasn't moving at the terrific speed flying saucers were allegedly capable of. Then he saw the church steeple on it. A few minutes later he had relayed a message from Superior, formerly of Ohio, addressed to whom it might concern: It said that Superior had seceded from Earth. One other radio message came from Superior, now airborne, on that first day. A ham radio operator reported an unidentified voice as saying plaintively: " Cold up here!" Don Cort had been dozing in what passed for the club car on the Buckeye Cannonball when the train braked to a stop. He looked out the window, hoping this was Columbus, where he planned to catch a plane east. But it wasn't Columbus. All he could see were some lanterns jogging as trainmen hurried along the tracks. The conductor looked into the car. The redhead across the aisle in whom Don had taken a passing interest earlier in the evening asked, "Why did we stop?" "Somebody flagged us down," the conductor said. "We don't make a station stop at Superior on this run." The girl's hair was a subtle red, but false. When Don had entered the club car he'd seen her hatless head from above and noticed that the hair along the part was dark. Her eyes had been on a book and Don had the opportunity for a brief study of her face. The cheeks were full and untouched by make-up. There were lines at the corners of her mouth which indicated a tendency to arrange her expression into one of disapproval. The lips were full, like the cheeks, but it was obvious that the scarlet lipstick had contrived a mouth a trifle bigger than the one nature had given her. Her glance upward at that moment interrupted his examination, which had been about to go on to her figure. Later, though, he was able to observe that it was more than adequate. If the girl had given Don Cort more than that one glance, or if it had been a trained, all-encompassing glance, she would have seen a man in his mid-twenties—about her age—lean, tall and straight-shouldered, with once-blond hair now verging on dark brown, a face neither handsome nor ugly, and a habit of drawing the inside of his left cheek between his teeth and nibbling at it thoughtfully. But it was likely that all she noticed then was the brief case he carried, attached by a chain to a handcuff on his left wrist. "Will we be here long?" Don asked the conductor. He didn't want to miss his plane at Columbus. The sooner he got to Washington, the sooner he'd get rid of the brief case. The handcuff it was attached to was one reason why his interest in the redhead had been only passing. "Can't say," the conductor told him. He let the door close again and went down to the tracks. Don hesitated, shrugged at the redhead, said, "Excuse me," and followed the conductor. About a dozen people were milling around the train as it sat in the dark, hissing steam. Don made his way up to the locomotive and found a bigger knot of people gathered in front of the cowcatcher. Some sort of barricade had been put up across the tracks and it was covered with every imaginable kind of warning device. There were red lanterns, both battery and electric; flashlights; road flares; and even an old red shirt. Don saw two men who must have been the engineer and the fireman talking to an old bearded gentleman wearing a civil defense helmet, a topcoat and riding boots. "You'd go over the edge, I tell you," the old gentleman was saying. "If you don't get this junk off the line," the engineer said, "I'll plow right through it. Off the edge! you crazy or something?" "Look for yourself," the old man in the white helmet said. "Go ahead. Look." The engineer was exasperated. He turned to the fireman. "You look. Humor the old man. Then let's go." The bearded man—he called himself Professor Garet—went off with the fireman. Don followed them. They had tramped a quarter of a mile along the gravel when the fireman stopped. "Okay," he said "where's the edge? I don't see nothing." The tracks seemed to stretch forever into the darkness. "It's another half mile or so," the professor said. "Well, let's hurry up. We haven't got all night." The old man chuckled. "I'm afraid you have." They came to it at last, stopping well back from it. Professor Garet swelled with pride, it seemed, as he made a theatrical gesture. "Behold," he said. "Something even Columbus couldn't find. The edge of the world." True, everything seemed to stop, and they could see stars shining low on the horizon where stars could not properly be expected to be seen. Don Cort and the fireman walked cautiously toward the edge while the professor ambled ahead with the familiarity of one who had been there before. But there was a wind and they did not venture too close. Nevertheless, Don could see that it apparently was a neat, sharp edge, not one of your old ragged, random edges such as might have been caused by an explosion. This one had the feeling of design behind it. Standing on tiptoe and repressing a touch of giddiness, Don looked over the edge. He didn't have to stand on tiptoe any more than he had to sit on the edge of his seat during the exciting part of a movie, but the situation seemed to call for it. Over the edge could be seen a big section of Ohio. At least he supposed it was Ohio. Don looked at the fireman, who had an unbelieving expression on his face, then at the bearded old man, who was smiling and nodding. "You see what I mean," he said. "You would have gone right over. I believe you would have had a two-mile fall." "Of course you could have stayed aboard the train," the man driving the old Pontiac said, "but I really think you'll be more comfortable at Cavalier." Don Cort, sitting in the back seat of the car with the redhead from the club car, asked, "Cavalier?" "The college. The institute, really; it's not accredited. What did you say your name was, miss?" "Jen Jervis," she said. "Geneva Jervis, formally." "Miss Jervis. I'm Civek. You know Mr. Cort, I suppose." The girl smiled sideways. "We have a nodding acquaintance." Don nodded and grinned. "There's plenty of room in the dormitories," Civek said. "People don't exactly pound on the gates and scream to be admitted to Cavalier." "Are you connected with the college?" Don asked. "Me? No. I'm the mayor of Superior. The old town's really come up in the world, hasn't it?" "Overnight," Geneva Jervis said. "If what Mr. Cort and the fireman say is true. I haven't seen the edge myself." "You'll have a better chance to look at it in the morning," the mayor said, "if we don't settle back in the meantime." "Was there any sort of explosion?" Don asked. "No. There wasn't any sensation at all, as far as I noticed. I was watching the late show—or trying to. My house is down in a hollow and reception isn't very good, especially with old English movies. Well, all of a sudden the picture sharpened up and I could see just as plain. Then the phone rang and it was Professor Garet." "The old fellow with the whiskers and the riding boots?" Jen Jervis asked. "Yes. Osbert Garet, Professor of Magnology at the Cavalier Institute of Applied Sciences." "Professor of what?" "Magnology. As I say, the school isn't accredited. Well, Professor Garet telephoned and said, 'Hector'—that's my name, Hector Civek—'everything's up in the air.' He was having his little joke, of course. I said, 'What?' and then he told me." "Told you what?" Jen Jervis asked. "I mean, does he have any theory about it?" "He has a theory about everything. I think what he was trying to convey was that this—this levitation confirmed his magnology principle." "What's that?" Don asked. "I haven't the faintest idea. I'm a politician, not a scientist. Professor Garet went on about it for a while, on the telephone, about magnetism and gravity, but I think he was only calling as a courtesy, so the mayor wouldn't look foolish the next morning, not knowing his town had flown the coop." "What's the population of Superior?" "Three thousand, including the students at the institute. Three thousand and forty, counting you people from the train. I guess you'll be with us for a while." "What do you mean by that?" Jen Jervis asked. "Well, I don't see how you can get down. Do you?" "Does Superior have an airport?" Don asked. "I've got to get back to—to Earth." It sounded odd to put it that way. "Nope," Civek said. "No airport. No place for a plane to land, either." "Maybe not a plane," Don said, "but a helicopter could land just about anywhere." "No helicopters here, either." "Maybe not. But I'll bet they're swarming all over you by morning." "Hm," said Hector Civek. Don couldn't quite catch his expression in the rearview mirror. "I suppose they could, at that. Well, here's Cavalier. You go right in that door, where the others are going. There's Professor Garet. I've got to see him—excuse me." The mayor was off across the campus. Don looked at Geneva Jervis, who was frowning. "Are you thinking," he asked, "that Mayor Civek was perhaps just a little less than completely honest with us?" "I'm thinking," she said, "that I should have stayed with Aunt Hattie another night, then taken a plane to Washington." "Washington?" Don said. "That's where I'm going. I mean where I was going before Superior became airborne. What do you do in Washington, Miss Jervis?" "I work for the Government. Doesn't everybody?" "Not everybody. Me, for instance." "No?" she said. "Judging by that satchel you're handcuffed to, I'd have thought you were a courier for the Pentagon. Or maybe State." He laughed quickly and loudly because she was getting uncomfortably close. "Oh, no. Nothing so glamorous. I'm a messenger for the Riggs National Bank, that's all. Where do you work?" "I'm with Senator Bobby Thebold, S.O.B." Don laughed again. "He sure is." " Mister Cort!" she said, annoyed. "You know as well as I do that S.O.B. stands for Senate Office Building. I'm his secretary." "I'm sorry. We'd better get out and find a place to sleep. It's getting late." " Places to sleep," she corrected. She looked angry. "Of course," Don said, puzzled by her emphasis. "Come on. Where they put you, you'll probably be surrounded by co-eds, even if I could get out of this cuff." He took her bag in his free hand and they were met by a gray-haired woman who introduced herself as Mrs. Garet. "We'll try to make you comfortable," she said. "What a night, eh? The professor is simply beside himself. We haven't had so much excitement since the cosmolineator blew up." They had a glimpse of the professor, still in his CD helmet, going around a corner, gesticulating wildly to someone wearing a white laboratory smock. II Don Cort had slept, but not well. He had tried to fold the brief case to pull it through his sleeve so he could take his coat off, but whatever was inside the brief case was too big. Cavalier had given him a room to himself at one end of a dormitory and he'd taken his pants off but had had to sleep with his coat and shirt on. He got up, feeling gritty, and did what little dressing was necessary. It was eight o'clock, according to the watch on the unhandcuffed wrist, and things were going on. He had a view of the campus from his window. A bright sun shone on young people moving generally toward a squat building, and other people going in random directions. The first were students going to breakfast, he supposed, and the others were faculty members. The air was very clear and the long morning shadows distinct. Only then did he remember completely that he and the whole town of Superior were up in the air. He went through the dormitory. A few students were still sleeping. The others had gone from their unmade beds. He shivered as he stepped outdoors. It was crisp, if not freezing, and his breath came out visibly. First he'd eat, he decided, so he'd be strong enough to go take a good look over the edge, in broad daylight, to the Earth below. The mess hall, or whatever they called it, was cafeteria style and he got in line with a tray for juice, eggs and coffee. He saw no one he knew, but as he was looking for a table a willowy blonde girl smiled and gestured to the empty place opposite her. "You're Mr. Cort," she said. "Won't you join me?" "Thanks," he said, unloading his tray. "How did you know?" "The mystery man with the handcuff. You'd be hard to miss. I'm Alis—that's A-l-i-s, not A-l-i-c-e—Garet. Are you with the FBI? Or did you escape from jail?" "How do you do. No, just a bank messenger. What an unusual name. Professor Garet's daughter?" "The same," she said. "Also the only. A pity, because if there'd been two of us I'd have had a fifty-fifty chance of going to OSU. As it is, I'm duty-bound to represent the second generation at the nut factory." "Nut factory? You mean Cavalier?" Don struggled to manipulate knife and fork without knocking things off the table with his clinging brief case. "Here, let me cut your eggs for you," Alis said. "You'd better order them scrambled tomorrow. Yes, Cavalier. Home of the crackpot theory and the latter-day alchemist." "I'm sure it's not that bad. Thanks. As for tomorrow, I hope to be out of here by then." "How do you get down from an elephant? Old riddle. You don't; you get down from ducks. How do you plan to get down from Superior?" "I'll find a way. I'm more interested at the moment in how I got up here." "You were levitated, like everybody else." "You make it sound deliberate, Miss Garet, as if somebody hoisted a whole patch of real estate for some fell purpose." "Scarcely fell , Mr. Cort. As for it being deliberate, that seems to be a matter of opinion. Apparently you haven't seen the papers." "I didn't know there were any." "Actually there's only one, the Superior Sentry , a weekly. This is an extra. Ed Clark must have been up all night getting it out." She opened her purse and unfolded a four-page tabloid. Don blinked at the headline: Town Gets High "Ed Clark's something of an eccentric, like everybody else in Superior," Alis said. Don read the story, which seemed to him a capricious treatment of an apparently grave situation. Residents having business beyond the outskirts of town today are advised not to. It's a long way down. Where Superior was surrounded by Ohio, as usual, today Superior ends literally at the town line. A Citizens' Emergency Fence-Building Committee is being formed, but in the meantime all are warned to stay well away from the edge. The law of gravity seems to have been repealed for the town but it is doubtful if the same exemption would apply to a dubious individual bent on investigating.... Don skimmed the rest. "I don't see anything about it being deliberate." Alis had been creaming and sugaring Don's coffee. She pushed it across to him and said, "It's not on page one. Ed Clark and Mayor Civek don't get along, so you'll find the mayor's statement in a box on page three, bottom." Don creased the paper the other way, took a sip of coffee, nodded his thanks, and read: Mayor Claims Secession From Earth Mayor Hector Civek, in a proclamation issued locally by hand and dropped to the rest of the world in a plastic shatter-proof bottle, said today that Superior has seceded from Earth. His reasons were as vague as his explanation. The "reasons" include these: (1) Superior has been discriminated against by county, state and federal agencies; (2) Cavalier Institute has been held up to global derision by orthodox (presumably meaning accredited) colleges and universities; and (3) chicle exporters have conspired against the Superior Bubble Gum Company by unreasonably raising prices. The "explanation" consists of a 63-page treatise on applied magnology by Professor Osbert Garet of Cavalier which the editor (a) does not understand; (b) lacks space to publish; and which (it being atrociously handwritten) he (c) has not the temerity to ask his linotype operator to set. Don said, "I'm beginning to like this Ed Clark." "He's a doll," Alis said. "He's about the only one in town who stands up to Father." "Does your father claim that he levitated Superior off the face of the Earth?" "Not to me he doesn't. I'm one of those banes of his existence, a skeptic. He gave up trying to magnolize me when I was sixteen. I had a science teacher in high school—not in Superior, incidentally—who gave me all kinds of embarrassing questions to ask Father. I asked them, being a natural-born needler, and Father has disowned me intellectually ever since." "How old are you, Miss Garet, if I may ask?" She sat up straight and tucked her sweater tightly into her skirt, emphasizing her good figure. To a male friend Don would have described the figure as outstanding. She had mocking eyes, a pert nose and a mouth of such moist red softness that it seemed perpetually waiting to be kissed. All in all she could have been the queen of a campus much more densely populated with co-eds than Cavalier was. "You may call me Alis," she said. "And I'm nineteen." Don grinned. "Going on?" "Three months past. How old are you , Mr. Cort?" "Don's the name I've had for twenty-six years. Please use it." "Gladly. And now, Don, unless you want another cup of coffee, I'll go with you to the end of the world." "On such short notice?" Don was intrigued. Last night the redhead from the club car had repelled an advance that hadn't been made, and this morning a blonde was apparently making an advance that hadn't been solicited. He wondered where Geneva Jervis was, but only vaguely. "I'll admit to the double entendre ," Alis said. "What I meant—for now—was that we can stroll out to where Superior used to be attached to the rest of Ohio and see how the Earth is getting along without us." "Delighted. But don't you have any classes?" "Sure I do. Non-Einsteinian Relativity 1, at nine o'clock. But I'm a demon class-cutter, which is why I'm still a Senior at my advanced age. On to the brink!" They walked south from the campus and came to the railroad track. The train was standing there with nowhere to go. It had been abandoned except for the conductor, who had dutifully spent the night aboard. "What's happening?" he asked when he saw them. "Any word from down there?" "Not that I know of," Don said. He introduced him to Alis Garet. "What are you going to do?" "What can I do?" the conductor asked. "You can go over to Cavalier and have breakfast," Alis said. "Nobody's going to steal your old train." The conductor reckoned as how he might just do that, and did. "You know," Don said, "I was half-asleep last night but before the train stopped I thought it was running alongside a creek for a while." "South Creek," Alis said. "That's right. It's just over there." "Is it still? I mean hasn't it all poured off the edge by now? Was that Superior's water supply?" Alis shrugged. "All I know is you turn on the faucet and there's water. Let's go look at the creek." They found it coursing along between the banks. "Looks just about the same," she said. "That's funny. Come on; let's follow it to the edge." The brink, as Alis called it, looked even more awesome by daylight. Everything stopped short. There were the remnants of a cornfield, with the withered stalks cut down, then there was nothing. There was South Creek surging along, then nothing. In the distance a clump of trees, with a few autumn leaves still clinging to their branches, simply ended. "Where is the water going?" Don asked. "I can't make it out." "Down, I'd say. Rain for the Earth-people." "I should think it'd be all dried up by now. I'm going to have a look." "Don't! You'll fall off!" "I'll be careful." He walked cautiously toward the edge. Alis followed him, a few feet behind. He stopped a yard from the brink and waited for a spell of dizziness to pass. The Earth was spread out like a topographer's map, far below. Don took another wary step, then sat down. "Chicken," said Alis. She laughed uncertainly, then she sat down, too. "I still can't see where the water goes," Don said. He stretched out on his stomach and began to inch forward. "You stay there." Finally he had inched to a point where, by stretching out a hand, he could almost reach the edge. He gave another wriggle and the fingers of his right hand closed over the brink. For a moment he lay there, panting, head pressed to the ground. "How do you feel?" Alis asked. "Scared. When I get my courage back I'll pick up my head and look." Alis put a hand out tentatively, then purposefully took hold of his ankle and held it tight. "Just in case a high wind comes along," she said. "Thanks. It helps. Okay, here we go." He lifted his head. "Damn." "What?" "It still isn't clear. Do you have a pocket mirror?" "I have a compact." She took it out of her bag with her free hand and tossed it to him. It rolled and Don had to grab to keep it from going over the edge. Alis gave a little shriek. Don was momentarily unnerved and had to put his head back on the ground. "Sorry," she said. Don opened the compact and carefully transferred it to his right hand. He held it out beyond the edge and peered into it, focusing it on the end of the creek. "Now I've got it. The water isn't going off the edge!" "It isn't? Then where is it going?" "Down, of course, but it's as if it's going into a well, or a vertical tunnel, just short of the edge." "Why? How?" "I can't see too well, but that's my impression. Hold on now. I'm coming back." He inched away from the edge, then got up and brushed himself off. He returned her compact. "I guess you know where we go next." "The other end of the creek?" "Exactly." South Creek did not bisect Superior, as Don thought it might, but flowed in an arc through a southern segment of it. They had about two miles to go, past South Creek Bridge—which used to lead to Ladenburg, Alis said—past Raleigh Country Club (a long drive would really put the ball out of play, Don thought) and on to the edge again. But as they approached what they were forced to consider the source of the creek, they found a wire fence at the spot. "This is new," Alis said. The fence, which had a sign on it, warning—electrified , was semicircular, with each end at the edge and tarpaulins strung behind it so they could see the mouth of the creek. The water flowed from under the tarp and fence. "Look how it comes in spurts," Alis said. "As if it's being pumped." Smaller print on the sign said: Protecting mouth of South Creek, one of two sources of water for Superior. Electrical charge in fence is sufficient to kill. It was signed: Vincent Grande, Chief of Police, Hector Civek, Mayor . "What's the other source, besides the faucet in your bathroom?" Don asked. "North Lake, maybe," Alis said. "People fish there but nobody's allowed to swim." "Is the lake entirely within the town limits?" "I don't know." "If it were on the edge, and if I took a rowboat out on it, I wonder what would happen?" "I know one thing—I wouldn't be there holding your ankle while you found out." She took his arm as they gazed past the electrified fence at the Earth below and to the west. "It's impressive, isn't it?" she said. "I wonder if that's Indiana way over there?" He patted her hand absent-mindedly. "I wonder if it's west at all. I mean, how do we know Superior is maintaining the same position up here as it used to down there?" "We could tell by the sun, silly." "Of course," he said, grinning at his stupidity. "And I guess we're not high enough to see very far. If we were we'd be able to see the Great Lakes—or Lake Erie, anyway." They were musing about the geography when a plane came out of a cloudbank and, a second later, veered sharply. They could make out UAL on the underside of a wing. As it turned they imagined they could see faces peering out of the windows. They waved and thought they saw one or two people wave back. Then the plane climbed toward the east and was gone. "Well," Don said as they turned to go back to Cavalier, "now we know that they know. Maybe we'll begin to get some answers. Or, if not answers, then transportation." "Transportation?" Alis squeezed the arm she was holding. "Why? Don't you like it here?" "If you mean don't I like you, the answer is yes, of course I do. But if I don't get out of this handcuff soon so I can take a bath and get into clean clothes, you're not going to like me." "You're still quite acceptable, if a bit whiskery." She stopped, still holding his arm, and he turned so they were face to face. "So kiss me," she said, "before you deteriorate." They were in the midst of an extremely pleasant kiss when the brief case at the end of Don's handcuff began to talk to him.
|
D. Because he was handcuffed to a briefcase.
|
Which “Joe” faces the brunt of Colonel Walsh’s racism?
A. Bartender Joe
B. Trader Joe
C. Military Joe
D. Jungle Guide Joe
|
A PLANET NAMED JOE By S. A. LOMBINO There were more Joes on Venus than you could shake a ray-gun at. Perhaps there was method in Colonel Walsh's madness—murder-madness—when he ordered Major Polk to scan the planet for a guy named Joe. [Transcriber's Note: This etext was produced from Planet Stories November 1952. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] Colonel Walsh had a great sense of humor. I hated his guts ever since we went through the Academy together, but he had a great sense of humor. For example, he could have chosen a Second Looie for the job on Venus. He might even have picked a Captain. But he liked me about as much as I liked him, and so he decided the job was just right for a Major. At least, that's what he told me. I stood at attention before his desk in the Patrol Station. We were somewhere in Area Two on Earth, takeoff point for any operations in Space II. The duty was fine, and I liked it a lot. Come to think of it, the most I ever did was inspect a few defective tubes every now and then. The rest was gravy, and Colonel Walsh wasn't going to let me get by with gravy. "It will be a simple assignment, Major," he said to me, peering over his fingers. He held them up in front of him like a cathedral. "Yes, sir," I said. "It will involve finding one man, a Venusian native." I wanted to say, "Then why the hell don't you send a green kid on the job? Why me?" Instead, I nodded and watched him playing with his fingers. "The man is a trader of sorts. Rather intelligent." He paused, then added, "For a native, that is." I had never liked Walsh's attitude toward natives. I hadn't liked the way he'd treated the natives on Mars ever since he'd taken over there. Which brought to mind an important point. "I always figured Venus was under the jurisdiction of Space III, sir. I thought our activities were confined to Mars." He folded his fingers like a deck of cards and dropped them on his desk as if he were waiting for me to cut. "Mmmm," he said, "yes, that's true. But this is a special job. It so happens this Venusian is the one man who can help us understand just what's happening on Mars." I tried to picture a Venusian understanding Mars and I didn't get very far. "He's had many dealings with the natives there," Walsh explained. "If anyone can tell us the reasons for the revolt, he can." If Walsh really wanted to know the reasons for the revolt, I could give them to him in one word: Walsh. I had to laugh at the way he called it "revolt." It had been going on for six months now and we'd lost at least a thousand men from Space II. Revolt. "And this man is on Venus now?" I asked for confirmation. I'd never been to Venus, being in Space II ever since I'd left the Moon run. It was just like Walsh to ship me off to a strange place. "Yes, Major," he said. "This man is on Venus." At the Academy he had called me Fred. That was before I'd reported him for sleeping on Boiler Watch. He'd goofed off on a pile of uranium that could've, and almost did, blow the barracks sky-high that night. He still thought it was my fault, as if I'd done the wrong thing by reporting him. And now, through the fouled-up machinery that exists in any military organization, he outranked me. "And the man's name, sir?" "Joe." A tight smile played on his face. "Joe what?" I asked. "Just Joe." "Just Joe?" "Yes," Walsh said. "A native, you know. They rarely go in for more than first names. But then, it should be simple to find a man with a name like Joe. Among the natives, I mean." "I don't know, sir." "A relatively simple assignment," Walsh said. "Can you tell me anything else about this man? Physical appearance? Personal habits? Anything?" Walsh seemed to consider this for a moment. "Well, physically he's like any of the other Venusians, so I can't give you much help there. He does have a peculiar habit, though." "What's that?" "He has an affinity for Terran cigarettes." I sighed. "Well, it's not very much to go on." "You'll find him," Walsh said, grinning. "I'm sure of it." The trip to Venus came off without a hitch. I did a lot of thinking on that trip. I thought about Mars and the revolt there. And I thought about Colonel Leonard Walsh and how he was supposed to be quelling that revolt. Ever since Walsh had taken command, ever since he'd started pushing the natives around, there'd been trouble. It was almost as if the whole damned planet had blown up in our faces the moment he took over. Swell guy, Walsh. Venus was hotter than I'd expected it to be. Much too hot for the tunic I was wearing. It smelled, too. A funny smell I couldn't place. Like a mixture of old shoe and after-shave. There were plants everywhere I looked. Big plants and small ones, some blooming with flowers I'd never seen before, and some as bare as cactus. I recognized a blue figure as one of the natives the pilot had told me about. He was tall, looking almost human except that everything about him was elongated. His features, his muscles, everything seemed to have been stretched like a rubber band. I kept expecting him to pop back to normal. Instead, he flashed a double row of brilliant teeth at me. I wondered if he spoke English. "Hey, boy," I called. He ambled over with long-legged strides that closed the distance between us in seconds. "Call me Joe," he said. I dropped my bags and stared at him. Maybe this was going to be a simple assignment after all. "I sure am glad to see you, Joe," I said. "Same here, Toots," he answered. "The guys back in Space II are searching high and low for you," I told him. "You've got the wrong number," he said, and I was a little surprised at his use of Terran idiom. "You are Joe, aren't you? Joe the trader?" "I'm Joe, all right," he said. "Only thing I ever traded, though, was a pocketknife. Got a set of keys for it." "Oh," I said, my voice conveying my disappointment. I sighed and began wondering just how I should go about contacting the Joe I was looking for. My orders said I was to report to Captain Bransten immediately upon arrival. I figured the hell with Captain Bransten. I outranked him anyway, and there wasn't much he could do if I decided to stop for a drink first. "Where's the Officer's Club?" I asked the Venusian. "Are you buying information or are you just curious?" "Can you take me there?" I asked. "Sure thing, Toots." He picked up my bags and started walking up a heavily overgrown path. We'd probably walked for about ten minutes when he dropped my bags and said, "There it is." The Officer's Club was a plasteel hut with window shields that protected it from the heat of the sun. It didn't look too comfortable but I really wanted that drink. I reached into my tunic and slipped the native thirty solars. He stared at the credits curiously and then shrugged his shoulders. "Oh well, you're new here. We'll let it go." He took off then, while I stared after him, wondering just what he'd meant. Had I tipped him too little? I shrugged and looked over at the Officer's Club. From the outside it looked as hot as hell. On the inside it was about two degrees short of that mark. I began to curse Walsh for taking me away from my nice soft job in Space II. There wasn't much inside the club. A few tables and chairs, a dart game and a bar. Behind the bar a tall Venusian lounged. I walked over and asked, "What are you serving, pal?" "Call me Joe," he answered. He caught me off balance. "What?" "Joe," he said again. A faint glimmer of understanding began to penetrate my thick skull. "You wouldn't happen to be Joe the trader? The guy who knows all about Mars, would you?" "I never left home," he said simply. "What are you drinking?" That rat! That dirty, filthy, stinking, unprincipled.... But then, it should be simple to find a man with a name like Joe. Among the natives, I mean. Sure. Oh sure. Real simple. Walsh was about the lowest, most contemptible.... "What are you drinking, pal?" the Venusian asked again. "Skip it," I said. "How do I get to the captain's shack?" "Follow your nose, pal. Can't miss it." I started to pick up my bag as another Venusian entered. He waved at the bartender. "Hello, Joe," he said. "How's it going?" "Not so hot, Joe," the bartender replied. I listened in fascination. Joe, Joe, Joe. So this was Walsh's idea of a great gag. Very funny. Very.... "You Major Polk, sweetheart?" the Venusian who'd just come in asked. "Yes," I said, still thinking of Colonel Walsh. "You better get your butt over to the captain's shack," he said. "He's about ready to post you as overdue." "Sure," I said wearily. "Will you take my bags, please?" "Roger," he answered. He picked up the bags and nodded at the bar. "So long, Joe," he said to the bartender. "See you, Joe," the bartender called back. Captain Bransten was a mousey, unimpressive sort of man. He was wearing a tropical tunic, but he still resembled a wilted lily more than he did an officer. "Have a seat, Major," he offered. He reached for a cigarette box on the desk and extended it to me. He coughed in embarrassment when he saw it was empty. Quickly, he pressed a button on his desk and the door popped open. A tall, blue Venusian stepped lithely into the room. "Sir?" the Venusian asked. "We're out of cigarettes, Joe," the Captain said. "Will you get us some, please?" "Sure thing," the Venusian answered. He smiled broadly and closed the door behind him. Another Joe , I thought. Another damned Joe. "They steal them," Captain Bransten said abruptly. "Steal what?" I asked. "Cigarettes. I sometimes think the cigarette is one of the few things they like about Terran culture." So Walsh had taken care of that angle too. He does have a peculiar habit, though. He has an affinity for Terran cigarettes. Cigarettes was the tip I should have given; not solars. "All right," I said, "suppose we start at the beginning." Captain Bransten opened his eyes wide. "Sir?" he asked. "What's with all this Joe business? It may be a very original name but I think its popularity here is a little outstanding." Captain Bransten began to chuckle softly. I personally didn't think it was so funny. I tossed him my withering Superior Officer's gaze and waited for his explanation. "I hadn't realized this was your first time on Venus," he said. "Is there a local hero named Joe?" I asked. "No, no, nothing like that," he assured me. "It's a simple culture, you know. Not nearly as developed as Mars." "I can see that," I said bitingly. "And the natives are only now becoming acquainted with Terran culture. Lots of enlisted men, you know." I began to get the idea. And I began to appreciate Walsh's doubtful ancestry more keenly. "It's impossible to tell exactly where it all started, of course," Bransten was saying. I was beginning to get angry. Very angry. I was thinking of Walsh sitting back in a nice cozy foam chair back on Earth. "Get to the point, Captain!" I barked. "Easy, sir," Bransten said, turning pale. I could see that the Captain wasn't used to entertaining Majors. "The enlisted men. You know how they are. They'll ask a native to do something and they'll call him Joe. 'Hey, Joe, give me a hand with this.' Or 'Listen, Joe, how'd you like to earn some cigarettes?' Do you follow?" "I follow, all right," I said bitterly. "Well," Bransten went on, "that sort of thing mushrooms. The natives are a simple, almost childish people. It appealed to them—the Joe business, I mean. Now they're all Joe. They like it. That and the cigarettes." He cleared his throat and looked at me apologetically as if he were personally responsible for Venusian culture. In fact, he looked as if he were responsible for having put Venus in the heavens in the first place. "Do you understand, Major? Just a case of extended idiom, that's all." Just a case of extended idiot , I thought. An idiot on a wild goose chase a hell of a long way from home. "I understand perfectly," I snapped. "Where are my quarters?" Bransten asked a Venusian named Joe to show me my quarters, reminding me that chow was at thirteen hundred. As I was leaving, the first Venusian came back with the cigarettes Bransten had ordered. I could tell by the look on his face that he probably had half a carton stuffed into his pockets. I shrugged and went to change into a tropical tunic. I called Earth right after chow. The Captain assured me that this sort of thing was definitely against regulations, but he submitted when I twinkled my little gold leaf under his nose. Walsh's face appeared on the screen. He was smiling, looking like a fat pussy cat. "What is it, Major?" he asked. "This man Joe," I said. "Can you give me any more on him?" Walsh's grin grew wider. "Why, Major," he said, "you're not having any difficulties, are you?" "None at all," I snapped back. "I just thought I'd be able to find him a lot sooner if...." "Take your time, Major," Walsh beamed. "There's no rush at all." "I thought...." "I'm sure you can do the job," Walsh cut in. "I wouldn't have sent you otherwise." Hell, I was through kidding around. "Look...." "He's somewhere in the jungle, you know," Walsh said. I wanted to ram my fist into the screen, right smack up against those big white teeth. Instead, I cut off the transmission and watched the surprised look on his face as his screen went blank millions of miles away. He blinked at the screen, trying to realize I'd deliberately hung up on him. "Polk!" he shouted, "can you hear me?" I smiled, saw the twisted hatred on his features, and then the screen on my end went blank, too. He's somewhere in the jungle, you know. I thanked Captain Bransten for his hospitality and went back to my quarters. As I saw it, there were two courses for me to follow. One: I could say the hell with Walsh and Venus. That would mean hopping the next ship back to Earth. It would also mean disobeying the direct order of a superior officer. It might mean demotion, and it might mean getting bounced out of the Service altogether. Two: I could assume there really was a guy name Joe somewhere in that jungle, a Joe separate and apart from the other Joes on this planet, a trader Joe who knew the Martians well. I could always admit failure, of course, and return empty handed. Mission not accomplished. Or, I might really find a guy who was trader Joe. I made my decision quickly. I wanted to stay in the Service, and besides Walsh may have been on the level for the first time in his life. Maybe there was a Joe here who could help us on Mars. If there was I'd try to find him. It was still a hell of a trick though. I cursed Walsh again and pushed the buzzer near my bed. A tall Venusian stepped into the room. "Joe?" I asked, just to be sure. "Who else, boss?" he answered. "I'm trying to locate someone," I said. "I'll need a guide to take me into the jungle. Can you get me one?" "It'll cost you, boss," the Venusian said. "How much?" "Two cartons of cigarettes at least." "Who's the guide?" I asked. "How's the price sound?" "Fine, fine," I said impatiently. And the Captain had said they were almost a childish people! "His name is Joe," the Venusian told me. "Best damn guide on the planet. Take you anywhere you want to go, do anything you want to do. Courageous. Doesn't know the meaning of fear. I've known him to...." "Skip it," I said, cutting the promotion short. "Tell him to show up around fifteen hundred with a complete list of what we'll need." The Venusian started to leave. "And Joe," I said, stopping him at the door, "I hope you're not overlooking your commission on the deal." His face broke into a wide grin. "No danger of that, boss," he said. When he was gone I began figuring out a plan of action. Obviously, I'd just have to traipse through the jungle looking for a guy named Joe on a planet where everyone was named Joe. Everybody, at least, but the Captain, the small garrison attached to the Station, and me. I began wondering why Walsh had gone to so much trouble to get rid of me. The job, as I saw it, would take a hell of a long time. It seemed like a silly thing to do, just to get even with a guy for something that had happened years ago. He surely must have realized that I'd be back again, sooner or later. Maybe he had another little junket all set for me. Or maybe he didn't expect me to come back. The thought hadn't occurred to me before this, and I began to consider it seriously. Walsh was no good, rotten clear through. He was failing at the job of keeping Mars in hand, and he probably realized that a few more mistakes on his part would mean the end of his career with Space II. I chuckled as I thought of him isolated in some God-forsaken place like Space V or Space VII. This probably bothered him a lot, too. But what probably bothered him more was the fact that I was next in command. If he were transferred, I'd be in charge of Space II, and I could understand how much that would appeal to Walsh. I tried to figure the thing out sensibly, tried to weigh his good points against his bad. But it all came back to the same thing. A guy who would deliberately go to sleep on Boiler Watch with a ton of uranium ready to blast a barracks to smithereens if it wasn't watched, would deliberately do just about anything. Sending me off on a wild goose chase after a character named Joe may have been a gag. But it may have been something a little grimmer than a gag, and I made up my mind to be extremely careful from here on in. The guide arrived at fifteen hundred on the dot. He was tall, elongated, looked almost like all the other Venusians I'd seen so far. "I understand you need a Grade A guide, sir," he said. "Are you familiar with the jungle?" I asked him. "Born and raised there, sir. Know it like the back of my hand." "Has Joe told you what the payment will be?" "Yes, sir. A carton and a half of cigarettes." I thought about Joe deducting his commission and smiled. "When can we leave?" "Right away, sir. We won't need much really. I've made a list of supplies and I can get them in less than an hour. I suggest you wear light clothing, boots, and a hat." "Will I need a weapon?" He looked at me, his eyes faintly amused. "Why, what for, sir?" "Never mind," I said. "What's your name, by the way?" He lifted his eyebrows, and his eyes widened in his narrow face. He was definitely surprised. "Joe," he said. "Didn't you know?" When we'd been out for a while I discovered why Joe had suggested the boots and the hat. The undergrowth was often sharp and jagged and it would have sliced my legs to ribbons were they not protected by the high boots. The hat kept the strong sun off my head. Joe was an excellent guide and a pleasant companion. He seemed to be enjoying a great romp, seemed to love the jungle and take a secret pleasure in the work he was doing. There were times when I couldn't see three feet ahead of me. He'd stand stock still for a few minutes, his head barely moving, his eyes darting from one plant to another. Then he'd say, "This way," and take off into what looked like more impenetrable jungle invariably to find a little path leading directly to another village. Each village was the same. The natives would come running out of their huts, tall and blue, shouting, "Cigarettes, Joe? Cigarettes?" It took me a while to realize they were addressing me and not my guide. Everybody was Joe. It was one beautiful, happy, joyous round of stinking, hot jungle. And I wasn't getting any nearer my man. Nor had I any idea how I was supposed to find him. I began to feel pretty low about the whole affair. Joe, on the other hand, enjoyed every moment of the trip. In each village he greeted the natives cheerfully, told them stories, swapped gossip and jokes. And when it was time to leave, he would say goodbye to all his friends and we would plunge into the twisted foliage again. His spirits were always high and he never failed to say the right thing that would give a momentary lift to my own depressed state of mind. He would talk for hours on end as we hacked our way through the jungle. "I like Venus," he said once. "I would never leave it." "Have you ever been to Earth?" I asked. "No," Joe replied. "I like Terrans too, you understand. They are good for Venus. And they are fun." "Fun?" I asked, thinking of a particular species of Terran: species Leonard Walsh. "Yes, yes," he said wholeheartedly. "They joke and they laugh and ... well, you know." "I suppose so," I admitted. Joe smiled secretly, and we pushed on. I began to find, more and more, that I had started to talk freely to Joe. In the beginning he had been just my guide. There had been the strained relationship of employer and employee. But as the days lengthened into weeks, the formal atmosphere began to crumble. I found myself telling him all about Earth, about the people there, about my decision to attend the Academy, the rigid tests, the grind, even the Moon run. Joe was a good listener, nodding sympathetically, finding experiences in his own life to parallel my own. And as our relationship progressed from a casual one to a definitely friendly one, Joe seemed more enthusiastic than ever to keep up our grinding pace to find what we were looking for. Once we stopped in a clearing to rest. Joe lounged on the matted greenery, his long body stretched out in front of him, the knife gleaming in his belt. I'd seen him slash his way through thick, tangled vines with that knife, his long, muscular arms powerfully slicing through them like strips of silk. "How far are we from the Station?" I asked. "Three or four Earth weeks," he replied. I sighed wearily. "Where do we go from here?" "There are more villages," he said. "We'll never find him." "Possibly," Joe mused, the smile creeping over his face again. "A wild goose chase. A fool's errand." "We'd better get started," Joe said simply. I got to my feet and we started the march again. Joe was still fresh, a brilliant contrast to me, weary and dejected. Somehow, I had the same feeling I'd had a long time ago on my sixteenth birthday. One of my friends had taken me all over the city, finally dropping me off at my own house where the whole gang was gathered for a surprise party. Joe reminded me of that friend. "There's a village ahead," he said, and the grin on his face was large now, his eyes shining. Something was missing here. Natives. There were no natives rushing out to greet us. No cries of "Cigarettes? Cigarettes?" I caught up with Joe. "What's the story?" I whispered. He shrugged knowingly and continued walking. And then I saw the ship, nose pointing into space, catching the rays of the sun like a great silver bullet. "What...?" I started. "It's all right," Joe said, smiling. The ship looked vaguely familiar. I noticed the crest of Space II near the nose, and a lot of things became clear then. I also saw Walsh standing near one of the huts, a stun gun in his hand. "Hello, Major," he called, almost cheerfully. The gun didn't look cheerful, though. It was pointed at my head. "Fancy meeting you here, Colonel," I said, trying to match his joviality. Somehow it didn't quite come off. Joe was walking beside me, waving at the colonel, beaming all over with happiness. "I see you found your man," Walsh said. I turned rapidly. Joe nodded and kept grinning, a grin that told me he was getting a big kick out of all this. Like a kid playing a game. I faced Walsh again. "Okay, what's it all about, pal?" "Colonel," Walsh corrected me. "You mustn't forget to say Colonel, Major ." He emphasized my rank, and he said it with a sort of ruthless finality. I waited. I could see he was just busting to tell me how clever he'd been. Besides, there wasn't much I could do but wait. Not with Walsh pointing the stun gun at my middle. "We've come a long way since the Academy, haven't we, Major?" "If you mean in miles," I said, looking around at the plants, "we sure have." Walsh grinned a little. "Always the wit," he said drily. And then the smile faded from his lips and his eyes took on a hard lustre. "I'm going to kill you, you know." He said it as if he were saying, "I think it'll rain tomorrow." Joe almost clapped his hands together with glee. He was really enjoying this. Another of those funny Terran games. "You gave me a powerful handicap to overcome," Walsh said. "I suppose I should thank you, really." "You're welcome," I said. "It wasn't easy living down the disgrace you caused me." "It was your own damn fault," I said. "You knew what you were doing when you decided to cork off." Beside me, Joe chuckled a little, enjoying the game immensely. "You didn't have to report me," Walsh said. "No? Maybe I should have forgotten all about it? Maybe I should have nudged you and served you orange juice? So you could do it again sometime and maybe blow up the whole damn Academy!" Walsh was silent for a long time. When he spoke his voice was barely audible. The heat was oppressive, as if it were concentrated on this little spot in the jungle, focusing all its penetration on a small, unimportant drama. I could hear Joe breathing beside me. "I'm on my way out," Walsh rasped. "Finished, do you understand?" "Good," I said. And I meant it. "This Mars thing. A terrible fix. Terrible." Beside me, a slight frown crossed Joe's face. Apparently he couldn't understand the seriousness of our voices. What had happened to the game, the fun? "You brought the Mars business on yourself," I told Walsh. "There was never any trouble before you took command." "The natives," he practically shouted. "They ... they...." Joe caught his breath sharply, and I wondered what Walsh was going to say about the natives. Apparently he'd realized that Joe was a native. Or maybe Joe's knife had something to do with it. "What about the natives?" I asked. "Nothing," Walsh said. "Nothing." He was silent for a while. "A man of my calibre," he said then, his face grim. "Dealing with savages." He caught himself again and threw a hasty glance at Joe. The perplexed frown had grown heavier on Joe's face. He looked at the colonel in puzzlement.
|
D. Jungle Guide Joe
|
In which chapter can we find more on OA policies?
A. Chapter 4
B. Chapter 5
C. Chapter 9
D. Chapter 2
|
What Is Open Access? Shifting from ink on paper to digital text suddenly allows us to make perfect copies of our work. Shifting from isolated computers to a globe-spanning network of connected computers suddenly allows us to share perfect copies of our work with a worldwide audience at essentially no cost. About thirty years ago this kind of free global sharing became something new under the sun. Before that, it would have sounded like a quixotic dream. Digital technologies have created more than one revolution. Let’s call this one the access revolution. Why don’t more authors take advantage of the access revolution to reach more readers? The answer is pretty clear. Authors who share their works in this way aren’t selling them, and even authors with purposes higher than money depend on sales to make a living. Or at least they appreciate sales. Let’s sharpen the question, then, by putting to one side authors who want to sell their work. We can even acknowledge that we’re putting aside the vast majority of authors. Imagine a tribe of authors who write serious and useful work, and who follow a centuries-old custom of giving it away without charge. I don’t mean a group of rich authors who don’t need money. I mean a group of authors defined by their topics, genres, purposes, incentives, and institutional circumstances, not by their wealth. In fact, very few are wealthy. For now, it doesn’t matter who these authors are, how rare they are, what they write, or why they follow this peculiar custom. It’s enough to know that their employers pay them salaries, freeing them to give away their work, that they write for impact rather than money, and that they score career points when they make the kind of impact they hoped to make. Suppose that selling their work would actually harm their interests by shrinking their audience, reducing their impact, and distorting their professional goals by steering them toward popular topics and away from the specialized questions on which they are experts. If authors like that exist, at least they should take advantage of the access revolution. The dream of global free access can be a reality for them, even if most other authors hope to earn royalties and feel obliged to sit out this particular revolution. These lucky authors are scholars, and the works they customarily write and publish without payment are peer-reviewed articles in scholarly journals. Open access is the name of the revolutionary kind of access these authors, unencumbered by a motive of financial gain, are free to provide to their readers. Open access (OA) literature is digital, online, free of charge, and free of most copyright and licensing restrictions. We could call it “barrier-free” access, but that would emphasize the negative rather than the positive. In any case, we can be more specific about which access barriers OA removes. A price tag is a significant access barrier. Most works with price tags are individually affordable. But when a scholar needs to read or consult hundreds of works for one research project, or when a library must provide access for thousands of faculty and students working on tens of thousands of topics, and when the volume of new work grows explosively every year, price barriers become insurmountable. The resulting access gaps harm authors by limiting their audience and impact, harm readers by limiting what they can retrieve and read, and thereby harm research from both directions. OA removes price barriers. Copyright can also be a significant access barrier. If you have access to a work for reading but want to translate it into another language, distribute copies to colleagues, copy the text for mining with sophisticated software, or reformat it for reading with new technology, then you generally need the permission of the copyright holder. That makes sense when the author wants to sell the work and when the use you have in mind could undermine sales. But for research articles we’re generally talking about authors from the special tribe who want to share their work as widely as possible. Even these authors, however, tend to transfer their copyrights to intermediaries—publishers—who want to sell their work. As a result, users may be hampered in their research by barriers erected to serve intermediaries rather than authors. In addition, replacing user freedom with permission-seeking harms research authors by limiting the usefulness of their work, harms research readers by limiting the uses they may make of works even when they have access, and thereby harms research from both directions. OA removes these permission barriers. Removing price barriers means that readers are not limited by their own ability to pay, or by the budgets of the institutions where they may have library privileges. Removing permission barriers means that scholars are free to use or reuse literature for scholarly purposes. These purposes include reading and searching, but also redistributing, translating, text mining, migrating to new media, long-term archiving, and innumerable new forms of research, analysis, and processing we haven’t yet imagined. OA makes work more useful in both ways, by making it available to more people who can put it to use, and by freeing those people to use and reuse it. Terminology When we need to, we can be more specific about access vehicles and access barriers. In the jargon, OA delivered by journals is called gold OA , and OA delivered by repositories is called green OA . Work that is not open access, or that is available only for a price, is called toll access (TA). Over the years I’ve asked publishers for a neutral, nonpejorative and nonhonorific term for toll-access publishers, and conventional publishers is the suggestion I hear most often. While every kind of OA removes price barriers, there are many different permission barriers we could remove if we wanted to. If we remove price barriers alone, we provide gratis OA , and if we remove at least some permission barriers as well, we provide libre OA . (Also see section 3.1 on green/gold and section 3.3 on gratis/libre.) OA was defined in three influential public statements: the Budapest Open Access Initiative (February 2002), the Bethesda Statement on Open Access Publishing (June 2003), and the Berlin Declaration on Open Access to Knowledge in the Sciences and Humanities (October 2003). I sometimes refer to their overlap or common ground as the BBB definition of OA. My definition here is the BBB definition reduced to its essential elements and refined with some post-BBB terminology (green, gold, gratis, libre) for speaking precisely about subspecies of OA. Here’s how the Budapest statement defined OA: There are many degrees and kinds of wider and easier access to [research] literature. By “open access” to this literature, we mean its free availability on the public internet, permitting any users to read, download, copy, distribute, print, search, or link to the full texts of these articles, crawl them for indexing, pass them as data to software, or use them for any other lawful purpose, without financial, legal, or technical barriers other than those inseparable from gaining access to the internet itself. The only constraint on reproduction and distribution, and the only role for copyright in this domain, should be to give authors control over the integrity of their work and the right to be properly acknowledged and cited. Here’s how the Bethesda and Berlin statements put it: For a work to be OA, the copyright holder must consent in advance to let users “copy, use, distribute, transmit and display the work publicly and to make and distribute derivative works, in any digital medium for any responsible purpose, subject to proper attribution of authorship.” Note that all three legs of the BBB definition go beyond removing price barriers to removing permission barriers, or beyond gratis OA to libre OA. But at the same time, all three allow at least one limit on user freedom: an obligation to attribute the work to the author. The purpose of OA is to remove barriers to all legitimate scholarly uses for scholarly literature, but there’s no legitimate scholarly purpose in suppressing attribution to the texts we use. (That’s why my shorthand definition says that OA literature is free of “most” rather than “all” copyright and licensing restrictions.) The basic idea of OA is simple: Make research literature available online without price barriers and without most permission barriers. Even the implementation is simple enough that the volume of peer-reviewed OA literature and the number of institutions providing it have grown at an increasing rate for more than a decade. If there are complexities, they lie in the transition from where we are now to a world in which OA is the default for new research. This is complicated because the major obstacles are not technical, legal, or economic, but cultural. (More in chapter 9 on the future.) In principle, any kind of digital content can be OA, since any digital content can be put online without price or permission barriers. Moreover, any kind of content can be digital: texts, data, images, audio, video, multimedia, and executable code. We can have OA music and movies, news and novels, sitcoms and software—and to different degrees we already do. But the term “open access” was coined by researchers trying to remove access barriers to research. The next section explains why. 1.1 What Makes OA Possible? OA is made possible by the internet and copyright-holder consent. But why would a copyright holder consent to OA? Two background facts suggest the answer. First, authors are the copyright holders for their work until or unless they transfer rights to someone else, such as a publisher. Second, scholarly journals generally don’t pay authors for their research articles, which frees this special tribe of authors to consent to OA without losing revenue. This fact distinguishes scholars decisively from musicians and moviemakers, and even from most other kinds of authors. This is why controversies about OA to music and movies don’t carry over to OA for research articles. Both facts are critical, but the second is nearly unknown outside the academic world. It’s not a new fact of academic life, arising from a recent economic downturn in the publishing industry. Nor is it a case of corporate exploitation of unworldly academics. Scholarly journals haven’t paid authors for their articles since the first scholarly journals, the Philosophical Transactions of the Royal Society of London and the Journal des sçavans , launched in London and Paris in 1665. The academic custom to write research articles for impact rather than money may be a lucky accident that could have been otherwise. Or it may be a wise adaptation that would eventually evolve in any culture with a serious research subculture. (The optimist in me wants to believe the latter, but the evolution of copyright law taunts that optimism.) This peculiar custom does more than insulate cutting-edge research from the market and free scholars to consent to OA without losing revenue. It also supports academic freedom and the kinds of serious inquiry that advance knowledge. It frees researchers to challenge conventional wisdom and defend unpopular ideas, which are essential to academic freedom. At the same time it frees them to microspecialize and defend ideas of immediate interest to just a handful people in the world, which are essential to pushing the frontiers of knowledge. This custom doesn’t guarantee that truth-seeking won’t be derailed by profit-seeking, and it doesn’t guarantee that we’ll eventually fill the smallest gaps in our collaborative understanding of the world. It doesn’t even guarantee that scholars won’t sometimes play for the crowd and detour into fad thinking. But it removes a major distraction by allowing them, if they wish, to focus on what is likely to be true rather than what is likely to sell. It’s a payment structure we need for good research itself, not just for good access to research, and it’s the key to the legal and economic lock that would otherwise shackle steps toward OA. Creative people who live by royalties, such as novelists, musicians, and moviemakers, may consider this scholarly tradition a burden and sacrifice for scholars. We might even agree, provided we don’t overlook a few facts. First, it’s a sacrifice that scholars have been making for nearly 350 years. OA to research articles doesn’t depend on asking royalty-earning authors to give up their royalties. Second, academics have salaries from universities, freeing them to dive deeply into their research topics and publish specialized articles without market appeal. Many musicians and moviemakers might envy that freedom to disregard sales and popular taste. Third, academics receive other, less tangible rewards from their institutions—like promotion and tenure—when their research is recognized by others, accepted, cited, applied, and built upon. It’s no accident that faculty who advance knowledge in their fields also advance their careers. Academics are passionate about certain topics, ideas, questions, inquiries, or disciplines. They feel lucky to have jobs in which they may pursue these passions and even luckier to be rewarded for pursuing them. Some focus single-mindedly on carrying an honest pebble to the pile of knowledge (as John Lange put it), having an impact on their field, or scooping others working on the same questions. Others focus strategically on building the case for promotion and tenure. But the two paths converge, which is not a fortuitous fact of nature but an engineered fact of life in the academy. As incentives for productivity, these intangible career benefits may be stronger for the average researcher than royalties are for the average novelist or musician. (In both domains, bountiful royalties for superstars tell us nothing about effective payment models for the long tail of less stellar professionals.) There’s no sense in which research would be more free, efficient, or effective if academics took a more “businesslike” position, behaved more like musicians and moviemakers, abandoned their insulation from the market, and tied their income to the popularity of their ideas. Nonacademics who urge academics to come to their senses and demand royalties even for journal articles may be more naive about nonprofit research than academics are about for-profit business. We can take this a step further. Scholars can afford to ignore sales because they have salaries and research grants to take the place of royalties. But why do universities pay salaries and why do funding agencies award grants? They do it to advance research and the range of public interests served by research. They don’t do it to earn profits from the results. They are all nonprofit. They certainly don’t do it to make scholarly writings into gifts to enrich publishers, especially when conventional publishers erect access barriers at the expense of research. Universities and funding agencies pay researchers to make their research into gifts to the public in the widest sense. Public and private funding agencies are essentially public and private charities, funding research they regard as useful or beneficial. Universities have a public purpose as well, even when they are private institutions. We support the public institutions with public funds, and we support the private ones with tax exemptions for their property and tax deductions for their donors. We’d have less knowledge, less academic freedom, and less OA if researchers worked for royalties and made their research articles into commodities rather than gifts. It should be no surprise, then, that more and more funding agencies and universities are adopting strong OA policies. Their mission to advance research leads them directly to logic of OA: With a few exceptions, such as classified research, research that is worth funding or facilitating is worth sharing with everyone who can make use of it. (See chapter 4 on OA policies.) Newcomers to OA often assume that OA helps readers and hurts authors, and that the reader side of the scholarly soul must beg the author side to make the necessary sacrifice. But OA benefits authors as well as readers. Authors want access to readers at least as much as readers want access to authors. All authors want to cultivate a larger audience and greater impact. Authors who work for royalties have reason to compromise and settle for the smaller audience of paying customers. But authors who aren’t paid for their writing have no reason to compromise. It takes nothing away from a disinterested desire to advance knowledge to recognize that scholarly publication is accompanied by a strong interest in impact and career building. The result is a mix of interested and disinterested motives. The reasons to make work OA are essentially the same as the reasons to publish. Authors who make their work OA are always serving others but not always acting from altruism. In fact, the idea that OA depends on author altruism slows down OA progress by hiding the role of author self-interest. Another aspect of author self-interest emerges from the well-documented phenomenon that OA articles are cited more often than non-OA articles, even when they are published in the same issue of the same journal. There’s growing evidence that OA articles are downloaded more often as well, and that journals converting to OA see a rise in their submissions and citation impact. There are many hypotheses to explain the correlation between OA and increased citations, but it’s likely that ongoing studies will show that much of the correlation is simply due to the larger audience and heightened visibility provided by OA itself. When you enlarge the audience for an article, you also enlarge the subset of the audience that will later cite it, including professionals in the same field at institutions unable to afford subscription access. OA enlarges the potential audience, including the potential professional audience, far beyond that for even the most prestigious and popular subscription journals. In any case, these studies bring a welcome note of author self-interest to the case for OA. OA is not a sacrifice for authors who write for impact rather than money. It increases a work’s visibility, retrievability, audience, usage, and citations, which all convert to career building. For publishing scholars, it would be a bargain even if it were costly, difficult, and time-consuming. But as we’ll see, it’s not costly, not difficult, and not time-consuming. My colleague Stevan Harnad frequently compares research articles to advertisements. They advertise the author’s research. Try telling advertisers that they’re making a needless sacrifice by allowing people to read their ads without having to pay for the privilege. Advertisers give away their ads and even pay to place them where they might be seen. They do this to benefit themselves, and scholars have the same interest in sharing their message as widely as possible. Because any content can be digital, and any digital content can be OA, OA needn’t be limited to royalty-free literature like research articles. Research articles are just ripe examples of low-hanging fruit. OA could extend to royalty-producing work like monographs, textbooks, novels, news, music, and movies. But as soon as we cross the line into OA for royalty-producing work, authors will either lose revenue or fear that they will lose revenue. Either way, they’ll be harder to persuade. But instead of concluding that royalty-producing work is off limits to OA, we should merely conclude that it’s higher-hanging fruit. In many cases we can still persuade royalty-earning authors to consent to OA. (See section 5.3 on OA for books.) Authors of scholarly research articles aren’t the only players who work without pay in the production of research literature. In general, scholarly journals don’t pay editors or referees either. In general, editors and referees are paid salaries by universities to free them, like authors, to donate their time and labor to ensure the quality of new work appearing in scholarly journals. An important consequence follows. All the key players in peer review can consent to OA without losing revenue. OA needn’t dispense with peer review or favor unrefereed manuscripts over refereed articles. We can aim for the prize of OA to peer-reviewed scholarship. (See section 5.1 on peer review.) Of course, conventional publishers are not as free as authors, editors, and referees to forgo revenue. This is a central fact in the transition to OA, and it explains why the interests of scholars and conventional publishers diverge more in the digital age than they diverged earlier. But not all publishers are conventional, and not all conventional publishers will carry print-era business models into the digital age. Academic publishers are not monolithic. Some new ones were born OA and some older ones have completely converted to OA. Many provide OA to some of their work but not all of it. Some are experimenting with OA, and some are watching the experiments of others. Most allow green OA (through repositories) and a growing number offer at least some kind of gold OA (through journals). Some are supportive, some undecided, some opposed. Among the opposed, some have merely decided not to provide OA themselves, while others lobby actively against policies to encourage or require OA. Some oppose gold but not green OA, while others oppose green but not gold OA. OA gains nothing and loses potential allies by blurring these distinctions. This variety reminds us (to paraphrase Tim O’Reilly) that OA doesn’t threaten publishing; it only threatens existing publishers who do not adapt. A growing number of journal publishers have chosen business models allowing them to dispense with subscription revenue and offer OA. They have expenses but they also have revenue to cover their expenses. In fact, some OA publishers are for-profit and profitable. (See chapter 7 on economics.) Moreover, peer review is done by dedicated volunteers who don’t care how a journal pays its bills, or even whether the journal is in the red or the black. If all peer-reviewed journals converted to OA overnight, the authors, editors, and referees would have the same incentives to participate in peer review that they had the day before. They needn’t stop offering their services, needn’t lower their standards, and needn’t make sacrifices they weren’t already making. They volunteer their time not because of a journal’s choice of business model but because of its contribution to research. They could carry on with solvent or insolvent subscription publishers, with solvent or insolvent OA publishers, or even without publishers. The Budapest Open Access Initiative said in February 2002: “An old tradition and a new technology have converged to make possible an unprecedented public good. The old tradition is the willingness of scientists and scholars to publish the fruits of their research in scholarly journals without payment. . . . The new technology is the internet.” To see what this willingness looks like without the medium to give it effect, look at scholarship in the age of print. Author gifts turned into publisher commodities, and access gaps for readers were harmfully large and widespread. (Access gaps are still harmfully large and widespread, but only because OA is not yet the default for new research.) To see what the medium looks like without the willingness, look at music and movies in the age of the internet. The need for royalties keeps creators from reaching everyone who would enjoy their work. A beautiful opportunity exists where the willingness and the medium overlap. A scholarly custom that evolved in the seventeenth century frees scholars to take advantage of the access revolution in the twentieth and twenty-first. Because scholars are nearly unique in following this custom, they are nearly unique in their freedom to take advantage of this revolution without financial risk. In this sense, the planets have aligned for scholars. Most other authors are constrained to fear rather than seize the opportunities created by the internet. 1.2 What OA Is Not We can dispel a cloud of objections and misunderstandings simply by pointing out a few things that OA is not. (Many of these points will be elaborated in later chapters.) OA isn’t an attempt to bypass peer review. OA is compatible with every kind of peer review, from the most conservative to the most innovative, and all the major public statements on OA insist on its importance. Because scholarly journals generally don’t pay peer-reviewing editors and referees, just as they don’t pay authors, all the participants in peer review can consent to OA without losing revenue. While OA to unrefereed preprints is useful and widespread, the OA movement isn’t limited to unrefereed preprints and, if anything, focuses on OA to peer-reviewed articles. (More in section 5.1 on peer review.) OA isn’t an attempt to reform, violate, or abolish copyright. It’s compatible with copyright law as it is. OA would benefit from the right kinds of copyright reforms, and many dedicated people are working on them. But it needn’t wait for reforms and hasn’t waited. OA literature avoids copyright problems in exactly the same way that conventional toll-access literature does. For older works, it takes advantage of the public domain, and for newer works, it rests on copyright-holder consent. (More in chapter 4 on policies and chapter 6 on copyright.) OA isn’t an attempt to deprive royalty-earning authors of income. The OA movement focuses on research articles precisely because they don’t pay royalties. In any case, inside and outside that focus, OA for copyrighted work depends on copyright-holder consent. Hence, royalty-earning authors have nothing to fear but persuasion that the benefits of OA might outweigh the risks to royalties. (More in section 5.3 on OA for books.) OA isn’t an attempt to deny the reality of costs. No serious OA advocate has ever argued that OA literature is costless to produce, although many argue that it is less expensive to produce than conventionally published literature, even less expensive than born-digital toll-access literature. The question is not whether research literature can be made costless, but whether there are better ways to pay the bills than charging readers and creating access barriers. (More in chapter 7 on economics.) Terminology We could talk about vigilante OA, infringing OA, piratical OA, or OA without consent. That sort of OA could violate copyrights and deprive royalty-earning authors of royalties against their will. But we could also talk about vigilante publishing, infringing publishing, piratical publishing, or publishing without consent. Both happen. However, we generally reserve the term “publishing” for lawful publishing, and tack on special adjectives to describe unlawful variations on the theme. Likewise, I’ll reserve the term “open access” for lawful OA that carries the consent of the relevant rightsholder. OA isn’t an attempt to reduce authors’ rights over their work. On the contrary, OA depends on author decisions and requires authors to exercise more rights or control over their work than they are allowed to exercise under traditional publishing contracts. One OA strategy is for authors to retain some of the rights they formerly gave publishers, including the right to authorize OA. Another OA strategy is for publishers to permit more uses than they formerly permitted, including permission for authors to make OA copies of their work. By contrast, traditional journal-publishing contracts demand that authors transfer all rights to publishers, and author rights or control cannot sink lower than that. (See chapters 4 on policies and 6 on copyright.) OA isn’t an attempt to reduce academic freedom. Academic authors remain free to submit their work to the journals or publishers of their choice. Policies requiring OA do so conditionally, for example, for researchers who choose to apply for a certain kind of grant. In addition, these policies generally build in exceptions, waiver options, or both. Since 2008 most university OA policies have been adopted by faculty deeply concerned to preserve and even enhance their prerogatives. (See chapter 4 on OA policies.) OA isn’t an attempt to relax rules against plagiarism. All the public definitions of OA support author attribution, even construed as a “restriction” on users. All the major open licenses require author attribution. Moreover, plagiarism is typically punished by the plagiarist’s institution rather than by courts, that is, by social norms rather than by law. Hence, even when attribution is not legally required, plagiarism is still a punishable offense and no OA policy anywhere interferes with those punishments. In any case, if making literature digital and online makes plagiarism easier to commit, then OA makes plagiarism easier to detect. Not all plagiarists are smart, but the smart ones will not steal from OA sources indexed in every search engine. In this sense, OA deters plagiarism. OA isn’t an attempt to punish or undermine conventional publishers. OA is an attempt to advance the interests of research, researchers, and research institutions. The goal is constructive, not destructive. If OA does eventually harm toll-access publishers, it will be in the way that personal computers harmed typewriter manufacturers. The harm was not the goal, but a side effect of developing something better. Moreover, OA doesn’t challenge publishers or publishing per se, just one business model for publishing, and it’s far easier for conventional publishers to adapt to OA than for typewriter manufacturers to adapt to computers. In fact, most toll-access publishers are already adapting, by allowing author-initiated OA, providing some OA themselves, or experimenting with OA. (See section 3.1 on green OA and chapter 8 on casualties.) OA doesn’t require boycotting any kind of literature or publisher. It doesn’t require boycotting toll-access research any more than free online journalism requires boycotting priced online journalism. OA doesn’t require us to strike toll-access literature from our personal reading lists, course syllabi, or libraries. Some scholars who support OA decide to submit new work only to OA journals, or to donate their time as editors or referees only to OA journals, in effect boycotting toll-access journals as authors, editors, and referees. But this choice is not forced by the definition of OA, by a commitment to OA, or by any OA policy, and most scholars who support OA continue to work with toll-access journals. In any case, even those scholars who do boycott toll-access journals as authors, editors, or referees don’t boycott them as readers. (Here we needn’t get into the complexity that some toll-access journals effectively create involuntary reader boycotts by pricing their journals out of reach of readers who want access.) OA isn’t primarily about bringing access to lay readers. If anything, the OA movement focuses on bringing access to professional researchers whose careers depend on access. But there’s no need to decide which users are primary and which are secondary. The publishing lobby sometimes argues that the primary beneficiaries of OA are lay readers, perhaps to avoid acknowledging how many professional researchers lack access, or perhaps to set up the patronizing counter-argument that lay people don’t care to read research literature and wouldn’t understand it if they tried. OA is about bringing access to everyone with an internet connection who wants access, regardless of their professions or purposes. There’s no doubt that if we put “professional researchers” and “everyone else” into separate categories, a higher percentage of researchers will want access to research literature, even after taking into account that many already have paid access through their institutions. But it’s far from clear why that would matter, especially when providing OA to all internet users is cheaper and simpler than providing OA to just a subset of worthy internet users. If party-goers in New York and New Jersey can both enjoy the Fourth of July fireworks in New York Harbor, then the sponsors needn’t decide that one group is primary, even if a simple study could show which group is more numerous. If this analogy breaks down, it’s because New Jersey residents who can’t see the fireworks gain nothing from New Yorkers who can. But research does offer this double or indirect benefit. When OA research directly benefits many lay readers, so much the better. But when it doesn’t, it still benefits everyone indirectly by benefiting researchers directly. (Also see section 5.5.1 on access for lay readers.) Finally, OA isn’t universal access. Even when we succeed at removing price and permission barriers, four other kinds of access barrier might remain in place: Filtering and censorship barriers Many schools, employers, ISPs, and governments want to limit what users can see. Language barriers Most online literature is in English, or another single language, and machine translation is still very weak. Handicap access barriers Most web sites are not yet as accessible to handicapped users as they should be. Connectivity barriers The digital divide keeps billions of people offline, including millions of scholars, and impedes millions of others with slow, flaky, or low-bandwidth internet connections. Most us want to remove all four of these barriers. But there’s no reason to save the term open access until we succeed. In the long climb to universal access, removing price and permission barriers is a significant plateau worth recognizing with a special name.
|
A. Chapter 4
|
What were the lights Lowry saw in the dark?
A. Svan and his conspirators
B. The guards
C. The delegation
D. Another spy-ray
|
DOUBLECROSS by JAMES Mac CREIGH Revolt was brewing on Venus, led by the descendant of the first Earthmen to land. Svan was the leader making the final plans—plotting them a bit too well. [Transcriber's Note: This etext was produced from Planet Stories Winter 1944. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The Officer of the Deck was pleased as he returned to the main lock. There was no reason why everything shouldn't have been functioning perfectly, of course, but he was pleased to have it confirmed, all the same. The Executive Officer was moodily smoking a cigarette in the open lock, staring out over the dank Venusian terrain at the native town. He turned. "Everything shipshape, I take it!" he commented. The OD nodded. "I'll have a blank log if this keeps up," he said. "Every man accounted for except the delegation, cargo stowed, drivers ready to lift as soon as they come back." The Exec tossed away his cigarette. " If they come back." "Is there any question?" The Exec shrugged. "I don't know, Lowry," he said. "This is a funny place. I don't trust the natives." Lowry lifted his eyebrows. "Oh? But after all, they're human beings, just like us—" "Not any more. Four or five generations ago they were. Lord, they don't even look human any more. Those white, flabby skins—I don't like them." "Acclimation," Lowry said scientifically. "They had to acclimate themselves to Venus's climate. They're friendly enough." The Exec shrugged again. He stared at the wooden shacks that were the outskirts of the native city, dimly visible through the ever-present Venusian mist. The native guard of honor, posted a hundred yards from the Earth-ship, stood stolidly at attention with their old-fashioned proton-rifles slung over their backs. A few natives were gazing wonderingly at the great ship, but made no move to pass the line of guards. "Of course," Lowry said suddenly, "there's a minority who are afraid of us. I was in town yesterday, and I talked with some of the natives. They think there will be hordes of immigrants from Earth, now that we know Venus is habitable. And there's some sort of a paltry underground group that is spreading the word that the immigrants will drive the native Venusians—the descendants of the first expedition, that is—right down into the mud. Well—" he laughed—"maybe they will. After all, the fittest survive. That's a basic law of—" The annunciator over the open lock clanged vigorously, and a metallic voice rasped: "Officer of the Deck! Post Number One! Instruments reports a spy ray focused on the main lock!" Lowry, interrupted in the middle of a word, jerked his head back and stared unbelievingly at the tell-tale next to the annunciator. Sure enough, it was glowing red—might have been glowing for minutes. He snatched at the hand-phone dangling from the wall, shouted into it. "Set up a screen! Notify the delegation! Alert a landing party!" But even while he was giving orders, the warning light flickered suddenly and went out. Stricken, Lowry turned to the Exec. The Executive Officer nodded gloomily. He said, "You see!" "You see?" Svan clicked off the listening-machine and turned around. The five others in the room looked apprehensive. "You see?" Svan repeated. "From their own mouths you have heard it. The Council was right." The younger of the two women sighed. She might have been beautiful, in spite of her dead-white skin, if there had been a scrap of hair on her head. "Svan, I'm afraid," she said. "Who are we to decide if this is a good thing? Our parents came from Earth. Perhaps there will be trouble at first, if colonists come, but we are of the same blood." Svan laughed harshly. " They don't think so. You heard them. We are not human any more. The officer said it." The other woman spoke unexpectedly. "The Council was right," she agreed. "Svan, what must we do?" Svan raised his hand, thoughtfully. "One moment. Ingra, do you still object?" The younger woman shrank back before the glare in his eyes. She looked around at the others, found them reluctant and uneasy, but visibly convinced by Svan. "No," she said slowly. "I do not object." "And the rest of us? Does any of us object?" Svan eyed them, each in turn. There was a slow but unanimous gesture of assent. "Good," said Svan. "Then we must act. The Council has told us that we alone will decide our course of action. We have agreed that, if the Earth-ship returns, it means disaster for Venus. Therefore, it must not return." An old man shifted restlessly. "But they are strong, Svan," he complained. "They have weapons. We cannot force them to stay." Svan nodded. "No. They will leave. But they will never get back to Earth." "Never get back to Earth?" the old man gasped. "Has the Council authorized—murder?" Svan shrugged. "The Council did not know what we would face. The Councilmen could not come to the city and see what strength the Earth-ship has." He paused dangerously. "Toller," he said, "do you object?" Like the girl, the old man retreated before his eyes. His voice was dull. "What is your plan?" he asked. Svan smiled, and it was like a dark flame. He reached to a box at his feet, held up a shiny metal globe. "One of us will plant this in the ship. It will be set by means of this dial—" he touched a spot on the surface of the globe with a pallid finger—"to do nothing for forty hours. Then—it will explode. Atomite." He grinned triumphantly, looking from face to face. The grin faded uncertainly as he saw what was in their eyes—uncertainty, irresolution. Abruptly he set the bomb down, savagely ripped six leaves off a writing tablet on the table next him. He took a pencil and made a mark on one of them, held it up. "We will let chance decide who is to do the work," he said angrily. "Is there anyone here who is afraid? There will be danger, I think...." No answer. Svan jerked his head. "Good," he said. "Ingra, bring me that bowl." Silently the girl picked up an opaque glass bowl from the broad arm of her chair. It had held Venus-tobacco cigarettes; there were a few left. She shook them out and handed the bowl to Svan, who was rapidly creasing the six fatal slips. He dropped them in the bowl, stirred it with his hand, offered it to the girl. "You first, Ingra," he said. She reached in mechanically, her eyes intent on his, took out a slip and held it without opening it. The bowl went the rounds, till Svan himself took the last. All eyes were on him. No one had looked at their slips. Svan, too, had left his unopened. He sat at the table, facing them. "This is the plan," he said. "We will go, all six of us, in my ground car, to look at the Earth-ship. No one will suspect—the whole city has been to see it already. One will get out, at the best point we can find. It is almost dusk now. He can hide, surely, in the vegetation. The other five will start back. Something will go wrong with the car—perhaps it will run off the road, start to sink in the swamp. The guards will be called. There will be commotion—that is easy enough, after all; a hysterical woman, a few screams, that's all there is to it. And the sixth person will have his chance to steal to the side of the ship. The bomb is magnetic. It will not be noticed in the dark—they will take off before sunrise, because they must travel away from the sun to return—in forty hours the danger is removed." There was comprehension in their eyes, Svan saw ... but still that uncertainty. Impatiently, he crackled: "Look at the slips!" Though he had willed his eyes away from it, his fingers had rebelled. Instinctively they had opened the slip, turned it over and over, striving to detect if it was the fatal one. They had felt nothing.... And his eyes saw nothing. The slip was blank. He gave it but a second's glance, then looked up to see who had won the lethal game of chance. Almost he was disappointed. Each of the others had looked in that same second. And each was looking up now, around at his neighbors. Svan waited impatiently for the chosen one to announce it—a second, ten seconds.... Then gray understanding came to him. A traitor! his subconscious whispered. A coward! He stared at them in a new light, saw their indecision magnified, became opposition. Svan thought faster than ever before in his life. If there was a coward, it would do no good to unmask him. All were wavering, any might be the one who had drawn the fatal slip. He could insist on inspecting every one, but—suppose the coward, cornered, fought back? In fractions of a second, Svan had considered the evidence and reached his decision. Masked by the table, his hand, still holding the pencil, moved swiftly beneath the table, marked his own slip. In the palm of his hand, Svan held up the slip he had just marked in secret. His voice was very tired as he said, "I will plant the bomb." The six conspirators in Svan's old ground car moved slowly along the main street of the native town. Two Earth-ship sailors, unarmed except for deceptively flimsy-looking pistols at their hips, stood before the entrance to the town's Hall of Justice. "Good," said Svan, observing them. "The delegation is still here. We have ample time." He half turned in the broad front seat next to the driver, searching the faces of the others in the car. Which was the coward? he wondered. Ingra? Her aunt? One of the men? The right answer leaped up at him. They all are , he thought. Not one of them understands what this means. They're afraid. He clamped his lips. "Go faster, Ingra," he ordered the girl who was driving. "Let's get this done with." She looked at him, and he was surprised to find compassion in her eyes. Silently she nodded, advanced the fuel-handle so that the clumsy car jolted a trace more rapidly over the corduroy road. It was quite dark now. The car's driving light flared yellowishly in front of them, illuminating the narrow road and the pale, distorted vegetation of the jungle that surrounded them. Svan noticed it was raining a little. The present shower would deepen and intensify until midnight, then fall off again, to halt before morning. But before then they would be done. A proton-bolt lanced across the road in front of them. In the silence that followed its thunderous crash, a man's voice bellowed: "Halt!" The girl, Ingra, gasped something indistinguishable, slammed on the brakes. A Venusian in the trappings of the State Guard advanced on them from the side of the road, proton-rifle held ready to fire again. "Where are you going?" he growled. Svan spoke up. "We want to look at the Earth-ship," he said. He opened the door beside him and stepped out, careless of the drizzle. "We heard it was leaving tonight," he continued, "and we have not seen it. Is that not permitted?" The guard shook his head sourly. "No one is allowed near the ship. The order was just issued. It is thought there is danger." Svan stepped closer, his teeth bared in what passed for a smile. "It is urgent," he purred. His right hand flashed across his chest in a complicated gesture. "Do you understand?" Confusion furrowed the guard's hairless brows, then was replaced by a sudden flare of understanding—and fear. "The Council!" he roared. "By heaven, yes, I understand! You are the swine that caused this—" He strove instinctively to bring the clumsy rifle up, but Svan was faster. His gamble had failed; there was only one course remaining. He hurled his gross white bulk at the guard, bowled him over against the splintery logs of the road. The proton-rifle went flying, and Svan savagely tore at the throat of the guard. Knees, elbows and claw-like nails—Svan battered at the astonished man with every ounce of strength in his body. The guard was as big as Svan, but Svan had the initial advantage ... and it was only a matter of seconds before the guard lay unconscious, his skull a mass of gore at the back where Svan had ruthlessly pounded it against the road. Svan grunted as his fingers constricted brutally. Svan rose, panting, stared around. No one else was in sight, save the petrified five and the ground car. Svan glared at them contemptuously, then reached down and heaved on the senseless body of the guard. Over the shoulder of the road the body went, onto the damp swampland of the jungle. Even while Svan watched the body began to sink. There would be no trace. Svan strode back to the car. "Hurry up," he gasped to the girl. "Now there is danger for all of us, if they discover he is missing. And keep a watch for other guards." Venus has no moon, and no star can shine through its vast cloud layer. Ensign Lowry, staring anxiously out through the astro-dome in the bow of the Earth-ship, cursed the blackness. "Can't see a thing," he complained to the Exec, steadily writing away at the computer's table. "Look—are those lights over there?" The Exec looked up wearily. He shrugged. "Probably the guards. Of course, you can't tell. Might be a raiding party." Lowry, stung, looked to see if the Exec was smiling, but found no answer in his stolid face. "Don't joke about it," he said. "Suppose something happens to the delegation?" "Then we're in the soup," the Exec said philosophically. "I told you the natives were dangerous. Spy-rays! They've been prohibited for the last three hundred years." "It isn't all the natives," Lowry said. "Look how they've doubled the guard around us. The administration is co-operating every way they know how. You heard the delegation's report on the intercom. It's this secret group they call the Council." "And how do you know the guards themselves don't belong to it?" the Exec retorted. "They're all the same to me.... Look, your light's gone out now. Must have been the guard. They're on the wrong side to be coming from the town, anyhow...." Svan hesitated only a fraction of a second after the girl turned the lights out and stopped the car. Then he reached in the compartment under the seat. If he took a little longer than seemed necessary to get the atomite bomb out of the compartment, none of the others noticed. Certainly it did not occur to them that there had been two bombs in the compartment, though Svan's hand emerged with only one. He got out of the car, holding the sphere. "This will do for me," he said. "They won't be expecting anyone to come from behind the ship—we were wise to circle around. Now, you know what you must do?" Ingra nodded, while the others remained mute. "We must circle back again," she parroted. "We are to wait five minutes, then drive the car into the swamp. We will create a commotion, attract the guards." Svan, listening, thought: It's not much of a plan. The guards would not be drawn away. I am glad I can't trust these five any more. If they must be destroyed, it is good that their destruction will serve a purpose. Aloud, he said, "You understand. If I get through, I will return to the city on foot. No one will suspect anything if I am not caught, because the bomb will not explode until the ship is far out in space. Remember, you are in no danger from the guards." From the guards , his mind echoed. He smiled. At least, they would feel no pain, never know what happened. With the amount of atomite in that bomb in the compartment, they would merely be obliterated in a ground-shaking crash. Abruptly he swallowed, reminded of the bomb that was silently counting off the seconds. "Go ahead," he ordered. "I will wait here." "Svan." The girl, Ingra, leaned over to him. Impulsively she reached for him, kissed him. "Good luck to you, Svan," she said. "Good luck," repeated the others. Then silently the electric motor of the car took hold. Skilfully the girl backed it up, turned it around, sent it lumbering back down the road. Only after she had traveled a few hundred feet by the feel of the road did she turn the lights on again. Svan looked after them. The kiss had surprised him. What did it mean? Was it an error that the girl should die with the others? There was an instant of doubt in his steel-shackled mind, then it was driven away. Perhaps she was loyal, yet certainly she was weak. And since he could not know which was the one who had received the marked slip, and feared to admit it, it was better they all should die. He advanced along the midnight road to where the ground rose and the jungle plants thinned out. Ahead, on an elevation, were the rain-dimmed lights of the Earth-ship, set down in the center of a clearing made by its own fierce rockets. Svan's mist-trained eyes spotted the circling figures of sentries, and knew that these would be the ship's own. They would not be as easily overcome as the natives, not with those slim-shafted blasters they carried. Only deceit could get him to the side of the ship. Svan settled himself at the side of the road, waiting for his chance. He had perhaps three minutes to wait; he reckoned. His fingers went absently to the pouch in his wide belt, closed on the slip of paper. He turned it over without looking at it, wondering who had drawn the first cross, and been a coward. Ingra? One of the men? He became abruptly conscious of a commotion behind him. A ground car was racing along the road. He spun around and was caught in the glare of its blinding driving-light, as it bumped to a slithering stop. Paralyzed, he heard the girl's voice. "Svan! They're coming! They found the guard's rifle, and they're looking for us! Thirty Earthmen, Svan, with those frightful guns. They fired at us, but we got away and came for you. We must flee!" He stared unseeingly at the light. "Go away!" he croaked unbelievingly. Then his muscles jerked into action. The time was almost up—the bomb in the car— "Go away!" he shrieked, and turned to run. His fists clenched and swinging at his side, he made a dozen floundering steps before something immense pounded at him from behind. He felt himself lifted from the road, sailing, swooping, dropping with annihilating force onto the hard, charred earth of the clearing. Only then did he hear the sound of the explosion, and as the immense echoes died away he began to feel the pain seeping into him from his hideously racked body.... The Flight Surgeon rose from beside him. "He's still alive," he said callously to Lowry, who had just come up. "It won't last long, though. What've you got there?" Lowry, a bewildered expression on his beardless face, held out the two halves of a metallic sphere. Dangling ends of wires showed where a connection had been broken. "He had a bomb," he said. "A magnetic-type, delayed-action atomite bomb. There must have been another in the car, and it went off. They—they were planning to bomb us." "Amazing," the surgeon said dryly. "Well, they won't do any bombing now." Lowry was staring at the huddled, mutilated form of Svan. He shuddered. The surgeon, seeing the shudder, grasped his shoulder. "Better them than us," he said. "It's poetic justice if I ever saw it. They had it coming...." He paused thoughtfully, staring at a piece of paper between his fingers. "This is the only part I don't get," he said. "What's that?" Lowry craned his neck. "A piece of paper with a cross on it? What about it?" The surgeon shrugged. "He had it clenched in his hand," he said. "Had the devil of a time getting it loose from him." He turned it over slowly, displayed the other side. "Now what in the world would he be doing carrying a scrap of paper with a cross marked on both sides?"
|
A. Svan and his conspirators
|
What is Myles Cabot’s relationship to the narrator, Mr. Farley? Evidence of this?
A. They are both radio engineers, and presumably bothers. Cabot built a radio set and natter-transmitting device on Farley’s rooftop.
B. They met on Venus and became fast friends. Cabot helped Farley to plan a coup to usurp the arch-fiend Yuri, King of both Formia and Cupia.
C. They are both radio engineers, and presumably friends. Farley allowed Cabot to built a radio set and natter-transmitting device on his farm.
D. They met on Venus and became fast friends. Farley allowed Cabot to built a radio set and natter-transmitting device on his farm.
|
THE RADIO PLANET Ralph Milne Farley I “It’s too bad that Myles Cabot can’t see this!” I exclaimed, as my eye fell on the following item: SIGNALS FROM MARS FAIL TO REACH HARVARD Cambridge, Massachusetts, Wednesday. The Harvard College Radio Station has for several weeks been in receipt of fragmentary signals of extraordinarily long wave-length, Professor Hammond announced yesterday. So far as it has been possible to test the direction of the source of these waves, it appears that the direction has a twenty-four hour cycle, thus indicating that the origin of these waves is some point outside the earth. The university authorities will express no opinion as to whether or not these messages come from Mars. Myles, alone of all the radio engineers of my acquaintance, was competent to surmount these difficulties, and thus enable the Cambridge savants to receive with clearness the message from another planet. 6 Twelve months ago he would have been available, for he was then quietly visiting at my farm, after five earth-years spent on the planet Venus, where, by the aid of radio, he had led the Cupians to victory over their oppressors, a human-brained race of gigantic black ants. He had driven the last ant from the face of continental Poros, and had won and wed the Princess Lilla, who had borne him a son to occupy the throne of Cupia. While at my farm Cabot had rigged up a huge radio set and a matter-transmitting apparatus, with which he had (presumably) shot himself back to Poros on the night of the big October storm which had wrecked his installation. I showed the newspaper item to Mrs. Farley, and lamented on Cabot’s absence. Her response opened up an entirely new line of thought. Said she: “Doesn’t the very fact that Mr. Cabot isn’t here suggest to you that this may be a message, not from Mars, but from him? Or perhaps from the Princess Lilla, inquiring about him in case he has failed in his attempted return?” That had never occurred to me! How stupid! “What had I better do about it, if anything?” I asked. “Drop Professor Hammond a line?” But Mrs. Farley was afraid that I would be taken for a crank. That evening, when I was over in town, the clerk in the drug store waylaid me to say that there had been a long-distance phone call for me, and would I please call a certain Cambridge number. So, after waiting an interminable time in the stuffy booth with my hands full of dimes, nickels, and quarters, I finally got my party. “Mr. Farley?” “Speaking.” “This is Professor Kellogg, O. D. Kellogg,” the voice replied. 7 It was my friend of the Harvard math faculty, the man who had analyzed the measurements of the streamline projectile in which Myles Cabot had shot to earth the account of the first part of his adventures on Venus. Some further adventures Myles had told me in person during his stay on my farm. “Professor Hammond thinks that he is getting Mars on the air,” the voice continued. “Yes,” I replied. “I judged as much from what I read in this morning’s paper. But what do you think?” Kellogg’s reply gave my sluggish mind the second jolt which it had received that day. “Well,” he said, “in view of the fact that I am one of the few people among your readers who take your radio stories seriously, I think that Hammond is getting Venus. Can you run up here and help me try and convince him?” And so it was that I took the early boat next morning for Boston, and had lunch with the two professors. As a result of our conference, a small committee of engineers returned with me to Edgartown that evening for the purpose of trying to repair the wrecked radio set which Myles Cabot had left on my farm. They utterly failed to comprehend the matter-transmitting apparatus, and so—after the fallen tower had been reerected and the rubbish cleared away—they had devoted their attention to the restoration of the conversational part of the set. To make a long story short, we finally restored it, with the aid of some old blue prints of Cabot’s which Mrs. Farley, like Swiss Family Robinson’s wife, produced from somewhere. I was the first to try the earphones, and was rewarded by a faint “bzt-bzt” like the song of a north woods blackfly. In conventional radioese, I repeated the sounds to the Harvard group: “Dah-dit-dah-dit dah-dah-dit-dah. Dah-dit-dah-dit dah-dah-dit-dah. Dah-dit-dah-dit dah-dah-dit-dah. Dah-dit-dit dit. Dah-dit-dah-dit dit-dah dah-dit dit dit dah-dah-dah dah. Dah-dit-dah-dit dit-dah dah-dit-dit-dit dah-dah-dah dah. Dah-dit-dah-dit dit-dah dah-dit-dit-dit-dah dah-dah-dah.” 8 A look of incredulity spread over their faces. Again came the same message, and again I repeated it. “You’re spoofing us!” one of them shouted. “Give me the earphones.” And he snatched them from my head. Adjusting them on his own head, he spelled out to us, “C-Q C-Q C-Q D-E C-A-B-O-T C-A-B-O-T C-A-B-O-T—” Seizing the big leaf-switch, he threw it over. The motor-generator began to hum. Grasping the key, the Harvard engineer ticked off into space: “Cabot Cabot Cabot D-E—” “Has this station a call letter?” he hurriedly asked me. “Yes,” I answered quickly, “One-X-X-B.” “One-X-X-B,” he continued the ticking “K.” Interplanetary communication was an established fact at last! And not with Mars after all these years of scientific speculations. But what meant more to me was that I was again in touch with my classmate Myles Standish Cabot, the radio man. The next day a party of prominent scientists, accompanied by a telegrapher and two stenographers, arrived at my farm. During the weeks that followed there was recorded Myles’s own account of the amazing adventures on the planet Venus (or Poros, as its own inhabitants call it,) which befell him upon his return there after his brief visit to the earth. I have edited those notes into the following coherent story. II TOO MUCH STATIC Myles Cabot had returned to the earth to study the latest developments of modern terrestrial science for the benefit of the Cupian nation. He was the regent of Cupia during the minority of his baby son, King Kew the Thirteenth. The loyal Prince Toron occupied the throne in his absence. The last of the ant-men and their ally, the renegade Cupian Prince Yuri, had presumably perished in an attempt to escape by flying through the steam-clouds which completely hem in continental Poros. What lay beyond the boiling seas no man knew. 9 During his stay on my farm, Cabot had built the matter-transmitting apparatus, with which he had shot himself off into space on that October night on which he had received the message from the skies: “S O S, Lilla.” A thunderstorm had been brewing all that evening, and just as Myles had placed himself between the coordinate axes of his machine and had gathered up the strings which ran from his control levers to within the apparatus, there had come a blinding flash. Lightning had struck his aerial. How long his unconsciousness lasted he knew not. He was some time in regaining his senses. But when he had finally and fully recovered, he found himself lying on a sandy beach beside a calm and placid lake beneath a silver sky. He fell to wondering, vaguely and pleasantly, where he was and how he had got here. Suddenly, however, his ears were jarred by a familiar sound. At once his senses cleared, and he listened intently to the distant purring of a motor. Yes, there could be no mistake; an airplane was approaching. Now he could see it, a speck in the sky, far down the beach. Nearer and nearer it came. Myles sprang to his feet. To his intense surprise, he found that the effort threw him quite a distance into the air. Instantly the idea flashed through his mind: “I must be on Mars! Or some other strange planet.” This idea was vaguely reminiscent of something. But while he was trying to catch this vaguely elusive train of thought, his attention was diverted by the fact that, for some unaccountable reason, his belt buckle and most of the buttons which had held his clothes together were missing, so that his clothing came to pieces as he rose, and that he had to shed it rapidly in order to avoid impeding his movements. He wondered at the cause of this. 10 But his speculations were cut short by the alighting of the plane a hundred yards down the beach. What was his horror when out of it clambered, not men but ants! Ants, six-footed, and six feet high. Huge ants, four of them, running toward him over the glistening sands. Gone was all his languor, as he seized a piece of driftwood and prepared to defend himself. As he stood thus expectant, Myles realized that his present position and condition, the surrounding scenery, and the advance of the ant-men were exactly, item for item, like the opening events of his first arrival on the planet Poros. He even recognized one of the ant-men as old Doggo, who had befriended him on his previous visit. Could it be that all his adventures in Cupia had been naught but a dream; a recurring dream, in fact? Were his dear wife Lilla and his little son Kew merely figments of his imagination? Horrible thought! And then events began to differ from those of the past; for the three other Formians halted, and Doggo advanced alone. By the agitation of the beast’s antennae the earth man could see that it was talking to him. But Myles no longer possessed the wonderful electrical headset which he had contrived and built during his previous visit to that planet, so as to talk with Cupians and Formians, both of which races are earless and converse by means of radiations from their antennae. So he picked up two sticks from the beach, and held them projecting from his forehead; then threw them to the ground with a grimace of disgust and pointed to his ears. Doggo understood, and scratched with his paw in Cupian shorthand on the silver sands the message: “Myles Cabot, you are our prisoner.” “What, again?” scratched Myles, then made a sign of submission. 11 He dreaded the paralyzing bite which Formians usually administer to their victims, and which he had twice experienced in the past; but, fortunately, it was not now forthcoming. The other three ants kept away from him as Doggo led him to the beached airplane, and soon they were scudding along beneath silver skies, northward as it later turned out. Far below them were silver-green fields and tangled tropical woods, interspersed with rivulets and little ponds. This was Cupia, his Cupia. He was home once more, back again upon the planet which held all that was dear to him in two worlds. His heart glowed with the warmth of homecoming. What mattered it that he was now a prisoner, in the hands (or, rather, claws) of his old enemies, the Formians? He had been their prisoner before, and had escaped. Once more he could escape, and rescue the Princess Lilla. Poor girl! How eager he was to reach her side, and save her from that peril, whatever it was, which had caused her to flash that “S O S” a hundred million miles across the solar system from Poros to the earth. He wondered what could have happened in Cupia since his departure, only a few sangths ago. How was it that the ant-men had survived their airplane journey across the boiling seas? What had led them to return? Or perhaps these ants were a group who had hidden somewhere and thus had escaped the general extermination of their race. In either event, how had they been able to reconquer Cupia? And where was their former leader, Yuri, the renegade Cupian prince? These and a hundred other similar questions flooded in upon the earth-man, as the Formian airship carried him, a captive, through the skies. He gazed again at the scene below, and now noted one difference from the accustomed Porovian landscape, for nowhere ran the smooth concrete roads which bear the swift two-wheeled kerkools of the Cupians to all parts of their continent. What uninhabited portion of Cupia could this be, over which they were now passing? 12 Turning to Doggo, Myles extended his left palm, and made a motion as though writing on it with the thumb and forefinger of his right hand. But the ant-man waved a negative with one of his forepaws. It was evident that there were no writing materials aboard the ship. Myles would have to wait until they reached their landing place; for doubtless they would soon hover down in some city or town, though just which one he could not guess, as the country below was wholly unfamiliar. Finally a small settlement loomed ahead. It was of the familiar style of toy-building-block architecture affected by the ant-men, and, from its appearance, was very new. On its outskirts further building operations were actively in progress. Apparently a few survivors of the accursed race of Formians were consolidating their position and attempting to build up a new empire in some out-of-the-way portion of the continent. As the earth-man was turning these thoughts over in his mind the plane softly settled down upon one of the flat roofs, and its occupants disembarked. Three of the ants advanced menacingly toward Myles, but Doggo held them off. Then all of the party descended down one of the ramps to the lower levels of the building. Narrow slitlike window openings gave onto courtyards, where fountains played and masses of blue and yellow flowers bloomed, amid gray-branched lichens with red and purple twig-knobs. It was in just such a garden, through just such a window, that he had first looked upon the lovely blue-eyed, golden-haired Lilla, Crown Princess of Cupia. The earth-man sighed. Where was his beloved wife now? That she needed his help was certain. He must therefore get busy. So once again he made motions of writing on the palm of his left hand with the thumb and forefinger of his right; and this time the sign language produced results, for Doggo halted the procession and led Cabot into a room. 13 It was a plain bare room, devoid of any furniture except a small table, for ant-men have no use for chairs and couches. The sky outside was already beginning to pinken with the unseen sun. With a sweep of his paw, Doggo indicated that this was to be Cabot’s quarters. Then, with another wave, he pointed to the table, where lay a pad of paper and stylus, not a pencil-like stylus as employed by the Cupians, but rather one equipped with straps for attaching it to the claw of a Formian. Even so, it was better than nothing. The earth-man seized it eagerly, but before he could begin writing an ant entered bearing a Cupian toga, short-sleeved and bordered with Grecian wave designs in blue. Myles put on this garment, and then quickly filled a sheet with questions: “How is my princess and my son, the baby king? Whence come all you Formians, whose race I thought had been exterminated? What part of Cupia is this? What is this city? Where is Prince Yuri? And what do you intend to do with me this time?” Then he passed the paper and stylus over to his old friend Doggo. They were alone together at last. The ant-man’s reply consumed sheet after sheet of paper; but, owning to the rapidity of Porovian shorthand, did not take so very much more time than speaking would have required. As he completed each sheet he passed it over to Myles, who read as follows: “As to your princess and your son, I know not, for this is not Cupia. Do you remember how, when your victorious army and air navy swept to the southern extremity of what had been Formia, a few of our survivors rose in planes from the ruins of our last stronghold and braved the dangers of the steam clouds which overhang the boiling seas? Our leader was Prince Yuri, erstwhile contender for the throne of Cupia, splendid even in defeat. “It was his brain that conceived our daring plan of escape. If there were other lands beyond the boiling seas, the lands which tradition taught were the origin of the Cupian race, then there we might prosper and raise up a new empire. At the worst we should merely meet death in another form, rather than at your hands. So we essayed. 14 “Your planes followed us, but turned back as we neared the area of terrific heat. Soon the vapor closed over us, blotting our enemies and our native land from view.” For page after page Doggo, the ant-man, related the harrowing details of that perilous flight across the boiling seas, ending with the words: “Here we are, and here are you, in Yuriana, capitol of New Formia. But how is it that you, Myles Cabot, have arrived here on this continent in exactly the same manner and condition in which I discovered you in old Formia eight years ago?” When Myles reached the end of reading this narrative, he in turn took the pad and stylus and related how he had gone to the planet Minos (which we call the Earth) to learn the latest discoveries and inventions there, and how his calculations for his return to Poros had been upset by some static conditions just as he had been about to transmit himself back. Oh, if only he had landed by chance upon the same beach as on his first journey through the skies! Wisely he refrained from mentioning the “S O S” message from Lilla. But his recollection of her predicament spurred him to be anxious about her rescue. His immediate problem was to learn what the ant-men planned for him; so the concluding words which he wrote upon the pad were: “And, now that you have me in your power, what shall you do with me?” “Old friend,” Doggo wrote in reply, “that depends entirely upon Yuri, our king, whose toga you now have on.” III YURI OR FORMIS? The earth-man grimaced, but then smiled. Perhaps, his succeeding to the toga of King Yuri might prove to be an omen. 15 “So Yuri is king of the ants?” he asked. “Yes,” his captor replied, “for Queen Formis did not survive the trip across the boiling seas.” “Then what of your empire?” Myles inquired. “No queen. No eggs. How can your race continue? For you Formians are like the ants on my own planet Minos.” Doggo’s reply astounded him. “Do you remember back at Wautoosa, I told you that some of us lesser Formians had occasionally laid eggs? So now behold before you Doggo, Admiral of the Formian Air Navy, and mother of a new Queen Formis.” This was truly a surprise! All along Cabot had always regarded the Formians as mannish. And rightly so, for they performed in their own country the duties assigned to men among the Cupians. Furthermore, all Formians, save only the reigning Formis herself, were called by the Porovian pronoun, which corresponds to “he” in English. When Myles had somewhat recovered from his astonishment, he warmly congratulated his friend by patting him on the side of the head, as is the Porovian custom. “Doggo,” he wrote, “this ought to constitute you a person of some importance among the Formians.” “It ought to,” the ant-man replied, “but as a matter of fact, it merely intensifies Yuri’s mistrust and hatred of me. Now that I am mother of the queen, he fears that I may turn against him and establish Formis in his place as the head of an empire of the Formians, by the Formians, and for the Formians exclusively.” “Why don’t you?” Myles wrote. It seemed to him to be a bully good idea, and incidentally a solution of his own difficulties. But Doggo wrote in horror, “It would be treason!” Then tore up all the correspondence. It is difficult to inculcate the thought of independence in the mind of one reared in an autocracy. The earth-man, however, persisted. “How many of the council can you count on, if the interests of Yuri should clash with those of Formis?” 16 “Only one—myself.” And again Doggo tore up the correspondence. Myles tactfully changed the subject. “Where is the arch-fiend now?” he asked. “We know not,” the Formian wrote in reply. “Six days ago he left us in his airship and flew westward. When he failed to return, we sent out scout planes to search for him, and we have been hunting ever since. When we sighted you on the beach this morning we thought that you might be our lost leader, and that is why we landed and approached you.” At about this point the conversation was interrupted by a worker ant who brought food: roast alta and green aphid milk. With what relish did the earth-man plunge into the feast, his first taste of Porovian delicacies in many months. During the meal conversation lagged, owing to the difficulty of writing and eating at the same time. But now Myles Cabot seized his pad and stylus and wrote: “Have you ever known me to fail in any undertaking on the planet Poros?” “No,” the ant-man wrote in reply. “Have you ever known me to be untrue to a principle, a cause, or a friend?” “No,” Doggo replied. “Then,” Myles wrote, “let us make your daughter queen in fact as well as in name.” “It is treason,” Doggo wrote in reply, but this time he did not tear up the correspondence. “Treason?” Myles asked. If he had spoken the word, he would have spoken it with scorn and derision. “Treason? Is it treason to support your own queen? What has become of the national pride of the once great Formians? Look! I pledge myself to the cause of Formis, rightful Queen of Formia. Formis, daughter of Doggo! What say you?” This time, as he tore up the correspondence, Doggo signified an affirmative. And thus there resulted further correspondence. 17 “Doggo,” Myles wrote, “can you get to the antenna of the queen?” The ant-man indicated that he could. “If she has inherited any of your character,” Myles continued, “she will assert herself, if given half a chance.” So the Pitmanesque conversation continued. Long since had the pink light of Porovian evening faded from the western sky. The ceiling vapor-lamps were lit. The night showed velvet-black through the slit-like windows. And still the two old friends wrote on, Myles Standish Cabot, the Bostonian, and Doggo, No. 334-2-18, the only really humanlike ant-man whom Myles had ever known among the once dominant race of Poros. Finally, as the dials indicated midnight, the two conspirators ceased their labors. All was arranged for the coup d’ etat . They tore into shreds every scrap of used paper, leaving extant merely the ant-man’s concluding words: “Meanwhile you are my prisoner.” Doggo then rang a soundless bell, which was answered by a worker ant, whom he inaudibly directed to bring sufficient draperies to form a bed for the earth-man. These brought, the two friends patted each other a fond good night, and the tired earth-man lay down for the first sleep which he had had in over forty earth hours. It hardly seemed possible! Night before last he had slept peacefully on a conventional feather-bed in a little New England farmhouse. Then had come the S O S message from the skies; and here he was now, millions of miles away through space retiring on matted silver felting on the concrete floor of a Porovian ant-house. Such are the mutations of fortune! With these thoughts the returned wanderer lapsed into a deep and dreamless sleep. When he awakened in the morning there was a guard posted at the door. 18 Doggo did not show up until nearly noon, when he rattled in, bristling with excitement. Seizing the pad he wrote: “A stormy session of the Council of Twelve! We are all agreed that you must be indicted for high crimes and misdemeanors. But the great question is as to just what we can charge you with.” “Sorry I can’t assist you,” the earth-man wrote. “How would it be if I were to slap your daughter’s face, or something? Or why not try me for general cussedness?” “That is just what we finally decided to do,” the ant-man wrote in reply. “We shall try you on general principles, and let the proper accusation develop from the evidence. “At some stage of the proceedings it will inevitably occur to some member of the council to suggest that you be charged with treason to Yuri, whereupon two members of the council, whom I have won over to the cause of my daughter, will raise the objection that Yuri is not our king. This will be the signal for the proclaiming of Queen Formis. If you will waive counsel the trial can take place to-morrow.” “I will waive anything,” Myles replied, “counsel, immunity, extradition, anything in order to speed up my return to Cupia, where Lilla awaits in some dire extremity.” “All right,” Doggo wrote, and the conference was at an end. The morrow would decide the ascendancy of Myles Cabot or the Prince Yuri over the new continent. IV THE COUP D’ETAT The next morning Myles Cabot was led under guard to the council chamber of the dread thirteen: Formis and her twelve advisers. The accused was placed in a wicker cage, from which he surveyed his surroundings as the proceedings opened. 19 On a raised platform stood the ant queen, surmounted by a scarlet canopy, which set off the perfect proportions of her jet-black body. On each side of her stood six refined and intelligent ant-men, her councillors. One of the twelve was Doggo. Messenger ants hurried hither and thither. First the accusation was read, Myles being furnished with a written copy. The witnesses were then called. They were veterans who had served in the wars in which Cabot had twice freed Cupia from the domination of its Formian oppressors. They spoke with bitterness of the downfall of their beloved Formia. Their testimony was brief. Then the accused was asked if he wished to say anything in his own behalf. Myles rose, then shrugged his shoulders, sat down again, and wrote: “I fully realize the futility of making an argument through the antennae of another.” Whereupon the queen and the council went into executive session. Their remarks were not intended for the eyes of the prisoner, but he soon observed that some kind of a dispute was on between Doggo, supported by two councillors named Emu and Fum on one side, and a councillor named Barth on the other. As this dispute reached its height, a messenger ant rushed in and held up one paw. Cabot’s interpreter, not deeming this a part of the executive session, obligingly translated the following into writing: The messenger: “Yuri lives and reigns over Cupia. It is his command that Cabot die.” Barth: “It is the radio. Know then, O Queen, and ye, members of the council, that when we fled across the boiling seas under the gallant leadership of Prince Yuri, the man with the heart of a Formian, he brought with him one of those powerful radio sets invented by the beast who is our prisoner here to-day. “Supporters of Yuri still remained among the Cupians, and he has been in constant communication with these ever since shortly after our arrival here. From them he learned of the return of Myles Cabot to the planet Minos. 20 “Then Yuri disappeared. Those of us who were closest to him suspected that he had gone back across the boiling seas to claim as his own the throne of Cupia. But we hesitated to announce this until we were sure, for we feared that some of our own people would regard his departure as desertion. Yet who can blame him for returning to his father-land and to the throne which is his by rights?” To which the messenger added: “And he offers to give us back our own old country, if we too will return across the boiling seas again.” “It is a lie!” Doggo shouted. “Yuri, usurper of the thrones of two continents. Bah!” shouted Emu. “Yuri, our rightful leader,” shouted Barth. “Give us a queen of our own race,” shouted Fum. “Release the prisoner,” shouted the Queen. And that is all that Myles learned of the conversation, for his interpreter at this juncture stopped writing and obeyed the queen. The earth-man was free! With one bound he gained the throne, where fighting was already in progress between the two factions. Barth and Doggo were rolling over and over on the floor in a death grapple, while the ant-queen had backed to the rear of the stage, closely guarded by Emu and Fum. Seizing one of the pikes which supported the scarlet canopy, Myles wrenched it loose and drove it into the thorax of Barth. In another instant the earth-man and Doggo stood beside the queen. Ant-men now came pouring into the chamber through all the entrances, taking sides as they entered and sized up the situation. If it had still been in vogue among the Formians to be known by numbers rather than names, and to have these identifying numbers painted on the backs of their abdomens followed by the numbers of those whom they had defeated in the duels so common among them, then many a Formian would have “got the number” of many another, that day.
|
C. They are both radio engineers, and presumably friends. Farley allowed Cabot to built a radio set and natter-transmitting device on his farm.
|
What classifier do they use?
|
### Introduction
Currently, social networks are so popular. Some of the biggest ones include Facebook, Twitter, Youtube,... with extremely number of users. Thus, controlling content of those platforms is essential. For years, social media companies such as Twitter, Facebook, and YouTube have been investing hundreds of millions euros on this task BIBREF0, BIBREF1. However, their effort is not enough since such efforts are primarily based on manual moderation to identify and delete offensive materials. The process is labour intensive, time consuming, and not sustainable or scalable in reality BIBREF2, BIBREF0, BIBREF3. In the sixth international workshop on Vietnamese Language and Speech Processing (VLSP 2019), the Hate Speech Detection (HSD) task is proposed as one of the shared-tasks to handle the problem related to controlling content in SNSs. HSD is required to build a multi-class classification model that is capable of classifying an item to one of 3 classes (hate, offensive, clean). Hate speech (hate): an item is identified as hate speech if it (1) targets individuals or groups on the basis of their characteristics; (2) demonstrates a clear intention to incite harm, or to promote hatred; (3) may or may not use offensive or profane words. Offensive but not hate speech (offensive): an item (posts/comments) may contain offensive words but it does not target individuals or groups on the basis of their characteristics. Neither offensive nor hate speech (clean): normal item, it does not contain offensive language or hate speech. The term `hate speech' was formally defined as `any communication that disparages a person or a group on the basis of some characteristics (to be referred to as types of hate or hate classes) such as race, colour, ethnicity, gender, sexual orientation, nationality, religion, or other characteristics' BIBREF4. Many researches have been conducted in recent years to develop automatic methods for hate speech detection in the social media domain. These typically employ semantic content analysis techniques built on Natural Language Processing (NLP) and Machine Learning (ML) methods. The task typically involves classifying textual content into non-hate or hateful. This HSD task is much more difficult when it requires classify text in three classes, with hate and offensive class quite hard to classify even with humans. In this paper, we propose a method to handle this HSD problem. Our system combines multiple text representations and models architecture in order to make diverse predictions. The system is heavily based on the ensemble method. The next section will present detail of our system including data preparation (how we clean text and build text representation), architecture of the model using in the system, and how we combine them together. The third section is our experiment and result report in HSD shared-task VLSP 2019. The final section is our conclusion with advantages and disadvantages of the system following by our perspective. ### System description
In this section, we present the system architecture. It includes how we pre-process text, what types of text representation we use and models used in our system. In the end, we combine model results by using an ensemble technique. ### System description ::: System overview
The fundamental idea of this system is how to make a system that has the diversity of viewing an input. That because of the variety of the meaning in Vietnamese language especially with the acronym, teen code type. To make this diversity, after cleaning raw text input, we use multiple types of word tokenizers. Each one of these tokenizers, we combine with some types of representation methods, including word to vector methods such as continuous bag of words BIBREF5, pre-trained embedding as fasttext (trained on Wiki Vietnamese language) BIBREF6 and sonvx (trained on Vietnamese newspaper) BIBREF7. Each sentence has a set of words corresponding to a set of word vectors, and that set of word vectors is a representation of a sentence. We also make a sentence embedding by using RoBERTa architecture BIBREF8. CBOW and RoBERTa models trained on text from some resources including VLSP 2016 Sentiment Analysis, VLSP 2018 Sentiment Analysis, VLSP 2019 HSD and text crawled from Facebook. After having sentence representation, we use some classification models to classify input sentences. Those models will be described in detail in the section SECREF13. With the multiply output results, we will use an ensemble method to combine them and output the final result. Ensemble method we use here is Stacking method will be introduced in the section SECREF16. ### System description ::: Data pre-processing
Content in the dataset that provided in this HSD task is very diverse. Words having the same meaning were written in various types (teen code, non tone, emojis,..) depending on the style of users. Dataset was crawled from various sources with multiple text encodes. In order to make it easy for training, all types of encoding need to be unified. This cleaning module will be used in two processes: cleaning data before training and cleaning input in inferring phase. Following is the data processing steps that we use: Step 1: Format encoding. Vietnamese has many accents, intonations with different Unicode typing programs which may have different outputs with the same typing type. To make it unified, we build a library named visen. For example, the input "thíêt kê will be normalized to "thiết kế" as the output. Step 2: In social networks, people show their feelings a lot by emojis. Emoticon is often a special Unicode character, but sometimes, it is combined by multiple normal characters like `: ( = ]'. We make a dictionary mapping this emoji (combined by some characters) to a single Unicode character like other emojis to make it unified. Step 3: Remove unseen characters. For human, unseen character is invisible but for a computer, it makes the model harder to process and inserts space between words, punctuation and emoji. This step aims at reducing the number of words in the dictionary which is important task, especially with low dataset resources like this HSD task. Step 4: With model requiring Vietnamese word segmentation as the input, we use BIBREF9, BIBREF10 to tokenize the input text. Step 5: Make all string lower. We experimented and found that lower-case or upper-case are not a significant impact on the result, but with lower characters, the number of words in the dictionary is reduced. RoBERTa proposed in BIBREF8 an optimized method for pretraining self-supervised NLP systems. In our system, we use RoBERTa not only to make sentence representation but also to augment data. With mask mechanism, we replace a word in the input sentence with another word that RoBERTa model proposes. To reduce the impact of replacement word, the chosen words are all common words that appear in almost three classes of the dataset. For example, with input `nhổn làm gắt vl', we can augment to other outputs: `vl làm gắt qá', `còn làm vl vậy', `vl làm đỉnh vl' or `thanh chút gắt vl'. british ### System description ::: Models architecture
Social comment dataset has high variety, the core idea is using multiple model architectures to handle data in many viewpoints. In our system, we use five different model architectures combining many types of CNN, and RNN. Each model will use some types of word embedding or handle directly sentence embedding to achieve the best general result. Source code of five models is extended from the GitHub repository The first model is TextCNN (figure FIGREF2) proposed in BIBREF11. It only contains CNN blocks following by some Dense layers. The output of multiple CNN blocks with different kernel sizes is connected to each other. The second model is VDCNN (figure FIGREF5) inspired by the research in BIBREF12. Like the TextCNN model, it contains multiple CNN blocks. The addition in this model is its residual connection. The third model is a simple LSTM bidirectional model (figure FIGREF15). It contains multiple LSTM bidirectional blocks stacked to each other. The fourth model is LSTMCNN (figure FIGREF24). Before going through CNN blocks, series of word embedding will be transformed by LSTM bidirectional block. The final model is the system named SARNN (figure FIGREF25). It adds an attention block between LTSM blocks. ### System description ::: Ensemble method
Ensemble methods is a machine learning technique that combines several base models in order to produce one optimal predictive model. Have the main three types of ensemble methods including Bagging, Boosting and Stacking. In this system, we use the Stacking method. In this method, the output of each model is not only class id but also the probability of each class in the set of three classes. This probability will become a feature for the ensemble model. The stacking ensemble model here is a simple full-connection model with input is all of probability that output from sub-model. The output is the probability of each class. ### Experiment
The dataset in this HSD task is really imbalance. Clean class dominates with 91.5%, offensive class takes 5% and the rest belongs to hate class with 3.5%. To make model being able to learn with this imbalance data, we inject class weight to the loss function with the corresponding ratio (clean, offensive, hate) is $(0.09, 0.95, 0.96)$. Formular DISPLAY_FORM17 is the loss function apply for all models in our system. $w_i$ is the class weight, $y_i$ is the ground truth and $\hat{y}_i$ is the output of the model. If the class weight is not set, we find that model cannot adjust parameters. The model tends to output all clean classes. We experiment 8 types of embedding in total: comment: CBOW embedding training in all dataset comment, each word is splited by space. Embedding size is 200. comment_bpe: CBOW embedding training in all dataset comment, each word is splited by subword bpe. Embedding size is 200. comment_tokenize: CBOW embedding training in all dataset comment, each word is splited by space. Before split by space, word is concatenated by using BIBREF9, BIBREF13, BIBREF10. Embedding size is 200. roberta: sentence embedding training in all dataset comment, training by using RoBERTa architecture. Embedding size is 256. fasttext, sonvx* is all pre-trained word embedding in general domain. Before mapping word to vector, word is concatenated by using BIBREF9, BIBREF13, BIBREF10. Embedding size of fasttext is 300. (sonvx_wiki, sonvx_baomoi_w2, sonvx_baomoi_w5) have embedding size corresponding is (400, 300, 400). In our experiment, the dataset is split into two-part: train set and dev set with the corresponding ratio $(0.9, 0.1)$. Two subsets have the same imbalance ratio like the root set. For each combination of model and word embedding, we train model in train set until it achieve the best result of loss score in the dev set. The table TABREF12 shows the best result of each combination on the f1_macro score. For each model having the best fit on the dev set, we export the probability distribution of classes for each sample in the dev set. In this case, we only use the result of model that has f1_macro score that larger than 0.67. The probability distribution of classes is then used as feature to input into a dense model with only one hidden layer (size 128). The training process of the ensemble model is done on samples of the dev set. The best fit result is 0.7356. The final result submitted in public leaderboard is 0.73019 and in private leaderboard is 0.58455. It is quite different in bad way. That maybe is the result of the model too overfit on train set tuning on public test set. Statistics of the final result on the dev set shows that almost cases have wrong prediction from offensive and hate class to clean class belong to samples containing the word `vl'. (62% in the offensive class and 48% in the hate class). It means that model overfit the word `vl' to the clean class. This makes sense because `vl' appears too much in the clean class dataset. In case the model predicts wrong from the clean class to the offensive class and the hate class, the model tends to decide case having sensitive words to be wrong class. The class offensive and the hate are quite difficult to distinguish even with human. ### Conclusion
In this study, we experiment the combination of multiple embedding types and multiple model architecture to solve a part of the problem Hate Speech Detection with a signification good classification results. Our system heavily based on the ensemble technique so the weakness of the system is slow processing speed. But in fact, it is not big trouble with this HSD problem when human usually involve handling directly in the before. HSD is a hard problem even with human. In order to improve classification quality, in the future, we need to collect more data especially social networks content. This will make building text representation more correct and help model easier to classify. british Figure 1. Hate Speech Detection System Overview Figure 2. TextCNN model architecture Figure 4. LSTM model architecture Figure 3. VDCNN model architecture Table I F1_MACRO SCORE OF DIFFERENT MODEL Figure 5. LSTMCNN model architecture Figure 6. SARNN model architecture
|
Stacking method, LSTMCNN, SARNN, simple LSTM bidirectional model, TextCNN
|
What is the overall tone of the article?
A. Serious
B. Grim
C. Violent
D. Objective
|
HOMECOMING BY MIGUEL HIDALGO What lasts forever? Does love? Does death?... Nothing lasts forever.... Not even forever [Transcriber's Note: This etext was produced from Worlds of If Science Fiction, April 1958. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The large horse plodded slowly over the shifting sand. The rider was of medium size, with huge, strong hands and seemingly hollow eyes. Strange eyes, alive and aflame. They had no place in the dust-caked, tired body, yet there they were, seeking, always seeking—searching the clear horizon, and never seeming to find what they sought. The horse moved faster now. They were nearing a river; the water would be welcome on tired bodies and dry throats. He spurred his horse, and when they reached the water's edge, he dismounted and unsaddled the horse. Then both man and horse plunged headlong into the waiting torrent, deep into the cool embrace of the clear liquid. They soaked it into their pores and drank deeply of it, feeling life going once more through their veins. Satisfied, they lifted themselves from the water, and the man lay down on the yellow sand of the river bank to sleep. When he awoke, the sun was almost setting. The bright shafts of red light spilled across the sky, making the mountains silent scarlet shadows on the face of the rippling water. Quickly he gathered driftwood, and built a small fire. From his pack he removed some of the coffee he had found in one of the ruined cities. He brought water from the river in the battered coffee-pot he had salvaged, and while he waited for it to boil, he went to his horse, Conqueror, stroking his mane and whispering in his ear. Then he led him silently to a grassy slope where he hobbled him and left him for the night. In the fading light, he ate the hard beef jerky and drank the scalding coffee. Refreshed and momentarily content, he sat staring into the dying fire, seeing the bright glowing coals as living fingers clutching at the wood in consuming embrace, taking all and returning nothing but ashes. Slowly his eyelids yielded. His body sagged, and blood seemed to fill his brain, bathing it in a gentle, warm flood. He slept. His brain slept. But the portion of his brain called memory stirred. It was all alone; all else was at rest. Images began to appear, drawn from inexhaustible files, wherein are kept all thoughts, past, present, and future.... It was the night before he was to go overseas. World War III had been declared, and he had enlisted, receiving his old rank of captain. He was with his wife in the living room of their home. They had put the children to bed—their sons—and now sat on the couch, watching the blazing fire. It was then that he had showed it to her. "I've got something to tell you, and something to show you." He had removed the box from his pocket and opened it. And heard her cry of surprised joy. "Oh, a ring, and it's a diamond, too!" she cried in her rich, happy voice which always seemed to send a thrill through his body. "It's for you; so long as you wear it, I'll come back, even from the dead, if need be. Read the inscription." She held the ring up to the light and read aloud, "It is forever." Then she had slipped the ring on her finger and her arms around him. He held her very close, feeling the warmth from her body flowing into his and making him oblivious to everything except that she was there in his arms and that he was sinking deep, deep into a familiar sea, where he had been many times before but each time found something new and unexplored, some vastly different emotion he could never quite explain. "Wait!" she cried. "I've something for you, too." She took off the locket she wore about her neck and held it up to the shimmering light, letting it spin at the end of its chain. It caught the shadows of the fire and reflected them, greatly magnified, over the room. It was in the shape of a star, encrusted with emeralds, with one large ruby in the center. When he opened it, he found a picture of her in one side, and in the other a picture of the children. He took her in his arms again, and loosened her long, black hair, burying his face in it for a moment. Then he kissed her, and instantly was drawn down into the abyss which seemed to have no beginning or any end. The next morning had been bleak and gray. The mist clung to the wet, sodden ground, and the air was heavy in his lungs. He had driven off in the jeep the army had sent for him, watching her there on the porch until the mist swirled around her feet and she ran back into the house and slammed the door. His cold fingers found the locket, making a little bulge under his uniform, and the touch of it seemed to warm the blood in his veins. Three days later they had landed in Spain, merged with another division, then crossed the Pyrenees into France, and finally to Paris where the fighting had begun. Already the city was a silent graveyard, littered with the rubble of towers and cathedrals which had once been great. Three years later they were on the road to Moscow. Over a thousand miles lay behind, a dead man on every foot of those miles. Yet victory was near. The Russians had not yet used the H-bomb; the threat of annihilation by the retaliation forces had been too great. He had done well in the war, and had been decorated many times for bravery in action. Now he felt the victory that seemed to be in the air, and he had wished it would come quickly, so that he might return to her. Home. The very feel of the word was everything a battle-weary soldier needed to make him fight harder and live longer. Suddenly he had become aware of a droning, wooshing sound above him. It grew louder and louder until he knew what it was. "Heavy bombers!" The alarm had sounded, and the men had headed for their foxholes. But the planes had passed over, the sun glinting on their bellies, reflecting a blinding light. They were bound for bigger, more important targets. When the all-clear had sounded, the men clambered from their shelters. An icy wind swept the field, bringing with it clouds which covered the sun. A strange fear had gripped him then.... Across the Atlantic, over the pole, via Alaska, the great bombers flew. In cities, great and small, the air raid sirens sounded, high screaming noises which had jarred the people from sleep in time to die. The defending planes roared into the sky to intercept the on-rushing bombers. The horrendous battle split the universe. Many bombers fell, victims of fanatical suicide planes, or of missiles that streaked across the sky which none could escape. But too many bombers got through, dropping their deadly cargo upon the helpless cities. And not all the prayers or entreaties to any God had stopped their carnage. First there had been the red flashes that melted buildings into molten streams, and then the great triple-mushroom cloud filled with the poisonous gases that the wind swept away to other cities, where men had not died quickly and mercifully, but had rotted away, leaving shreds of putrid flesh behind to mark the places where they had crawled. The retaliatory forces had roared away to bomb the Russian cities. Few, if any, had returned. Too much blood and life were on their hands. Those who had remained alive had found a resting place on the crown of some distant mountain. Others had preferred the silent peaceful sea, where flesh stayed not long on bones, and only darting fishes and merciful beams of filtered light found their aluminum coffins. The war had ended. To no avail. Neither side had won. Most of the cities and the majority of the population of both countries had been destroyed. Even their governments had vanished, leaving a silent nothingness. The armies that remained were without leaders, without sources of supplies, save what they could forage and beg from an unfriendly people. They were alone now, a group of tired, battered men, for whom life held nothing. Their families had long since died, their bodies turned to dust, their spirits fled on the winds to a new world. Yet these remnants of an army must return—or at least try. Their exodus was just beginning. Somehow he had managed to hold together the few men left from his force. He had always nourished the hope that she might still be alive. And now that the war was over he had to return—had to know whether she was still waiting for him. They had started the long trek. Throughout Europe anarchy reigned. He and his men were alone. All they could do now was fight. Finally they reached the seaport city of Calais. With what few men he had left, he had commandeered a small yacht, and they had taken to the sea. After months of storms and bad luck, they had been shipwrecked somewhere off the coast of Mexico. He had managed to swim ashore, and had been found by a fisherman's family. Many months he had spent swimming and fishing, recovering his strength, inquiring about the United States. The Mexicans had spoken with fear of the land across the Rio Grande. All its great cities had been destroyed, and those that had been only partially destroyed were devoid of people. The land across the Rio Grande had become a land of shadows. The winds were poisoned, and the few people who might have survived, were crazed and maimed by the blasts. Few men had dared cross the Rio Grande into "El Mundo gris de Noviembre"—the November world. Those who had, had never returned. In time he had traveled north until he reached the Rio Grande. He had waded into the muddy waters and somehow landed on the American side. In the November world. It was rightly called. The deserts were long. All plant life had died, leaving to those once great fertile stretches, nothing but the sad, temporal beauty that comes with death. No people had he seen. Only the ruins of what had once been their cities. He had walked through them, and all that he had seen were the small mutant rodents, and all that he had heard was the occasional swish of the wind as it whisked along what might have been dead leaves, but wasn't. He had been on the trail for a long time. His food was nearly exhausted. The mountains were just beginning, and he hoped to find food there. He had not found food, but his luck had been with him. He had found a horse. Not a normal horse, but a mutation. It was almost twice as large as a regular horse. Its skin seemed to shimmer and was like glassy steel to the touch. From the center of its forehead grew a horn, straight out, as the horn of a unicorn. But most startling of all were the animal's eyes which seemed to speak—a silent mental speech, which he could understand. The horse had looked up as he approached it and seemed to say: "Follow me." And he had followed. Over a mountain, until they came to a pass, and finally to a narrow path which led to an old cabin. He had found it empty, but there were cans of food and a rifle and many shells. He had remained there a long time—how long he could not tell, for he could only measure time by the cycles of the sun and the moon. Finally he had taken the horse, the rifle and what food was left, and once again started the long journey home. The farther north he went, the more life seemed to have survived. He had seen great herds of horses like his own, stampeding across the plains, and strange birds which he could not identify. Yet he had seen no human beings. But he knew he was closer now. Closer to home. He recognized the land. How, he did not know, for it was much changed. A sensing, perhaps, of what it had once been. He could not be more than two days' ride away. Once he was through this desert, he would find her, he would be with her once again; all would be well, and his long journey would be over. The images faded. Even memory slept in a flow of warm blood. Body and mind slept into the shadows of the dawn. He awoke and stretched the cramped muscles of his body. At the edge of the water he removed his clothes and stared at himself in the rippling mirror. His muscles were lean and hard, evenly placed throughout the length of his frame. A deep ridge ran down the length of his torso, separating the muscles, making the chest broad. Well satisfied with his body, he plunged into the cold water, deep down, until he thought his lungs would burst; then swiftly returned to the clean air, tingling in every pore. He dried himself and dressed. Conqueror was eating the long grass near the stream. Quickly he saddled him. No time for breakfast. He would ride all day and the next night. And he would be home. Still northward. The hours crawled slower than a dying man. The sun was a torch that pierced his skin, seeming to melt his bones into a burning stream within his body. But day at last gave way to night, and the sun to the moon. The torch became a white pock-marked goddess, with streaming hair called stars. In the moonlight he had not seen the crater until he was at its very edge. Even then he might not have seen it had not the horse stopped suddenly. The wind swirled through its vast emptiness, slapping his face with dusty hands. For a moment he thought he heard voices—mournful, murmuring voices, echoing up from the misty depths. He turned quickly away and did not look back. Night paled into day; day burned into night. There were clouds in the sky now, and a gentle wind caressed the sweat from his tired body. He stopped. There it was! Barely discernible through the moonlight, he saw it. Home. Quickly he dismounted and ran. Now he could see a small light in the window, and he knew they were there. His breath came in hard ragged gulps. At the window he peered in, and as his eyes became accustomed to the inner gloom, he saw how bare the room was. No matter. Now that he was home he would build new furniture, and the house would be even better than it had been before. Then he saw her. She was sitting motionless in a straight wooden chair beside the fireplace, the feeble light cast by the embers veiling her in mauve shadows. He waited, wondering if she were.... Presently she stirred like a restless child in sleep, then moved from the chair to the pile of wood near the hearth, and replenished the fire. The wood caught quickly, sending up long tongues of flame, and forming a bright pool of light around her. His blood froze. The creature illuminated by the firelight was a monster. Large greasy scales covered its face and arms, and there was no hair on its head. Its gums were toothless cavities in a sunken, mumbling mouth. The eyes, turned momentarily toward the window, were empty of life. "No, no!" he cried soundlessly. This was not his house. In his delirium he had only imagined he had found it. He had been searching so long. He would go on searching. He was turning wearily away from the window when the movement of the creature beside the fire held his attention. It had taken a ring from one skeleton-like finger and stood, turning the ring slowly as if trying to decipher some inscription inside it. He knew then. He had come home. Slowly he moved toward the door. A great weakness was upon him. His feet were stones, reluctant to leave the earth. His body was a weed, shriveled by thirst. He grasped the doorknob and clung to it, looking up at the night sky and trying to draw strength from the wind that passed over him. It was no use. There was no strength. Only fear—a kind of fear he had never known. He fumbled at his throat, his fingers crawling like cold worms around his neck until he found the locket and the clasp which had held it safely through endless nightmare days and nights. He slipped the clasp and the locket fell into his waiting hand. As one in a dream, he opened it, and stared at the pictures, now in the dim moonlight no longer faces of those he loved, but grey ghosts from the past. Even the ruby had lost its glow. What had once been living fire was now a dull glob of darkness. "Nothing is forever!" He thought he had shouted the words, but only a thin sound, the sound of leaves ruffled by the wind, came back to him. He closed the locket and fastened the clasp, and hung it on the doorknob. It moved slowly in the wind, back and forth, like a pendulum. "Forever—forever. Only death is forever." He could have sworn he heard the words. He ran. Away from the house. To the large horse with a horn in the center of its forehead, like a unicorn. Once in the saddle, the spurt of strength left him. His shoulders slumped, his head dropped onto his chest. Conqueror trotted away, the sound of his hooves echoing hollowly in the vast emptiness.
|
B. Grim
|
Why do Reston-Farrell and Brett-James bring Joe to the future?
A. Joe was going to kill Al Rossi. Reston-Farrell and Brett James need Rossi alive.
B. Joe is a caregiver. They want him to take care of someone.
C. Joe is a hitman. They want him to kill someone.
D. Joe is a variant. They removed him from 1960 to correct the timeline.
|
Illustrated by van Dongen A gun is an interesting weapon; it can be hired, of course, and naturally doesn't care who hires it. Something much the same can be said of the gunman, too.... GUN FOR HIRE By MACK REYNOLDS Joe Prantera called softly, "Al." The pleasurable, comfortable, warm feeling began spreading over him, the way it always did. The older man stopped and squinted, but not suspiciously, even now. The evening was dark, it was unlikely that the other even saw the circle of steel that was the mouth of the shotgun barrel, now resting on the car's window ledge. "Who's it?" he growled. Joe Prantera said softly, "Big Louis sent me, Al." And he pressed the trigger. And at that moment, the universe caved inward upon Joseph Marie Prantera. There was nausea and nausea upon nausea. There was a falling through all space and through all time. There was doubling and twisting and twitching of every muscle and nerve. There was pain, horror and tumultuous fear. And he came out of it as quickly and completely as he'd gone in. He was in, he thought, a hospital and his first reaction was to think, This here California. Everything different. Then his second thought was Something went wrong. Big Louis, he ain't going to like this. He brought his thinking to the present. So far as he could remember, he hadn't completely pulled the trigger. That at least meant that whatever the rap was it wouldn't be too tough. With luck, the syndicate would get him off with a couple of years at Quentin. A door slid open in the wall in a way that Joe had never seen a door operate before. This here California. The clothes on the newcomer were wrong, too. For the first time, Joe Prantera began to sense an alienness—a something that was awfully wrong. The other spoke precisely and slowly, the way a highly educated man speaks a language which he reads and writes fluently but has little occasion to practice vocally. "You have recovered?" Joe Prantera looked at the other expressionlessly. Maybe the old duck was one of these foreign doctors, like. The newcomer said, "You have undoubtedly been through a most harrowing experience. If you have any untoward symptoms, possibly I could be of assistance." Joe couldn't figure out how he stood. For one thing, there should have been some kind of police guard. The other said, "Perhaps a bit of stimulant?" Joe said flatly, "I wanta lawyer." The newcomer frowned at him. "A lawyer?" "I'm not sayin' nothin'. Not until I get a mouthpiece." The newcomer started off on another tack. "My name is Lawrence Reston-Farrell. If I am not mistaken, you are Joseph Salviati-Prantera." Salviati happened to be Joe's mother's maiden name. But it was unlikely this character could have known that. Joe had been born in Naples and his mother had died in childbirth. His father hadn't brought him to the States until the age of five and by that time he had a stepmother. "I wanta mouthpiece," Joe said flatly, "or let me outta here." Lawrence Reston-Farrell said, "You are not being constrained. There are clothes for you in the closet there." Joe gingerly tried swinging his feet to the floor and sitting up, while the other stood watching him, strangely. He came to his feet. With the exception of a faint nausea, which brought back memories of that extreme condition he'd suffered during ... during what? He hadn't the vaguest idea of what had happened. He was dressed in a hospital-type nightgown. He looked down at it and snorted and made his way over to the closet. It opened on his approach, the door sliding back into the wall in much the same manner as the room's door had opened for Reston-Farrell. Joe Prantera scowled and said, "These ain't my clothes." "No, I am afraid not." "You think I'd be seen dead wearing this stuff? What is this, some religious crackpot hospital?" Reston-Farrell said, "I am afraid, Mr. Salviati-Prantera, that these are the only garments available. I suggest you look out the window there." Joe gave him a long, chill look and then stepped to the window. He couldn't figure the other. Unless he was a fruitcake. Maybe he was in some kind of pressure cooker and this was one of the fruitcakes. He looked out, however, not on the lawns and walks of a sanitarium but upon a wide boulevard of what was obviously a populous city. And for a moment again, Joe Prantera felt the depths of nausea. This was not his world. He stared for a long, long moment. The cars didn't even have wheels, he noted dully. He turned slowly and faced the older man. Reston-Farrell said compassionately, "Try this, it's excellent cognac." Joe Prantera stared at him, said finally, flatly, "What's it all about?" The other put down the unaccepted glass. "We were afraid first realization would be a shock to you," he said. "My colleague is in the adjoining room. We will be glad to explain to you if you will join us there." "I wanta get out of here," Joe said. "Where would you go?" The fear of police, of Al Rossi's vengeance, of the measures that might be taken by Big Louis on his failure, were now far away. Reston-Farrell had approached the door by which he had entered and it reopened for him. He went through it without looking back. There was nothing else to do. Joe dressed, then followed him. In the adjoining room was a circular table that would have accommodated a dozen persons. Two were seated there now, papers, books and soiled coffee cups before them. There had evidently been a long wait. Reston-Farrell, the one Joe had already met, was tall and drawn of face and with a chainsmoker's nervousness. The other was heavier and more at ease. They were both, Joe estimated, somewhere in their middle fifties. They both looked like docs. He wondered, all over again, if this was some kind of pressure cooker. But that didn't explain the view from the window. Reston-Farrell said, "May I present my colleague, Citizen Warren Brett-James? Warren, this is our guest from ... from yesteryear, Mr. Joseph Salviati-Prantera." Brett-James nodded to him, friendly, so far as Joe could see. He said gently, "I think it would be Mr. Joseph Prantera, wouldn't it? The maternal linage was almost universally ignored." His voice too gave the impression he was speaking a language not usually on his tongue. Joe took an empty chair, hardly bothering to note its alien qualities. His body seemed to fit into the piece of furniture, as though it had been molded to his order. Joe said, "I think maybe I'll take that there drink, Doc." Reston-Farrell said, "Of course," and then something else Joe didn't get. Whatever the something else was, a slot opened in the middle of the table and a glass, so clear of texture as to be all but invisible, was elevated. It contained possibly three ounces of golden fluid. Joe didn't allow himself to think of its means of delivery. He took up the drink and bolted it. He put the glass down and said carefully, "What's it all about, huh?" Warren Brett-James said soothingly, "Prepare yourself for somewhat of a shock, Mr. Prantera. You are no longer in Los Angeles—" "Ya think I'm stupid? I can see that." "I was about to say, Los Angeles of 1960. Mr. Prantera, we welcome you to Nuevo Los Angeles." "Ta where?" "To Nuevo Los Angeles and to the year—" Brett-James looked at his companion. "What is the date, Old Calendar?" "2133," Reston-Farrell said. "2133 A.D. they would say." Joe Prantera looked from one of them to the other, scowling. "What are you guys talking about?" Warren Brett-James said softly, "Mr. Prantera, you are no longer in the year 1960, you are now in the year 2133." He said, uncomprehendingly, "You mean I been, like, unconscious for—" He let the sentence fall away as he realized the impossibility. Brett-James said gently, "Hardly for one hundred and seventy years, Mr. Prantera." Reston-Farrell said, "I am afraid we are confusing you. Briefly, we have transported you, I suppose one might say, from your own era to ours." Joe Prantera had never been exposed to the concept of time travel. He had simply never associated with anyone who had ever even remotely considered such an idea. Now he said, "You mean, like, I been asleep all that time?" "Not exactly," Brett-James said, frowning. Reston-Farrell said, "Suffice to say, you are now one hundred and seventy-three years after the last memory you have." Joe Prantera's mind suddenly reverted to those last memories and his eyes narrowed dangerously. He felt suddenly at bay. He said, "Maybe you guys better let me in on what's this all about." Reston-Farrell said, "Mr. Prantera, we have brought you from your era to perform a task for us." Joe stared at him, and then at the other. He couldn't believe he was getting through to them. Or, at least, that they were to him. Finally he said, "If I get this, you want me to do a job for you." "That is correct." Joe said, "You guys know the kind of jobs I do?" "That is correct." "Like hell you do. You think I'm stupid? I never even seen you before." Joe Prantera came abruptly to his feet. "I'm gettin' outta here." For the second time, Reston-Farrell said, "Where would you go, Mr. Prantera?" Joe glared at him. Then sat down again, as abruptly as he'd arisen. "Let's start all over again. I got this straight, you brought me, some screwy way, all the way ... here. O.K., I'll buy that. I seen what it looks like out that window—" The real comprehension was seeping through to him even as he talked. "Everybody I know, Jessie, Tony, the Kid, Big Louis, everybody, they're dead. Even Big Louis." "Yes," Brett-James said, his voice soft. "They are all dead, Mr. Prantera. Their children are all dead, and their grandchildren." The two men of the future said nothing more for long minutes while Joe Prantera's mind whirled its confusion. Finally he said, "What's this bit about you wanting me to give it to some guy." "That is why we brought you here, Mr. Prantera. You were ... you are, a professional assassin." "Hey, wait a minute, now." Reston-Farrell went on, ignoring the interruption. "There is small point in denying your calling. Pray remember that at the point when we ... transported you, you were about to dispose of a contemporary named Alphonso Annunziata-Rossi. A citizen, I might say, whose demise would probably have caused small dismay to society." They had him pegged all right. Joe said, "But why me? Why don't you get some heavy from now? Somebody knows the ropes these days." Brett-James said, "Mr. Prantera, there are no professional assassins in this age, nor have there been for over a century and a half." "Well, then do it yourself." Joe Prantera's irritation over this whole complicated mess was growing. And already he was beginning to long for the things he knew—for Jessie and Tony and the others, for his favorite bar, for the lasagne down at Papa Giovanni's. Right now he could have welcomed a calling down at the hands of Big Louis. Reston-Farrell had come to his feet and walked to one of the large room's windows. He looked out, as though unseeing. Then, his back turned, he said, "We have tried, but it is simply not in us, Mr. Prantera." "You mean you're yella?" "No, if by that you mean afraid. It is simply not within us to take the life of a fellow creature—not to speak of a fellow man." Joe snapped: "Everything you guys say sounds crazy. Let's start all over again." Brett-James said, "Let me do it, Lawrence." He turned his eyes to Joe. "Mr. Prantera, in your own era, did you ever consider the future?" Joe looked at him blankly. "In your day you were confronted with national and international, problems. Just as we are today and just as nations were a century or a millennium ago." "Sure, O.K., so we had problems. I know whatcha mean—like wars, and depressions and dictators and like that." "Yes, like that," Brett-James nodded. The heavy-set man paused a moment. "Yes, like that," he repeated. "That we confront you now indicates that the problems of your day were solved. Hadn't they been, the world most surely would have destroyed itself. Wars? Our pedagogues are hard put to convince their students that such ever existed. More than a century and a half ago our society eliminated the reasons for international conflict. For that matter," he added musingly, "we eliminated most international boundaries. Depressions? Shortly after your own period, man awoke to the fact that he had achieved to the point where it was possible to produce an abundance for all with a minimum of toil. Overnight, for all practical purposes, the whole world was industrialized, automated. The second industrial revolution was accompanied by revolutionary changes in almost every field, certainly in every science. Dictators? Your ancestors found, Mr. Prantera, that it is difficult for a man to be free so long as others are still enslaved. Today the democratic ethic has reached a pinnacle never dreamed of in your own era." "O.K., O.K.," Joe Prantera growled. "So everybody's got it made. What I wanta know is what's all this about me giving it ta somebody? If everything's so great, how come you want me to knock this guy off?" Reston-Farrell bent forward and thumped his right index finger twice on the table. "The bacterium of hate—a new strain—has found the human race unprotected from its disease. We had thought our vaccines immunized us." "What's that suppose to mean?" Brett-James took up the ball again. "Mr. Prantera, have you ever heard of Ghengis Khan, of Tamerlane, Alexander, Caesar?" Joe Prantera scowled at him emptily. "Or, more likely, of Napoleon, Hitler, Stalin?" "Sure I heard of Hitler and Stalin," Joe growled. "I ain't stupid." The other nodded. "Such men are unique. They have a drive ... a drive to power which exceeds by far the ambitions of the average man. They are genii in their way, Mr. Prantera, genii of evil. Such a genius of evil has appeared on the current scene." "Now we're getting somewheres," Joe snorted. "So you got a guy what's a little ambitious, like, eh? And you guys ain't got the guts to give it to him. O.K. What's in it for me?" The two of them frowned, exchanged glances. Reston-Farrell said, "You know, that is one aspect we had not considered." Brett-James said to Joe Prantera, "Had we not, ah, taken you at the time we did, do you realize what would have happened?" "Sure," Joe grunted. "I woulda let old Al Rossi have it right in the guts, five times. Then I woulda took the plane back to Chi." Brett-James was shaking his head. "No. You see, by coincidence, a police squad car was coming down the street just at that moment to arrest Mr. Rossi. You would have been apprehended. As I understand Californian law of the period, your life would have been forfeit, Mr. Prantera." Joe winced. It didn't occur to him to doubt their word. Reston-Farrell said, "As to reward, Mr. Prantera, we have already told you there is ultra-abundance in this age. Once this task has been performed, we will sponsor your entry into present day society. Competent psychiatric therapy will soon remove your present—" "Waita minute, now. You figure on gettin' me candled by some head shrinker, eh? No thanks, Buster. I'm going back to my own—" Brett-James was shaking his head again. "I am afraid there is no return, Mr. Prantera. Time travel works but in one direction, with the flow of the time stream. There can be no return to your own era." Joe Prantera had been rocking with the mental blows he had been assimilating, but this was the final haymaker. He was stuck in this squaresville of a world. Joe Prantera on a job was thorough. Careful, painstaking, competent. He spent the first three days of his life in the year 2133 getting the feel of things. Brett-James and Reston-Farrell had been appointed to work with him. Joe didn't meet any of the others who belonged to the group which had taken the measures to bring him from the past. He didn't want to meet them. The fewer persons involved, the better. He stayed in the apartment of Reston-Farrell. Joe had been right, Reston-Farrell was a medical doctor. Brett-James evidently had something to do with the process that had enabled them to bring Joe from the past. Joe didn't know how they'd done it, and he didn't care. Joe was a realist. He was here. The thing was to adapt. There didn't seem to be any hurry. Once the deal was made, they left it up to him to make the decisions. They drove him around the town, when he wished to check the traffic arteries. They flew him about the whole vicinity. From the air, Southern California looked much the same as it had in his own time. Oceans, mountains, and to a lesser extent, deserts, are fairly permanent even against man's corroding efforts. It was while he was flying with Brett-James on the second day that Joe said, "How about Mexico? Could I make the get to Mexico?" The physicist looked at him questioningly. "Get?" he said. Joe Prantera said impatiently, "The getaway. After I give it to this Howard Temple-Tracy guy, I gotta go on the run, don't I?" "I see." Brett-James cleared his throat. "Mexico is no longer a separate nation, Mr. Prantera. All North America has been united into one unit. Today, there are only eight nations in the world." "Where's the nearest?" "South America." "That's a helluva long way to go on a get." "We hadn't thought of the matter being handled in that manner." Joe eyed him in scorn. "Oh, you didn't, huh? What happens after I give it to this guy? I just sit around and wait for the cops to put the arm on me?" Brett-James grimaced in amusement. "Mr. Prantera, this will probably be difficult for you to comprehend, but there are no police in this era." Joe gaped at him. "No police! What happens if you gotta throw some guy in stir?" "If I understand your idiom correctly, you mean prison. There are no prisons in this era, Mr. Prantera." Joe stared. "No cops, no jails. What stops anybody? What stops anybody from just going into some bank, like, and collecting up all the bread?" Brett-James cleared his throat. "Mr. Prantera, there are no banks." "No banks! You gotta have banks!" "And no money to put in them. We found it a rather antiquated method of distribution well over a century ago." Joe had given up. Now he merely stared. Brett-James said reasonably, "We found we were devoting as much time to financial matters in all their endless ramifications—including bank robberies—as we were to productive efforts. So we turned to more efficient methods of distribution." On the fourth day, Joe said, "O.K., let's get down to facts. Summa the things you guys say don't stick together so good. Now, first place, where's this guy Temple-Tracy you want knocked off?" Reston-Farrell and Brett-James were both present. The three of them sat in the living room of the latter's apartment, sipping a sparkling wine which seemed to be the prevailing beverage of the day. For Joe's taste it was insipid stuff. Happily, rye was available to those who wanted it. Reston-Farrell said, "You mean, where does he reside? Why, here in this city." "Well, that's handy, eh?" Joe scratched himself thoughtfully. "You got somebody can finger him for me?" "Finger him?" "Look, before I can give it to this guy I gotta know some place where he'll be at some time. Get it? Like Al Rossi. My finger, he works in Rossi's house, see? He lets me know every Wednesday night, eight o'clock, Al leaves the house all by hisself. O.K., so I can make plans, like, to give it to him." Joe Prantera wound it up reasonably. "You gotta have a finger." Brett-James said, "Why not just go to Temple-Tracy's apartment and, ah, dispose of him?" "Jest walk in, eh? You think I'm stupid? How do I know how many witnesses hangin' around? How do I know if the guy's carryin' heat?" "Heat?" "A gun, a gun. Ya think I'm stupid? I come to give it to him and he gives it to me instead." Dr. Reston-Farrell said, "Howard Temple-Tracy lives alone. He customarily receives visitors every afternoon, largely potential followers. He is attempting to recruit members to an organization he is forming. It would be quite simple for you to enter his establishment and dispose of him. I assure you, he does not possess weapons." Joe was indignant. "Just like that, eh?" he said sarcastically. "Then what happens? How do I get out of the building? Where's my get car parked? Where do I hide out? Where do I dump the heat?" "Dump the heat?" "Get rid of the gun. You want I should get caught with the gun on me? I'd wind up in the gas chamber so quick—" "See here, Mr. Prantera," Brett-James said softly. "We no longer have capital punishment, you must realize." "O.K. I still don't wanta get caught. What is the rap these days, huh?" Joe scowled. "You said they didn't have no jails any more." "This is difficult for you to understand, I imagine," Reston-Farrell told him, "but, you see, we no longer punish people in this era." That took a long, unbelieving moment to sink in. "You mean, like, no matter what they do? That's crazy. Everybody'd be running around giving it to everybody else." "The motivation for crime has been removed, Mr. Prantera," Reston-Farrell attempted to explain. "A person who commits a violence against another is obviously in need of medical care. And, consequently, receives it." "You mean, like, if I steal a car or something, they just take me to a doctor?" Joe Prantera was unbelieving. "Why would anybody wish to steal a car?" Reston-Farrell said easily. "But if I give it to somebody?" "You will be turned over to a medical institution. Citizen Howard Temple-Tracy is the last man you will ever kill, Mr. Prantera." A chillness was in the belly of Joe Prantera. He said very slowly, very dangerously, "You guys figure on me getting caught, don't you?" "Yes," Brett-James said evenly. "Well then, figure something else. You think I'm stupid?" "Mr. Prantera," Dr. Reston-Farrell said, "there has been as much progress in the field of psychiatry in the past two centuries as there has in any other. Your treatment would be brief and painless, believe me." Joe said coldly, "And what happens to you guys? How do you know I won't rat on you?" Brett-James said gently, "The moment after you have accomplished your mission, we plan to turn ourselves over to the nearest institution to have determined whether or not we also need therapy." "Now I'm beginning to wonder about you guys," Joe said. "Look, all over again, what'd'ya wanta give it to this guy for?" The doctor said, "We explained the other day, Mr. Prantera. Citizen Howard Temple-Tracy is a dangerous, atavistic, evil genius. We are afraid for our institutions if his plans are allowed to mature." "Well if you got things so good, everybody's got it made, like, who'd listen to him?" The doctor nodded at the validity of the question. "Mr. Prantera, Homo sapiens is a unique animal. Physically he matures at approximately the age of thirteen. However, mental maturity and adjustment is often not fully realized until thirty or even more. Indeed, it is sometimes never achieved. Before such maturity is reached, our youth are susceptible to romantic appeal. Nationalism, chauvinism, racism, the supposed glory of the military, all seem romantic to the immature. They rebel at the orderliness of present society. They seek entertainment in excitement. Citizen Temple-Tracy is aware of this and finds his recruits among the young." "O.K., so this guy is dangerous. You want him knocked off before he screws everything up. But the way things are, there's no way of making a get. So you'll have to get some other patsy. Not me." "I am afraid you have no alternative," Brett-James said gently. "Without us, what will you do? Mr. Prantera, you do not even speak the language." "What'd'ya mean? I don't understand summa the big words you eggheads use, but I get by O.K." Brett-James said, "Amer-English is no longer the language spoken by the man in the street, Mr. Prantera. Only students of such subjects any longer speak such tongues as Amer-English, French, Russian or the many others that once confused the race with their limitations as a means of communication." "You mean there's no place in the whole world where they talk American?" Joe demanded, aghast. Dr. Reston-Farrell controlled the car. Joe Prantera sat in the seat next to him and Warren Brett-James sat in the back. Joe had, tucked in his belt, a .45 caliber automatic, once displayed in a museum. It had been more easily procured than the ammunition to fit it, but that problem too had been solved. The others were nervous, obviously repelled by the very conception of what they had planned. Inwardly, Joe was amused. Now that they had got in the clutch, the others were on the verge of chickening out. He knew it wouldn't have taken much for them to cancel the project. It wasn't any answer though. If they allowed him to call it off today, they'd talk themselves into it again before the week was through. Besides, already Joe was beginning to feel the comfortable, pleasurable, warm feeling that came to him on occasions like this. He said, "You're sure this guy talks American, eh?" Warren Brett-James said, "Quite sure. He is a student of history." "And he won't think it's funny I talk American to him, eh?" "He'll undoubtedly be intrigued." They pulled up before a large apartment building that overlooked the area once known as Wilmington. Joe was coolly efficient now. He pulled out the automatic, held it down below his knees and threw a shell into the barrel. He eased the hammer down, thumbed on the safety, stuck the weapon back in his belt and beneath the jacketlike garment he wore. He said, "O.K. See you guys later." He left them and entered the building. An elevator—he still wasn't used to their speed in this era—whooshed him to the penthouse duplex occupied by Citizen Howard Temple-Tracy. There were two persons in the reception room but they left on Joe's arrival, without bothering to look at him more than glancingly. He spotted the screen immediately and went over and stood before it. The screen lit and revealed a heavy-set, dour of countenance man seated at a desk. He looked into Joe Prantera's face, scowled and said something. Joe said, "Joseph Salviati-Prantera to interview Citizen Howard Temple-Tracy." The other's shaggy eyebrows rose. "Indeed," he said. "In Amer-English?" Joe nodded. "Enter," the other said. A door had slid open on the other side of the room. Joe walked through it and into what was obviously an office. Citizen Temple-Tracy sat at a desk. There was only one other chair in the room. Joe Prantera ignored it and remained standing. Citizen Temple-Tracy said, "What can I do for you?" Joe looked at him for a long, long moment. Then he reached down to his belt and brought forth the .45 automatic. He moistened his lips. Joe said softly, "You know what this here is?" Temple-Tracy stared at the weapon. "It's a handgun, circa, I would say, about 1925 Old Calendar. What in the world are you doing with it?" Joe said, very slowly, "Chief, in the line you're in these days you needa heavy around with wunna these. Otherwise, Chief, you're gunna wind up in some gutter with a lotta holes in you. What I'm doin', I'm askin' for a job. You need a good man knows how to handle wunna these, Chief." Citizen Howard Temple-Tracy eyed him appraisingly. "Perhaps," he said, "you are right at that. In the near future, I may well need an assistant knowledgeable in the field of violence. Tell me more about yourself. You surprise me considerably." "Sure, Chief. It's kinda a long story, though. First off, I better tell you you got some bad enemies, Chief. Two guys special, named Brett-James and Doc Reston-Farrell. I think one of the first jobs I'm gunna hafta do for you, Chief, is to give it to those two." THE END Transcriber's Note: This etext was produced from Analog December 1960. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
C. Joe is a hitman. They want him to kill someone.
|
Why was mountain climbing prohibited on the Eastern Slope during the time?
A. The rocks were shifting too fast and the paths could be confusing
B. The flooding was too substantial
C. They feared the danger of rock slides
D. Rescue missions were too dangerous due to the sand storms
|
THE GREAT NEBRASKA SEA By ALLAN DANZIG Illustrated by WOOD [Transcriber's Note: This etext was produced from Galaxy Magazine August 1963. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] It has happened a hundred times in the long history of Earth—and, sooner or later, will happen again! Everyone—all the geologists, at any rate—had known about the Kiowa Fault for years. That was before there was anything very interesting to know about it. The first survey of Colorado traced its course north and south in the narrow valley of Kiowa Creek about twenty miles east of Denver; it extended south to the Arkansas River. And that was about all even the professionals were interested in knowing. There was never so much as a landslide to bring the Fault to the attention of the general public. It was still a matter of academic interest when in the late '40s geologists speculated on the relationship between the Kiowa Fault and the Conchas Fault farther south, in New Mexico, and which followed the Pecos as far south as Texas. Nor was there much in the papers a few years later when it was suggested that the Niobrara Fault (just inside and roughly parallel to the eastern border of Wyoming) was a northerly extension of the Kiowa. By the mid sixties it was definitely established that the three Faults were in fact a single line of fissure in the essential rock, stretching almost from the Canadian border well south of the New Mexico-Texas line. It is not really surprising that it took so long to figure out the connection. The population of the states affected was in places as low as five people per square mile! The land was so dry it seemed impossible that it could ever be used except for sheep-farming. It strikes us today as ironic that from the late '50s there was grave concern about the level of the water table throughout the entire area. The even more ironic solution to the problem began in the summer of 1973. It had been a particularly hot and dry August, and the Forestry Service was keeping an anxious eye out for the fires it knew it could expect. Dense smoke was reported rising above a virtually uninhabited area along Black Squirrel Creek, and a plane was sent out for a report. The report was—no fire at all. The rising cloud was not smoke, but dust. Thousands of cubic feet of dry earth rising lazily on the summer air. Rock slides, they guessed; certainly no fire. The Forestry Service had other worries at the moment, and filed the report. But after a week had gone by, the town of Edison, a good twenty miles away from the slides, was still complaining of the dust. Springs was going dry, too, apparently from underground disturbances. Not even in the Rockies could anyone remember a series of rock slides as bad as this. Newspapers in the mountain states gave it a few inches on the front page; anything is news in late August. And the geologists became interested. Seismologists were reporting unusual activity in the area, tremors too severe to be rock slides. Volcanic activity? Specifically, a dust volcano? Unusual, they knew, but right on the Kiowa Fault—could be. Labor Day crowds read the scientific conjectures with late summer lassitude. Sunday supplements ran four-color artists' conceptions of the possible volcano. "Only Active Volcano in U. S.?" demanded the headlines, and some papers even left off the question mark. It may seem odd that the simplest explanation was practically not mentioned. Only Joseph Schwartzberg, head geographer of the Department of the Interior, wondered if the disturbance might not be a settling of the Kiowa Fault. His suggestion was mentioned on page nine or ten of the Monday newspapers (page 27 of the New York Times ). The idea was not nearly so exciting as a volcano, even a lava-less one, and you couldn't draw a very dramatic picture of it. To excuse the other geologists, it must be said that the Kiowa Fault had never acted up before. It never sidestepped, never jiggled, never, never produced the regular shows of its little sister out in California, which almost daily bounced San Francisco or Los Angeles, or some place in between. The dust volcano was on the face of it a more plausible theory. Still, it was only a theory. It had to be proved. As the tremors grew bigger, along with the affected area, as several towns including Edison were shaken to pieces by incredible earthquakes, whole bus- and plane-loads of geologists set out for Colorado, without even waiting for their university and government department to approve budgets. They found, of course, that Schwartzberg had been perfectly correct. They found themselves on the scene of what was fast becoming the most violent and widespread earthquake North America—probably the world—has ever seen in historic times. To describe it in the simplest terms, land east of the Fault was settling, and at a precipitous rate. Rock scraped rock with a whining roar. Shuddery as a squeaky piece of chalk raked across a blackboard, the noise was deafening. The surfaces of the land east and west of the Fault seemed no longer to have any relation to each other. To the west, tortured rock reared into cliffs. East, where sharp reports and muffled wheezes told of continued buckling and dropping, the earth trembled downward. Atop the new cliffs, which seemed to grow by sudden inches from heaving rubble, dry earth fissured and trembled, sliding acres at a time to fall, smoking, into the bucking, heaving bottom of the depression. There the devastation was even more thorough, if less spectacular. Dry earth churned like mud, and rock shards weighing tons bumped and rolled about like pebbles as they shivered and cracked into pebbles themselves. "It looks like sand dancing in a child's sieve," said the normally impassive Schwartzberg in a nationwide broadcast from the scene of disaster. "No one here has ever seen anything like it." And the landslip was growing, north and south along the Fault. "Get out while you can," Schwartzberg urged the population of the affected area. "When it's over you can come back and pick up the pieces." But the band of scientists who had rallied to his leadership privately wondered if there would be any pieces. The Arkansas River, at Avondale and North Avondale, was sluggishly backing north into the deepening trough. At the rate things were going, there might be a new lake the entire length of El Paso and Pueblo Counties. And, warned Schwartzberg, this might only be the beginning. By 16 September the landslip had crept down the Huerfano River past Cedarwood. Avondale, North Avondale and Boone had totally disappeared. Land west of the Fault was holding firm, though Denver had recorded several small tremors; everywhere east of the Fault, to almost twenty miles away, the now-familiar lurch and steady fall had already sent several thousand Coloradans scurrying for safety. All mountain climbing was prohibited on the Eastern Slope because of the danger of rock slides from minor quakes. The geologists went home to wait. There wasn't much to wait for. The news got worse and worse. The Platte River, now, was creating a vast mud puddle where the town of Orchard had been. Just below Masters, Colorado, the river leaped 70-foot cliffs to add to the heaving chaos below. And the cliffs were higher every day as the land beneath them groaned downward in mile-square gulps. As the Fault moved north and south, new areas quivered into unwelcome life. Fields and whole mountainsides moved with deceptive sloth down, down. They danced "like sand in a sieve"; dry, they boiled into rubble. Telephone lines, railroad tracks, roads snapped and simply disappeared. Virtually all east-west land communication was suspended and the President declared a national emergency. By 23 September the Fault was active well into Wyoming on the north, and rapidly approaching the border of New Mexico to the south. Trinchera and Branson were totally evacuated, but even so the over-all death toll had risen above 1,000. Away to the east the situation was quiet but even more ominous. Tremendous fissures opened up perpendicular to the Fault, and a general subsidence of the land was noticeable well into Kansas and Nebraska. The western borders of these states, and soon of the Dakotas and Oklahoma as well, were slowly sinking. On the actual scene of the disaster (or the scenes ; it is impossible to speak of anything this size in the singular) there was a horrifying confusion. Prairie and hill cracked open under intolerable strains as the land shuddered downward in gasps and leaps. Springs burst to the surface in hot geysers and explosions of steam. The downtown section of North Platte, Nebraska, dropped eight feet, just like that, on the afternoon of 4 October. "We must remain calm," declared the Governor of Nebraska. "We must sit this thing out. Be assured that everything possible is being done." But what could be done, with his state dropping straight down at a mean rate of a foot a day? The Fault nicked off the south-east corner of Montana. It worked its way north along the Little Missouri. South, it ripped past Roswell, New Mexico, and tore down the Pecos toward Texas. All the upper reaches of the Missouri were standing puddles by now, and the Red River west of Paris, Texas, had begun to run backward. Soon the Missouri began slowly slipping away westward over the slowly churning land. Abandoning its bed, the river spread uncertainly across farmland and prairie, becoming a sea of mud beneath the sharp new cliffs which rose in rending line, ever taller as the land continued to sink, almost from Canada to the Mexican border. There were virtually no floods, in the usual sense. The water moved too slowly, spread itself with no real direction or force. But the vast sheets of sluggish water and jelly-like mud formed death-traps for the countless refugees now streaming east. Perhaps the North Platte disaster had been more than anyone could take. 193 people had died in that one cave-in. Certainly by 7 October it had to be officially admitted that there was an exodus of epic proportion. Nearly two million people were on the move, and the U. S. was faced with a gigantic wave of refugees. Rails, roads and air-lanes were jammed with terrified hordes who had left everything behind to crowd eastward. All through October hollow-eyed motorists flocked into Tulsa, Topeka, Omaha, Sioux Falls and Fargo. St. Louis was made distributing center for emergency squads which flew everywhere with milk for babies and dog food for evacuating pets. Gasoline trucks boomed west to meet the demand for gas, but once inside the "zone of terror," as the newspapers now called it, they found their route blocked by eastbound cars on the wrong side of the road. Shops left by their fleeing owners were looted by refugees from further west; an American Airlines plane was wrecked by a mob of would-be passengers in Bismarck, North Dakota. Federal and State troops were called out, but moving two million people was not to be done in an orderly way. And still the landslip grew larger. The new cliffs gleamed in the autumn sunshine, growing higher as the land beneath them continued its inexorable descent. On 21 October, at Lubbock, Texas, there was a noise variously described as a hollow roar, a shriek and a deep musical vibration like a church bell. It was simply the tortured rock of the substrata giving way. The second phase of the national disaster was beginning. The noise traveled due east at better than 85 miles per hour. In its wake the earth to the north "just seemed to collapse on itself like a punctured balloon," read one newspaper report. "Like a cake that's failed," said a Texarkana housewife who fortunately lived a block south of Thayer Street, where the fissure raced through. There was a sigh and a great cloud of dust, and Oklahoma subsided at the astounding rate of about six feet per hour. At Biloxi, on the Gulf, there had been uneasy shufflings under foot all day. "Not tremors, exactly," said the captain of a fishing boat which was somehow to ride out the coming flood, "but like as if the land wanted to be somewhere else." Everyone in doomed Biloxi would have done well to have been somewhere else that evening. At approximately 8:30 p.m. the town shuddered, seemed to rise a little like the edge of a hall carpet caught in a draft, and sank. So did the entire Mississippi and Alabama coast, at about the same moment. The tidal wave which was to gouge the center from the U. S. marched on the land. From the north shore of Lake Ponchartrain to the Appalachicola River in Florida, the Gulf coast simply disappeared. Gulfport, Biloxi, Mobile, Pensacola, Panama City: 200 miles of shoreline vanished, with over two and a half million people. An hour later a wall of water had swept over every town from Dothan, Alabama, to Bogalusa on the Louisiana-Mississippi border. "We must keep panic from our minds," said the Governor of Alabama in a radio message delivered from a hastily arranged all-station hookup. "We of the gallant southland have faced and withstood invasion before." Then, as ominous creakings and groanings of the earth announced the approach of the tidal wave, he flew out of Montgomery half an hour before the town disappeared forever. One head of the wave plunged north, eventually to spend itself in the hills south of Birmingham. The main sweep followed the lowest land. Reaching west, it swallowed Vicksburg and nicked the corner of Louisiana. The whole of East Carroll Parish was scoured from the map. The Mississippi River now ended at about Eudora, Arkansas, and minute by minute the advancing flood bit away miles of river bed, swelling north. Chicot, Jennie, Lake Village, Arkansas City, Snow Lake, Elaine, Helena and Memphis felt the tremors. The tormented city shuddered through the night. The earth continued its descent, eventually tipping 2-1/2 degrees down to the west. The "Memphis Tilt" is today one of the unique and charming characteristics of the gracious Old Town, but during the night of panic Memphis residents were sure they were doomed. South and west the waters carved deeply into Arkansas and Oklahoma. By morning it was plain that all of Arkansas was going under. Waves advanced on Little Rock at almost 100 miles an hour, new crests forming, overtopping the wave's leading edge as towns, hills and the thirst of the soil temporarily broke the furious charge. Washington announced the official hope that the Ozarks would stop the wild gallop of the unleashed Gulf, for in northwest Arkansas the land rose to over 2,000 feet. But nothing could save Oklahoma. By noon the water reached clutching fingers around Mt. Scott and Elk Mountain, deluging Hobart and almost all of Greer County. Despite hopeful announcements that the wave was slowing, had virtually stopped after inundating Oklahoma City, was being swallowed up in the desert near Amarillo, the wall of water continued its advance. For the land was still sinking, and the floods were constantly replenished from the Gulf. Schwartzberg and his geologists advised the utmost haste in evacuating the entire area between Colorado and Missouri, from Texas to North Dakota. Lubbock, Texas, went under. On a curling reflex the tidal wave blotted out Sweetwater and Big Spring. The Texas panhandle disappeared in one great swirl. Whirlpools opened. A great welter of smashed wood and human debris was sucked under, vomited up and pounded to pieces. Gulf-water crashed on the cliffs of New Mexico and fell back on itself in foam. Would-be rescuers on the cliffs along what had been the west bank of the Pecos River afterwards recalled the hiss and scream like tearing silk as the water broke furiously on the newly exposed rock. It was the most terrible sound they had ever heard. "We couldn't hear any shouts, of course, not that far away and with all the noise," said Dan Weaver, Mayor of Carlsbad. "But we knew there were people down there. When the water hit the cliffs, it was like a collision between two solid bodies. We couldn't see for over an hour, because of the spray." Salt spray. The ocean had come to New Mexico. The cliffs proved to be the only effective barrier against the westward march of the water, which turned north, gouging out lumps of rock and tumbling down blocks of earth onto its own back. In places scoops of granite came out like ice cream. The present fishing town of Rockport, Colorado, is built on a harbor created in such a way. The water had found its farthest westering. But still it poured north along the line of the original Fault. Irresistible fingers closed on Sterling, Colorado, on Sidney, Nebraska, on Hot Springs, South Dakota. The entire tier of states settled, from south to north, down to its eventual place of stability one thousand feet below the level of the new sea. Memphis was by now a seaport. The Ozarks, islands in a mad sea, formed precarious havens for half-drowned humanity. Waves bit off a corner of Missouri, flung themselves on Wichita. Topeka, Lawrence and Belleville were the last Kansas towns to disappear. The Governor of Kansas went down with his State. Daniel Bernd of Lincoln, Nebraska, was washed up half-drowned in a cove of the Wyoming cliffs, having been sucked from one end of vanished Nebraska to the other. Similar hair-breadth escapes were recounted on radio and television. Virtually the only people saved out of the entire population of Pierre, South Dakota were the six members of the Creeth family. Plucky Timothy Creeth carried and dragged his aged parents to the loft of their barn on the outskirts of town. His brother Geoffrey brought along the younger children and what provisions they could find—"Mostly a ham and about half a ton of vanilla cookies," he explained to his eventual rescuers. The barn, luckily collapsing in the vibrations as the waves bore down on them, became an ark in which they rode out the disaster. "We must of played cards for four days straight," recalled genial Mrs. Creeth when she afterwards appeared on a popular television spectacular. Her rural good-humor undamaged by an ordeal few women can ever have been called on to face, she added, "We sure wondered why flushes never came out right. Jimanettly, we'd left the king of hearts behind, in the rush!" But such lightheartedness and such happy endings were by no means typical. The world could only watch aghast as the water raced north under the shadow of the cliffs which occasionally crumbled, roaring, into the roaring waves. Day by day the relentless rush swallowed what had been dusty farmland, cities and towns. Some people were saved by the helicopters which flew mercy missions just ahead of the advancing waters. Some found safety in the peaks of western Nebraska and the Dakotas. But when the waters came to rest along what is roughly the present shoreline of our inland sea, it was estimated that over fourteen million people had lost their lives. No one could even estimate the damage to property; almost the entirety of eight states, and portions of twelve others, had simply vanished from the heart of the North American continent forever. It was in such a cataclysmic birth that the now-peaceful Nebraska Sea came to America. Today, nearly one hundred years after the unprecedented—and happily unrepeated—disaster, it is hard to remember the terror and despair of those weeks in October and November, 1973. It is inconceivable to think of the United States without its beautiful and economically essential curve of interior ocean. Two-thirds as long as the Mediterranean, it graduates from the warm waters of the Gulf of Mexico through the equally blue waves of the Mississippi Bight, becoming cooler and greener north and west of the pleasant fishing isles of the Ozark Archipelago, finally shading into the gray-green chop of the Gulf of Dakota. What would the United States have become without the 5600-mile coastline of our inland sea? It is only within the last twenty years that any but the topmost layer of water has cleared sufficiently to permit a really extensive fishing industry. Mud still held in suspension by the restless waves will not precipitate fully even in our lifetimes. Even so, the commercial fisheries of Missouri and Wyoming contribute no small part to the nation's economy. Who can imagine what the middle west must have been like before the amelioration of climate brought about by the proximity of a warm sea? The now-temperate state of Minnesota (to say nothing of the submerged Dakotas) must have been Siberian. From contemporary accounts Missouri, our second California, was unbelievably muggy, almost uninhabitable during the summer months. Our climate today, from Ohio and North Carolina to the rich fields of New Mexico and the orchards of Montana, is directly ameliorated by the marine heart of the continent. Who today could imagine the United States without the majestic sea-cliffs in stately parade from New Mexico to Montana? The beaches of Wyoming, the American Riviera, where fruit trees grow almost to the water's edge? Or incredible Colorado, where the morning skier is the afternoon bather, thanks to the monorail connecting the highest peaks with the glistening white beaches? Of course there have been losses to balance slightly these strong gains. The Mississippi was, before 1973, one of the great rivers of the world. Taken together with its main tributary, the Missouri, it vied favorably with such giant systems as the Amazon and the Ganges. Now, ending as it does at Memphis and drawing its water chiefly from the Appalachian Mountains, it is only a slight remnant of what it was. And though the Nebraska Sea today carries many times the tonnage of shipping in its ceaseless traffic, we have lost the old romance of river shipping. We may only guess what it was like when we look upon the Ohio and the truncated Mississippi. And transcontinental shipping is somewhat more difficult, with trucks and the freight-railroads obliged to take the sea-ferries across the Nebraska Sea. We shall never know what the United States was like with its numerous coast-to-coast highways busy with trucks and private cars. Still, the ferry ride is certainly a welcome break after days of driving, and for those who wish a glimpse of what it must have been like, there is always the Cross-Canada Throughway and the magnificent U. S. Highway 73 looping north through Minnesota and passing through the giant port of Alexis, North Dakota, shipping center for the wheat of Manitoba and crossroad of a nation. The political situation has long been a thorny problem. Only tattered remnants of the eight submerged states remained after the flood, but none of them wanted to surrender its autonomy. The tiny fringe of Kansas seemed, for a time, ready to merge with contiguous Missouri, but following the lead of the Arkansas Forever faction, the remaining population decided to retain political integrity. This has resulted in the continuing anomaly of the seven "fringe States" represented in Congress by the usual two Senators each, though the largest of them is barely the size of Connecticut and all are economically indistinguishable from their neighboring states. Fortunately it was decided some years ago that Oklahoma, only one of the eight to have completely disappeared, could not in any sense be considered to have a continuing political existence. So, though there are still families who proudly call themselves Oklahomans, and the Oklahoma Oil Company continues to pump oil from its submerged real estate, the state has in fact disappeared from the American political scene. But this is by now no more than a petty annoyance, to raise a smile when the talk gets around to the question of State's Rights. Not even the tremendous price the country paid for its new sea—fourteen million dead, untold property destroyed—really offsets the asset we enjoy today. The heart of the continent, now open to the shipping of the world, was once dry and land-locked, cut off from the bustle of trade and the ferment of world culture. It would indeed seem odd to an American of the '50s or '60s of the last century to imagine sailors from the merchant fleets of every nation walking the streets of Denver, fresh ashore at Newport, only fifteen miles away. Or to imagine Lincoln, Fargo, Kansas City and Dallas as world ports and great manufacturing centers. Utterly beyond their ken would be Roswell, New Mexico; Benton, Wyoming; Westport, Missouri, and the other new ports of over a million inhabitants each which have developed on the new harbors of the inland sea. Unimaginable too would have been the general growth of population in the states surrounding the new sea. As the water tables rose and manufacturing and trade moved in to take advantage of the just-created axis of world communication, a population explosion was touched off of which we are only now seeing the diminution. This new westering is to be ranked with the first surge of pioneers which created the American west. But what a difference! Vacation paradises bloom, a new fishing industry thrives; her water road is America's main artery of trade, and fleets of all the world sail ... where once the prairie schooner made its laborious and dusty way west!
|
C. They feared the danger of rock slides
|
What improvements they did for DMN?
|
### Introduction
Neural network based methods have made tremendous progress in image and text classification BIBREF0 , BIBREF1 . However, only recently has progress been made on more complex tasks that require logical reasoning. This success is based in part on the addition of memory and attention components to complex neural networks. For instance, memory networks BIBREF2 are able to reason over several facts written in natural language or (subject, relation, object) triplets. Attention mechanisms have been successful components in both machine translation BIBREF3 , BIBREF4 and image captioning models BIBREF5 . The dynamic memory network BIBREF6 (DMN) is one example of a neural network model that has both a memory component and an attention mechanism. The DMN yields state of the art results on question answering with supporting facts marked during training, sentiment analysis, and part-of-speech tagging. We analyze the DMN components, specifically the input module and memory module, to improve question answering. We propose a new input module which uses a two level encoder with a sentence reader and input fusion layer to allow for information flow between sentences. For the memory, we propose a modification to gated recurrent units (GRU) BIBREF7 . The new GRU formulation incorporates attention gates that are computed using global knowledge over the facts. Unlike before, the new DMN+ model does not require that supporting facts (i.e. the facts that are relevant for answering a particular question) are labeled during training. The model learns to select the important facts from a larger set. In addition, we introduce a new input module to represent images. This module is compatible with the rest of the DMN architecture and its output is fed into the memory module. We show that the changes in the memory module that improved textual question answering also improve visual question answering. Both tasks are illustrated in Fig. 1 . ### Dynamic Memory Networks
We begin by outlining the DMN for question answering and the modules as presented in BIBREF6 . The DMN is a general architecture for question answering (QA). It is composed of modules that allow different aspects such as input representations or memory components to be analyzed and improved independently. The modules, depicted in Fig. 1 , are as follows: Input Module: This module processes the input data about which a question is being asked into a set of vectors termed facts, represented as $F=[f_1,\hdots ,f_N]$ , where $N$ is the total number of facts. These vectors are ordered, resulting in additional information that can be used by later components. For text QA in BIBREF6 , the module consists of a GRU over the input words. As the GRU is used in many components of the DMN, it is useful to provide the full definition. For each time step $i$ with input $x_i$ and previous hidden state $h_{i-1}$ , we compute the updated hidden state $h_i = GRU(x_i,h_{i-1})$ by $$u_i &=& \sigma \left(W^{(u)}x_{i} + U^{(u)} h_{i-1} + b^{(u)} \right)\\
r_i &=& \sigma \left(W^{(r)}x_{i} + U^{(r)} h_{i-1} + b^{(r)} \right)\\
\tilde{h}_i &=& \tanh \left(Wx_{i} + r_i \circ U h_{i-1} + b^{(h)}\right)\\
h_i &=& u_i\circ \tilde{h}_i + (1-u_i) \circ h_{i-1}$$ (Eq. 2) where $\sigma $ is the sigmoid activation function, $\circ $ is an element-wise product, $W^{(z)}, W^{(r)}, W \in \mathbb {R}^{n_H \times n_I}$ , $U^{(z)}, U^{(r)}, U \in \mathbb {R}^{n_H \times n_H}$ , $n_H$ is the hidden size, and $n_I$ is the input size. Question Module: This module computes a vector representation $q$ of the question, where $q \in \mathbb {R}^{n_H}$ is the final hidden state of a GRU over the words in the question. Episodic Memory Module: Episode memory aims to retrieve the information required to answer the question $q$ from the input facts. To improve our understanding of both the question and input, especially if questions require transitive reasoning, the episode memory module may pass over the input multiple times, updating episode memory after each pass. We refer to the episode memory on the $t^{th}$ pass over the inputs as $m^t$ , where $m^t \in \mathbb {R}^{n_H}$ , the initial memory vector is set to the question vector: $m^0 = q$ . The episodic memory module consists of two separate components: the attention mechanism and the memory update mechanism. The attention mechanism is responsible for producing a contextual vector $c^t$ , where $c^t \in \mathbb {R}^{n_H}$ is a summary of relevant input for pass $t$ , with relevance inferred by the question $q$ and previous episode memory $m^{t-1}$ . The memory update mechanism is responsible for generating the episode memory $m^t$ based upon the contextual vector $c^t$ and previous episode memory $m^{t-1}$ . By the final pass $T$ , the episodic memory $m^T$ should contain all the information required to answer the question $c^t \in \mathbb {R}^{n_H}$0 . Answer Module: The answer module receives both $q$ and $m^T$ to generate the model's predicted answer. For simple answers, such as a single word, a linear layer with softmax activation may be used. For tasks requiring a sequence output, an RNN may be used to decode $a = [q ; m^T]$ , the concatenation of vectors $q$ and $m^T$ , to an ordered set of tokens. The cross entropy error on the answers is used for training and backpropagated through the entire network. ### Improved Dynamic Memory Networks: DMN+
We propose and compare several modeling choices for two crucial components: input representation, attention mechanism and memory update. The final DMN+ model obtains the highest accuracy on the bAbI-10k dataset without supporting facts and the VQA dataset BIBREF8 . Several design choices are motivated by intuition and accuracy improvements on that dataset. ### Input Module for Text QA
In the DMN specified in BIBREF6 , a single GRU is used to process all the words in the story, extracting sentence representations by storing the hidden states produced at the end of sentence markers. The GRU also provides a temporal component by allowing a sentence to know the content of the sentences that came before them. Whilst this input module worked well for bAbI-1k with supporting facts, as reported in BIBREF6 , it did not perform well on bAbI-10k without supporting facts (Sec. "Model Analysis" ). We speculate that there are two main reasons for this performance disparity, all exacerbated by the removal of supporting facts. First, the GRU only allows sentences to have context from sentences before them, but not after them. This prevents information propagation from future sentences. Second, the supporting sentences may be too far away from each other on a word level to allow for these distant sentences to interact through the word level GRU. Input Fusion Layer For the DMN+, we propose replacing this single GRU with two different components. The first component is a sentence reader, responsible only for encoding the words into a sentence embedding. The second component is the input fusion layer, allowing for interactions between sentences. This resembles the hierarchical neural auto-encoder architecture of BIBREF9 and allows content interaction between sentences. We adopt the bi-directional GRU for this input fusion layer because it allows information from both past and future sentences to be used. As gradients do not need to propagate through the words between sentences, the fusion layer also allows for distant supporting sentences to have a more direct interaction. Fig. 2 shows an illustration of an input module, where a positional encoder is used for the sentence reader and a bi-directional GRU is adopted for the input fusion layer. Each sentence encoding $f_i$ is the output of an encoding scheme taking the word tokens $[w^i_1, \hdots , w^i_{M_i}]$ , where $M_i$ is the length of the sentence. The sentence reader could be based on any variety of encoding schemes. We selected positional encoding described in BIBREF10 to allow for a comparison to their work. GRUs and LSTMs were also considered but required more computational resources and were prone to overfitting if auxiliary tasks, such as reconstructing the original sentence, were not used. For the positional encoding scheme, the sentence representation is produced by $f_i = \sum ^{j=1}_M l_j \circ w^i_j$ , where $\circ $ is element-wise multiplication and $l_j$ is a column vector with structure $l_{jd} = (1 - j / M) - (d / D) (1 - 2j / M)$ , where $d$ is the embedding index and $D$ is the dimension of the embedding. The input fusion layer takes these input facts and enables an information exchange between them by applying a bi-directional GRU. $$\overrightarrow{f_i} = GRU_{fwd}(f_i, \overrightarrow{f_{i-1}}) \\
\overleftarrow{f_{i}} = GRU_{bwd}(f_{i}, \overleftarrow{f_{i+1}}) \\
\overleftrightarrow{f_i} = \overleftarrow{f_i} + \overrightarrow{f_i}$$ (Eq. 5) where $f_i$ is the input fact at timestep $i$ , $ \overrightarrow{f_i}$ is the hidden state of the forward GRU at timestep $i$ , and $\overleftarrow{f_i}$ is the hidden state of the backward GRU at timestep $i$ . This allows contextual information from both future and past facts to impact $\overleftrightarrow{f_i}$ . We explored a variety of encoding schemes for the sentence reader, including GRUs, LSTMs, and the positional encoding scheme described in BIBREF10 . For simplicity and speed, we selected the positional encoding scheme. GRUs and LSTMs were also considered but required more computational resources and were prone to overfitting if auxiliary tasks, such as reconstructing the original sentence, were not used. ### Input Module for VQA
To apply the DMN to visual question answering, we introduce a new input module for images. The module splits an image into small local regions and considers each region equivalent to a sentence in the input module for text. The input module for VQA is composed of three parts, illustrated in Fig. 3 : local region feature extraction, visual feature embedding, and the input fusion layer introduced in Sec. "Input Module for Text QA" . Local region feature extraction: To extract features from the image, we use a convolutional neural network BIBREF0 based upon the VGG-19 model BIBREF11 . We first rescale the input image to $448 \times 448$ and take the output from the last pooling layer which has dimensionality $d = 512 \times 14 \times 14$ . The pooling layer divides the image into a grid of $14 \times 14$ , resulting in 196 local regional vectors of $d = 512$ . Visual feature embedding: As the VQA task involves both image features and text features, we add a linear layer with tanh activation to project the local regional vectors to the textual feature space used by the question vector $q$ . Input fusion layer: The local regional vectors extracted from above do not yet have global information available to them. Without global information, their representational power is quite limited, with simple issues like object scaling or locational variance causing accuracy problems. To solve this, we add an input fusion layer similar to that of the textual input module described in Sec. "Input Module for Text QA" . First, to produce the input facts $F$ , we traverse the image in a snake like fashion, as seen in Figure 3 . We then apply a bi-directional GRU over these input facts $F$ to produce the globally aware input facts $\overleftrightarrow{F}$ . The bi-directional GRU allows for information propagation from neighboring image patches, capturing spatial information. ### The Episodic Memory Module
The episodic memory module, as depicted in Fig. 4 , retrieves information from the input facts $\overleftrightarrow{F} = [\overleftrightarrow{f_1}, \hdots , \overleftrightarrow{f_N}]$ provided to it by focusing attention on a subset of these facts. We implement this attention by associating a single scalar value, the attention gate $g^t_i$ , with each fact $\overleftrightarrow{f}_i$ during pass $t$ . This is computed by allowing interactions between the fact and both the question representation and the episode memory state. $$z^t_i &=& [\overleftrightarrow{f_i} \circ q; \overleftrightarrow{f_i} \circ m^{t-1}; \vert \overleftrightarrow{f_i} - q \vert ; \vert \overleftrightarrow{f_i} - m^{t-1} \vert ] \\
Z^t_i &=& W^{(2)} \tanh \left(W^{(1)}z^t_i + b^{(1)} \right)+ b^{(2)} \\
g^t_i &=& \frac{\exp (Z^t_i)}{\sum _{k=1}^{M_i} \exp (Z^t_k)} $$ (Eq. 10) where $\overleftrightarrow{f_i}$ is the $i^{th}$ fact, $m^{t-1}$ is the previous episode memory, $q$ is the original question, $\circ $ is the element-wise product, $|\cdot |$ is the element-wise absolute value, and $;$ represents concatenation of the vectors. The DMN implemented in BIBREF6 involved a more complex set of interactions within $z$ , containing the additional terms $[f; m^{t-1}; q; f^T W^{(b)} q; f^T W^{(b)} m^{t-1}]$ . After an initial analysis, we found these additional terms were not required. Attention Mechanism Once we have the attention gate $g^t_i$ we use an attention mechanism to extract a contextual vector $c^t$ based upon the current focus. We focus on two types of attention: soft attention and a new attention based GRU. The latter improves performance and is hence the final modeling choice for the DMN+. Soft attention: Soft attention produces a contextual vector $c^t$ through a weighted summation of the sorted list of vectors $\overleftrightarrow{F}$ and corresponding attention gates $g_i^t$ : $c^t = \sum _{i=1}^N g^t_i \overleftrightarrow{f}_i$ This method has two advantages. First, it is easy to compute. Second, if the softmax activation is spiky it can approximate a hard attention function by selecting only a single fact for the contextual vector whilst still being differentiable. However the main disadvantage to soft attention is that the summation process loses both positional and ordering information. Whilst multiple attention passes can retrieve some of this information, this is inefficient. Attention based GRU: For more complex queries, we would like for the attention mechanism to be sensitive to both the position and ordering of the input facts $\overleftrightarrow{F}$ . An RNN would be advantageous in this situation except they cannot make use of the attention gate from Equation . We propose a modification to the GRU architecture by embedding information from the attention mechanism. The update gate $u_i$ in Equation 2 decides how much of each dimension of the hidden state to retain and how much should be updated with the transformed input $x_i$ from the current timestep. As $u_i$ is computed using only the current input and the hidden state from previous timesteps, it lacks any knowledge from the question or previous episode memory. By replacing the update gate $u_i$ in the GRU (Equation 2 ) with the output of the attention gate $g^t_i$ (Equation ) in Equation , the GRU can now use the attention gate for updating its internal state. This change is depicted in Fig 5 . $$h_i &=& g^t_i \circ \tilde{h}_i + (1-g^t_i) \circ h_{i-1}$$ (Eq. 12) An important consideration is that $g^t_i$ is a scalar, generated using a softmax activation, as opposed to the vector $u_i \in \mathbb {R}^{n_H}$ , generated using a sigmoid activation. This allows us to easily visualize how the attention gates activate over the input, later shown for visual QA in Fig. 6 . Though not explored, replacing the softmax activation in Equation with a sigmoid activation would result in $g^t_i \in \mathbb {R}^{n_H}$ . To produce the contextual vector $c^t$ used for updating the episodic memory state $m^t$ , we use the final hidden state of the attention based GRU. Episode Memory Updates After each pass through the attention mechanism, we wish to update the episode memory $m^{t-1}$ with the newly constructed contextual vector $c^t$ , producing $m^t$ . In the DMN, a GRU with the initial hidden state set to the question vector $q$ is used for this purpose. The episodic memory for pass $t$ is computed by $$m^t = GRU(c^t, m^{t-1})$$ (Eq. 13) The work of BIBREF10 suggests that using different weights for each pass through the episodic memory may be advantageous. When the model contains only one set of weights for all episodic passes over the input, it is referred to as a tied model, as in the “Mem Weights” row in Table 1 . Following the memory update component used in BIBREF10 and BIBREF12 we experiment with using a ReLU layer for the memory update, calculating the new episode memory state by $$m^t = ReLU\left(W^t [m^{t-1} ; c^t ; q] + b\right)$$ (Eq. 14) where $;$ is the concatenation operator, $W^t \in \mathbb {R}^{n_H \times n_H}$ , $b \in \mathbb {R}^{n_H}$ , and $n_H$ is the hidden size. The untying of weights and using this ReLU formulation for the memory update improves accuracy by another 0.5% as shown in Table 1 in the last column. The final output of the memory network is passed to the answer module as in the original DMN. ### Related Work
The DMN is related to two major lines of recent work: memory and attention mechanisms. We work on both visual and textual question answering which have, until now, been developed in separate communities. Neural Memory Models The earliest recent work with a memory component that is applied to language processing is that of memory networks BIBREF2 which adds a memory component for question answering over simple facts. They are similar to DMNs in that they also have input, scoring, attention and response mechanisms. However, unlike the DMN their input module computes sentence representations independently and hence cannot easily be used for other tasks such as sequence labeling. Like the original DMN, this memory network requires that supporting facts are labeled during QA training. End-to-end memory networks BIBREF10 do not have this limitation. In contrast to previous memory models with a variety of different functions for memory attention retrieval and representations, DMNs BIBREF6 have shown that neural sequence models can be used for input representation, attention and response mechanisms. Sequence models naturally capture position and temporality of both the inputs and transitive reasoning steps. Neural Attention Mechanisms Attention mechanisms allow neural network models to use a question to selectively pay attention to specific inputs. They can benefit image classification BIBREF13 , generating captions for images BIBREF5 , among others mentioned below, and machine translation BIBREF14 , BIBREF3 , BIBREF4 . Other recent neural architectures with memory or attention which have proposed include neural Turing machines BIBREF15 , neural GPUs BIBREF16 and stack-augmented RNNs BIBREF17 . Question Answering in NLP Question answering involving natural language can be solved in a variety of ways to which we cannot all do justice. If the potential input is a large text corpus, QA becomes a combination of information retrieval and extraction BIBREF18 . Neural approaches can include reasoning over knowledge bases, BIBREF19 , BIBREF20 or directly via sentences for trivia competitions BIBREF21 . Visual Question Answering (VQA) In comparison to QA in NLP, VQA is still a relatively young task that is feasible only now that objects can be identified with high accuracy. The first large scale database with unconstrained questions about images was introduced by BIBREF8 . While VQA datasets existed before they did not include open-ended, free-form questions about general images BIBREF22 . Others are were too small to be viable for a deep learning approach BIBREF23 . The only VQA model which also has an attention component is the stacked attention network BIBREF24 . Their work also uses CNN based features. However, unlike our input fusion layer, they use a single layer neural network to map the features of each patch to the dimensionality of the question vector. Hence, the model cannot easily incorporate adjacency of local information in its hidden state. A model that also uses neural modules, albeit logically inspired ones, is that by BIBREF25 who evaluate on knowledgebase reasoning and visual question answering. We compare directly to their method on the latter task and dataset. Related to visual question answering is the task of describing images with sentences BIBREF26 . BIBREF27 used deep learning methods to map images and sentences into the same space in order to describe images with sentences and to find images that best visualize a sentence. This was the first work to map both modalities into a joint space with deep learning methods, but it could only select an existing sentence to describe an image. Shortly thereafter, recurrent neural networks were used to generate often novel sentences based on images BIBREF28 , BIBREF29 , BIBREF30 , BIBREF5 . ### Datasets
To analyze our proposed model changes and compare our performance with other architectures, we use three datasets. ### bAbI-10k
For evaluating the DMN on textual question answering, we use bAbI-10k English BIBREF31 , a synthetic dataset which features 20 different tasks. Each example is composed of a set of facts, a question, the answer, and the supporting facts that lead to the answer. The dataset comes in two sizes, referring to the number of training examples each task has: bAbI-1k and bAbI-10k. The experiments in BIBREF10 found that their lowest error rates on the smaller bAbI-1k dataset were on average three times higher than on bAbI-10k. ### DAQUAR-ALL visual dataset
The DAtaset for QUestion Answering on Real-world images (DAQUAR) BIBREF23 consists of 795 training images and 654 test images. Based upon these images, 6,795 training questions and 5,673 test questions were generated. Following the previously defined experimental method, we exclude multiple word answers BIBREF32 , BIBREF33 . The resulting dataset covers 90% of the original data. The evaluation method uses classification accuracy over the single words. We use this as a development dataset for model analysis (Sec. "Model Analysis" ). ### Visual Question Answering
The Visual Question Answering (VQA) dataset was constructed using the Microsoft COCO dataset BIBREF34 which contained 123,287 training/validation images and 81,434 test images. Each image has several related questions with each question answered by multiple people. This dataset contains 248,349 training questions, 121,512 validation questions, and 244,302 for testing. The testing data was split into test-development, test-standard and test-challenge in BIBREF8 . Evaluation on both test-standard and test-challenge are implemented via a submission system. test-standard may only be evaluated 5 times and test-challenge is only evaluated at the end of the competition. To the best of our knowledge, VQA is the largest and most complex image dataset for the visual question answering task. ### Model Analysis
To understand the impact of the proposed module changes, we analyze the performance of a variety of DMN models on textual and visual question answering datasets. The original DMN (ODMN) is the architecture presented in BIBREF6 without any modifications. DMN2 only replaces the input module with the input fusion layer (Sec. "Input Module for Text QA" ). DMN3, based upon DMN2, replaces the soft attention mechanism with the attention based GRU proposed in Sec. "The Episodic Memory Module" . Finally, DMN+, based upon DMN3, is an untied model, using a unique set of weights for each pass and a linear layer with a ReLU activation to compute the memory update. We report the performance of the model variations in Table 1 . A large improvement to accuracy on both the bAbI-10k textual and DAQUAR visual datasets results from updating the input module, seen when comparing ODMN to DMN2. On both datasets, the input fusion layer improves interaction between distant facts. In the visual dataset, this improvement is purely from providing contextual information from neighboring image patches, allowing it to handle objects of varying scale or questions with a locality aspect. For the textual dataset, the improved interaction between sentences likely helps the path finding required for logical reasoning when multiple transitive steps are required. The addition of the attention GRU in DMN3 helps answer questions where complex positional or ordering information may be required. This change impacts the textual dataset the most as few questions in the visual dataset are likely to require this form of logical reasoning. Finally, the untied model in the DMN+ overfits on some tasks compared to DMN3, but on average the error rate decreases. From these experimental results, we find that the combination of all the proposed model changes results, culminating in DMN+, achieves the highest performance across both the visual and textual datasets. ### Comparison to state of the art using bAbI-10k
We trained our models using the Adam optimizer BIBREF35 with a learning rate of 0.001 and batch size of 128. Training runs for up to 256 epochs with early stopping if the validation loss had not improved within the last 20 epochs. The model from the epoch with the lowest validation loss was then selected. Xavier initialization was used for all weights except for the word embeddings, which used random uniform initialization with range $[-\sqrt{3}, \sqrt{3}]$ . Both the embedding and hidden dimensions were of size $d = 80$ . We used $\ell _2$ regularization on all weights except bias and used dropout on the initial sentence encodings and the answer module, keeping the input with probability $p=0.9$ . The last 10% of the training data on each task was chosen as the validation set. For all tasks, three passes were used for the episodic memory module, allowing direct comparison to other state of the art methods. Finally, we limited the input to the last 70 sentences for all tasks except QA3 for which we limited input to the last 130 sentences, similar to BIBREF10 . On some tasks, the accuracy was not stable across multiple runs. This was particularly problematic on QA3, QA17, and QA18. To solve this, we repeated training 10 times using random initializations and evaluated the model that achieved the lowest validation set loss. Text QA Results We compare our best performing approach, DMN+, to two state of the art question answering architectures: the end to end memory network (E2E) BIBREF10 and the neural reasoner framework (NR) BIBREF12 . Neither approach use supporting facts for training. The end-to-end memory network is a form of memory network BIBREF2 tested on both textual question answering and language modeling. The model features both explicit memory and a recurrent attention mechanism. We select the model from the paper that achieves the lowest mean error over the bAbI-10k dataset. This model utilizes positional encoding for input, RNN-style tied weights for the episode module, and a ReLU non-linearity for the memory update component. The neural reasoner framework is an end-to-end trainable model which features a deep architecture for logical reasoning and an interaction-pooling mechanism for allowing interaction over multiple facts. While the neural reasoner framework was only tested on QA17 and QA19, these were two of the most challenging question types at the time. In Table 2 we compare the accuracy of these question answering architectures, both as mean error and error on individual tasks. The DMN+ model reduces mean error by 1.4% compared to the the end-to-end memory network, achieving a new state of the art for the bAbI-10k dataset. One notable deficiency in our model is that of QA16: Basic Induction. In BIBREF10 , an untied model using only summation for memory updates was able to achieve a near perfect error rate of $0.4$ . When the memory update was replaced with a linear layer with ReLU activation, the end-to-end memory network's overall mean error decreased but the error for QA16 rose sharply. Our model experiences the same difficulties, suggesting that the more complex memory update component may prevent convergence on certain simpler tasks. The neural reasoner model outperforms both the DMN and end-to-end memory network on QA17: Positional Reasoning. This is likely as the positional reasoning task only involves minimal supervision - two sentences for input, yes/no answers for supervision, and only 5,812 unique examples after removing duplicates from the initial 10,000 training examples. BIBREF12 add an auxiliary task of reconstructing both the original sentences and question from their representations. This auxiliary task likely improves performance by preventing overfitting. ### Comparison to state of the art using VQA
For the VQA dataset, each question is answered by multiple people and the answers may not be the same, the generated answers are evaluated using human consensus. For each predicted answer $a_i$ for the $i_{th}$ question with target answer set $T^{i}$ , the accuracy of VQA: $Acc_{VQA} = \frac{1}{N}\sum _{i=1}^Nmin(\frac{\sum _{t\in T^i}{1}_{(a_i==t)}}{3},1)$ where ${1}_{(\cdot )}$ is the indicator function. Simply put, the answer $a_i$ is only 100 $\%$ accurate if at least 3 people provide that exact answer. Training Details We use the Adam optimizer BIBREF35 with a learning rate of 0.003 and batch size of 100. Training runs for up to 256 epochs with early stopping if the validation loss has not improved in the last 10 epochs. For weight initialization, we sampled from a random uniform distribution with range $[-0.08, 0.08]$ . Both the word embedding and hidden layers were vectors of size $d=512$ . We apply dropout on the initial image output from the VGG convolutional neural network BIBREF11 as well as the input to the answer module, keeping input with probability $p=0.5$ . Results and Analysis The VQA dataset is composed of three question domains: Yes/No, Number, and Other. This enables us to analyze the performance of the models on various tasks that require different reasoning abilities. The comparison models are separated into two broad classes: those that utilize a full connected image feature for classification and those that perform reasoning over multiple small image patches. Only the SAN and DMN approach use small image patches, while the rest use the fully-connected whole image feature approach. Here, we show the quantitative and qualitative results in Table 3 and Fig. 6 , respectively. The images in Fig. 6 illustrate how the attention gate $g^t_i$ selectively activates over relevant portions of the image according to the query. In Table 3 , our method outperforms baseline and other state-of-the-art methods across all question domains (All) in both test-dev and test-std, and especially for Other questions, achieves a wide margin compared to the other architectures, which is likely as the small image patches allow for finely detailed reasoning over the image. However, the granularity offered by small image patches does not always offer an advantage. The Number questions may be not solvable for both the SAN and DMN architectures, potentially as counting objects is not a simple task when an object crosses image patch boundaries. ### Conclusion
We have proposed new modules for the DMN framework to achieve strong results without supervision of supporting facts. These improvements include the input fusion layer to allow interactions between input facts and a novel attention based GRU that allows for logical reasoning over ordered inputs. Our resulting model obtains state of the art results on both the VQA dataset and the bAbI-10k text question-answering dataset, proving the framework can be generalized across input domains. Figure 1. Question Answering over text and images using a Dynamic Memory Network. Figure 2. The input module with a “fusion layer”, where the sentence reader encodes the sentence and the bi-directional GRU allows information to flow between sentences. Figure 3. VQA input module to represent images for the DMN. Figure 5. (a) The traditional GRU model, and (b) the proposed attention-based GRU model Figure 4. The episodic memory module of the DMN+ when using two passes. The ←→ F is the output of the input module. Table 2. Test error rates of various model architectures on tasks from the the bAbI English 10k dataset. E2E = End-To-End Memory Network results from Sukhbaatar et al. (2015). NR = Neural Reasoner with original auxiliary task from Peng et al. (2015). DMN+ and E2E achieve an error of 0 on bAbI question sets (1,4,10,12,13,15,20). Table 1. Test error rates of various model architectures on the bAbI-10k dataset, and accuracy performance on the DAQUAR-ALL visual dataset. The skipped bAbI questions (1,4,11,12,13,15,19) achieved 0 error across all models. Table 3. Performance of various architectures and approaches on VQA test-dev and test-standard data. Baseline only uses the spatial mean of the last pooling layer without input fusion and episoidic memory; VQA numbers are from Antol et al. (2015); ACK Wu et al. (2015); iBOWIMG -Zhou et al. (2015); DPPnet - Noh et al. (2015); D-NMN - Andreas et al. (2016); SMem-VQA -Xu & Saenko (2015); SAN -Yang et al. (2015) Figure 6. Examples of qualitative results of attention for VQA. The original images are shown on the left. On the right we show how the attention gate gti activates given one pass over the image and query. White regions are the most active. Answers are given by the DMN+.
|
the new DMN+ model does not require that supporting facts (i.e. the facts that are relevant for answering a particular question) are labeled during training., In addition, we introduce a new input module to represent images.
|
What about Haron didn't excite Matilda?
A. he was egotistical
B. he lived nearby
C. his physical appearance
D. his name
|
PEN PAL Illustrated by DON SIBLEY By MILTON LESSER [Transcriber's Note: This etext was produced from Galaxy Science Fiction July 1951. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] All she wanted was a mate and she had the gumption to go out and hunt one down. But that meant poaching in a strictly forbidden territory! The best that could be said for Matilda Penshaws was that she was something of a paradox. She was thirty-three years old, certainly not aged when you consider the fact that the female life expectancy is now up in the sixties, but the lines were beginning to etch their permanent paths across her face and now she needed certain remedial undergarments at which she would have scoffed ten or even five years ago. Matilda was also looking for a husband. This, in itself, was not unusual—but Matilda was so completely wrapped up in the romantic fallacy of her day that she sought a prince charming, a faithful Don Juan, a man who had been everywhere and tasted of every worldly pleasure and who now wanted to sit on a porch and talk about it all to Matilda. The fact that in all probability such a man did not exist disturbed Matilda not in the least. She had been known to say that there are over a billion men in the world, a goodly percentage of whom are eligible bachelors, and that the right one would come along simply because she had been waiting for him. Matilda, you see, had patience. She also had a fetish. Matilda had received her A.B. from exclusive Ursula Johns College and Radcliff had yielded her Masters degree, yet Matilda was an avid follower of the pen pal columns. She would read them carefully and then read them again, looking for the masculine names which, through a system known only to Matilda, had an affinity to her own. To the gentlemen upon whom these names were affixed, Matilda would write, and she often told her mother, the widow Penshaws, that it was in this way she would find her husband. The widow Penshaws impatiently told her to go out and get dates. That particular night, Matilda pulled her battered old sedan into the garage and walked up the walk to the porch. The widow Penshaws was rocking on the glider and Matilda said hello. The first thing the widow Penshaws did was to take Matilda's left hand in her own and examine the next-to-the-last finger. "I thought so," she said. "I knew this was coming when I saw that look in your eye at dinner. Where is Herman's engagement ring?" Matilda smiled. "It wouldn't have worked out, Ma. He was too darned stuffy. I gave him his ring and said thanks anyway and he smiled politely and said he wished I had told him sooner because his fifteenth college reunion was this weekend and he had already turned down the invitation." The widow Penshaws nodded regretfully. "That was thoughtful of Herman to hide his feelings." "Hogwash!" said her daughter. "He has no true feelings. He's sorry that he had to miss his college reunion. That's all he has to hide. A stuffy Victorian prude and even less of a man than the others." "But, Matilda, that's your fifth broken engagement in three years. It ain't that you ain't popular, but you just don't want to cooperate. You don't fall in love, Matilda—no one does. Love osmoses into you slowly, without you even knowing, and it keeps growing all the time." Matilda admired her mother's use of the word osmosis, but she found nothing which was not objectionable about being unaware of the impact of love. She said good-night and went upstairs, climbed out of her light summer dress and took a cold shower. She began to hum to herself. She had not yet seen the pen pal section of the current Literary Review , and because the subject matter of that magazine was somewhat highbrow and cosmopolitan, she could expect a gratifying selection of pen pals. She shut off the shower, brushed her teeth, gargled, patted herself dry with a towel, and jumped into bed, careful to lock the door of her bedroom. She dared not let the widow Penshaws know that she slept in the nude; the widow Penshaws would object to a girl sleeping in the nude, even if the nearest neighbor was three hundred yards away. Matilda switched her bed lamp on and dabbed some citronella on each ear lobe and a little droplet on her chin (how she hated insects!). Then she propped up her pillows—two pillows partially stopped her post-nasal drip; and took the latest issue of the Literary Review off the night table. She flipped through the pages and came to personals. Someone in Nebraska wanted to trade match books; someone in New York needed a midwestern pen pal, but it was a woman; an elderly man interested in ornithology wanted a young chick correspondent interested in the same subject; a young, personable man wanted an editorial position because he thought he had something to offer the editorial world; and— Matilda read the next one twice. Then she held it close to the light and read it again. The Literary Review was one of the few magazines which printed the name of the advertiser rather than a box number, and Matilda even liked the sound of the name. But mostly, she had to admit to herself, it was the flavor of the wording. This very well could be it . Or, that is, him . Intelligent, somewhat egotistical male who's really been around, whose universal experience can make the average cosmopolite look like a provincial hick, is in need of several female correspondents: must be intelligent, have gumption, be capable of listening to male who has a lot to say and wants to say it. All others need not apply. Wonderful opportunity cultural experience ... Haron Gorka, Cedar Falls, Ill. The man was egotistical, all right; Matilda could see that. But she had never minded an egotistical man, at least not when he had something about which he had a genuine reason to be egotistical. The man sounded as though he would have reason indeed. He only wanted the best because he was the best. Like calls to like. The name—Haron Gorka: its oddness was somehow beautiful to Matilda. Haron Gorka—the nationality could be anything. And that was it. He had no nationality for all intents and purposes; he was an international man, a figure among figures, a paragon.... Matilda sighed happily as she put out the light. The moon shone in through the window brightly, and at such times Matilda generally would get up, go to the cupboard, pull out a towel, take two hairpins from her powder drawer, pin the towel to the screen of her window, and hence keep the disturbing moonlight from her eyes. But this time it did not disturb her, and she would let it shine. Cedar Falls was a small town not fifty miles from her home, and she'd get there a hop, skip, and jump ahead of her competitors, simply by arriving in person instead of writing a letter. Matilda was not yet that far gone in years or appearance. Dressed properly, she could hope to make a favorable impression in person, and she felt it was important to beat the influx of mail to Cedar Falls. Matilda got out of bed at seven, tiptoed into the bathroom, showered with a merest wary trickle of water, tiptoed back into her bedroom, dressed in her very best cotton over the finest of uplifting and figure-moulding underthings, made sure her stocking seams were perfectly straight, brushed her suede shoes, admired herself in the mirror, read the ad again, wished for a moment she were a bit younger, and tiptoed downstairs. The widow Penshaws met her at the bottom of the stairwell. "Mother," gasped Matilda. Matilda always gasped when she saw something unexpected. "What on earth are you doing up?" The widow Penshaws smiled somewhat toothlessly, having neglected to put in both her uppers and lowers this early in the morning. "I'm fixing breakfast, of course...." Then the widow Penshaws told Matilda that she could never hope to sneak about the house without her mother knowing about it, and that even if she were going out in response to one of those foolish ads in the magazines, she would still need a good breakfast to start with like only mother could cook. Matilda moodily thanked the widow Penshaws. Driving the fifty miles to Cedar Falls in a little less than an hour, Matilda hummed Mendelssohn's Wedding March all the way. It was her favorite piece of music. Once, she told herself: Matilda Penshaws, you are being premature about the whole thing. But she laughed and thought that if she was, she was, and, meanwhile, she could only get to Cedar Falls and find out. And so she got there. The man in the wire cage at the Cedar Falls post office was a stereotype. Matilda always liked to think in terms of stereotypes. This man was small, roundish, florid of face, with a pair of eyeglasses which hung too far down on his nose. Matilda knew he would peer over his glasses and answer questions grudgingly. "Hello," said Matilda. The stereotype grunted and peered at her over his glasses. Matilda asked him where she could find Haron Gorka. "What?" "I said, where can I find Haron Gorka?" "Is that in the United States?" "It's not a that; it's a he. Where can I find him? Where does he live? What's the quickest way to get there?" The stereotype pushed up his glasses and looked at her squarely. "Now take it easy, ma'am. First place, I don't know any Haron Gorka—" Matilda kept the alarm from creeping into her voice. She muttered an oh under her breath and took out the ad. This she showed to the stereotype, and he scratched his bald head. Then he told Matilda almost happily that he was sorry he couldn't help her. He grudgingly suggested that if it really were important, she might check with the police. Matilda did, only they didn't know any Haron Gorka, either. It turned out that no one did: Matilda tried the general store, the fire department, the city hall, the high school, all three Cedar Falls gas stations, the livery stable, and half a dozen private dwellings at random. As far us the gentry of Cedar Falls was concerned, Haron Gorka did not exist. Matilda felt bad, but she had no intention of returning home this early. If she could not find Haron Gorka, that was one thing; but she knew that she'd rather not return home and face the widow Penshaws, at least not for a while yet. The widow Penshaws meant well, but she liked to analyze other people's mistakes, especially Matilda's. Accordingly, Matilda trudged wearily toward Cedar Falls' small and unimposing library. She could release some of her pent-up aggression by browsing through the dusty slacks. This she did, but it was unrewarding. Cedar Falls had what might be called a microscopic library, and Matilda thought that if this small building were filled with microfilm rather than books, the library still would be lacking. Hence she retraced her steps and nodded to the old librarian as she passed. Then Matilda frowned. Twenty years from now, this could be Matilda Penshaws—complete with plain gray dress, rimless spectacles, gray hair, suspicious eyes, and a broom-stick figure.... On the other hand—why not? Why couldn't the librarian help her? Why hadn't she thought of it before? Certainly a man as well-educated as Haron Gorka would be an avid reader, and unless he had a permanent residence here in Cedar Palls, one couldn't expect that he'd have his own library with him. This being the case, a third-rate collection of books was far better than no collection at all, and perhaps the librarian would know Mr. Haron Gorka. Matilda cleared her throat. "Pardon me," she began. "I'm looking for—" "Haron Gorka." The librarian nodded. "How on earth did you know?" "That's easy. You're the sixth young woman who came here inquiring about that man today. Six of you—five others in the morning, and now you in the afternoon. I never did trust this Mr. Gorka...." Matilda jumped as if she had been struck strategically from the rear. "You know him? You know Haron Gorka?" "Certainly. Of course I know him. He's our steadiest reader here at the library. Not a week goes by that he doesn't take out three, four books. Scholarly gentleman, but not without charm. If I were twenty years younger—" Matilda thought a little flattery might be effective. "Only ten," she assured the librarian. "Ten years would be more than sufficient, I'm sure." "Are you? Well. Well, well." The librarian did something with the back of her hair, but it looked the same as before. "Maybe you're right. Maybe you're right at that." Then she sighed. "But I guess a miss is as good as a mile." "What do you mean?" "I mean anyone would like to correspond with Haron Gorka. Or to know him well. To be considered his friend. Haron Gorka...." The librarian seemed about to soar off into the air someplace, and if five women had been here first, Matilda was now definitely in a hurry. "Um, where can I find Mr. Gorka?" "I'm not supposed to do this, you know. We're not permitted to give the addresses of any of our people. Against regulations, my dear." "What about the other five women?" "They convinced me that I ought to give them his address." Matilda reached into her pocket-book and withdrew a five dollar bill. "Was this the way?" she demanded. Matilda was not very good at this sort of thing. The librarian shook her head. Matilda nodded shrewdly and added a twin brother to the bill in her hand. "Then is this better?" "That's worse. I wouldn't take your money—" "Sorry. What then?" "If I can't enjoy an association with Haron Gorka directly, I still could get the vicarious pleasure of your contact with him. Report to me faithfully and you'll get his address. That's what the other five will do, and with half a dozen of you, I'll get an overall picture. Each one of you will tell me about Haron Gorka, sparing no details. You each have a distinct personality, of course, and it will color each picture considerably. But with six of you reporting, I should receive my share of vicarious enjoyment. Is it—ah—a deal?" Matilda assured her that it was, and, breathlessly, she wrote down the address. She thanked the librarian and then she went out to her car, whistling to herself. Haron Gorka lived in what could have been an agrarian estate, except that the land no longer was being tilled. The house itself had fallen to ruin. This surprised Matilda, but she did not let it keep her spirits in check. Haron Gorka, the man, was what counted, and the librarian's account of him certainly had been glowing enough. Perhaps he was too busy with his cultural pursuits to pay any real attention to his dwelling. That was it, of course: the conspicuous show of wealth or personal industry meant nothing at all to Haron Gorka. Matilda liked him all the more for it. There were five cars parked in the long driveway, and now Matilda's made the sixth. In spite of herself, she smiled. She had not been the only one with the idea to visit Haron Gorka in person. With half a dozen of them there, the laggards who resorted to posting letters would be left far behind. Matilda congratulated herself for what she thought had been her ingenuity, and which now turned out to be something which she had in common with five other women. You live and learn, thought Matilda. And then, quite annoyedly, she berated herself for not having been the first. Perhaps the other five all were satisfactory; perhaps she wouldn't be needed; perhaps she was too late.... As it turned out, she wasn't. Not only that, she was welcomed with open arms. Not by Haron Gorka; that she really might have liked. Instead, someone she could only regard as a menial met her, and when he asked had she come in response to the advertisement, she nodded eagerly. He told her that was fine and he ushered her straight into a room which evidently was to be her living quarters. It contained a small undersized bed, a table, and a chair, and, near a little slot in the wall, there was a button. "You want any food or drink," the servant told her, "and you just press that button. The results will surprise you." "What about Mr. Gorka?" "When he wants you, he will send for you. Meanwhile, make yourself to home, lady, and I will tell him you are here." A little doubtful now, Matilda thanked him and watched him leave. He closed the door softly behind his retreating feet, but Matilda's ears had not missed the ominous click. She ran to the door and tried to open it, but it would not budge. It was locked—from the outside. It must be said to Matilda's favor that she sobbed only once. After that she realized that what is done is done and here, past thirty, she wasn't going to be girlishly timid about it. Besides, it was not her fault if, in his unconcern, Haron Gorka had unwittingly hired a neurotic servant. For a time Matilda paced back and forth in her room, and of what was going on outside she could hear nothing. In that case, she would pretend that there was nothing outside the little room, and presently she lay down on the bed to take a nap. This didn't last long, however: she had a nightmare in which Haron Gorka appeared as a giant with two heads, but, upon awaking with a start, she immediately ascribed that to her overwrought nerves. At that point she remembered what the servant had said about food and she thought at once of the supreme justice she could do to a juicy beefsteak. Well, maybe they didn't have a beefsteak. In that case, she would take what they had, and, accordingly, she walked to the little slot in the wall and pressed the button. She heard the whir of machinery. A moment later there was a soft sliding sound. Through the slot first came a delicious aroma, followed almost instantly by a tray. On the tray were a bowl of turtle soup, mashed potatoes, green peas, bread, a strange cocktail, root-beer, a parfait—and a thick tenderloin sizzling in hot butter sauce. Matilda gasped once and felt about to gasp again—but by then her salivary glands were working overtime, and she ate her meal. The fact that it was precisely what she would have wanted could, of course, be attributed to coincidence, and the further fact that everything was extremely palatable made her forget all about Haron Gorka's neurotic servant. When she finished her meal a pleasant lethargy possessed her, and in a little while Matilda was asleep again. This time she did not dream at all. It was a deep sleep and a restful one, and when she awoke it was with the wonderful feeling that everything was all right. The feeling did not last long. Standing over her was Haron Gorka's servant, and he said, "Mr. Gorka will see you now." "Now?" "Now. That's what you're here for, isn't it?" He had a point there, but Matilda hardly even had time to fix her hair. She told the servant so. "Miss," he replied, "I assure you it will not matter in the least to Haron Gorka. You are here and he is ready to see you and that is all that matters." "You sure?" Matilda wanted to take no chances. "Yes. Come." She followed him out of the little room and across what should have been a spacious dining area, except that everything seemed covered with dust. Of the other women Matilda could see nothing, and she suddenly realized that each of them probably had a cubicle of a room like her own, and that each in her turn had already had her first visit with Haron Gorka. Well, then, she must see to it that she impressed him better than did all the rest, and, later, when she returned to tell the old librarian of her adventures, she could perhaps draw her out and compare notes. She would not admit even to herself that she was disappointed with Haron Gorka. It was not that he was homely and unimpressive; it was just that he was so ordinary -looking. She almost would have preferred the monster of her dreams. He wore a white linen suit and he had mousy hair, drab eyes, an almost-Roman nose, a petulant mouth with the slight arch of the egotist at each corner. He said, "Greetings. You have come—" "In response to your ad. How do you do, Mr. Gorka?" She hoped she wasn't being too formal. But, then, there was no sense in assuming that he would like informality. She could only wait and see and adjust her own actions to suit him. Meanwhile, it would be best to keep on the middle of the road. "I am fine. Are you ready?" "Ready?" "Certainly. You came in response to my ad. You want to hear me talk, do you not?" "I—do." Matilda had had visions of her prince charming sitting back and relaxing with her, telling her of the many things he had done and seen. But first she certainly would have liked to get to know the man. Well, Haron Gorka obviously had more experience along these lines than she did. He waited, however, as if wondering what to say, and Matilda, accustomed to social chatter, gave him a gambit. "I must admit I was surprised when I got exactly what I wanted for dinner," she told him brightly. "Eh? What say? Oh, yes, naturally. A combination of telepathy and teleportation. The synthetic cookery is attuned to your mind when you press the buzzer, and the strength of your psychic impulses determines how closely the meal will adjust to your desires. The fact that the adjustment here was near perfect is commendable. It means either that you have a high psi-quotient, or that you were very hungry." "Yes," said Matilda vaguely. Perhaps it might be better, after all, if Haron Gorka were to talk to her as he saw fit. "Ready?" "Uh—ready." "Well?" "Well, what, Mr. Gorka?" "What would you like me to talk about?" "Oh, anything." "Please. As the ad read, my universal experience—is universal. Literally. You'll have to be more specific." "Well, why don't you tell me about some of your far travels? Unfortunately, while I've done a lot of reading, I haven't been to all the places I would have liked—" "Good enough. You know, of course, how frigid Deneb VII is?" Matilda said, "Beg pardon?" "Well, there was the time our crew—before I had retired, of course—made a crash landing there. We could survive in the vac-suits, of course, but the thlomots were after us almost at once. They go mad over plastic. They will eat absolutely any sort of plastic. Our vac-suits—" "—were made of plastic," Matilda suggested. She did not understand a thing he was talking about, but she felt she had better act bright. "No, no. Must you interrupt? The air-hose and the water feed, these were plastic. Not the rest of the suit. The point is that half of us were destroyed before the rescue ship could come, and the remainder were near death. I owe my life to the mimicry of a flaak from Capella III. It assumed the properties of plastic and led the thlomots a merry chase across the frozen surface of D VII. You travel in the Deneb system now and Interstellar Ordinance makes it mandatory to carry flaaks with you. Excellent idea, really excellent." Almost at once, Matilda's educational background should have told her that Haron Gorka was mouthing gibberish. But on the other hand she wanted to believe in him and the result was that it took until now for her to realize it. "Stop making fun of me," she said. "So, naturally, you'll see flaaks all over that system—" "Stop!" "What's that? Making fun of you?" Haron Gorka's voice had been so eager as he spoke, high-pitched, almost like a child's, and now he seemed disappointed. He smiled, but it was a sad smile, a smile of resignation, and he said, "Very well. I'm wrong again. You are the sixth, and you're no better than the other five. Perhaps you are even more outspoken. When you see my wife, tell her to come back. Again she is right and I am wrong...." Haron Gorka turned his back. Matilda could do nothing but leave the room, walk back through the house, go outside and get into her car. She noticed not without surprise that the other five cars were now gone. She was the last of Haron Gorka's guests to depart. As she shifted into reverse and pulled out of the driveway, she saw the servant leaving, too. Far down the road, he was walking slowly. Then Haron Gorka had severed that relationship, too, and now he was all alone. As she drove back to town, the disappointment melted slowly away. There were, of course, two alternatives. Either Haron Gorka was an eccentric who enjoyed this sort of outlandish tomfoolery, or else he was plainly insane. She could still picture him ranting on aimlessly to no one in particular about places which had no existence outside of his mind, his voice high-pitched and eager. It was not until she had passed the small library building that she remembered what she had promised the librarian. In her own way, the aging woman would be as disappointed as Matilda, but a promise was a promise, and Matilda turned the car in a wide U-turn and parked it outside the library. The woman sat at her desk as Matilda had remembered her, gray, broom-stick figure, rigid. But now when she saw Matilda she perked up visibly. "Hello, my dear," she said. "Hi." "You're back a bit sooner than I expected. But, then, the other five have returned, too, and I imagine your story will be similar." "I don't know what they told you," Matilda said. "But this is what happened to me." She quickly then related everything which had happened, completely and in detail. She did this first because it was a promise, and second because she knew it would make her feel better. "So," she finished, "Haron Gorka is either extremely eccentric or insane. I'm sorry." "He's neither," the librarian contradicted. "Perhaps he is slightly eccentric by your standards, but really, my dear, he is neither." "What do you mean?" "Did he leave a message for his wife?" "Why, yes. Yes, he did. But how did you know? Oh, I suppose he told the five." "No. He didn't. But you were the last and I thought he would give you a message for his wife—" Matilda didn't understand. She didn't understand at all, but she told the little librarian what the message was. "He wanted her to return," she said. The librarian nodded, a happy smile on her lips. "You wouldn't believe me if I told you something." "What's that?" "I am Mrs. Gorka." The librarian stood up and came around the desk. She opened a drawer and took out her hat and perched it jauntily atop her gray hair. "You see, my dear, Haron expects too much. He expects entirely too much." Matilda did not say a word. One madman a day would be quite enough for anybody, but here she found herself confronted with two. "We've been tripping for centuries, visiting every habitable star system from our home near Canopus. But Haron is too demanding. He says I am a finicky traveler, that he could do much better alone, the accommodations have to be just right for me, and so forth. When he loses his temper, he tries to convince me that any number of females of the particular planet would be more than thrilled if they were given the opportunity just to listen to him. "But he's wrong. It's a hard life for a woman. Someday—five thousand, ten thousand years from now—I will convince him. And then we will settle down on Canopus XIV and cultivate torgas . That would be so nice—" "I'm sure." "Well, if Haron wants me back, then I have to go. Have a care, my dear. If you marry, choose a home-body. I've had the experience and you've seen my Haron for yourself." And then the woman was gone. Numbly, Matilda walked to the doorway and watched her angular figure disappear down the road. Of all the crazy things.... Deneb and Capella and Canopus, these were stars. Add a number and you might have a planet revolving about each star. Of all the insane— They were mad, all right, and now Matilda wondered if, actually, they were husband and wife. It could readily be; maybe the madness was catching. Maybe if you thought too much about such things, such travels, you could get that way. Of course, Herman represented the other extreme, and Herman was even worse in his own way—but hereafter Matilda would seek the happy medium. And, above all else, she had had enough of her pen pal columns. They were, she realized, for kids. She ate dinner in Cedar Falls and then she went out to her car again, preparing for the journey back home. The sun had set and it was a clear night, and overhead the great broad sweep of the Milky Way was a pale rainbow bridge in the sky. Matilda paused. Off in the distance there was a glow on the horizon, and that was the direction of Haron Gorka's place. The glow increased; soon it was a bright red pulse pounding on the horizon. It flickered. It flickered again, and finally it was gone. The stars were white and brilliant in the clear country air. That was why Matilda liked the country better than the city, particularly on a clear summer night when you could see the span of the Milky Way. But abruptly the stars and the Milky Way were paled by the brightest shooting star Matilda had ever seen. It flashed suddenly and it remained in view for a full second, searing a bright orange path across the night sky. Matilda gasped and ran into her car. She started the gears and pressed the accelerator to the floor, keeping it there all the way home. It was the first time she had ever seen a shooting star going up .
|
C. his physical appearance
|
What are the categories of code-mixed puns?
|
### Introduction
Humour is one of the most complex and intriguing phenomenon of the human language. It exists in various forms, across space and time, in literature and culture, and is a valued part of human interactions. Puns are one of the simplest and most common forms of humour in the English language. They are also one of the most widespread forms of spontaneous humour BIBREF0 and have found their place in casual conversations, literature, online comments, tweets and advertisements BIBREF1 , BIBREF2 . Puns are a hugely versatile and commonly used literary device and it is essential to include them in any comprehensive approach to computational humour. In this paper, we consider Hindi-English code-mixed puns and aim to automatically recover their targets. The target of a pun is its phonologically similar counterpart, the relationship to which and whose resolution (recovery) in the mind of the listener/hearer induces humour. For example, in the pun “The life of a patient of hypertension is always at steak." the word “steak" is the pun with target “stake". With India being a diverse linguistic region, there is an ever increasing usage of code-mixed Hindi-English language (along with various others) because bilingualism and even multilingualism are quite common. Consequently, we have also seen an increase in the usage of code-mixed language in online forums, advertisements etc. Code-mixed humour, especially puns have become increasingly popular because being able to use the same punning techniques but with two languages in play has opened up numerous avenues for new and interesting wordplays. With the increasing popularity and acceptance for the usage of code-mixed language, it has become important that computers are also able to process it and even decipher complex phenomena like humour. Traditional Word Sense Disambiguation (WSD) based methods cannot be used in target recovery of code-mixed puns, because they are no longer about multiple senses of a single word but about two words from two different languages. Code-switching comes with no markers, and the punning word may not even be a word in either of the languages being used. Sometimes words from the two languages can be combined to form a word which only a bilingual speaker would understand. Hence, this task on such data calls for a different set of strategies altogether. We approach this problem in two parts. First, we analyze the types of structures in code-mixed puns and classify them into two categories namely intra-sequential and intra-word. Second, we develop a four stage pipeline to achieve our goal - Language Identification, Pun Candidate Identification, Context Lookup and Phonetic Distance Minimization. We then test our approach on a small dataset and note that our method is successfully able to recover targets for a majority of the puns. To the best of our knowledge, this is a first attempt at dealing with code-mixed puns. The outline of the paper is as follows: Section 2 gives a brief description of the background and prior work on puns - both in the field of linguistics and in the field of computational humour, along with a brief introduction to the field of code-mixing. Section 3 defines our problem statement, our classification model on code-mixed puns, the dataset we use to test our approach, and our proposed model for the task of automatic target recovery of Hindi-English code-mixed puns. In Section 4, we analyse the performance of our model on a set of puns, and discuss the various error cases. Finally, we conclude in Section 5 with a review of our research contributions and an outline of our plans for future work. ### Puns
Puns are a form of wordplay jokes in which one sign (e.g. a word or a phrase) suggests two or more meanings by exploiting polysemy, homonymy, or phonological similarity to another sign, for an intended humorous or rhetorical effect BIBREF3 . Puns where the two meanings share the same pronunciation are known as homophonic or perfect puns, while those relying on similar but non-identical sounding words are known as heterophonic BIBREF4 or imperfect puns BIBREF5 . In this paper, we study automatic target recoverability of English-Hindi code mixed puns - which are more commonly imperfect puns, but may also be perfect puns in some cases. Zwicky and Zwicky zwicky1986imperfect, Sobkowiak sobkowiak1991metaphonology extensively studied various phonological variations in imperfect puns such as strong asymmetry in phoneme substitution. They note that puns show more frequent changes in vowels than in consonants because of their smaller role in target recoverability. Puns have received attention in the field of computational humour, both in generation of puns and their understanding. Generation: One of the earliest attempts at generating humour was by Lessard and Levin lessard1992computational, when they built an antonym-based system to generate Tom Swifties. Since then, we have seen various other attempts at the task with different strategies. JAPE was a system which exploited framing and phonetic relationships to automatically generate funny punning riddles, or more specifically phonologically ambiguous riddles, having noun phrase punchlines BIBREF6 . Venour venour1999computational built a system which generated HCPPs (Homonym Common Phrase Pun), simple 2 sentence puns based on associations between words occurring in common phrases. WisCraic was a system built by McKay mckay2002generation, which generated simple one-sentence puns based on semantic associations of words. Valitutti et al. valitutti2008textual attempted to automatically generate advertisements by punning on familiar expressions, with an affective connotation. Identification and understanding: Hempelmann hempelmann2003paronomasic studied target recoverability, arguing that a good model for it provides necessary groundwork for effective automatic pun generation. He worked on a theory which models prominent factors in punning such as phonological similarity and studied how these measures could be used to evaluate possible imperfect puns given an input word and a set of target words. Yokogawa yokogawa2002japanese analyzed ungrammatical Japanese puns and generated target candidates by replacing ungrammatical parts of the sentence by similar expressions. Taylor and Mazlack taylor2004computationally worked on computational recognition of word-play in the restricted domain of Knock-Knock jokes. Jaech et al. jaech2016phonological developed a computational model for target recovery of puns using techniques for automatic speech recognition, and learned phone edit probabilities in puns. Miller and Gurevych Miller2015AutomaticDO, Miller et al.miller2017semeval describe different methods on pun identification and disambiguation. Word Sense Disambiguation (WSD) based techniques are most common among the methods used. To the best of our knowledge no prior work has been attempted on code-mixed puns. ### Code-mixing
Code-mixing is the mixing of two or more languages or language varieties. Code-mixing is now recognized as a natural part of bilingual and multilingual language use. Significant linguistic efforts have been made to understand the sociological and conversational necessity behind code-switching BIBREF7 ; for example, to understand whether it is an act of identity in a social group, or a consequence of a lack of competence in either of the languages. These papers distinguish between inter-sentence, intra-sentence and intra-word code mixing. Different types of language mixing phenomena have been discussed and defined by several linguists, with some making clear distinctions between phenomena based on certain criteria, while others use `code-mixing’ or `code-switching’ as umbrella terms to include any type of language mixing — see, e.g., Muysken muysken1995code or Gafaranga and Torras gafaranga2002interactional. In this paper, we use both these terms ‘code-mixing’ and `code-switching' interchangeably. Coming to the work on automatic analysis of code-mixed languages, there have been studies on detecting code mixing in spoken language as well as different types of short texts, such as information retrieval queries BIBREF8 , SMS messages BIBREF9 , BIBREF10 , social media data BIBREF11 and online conversations BIBREF12 . These scholars have carried out experiments for the task of language identification using language models, dictionaries, logistic regression classification, Conditional Random Fields, SVMs, and noted that approaches using contextual knowledge were most robust. King and Abney king2013labeling used weakly semi-supervised methods to perform word-level language identification. We however, use a dictionary based approach for the language identification task. While working with puns, ambiguity in language identification can be an important marker for identifying the pun, so it is more important for us to recognize all possible ambiguities rather than picking just one depending on probabilities. This ability to recognize ambiguities, and the simplicity of a dictionary-based language identification model makes it suited for this task. ### Methodology
We focus on the task of automatically disambiguating or recovering Hindi-English code mixed puns. For this purpose, it is first necessary to understand what these puns are. ### Classification
For the purposes of this research, we only consider puns where the ambiguity or the wordplay lies in the code-switching i.e, the pun word and its target are from different languages. For example the pun "Rivers can't hear because woh behri hoti hai." is a sentence with the pun being behri (meaning deaf) and its target being beh rahi (meaning flowing). Here, while the sentence is code-mixed, the pun word and the target both belong to the same language. We do not consider such puns for the present study. We analyze the structure of code-mixed puns with the pun word and its target belonging to different languages and propose two broad categories to classify them in - puns where the code-mixing is intra-sentential and the other where it is intra-word. Both these categories are explained below, while we evaluate only on the former category. Intra-sentential code-mixing is where code-switching occurs within a sentence. Here, the language varies at the word level. Also, each word of the sentence belongs to one or the other language. Table 1 gives examples of puns belonging to this category. In this category, code mixing is present within a word. New words are formed using Portmanteau or Blending where two or more syllables/phonemes from different languages are blended together to form a single word, resulting in a word which is phonetically similar to the target word. Table 2 illustrates examples of intra-word code-mixed puns. ### Dataset
Most puns we hear or use in everyday conversations are rarely recorded. One of the most common resources to find recorded puns are advertisements, for example the highly creative and frequently released Amul advertisements in India BIBREF1 . Most of these are contextually integrated BIBREF0 with an image. While such puns may lose their humour out of context, it is still possible to recover their targets, so using these does not affect our task in any way To create a dataset to test our model on, we collected 518 advertisements released by Amul in the years 2014, 2015, 2017 and 2018, from their official web page. Of these, 333 were puns, including 121 code-mixed puns as defined in Section 3.1. We extracted the text of these 121 code-mixed puns and asked 3 people to disambiguate them, given just the advertisement text. All three annotators were university students in 22-23 years age group, native Hindi speakers with bilingual fluency in English. The annotators were asked to identify the location of the pun in each of the advertisements and write down the target of the pun. Any disagreements between annotators were resolved by mutual discussion. In a few cases where puns were identified to have multiple targets, we kept all such possibilities in our dataset. A few puns were identified to be non-recoverable because of the lack of contextual knowledge, while a few puns had multiple pun locations. We removed both these types from our dataset, which left us with 110 puns. Finally, we divided these 110 annotated puns into the two categories as defined in Section 3.1 thereby getting 51 advertisements categorized as intra-sentential code-mixed puns, and the rest as intra-word code-mixed puns. We use the former as our test data. ### Model
For preprocessing the text we give as input to our system, we first tokenize the advertisement text using NLTK's BIBREF13 tokenizer and remove all punctuations. We then give the resultant tokens as input to our model, which is a 4 step process as described below: At this step, we aim to identify the language of each of the tokens in the input text by classifying them into one of the 5 categories: English, Hindi, Named Entity (NE), Out of Vocabulary (OOV), or Ambiguous (words that could belong to both English and Hindi). We use a dictionary-based lookup method to classify a word in English or Hindi. Since the input is in Roman script, to recognize Hindi words, we use a list of 30k transliterated Hindi words in Roman to their Devanagari counterparts BIBREF14 . For the English language, we collected news data from the archives of a leading Indian Newspaper, The Hindu. Data from 2012-2018 under the tags National, International, Sports, Cinema, Television was collected, amounting to 12,600 articles with 200k sentences and around 38k unique words. We use this data to build an English dictionary. Also, we used NLTK's BIBREF13 Named Entity Recognition module on the same data to get a dictionary of Named Entities. We first try to classify all tokens as English, Hindi and NE using these dictionaries. Then, words which are found in both English and Hindi are marked as Ambiguous. The words which do not fall into any of these are classified as OOV. We now identify all possible punning locations in the text. For this, we consider words on the boundaries of language change as candidates for pun locations. Then, all NEs and OOV words are added to the list of pun candidates as well. Third, if any Ambiguous words exist in the text, we consider it once as English and once as Hindi for the next steps. In this step, we contextually lookup all the candidate locations using left context and right context to get a list of all words that may occur at that position. We use bi-gram language models we built using Knesser-Ney smoothing BIBREF15 . We used the data mentioned in the previous step to build the language model for English, and 100k sentences from Hindi monolingual data from BIBREF16 to build the language models for English and Hindi respectively. As it is highly likely that the left and the right context at a pun location belong to different languages, we look at each of those separately instead of taking an intersection of the left and the right context. Lastly, at each pun location, we calculate the similarity of the word at that location with all the words that can occur at that location depending on the context and pick the most similar words as the possible targets. To compare words belonging to two different languages on a phonetic basis, we convert both of them to WX notation BIBREF17 , which denotes a standard way to represent Indian languages in the Roman script. We transliterate our identified Hindi words from Devanagari to WX notation. To convert English words to the same notation, we use the CMU phonetic dictionary , which uses a 39 phoneme set to represent North American pronunciations of English words. We build a mapping between this phoneme set and WX notation. Whenever there was no exact parallel between CMU pronouncing dictionary's notation and WX, we used the word's Indian English pronunciation to find the closest match. Once we converted all to WX notation, we use a modified version of Levenshtein Distance BIBREF18 to find most similar words. In this normalized version of Levenshtein distance, we account for a few features like aspirations (for example, /p/,/ph/) which are non-phonemic in English, vowel elongations, rhyme, same beginning or ending sounds. In case of an OOV word, since it cannot be converted to WX notation due to non-availability of any phonetic transcription, we simply find the words with the least orthographic distance when written in Roman script, using a similar measure as used for phonetic distance with a few more normalizations (for example, considering 'w' and 'v' as similar). ### Results and discussion
We test the model explained in the previous section on our test dataset described in Section 3.2 and note that this method is correctly able to recover targets for 34 out of these 51 puns, or around 67% of the puns, which are very encouraging results for this complex task. Examples where the system performed successfully are given in Table 3 . We do a thorough error analysis below for the cases our method fails for. ### Conclusion and Future work
To conclude, in this paper, we present a first-ever work on target recovery code-mixed puns. We study various puns where the word-play is a result of code-switching, and classify them into 2 categories - puns with intra-sentential code mixing and those with intra-word code mixing. We then propose a methodology to recover the targets for puns belonging to the former category, using only monolingual language data. We test our proposed approach on a small manually annotated dataset, and we see that our system was able to successfully recover 67% of the puns from the set. In the future, we want to perform a more comprehensive evaluation of this approach on a larger, more diverse set of puns. We want to improve and extend our approach to be able to recover intra-word code-mixed puns along with the intra-sentential ones that it handles right now. After that, the system should be extended to be able to recover all kinds of puns in code-mixed language, regardless of whether the pun itself is monolingual or code-mixed. ### Acknowledgements
We thank the anonymous reviewers for their comments that helped improve this paper. Table 2: Examples of intra-word code-mixed puns Table 1: Examples of intra-sentential code-mixed puns Figure 1: This figure illustrates, taking Pun1 as example, our model and the 4 major steps it comprises: 1. Language Identification, 2. Identification of Candidate Pun Locations, 3. Context Lookup and 4. Phonetic Distance minimization. Table 3: Examples of puns successfully recovered by our system Table 5: Example for error case 2, where the pun is based on the pronunciation of an abbreviation. Table 6: Example for error case 3, where the target does not exist in the language model.
|
intra-sequential and intra-word
|
Why would the band of scientists fear that there might not be pieces to pick up once the affected area population returned?
A. They were doubting the theory by Schwartzberg.
B. The damage would be too substantial and there would nothing left to salvage.
C. They feared that no one would escape alive.
D. Theft in the area was also on the rise.
|
THE GREAT NEBRASKA SEA By ALLAN DANZIG Illustrated by WOOD [Transcriber's Note: This etext was produced from Galaxy Magazine August 1963. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] It has happened a hundred times in the long history of Earth—and, sooner or later, will happen again! Everyone—all the geologists, at any rate—had known about the Kiowa Fault for years. That was before there was anything very interesting to know about it. The first survey of Colorado traced its course north and south in the narrow valley of Kiowa Creek about twenty miles east of Denver; it extended south to the Arkansas River. And that was about all even the professionals were interested in knowing. There was never so much as a landslide to bring the Fault to the attention of the general public. It was still a matter of academic interest when in the late '40s geologists speculated on the relationship between the Kiowa Fault and the Conchas Fault farther south, in New Mexico, and which followed the Pecos as far south as Texas. Nor was there much in the papers a few years later when it was suggested that the Niobrara Fault (just inside and roughly parallel to the eastern border of Wyoming) was a northerly extension of the Kiowa. By the mid sixties it was definitely established that the three Faults were in fact a single line of fissure in the essential rock, stretching almost from the Canadian border well south of the New Mexico-Texas line. It is not really surprising that it took so long to figure out the connection. The population of the states affected was in places as low as five people per square mile! The land was so dry it seemed impossible that it could ever be used except for sheep-farming. It strikes us today as ironic that from the late '50s there was grave concern about the level of the water table throughout the entire area. The even more ironic solution to the problem began in the summer of 1973. It had been a particularly hot and dry August, and the Forestry Service was keeping an anxious eye out for the fires it knew it could expect. Dense smoke was reported rising above a virtually uninhabited area along Black Squirrel Creek, and a plane was sent out for a report. The report was—no fire at all. The rising cloud was not smoke, but dust. Thousands of cubic feet of dry earth rising lazily on the summer air. Rock slides, they guessed; certainly no fire. The Forestry Service had other worries at the moment, and filed the report. But after a week had gone by, the town of Edison, a good twenty miles away from the slides, was still complaining of the dust. Springs was going dry, too, apparently from underground disturbances. Not even in the Rockies could anyone remember a series of rock slides as bad as this. Newspapers in the mountain states gave it a few inches on the front page; anything is news in late August. And the geologists became interested. Seismologists were reporting unusual activity in the area, tremors too severe to be rock slides. Volcanic activity? Specifically, a dust volcano? Unusual, they knew, but right on the Kiowa Fault—could be. Labor Day crowds read the scientific conjectures with late summer lassitude. Sunday supplements ran four-color artists' conceptions of the possible volcano. "Only Active Volcano in U. S.?" demanded the headlines, and some papers even left off the question mark. It may seem odd that the simplest explanation was practically not mentioned. Only Joseph Schwartzberg, head geographer of the Department of the Interior, wondered if the disturbance might not be a settling of the Kiowa Fault. His suggestion was mentioned on page nine or ten of the Monday newspapers (page 27 of the New York Times ). The idea was not nearly so exciting as a volcano, even a lava-less one, and you couldn't draw a very dramatic picture of it. To excuse the other geologists, it must be said that the Kiowa Fault had never acted up before. It never sidestepped, never jiggled, never, never produced the regular shows of its little sister out in California, which almost daily bounced San Francisco or Los Angeles, or some place in between. The dust volcano was on the face of it a more plausible theory. Still, it was only a theory. It had to be proved. As the tremors grew bigger, along with the affected area, as several towns including Edison were shaken to pieces by incredible earthquakes, whole bus- and plane-loads of geologists set out for Colorado, without even waiting for their university and government department to approve budgets. They found, of course, that Schwartzberg had been perfectly correct. They found themselves on the scene of what was fast becoming the most violent and widespread earthquake North America—probably the world—has ever seen in historic times. To describe it in the simplest terms, land east of the Fault was settling, and at a precipitous rate. Rock scraped rock with a whining roar. Shuddery as a squeaky piece of chalk raked across a blackboard, the noise was deafening. The surfaces of the land east and west of the Fault seemed no longer to have any relation to each other. To the west, tortured rock reared into cliffs. East, where sharp reports and muffled wheezes told of continued buckling and dropping, the earth trembled downward. Atop the new cliffs, which seemed to grow by sudden inches from heaving rubble, dry earth fissured and trembled, sliding acres at a time to fall, smoking, into the bucking, heaving bottom of the depression. There the devastation was even more thorough, if less spectacular. Dry earth churned like mud, and rock shards weighing tons bumped and rolled about like pebbles as they shivered and cracked into pebbles themselves. "It looks like sand dancing in a child's sieve," said the normally impassive Schwartzberg in a nationwide broadcast from the scene of disaster. "No one here has ever seen anything like it." And the landslip was growing, north and south along the Fault. "Get out while you can," Schwartzberg urged the population of the affected area. "When it's over you can come back and pick up the pieces." But the band of scientists who had rallied to his leadership privately wondered if there would be any pieces. The Arkansas River, at Avondale and North Avondale, was sluggishly backing north into the deepening trough. At the rate things were going, there might be a new lake the entire length of El Paso and Pueblo Counties. And, warned Schwartzberg, this might only be the beginning. By 16 September the landslip had crept down the Huerfano River past Cedarwood. Avondale, North Avondale and Boone had totally disappeared. Land west of the Fault was holding firm, though Denver had recorded several small tremors; everywhere east of the Fault, to almost twenty miles away, the now-familiar lurch and steady fall had already sent several thousand Coloradans scurrying for safety. All mountain climbing was prohibited on the Eastern Slope because of the danger of rock slides from minor quakes. The geologists went home to wait. There wasn't much to wait for. The news got worse and worse. The Platte River, now, was creating a vast mud puddle where the town of Orchard had been. Just below Masters, Colorado, the river leaped 70-foot cliffs to add to the heaving chaos below. And the cliffs were higher every day as the land beneath them groaned downward in mile-square gulps. As the Fault moved north and south, new areas quivered into unwelcome life. Fields and whole mountainsides moved with deceptive sloth down, down. They danced "like sand in a sieve"; dry, they boiled into rubble. Telephone lines, railroad tracks, roads snapped and simply disappeared. Virtually all east-west land communication was suspended and the President declared a national emergency. By 23 September the Fault was active well into Wyoming on the north, and rapidly approaching the border of New Mexico to the south. Trinchera and Branson were totally evacuated, but even so the over-all death toll had risen above 1,000. Away to the east the situation was quiet but even more ominous. Tremendous fissures opened up perpendicular to the Fault, and a general subsidence of the land was noticeable well into Kansas and Nebraska. The western borders of these states, and soon of the Dakotas and Oklahoma as well, were slowly sinking. On the actual scene of the disaster (or the scenes ; it is impossible to speak of anything this size in the singular) there was a horrifying confusion. Prairie and hill cracked open under intolerable strains as the land shuddered downward in gasps and leaps. Springs burst to the surface in hot geysers and explosions of steam. The downtown section of North Platte, Nebraska, dropped eight feet, just like that, on the afternoon of 4 October. "We must remain calm," declared the Governor of Nebraska. "We must sit this thing out. Be assured that everything possible is being done." But what could be done, with his state dropping straight down at a mean rate of a foot a day? The Fault nicked off the south-east corner of Montana. It worked its way north along the Little Missouri. South, it ripped past Roswell, New Mexico, and tore down the Pecos toward Texas. All the upper reaches of the Missouri were standing puddles by now, and the Red River west of Paris, Texas, had begun to run backward. Soon the Missouri began slowly slipping away westward over the slowly churning land. Abandoning its bed, the river spread uncertainly across farmland and prairie, becoming a sea of mud beneath the sharp new cliffs which rose in rending line, ever taller as the land continued to sink, almost from Canada to the Mexican border. There were virtually no floods, in the usual sense. The water moved too slowly, spread itself with no real direction or force. But the vast sheets of sluggish water and jelly-like mud formed death-traps for the countless refugees now streaming east. Perhaps the North Platte disaster had been more than anyone could take. 193 people had died in that one cave-in. Certainly by 7 October it had to be officially admitted that there was an exodus of epic proportion. Nearly two million people were on the move, and the U. S. was faced with a gigantic wave of refugees. Rails, roads and air-lanes were jammed with terrified hordes who had left everything behind to crowd eastward. All through October hollow-eyed motorists flocked into Tulsa, Topeka, Omaha, Sioux Falls and Fargo. St. Louis was made distributing center for emergency squads which flew everywhere with milk for babies and dog food for evacuating pets. Gasoline trucks boomed west to meet the demand for gas, but once inside the "zone of terror," as the newspapers now called it, they found their route blocked by eastbound cars on the wrong side of the road. Shops left by their fleeing owners were looted by refugees from further west; an American Airlines plane was wrecked by a mob of would-be passengers in Bismarck, North Dakota. Federal and State troops were called out, but moving two million people was not to be done in an orderly way. And still the landslip grew larger. The new cliffs gleamed in the autumn sunshine, growing higher as the land beneath them continued its inexorable descent. On 21 October, at Lubbock, Texas, there was a noise variously described as a hollow roar, a shriek and a deep musical vibration like a church bell. It was simply the tortured rock of the substrata giving way. The second phase of the national disaster was beginning. The noise traveled due east at better than 85 miles per hour. In its wake the earth to the north "just seemed to collapse on itself like a punctured balloon," read one newspaper report. "Like a cake that's failed," said a Texarkana housewife who fortunately lived a block south of Thayer Street, where the fissure raced through. There was a sigh and a great cloud of dust, and Oklahoma subsided at the astounding rate of about six feet per hour. At Biloxi, on the Gulf, there had been uneasy shufflings under foot all day. "Not tremors, exactly," said the captain of a fishing boat which was somehow to ride out the coming flood, "but like as if the land wanted to be somewhere else." Everyone in doomed Biloxi would have done well to have been somewhere else that evening. At approximately 8:30 p.m. the town shuddered, seemed to rise a little like the edge of a hall carpet caught in a draft, and sank. So did the entire Mississippi and Alabama coast, at about the same moment. The tidal wave which was to gouge the center from the U. S. marched on the land. From the north shore of Lake Ponchartrain to the Appalachicola River in Florida, the Gulf coast simply disappeared. Gulfport, Biloxi, Mobile, Pensacola, Panama City: 200 miles of shoreline vanished, with over two and a half million people. An hour later a wall of water had swept over every town from Dothan, Alabama, to Bogalusa on the Louisiana-Mississippi border. "We must keep panic from our minds," said the Governor of Alabama in a radio message delivered from a hastily arranged all-station hookup. "We of the gallant southland have faced and withstood invasion before." Then, as ominous creakings and groanings of the earth announced the approach of the tidal wave, he flew out of Montgomery half an hour before the town disappeared forever. One head of the wave plunged north, eventually to spend itself in the hills south of Birmingham. The main sweep followed the lowest land. Reaching west, it swallowed Vicksburg and nicked the corner of Louisiana. The whole of East Carroll Parish was scoured from the map. The Mississippi River now ended at about Eudora, Arkansas, and minute by minute the advancing flood bit away miles of river bed, swelling north. Chicot, Jennie, Lake Village, Arkansas City, Snow Lake, Elaine, Helena and Memphis felt the tremors. The tormented city shuddered through the night. The earth continued its descent, eventually tipping 2-1/2 degrees down to the west. The "Memphis Tilt" is today one of the unique and charming characteristics of the gracious Old Town, but during the night of panic Memphis residents were sure they were doomed. South and west the waters carved deeply into Arkansas and Oklahoma. By morning it was plain that all of Arkansas was going under. Waves advanced on Little Rock at almost 100 miles an hour, new crests forming, overtopping the wave's leading edge as towns, hills and the thirst of the soil temporarily broke the furious charge. Washington announced the official hope that the Ozarks would stop the wild gallop of the unleashed Gulf, for in northwest Arkansas the land rose to over 2,000 feet. But nothing could save Oklahoma. By noon the water reached clutching fingers around Mt. Scott and Elk Mountain, deluging Hobart and almost all of Greer County. Despite hopeful announcements that the wave was slowing, had virtually stopped after inundating Oklahoma City, was being swallowed up in the desert near Amarillo, the wall of water continued its advance. For the land was still sinking, and the floods were constantly replenished from the Gulf. Schwartzberg and his geologists advised the utmost haste in evacuating the entire area between Colorado and Missouri, from Texas to North Dakota. Lubbock, Texas, went under. On a curling reflex the tidal wave blotted out Sweetwater and Big Spring. The Texas panhandle disappeared in one great swirl. Whirlpools opened. A great welter of smashed wood and human debris was sucked under, vomited up and pounded to pieces. Gulf-water crashed on the cliffs of New Mexico and fell back on itself in foam. Would-be rescuers on the cliffs along what had been the west bank of the Pecos River afterwards recalled the hiss and scream like tearing silk as the water broke furiously on the newly exposed rock. It was the most terrible sound they had ever heard. "We couldn't hear any shouts, of course, not that far away and with all the noise," said Dan Weaver, Mayor of Carlsbad. "But we knew there were people down there. When the water hit the cliffs, it was like a collision between two solid bodies. We couldn't see for over an hour, because of the spray." Salt spray. The ocean had come to New Mexico. The cliffs proved to be the only effective barrier against the westward march of the water, which turned north, gouging out lumps of rock and tumbling down blocks of earth onto its own back. In places scoops of granite came out like ice cream. The present fishing town of Rockport, Colorado, is built on a harbor created in such a way. The water had found its farthest westering. But still it poured north along the line of the original Fault. Irresistible fingers closed on Sterling, Colorado, on Sidney, Nebraska, on Hot Springs, South Dakota. The entire tier of states settled, from south to north, down to its eventual place of stability one thousand feet below the level of the new sea. Memphis was by now a seaport. The Ozarks, islands in a mad sea, formed precarious havens for half-drowned humanity. Waves bit off a corner of Missouri, flung themselves on Wichita. Topeka, Lawrence and Belleville were the last Kansas towns to disappear. The Governor of Kansas went down with his State. Daniel Bernd of Lincoln, Nebraska, was washed up half-drowned in a cove of the Wyoming cliffs, having been sucked from one end of vanished Nebraska to the other. Similar hair-breadth escapes were recounted on radio and television. Virtually the only people saved out of the entire population of Pierre, South Dakota were the six members of the Creeth family. Plucky Timothy Creeth carried and dragged his aged parents to the loft of their barn on the outskirts of town. His brother Geoffrey brought along the younger children and what provisions they could find—"Mostly a ham and about half a ton of vanilla cookies," he explained to his eventual rescuers. The barn, luckily collapsing in the vibrations as the waves bore down on them, became an ark in which they rode out the disaster. "We must of played cards for four days straight," recalled genial Mrs. Creeth when she afterwards appeared on a popular television spectacular. Her rural good-humor undamaged by an ordeal few women can ever have been called on to face, she added, "We sure wondered why flushes never came out right. Jimanettly, we'd left the king of hearts behind, in the rush!" But such lightheartedness and such happy endings were by no means typical. The world could only watch aghast as the water raced north under the shadow of the cliffs which occasionally crumbled, roaring, into the roaring waves. Day by day the relentless rush swallowed what had been dusty farmland, cities and towns. Some people were saved by the helicopters which flew mercy missions just ahead of the advancing waters. Some found safety in the peaks of western Nebraska and the Dakotas. But when the waters came to rest along what is roughly the present shoreline of our inland sea, it was estimated that over fourteen million people had lost their lives. No one could even estimate the damage to property; almost the entirety of eight states, and portions of twelve others, had simply vanished from the heart of the North American continent forever. It was in such a cataclysmic birth that the now-peaceful Nebraska Sea came to America. Today, nearly one hundred years after the unprecedented—and happily unrepeated—disaster, it is hard to remember the terror and despair of those weeks in October and November, 1973. It is inconceivable to think of the United States without its beautiful and economically essential curve of interior ocean. Two-thirds as long as the Mediterranean, it graduates from the warm waters of the Gulf of Mexico through the equally blue waves of the Mississippi Bight, becoming cooler and greener north and west of the pleasant fishing isles of the Ozark Archipelago, finally shading into the gray-green chop of the Gulf of Dakota. What would the United States have become without the 5600-mile coastline of our inland sea? It is only within the last twenty years that any but the topmost layer of water has cleared sufficiently to permit a really extensive fishing industry. Mud still held in suspension by the restless waves will not precipitate fully even in our lifetimes. Even so, the commercial fisheries of Missouri and Wyoming contribute no small part to the nation's economy. Who can imagine what the middle west must have been like before the amelioration of climate brought about by the proximity of a warm sea? The now-temperate state of Minnesota (to say nothing of the submerged Dakotas) must have been Siberian. From contemporary accounts Missouri, our second California, was unbelievably muggy, almost uninhabitable during the summer months. Our climate today, from Ohio and North Carolina to the rich fields of New Mexico and the orchards of Montana, is directly ameliorated by the marine heart of the continent. Who today could imagine the United States without the majestic sea-cliffs in stately parade from New Mexico to Montana? The beaches of Wyoming, the American Riviera, where fruit trees grow almost to the water's edge? Or incredible Colorado, where the morning skier is the afternoon bather, thanks to the monorail connecting the highest peaks with the glistening white beaches? Of course there have been losses to balance slightly these strong gains. The Mississippi was, before 1973, one of the great rivers of the world. Taken together with its main tributary, the Missouri, it vied favorably with such giant systems as the Amazon and the Ganges. Now, ending as it does at Memphis and drawing its water chiefly from the Appalachian Mountains, it is only a slight remnant of what it was. And though the Nebraska Sea today carries many times the tonnage of shipping in its ceaseless traffic, we have lost the old romance of river shipping. We may only guess what it was like when we look upon the Ohio and the truncated Mississippi. And transcontinental shipping is somewhat more difficult, with trucks and the freight-railroads obliged to take the sea-ferries across the Nebraska Sea. We shall never know what the United States was like with its numerous coast-to-coast highways busy with trucks and private cars. Still, the ferry ride is certainly a welcome break after days of driving, and for those who wish a glimpse of what it must have been like, there is always the Cross-Canada Throughway and the magnificent U. S. Highway 73 looping north through Minnesota and passing through the giant port of Alexis, North Dakota, shipping center for the wheat of Manitoba and crossroad of a nation. The political situation has long been a thorny problem. Only tattered remnants of the eight submerged states remained after the flood, but none of them wanted to surrender its autonomy. The tiny fringe of Kansas seemed, for a time, ready to merge with contiguous Missouri, but following the lead of the Arkansas Forever faction, the remaining population decided to retain political integrity. This has resulted in the continuing anomaly of the seven "fringe States" represented in Congress by the usual two Senators each, though the largest of them is barely the size of Connecticut and all are economically indistinguishable from their neighboring states. Fortunately it was decided some years ago that Oklahoma, only one of the eight to have completely disappeared, could not in any sense be considered to have a continuing political existence. So, though there are still families who proudly call themselves Oklahomans, and the Oklahoma Oil Company continues to pump oil from its submerged real estate, the state has in fact disappeared from the American political scene. But this is by now no more than a petty annoyance, to raise a smile when the talk gets around to the question of State's Rights. Not even the tremendous price the country paid for its new sea—fourteen million dead, untold property destroyed—really offsets the asset we enjoy today. The heart of the continent, now open to the shipping of the world, was once dry and land-locked, cut off from the bustle of trade and the ferment of world culture. It would indeed seem odd to an American of the '50s or '60s of the last century to imagine sailors from the merchant fleets of every nation walking the streets of Denver, fresh ashore at Newport, only fifteen miles away. Or to imagine Lincoln, Fargo, Kansas City and Dallas as world ports and great manufacturing centers. Utterly beyond their ken would be Roswell, New Mexico; Benton, Wyoming; Westport, Missouri, and the other new ports of over a million inhabitants each which have developed on the new harbors of the inland sea. Unimaginable too would have been the general growth of population in the states surrounding the new sea. As the water tables rose and manufacturing and trade moved in to take advantage of the just-created axis of world communication, a population explosion was touched off of which we are only now seeing the diminution. This new westering is to be ranked with the first surge of pioneers which created the American west. But what a difference! Vacation paradises bloom, a new fishing industry thrives; her water road is America's main artery of trade, and fleets of all the world sail ... where once the prairie schooner made its laborious and dusty way west!
|
B. The damage would be too substantial and there would nothing left to salvage.
|
Lanfierre has all of the following beliefs about humanity EXCEPT:
A. They are oblivious
B. They are self-absorbed
C. They are untrustworthy
D. They are vapid
|
A FALL OF GLASS By STANLEY R. LEE Illustrated by DILLON [Transcriber's Note: This etext was produced from Galaxy Magazine October 1960. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] The weatherman was always right: Temperature, 59; humidity, 47%; occasional light showers—but of what? The pockets of Mr. Humphrey Fownes were being picked outrageously. It was a splendid day. The temperature was a crisp 59 degrees, the humidity a mildly dessicated 47%. The sun was a flaming orange ball in a cloudless blue sky. His pockets were picked eleven times. It should have been difficult. Under the circumstances it was a masterpiece of pocket picking. What made it possible was Humphrey Fownes' abstraction; he was an uncommonly preoccupied individual. He was strolling along a quiet residential avenue: small private houses, one after another, a place of little traffic and minimum distractions. But he was thinking about weather, which was an unusual subject to begin with for a person living in a domed city. He was thinking so deeply about it that it never occurred to him that entirely too many people were bumping into him. He was thinking about Optimum Dome Conditions (a crisp 59 degrees, a mildly dessicated 47%) when a bogus postman, who pretended to be reading a postal card, jostled him. In the confusion of spilled letters and apologies from both sides, the postman rifled Fownes's handkerchief and inside jacket pockets. He was still thinking about temperature and humidity when a pretty girl happened along with something in her eye. They collided. She got his right and left jacket pockets. It was much too much for coincidence. The sidewalk was wide enough to allow four people to pass at one time. He should surely have become suspicious when two men engaged in a heated argument came along. In the ensuing contretemps they emptied his rear pants pockets, got his wristwatch and restored the contents of the handkerchief pocket. It all went off very smoothly, like a game of put and take—the sole difference being that Humphrey Fownes had no idea he was playing. There was an occasional tinkle of falling glass. It fell on the streets and houses, making small geysers of shiny mist, hitting with a gentle musical sound, like the ephemeral droppings of a celesta. It was precipitation peculiar to a dome: feather-light fragments showering harmlessly on the city from time to time. Dome weevils, their metal arms reaching out with molten glass, roamed the huge casserole, ceaselessly patching and repairing. Humphrey Fownes strode through the puffs of falling glass still intrigued by a temperature that was always 59 degrees, by a humidity that was always 47%, by weather that was always Optimum. It was this rather than skill that enabled the police to maintain such a tight surveillance on him, a surveillance that went to the extent of getting his fingerprints off the postman's bag, and which photographed, X-rayed and chemically analyzed the contents of his pockets before returning them. Two blocks away from his home a careless housewife spilled a five-pound bag of flour as he was passing. It was really plaster of Paris. He left his shoe prints, stride measurement, height, weight and handedness behind. By the time Fownes reached his front door an entire dossier complete with photographs had been prepared and was being read by two men in an orange patrol car parked down the street. Lanfierre had undoubtedly been affected by his job. Sitting behind the wheel of the orange car, he watched Humphrey Fownes approach with a distinct feeling of admiration, although it was an odd, objective kind of admiration, clinical in nature. It was similar to that of a pathologist observing for the first time a new and particularly virulent strain of pneumococcus under his microscope. Lanfierre's job was to ferret out aberration. It couldn't be tolerated within the confines of a dome. Conformity had become more than a social force; it was a physical necessity. And, after years of working at it, Lanfierre had become an admirer of eccentricity. He came to see that genuine quirks were rare and, as time went on, due partly to his own small efforts, rarer. Fownes was a masterpiece of queerness. He was utterly inexplicable. Lanfierre was almost proud of Humphrey Fownes. "Sometimes his house shakes ," Lanfierre said. "House shakes," Lieutenant MacBride wrote in his notebook. Then he stopped and frowned. He reread what he'd just written. "You heard right. The house shakes ," Lanfierre said, savoring it. MacBride looked at the Fownes house through the magnifying glass of the windshield. "Like from ... side to side ?" he asked in a somewhat patronizing tone of voice. "And up and down." MacBride returned the notebook to the breast pocket of his orange uniform. "Go on," he said, amused. "It sounds interesting." He tossed the dossier carelessly on the back seat. Lanfierre sat stiffly behind the wheel, affronted. The cynical MacBride couldn't really appreciate fine aberrations. In some ways MacBride was a barbarian. Lanfierre had held out on Fownes for months. He had even contrived to engage him in conversation once, a pleasantly absurd, irrational little chat that titillated him for weeks. It was only with the greatest reluctance that he finally mentioned Fownes to MacBride. After years of searching for differences Lanfierre had seen how extraordinarily repetitious people were, echoes really, dimly resounding echoes, each believing itself whole and separate. They spoke in an incessant chatter of cliches, and their actions were unbelievably trite. Then a fine robust freak came along and the others—the echoes—refused to believe it. The lieutenant was probably on the point of suggesting a vacation. "Why don't you take a vacation?" Lieutenant MacBride suggested. "It's like this, MacBride. Do you know what a wind is? A breeze? A zephyr?" "I've heard some." "They say there are mountain-tops where winds blow all the time. Strong winds, MacBride. Winds like you and I can't imagine. And if there was a house sitting on such a mountain and if winds did blow, it would shake exactly the way that one does. Sometimes I get the feeling the whole place is going to slide off its foundation and go sailing down the avenue." Lieutenant MacBride pursed his lips. "I'll tell you something else," Lanfierre went on. "The windows all close at the same time. You'll be watching and all of a sudden every single window in the place will drop to its sill." Lanfierre leaned back in the seat, his eyes still on the house. "Sometimes I think there's a whole crowd of people in there waiting for a signal—as if they all had something important to say but had to close the windows first so no one could hear. Why else close the windows in a domed city? And then as soon as the place is buttoned up they all explode into conversation—and that's why the house shakes." MacBride whistled. "No, I don't need a vacation." A falling piece of glass dissolved into a puff of gossamer against the windshield. Lanfierre started and bumped his knee on the steering wheel. "No, you don't need a rest," MacBride said. "You're starting to see flying houses, hear loud babbling voices. You've got winds in your brain, Lanfierre, breezes of fatigue, zephyrs of irrationality—" At that moment, all at once, every last window in the house slammed shut. The street was deserted and quiet, not a movement, not a sound. MacBride and Lanfierre both leaned forward, as if waiting for the ghostly babble of voices to commence. The house began to shake. It rocked from side to side, it pitched forward and back, it yawed and dipped and twisted, straining at the mooring of its foundation. The house could have been preparing to take off and sail down the.... MacBride looked at Lanfierre and Lanfierre looked at MacBride and then they both looked back at the dancing house. "And the water ," Lanfierre said. "The water he uses! He could be the thirstiest and cleanest man in the city. He could have a whole family of thirsty and clean kids, and he still wouldn't need all that water." The lieutenant had picked up the dossier. He thumbed through the pages now in amazement. "Where do you get a guy like this?" he asked. "Did you see what he carries in his pockets?" "And compasses won't work on this street." The lieutenant lit a cigarette and sighed. He usually sighed when making the decision to raid a dwelling. It expressed his weariness and distaste for people who went off and got neurotic when they could be enjoying a happy, normal existence. There was something implacable about his sighs. "He'll be coming out soon," Lanfierre said. "He eats supper next door with a widow. Then he goes to the library. Always the same. Supper at the widow's next door and then the library." MacBride's eyebrows went up a fraction of an inch. "The library?" he said. "Is he in with that bunch?" Lanfierre nodded. "Should be very interesting," MacBride said slowly. "I can't wait to see what he's got in there," Lanfierre murmured, watching the house with a consuming interest. They sat there smoking in silence and every now and then their eyes widened as the house danced a new step. Fownes stopped on the porch to brush the plaster of paris off his shoes. He hadn't seen the patrol car and this intense preoccupation of his was also responsible for the dancing house—he simply hadn't noticed. There was a certain amount of vibration, of course. He had a bootleg pipe connected into the dome blower system, and the high-pressure air caused some buffeting against the thin walls of the house. At least, he called it buffeting; he'd never thought to watch from outside. He went in and threw his jacket on the sofa, there being no room left in the closets. Crossing the living room he stopped to twist a draw-pull. Every window slammed shut. "Tight as a kite," he thought, satisfied. He continued on toward the closet at the foot of the stairs and then stopped again. Was that right? No, snug as a hug in a rug . He went on, thinking: The old devils. The downstairs closet was like a great watch case, a profusion of wheels surrounding the Master Mechanism, which was a miniature see-saw that went back and forth 365-1/4 times an hour. The wheels had a curious stateliness about them. They were all quite old, salvaged from grandfather's clocks and music boxes and they went around in graceful circles at the rate of 30 and 31 times an hour ... although there was one slightly eccentric cam that vacillated between 28 and 29. He watched as they spun and flashed in the darkness, and then set them for seven o'clock in the evening, April seventh, any year. Outside, the domed city vanished. It was replaced by an illusion. Or, as Fownes hoped it might appear, the illusion of the domed city vanished and was replaced by a more satisfactory, and, for his specific purpose, more functional, illusion. Looking through the window he saw only a garden. Instead of an orange sun at perpetual high noon, there was a red sun setting brilliantly, marred only by an occasional arcover which left the smell of ozone in the air. There was also a gigantic moon. It hid a huge area of sky, and it sang. The sun and moon both looked down upon a garden that was itself scintillant, composed largely of neon roses. Moonlight, he thought, and roses. Satisfactory. And cocktails for two. Blast, he'd never be able to figure that one out! He watched as the moon played, Oh, You Beautiful Doll and the neon roses flashed slowly from red to violet, then went back to the closet and turned on the scent. The house began to smell like an immensely concentrated rose as the moon shifted to People Will Say We're In Love . He rubbed his chin critically. It seemed all right. A dreamy sunset, an enchanted moon, flowers, scent. They were all purely speculative of course. He had no idea how a rose really smelled—or looked for that matter. Not to mention a moon. But then, neither did the widow. He'd have to be confident, assertive. Insist on it. I tell you, my dear, this is a genuine realistic romantic moon. Now, does it do anything to your pulse? Do you feel icy fingers marching up and down your spine? His own spine didn't seem to be affected. But then he hadn't read that book on ancient mores and courtship customs. How really odd the ancients were. Seduction seemed to be an incredibly long and drawn-out process, accompanied by a considerable amount of falsification. Communication seemed virtually impossible. "No" meant any number of things, depending on the tone of voice and the circumstances. It could mean yes, it could mean ask me again later on this evening. He went up the stairs to the bedroom closet and tried the rain-maker, thinking roguishly: Thou shalt not inundate. The risks he was taking! A shower fell gently on the garden and a male chorus began to chant Singing in the Rain . Undiminished, the yellow moon and the red sun continued to be brilliant, although the sun occasionally arced over and demolished several of the neon roses. The last wheel in the bedroom closet was a rather elegant steering wheel from an old 1995 Studebaker. This was on the bootleg pipe; he gingerly turned it. Far below in the cellar there was a rumble and then the soft whistle of winds came to him. He went downstairs to watch out the living room window. This was important; the window had a really fixed attitude about air currents. The neon roses bent and tinkled against each other as the wind rose and the moon shook a trifle as it whispered Cuddle Up a Little Closer . He watched with folded arms, considering how he would start. My dear Mrs. Deshazaway. Too formal. They'd be looking out at the romantic garden; time to be a bit forward. My very dear Mrs. Deshazaway. No. Contrived. How about a simple, Dear Mrs. Deshazaway . That might be it. I was wondering, seeing as how it's so late, if you wouldn't rather stay over instead of going home.... Preoccupied, he hadn't noticed the winds building up, didn't hear the shaking and rattling of the pipes. There were attic pipes connected to wall pipes and wall pipes connected to cellar pipes, and they made one gigantic skeleton that began to rattle its bones and dance as high-pressure air from the dome blower rushed in, slowly opening the Studebaker valve wider and wider.... The neon roses thrashed about, extinguishing each other. The red sun shot off a mass of sparks and then quickly sank out of sight. The moon fell on the garden and rolled ponderously along, crooning When the Blue of the Night Meets the Gold of the Day . The shaking house finally woke him up. He scrambled upstairs to the Studebaker wheel and shut it off. At the window again, he sighed. Repairs were in order. And it wasn't the first time the winds got out of line. Why didn't she marry him and save all this bother? He shut it all down and went out the front door, wondering about the rhyme of the months, about stately August and eccentric February and romantic April. April. Its days were thirty and it followed September. And all the rest have thirty-one. What a strange people, the ancients! He still didn't see the orange car parked down the street. "Men are too perishable," Mrs. Deshazaway said over dinner. "For all practical purposes I'm never going to marry again. All my husbands die." "Would you pass the beets, please?" Humphrey Fownes said. She handed him a platter of steaming red beets. "And don't look at me that way," she said. "I'm not going to marry you and if you want reasons I'll give you four of them. Andrew. Curt. Norman. And Alphonse." The widow was a passionate woman. She did everything passionately—talking, cooking, dressing. Her beets were passionately red. Her clothes rustled and her high heels clicked and her jewelry tinkled. She was possessed by an uncontrollable dynamism. Fownes had never known anyone like her. "You forgot to put salt on the potatoes," she said passionately, then went on as calmly as it was possible for her to be, to explain why she couldn't marry him. "Do you have any idea what people are saying? They're all saying I'm a cannibal! I rob my husbands of their life force and when they're empty I carry their bodies outside on my way to the justice of the peace." "As long as there are people," he said philosophically, "there'll be talk." "But it's the air! Why don't they talk about that? The air is stale, I'm positive. It's not nourishing. The air is stale and Andrew, Curt, Norman and Alphonse couldn't stand it. Poor Alphonse. He was never so healthy as on the day he was born. From then on things got steadily worse for him." "I don't seem to mind the air." She threw up her hands. "You'd be the worst of the lot!" She left the table, rustling and tinkling about the room. "I can just hear them. Try some of the asparagus. Five. That's what they'd say. That woman did it again. And the plain fact is I don't want you on my record." "Really," Fownes protested. "I feel splendid. Never better." He could hear her moving about and then felt her hands on his shoulders. "And what about those very elaborate plans you've been making to seduce me?" Fownes froze with three asparagus hanging from his fork. "Don't you think they'll find out? I found out and you can bet they will. It's my fault, I guess. I talk too much. And I don't always tell the truth. To be completely honest with you, Mr. Fownes, it wasn't the old customs at all standing between us, it was air. I can't have another man die on me, it's bad for my self-esteem. And now you've gone and done something good and criminal, something peculiar." Fownes put his fork down. "Dear Mrs. Deshazaway," he started to say. "And of course when they do find out and they ask you why, Mr. Fownes, you'll tell them. No, no heroics, please! When they ask a man a question he always answers and you will too. You'll tell them I wanted to be courted and when they hear that they'll be around to ask me a few questions. You see, we're both a bit queer." "I hadn't thought of that," Fownes said quietly. "Oh, it doesn't really matter. I'll join Andrew, Curt, Norman—" "That won't be necessary," Fownes said with unusual force. "With all due respect to Andrew, Curt, Norman and Alphonse, I might as well state here and now I have other plans for you, Mrs. Deshazaway." "But my dear Mr. Fownes," she said, leaning across the table. "We're lost, you and I." "Not if we could leave the dome," Fownes said quietly. "That's impossible! How?" In no hurry, now that he had the widow's complete attention, Fownes leaned across the table and whispered: "Fresh air, Mrs. Deshazaway? Space? Miles and miles of space where the real-estate monopoly has no control whatever? Where the wind blows across prairies ; or is it the other way around? No matter. How would you like that , Mrs. Deshazaway?" Breathing somewhat faster than usual, the widow rested her chin on her two hands. "Pray continue," she said. "Endless vistas of moonlight and roses? April showers, Mrs. Deshazaway. And June, which as you may know follows directly upon April and is supposed to be the month of brides, of marrying. June also lies beyond the dome." "I see." " And ," Mr. Fownes added, his voice a honeyed whisper, "they say that somewhere out in the space and the roses and the moonlight, the sleeping equinox yawns and rises because on a certain day it's vernal and that's when it roams the Open Country where geigers no longer scintillate." " My. " Mrs. Deshazaway rose, paced slowly to the window and then came back to the table, standing directly over Fownes. "If you can get us outside the dome," she said, "out where a man stays warm long enough for his wife to get to know him ... if you can do that, Mr. Fownes ... you may call me Agnes." When Humphrey Fownes stepped out of the widow's house, there was a look of such intense abstraction on his features that Lanfierre felt a wistful desire to get out of the car and walk along with the man. It would be such a deliciously insane experience. ("April has thirty days," Fownes mumbled, passing them, "because thirty is the largest number such that all smaller numbers not having a common divisor with it are primes ." MacBride frowned and added it to the dossier. Lanfierre sighed.) Pinning his hopes on the Movement, Fownes went straight to the library several blocks away, a shattered depressing place given over to government publications and censored old books with holes in them. It was used so infrequently that the Movement was able to meet there undisturbed. The librarian was a yellowed, dog-eared woman of eighty. She spent her days reading ancient library cards and, like the books around her, had been rendered by time's own censor into near unintelligibility. "Here's one," she said to him as he entered. " Gulliver's Travels. Loaned to John Wesley Davidson on March 14, 1979 for five days. What do you make of it?" In the litter of books and cards and dried out ink pads that surrounded the librarian, Fownes noticed a torn dust jacket with a curious illustration. "What's that?" he said. "A twister," she replied quickly. "Now listen to this . Seven years later on March 21, 1986, Ella Marshall Davidson took out the same book. What do you make of that ?" "I'd say," Humphrey Fownes said, "that he ... that he recommended it to her, that one day they met in the street and he told her about this book and then they ... they went to the library together and she borrowed it and eventually, why eventually they got married." "Hah! They were brother and sister!" the librarian shouted in her parched voice, her old buckram eyes laughing with cunning. Fownes smiled weakly and looked again at the dust jacket. The twister was unquestionably a meteorological phenomenon. It spun ominously, like a malevolent top, and coursed the countryside destructively, carrying a Dorothy to an Oz. He couldn't help wondering if twisters did anything to feminine pulses, if they could possibly be a part of a moonlit night, with cocktails and roses. He absently stuffed the dust jacket in his pocket and went on into the other rooms, the librarian mumbling after him: "Edna Murdoch Featherstone, April 21, 1991," as though reading inscriptions on a tombstone. The Movement met in what had been the children's room, where unpaid ladies of the afternoon had once upon a time read stories to other people's offspring. The members sat around at the miniature tables looking oddly like giants fled from their fairy tales, protesting. "Where did the old society fail?" the leader was demanding of them. He stood in the center of the room, leaning on a heavy knobbed cane. He glanced around at the group almost complacently, and waited as Humphrey Fownes squeezed into an empty chair. "We live in a dome," the leader said, "for lack of something. An invention! What is the one thing that the great technological societies before ours could not invent, notwithstanding their various giant brains, electronic and otherwise?" Fownes was the kind of man who never answered a rhetorical question. He waited, uncomfortable in the tight chair, while the others struggled with this problem in revolutionary dialectics. " A sound foreign policy ," the leader said, aware that no one else had obtained the insight. "If a sound foreign policy can't be created the only alternative is not to have any foreign policy at all. Thus the movement into domes began— by common consent of the governments . This is known as self-containment." Dialectically out in left field, Humphrey Fownes waited for a lull in the ensuing discussion and then politely inquired how it might be arranged for him to get out. "Out?" the leader said, frowning. "Out? Out where?" "Outside the dome." "Oh. All in good time, my friend. One day we shall all pick up and leave." "And that day I'll await impatiently," Fownes replied with marvelous tact, "because it will be lonely out there for the two of us. My future wife and I have to leave now ." "Nonsense. Ridiculous! You have to be prepared for the Open Country. You can't just up and leave, it would be suicide, Fownes. And dialectically very poor." "Then you have discussed preparations, the practical necessities of life in the Open Country. Food, clothing, a weapon perhaps? What else? Have I left anything out?" The leader sighed. "The gentleman wants to know if he's left anything out," he said to the group. Fownes looked around at them, at some dozen pained expressions. "Tell the man what he's forgotten," the leader said, walking to the far window and turning his back quite pointedly on them. Everyone spoke at the same moment. " A sound foreign policy ," they all said, it being almost too obvious for words. On his way out the librarian shouted at him: " A Tale of a Tub , thirty-five years overdue!" She was calculating the fine as he closed the door. Humphrey Fownes' preoccupation finally came to an end when he was one block away from his house. It was then that he realized something unusual must have occurred. An orange patrol car of the security police was parked at his front door. And something else was happening too. His house was dancing. It was disconcerting, and at the same time enchanting, to watch one's residence frisking about on its foundation. It was such a strange sight that for the moment he didn't give a thought to what might be causing it. But when he stepped gingerly onto the porch, which was doing its own independent gavotte, he reached for the doorknob with an immense curiosity. The door flung itself open and knocked him back off the porch. From a prone position on his miniscule front lawn, Fownes watched as his favorite easy chair sailed out of the living room on a blast of cold air and went pinwheeling down the avenue in the bright sunshine. A wild wind and a thick fog poured out of the house. It brought chairs, suits, small tables, lamps trailing their cords, ashtrays, sofa cushions. The house was emptying itself fiercely, as if disgorging an old, spoiled meal. From deep inside he could hear the rumble of his ancient upright piano as it rolled ponderously from room to room. He stood up; a wet wind swept over him, whipping at his face, toying with his hair. It was a whistling in his ears, and a tingle on his cheeks. He got hit by a shoe. As he forced his way back to the doorway needles of rain played over his face and he heard a voice cry out from somewhere in the living room. "Help!" Lieutenant MacBride called. Standing in the doorway with his wet hair plastered down on his dripping scalp, the wind roaring about him, the piano rumbling in the distance like thunder, Humphrey Fownes suddenly saw it all very clearly. " Winds ," he said in a whisper. "What's happening?" MacBride yelled, crouching behind the sofa. " March winds," he said. "What?!" "April showers!" The winds roared for a moment and then MacBride's lost voice emerged from the blackness of the living room. "These are not Optimum Dome Conditions!" the voice wailed. "The temperature is not 59 degrees. The humidity is not 47%!" Fownes held his face up to let the rain fall on it. "Moonlight!" he shouted. "Roses! My soul for a cocktail for two!" He grasped the doorway to keep from being blown out of the house. "Are you going to make it stop or aren't you!" MacBride yelled. "You'll have to tell me what you did first!" "I told him not to touch that wheel! Lanfierre. He's in the upstairs bedroom!" When he heard this Fownes plunged into the house and fought his way up the stairs. He found Lanfierre standing outside the bedroom with a wheel in his hand. "What have I done?" Lanfierre asked in the monotone of shock. Fownes took the wheel. It was off a 1995 Studebaker. "I'm not sure what's going to come of this," he said to Lanfierre with an astonishing amount of objectivity, "but the entire dome air supply is now coming through my bedroom." The wind screamed. "Is there something I can turn?" Lanfierre asked. "Not any more there isn't." They started down the stairs carefully, but the wind caught them and they quickly reached the bottom in a wet heap. Recruiting Lieutenant MacBride from behind his sofa, the men carefully edged out of the house and forced the front door shut. The wind died. The fog dispersed. They stood dripping in the Optimum Dome Conditions of the bright avenue. "I never figured on this ," Lanfierre said, shaking his head. With the front door closed the wind quickly built up inside the house. They could see the furnishing whirl past the windows. The house did a wild, elated jig. "What kind of a place is this?" MacBride said, his courage beginning to return. He took out his notebook but it was a soggy mess. He tossed it away. "Sure, he was different ," Lanfierre murmured. "I knew that much." When the roof blew off they weren't really surprised. With a certain amount of equanimity they watched it lift off almost gracefully, standing on end for a moment before toppling to the ground. It was strangely slow motion, as was the black twirling cloud that now rose out of the master bedroom, spewing shorts and socks and cases every which way. " Now what?" MacBride said, thoroughly exasperated, as this strange black cloud began to accelerate, whirling about like some malevolent top.... Humphrey Fownes took out the dust jacket he'd found in the library. He held it up and carefully compared the spinning cloud in his bedroom with the illustration. The cloud rose and spun, assuming the identical shape of the illustration. "It's a twister," he said softly. "A Kansas twister!" "What," MacBride asked, his bravado slipping away again, "what ... is a twister?" The twister roared and moved out of the bedroom, out over the rear of the house toward the side of the dome. "It says here," Fownes shouted over the roaring, "that Dorothy traveled from Kansas to Oz in a twister and that ... and that Oz is a wonderful and mysterious land beyond the confines of everyday living ." MacBride's eyes and mouth were great zeros. "Is there something I can turn?" Lanfierre asked. Huge chunks of glass began to fall around them. "Fownes!" MacBride shouted. "This is a direct order! Make it go back!" But Fownes had already begun to run on toward the next house, dodging mountainous puffs of glass as he went. "Mrs. Deshazaway!" he shouted. "Yoo-hoo, Mrs. Deshazaway!" The dome weevils were going berserk trying to keep up with the precipitation. They whirred back and forth at frightful speed, then, emptied of molten glass, rushed to the Trough which they quickly emptied and then rushed about empty-handed. "Yoo-hoo!" he yelled, running. The artificial sun vanished behind the mushrooming twister. Optimum temperature collapsed. "Mrs. Deshazaway! Agnes , will you marry me? Yoo-hoo!" Lanfierre and Lieutenant MacBride leaned against their car and waited, dazed. There was quite a large fall of glass.
|
C. They are untrustworthy
|
What is their baseline?
|
### Introduction
Following a turbulent election season, 2016's cyber world is awash with hate speech. Automatic detection of hate speech has become an urgent need since human supervision is unable to deal with large quantities of emerging texts. Context information, by our definition, is the text, symbols or any other kind of information related to the original text. While intuitively, context accompanying hate speech is useful for detecting hate speech, context information of hate speech has been overlooked in existing datasets and automatic detection models. Online hate speech tends to be subtle and creative, which makes context especially important for automatic hate speech detection. For instance, (1) barryswallows: Merkel would never say NO This comment is posted for the News titled by "German lawmakers approve 'no means no' rape law after Cologne assaults". With context, it becomes clear that this comment is a vicious insult towards female politician. However, almost all the publicly available hate speech annotated datasets do not contain context information. BIBREF0 , BIBREF1 , BIBREF2 , BIBREF3 . We have created a new dataset consisting of 1528 Fox News user comments, which were taken from 10 complete discussion threads for 10 widely read Fox News articles. It is different from previous datasets from the following two perspectives. First, it preserves rich context information for each comment, including its user screen name, all comments in the same thread and the news article the comment is written for. Second, there is no biased data selection and all comments in each news comment thread were annotated. In this paper, we explored two types of models, feature-based logistic regression models and neural network models, in order to incorporate context information in automatic hate speech detection. First, logistic regression models have been used in several prior hate speech detection studies BIBREF4 , BIBREF5 , BIBREF6 , BIBREF7 , BIBREF8 , BIBREF0 , BIBREF2 , BIBREF9 and various features have been tried including character-level and word-level n-gram features, syntactic features, linguistic features, and comment embedding features. However, all the features were derived from the to-be-classified text itself. In contrast, we experiment with logistic regression models using features extracted from context text as well. Second, neural network models BIBREF10 , BIBREF11 , BIBREF12 have the potential to capture compositional meanings of text, but they have not been well explored for online hate speech detection until recently BIBREF13 . We experiment with neural net models containing separate learning components that model compositional meanings of context information. Furthermore, recognizing unique strengths of each type of models, we build ensemble models of the two types of models. Evaluation shows that context-aware logistic regression models and neural net models outperform their counterparts that are blind with context information. Especially, the final ensemble models outperform a strong baseline system by around 10% in F1-score. ### Related Works
Recently, a few datasets with human labeled hate speech have been created, however, most of existing datasets do not contain context information. Due to the sparsity of hate speech in everyday posts, researchers tend to sample candidates from bootstrapping instead of random sampling, in order to increase the chance of seeing hate speech. Therefore, the collected data instances are likely to be from distinct contexts. For instance, in the Primary Data Set described in BIBREF14 and later used by BIBREF9 , 10% of the dataset is randomly selected while the remaining consists of comments tagged by users and editors. BIBREF15 built a balanced data set of 24.5k tweets by selecting from Twitter accounts that claimed to be racist or were deemed racist using their followed news sources. BIBREF5 collected hateful tweets related to the murder of Drummer Lee Rigby in 2013. BIBREF0 provided a corpus of 16k annotated tweets in which 3.3k are labeled as sexist and 1.9k are labeled as racist. They created this corpus by bootstrapping from certain key words ,specific hashtags and certain prolific users. BIBREF16 created a dataset of 9000 human labeled paragraphs that were collected using regular expression matching in order to find hate speech targeting Judaism and Israel. BIBREF7 extracted data instances from instagram that were associated with certain user accounts. BIBREF2 presented a very large corpus containing over 115k Wikipedia comments that include around 37k randomly sampled comments and the remaining 78k comments were selected from Wikipedia blocked comments. Most of existing hate speech detection models are feature-based and use features derived from the target text itself. BIBREF5 experimented with different classification methods including Bayesian Logistic Regression, Random Forest Decision Trees and SVMs, using features such as n-grams, reduced n-grams, dependency paths, and hateful terms. BIBREF0 proposed a logistic regression model using character n-gram features. BIBREF14 used the paragraph2vec for joint modeling of comments and words, then the generated embeddings were used as feature in a logistic regression model. BIBREF9 experimented with various syntactic, linguistic and distributional semantic features including word length, sentence length, part of speech tags, and embedding features, in order to improve performance of logistic regression classifiers. Recently, BIBREF17 surveyed current approaches for hate speech detection, which interestingly also called to attention on modeling context information for resolving difficult hate speech instances. ### Corpus Overview
The Fox News User Comments corpus consists of 1528 annotated comments (435 labeled as hateful) that were posted by 678 different users in 10 complete news discussion threads in the Fox News website. The 10 threads were manually selected and represent popular discussion threads during August 2016. All of the comments included in these 10 threads were annotated. The number of comments in each of the 10 threads is roughly equal. Rich context information was kept for each comment, including its user screen name, the comments and their nested structure and the original news article. The data corpus along with annotation guidelines is posted on github. ### Annotation Guidelines
Our annotation guidelines are similar to the guidelines used by BIBREF9 . We define hateful speech to be the language which explicitly or implicitly threatens or demeans a person or a group based upon a facet of their identity such as gender, ethnicity, or sexual orientation. The labeling of hateful speech in our corpus is binary. A comment will be labeled as hateful or non-hateful. ### Annotation Procedure
We identified two native English speakers for annotating online user comments. The two annotators first discussed and practices before they started annotation. They achieved a surprisingly high Kappa score BIBREF18 of 0.98 on 648 comments from 4 threads. We think that thorough discussions in the training stage is the key for achieving this high inter-agreement. For those comments which annotators disagreed on, we label them as hateful as long as one annotator labeled them as hateful. Then one annotator continued to annotate the remaining 880 comments from the remaining six discussion threads. ### Characteristics in Fox News User Comments corpus
Hateful comments in the Fox News User Comments Corpus is often subtle, creative and implicit. Therefore, context information is necessary in order to accurately identify such hate speech. The hatefulness of many comments depended on understanding their contexts. For instance, (3) mastersundholm: Just remember no trabjo no cervesa This comment is posted for the news "States moving to restore work requirements for food stamp recipients". This comment implies that Latino immigrants abuse the usage of food stamp policy, which is clearly a stereotyping. Many hateful comments use implicit and subtle language, which contain no clear hate indicating word or phrase. In order to recognize such hard cases, we hypothesize that neural net models are more suitable by capturing overall composite meanings of a comment. For instance, the following comment is a typical implicit stereotyping against women. (4) MarineAssassin: Hey Brianne - get in the kitchen and make me a samich. Chop Chop 11% of our annotated comments have more than 50 words each. In such long comments, the hateful indicators usually appear in a small region of a comment while the majority of the comment is neutral. For example, (5) TMmckay: I thought ...115 words... Too many blacks winning, must be racist and needs affirmative action to make whites equally win! Certain user screen names indicate hatefulness, which imply that comments posted by these users are likely to contain hate speech. In the following example, commie is a slur for communists. (6)nocommie11: Blah blah blah. Israel is the only civilized nation in the region to keep the unwashed masses at bay. ### Logistic Regression Models
In logistic regression models, we extract four types of features, word-level and character-level n-gram features as well as two types of lexicon derived features. We extract these four types of features from the target comment first. Then we extract these features from two sources of context texts, specifically the title of the news article that the comment was posted for and the screen name of the user who posted the comment. For logistic regression model implementation, we use l2 loss. We adopt the balanced class weight as described in Scikit learn. Logistic regression model with character-level n-gram features is presented as a strong baseline for comparison since it was shown very effective. BIBREF0 , BIBREF9 For character level n-grams, we extract character level bigrams, tri-grams and four-grams. For word level n-grams, we extract unigrams and bigrams. Linguistic Inquiry and Word Count, also called LIWC, has been proven useful for text analysis and classification BIBREF19 . In the LIWC dictionary, each word is labeled with several semantic labels. In our experiment, we use the LIWC 2015 dictionary which contain 125 semantic categories. Each word is converted into a 125 dimension LIWC vector, one dimension per semantic category. The LIWC feature vector for a comment or its context is a 125 dimension vector as well, which is the sum of all its words' LIWC vectors. NRC emotion lexicon contains a list of English words that were labeled with eight basic emotions (anger, fear, anticipation, trust, surprise, sadness, joy, and disgust) and sentiment polarities (negative and positive) BIBREF20 . We use NRC emotion lexicon to capture emotion clues in text. Each word is converted into a 10 dimension emotion vector, corresponding to eight emotion types and two polarity labels. The emotion vector for a comment or its context is a 10 dimension vector as well, which is the sum of all its words' emotion vectors. As shown in table TABREF20 , given comment as the only input content, the combination of character n-grams, word n-grams, LIWC feature and NRC feature achieves the best performance. It shows that in addition to character level features, adding more features can improve hate speech detection performance. However, the improvement is limited. Compared with baseline model, the F1 score only improves 1.3%. In contrast, when context information was taken into account, the performance greatly improved. Specifically, after incorporating features extracted from the news title and username, the model performance was improved by around 4% in both F1 score and AUC score. This shows that using additional context based features in logistic regression models is useful for hate speech detection. ### Neural Network Models
Our neural network model mainly consists of three parallel LSTM BIBREF21 layers. It has three different inputs, including the target comment, its news title and its username. Comment and news title are encoded into a sequence of word embeddings. We use pre-trained word embeddings in word2vec. Username is encoded into a sequence of characters. We use one-hot encoding of characters. Comment is sent into a bi-directional LSTM with attention mechanism. BIBREF22 . News title and username are sent into a bi-directional LSTM. Note that we did not apply attention mechanism to the neural network models for username and news title because both types of context are relatively short and attention mechanism tends to be useful when text input is long. The three LSTM output layers are concatenated, then connected to a sigmoid layer, which outputs predictions. The number of hidden units in each LSTM used in our model is set to be 100. The recurrent dropout rate of LSTMs is set to 0.2. In addition, we use binary cross entropy as the loss function and a batch size of 128. The neural network models are trained for 30 epochs. As shown in table TABREF21 , given comment as the only input content, the bi-directional LSTM model with attention mechanism achieves the best performance. Note that the attention mechanism significantly improves the hate speech detection performance of the bi-directional LSTM model. We hypothesize that this is because hate indicator phrases are often concentrated in a small region of a comment, which is especially the case for long comments. ### Ensemble Models
To study the difference of logistic regression model and neural network model and potentially get performance improvement, we will build and evaluate ensemble models. As shown in table TABREF24 , both ensemble models significantly improved hate speech detection performance. Figure FIGREF28 shows the system prediction results of comments that were labeled as hateful in the dataset. It can be seen that the two models perform differently. We further examined predicted comments and find that both types of models have unique strengths in identifying certain types of hateful comments. The feature-based logistic regression models are capable of making good use of character-level n-gram features, which are powerful in identifying hateful comments that contains OOV words, capitalized words or misspelled words. We provide two examples from the hateful comments that were only labeled by the logistic regression model: (7)kmawhmf:FBLM. Here FBLM means fuck Black Lives Matter. This hateful comment contains only character information which can exactly be made use of by our logistic regression model. (8)SFgunrmn: what a efen loon, but most femanazis are. This comment deliberately misspelled feminazi for femanazis, which is a derogatory term for feminists. It shows that logistic regression model is capable in dealing with misspelling. The LSTM with attention mechanism are suitable for identifying specific small regions indicating hatefulness in long comments. In addition, the neural net models are powerful in capturing implicit hateful language as well. The following are two hateful comment examples that were only identified by the neural net model: (9)freedomscout: @LarJass Many religions are poisonous to logic and truth, that much is true...and human beings still remain fallen human beings even they are Redeemed by the Sacrifice of Jesus Christ. So there's that. But the fallacies of thinking cannot be limited or attributed to religion but to error inherent in human motivation, the motivation to utter self-centeredness as fallen sinful human beings. Nearly all of the world's many religions are expressions of that utter sinful nature...Christianity and Judaism being the sole exceptions. This comment is expressing the stereotyping against religions which are not Christian or Judaism. The hatefulness is concentrated within the two bolded segments. (10)mamahattheridge: blacks Love being victims. In this comment, the four words themselves are not hateful at all. But when combined together, it is clearly hateful against black people. ### Evaluation
We evaluate our model by 10 fold cross validation using our newly created Fox News User Comments Corpus. Both types of models use the exact same 10 folds of training data and test data. We report experimental results using multiple metrics, including accuracy, precision/recall/F1-score, and accuracy area under curve (AUC). ### Experimental Results
Table TABREF20 shows the performance of logistic regression models. The first section of table TABREF20 shows the performance of logistic regression models using features extracted from a target comment only. The result shows that the logistic regression model was improved in every metric after adding both word-level n-gram features and lexicon derived features. However, the improvements are moderate. The second section shows the performance of logistic regression models using the four types of features extracted from both a target comment and its contextsThe result shows that the logistic regression model using features extracted from a comment and both types of context achieved the best performance and obtained improvements of 2.8% and 2.5% in AUC score and F1-score respectively. Table TABREF21 shows the performance of neural network models. The first section of table TABREF21 shows the performance of several neural network models that use comments as the only input. The model names are self-explanatory. We can see that the attention mechanism coupled with the bi-directional LSTM neural net greatly improved the online hate speech detection, by 5.7% in AUC score. The second section of table TABREF21 shows performance of the best neural net model (bi-directional LSTM with attention) after adding additional learning components that take context as input. The results show that adding username and news title can both improve model performance. Using news title gives the best F1 score while using both news title and username gives the best AUC score. Table TABREF24 shows performance of ensemble models by combining prediction results of the best context-aware logistic regression model and the best context-aware neural network model. We used two strategies in combining prediction results of two types of models. Specifically, the Max Score Ensemble model made the final decisions based on the maximum of two scores assigned by the two separate models; instead, the Average Score Ensemble model used the average score to make final decisions. We can see that both ensemble models further improved hate speech detection performance compared with using one model only and achieved the best classification performance. Compared with the logistic regression baseline, the Max Score Ensemble model improved the recall by more than 20% with a comparable precision and improved the F1 score by around 10%, in addition, the Average Score Ensemble model improved the AUC score by around 7%. ### Conclusion
We demonstrated the importance of utilizing context information for online hate speech detection. We first presented a corpus of hateful speech consisting of full threads of online discussion posts. In addition, we presented two types of models, feature-based logistic regression models and neural network models, in order to incorporate context information for improving hate speech detection performance. Furthermore, we show that ensemble models leveraging strengths of both types of models achieve the best performance for automatic online hate speech detection. Table 1: Performance of Logistic Regression Models Table 2: Performance of Neural Network Models Table 3: Performance of Ensemble Models Figure 1: System Prediction Results of Comments that were Annotated as Hateful
|
Logistic regression model with character-level n-gram features
|
What are the three languages studied in the paper?
|
### Introduction
Neural machine translation (NMT) systems are conventionally trained based on the approach of maximizing the log-likelihood on a training corpus in order to learn distributed representations of words according to their sentence context, which is highly demanding in terms of training data as well as the network capacity. Under conditions of lexical sparsity, which may include the cases when the amount of training examples is insufficient to observe words in different context, and particularly in translation of morphologically-rich languages, where the same word can have exponentially many different surface realizations due to syntactic conditions, which are often rarely or ever observed in any set of collected examples, the model may suffer in learning accurate representations of words. The standard approach to overcome this limitation is to replace the word representations in the model with subword units that are shared among words, which are, in principle, more reliable as they are observed more frequently in varying context BIBREF0, BIBREF1. One drawback related to this approach, however, is that the estimation of the subword vocabulary relies on word segmentation methods optimized using corpus-dependent statistics, disregarding any linguistic notion and the translation objective, which may result in morphological errors during splitting, resulting in subword units that are semantically ambiguous as they might be used in far too many lexical contexts BIBREF2. Moreover, the words are generated predicting multiple subword units, which makes generalizing to unseen word forms more difficult, where some of the subword units that could be used to reconstruct a given word may be unlikely in the given context. To alleviate the sub-optimal effects of using explicit segmentation and generalize better to new morphological forms, recent studies explored the idea of extending the same approach to model translation directly at the level of characters BIBREF3, BIBREF4, which, in turn, have demonstrated the requirement of using comparably deeper networks, as the network would then need to learn longer distance grammatical dependencies BIBREF5. In this paper, we explore the benefit of explicitly modeling variations in the surface forms of words using methods from deep latent variable modeling in order to improve the translation accuracy in low-resource and morphologically-rich languages. Latent variable models allow us to inject inductive biases relevant to the task, which, in our case, is word formation, and we believe that follows a certain hierarchical procedure. Our model translates words one character at a time based on word representations learned compositionally from sub-lexical components, which are parameterized by a hierarchical latent variable model mimicking the process of morphological inflection, consisting of a continuous-space dense vector capturing the lexical semantics, and a set of (approximately) discrete features, representing the morphosyntactic role of the word in a given sentence. Each word representation during decoding is reformulated based on the shared latent morphological features, aiding in learning more reliable representations of words under sparse settings by generalizing across their different surface forms. We evaluate our method in translating English into three morphologically-rich languages each with a distinct morphological typology: Arabic, Czech and Turkish, and show that our model is able to obtain better translation accuracy and generalization capacity than conventional approaches to open-vocabulary NMT. ### Evaluation ::: Models
We evaluate our model by comparing it in machine translation against three baselines which constitute the conventional open-vocabulary NMT methods, including architectures using atomic parameterization either with subword units segmented with BPE BIBREF0 or characters, and the hierarchical parameterization method employed for generating all words in the output. We implement all architectures using Pytorch BIBREF6 within the OpenNMT-py framework BIBREF7. ### Evaluation ::: Data and Languages
In order to evaluate our model we design two sets of experiments. The experiments in §SECREF8 aim to evaluate different methods under low-resource settings, for languages with different morphological typology. We model the machine translation task from English into three languages with distinct morphological characteristics: Arabic (templatic), Czech (fusional), and Turkish (agglutinative). We use the TED Talks corpora BIBREF8 for training the NMT models for these experiments. In §SECREF10, we conduct more experiments in Turkish to demonstrate the case of increased data sparsity using multi-domain training corpora, where we extend the training set using corpora from EU Bookshop BIBREF9, Global Voices, Gnome, Tatoeba, Ubuntu BIBREF10, KDE4 BIBREF11, Open Subtitles BIBREF12 and SETIMES BIBREF13. The statistical characteristics of the training sets are given in Tables TABREF16 and TABREF17. We use the official evaluation sets of the IWSLT for validating and testing the accuracy of the models. In order to increase the number of unknown and rare words in the evaluation sets we measure accuracy on large test sets combining evaluation sets from many years (Table TABREF18 presents the evaluation sets used for development and testing). The accuracy of each model output is measured using BLEU BIBREF15 and chrF3 BIBREF16 metrics, whereas the significance of the improvements are computed using bootstrap hypothesis testing BIBREF17. ### Evaluation ::: Training Settings
All models are implemented using gated recurrent units (GRU) BIBREF18, and have a single-layer bi-RNN encoder. The source sides of the data used for training all NMT models, and the target sides of the data used in training the subword-level NMT models are segmented using BPE with 16,000 merge rules. We implement all decoders using a comparable number of GRU parameters, including 3-layer stacked-GRU subword and character-level decoders, where the attention is computed after the 1st layer BIBREF19 and a 3-layer hierarchical decoder which implements the attention mechanism after the 2nd layer. All models use an embedding dimension and GRU size of 512. The latent morphology model uses the same hierarchical GRU architecture, where the middle layer is augmented using 4 multi-layer perceptrons with 256 hidden units. We use a lemma vector dimension of 150, 10 inflectional features (See §SECREF21 for experiments conducted to tune the feature dimensions) and set the regularization constant to $\rho =0.4$. All models are trained using the Adam optimizer BIBREF20 with a batch size of 100, dropout rate of 0.2, learning rate of 0.0004 and learning rate decay of 0.8, applied when the perplexity does not decrease at a given epoch. Translations are generated with beam search with a beam size of 5, where the hierarchical models implement the hierarchical beam search BIBREF21. ### Evaluation ::: Results ::: The Effect of Morphological Typology
The experiment results given in Table TABREF9 shows the performance of each model in translating English into Arabic, Czech and Turkish. In Turkish, the most sparse target language in our benchmark, using character-based decoding shows to be more advantageous compared to the subword-level and hierarchical models, due to the fact that reduced granularity in the vocabulary units might aid in better predicting words under conditions of high data sparsity. In Arabic, on the other hand, using a hierarchical decoding model shows to be advantageous compared to the character-level decoder, as it might be useful in better learning syntactic dependencies, whereas it also outperforms the subword-level decoder. Using the latent morphology model provides improvements of 0.51 and 0.30 BLEU points in Arabic and Turkish over the best performing baselines, respectively. The fact that our model can efficiently work in both Arabic and Turkish suggests that it can handle the generation of both concatenative and non-concatenative morphological transformations. The results in the English-to-Czech translation direction do not indicate a specific advantage of using either method for generating fusional morphology, where morphemes are already optimized at the surface level, although our model is still able to achieve translation accuracy comparable to the character-level model. ### Evaluation ::: Results ::: The Effect of Data Size
The experiment conducted in the English-to-Turkish translation direction by increasing the amount of training data with multi-domain corpora demonstrates a more challenging case, where there is a greater possibility of observing rare words, either in the form of morphological inflections due to the complex agglutinative morphology of Turkish, or ambiguous terminology raising from the multi-domain characteristics. In this experiment, the character-level model experiences a drop in performance and its accuracy is much lower than the subword-level one, suggesting that its capacity cannot cope with the increased amount of sparsity. Empirical results suggest that with increased capacity, character-level models carry the potential to reach comparable performance to subword-level models BIBREF4. Our model reaches a much larger improvement of 0.82 BLEU points over the subword-level and 2.54 BLEU points over the character-level decoders, suggesting that it could make use of the increased sparsity in learning more accurate representations. ### Evaluation ::: Results ::: Predicting Unseen Words
In addition to general evaluation using automatic metrics, we perform a more focused analysis to illustrate the performance of different methods in predicting unseen words. We sample the sentences from the development sets which contain out-of-vocabulary words, and compute the average perplexity per character on these sentences using different NMT models, as suggested by BIBREF22. In general, the highest perplexities are obtained using the subword-based model, suggesting that generating unseen words using subword units is indeed increasing the difficulty of prediction, compared to the character-level which obtains the lowest perplexity. This result indicates that increased granularity aids in reducing the uncertainty during prediction. Similar to the results in §SECREF8, in Czech the values are almost comparable. Due to its stochastic nature, our model yields higher perplexity values compared to the hierarchical model, whereas the values range between subword and character-based models, possibly finding an optimal level of granularity between the two solutions. ### Evaluation ::: Results ::: Feature Variations
In order to understand whether the latent inflectional features in fact capture information about variations related to morphological transformations, we try generating different surface forms of the same lemma by assigning different values to the inflectional features. We use the latent morphology model based decoder to translate the English word `go', and after sampling the lemma, we fix its value and vary the values of the inflectional features at random positions for generating different outputs. Table TABREF14 presents different sets of feature values and the corresponding outputs generated by the decoder. The model generates different surface forms for different sets of features, confirming that latent variables encode information related to the infinitive form of the verb, as well as its formality conditions, prepositions, person, number and tense. We also observe that many trials based on different feature combinations may result in the same outputs, although some feature values may not be set in a single-word context. Varying the features individually does not necessarily yield distinct changes in the output, suggesting that some features may act jointly in determining the word form. ### Conclusion
In this paper we presented a novel decoding architecture for NMT employing a hierarchical latent variable model to promote sparsity in lexical representations, which demonstrated promising application for morphologically-rich and low-resource languages. Our model generates words one character at a time by composing two latent features representing their lemmas and inflectional features. We evaluate our model against conventional open-vocabulary NMT solutions such as subword and character-level decoding methods in translationg English into three morphologically-rich languages with different morphological typologies under low to mid-resource settings. Our results show that our model can significantly outperform subword-level NMT models, whereas demonstrates better capacity than character-level models in coping with increased amounts of data sparsity. We also conduct ablation studies on the effect of feature variations to the predictions, which prove that despite being completely unsupervised, our model can in fact capture morphosyntactic information and generalize to different surface forms of words. ### Acknowledgments
This project received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreements 825299 (GoURMET) and 688139 (SUMMA). ### Appendix ::: The Effect of Feature Dimensions
We investigate the optimal lemma and inflectional feature sizes by measuring the accuracy in English-to-Turkish translation using different feature vector dimensions. The results given in Figure FIGREF22 show that gradually compressing the word representations computed by recurrent hidden states, with an original dimension of 512, from 500 to 100, leads to increased output accuracy, suggesting that encoding more compact representations might provide the model with a better generalization capability. Our results also show that using a feature dimension of 10 is sufficient in reaching the best accuracy. Figure 1: The latent morphology model for computing word representations while translating the sentence ‘... went home’ into Turkish (‘eve gitti’). The character-level decoder is initialized with the attentional vector hi computed by the attention mechanism using current context ci and the word representation ti as in Luong & Manning (2016). Table 1: Machine translation accuracy in Arabic (AR), Czech (CS) and Turkish (TR) under lowresource settings using in-domain training data (middle column) and multi-domain training data (rightmost column). LMM representes the Latent Morphology Model. Best scores are in bold. All improvements over the baselines are statistically significant (p-value < 0.05). Table 2: Normalized perplexity measures per characters in different languages. Table 3: Outputs of the latent morphology model based on the lemma ‘git’ (‘go’) and different sets of inflectional features. Table 4: Training sets based on the TED Talks corpora (M: Million, K: Thousand). Table 6: Development and testing sets (K: Thousand). Figure 2: The top row shows the density function of the continuous base distribution over (0, 1). The middle row shows the result of stretching it to include 0 and 1 in its support. The bottom row shows the result of rectification: probability mass under (l, 0) collapses to 0 and probability mass under (1, r) collapses to 1, which cause sparse outcomes to have non-zero mass. Varying the shape parameters (a, b) of the underlying continuous distribution changes how much mass concentrates outside the support (0, 1) in the stretched density, and hence the probability of sampling sparse outcomes. Figure 3: The effect of feature dimensions on translation accuracy in Turkish.
|
Arabic, Czech and Turkish
|
Why was the large Kallerian not chosen for the journey?
A. There were already four Kallerians in inventory.
B. His species was too large to travel in the group.
C. He was argumentative during the interview process.
D. His payout demands exceeded their budget.
|
Birds of a Feather By ROBERT SILVERBERG Illustrated by WOOD [Transcriber's Note: This etext was produced from Galaxy Magazine November 1958. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] Getting specimens for the interstellar zoo was no problem—they battled for the honor—but now I had to fight like a wildcat to keep a display from making a monkey of me! It was our first day of recruiting on the planet, and the alien life-forms had lined up for hundreds of feet back from my rented office. As I came down the block from the hotel, I could hear and see and smell them with ease. My three staff men, Auchinleck, Stebbins and Ludlow, walked shieldwise in front of me. I peered between them to size the crop up. The aliens came in every shape and form, in all colors and textures—and all of them eager for a Corrigan contract. The Galaxy is full of bizarre beings, but there's barely a species anywhere that can resist the old exhibitionist urge. "Send them in one at a time," I told Stebbins. I ducked into the office, took my place back of the desk and waited for the procession to begin. The name of the planet was MacTavish IV (if you went by the official Terran listing) or Ghryne (if you called it by what its people were accustomed to calling it). I thought of it privately as MacTavish IV and referred to it publicly as Ghryne. I believe in keeping the locals happy wherever I go. Through the front window of the office, I could see our big gay tridim sign plastered to a facing wall: WANTED—EXTRATERRESTRIALS! We had saturated MacTavish IV with our promotional poop for a month preceding arrival. Stuff like this: Want to visit Earth—see the Galaxy's most glittering and exclusive world? Want to draw good pay, work short hours, experience the thrills of show business on romantic Terra? If you are a non-terrestrial, there may be a place for you in the Corrigan Institute of Morphological Science. No freaks wanted—normal beings only. J. F. Corrigan will hold interviews in person on Ghryne from Thirdday to Fifthday of Tenmonth. His last visit to the Caledonia Cluster until 2937, so don't miss your chance! Hurry! A life of wonder and riches can be yours! Broadsides like that, distributed wholesale in half a thousand languages, always bring them running. And the Corrigan Institute really packs in the crowds back on Earth. Why not? It's the best of its kind, the only really decent place where Earthmen can get a gander at the other species of the universe. The office buzzer sounded. Auchinleck said unctuously, "The first applicant is ready to see you, sir." "Send him, her or it in." The door opened and a timid-looking life-form advanced toward me on nervous little legs. He was a globular creature about the size of a big basketball, yellowish-green, with two spindly double-kneed legs and five double-elbowed arms, the latter spaced regularly around his body. There was a lidless eye at the top of his head and five lidded ones, one above each arm. Plus a big, gaping, toothless mouth. His voice was a surprisingly resounding basso. "You are Mr. Corrigan?" "That's right." I reached for a data blank. "Before we begin, I'll need certain information about—" "I am a being of Regulus II," came the grave, booming reply, even before I had picked up the blank. "I need no special care and I am not a fugitive from the law of any world." "Your name?" "Lawrence R. Fitzgerald." I throttled my exclamation of surprise, concealing it behind a quick cough. "Let me have that again, please?" "Certainly. My name is Lawrence R. Fitzgerald. The 'R' stands for Raymond." "Of course, that's not the name you were born with." The being closed his eyes and toddled around in a 360-degree rotation, remaining in place. On his world, that gesture is the equivalent of an apologetic smile. "My Regulan name no longer matters. I am now and shall evermore be Lawrence R. Fitzgerald. I am a Terraphile, you see." The little Regulan was as good as hired. Only the formalities remained. "You understand our terms, Mr. Fitzgerald?" "I'll be placed on exhibition at your Institute on Earth. You'll pay for my services, transportation and expenses. I'll be required to remain on exhibit no more than one-third of each Terran sidereal day." "And the pay will be—ah—$50 Galactic a week, plus expenses and transportation." The spherical creature clapped his hands in joy, three hands clapping on one side, two on the other. "Wonderful! I will see Earth at last! I accept the terms!" I buzzed for Ludlow and gave him the fast signal that meant we were signing this alien up at half the usual pay, and Ludlow took him into the other office to sign him up. I grinned, pleased with myself. We needed a green Regulan in our show; the last one had quit four years ago. But just because we needed him didn't mean we had to be extravagant in hiring him. A Terraphile alien who goes to the extent of rechristening himself with a Terran monicker would work for nothing, or even pay us, just so long as we let him get to Earth. My conscience won't let me really exploit a being, but I don't believe in throwing money away, either. The next applicant was a beefy ursinoid from Aldebaran IX. Our outfit has all the ursinoids it needs or is likely to need in the next few decades, and so I got rid of him in a couple of minutes. He was followed by a roly-poly blue-skinned humanoid from Donovan's Planet, four feet high and five hundred pounds heavy. We already had a couple of his species in the show, but they made good crowd-pleasers, being so plump and cheerful. I passed him along to Auchinleck to sign at anything short of top rate. Next came a bedraggled Sirian spider who was more interested in a handout than a job. If there's any species we have a real over-supply of, it's those silver-colored spiders, but this seedy specimen gave it a try anyway. He got the gate in half a minute, and he didn't even get the handout he was angling for. I don't approve of begging. The flora of applicants was steady. Ghryne is in the heart of the Caledonia Cluster, where the interstellar crossroads meet. We had figured to pick up plenty of new exhibits here and we were right. It was the isolationism of the late 29th century that turned me into the successful proprietor of Corrigan's Institute, after some years as an impoverished carnival man in the Betelgeuse system. Back in 2903, the World Congress declared Terra off-bounds for non-terrestrial beings, as an offshoot of the Terra for Terrans movement. Before then, anyone could visit Earth. After the gate clanged down, a non-terrestrial could only get onto Sol III as a specimen in a scientific collection—in short, as an exhibit in a zoo. That's what the Corrigan Institute of Morphological Science really is, of course. A zoo. But we don't go out and hunt for our specimens; we advertise and they come flocking to us. Every alien wants to see Earth once in his lifetime, and there's only one way he can do it. We don't keep too big an inventory. At last count, we had 690 specimens before this trip, representing 298 different intelligent life-forms. My goal is at least one member of at least 500 different races. When I reach that, I'll sit back and let the competition catch up—if it can. After an hour of steady work that morning, we had signed eleven new specimens. At the same time, we had turned away a dozen ursinoids, fifty of the reptilian natives of Ghryne, seven Sirian spiders, and no less than nineteen chlorine-breathing Procyonites wearing gas masks. It was also my sad duty to nix a Vegan who was negotiating through a Ghrynian agent. A Vegan would be a top-flight attraction, being some 400 feet long and appropriately fearsome to the eye, but I didn't see how we could take one on. They're gentle and likable beings, but their upkeep runs into literally tons of fresh meat a day, and not just any old kind of meat either. So we had to do without the Vegan. "One more specimen before lunch," I told Stebbins, "to make it an even dozen." He looked at me queerly and nodded. A being entered. I took a long close look at the life-form when it came in, and after that I took another one. I wondered what kind of stunt was being pulled. So far as I could tell, the being was quite plainly nothing but an Earthman. He sat down facing me without being asked and crossed his legs. He was tall and extremely thin, with pale blue eyes and dirty-blond hair, and though he was clean and reasonably well dressed, he had a shabby look about him. He said, in level Terran accents, "I'm looking for a job with your outfit, Corrigan." "There's been a mistake. We're interested in non-terrestrials only." "I'm a non-terrestrial. My name is Ildwar Gorb, of the planet Wazzenazz XIII." I don't mind conning the public from time to time, but I draw the line at getting bilked myself. "Look, friend, I'm busy, and I'm not known for my sense of humor. Or my generosity." "I'm not panhandling. I'm looking for a job." "Then try elsewhere. Suppose you stop wasting my time, bud. You're as Earthborn as I am." "I've never been within a dozen parsecs of Earth," he said smoothly. "I happen to be a representative of the only Earthlike race that exists anywhere in the Galaxy but on Earth itself. Wazzenazz XIII is a small and little-known planet in the Crab Nebula. Through an evolutionary fluke, my race is identical with yours. Now, don't you want me in your circus?" "No. And it's not a circus. It's—" "A scientific institute. I stand corrected." There was something glib and appealing about this preposterous phony. I guess I recognized a kindred spirit or I would have tossed him out on his ear without another word. Instead I played along. "If you're from such a distant place, how come you speak English so well?" "I'm not speaking. I'm a telepath—not the kind that reads minds, just the kind that projects. I communicate in symbols that you translate back to colloquial speech." "Very clever, Mr. Gorb." I grinned at him and shook my head. "You spin a good yarn—but for my money, you're really Sam Jones or Phil Smith from Earth, stranded here and out of cash. You want a free trip back to Earth. No deal. The demand for beings from Wazzenazz XIII is pretty low these days. Zero, in fact. Good-by, Mr. Gorb." He pointed a finger squarely at me and said, "You're making a big mistake. I'm just what your outfit needs. A representative of a hitherto utterly unknown race identical to humanity in every respect! Look here, examine my teeth. Absolutely like human teeth! And—" I pulled away from his yawning mouth. "Good-by, Mr. Gorb," I repeated. "All I ask is a contract, Corrigan. It isn't much. I'll be a big attraction. I'll—" " Good-by, Mr. Gorb! " He glowered at me reproachfully for a moment, stood up and sauntered to the door. "I thought you were a man of acumen, Corrigan. Well, think it over. Maybe you'll regret your hastiness. I'll be back to give you another chance." He slammed the door and I let my grim expression relax into a smile. This was the best con switch yet—an Earthman posing as an alien to get a job! But I wasn't buying it, even if I could appreciate his cleverness intellectually. There's no such place as Wazzenazz XIII and there's only one human race in the Galaxy—on Earth. I was going to need some real good reason before I gave a down-and-out grifter a free ticket home. I didn't know it then, but before the day was out, I would have that reason. And, with it, plenty of trouble on my hands. The first harbinger of woe turned up after lunch in the person of a Kallerian. The Kallerian was the sixth applicant that afternoon. I had turned away three more ursinoids, hired a vegetable from Miazan, and said no to a scaly pseudo-armadillo from one of the Delta Worlds. Hardly had the 'dillo scuttled dejectedly out of my office when the Kallerian came striding in, not even waiting for Stebbins to admit him officially. He was big even for his kind—in the neighborhood of nine feet high, and getting on toward a ton. He planted himself firmly on his three stocky feet, extended his massive arms in a Kallerian greeting-gesture, and growled, "I am Vallo Heraal, Freeman of Kaller IV. You will sign me immediately to a contract." "Sit down, Freeman Heraal. I like to make my own decisions, thanks." "You will grant me a contract!" "Will you please sit down?" He said sulkily, "I will remain standing." "As you prefer." My desk has a few concealed features which are sometimes useful in dealing with belligerent or disappointed life-forms. My fingers roamed to the meshgun trigger, just in case of trouble. The Kallerian stood motionless before me. They're hairy creatures, and this one had a coarse, thick mat of blue fur completely covering his body. Two fierce eyes glimmered out through the otherwise dense blanket of fur. He was wearing the kilt, girdle and ceremonial blaster of his warlike race. I said, "You'll have to understand, Freeman Heraal, that it's not our policy to maintain more than a few members of each species at our Institute. And we're not currently in need of any Kallerian males, because—" "You will hire me or trouble I will make!" I opened our inventory chart. I showed him that we were already carrying four Kallerians, and that was more than plenty. The beady little eyes flashed like beacons in the fur. "Yes, you have four representatives—of the Clan Verdrokh! None of the Clan Gursdrinn! For three years, I have waited for a chance to avenge this insult to the noble Clan Gursdrinn!" At the key-word avenge , I readied myself to ensnarl the Kallerian in a spume of tanglemesh the instant he went for his blaster, but he didn't move. He bellowed, "I have vowed a vow, Earthman. Take me to Earth, enroll a Gursdrinn, or the consequences will be terrible!" I'm a man of principles, like all straightforward double-dealers, and one of the most important of those principles is that I never let myself be bullied by anyone. "I deeply regret having unintentionally insulted your clan, Freeman Heraal. Will you accept my apologies?" He glared at me in silence. I went on, "Please be assured that I'll undo the insult at the earliest possible opportunity. It's not feasible for us to hire another Kallerian now, but I'll give preference to the Clan Gursdrinn as soon as a vacancy—" "No. You will hire me now." "It can't be done, Freeman Heraal. We have a budget, and we stick to it." "You will rue! I will take drastic measures!" "Threats will get you nowhere, Freeman Heraal. I give you my word I'll get in touch with you as soon as our organization has room for another Kallerian. And now, please, there are many applicants waiting—" You'd think it would be sort of humiliating to become a specimen in a zoo, but most of these races take it as an honor. And there's always the chance that, by picking a given member of a race, we're insulting all the others. I nudged the trouble-button on the side of my desk and Auchinleck and Ludlow appeared simultaneously from the two doors at right and left. They surrounded the towering Kallerian and sweet-talkingly led him away. He wasn't minded to quarrel physically, or he could have knocked them both into the next city with a backhand swipe of his shaggy paw, but he kept up a growling flow of invective and threats until he was out in the hall. I mopped sweat from my forehead and began to buzz Stebbins for the next applicant. But before my finger touched the button, the door popped open and a small being came scooting in, followed by an angry Stebbins. "Come here, you!" "Stebbins?" I said gently. "I'm sorry, Mr. Corrigan. I lost sight of this one for a moment, and he came running in—" "Please, please," squeaked the little alien pitifully. "I must see you, honored sir!" "It isn't his turn in line," Stebbins protested. "There are at least fifty ahead of him." "All right," I said tiredly. "As long as he's in here already, I might as well see him. Be more careful next time, Stebbins." Stebbins nodded dolefully and backed out. The alien was a pathetic sight: a Stortulian, a squirrely-looking creature about three feet high. His fur, which should have been a lustrous black, was a dull gray, and his eyes were wet and sad. His tail drooped. His voice was little more than a faint whimper, even at full volume. "Begging your most honored pardon most humbly, important sir. I am a being of Stortul XII, having sold my last few possessions to travel to Ghryne for the miserable purpose of obtaining an interview with yourself." I said, "I'd better tell you right at the outset that we're already carrying our full complement of Stortulians. We have both a male and a female now and—" "This is known to me. The female—is her name perchance Tiress?" I glanced down at the inventory chart until I found the Stortulian entry. "Yes, that's her name." The little being immediately emitted a soul-shaking gasp. "It is she! It is she!" "I'm afraid we don't have room for any more—" "You are not in full understanding of my plight. The female Tiress, she is—was—my own Fire-sent spouse, my comfort and my warmth, my life and my love." "Funny," I said. "When we signed her three years ago, she said she was single. It's right here on the chart." "She lied! She left my burrow because she longed to see the splendors of Earth. And I am alone, bound by our sacred customs never to remarry, languishing in sadness and pining for her return. You must take me to Earth!" "But—" "I must see her—her and this disgrace-bringing lover of hers. I must reason with her. Earthman, can't you see I must appeal to her inner flame? I must bring her back! " My face was expressionless. "You don't really intend to join our organization at all—you just want free passage to Earth?" "Yes, yes!" wailed the Stortulian. "Find some other member of my race, if you must! Let me have my wife again, Earthman! Is your heart a dead lump of stone?" It isn't, but another of my principles is to refuse to be swayed by sentiment. I felt sorry for this being's domestic troubles, but I wasn't going to break up a good act just to make an alien squirrel happy—not to mention footing the transportation. I said, "I don't see how we can manage it. The laws are very strict on the subject of bringing alien life to Earth. It has to be for scientific purposes only. And if I know in advance that your purpose in coming isn't scientific, I can't in all conscience lie for you, can I?" "Well—" "Of course not." I took advantage of his pathetic upset to steam right along. "Now if you had come in here and simply asked me to sign you up, I might conceivably have done it. But no—you had to go unburden your heart to me." "I thought the truth would move you." "It did. But in effect you're now asking me to conspire in a fraudulent criminal act. Friend, I can't do it. My reputation means too much to me," I said piously. "Then you will refuse me?" "My heart melts to nothingness for you. But I can't take you to Earth." "Perhaps you will send my wife to me here?" There's a clause in every contract that allows me to jettison an unwanted specimen. All I have to do is declare it no longer of scientific interest, and the World Government will deport the undesirable alien back to its home world. But I wouldn't pull a low trick like that on our female Stortulian. I said, "I'll ask her about coming home. But I won't ship her back against her will. And maybe she's happier where she is." The Stortulian seemed to shrivel. His eyelids closed half-way to mask his tears. He turned and shambled slowly to the door, walking like a living dishrag. In a bleak voice, he said, "There is no hope then. All is lost. I will never see my soulmate again. Good day, Earthman." He spoke in a drab monotone that almost, but not quite, had me weeping. I watched him shuffle out. I do have some conscience, and I had the uneasy feeling I had just been talking to a being who was about to commit suicide on my account. About fifty more applicants were processed without a hitch. Then life started to get complicated again. Nine of the fifty were okay. The rest were unacceptable for one reason or another, and they took the bad news quietly enough. The haul for the day so far was close to two dozen new life-forms under contract. I had just about begun to forget about the incidents of the Kallerian's outraged pride and the Stortulian's flighty wife when the door opened and the Earthman who called himself Ildwar Gorb of Wazzenazz XIII stepped in. "How did you get in here?" I demanded. "Your man happened to be looking the wrong way," he said cheerily. "Change your mind about me yet?" "Get out before I have you thrown out." Gorb shrugged. "I figured you hadn't changed your mind, so I've changed my pitch a bit. If you won't believe I'm from Wazzenazz XIII, suppose I tell you that I am Earthborn, and that I'm looking for a job on your staff." "I don't care what your story is! Get out or—" "—you'll have me thrown out. Okay, okay. Just give me half a second. Corrigan, you're no fool, and neither am I—but that fellow of yours outside is . He doesn't know how to handle alien beings. How many times today has a life-form come in here unexpectedly?" I scowled at him. "Too damn many." "You see? He's incompetent. Suppose you fire him, take me on instead. I've been living in the outworlds half my life; I know all there is to know about alien life-forms. You can use me, Corrigan." I took a deep breath and glanced all around the paneled ceiling of the office before I spoke. "Listen, Gorb, or whatever your name is, I've had a hard day. There's been a Kallerian in here who just about threatened murder, and there's been a Stortulian in here who's about to commit suicide because of me. I have a conscience and it's troubling me. But get this: I just want to finish off my recruiting, pack up and go home to Earth. I don't want you hanging around here bothering me. I'm not looking to hire new staff members, and if you switch back to claiming you're an unknown life-form from Wazzenazz XIII, the answer is that I'm not looking for any of those either. Now will you scram or—" The office door crashed open at that point and Heraal, the Kallerian, came thundering in. He was dressed from head to toe in glittering metalfoil, and instead of his ceremonial blaster, he was wielding a sword the length of a human being. Stebbins and Auchinleck came dragging helplessly along in his wake, hanging desperately to his belt. "Sorry, Chief," Stebbins gasped. "I tried to keep him out, but—" Heraal, who had planted himself in front of my desk, drowned him out with a roar. "Earthman, you have mortally insulted the Clan Gursdrinn!" Sitting with my hands poised near the meshgun trigger, I was ready to let him have it at the first sight of actual violence. Heraal boomed, "You are responsible for what is to happen now. I have notified the authorities and you prosecuted will be for causing the death of a life-form! Suffer, Earthborn ape! Suffer!" "Watch it, Chief," Stebbins yelled. "He's going to—" An instant before my numb fingers could tighten on the meshgun trigger, Heraal swung that huge sword through the air and plunged it savagely through his body. He toppled forward onto the carpet with the sword projecting a couple of feet out of his back. A few driblets of bluish-purple blood spread from beneath him. Before I could react to the big life-form's hara-kiri, the office door flew open again and three sleek reptilian beings entered, garbed in the green sashes of the local police force. Their golden eyes goggled down at the figure on the floor, then came to rest on me. "You are J. F. Corrigan?" the leader asked. "Y-yes." "We have received word of a complaint against you. Said complaint being—" "—that your unethical actions have directly contributed to the untimely death of an intelligent life-form," filled in the second of the Ghrynian policemen. "The evidence lies before us," intoned the leader, "in the cadaver of the unfortunate Kallerian who filed the complaint with us several minutes ago." "And therefore," said the third lizard, "it is our duty to arrest you for this crime and declare you subject to a fine of no less than $100,000 Galactic or two years in prison." "Hold on!" I stormed. "You mean that any being from anywhere in the Universe can come in here and gut himself on my carpet, and I'm responsible?" "This is the law. Do you deny that your stubborn refusal to yield to this late life-form's request lies at the root of his sad demise?" "Well, no, but—" "Failure to deny is admission of guilt. You are guilty, Earthman." Closing my eyes wearily, I tried to wish the whole babbling lot of them away. If I had to, I could pony up the hundred-grand fine, but it was going to put an awful dent in this year's take. And I shuddered when I remembered that any minute that scrawny little Stortulian was likely to come bursting in here to kill himself too. Was it a fine of $100,000 per suicide? At that rate, I could be out of business by nightfall. I was spared further such morbid thoughts by yet another unannounced arrival. The small figure of the Stortulian trudged through the open doorway and stationed itself limply near the threshold. The three Ghrynian policemen and my three assistants forgot the dead Kallerian for a moment and turned to eye the newcomer. I had visions of unending troubles with the law here on Ghryne. I resolved never to come here on a recruiting trip again—or, if I did come, to figure out some more effective way of screening myself against crackpots. In heart-rending tones, the Stortulian declared, "Life is no longer worth living. My last hope is gone. There is only one thing left for me to do." I was quivering at the thought of another hundred thousand smackers going down the drain. "Stop him, somebody! He's going to kill himself! He's—" Then somebody sprinted toward me, hit me amidships, and knocked me flying out from behind my desk before I had a chance to fire the meshgun. My head walloped the floor, and for five or six seconds, I guess I wasn't fully aware of what was going on. Gradually the scene took shape around me. There was a monstrous hole in the wall behind my desk; a smoking blaster lay on the floor, and I saw the three Ghrynian policemen sitting on the raving Stortulian. The man who called himself Ildwar Gorb was getting to his feet and dusting himself off. He helped me up. "Sorry to have had to tackle you, Corrigan. But that Stortulian wasn't here to commit suicide, you see. He was out to get you." I weaved dizzily toward my desk and dropped into my chair. A flying fragment of wall had deflated my pneumatic cushion. The smell of ashed plaster was everywhere. The police were effectively cocooning the struggling little alien in an unbreakable tanglemesh. "Evidently you don't know as much as you think you do about Stortulian psychology, Corrigan," Gorb said lightly. "Suicide is completely abhorrent to them. When they're troubled, they kill the person who caused their trouble. In this case, you." I began to chuckle—more of a tension-relieving snicker than a full-bodied laugh. "Funny," I said. "What is?" asked the self-styled Wazzenazzian. "These aliens. Big blustery Heraal came in with murder in his eye and killed himself , and the pint-sized Stortulian who looked so meek and pathetic damn near blew my head off." I shuddered. "Thanks for the tackle job." "Don't mention it," Gorb said. I glared at the Ghrynian police. "Well? What are you waiting for? Take that murderous little beast out of here! Or isn't murder against the local laws?" "The Stortulian will be duly punished," replied the leader of the Ghrynian cops calmly. "But there is the matter of the dead Kallerian and the fine of—" "—one hundred thousand dollars. I know." I groaned and turned to Stebbins. "Get the Terran Consulate on the phone, Stebbins. Have them send down a legal adviser. Find out if there's any way we can get out of this mess with our skins intact." "Right, Chief." Stebbins moved toward the visiphone. Gorb stepped forward and put a hand on his chest. "Hold it," the Wazzenazzian said crisply. "The Consulate can't help you. I can." "You?" I said. "I can get you out of this cheap." " How cheap?" Gorb grinned rakishly. "Five thousand in cash plus a contract as a specimen with your outfit. In advance, of course. That's a heck of a lot better than forking over a hundred grand, isn't it?" I eyed Gorb uncertainly. The Terran Consulate people probably wouldn't be much help; they tried to keep out of local squabbles unless they were really serious, and I knew from past experiences that no officials ever worried much about the state of my pocketbook. On the other hand, giving this slyster a contract might be a risky proposition. "Tell you what," I said finally. "You've got yourself a deal—but on a contingency basis. Get me out of this and you'll have five grand and the contract. Otherwise, nothing." Gorb shrugged. "What have I to lose?"
|
A. There were already four Kallerians in inventory.
|
Why didn't the Delegate have a robotic voice?
A. The robot was programmed by the Venusians to speak the many languages of Earth.
B. The robot was programmed by Professor Coltz and the group that helped him to speak eight languages.
C. The robot was voiced by Professor Coltz remotely.
D. The advanced Venusian technology allows for a natural-sounding voice.
|
The saucer was interesting, but where was the delegate? The DELEGATE FROM VENUS By HENRY SLESAR ILLUSTRATOR NOVICK Everybody was waiting to see what the delegate from Venus looked like. And all they got for their patience was the biggest surprise since David clobbered Goliath. " Let me put it this way," Conners said paternally. "We expect a certain amount of decorum from our Washington news correspondents, and that's all I'm asking for." Jerry Bridges, sitting in the chair opposite his employer's desk, chewed on his knuckles and said nothing. One part of his mind wanted him to play it cagey, to behave the way the newspaper wanted him to behave, to protect the cozy Washington assignment he had waited four years to get. But another part of him, a rebel part, wanted him to stay on the trail of the story he felt sure was about to break. "I didn't mean to make trouble, Mr. Conners," he said casually. "It just seemed strange, all these exchanges of couriers in the past two days. I couldn't help thinking something was up." "Even if that's true, we'll hear about it through the usual channels," Conners frowned. "But getting a senator's secretary drunk to obtain information—well, that's not only indiscreet, Bridges. It's downright dirty." Jerry grinned. "I didn't take that kind of advantage, Mr. Conners. Not that she wasn't a toothsome little dish ..." "Just thank your lucky stars that it didn't go any further. And from now on—" He waggled a finger at him. "Watch your step." Jerry got up and ambled to the door. But he turned before leaving and said: "By the way. What do you think is going on?" "I haven't the faintest idea." "Don't kid me, Mr. Conners. Think it's war?" "That'll be all, Bridges." The reporter closed the door behind him, and then strolled out of the building into the sunlight. He met Ruskin, the fat little AP correspondent, in front of the Pan-American Building on Constitution Avenue. Ruskin was holding the newspaper that contained the gossip-column item which had started the whole affair, and he seemed more interested in the romantic rather than political implications. As he walked beside him, he said: "So what really happened, pal? That Greta babe really let down her hair?" "Where's your decorum?" Jerry growled. Ruskin giggled. "Boy, she's quite a dame, all right. I think they ought to get the Secret Service to guard her. She really fills out a size 10, don't she?" "Ruskin," Jerry said, "you have a low mind. For a week, this town has been acting like the 39 Steps , and all you can think about is dames. What's the matter with you? Where will you be when the big mushroom cloud comes?" "With Greta, I hope," Ruskin sighed. "What a way to get radioactive." They split off a few blocks later, and Jerry walked until he came to the Red Tape Bar & Grill, a favorite hangout of the local journalists. There were three other newsmen at the bar, and they gave him snickering greetings. He took a small table in the rear and ate his meal in sullen silence. It wasn't the newsmen's jibes that bothered him; it was the certainty that something of major importance was happening in the capitol. There had been hourly conferences at the White House, flying visits by State Department officials, mysterious conferences involving members of the Science Commission. So far, the byword had been secrecy. They knew that Senator Spocker, chairman of the Congressional Science Committee, had been involved in every meeting, but Senator Spocker was unavailable. His secretary, however, was a little more obliging ... Jerry looked up from his coffee and blinked when he saw who was coming through the door of the Bar & Grill. So did every other patron, but for different reasons. Greta Johnson had that effect upon men. Even the confining effect of a mannishly-tailored suit didn't hide her outrageously feminine qualities. She walked straight to his table, and he stood up. "They told me you might be here," she said, breathing hard. "I just wanted to thank you for last night." "Look, Greta—" Wham! Her hand, small and delicate, felt like a slab of lead when it slammed into his cheek. She left a bruise five fingers wide, and then turned and stalked out. He ran after her, the restaurant proprietor shouting about the unpaid bill. It took a rapid dog-trot to reach her side. "Greta, listen!" he panted. "You don't understand about last night. It wasn't the way that lousy columnist said—" She stopped in her tracks. "I wouldn't have minded so much if you'd gotten me drunk. But to use me, just to get a story—" "But I'm a reporter , damn it. It's my job. I'd do it again if I thought you knew anything." She was pouting now. "Well, how do you suppose I feel, knowing you're only interested in me because of the Senator? Anyway, I'll probably lose my job, and then you won't have any use for me." "Good-bye, Greta," Jerry said sadly. "What?" "Good-bye. I suppose you won't want to see me any more." "Did I say that?" "It just won't be any use. We'll always have this thing between us." She looked at him for a moment, and then touched his bruised cheek with a tender, motherly gesture. "Your poor face," she murmured, and then sighed. "Oh, well. I guess there's no use fighting it. Maybe if I did tell you what I know, we could act human again." "Greta!" "But if you print one word of it, Jerry Bridges, I'll never speak to you again!" "Honey," Jerry said, taking her arm, "you can trust me like a brother." "That's not the idea," Greta said stiffly. In a secluded booth at the rear of a restaurant unfrequented by newsmen, Greta leaned forward and said: "At first, they thought it was another sputnik." " Who did?" "The State Department, silly. They got reports from the observatories about another sputnik being launched by the Russians. Only the Russians denied it. Then there were joint meetings, and nobody could figure out what the damn thing was." "Wait a minute," Jerry said dizzily. "You mean to tell me there's another of those metal moons up there?" "But it's not a moon. That's the big point. It's a spaceship." "A what ?" "A spaceship," Greta said coolly, sipping lemonade. "They have been in contact with it now for about three days, and they're thinking of calling a plenary session of the UN just to figure out what to do about it. The only hitch is, Russia doesn't want to wait that long, and is asking for a hurry-up summit meeting to make a decision." "A decision about what?" "About the Venusians, of course." "Greta," Jerry said mildly, "I think you're still a little woozy from last night." "Don't be silly. The spaceship's from Venus; they've already established that. And the people on it—I guess they're people—want to know if they can land their delegate." "Their what?" "Their delegate. They came here for some kind of conference, I guess. They know about the UN and everything, and they want to take part. They say that with all the satellites being launched, that our affairs are their affairs, too. It's kind of confusing, but that's what they say." "You mean these Venusians speak English?" "And Russian. And French. And German. And everything I guess. They've been having radio talks with practically every country for the past three days. Like I say, they want to establish diplomatic relations or something. The Senator thinks that if we don't agree, they might do something drastic, like blow us all up. It's kind of scary." She shivered delicately. "You're taking it mighty calm," he said ironically. "Well, how else can I take it? I'm not even supposed to know about it, except that the Senator is so careless about—" She put her fingers to her lips. "Oh, dear, now you'll really think I'm terrible." "Terrible? I think you're wonderful!" "And you promise not to print it?" "Didn't I say I wouldn't?" "Y-e-s. But you know, you're a liar sometimes, Jerry. I've noticed that about you." The press secretary's secretary, a massive woman with gray hair and impervious to charm, guarded the portals of his office with all the indomitable will of the U. S. Marines. But Jerry Bridges tried. "You don't understand, Lana," he said. "I don't want to see Mr. Howells. I just want you to give him something." "My name's not Lana, and I can't deliver any messages." "But this is something he wants to see." He handed her an envelope, stamped URGENT. "Do it for me, Hedy. And I'll buy you the flashiest pair of diamond earrings in Washington." "Well," the woman said, thawing slightly. "I could deliver it with his next batch of mail." "When will that be?" "In an hour. He's in a terribly important meeting right now." "You've got some mail right there. Earrings and a bracelet to match." She looked at him with exasperation, and then gathered up a stack of memorandums and letters, his own envelope atop it. She came out of the press secretary's office two minutes later with Howells himself, and Howells said: "You there, Bridges. Come in here." "Yes, sir !" Jerry said, breezing by the waiting reporters with a grin of triumph. There were six men in the room, three in military uniform. Howells poked the envelope towards Jerry, and snapped: "This note of yours. Just what do you think it means?" "You know better than I do, Mr. Howells. I'm just doing my job; I think the public has a right to know about this spaceship that's flying around—" His words brought an exclamation from the others. Howells sighed, and said: "Mr. Bridges, you don't make it easy for us. It's our opinion that secrecy is essential, that leakage of the story might cause panic. Since you're the only unauthorized person who knows of it, we have two choices. One of them is to lock you up." Jerry swallowed hard. "The other is perhaps more practical," Howells said. "You'll be taken into our confidence, and allowed to accompany those officials who will be admitted to the landing site. But you will not be allowed to relay the story to the press until such a time as all correspondents are informed. That won't give you a 'scoop' if that's what you call it, but you'll be an eyewitness. That should be worth something." "It's worth a lot," Jerry said eagerly. "Thanks, Mr. Howells." "Don't thank me, I'm not doing you any personal favor. Now about the landing tonight—" "You mean the spaceship's coming down?" "Yes. A special foreign ministers conference was held this morning, and a decision was reached to accept the delegate. Landing instructions are being given at Los Alamos, and the ship will presumably land around midnight tonight. There will be a jet leaving Washington Airport at nine, and you'll be on it. Meanwhile, consider yourself in custody." The USAF jet transport wasn't the only secrecy-shrouded aircraft that took off that evening from Washington Airport. But Jerry Bridges, sitting in the rear seat flanked by two Sphinx-like Secret Service men, knew that he was the only passenger with non-official status aboard. It was only a few minutes past ten when they arrived at the air base at Los Alamos. The desert sky was cloudy and starless, and powerful searchlights probed the thick cumulus. There were sleek, purring black autos waiting to rush the air passengers to some unnamed destination. They drove for twenty minutes across a flat ribbon of desert road, until Jerry sighted what appeared to be a circle of newly-erected lights in the middle of nowhere. On the perimeter, official vehicles were parked in orderly rows, and four USAF trailer trucks were in evidence, their radarscopes turning slowly. There was activity everywhere, but it was well-ordered and unhurried. They had done a good job of keeping the excitement contained. He was allowed to leave the car and stroll unescorted. He tried to talk to some of the scurrying officials, but to no avail. Finally, he contented himself by sitting on the sand, his back against the grill of a staff car, smoking one cigarette after another. As the minutes ticked off, the activity became more frenetic around him. Then the pace slowed, and he knew the appointed moment was approaching. Stillness returned to the desert, and tension was a tangible substance in the night air. The radarscopes spun slowly. The searchlights converged in an intricate pattern. Then the clouds seemed to part! "Here she comes!" a voice shouted. And in a moment, the calm was shattered. At first, he saw nothing. A faint roar was started in the heavens, and it became a growl that increased in volume until even the shouting voices could no longer be heard. Then the crisscrossing lights struck metal, glancing off the gleaming body of a descending object. Larger and larger the object grew, until it assumed the definable shape of a squat silver funnel, falling in a perfect straight line towards the center of the light-ringed area. When it hit, a dust cloud obscured it from sight. A loudspeaker blared out an unintelligible order, but its message was clear. No one moved from their position. Finally, a three-man team, asbestos-clad, lead-shielded, stepped out from the ring of spectators. They carried geiger counters on long poles before them. Jerry held his breath as they approached the object; only when they were yards away did he appreciate its size. It wasn't large; not more than fifteen feet in total circumference. One of the three men waved a gloved hand. "It's okay," a voice breathed behind him. "No radiation ..." Slowly, the ring of spectators closed tighter. They were twenty yards from the ship when the voice spoke to them. "Greetings from Venus," it said, and then repeated the phrase in six languages. "The ship you see is a Venusian Class 7 interplanetary rocket, built for one-passenger. It is clear of all radiation, and is perfectly safe to approach. There is a hatch which may be opened by an automatic lever in the side. Please open this hatch and remove the passenger." An Air Force General whom Jerry couldn't identify stepped forward. He circled the ship warily, and then said something to the others. They came closer, and he touched a small lever on the silvery surface of the funnel. A door slid open. "It's a box!" someone said. "A crate—" "Colligan! Moore! Schaffer! Lend a hand here—" A trio came forward and hoisted the crate out of the ship. Then the voice spoke again; Jerry deduced that it must have been activated by the decreased load of the ship. "Please open the crate. You will find our delegate within. We trust you will treat him with the courtesy of an official emissary." They set to work on the crate, its gray plastic material giving in readily to the application of their tools. But when it was opened, they stood aside in amazement and consternation. There were a variety of metal pieces packed within, protected by a filmy packing material. "Wait a minute," the general said. "Here's a book—" He picked up a gray-bound volume, and opened its cover. "'Instructions for assembling Delegate,'" he read aloud. "'First, remove all parts and arrange them in the following order. A-1, central nervous system housing. A-2 ...'" He looked up. "It's an instruction book," he whispered. "We're supposed to build the damn thing." The Delegate, a handsomely constructed robot almost eight feet tall, was pieced together some three hours later, by a team of scientists and engineers who seemed to find the Venusian instructions as elementary as a blueprint in an Erector set. But simple as the job was, they were obviously impressed by the mechanism they had assembled. It stood impassive until they obeyed the final instruction. "Press Button K ..." They found button K, and pressed it. The robot bowed. "Thank you, gentlemen," it said, in sweet, unmetallic accents. "Now if you will please escort me to the meeting place ..." It wasn't until three days after the landing that Jerry Bridges saw the Delegate again. Along with a dozen assorted government officials, Army officers, and scientists, he was quartered in a quonset hut in Fort Dix, New Jersey. Then, after seventy-two frustrating hours, he was escorted by Marine guard into New York City. No one told him his destination, and it wasn't until he saw the bright strips of light across the face of the United Nations building that he knew where the meeting was to be held. But his greatest surprise was yet to come. The vast auditorium which housed the general assembly was filled to its capacity, but there were new faces behind the plaques which designated the member nations. He couldn't believe his eyes at first, but as the meeting got under way, he knew that it was true. The highest echelons of the world's governments were represented, even—Jerry gulped at the realization—Nikita Khrushchev himself. It was a summit meeting such as he had never dreamed possible, a summit meeting without benefit of long foreign minister's debate. And the cause of it all, a placid, highly-polished metal robot, was seated blithely at a desk which bore the designation: VENUS. The robot delegate stood up. "Gentlemen," it said into the microphone, and the great men at the council tables strained to hear the translator's version through their headphones, "Gentlemen, I thank you for your prompt attention. I come as a Delegate from a great neighbor planet, in the interests of peace and progress for all the solar system. I come in the belief that peace is the responsibility of individuals, of nations, and now of worlds, and that each is dependent upon the other. I speak to you now through the electronic instrumentation which has been created for me, and I come to offer your planet not merely a threat, a promise, or an easy solution—but a challenge." The council room stirred. "Your earth satellites have been viewed with interest by the astronomers of our world, and we foresee the day when contact between our planets will be commonplace. As for ourselves, we have hitherto had little desire to explore beyond our realm, being far too occupied with internal matters. But our isolation cannot last in the face of your progress, so we believe that we must take part in your affairs. "Here, then, is our challenge. Continue your struggle of ideas, compete with each other for the minds of men, fight your bloodless battles, if you know no other means to attain progress. But do all this without unleashing the terrible forces of power now at your command. Once unleashed, these forces may or may not destroy all that you have gained. But we, the scientists of Venus, promise you this—that on the very day your conflict deteriorates into heedless violence, we will not stand by and let the ugly contagion spread. On that day, we of Venus will act swiftly, mercilessly, and relentlessly—to destroy your world completely." Again, the meeting room exploded in a babble of languages. "The vessel which brought me here came as a messenger of peace. But envision it, men of Earth, as a messenger of war. Unstoppable, inexorable, it may return, bearing a different Delegate from Venus—a Delegate of Death, who speaks not in words, but in the explosion of atoms. Think of thousands of such Delegates, fired from a vantage point far beyond the reach of your retaliation. This is the promise and the challenge that will hang in your night sky from this moment forward. Look at the planet Venus, men of Earth, and see a Goddess of Vengeance, poised to wreak its wrath upon those who betray the peace." The Delegate sat down. Four days later, a mysterious explosion rocked the quiet sands of Los Alamos, and the Venus spacecraft was no more. Two hours after that, the robot delegate, its message delivered, its mission fulfilled, requested to be locked inside a bombproof chamber. When the door was opened, the Delegate was an exploded ruin. The news flashed with lightning speed over the world, and Jerry Bridges' eyewitness accounts of the incredible event was syndicated throughout the nation. But his sudden celebrity left him vaguely unsatisfied. He tried to explain his feeling to Greta on his first night back in Washington. They were in his apartment, and it was the first time Greta had consented to pay him the visit. "Well, what's bothering you?" Greta pouted. "You've had the biggest story of the year under your byline. I should think you'd be tickled pink." "It's not that," Jerry said moodily. "But ever since I heard the Delegate speak, something's been nagging me." "But don't you think he's done good? Don't you think they'll be impressed by what he said?" "I'm not worried about that. I think that damn robot did more for peace than anything that's ever come along in this cockeyed world. But still ..." Greta snuggled up to him on the sofa. "You worry too much. Don't you ever think of anything else? You should learn to relax. It can be fun." She started to prove it to him, and Jerry responded the way a normal, healthy male usually does. But in the middle of an embrace, he cried out: "Wait a minute!" "What's the matter?" "I just thought of something! Now where the hell did I put my old notebooks?" He got up from the sofa and went scurrying to a closet. From a debris of cardboard boxes, he found a worn old leather brief case, and cackled with delight when he found the yellowed notebooks inside. "What are they?" Greta said. "My old school notebooks. Greta, you'll have to excuse me. But there's something I've got to do, right away!" "That's all right with me," Greta said haughtily. "I know when I'm not wanted." She took her hat and coat from the hall closet, gave him one last chance to change his mind, and then left. Five minutes later, Jerry Bridges was calling the airlines. It had been eleven years since Jerry had walked across the campus of Clifton University, heading for the ivy-choked main building. It was remarkable how little had changed, but the students seemed incredibly young. He was winded by the time he asked the pretty girl at the desk where Professor Martin Coltz could be located. "Professor Coltz?" She stuck a pencil to her mouth. "Well, I guess he'd be in the Holland Laboratory about now." "Holland Laboratory? What's that?" "Oh, I guess that was after your time, wasn't it?" Jerry felt decrepit, but managed to say: "It must be something new since I was here. Where is this place?" He followed her directions, and located a fresh-painted building three hundred yards from the men's dorm. He met a student at the door, who told him that Professor Coltz would be found in the physics department. The room was empty when Jerry entered, except for the single stooped figure vigorously erasing a blackboard. He turned when the door opened. If the students looked younger, Professor Coltz was far older than Jerry remembered. He was a tall man, with an unruly confusion of straight gray hair. He blinked when Jerry said: "Hello, Professor. Do you remember me? Jerry Bridges?" "Of course! I thought of you only yesterday, when I saw your name in the papers—" They sat at facing student desks, and chatted about old times. But Jerry was impatient to get to the point of his visit, and he blurted out: "Professor Coltz, something's been bothering me. It bothered me from the moment I heard the Delegate speak. I didn't know what it was until last night, when I dug out my old college notebooks. Thank God I kept them." Coltz's eyes were suddenly hooded. "What do you mean, Jerry?" "There was something about the Robot's speech that sounded familiar—I could have sworn I'd heard some of the words before. I couldn't prove anything until I checked my old notes, and here's what I found." He dug into his coat pocket and produced a sheet of paper. He unfolded it and read aloud. "'It's my belief that peace is the responsibility of individuals, of nations, and someday, even of worlds ...' Sound familiar, Professor?" Coltz shifted uncomfortably. "I don't recall every silly thing I said, Jerry." "But it's an interesting coincidence, isn't it, Professor? These very words were spoken by the Delegate from Venus." "A coincidence—" "Is it? But I also remember your interest in robotics. I'll never forget that mechanical homing pigeon you constructed. And you've probably learned much more these past eleven years." "What are you driving at, Jerry?" "Just this, Professor. I had a little daydream, recently, and I want you to hear it. I dreamed about a group of teachers, scientists, and engineers, a group who were suddenly struck by an exciting, incredible idea. A group that worked in the quiet and secrecy of a University on a fantastic scheme to force the idea of peace into the minds of the world's big shots. Does my dream interest you, Professor?" "Go on." "Well, I dreamt that this group would secretly launch an earth satellite of their own, and arrange for the nose cone to come down safely at a certain time and place. They would install a marvelous electronic robot within the cone, ready to be assembled. They would beam a radio message to earth from the cone, seemingly as if it originated from their 'spaceship.' Then, when the Robot was assembled, they would speak through it to demand peace for all mankind ..." "Jerry, if you do this—" "You don't have to say it, Professor, I know what you're thinking. I'm a reporter, and my business is to tell the world everything I know. But if I did it, there might not be a world for me to write about, would there? No, thanks, Professor. As far as I'm concerned, what I told you was nothing more than a daydream." Jerry braked the convertible to a halt, and put his arm around Greta's shoulder. She looked up at the star-filled night, and sighed romantically. Jerry pointed. "That one." Greta shivered closer to him. "And to think what that terrible planet can do to us!" "Oh, I dunno. Venus is also the Goddess of Love." He swung his other arm around her, and Venus winked approvingly. THE END Transcriber's Note: This etext was produced from Amazing Science Fiction Stories October 1958. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
C. The robot was voiced by Professor Coltz remotely.
|
What motivated the leprechauns to build a spaceship?
A. They desire to seek and add more riches to their already expansive collection
B. They believe that humans' obsession with technology will make the world inhabitable
C. They fear that their race will soon become extinct due to population decline
D. They wish to transport their riches to another location where humans will never steal it
|
Every writer must seek his own Flowery Kingdom in imagination's wide demesne, and if that search can begin and end on Earth his problem has been greatly simplified. In post-war Japan Walt Sheldon has found not only serenity, but complete freedom to write undisturbed about the things he treasures most. A one-time Air Force officer, he has turned to fantasy in his lighter moments, to bring us such brightly sparkling little gems as this. houlihan's equation by ... Walt Sheldon The tiny spaceship had been built for a journey to a star. But its small, mischievous pilots had a rendezvous with destiny—on Earth. I must admit that at first I wasn't sure I was hearing those noises. It was in a park near the nuclear propulsion center—a cool, green spot, with the leaves all telling each other to hush, be quiet, and the soft breeze stirring them up again. I had known precisely such a secluded little green sanctuary just over the hill from Mr. Riordan's farm when I was a boy. Now it was a place I came to when I had a problem to thrash out. That morning I had been trying to work out an equation to give the coefficient of discharge for the matter in combustion. You may call it gas, if you wish, for we treated it like gas at the center for convenience—as it came from the rocket tubes in our engine. Without this coefficient to give us control, we would have lacked a workable equation when we set about putting the first moon rocket around those extraordinary engines of ours, which were still in the undeveloped blueprint stage. I see I shall have to explain this, although I had hoped to get right along with my story. When you start from scratch, matter discharged from any orifice has a velocity directly proportional to the square root of the pressure-head driving it. But when you actually put things together, contractions or expansions in the gas, surface roughness and other factors make the velocity a bit smaller. At the terrible discharge speed of nuclear explosion—which is what the drive amounts to despite the fact that it is simply water in which nuclear salts have been previously dissolved—this small factor makes quite a difference. I had to figure everything into it—diameter of the nozzle, sharpness of the edge, the velocity of approach to the point of discharge, atomic weight and structure— Oh, there is so much of this that if you're not a nuclear engineer yourself it's certain to weary you. Perhaps you had better take my word for it that without this equation—correctly stated, mind you—mankind would be well advised not to make a first trip to the moon. And all this talk of coefficients and equations sits strangely, you might say, upon the tongue of a man named Kevin Francis Houlihan. But I am, after all, a scientist. If I had not been a specialist in my field I would hardly have found myself engaged in vital research at the center. Anyway, I heard these little noises in the park. They sounded like small working sounds, blending in eerily mysterious fashion with a chorus of small voices. I thought at first it might be children at play, but then at the time I was a bit absent-minded. I tiptoed to the edge of the trees, not wanting to deprive any small scalawags of their pleasure, and peered out between the branches. And what do you suppose I saw? Not children, but a group of little people, hard at work. There was a leader, an older one with a crank face. He was beating the air with his arms and piping: "Over here, now! All right, bring those electrical connections over here—and see you're not slow as treacle about it!" There were perhaps fifty of the little people. I was more than startled by it, too. I had not seen little people in—oh, close to thirty years. I had seen them first as a boy of eight, and then, very briefly again, on my tenth birthday. And I had become convinced they could never be seen here in America. I had never seen them so busy, either. They were building something in the middle of the glade. It was long and shiny and upright and a little over five feet in height. "Come along now, people!" said this crotchety one, looking straight at me. "Stop starin' and get to work! You'll not be needin' to mind that man standin' there! You know he can't see nor hear us!" Oh, it was good to hear the rich old tongue again. I smiled, and the foreman of the leprechauns—if that's what he was—saw me smile and became stiff and alert for a moment, as though suspecting that perhaps I actually could see him. Then he shrugged and turned away, clearly deeming such a thing impossible. I said, "Just a minute, friend, and I'll beg your pardon. It so happens I can see you." He whirled to face me again, staring open-mouthed. Then he said, "What? What's that, now?" "I can see you," I said. "Ohhh!" he said and put his palms to his cheekbones. "Saints be with us! He's a believer! Run everybody—run for your lives!" And they all began running, in as many directions as there were little souls. They began to scurry behind the trees and bushes, and a sloping embankment nearby. "No, wait!" I said. "Don't go away! I'll not be hurting you!" They continued to scurry. I knew what it was they feared. "I don't intend catching one of you!" I said. "Come back, you daft little creatures!" But the glade was silent, and they had all disappeared. They thought I wanted their crock of gold, of course. I'd be entitled to it if I could catch one and keep him. Or so the legends affirmed, though I've wondered often about the truth of them. But I was after no gold. I only wanted to hear the music of an Irish tongue. I was lonely here in America, even if I had latched on to a fine job of work for almost shamefully generous pay. You see, in a place as full of science as the nuclear propulsion center there is not much time for the old things. I very much wanted to talk to the little people. I walked over to the center of the glade where the curious shiny object was standing. It was as smooth as glass and shaped like a huge cigar. There were a pair of triangular fins down at the bottom, and stubby wings amidships. Of course it was a spaceship, or a miniature replica of one. I looked at it more closely. Everything seemed almost miraculously complete and workable. I shook my head in wonder, then stepped back from the spaceship and looked about the glade. I knew they were all hiding nearby, watching me apprehensively. I lifted my head to them. "Listen to me now, little people!" I called out. "My name's Houlihan of the Roscommon Houlihans. I am descended from King Niall himself—or so at least my father used to say! Come on out now, and pass the time o' day!" Then I waited, but they didn't answer. The little people always had been shy. Yet without reaching a decision in so many words I knew suddenly that I had to talk to them. I'd come to the glen to work out a knotty problem, and I was up against a blank wall. Simply because I was so lonely that my mind had become clogged. I knew that if I could just once hear the old tongue again, and talk about the old things, I might be able to think the problem through to a satisfactory conclusion. So I stepped back to the tiny spaceship, and this time I struck it a resounding blow with my fist. "Hear me now, little people! If you don't show yourselves and come out and talk to me, I'll wreck this spaceship from stem to stern!" I heard only the leaves rustling softly. "Do you understand? I'll give you until I count three to make an appearance! One!" The glade remained deathly silent. "Two!" I thought I heard a stirring somewhere, as if a small, brittle twig had snapped in the underbrush. " Three! " And with that the little people suddenly appeared. The leader—he seemed more wizened and bent than before—approached me slowly and warily as I stood there. The others all followed at a safe distance. I smiled to reassure them and then waved my arm in a friendly gesture of greeting. "Good morning," I said. "Good morning," the foreman said with some caution. "My name is Keech." "And mine's Houlihan, as I've told you. Are you convinced now that I have no intention of doing you any injury?" "Mr. Houlihan," said Keech, drawing a kind of peppered dignity up about himself, "in such matters I am never fully convinced. After living for many centuries I am all too acutely aware of the perversity of human nature." "Yes," I said. "Well, as you will quickly see, all I want to do is talk." I nodded as I spoke, and sat down cross-legged upon the grass. "Any Irishman wants to talk, Mr. Houlihan." "And often that's all he wants," I said. "Sit down with me now, and stop staring as if I were a snake returned to the Island." He shook his head and remained standing. "Have your say, Mr. Houlihan. And afterward we'll appreciate it if you'll go away and leave us to our work." "Well, now, your work," I said, and glanced at the spaceship. "That's exactly what's got me curious." The others had edged in a bit now and were standing in a circle, intently staring at me. I took out my pipe. "Why," I asked, "would a group of little people be building a spaceship here in America—out in this lonely place?" Keech stared back without much expression, and said, "I've been wondering how you guessed it was a spaceship. I was surprised enough when you told me you could see us but not overwhelmingly so. I've run into believers before who could see the little people. It happens every so often, though not as frequently as it did a century ago. But knowing a spaceship at first glance! Well, I must confess that does astonish me." "And why wouldn't I know a spaceship when I see one?" I said. "It just so happens I'm a doctor of science." "A doctor of science, now," said Keech. "Invited by the American government to work on the first moon rocket here at the nuclear propulsion center. Since it's no secret I can advise you of it." "A scientist, is it," said Keech. "Well, now, that's very interesting." "I'll make no apologies for it," I said. "Oh, there's no need for apology," said Keech. "Though in truth we prefer poets to scientists. But it has just now crossed my mind, Mr. Houlihan that you, being a scientist, might be of help to us." "How?" I asked. "Well, I might try starting at the beginning," he replied. "You might," I said. "A man usually does." Keech took out his own pipe—a clay dudeen—and looked hopeful. I gave him a pinch of tobacco from my pouch. "Well, now," he said, "first of all you're no doubt surprised to find us here in America." "I am surprised from time to time to find myself here," I said. "But continue." "We had to come here," said Keech, "to learn how to make a spaceship." "A spaceship, now," I said, unconsciously adopting some of the old manner. "Leprechauns are not really mechanically inclined," said Keech. "Their major passions are music and laughter and mischief, as anyone knows." "Myself included," I agreed. "Then why do you need a spaceship?" "Well, if I may use an old expression, we've had a feelin' lately that we're not long for this world. Or let me put it this way. We feel the world isn't long for itself." I scratched my cheek. "How would a man unravel a statement such as that?" "It's very simple. With all the super weapons you mortals have developed, there's the distinct possibility you might be blowin' us all up in the process of destroying yourselves." "There is that possibility," I said. "Well, then, as I say," said Keech, "the little people have decided to leave the planet in a spaceship. Which we're buildin' here and now. We've spied upon you and learned how to do it. Well—almost how to do it. We haven't learned yet how to control the power—" "Hold on, now," I said. "Leaving the planet, you say. And where would you be going?" "There's another committee working on that. 'Tis not our concern. I was inclined to suggest the constellation Orion, which sounds as though it has a good Irish name, but I was hooted down. Be that as it may, my own job was to go into your nuclear center, learn how to make the ship, and proceed with its construction. Naturally, we didn't understand all of your high-flyin' science, but some of our people are pretty clever at gettin' up replicas of things." "You mean you've been spying on us at the center all this time? Do you know, we often had the feeling we were being watched, but we thought it was by the Russians. There's one thing which puzzles me, though. If you've been constantly around us—and I'm still able to see the little people—why did I never see you before?" "It may be we never crossed your path. It may be you can only see us when you're thinkin' of us, and of course truly believin' in us. I don't know—'tis a thing of the mind, and not important at the moment. What's important is for us to get our first ship to workin' properly and then we'll be on our way." "You're determined to go." "Truly we are, Mr. Houlihan. Now—to business. Just during these last few minutes a certain matter has crossed my mind. That's why I'm wastin' all this time with you, sir. You say you are a scientist." "A nuclear engineer." "Well, then, it may be that you can help us—now that you know we're here." "Help you?" "The power control, Mr. Houlihan. As I understand it, 'tis necessary to know at any instant exactly how much thrust is bein' delivered through the little holes in back. And on paper it looks simple enough—the square of somethin' or other. I've got the figures jotted in a book when I need 'em. But when you get to doin' it it doesn't come out exactly as it does on paper." "You're referring to the necessity for a coefficient of discharge." "Whatever it might be named," said Keech, shrugging. "'Tis the one thing we lack. I suppose eventually you people will be gettin' around to it. But meanwhile we need it right now, if we're to make our ship move." "And you want me to help you with this?" "That is exactly what crossed my mind." I nodded and looked grave and kneaded my chin for a moment softly. "Well, now, Keech," I said finally, "why should I help you?" "Ha!" said Keech, grinning, but not with humor, "the avarice of humans! I knew it! Well, Mr. Houlihan, I'll give you reason enough. The pot o' gold, Mr. Houlihan!" "The one at the end of the rainbow?" "It's not at the end of the rainbow. That's a grandmother's tale. Nor is it actually in an earthen crock. But there's gold, all right, enough to make you rich for the rest of your life. And I'll make you a proposition." "Go ahead." "We'll not be needin' gold where we're goin'. It's yours if you show us how to make our ship work." "Well, now, that's quite an offer," I said. Keech had the goodness to be quiet while I sat and thought for a while. My pipe had gone out and I lit it again. I finally said, "Let's have a look at your ship's drive and see what we can see." "You accept the proposition then?" "Let's have a look," I said, and that was all. Well, we had a look, and then several looks, and before the morning was out we had half the spaceship apart, and were deep in argument about the whole project. It was a most fascinating session. I had often wished for a true working model at the center, but no allowance had been inserted in the budget for it. Keech brought me paper and pencil and I talked with the aid of diagrams, as engineers are wont to do. Although the pencils were small and I had to hold them between thumb and forefinger, as you would a needle, I was able to make many sensible observations and even a few innovations. I came back again the next day—and every day for the following two weeks. It rained several times, but Keech and his people made a canopy of boughs and leaves and I was comfortable enough. Every once in a while someone from the town or the center itself would pass by, and stop to watch me. But of course they wouldn't see the leprechauns or anything the leprechauns had made, not being believers. I would halt work, pass the time of day, and then, in subtle fashion, send the intruder on his way. Keech and the little people just stood by and grinned all the while. At the end of sixteen days I had the entire problem all but whipped. It is not difficult to understand why. The working model and the fact that the small people with their quick eyes and clever fingers could spot all sorts of minute shortcomings was a great help. And I was hearing the old tongue and talking of the old things every day, and truly that went far to take the clutter out of my mind. I was no longer so lonely that I couldn't think properly. On the sixteenth day I covered a piece of paper with tiny mathematical symbols and handed it to Keech. "Here is your equation," I said. "It will enable you to know your thrust at any given moment, under any circumstances, in or out of gravity, and under all conditions of friction and combustion." "Thank you, Mr. Houlihan," said Keech. All his people had gathered in a loose circle, as though attending a rite. They were all looking at me quietly. "Mr. Houlihan," said Keech, "you will not be forgotten by the leprechauns. If we ever meet again, upon another world perchance, you'll find our friendship always eager and ready." "Thank you," I said. "And now, Mr. Houlihan," said Keech, "I'll see that a quantity of gold is delivered to your rooms tonight, and so keep my part of the bargain." "I'll not be needing the gold," I said. Keech's eyebrows popped upward. "What's this now?" "I'll not be needing it," I repeated. "I don't feel it would be right to take it for a service of this sort." "Well," said Keech in surprise, and in some awe, too, "well, now, musha Lord help us! 'Tis the first time I ever heard such a speech from a mortal." He turned to his people. "We'll have three cheers now, do you hear, for Mr. Houlihan—friend of the little people as long as he shall live!" And they cheered. And little tears crept into the corners of some of their turned-up eyes. We shook hands, all of us, and I left. I walked through the park, and back to the nuclear propulsion center. It was another cool, green morning with the leaves making only soft noises as the breezes came along. It smelled exactly like a wood I had known in Roscommon. And I lit my pipe and smoked it slowly and chuckled to myself at how I had gotten the best of the little people. Surely it was not every mortal who could accomplish that. I had given them the wrong equation, of course. They would never get their spaceship to work now, and later, if they tried to spy out the right information I would take special measures to prevent it, for I had the advantage of being able to see them. As for our own rocket ship, it should be well on its way by next St. Patrick's Day. For I had indeed determined the true coefficient of discharge, which I never could have done so quickly without those sessions in the glade with Keech and his working model. It would go down in scientific literature now, I suppose, as Houlihan's Equation, and that was honor and glory enough for me. I could do without Keech's pot of gold, though it would have been pleasant to be truly rich for a change. There was no sense in cheating him out of the gold to boot, for leprechauns are most clever in matters of this sort and he would have had it back soon enough—or else made it a burden in some way. Indeed, I had done a piece of work greatly to my advantage, and also to the advantage of humankind, and when a man can do the first and include the second as a fortunate byproduct it is a most happy accident. For if I had shown the little people how to make a spaceship they would have left our world. And this world, as long as it lasts—what would it be in that event? I ask you now, wouldn't we be even more likely to blow ourselves to Kingdom Come without the little people here for us to believe in every now and then? Transcriber's Note: This etext was produced from Fantastic Universe September 1955. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
B. They believe that humans' obsession with technology will make the world inhabitable
|
How does Meyerhoff feel about Zeckler?
A. Meyerhoff thinks that Zeckler is a fool.
B. Meyerhoff thinks that Zeckler is a skilled con-man.
C. Meyerhoff thinks that Zeckler is misunderstood.
D. Meyerhoff thinks that Zeckler is an idiot.
|
Letter of the Law by Alan E. Nourse The place was dark and damp, and smelled like moldy leaves. Meyerhoff followed the huge, bear-like Altairian guard down the slippery flagstones of the corridor, sniffing the dead, musty air with distaste. He drew his carefully tailored Terran-styled jacket closer about his shoulders, shivering as his eyes avoided the black, yawning cell-holes they were passing. His foot slipped on the slimy flags from time to time, and finally he paused to wipe the caked mud from his trouser leg. "How much farther is it?" he shouted angrily. The guard waved a heavy paw vaguely into the blackness ahead. Quite suddenly the corridor took a sharp bend, and the Altairian stopped, producing a huge key ring from some obscure fold of his hairy hide. "I still don't see any reason for all the fuss," he grumbled in a wounded tone. "We've treated him like a brother." One of the huge steel doors clicked open. Meyerhoff peered into the blackness, catching a vaguely human outline against the back wall. "Harry?" he called sharply. There was a startled gasp from within, and a skinny, gnarled little man suddenly appeared in the guard's light, like a grotesque, twisted ghost out of the blackness. Wide blue eyes regarded Meyerhoff from beneath uneven black eyebrows, and then the little man's face broke into a crafty grin. "Paul! So they sent you ! I knew I could count on it!" He executed a deep, awkward bow, motioning Meyerhoff into the dark cubicle. "Not much to offer you," he said slyly, "but it's the best I can do under the circumstances." Meyerhoff scowled, and turned abruptly to the guard. "We'll have some privacy now, if you please. Interplanetary ruling. And leave us the light." The guard grumbled, and started for the door. "It's about time you showed up!" cried the little man in the cell. "Great day! Lucky they sent you, pal. Why, I've been in here for years—" "Look, Zeckler, the name is Meyerhoff, and I'm not your pal," Meyerhoff snapped. "And you've been here for two weeks, three days, and approximately four hours. You're getting as bad as your gentle guards when it comes to bandying the truth around." He peered through the dim light at the gaunt face of the prisoner. Zeckler's face was dark with a week's beard, and his bloodshot eyes belied the cocky grin on his lips. His clothes were smeared and sodden, streaked with great splotches of mud and moss. Meyerhoff's face softened a little. "So Harry Zeckler's in a jam again," he said. "You look as if they'd treated you like a brother." The little man snorted. "These overgrown teddy-bears don't know what brotherhood means, nor humanity, either. Bread and water I've been getting, nothing more, and then only if they feel like bringing it down." He sank wearily down on the rock bench along the wall. "I thought you'd never get here! I sent an appeal to the Terran Consulate the first day I was arrested. What happened? I mean, all they had to do was get a man over here, get the extradition papers signed, and provide transportation off the planet for me. Why so much time? I've been sitting here rotting—" He broke off in mid-sentence and stared at Meyerhoff. "You brought the papers, didn't you? I mean, we can leave now?" Meyerhoff stared at the little man with a mixture of pity and disgust. "You are a prize fool," he said finally. "Did you know that?" Zeckler's eyes widened. "What do you mean, fool? So I spend a couple of weeks in this pneumonia trap. The deal was worth it! I've got three million credits sitting in the Terran Consulate on Altair V, just waiting for me to walk in and pick them up. Three million credits—do you hear? That's enough to set me up for life!" Meyerhoff nodded grimly. " If you live long enough to walk in and pick them up, that is." "What do you mean, if?" Meyerhoff sank down beside the man, his voice a tense whisper in the musty cell. "I mean that right now you are practically dead. You may not know it, but you are. You walk into a newly opened planet with your smart little bag of tricks, walk in here with a shaky passport and no permit, with no knowledge of the natives outside of two paragraphs of inaccuracies in the Explorer's Guide, and even then you're not content to come in and sell something legitimate, something the natives might conceivably be able to use. No, nothing so simple for you. You have to pull your usual high-pressure stuff. And this time, buddy, you're paying the piper." " You mean I'm not being extradited? " Meyerhoff grinned unpleasantly. "I mean precisely that. You've committed a crime here—a major crime. The Altairians are sore about it. And the Terran Consulate isn't willing to sell all the trading possibilities here down the river just to get you out of a mess. You're going to stand trial—and these natives are out to get you. Personally, I think they're going to get you." Zeckler stood up shakily. "You can't believe anything the natives say," he said uneasily. "They're pathological liars. Why, you should see what they tried to sell me ! You've never seen such a pack of liars as these critters." He glanced up at Meyerhoff. "They'll probably drop a little fine on me and let me go." "A little fine of one Terran neck." Meyerhoff grinned nastily. "You've committed the most heinous crime these creatures can imagine, and they're going to get you for it if it's the last thing they do. I'm afraid, my friend, that your con-man days are over." Zeckler fished in the other man's pocket, extracted a cigarette, and lighted it with trembling fingers. "It's bad, then," he said finally. "It's bad, all right." Some shadow of the sly, elfin grin crept over the little con-man's face. "Well, at any rate, I'm glad they sent you over," he said weakly. "Nothing like a good lawyer to handle a trial." " Lawyer? Not me! Oh, no. Sorry, but no thanks." Meyerhoff chuckled. "I'm your advisor, old boy. Nothing else. I'm here to keep you from botching things up still worse for the Trading Commission, that's all. I wouldn't get tangled up in a mess with those creatures for anything!" He shook his head. "You're your own lawyer, Mr. Super-salesman. It's all your show. And you'd better get your head out of the sand, or you're going to lose a case like it's never been lost before!" Meyerhoff watched the man's pale face, and shook his head. In a way, he thought, it was a pity to see such a change in the rosy-cheeked, dapper, cocksure little man who had talked his way glibly in and out of more jams than Meyerhoff could count. Trading brought scalpers; it was almost inevitable that where rich and unexploited trading ground was uncovered, it would first fall prey to the fast-trading boys. They spread out from Terra with the first wave of exploration—the slick, fast-talking con-men who could work new territories unfettered by the legal restrictions that soon closed down the more established planets. The first men in were the richest out, and through some curious quirk of the Terrestrial mind, they knew they could count on Terran protection, however crooked and underhand their methods. But occasionally a situation arose where the civilization and social practices of the alien victims made it unwise to tamper with them. Altair I had been recognized at once by the Trading Commission as a commercial prize of tremendous value, but early reports had warned of the danger of wildcat trading on the little, musty, jungle-like planet with its shaggy, three-eyed inhabitants—warned specifically against the confidence tactics so frequently used—but there was always somebody, Meyerhoff reflected sourly, who just didn't get the word. Zeckler puffed nervously on his cigarette, his narrow face a study in troubled concentration. "But I didn't do anything!" he exploded finally. "So I pulled an old con game. So what? Why should they get so excited? So I clipped a few thousand credits, pulled a little fast business." He shrugged eloquently, spreading his hands. "Everybody's doing it. They do it to each other without batting an eye. You should see these critters operate on each other. Why, my little scheme was peanuts by comparison." Meyerhoff pulled a pipe from his pocket, and began stuffing the bowl with infinite patience. "And precisely what sort of con game was it?" he asked quietly. Zeckler shrugged again. "The simplest, tiredest, moldiest old racket that ever made a quick nickel. Remember the old Terran gag about the Brooklyn Bridge? The same thing. Only these critters didn't want bridges. They wanted land—this gooey, slimy swamp they call 'farm land.' So I gave them what they wanted. I just sold them some land." Meyerhoff nodded fiercely. "You sure did. A hundred square kilos at a swipe. Only you sold the same hundred square kilos to a dozen different natives." Suddenly he threw back his hands and roared. "Of all the things you shouldn't have done—" "But what's a chunk of land?" Meyerhoff shook his head hopelessly. "If you hadn't been so greedy, you'd have found out what a chunk of land was to these natives before you started peddling it. You'd have found out other things about them, too. You'd have learned that in spite of all their bumbling and fussing and squabbling they're not so dull. You'd have found out that they're marsupials, and that two out of five of them get thrown out of their mother's pouch before they're old enough to survive. You'd have realized that they have to start fighting for individual rights almost as soon as they're born. Anything goes, as long as it benefits them as individuals." Meyerhoff grinned at the little man's horrified face. "Never heard of that, had you? And you've never heard of other things, too. You've probably never heard that there are just too many Altairians here for the food their planet can supply, and their diet is so finicky that they just can't live on anything that doesn't grow here. And consequently, land is the key factor in their economy, not money; nothing but land. To get land, it's every man for himself, and the loser starves, and their entire legal and monetary system revolves on that principle. They've built up the most confusing and impossible system of barter and trade imaginable, aimed at individual survival, with land as the value behind the credit. That explains the lying—of course they're liars, with an economy like that. They've completely missed the concept of truth. Pathological? You bet they're pathological! Only a fool would tell the truth when his life depended on his being a better liar than the next guy! Lying is the time-honored tradition, with their entire legal system built around it." Zeckler snorted. "But how could they possibly have a legal system? I mean, if they don't recognize the truth when it slaps them in the face?" Meyerhoff shrugged. "As we understand legal systems, I suppose they don't have one. They have only the haziest idea what truth represents, and they've shrugged off the idea as impossible and useless." He chuckled maliciously. "So you went out and found a chunk of ground in the uplands, and sold it to a dozen separate, self-centered, half-starved natives! Encroachment on private property is legal grounds for murder on this planet, and twelve of them descended on the same chunk of land at the same time, all armed with title-deeds." Meyerhoff sighed. "You've got twelve mad Altairians in your hair. You've got a mad planet in your hair. And in the meantime, Terra's most valuable uranium source in five centuries is threatening to cut off supply unless they see your blood splattered liberally all the way from here to the equator." Zeckler was visibly shaken. "Look," he said weakly, "so I wasn't so smart. What am I going to do? I mean, are you going to sit quietly by and let them butcher me? How could I defend myself in a legal setup like this ?" Meyerhoff smiled coolly. "You're going to get your sly little con-man brain to working, I think," he said softly. "By Interplanetary Rules, they have to give you a trial in Terran legal form—judge, jury, court procedure, all that folderol. They think it's a big joke—after all, what could a judicial oath mean to them?—but they agreed. Only thing is, they're going to hang you, if they die trying. So you'd better get those stunted little wits of yours clicking—and if you try to implicate me , even a little bit, I'll be out of there so fast you won't know what happened." With that Meyerhoff walked to the door. He jerked it inward sharply, and spilled two guards over on their faces. "Privacy," he grunted, and started back up the slippery corridor. It certainly looked like a courtroom, at any rate. In the front of the long, damp stone room was a bench, with a seat behind it, and a small straight chair to the right. To the left was a stand with twelve chairs—larger chairs, with a railing running along the front. The rest of the room was filled almost to the door with seats facing the bench. Zeckler followed the shaggy-haired guard into the room, nodding approvingly. "Not such a bad arrangement," he said. "They must have gotten the idea fast." Meyerhoff wiped the perspiration from his forehead, and shot the little con-man a stony glance. "At least you've got a courtroom, a judge, and a jury for this mess. Beyond that—" He shrugged eloquently. "I can't make any promises." In the back of the room a door burst open with a bang. Loud, harsh voices were heard as half a dozen of the huge Altairians attempted to push through the door at once. Zeckler clamped on the headset to his translator unit, and watched the hubbub in the anteroom with growing alarm. Finally the question of precedent seemed to be settled, and a group of the Altairians filed in, in order of stature, stalking across the room in flowing black robes, pug-nosed faces glowering with self-importance. They descended upon the jury box, grunting and scrapping with each other for the first-row seats, and the judge took his place with obvious satisfaction behind the heavy wooden bench. Finally, the prosecuting attorney appeared, flanked by two clerks, who took their places beside him. The prosecutor eyed Zeckler with cold malevolence, then turned and delivered a sly wink at the judge. In a moment the room was a hubbub as it filled with the huge, bumbling, bear-like creatures, jostling each other and fighting for seats, growling and complaining. Two small fights broke out in the rear, but were quickly subdued by the group of gendarmes guarding the entrance. Finally the judge glared down at Zeckler with all three eyes, and pounded the bench top with a wooden mallet until the roar of activity subsided. The jurymen wriggled uncomfortably in their seats, exchanging winks, and finally turned their attention to the front of the court. "We are reading the case of the people of Altair I," the judge's voice roared out, "against one Harry Zeckler—" he paused for a long, impressive moment—"Terran." The courtroom immediately burst into an angry growl, until the judge pounded the bench five or six times more. "This—creature—is hereby accused of the following crimes," the judge bellowed. "Conspiracy to overthrow the government of Altair I. Brutal murder of seventeen law-abiding citizens of the village of Karzan at the third hour before dawn in the second period after his arrival. Desecration of the Temple of our beloved Goddess Zermat, Queen of the Harvest. Conspiracy with the lesser gods to cause the unprecedented drought in the Dermatti section of our fair globe. Obscene exposure of his pouch-marks in a public square. Four separate and distinct charges of jail-break and bribery—" The judge pounded the bench for order—"Espionage with the accursed scum of Altair II in preparation for interplanetary invasion." The little con-man's jaw sagged lower and lower, the color draining from his face. He turned, wide-eyed, to Meyerhoff, then back to the judge. "The Chairman of the Jury," said the Judge succinctly, "will read the verdict." The little native in the front of the jury-box popped up like a puppet on a string. "Defendant found guilty on all counts," he said. "Defendant is guilty! The court will pronounce sentence—" " Now wait a minute! " Zeckler was on his feet, wild-eyed. "What kind of railroad job—" The judge blinked disappointedly at Paul Meyerhoff. "Not yet?" he asked, unhappily. "No." Meyerhoff's hands twitched nervously. "Not yet, Your Honor. Later, Your Honor. The trial comes first ." The judge looked as if his candy had been stolen. "But you said I should call for the verdict." "Later. You have to have the trial before you can have the verdict." The Altairian shrugged indifferently. "Now—later—" he muttered. "Have the prosecutor call his first witness," said Meyerhoff. Zeckler leaned over, his face ashen. "These charges," he whispered. "They're insane!" "Of course they are," Meyerhoff whispered back. "But what am I going to—" "Sit tight. Let them set things up." "But those lies . They're liars, the whole pack of them—" He broke off as the prosecutor roared a name. The shaggy brute who took the stand was wearing a bright purple hat which sat rakishly over one ear. He grinned the Altairian equivalent of a hungry grin at the prosecutor. Then he cleared his throat and started. "This Terran riffraff—" "The oath," muttered the judge. "We've got to have the oath." The prosecutor nodded, and four natives moved forward, carrying huge inscribed marble slabs to the front of the court. One by one the chunks were reverently piled in a heap at the witness's feet. The witness placed a huge, hairy paw on the cairn, and the prosecutor said, "Do you swear to tell the truth, the whole truth, and nothing but the truth, so help you—" he paused to squint at the paper in his hand, and finished on a puzzled note, "—Goddess?" The witness removed the paw from the rock pile long enough to scratch his ear. Then he replaced it, and replied, "Of course," in an injured tone. "Then tell this court what you have seen of the activities of this abominable wretch." The witness settled back into the chair, fixing one eye on Zeckler's face, another on the prosecutor, and closing the third as if in meditation. "I think it happened on the fourth night of the seventh crossing of Altair II (may the Goddess cast a drought upon it)—or was it the seventh night of the fourth crossing?—" he grinned apologetically at the judge—"when I was making my way back through town toward my blessed land-plot, minding my own business, Your Honor, after weeks of bargaining for the crop I was harvesting. Suddenly from the shadow of the building, this creature—" he waved a paw at Zeckler—"stopped me in my tracks with a vicious cry. He had a weapon I'd never seen before, and before I could find my voice he forced me back against the wall. I could see by the cruel glint in his eyes that there was no warmth, no sympathy in his heart, that I was—" "Objection!" Zeckler squealed plaintively, jumping to his feet. "This witness can't even remember what night he's talking about!" The judge looked startled. Then he pawed feverishly through his bundle of notes. "Overruled," he said abruptly. "Continue, please." The witness glowered at Zeckler. "As I was saying before this loutish interruption," he muttered, "I could see that I was face to face with the most desperate of criminal types, even for Terrans. Note the shape of his head, the flabbiness of his ears. I was petrified with fear. And then, helpless as I was, this two-legged abomination began to shower me with threats of evil to my blessed home, dark threats of poisoning my land unless I would tell him where he could find the resting place of our blessed Goddess—" "I never saw him before in my life," Zeckler moaned to Meyerhoff. "Listen to him! Why should I care where their Goddess—" Meyerhoff gave him a stony look. "The Goddess runs things around here. She makes it rain. If it doesn't rain, somebody's insulted her. It's very simple." "But how can I fight testimony like that?" "I doubt if you can fight it." "But they can't prove a word of it—" He looked at the jury, who were listening enraptured to the second witness on the stand. This one was testifying regarding the butcherous slaughter of eighteen (or was it twenty-three? Oh, yes, twenty-three) women and children in the suburban village of Karzan. The pogrom, it seemed, had been accomplished by an energy weapon which ate great, gaping holes in the sides of buildings. A third witness took the stand, continuing the drone as the room grew hotter and muggier. Zeckler grew paler and paler, his eyes turning glassy as the testimony piled up. "But it's not true ," he whispered to Meyerhoff. "Of course it isn't! Can't you understand? These people have no regard for truth. It's stupid, to them, silly, a mark of low intelligence. The only thing in the world they have any respect for is a liar bigger and more skillful than they are." Zeckler jerked around abruptly as he heard his name bellowed out. "Does the defendant have anything to say before the jury delivers the verdict?" "Do I have—" Zeckler was across the room in a flash, his pale cheeks suddenly taking on a feverish glow. He sat down gingerly on the witness chair, facing the judge, his eyes bright with fear and excitement. "Your—Your Honor, I—I have a statement to make which will have a most important bearing on this case. You must listen with the greatest care." He glanced quickly at Meyerhoff, and back to the judge. "Your Honor," he said in a hushed voice. "You are in gravest of danger. All of you. Your lives—your very land is at stake." The judge blinked, and shuffled through his notes hurriedly as a murmur arose in the court. "Our land?" "Your lives, your land, everything you hold dear," Zeckler said quickly, licking his lips nervously. "You must try to understand me—" he glanced apprehensively over his shoulder "now, because I may not live long enough to repeat what I am about to tell you—" The murmur quieted down, all ears straining in their headsets to hear his words. "These charges," he continued, "all of them—they're perfectly true. At least, they seem to be perfectly true. But in every instance, I was working with heart and soul, risking my life, for the welfare of your beautiful planet." There was a loud hiss from the back of the court. Zeckler frowned and rubbed his hands together. "It was my misfortune," he said, "to go to the wrong planet when I first came to Altair from my homeland on Terra. I—I landed on Altair II, a grave mistake, but as it turned out, a very fortunate error. Because in attempting to arrange trading in that frightful place, I made certain contacts." His voice trembled, and sank lower. "I learned the horrible thing which is about to happen to this planet, at the hands of those barbarians. The conspiracy is theirs, not mine. They have bribed your Goddess, flattered her and lied to her, coerced her all-powerful goodness to their own evil interests, preparing for the day when they could persuade her to cast your land into the fiery furnace of a ten-year-drought—" Somebody in the middle of the court burst out laughing. One by one the natives nudged one another, and booed, and guffawed, until the rising tide of racket drowned out Zeckler's words. "The defendant is obviously lying," roared the prosecutor over the pandemonium. "Any fool knows that the Goddess can't be bribed. How could she be a Goddess if she could?" Zeckler grew paler. "But—perhaps they were very clever—" "And how could they flatter her, when she knows, beyond doubt, that she is the most exquisitely radiant creature in all the Universe? And you dare to insult her, drag her name in the dirt." The hisses grew louder, more belligerent. Cries of "Butcher him!" and "Scald his bowels!" rose from the courtroom. The judge banged for silence, his eyes angry. "Unless the defendant wishes to take up more of our precious time with these ridiculous lies, the jury—" "Wait! Your Honor, I request a short recess before I present my final plea." "Recess?" "A few moments to collect my thoughts, to arrange my case." The judge settled back with a disgusted snarl. "Do I have to?" he asked Meyerhoff. Meyerhoff nodded. The judge shrugged, pointing over his shoulder to the anteroom. "You can go in there," he said. Somehow, Zeckler managed to stumble from the witness stand, amid riotous boos and hisses, and tottered into the anteroom. Zeckler puffed hungrily on a cigarette, and looked up at Meyerhoff with haunted eyes. "It—it doesn't look so good," he muttered. Meyerhoff's eyes were worried, too. For some reason, he felt a surge of pity and admiration for the haggard con-man. "It's worse than I'd anticipated," he admitted glumly. "That was a good try, but you just don't know enough about them and their Goddess." He sat down wearily. "I don't see what you can do. They want your blood, and they're going to have it. They just won't believe you, no matter how big a lie you tell." Zeckler sat in silence for a moment. "This lying business," he said finally, "exactly how does it work?" "The biggest, most convincing liar wins. It's as simple as that. It doesn't matter how outlandish a whopper you tell. Unless, of course, they've made up their minds that you just naturally aren't as big a liar as they are. And it looks like that's just what they've done. It wouldn't make any difference to them what you say—unless, somehow, you could make them believe it." Zeckler frowned. "And how do they regard the—the biggest liar? I mean, how do they feel toward him?" Meyerhoff shifted uneasily. "It's hard to say. It's been my experience that they respect him highly—maybe even fear him a little. After all, the most convincing liar always wins in any transaction, so he gets more land, more food, more power. Yes, I think the biggest liar could go where he pleased without any interference." Zeckler was on his feet, his eyes suddenly bright with excitement. "Wait a minute," he said tensely. "To tell them a lie that they'd have to believe—a lie they simply couldn't help but believe—" He turned on Meyerhoff, his hands trembling. "Do they think the way we do? I mean, with logic, cause and effect, examining evidence and drawing conclusions? Given certain evidence, would they have to draw the same conclusions that we have to draw?" Meyerhoff blinked. "Well—yes. Oh, yes, they're perfectly logical." Zeckler's eyes flashed, and a huge grin broke out on his sallow face. His thin body fairly shook. He started hopping up and down on one foot, staring idiotically into space. "If I could only think—" he muttered. "Somebody—somewhere—something I read." "Whatever are you talking about?" "It was a Greek, I think—" Meyerhoff stared at him. "Oh, come now. Have you gone off your rocker completely? You've got a problem on your hands, man." "No, no, I've got a problem in the bag!" Zeckler's cheeks flushed. "Let's go back in there—I think I've got an answer!" The courtroom quieted the moment they opened the door, and the judge banged the gavel for silence. As soon as Zeckler had taken his seat on the witness stand, the judge turned to the head juryman. "Now, then," he said with happy finality. "The jury—" "Hold on! Just one minute more." The judge stared down at Zeckler as if he were a bug on a rock. "Oh, yes. You had something else to say. Well, go ahead and say it." Zeckler looked sharply around the hushed room. "You want to convict me," he said softly, "in the worst sort of way. Isn't that right?" Eyes swung toward him. The judge broke into an evil grin. "That's right." "But you can't really convict me until you've considered carefully any statement I make in my own defense. Isn't that right?" The judge looked uncomfortable. "If you've got something to say, go ahead and say it." "I've got just one statement to make. Short and sweet. But you'd better listen to it, and think it out carefully before you decide that you really want to convict me." He paused, and glanced slyly at the judge. "You don't think much of those who tell the truth, it seems. Well, put this statement in your record, then." His voice was loud and clear in the still room. " All Earthmen are absolutely incapable of telling the truth. " Puzzled frowns appeared on the jury's faces. One or two exchanged startled glances, and the room was still as death. The judge stared at him, and then at Meyerhoff, then back. "But you"—he stammered. "You're"—He stopped in mid-sentence, his jaw sagging. One of the jurymen let out a little squeak, and fainted dead away. It took, all in all, about ten seconds for the statement to soak in. And then pandemonium broke loose in the courtroom. "Really," said Harry Zeckler loftily, "it was so obvious I'm amazed that it didn't occur to me first thing." He settled himself down comfortably in the control cabin of the Interplanetary Rocket and grinned at the outline of Altair IV looming larger in the view screen. Paul Meyerhoff stared stonily at the controls, his lips compressed angrily. "You might at least have told me what you were planning." "And take the chance of being overheard? Don't be silly. It had to come as a bombshell. I had to establish myself as a liar—the prize liar of them all, but I had to tell the sort of lie that they simply could not cope with. Something that would throw them into such utter confusion that they wouldn't dare convict me." He grinned impishly at Meyerhoff. "The paradox of Epimenides the Cretan. It really stopped them cold. They knew I was an Earthmen, which meant that my statement that Earthmen were liars was a lie, which meant that maybe I wasn't a liar, in which case—oh, it was tailor-made." "It sure was." Meyerhoff's voice was a snarl. "Well, it made me out a liar in a class they couldn't approach, didn't it?" Meyerhoff's face was purple with anger. "Oh, indeed it did! And it put all Earthmen in exactly the same class, too." "So what's honor among thieves? I got off, didn't I?" Meyerhoff turned on him fiercely. "Oh, you got off just fine. You scared the living daylights out of them. And in an eon of lying they never have run up against a short-circuit like that. You've also completely botched any hope of ever setting up a trading alliance with Altair I, and that includes uranium, too. Smart people don't gamble with loaded dice. You scared them so badly they don't want anything to do with us." Zeckler's grin broadened, and he leaned back luxuriously. "Ah, well. After all, the Trading Alliance was your outlook, wasn't it? What a pity!" He clucked his tongue sadly. "Me, I've got a fortune in credits sitting back at the consulate waiting for me—enough to keep me on silk for quite a while, I might say. I think I'll just take a nice, long vacation." Meyerhoff turned to him, and a twinkle of malignant glee appeared in his eyes. "Yes, I think you will. I'm quite sure of it, in fact. Won't cost you a cent, either." "Eh?" Meyerhoff grinned unpleasantly. He brushed an imaginary lint fleck from his lapel, and looked up at Zeckler slyly. "That—uh—jury trial. The Altairians weren't any too happy to oblige. They wanted to execute you outright. Thought a trial was awfully silly—until they got their money back, of course. Not too much—just three million credits." Zeckler went white. "But that money was in banking custody!" "Is that right? My goodness. You don't suppose they could have lost those papers, do you?" Meyerhoff grinned at the little con-man. "And incidentally, you're under arrest, you know." A choking sound came from Zeckler's throat. " Arrest! " "Oh, yes. Didn't I tell you? Conspiring to undermine the authority of the Terran Trading Commission. Serious charge, you know. Yes, I think we'll take a nice long vacation together, straight back to Terra. And there I think you'll face a jury trial." Zeckler spluttered. "There's no evidence—you've got nothing on me! What kind of a frame are you trying to pull?" "A lovely frame. Airtight. A frame from the bottom up, and you're right square in the middle. And this time—" Meyerhoff tapped a cigarette on his thumb with happy finality—"this time I don't think you'll get off." Transcriber's Note: This etext was produced from "Tiger by the Tail and Other Science Fiction Stories by Alan E. Nourse" and was first published in If Magazine January 1954. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed. Minor spelling and typographical errors have been corrected without note.
|
A. Meyerhoff thinks that Zeckler is a fool.
|
Which term does NOT describe Moran's tone toward the other five crew members?
A. resigned
B. bitter
C. vindictive
D. sarcastic
|
PLANET of DREAD By MURRAY LEINSTER Illustrator ADKINS [Transcriber's Note: This etext was produced from Fantastic Stories of Imagination May 1962. Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed.] I. Moran cut apart the yard-long monstrosity with a slash of flame. The thing presumably died, but it continued to writhe senselessly. He turned to see other horrors crawling toward him. Then he knew he was being marooned on a planet of endless terrors. Moran, naturally, did not mean to help in the carrying out of the plans which would mean his destruction one way or another. The plans were thrashed out very painstakingly, in formal conference on the space-yacht Nadine , with Moran present and allowed to take part in the discussion. From the viewpoint of the Nadine's ship's company, it was simply necessary to get rid of Moran. In their predicament he might have come to the same conclusion; but he was not at all enthusiastic about their decision. He would die of it. The Nadine was out of overdrive and all the uncountable suns of the galaxy shone steadily, remotely, as infinitesimal specks of light of every color of the rainbow. Two hours since, the sun of this solar system had been a vast glaring disk off to port, with streamers and prominences erupting about its edges. Now it lay astern, and Moran could see the planet that had been chosen for his marooning. It was a cloudy world. There were some dim markings near one lighted limb, but nowhere else. There was an ice-cap in view. The rest was—clouds. The ice-cap, by its existence and circular shape, proved that the planet rotated at a not unreasonable rate. The fact that it was water-ice told much. A water-ice ice-cap said that there were no poisonous gases in the planet's atmosphere. Sulfur dioxide or chlorine, for example, would not allow the formation of water-ice. It would have to be sulphuric-acid or hydrochloric-acid ice. But the ice-cap was simple snow. Its size, too, told about temperature-distribution on the planet. A large cap would have meant a large area with arctic and sub-arctic temperatures, with small temperate and tropical climate-belts. A small one like this meant wide tropical and sub-tropical zones. The fact was verified by the thick, dense cloud-masses which covered most of the surface,—all the surface, in fact, outside the ice-cap. But since there were ice-caps there would be temperate regions. In short, the ice-cap proved that a man could endure the air and temperature conditions he would find. Moran observed these things from the control-room of the Nadine , then approaching the world on planetary drive. He was to be left here, with no reason ever to expect rescue. Two of the Nadine's four-man crew watched out the same ports as the planet seemed to approach. Burleigh said encouragingly; "It doesn't look too bad, Moran!" Moran disagreed, but he did not answer. He cocked an ear instead. He heard something. It was a thin, wabbling, keening whine. No natural radiation sounds like that. Moran nodded toward the all-band speaker. "Do you hear what I do?" he asked sardonically. Burleigh listened. A distinctly artificial signal came out of the speaker. It wasn't a voice-signal. It wasn't an identification beacon, such as are placed on certain worlds for the convenience of interstellar skippers who need to check their courses on extremely long runs. This was something else. Burleigh said: "Hm ... Call the others, Harper." Harper, prudently with him in the control-room, put his head into the passage leading away. He called. But Moran observed with grudging respect that he didn't give him a chance to do anything drastic. These people on the Nadine were capable. They'd managed to recapture the Nadine from him, but they were matter-of-fact about it. They didn't seem to resent what he'd tried to do, or that he'd brought them an indefinite distance in an indefinite direction from their last landing-point, and they had still to re-locate themselves. They'd been on Coryus Three and they'd gotten departure clearance from its space-port. With clearance-papers in order, they could land unquestioned at any other space-port and take off again—provided the other space-port was one they had clearance for. Without rigid control of space-travel, any criminal anywhere could escape the consequences of any crime simply by buying a ticket to another world. Moran couldn't have bought a ticket, but he'd tried to get off the planet Coryus on the Nadine . The trouble was that the Nadine had clearance papers covering five persons aboard—four men and a girl Carol. Moran made six. Wherever the yacht landed, such a disparity between its documents and its crew would spark an investigation. A lengthy, incredibly minute investigation. Moran, at least, would be picked out as a fugitive from Coryus Three. The others were fugitives too, from some unnamed world Moran did not know. They might be sent back where they came from. In effect, with six people on board instead of five, the Nadine could not land anywhere for supplies. With five on board, as her papers declared, she could. And Moran was the extra man whose presence would rouse space-port officials' suspicion of the rest. So he had to be dumped. He couldn't blame them. He'd made another difficulty, too. Blaster in hand, he'd made the Nadine take off from Coryus III with a trip-tape picked at random for guidance. But the trip-tape had been computed for another starting-point, and when the yacht came out of overdrive it was because the drive had been dismantled in the engine-room. So the ship's location was in doubt. It could have travelled at almost any speed in practically any direction for a length of time that was at least indefinite. A liner could re-locate itself without trouble. It had elaborate observational equipment and tri-di star-charts. But smaller craft had to depend on the Galactic Directory. The process would be to find a planet and check its climate and relationship to other planets, and its flora and fauna against descriptions in the Directory. That was the way to find out where one was, when one's position became doubtful. The Nadine needed to make a planet-fall for this. The rest of the ship's company came into the control-room. Burleigh waved his hand at the speaker. "Listen!" They heard it. All of them. It was a trilling, whining sound among the innumerable random noises to be heard in supposedly empty space. "That's a marker," Carol announced. "I saw a costume-story tape once that had that sound in it. It marked a first-landing spot on some planet or other, so the people could find that spot again. It was supposed to be a long time ago, though." "It's weak," observed Burleigh. "We'll try answering it." Moran stirred, and he knew that every one of the others was conscious of the movement. But they didn't watch him suspiciously. They were alert by long habit. Burleigh said they'd been Underground people, fighting the government of their native world, and they'd gotten away to make it seem the revolt had collapsed. They'd go back later when they weren't expected, and start it up again. Moran considered the story probable. Only people accustomed to desperate actions would have remained so calm when Moran had used desperate measures against them. Burleigh picked up the transmitter-microphone. "Calling ground," he said briskly. "Calling ground! We pick up your signal. Please reply." He repeated the call, over and over and over. There was no answer. Cracklings and hissings came out of the speaker as before, and the thin and reedy wabbling whine continued. The Nadine went on toward the enlarging cloudy mass ahead. Burleigh said; "Well?" "I think," said Carol, "that we should land. People have been here. If they left a beacon, they may have left an identification of the planet. Then we'd know where we are and how to get to Loris." Burleigh nodded. The Nadine had cleared for Loris. That was where it should make its next landing. The little yacht went on. All five of its proper company watched as the planet's surface enlarged. The ice-cap went out of sight around the bulge of the globe, but no markings appeared. There were cloud-banks everywhere, probably low down in the atmosphere. The darker vague areas previously seen might have been highlands. "I think," said Carol, to Moran, "that if it's too tropical where this signal's coming from, we'll take you somewhere near enough to the ice-cap to have an endurable climate. I've been figuring on food, too. That will depend on where we are from Loris because we have to keep enough for ourselves. But we can spare some. We'll give you the emergency-kit, anyhow." The emergency-kit contained antiseptics, seeds, and a weapon or two, with elaborate advice to castaways. If somebody were wrecked on an even possibly habitable planet, the especially developed seed-strains would provide food in a minimum of time. It was not an encouraging thought, though, and Moran grimaced. She hadn't said anything about being sorry that he had to be marooned. Maybe she was, but rebels learn to be practical or they don't live long. Moran wondered, momentarily, what sort of world they came from and why they had revolted, and what sort of set-back to the revolt had sent the five off in what they considered a strategic retreat but their government would think defeat. Moran's own situation was perfectly clear. He'd killed a man on Coryus III. His victim would not be mourned by anybody, and somebody formerly in very great danger would now be safe, which was the reason for what Moran had done. But the dead man had been very important, and the fact that Moran had forced him to fight and killed him in fair combat made no difference. Moran had needed to get off-planet, and fast. But space-travel regulations are especially designed to prevent such escapes. He'd made a pretty good try, at that. One of the controls on space-traffic required a ship on landing to deposit its fuel-block in the space-port's vaults. The fuel-block was not returned until clearance for departure had been granted. But Moran had waylaid the messenger carrying the Nadine's fuel-block back to that space-yacht. He'd knocked the messenger cold and presented himself at the yacht with the fuel. He was admitted. He put the block in the engine's gate. He duly took the plastic receipt-token the engine only then released, and he drew a blaster. He'd locked two of the Nadine's crew in the engine-room, rushed to the control-room without encountering the others, dogged the door shut, and threaded in the first trip-tape to come to hand. He punched the take-off button and only seconds later the overdrive. Then the yacht—and Moran—was away. But his present companions got the drive dismantled two days later and once the yacht was out of overdrive they efficiently gave him his choice of surrendering or else. He surrendered, stipulating that he wouldn't be landed back on Coryus; he still clung to hope of avoiding return—which was almost certain anyhow. Because nobody would want to go back to a planet from which they'd carried away a criminal, even though they'd done it unwillingly. Investigation of such a matter might last for months. Now the space-yacht moved toward a vast mass of fleecy whiteness without any visible features. Harper stayed with the direction-finder. From time to time he gave readings requiring minute changes of course. The wabbling, whining signal was louder now. It became louder than all the rest of the space-noises together. The yacht touched atmosphere and Burleigh said; "Watch our height, Carol." She stood by the echometer. Sixty miles. Fifty. Thirty. A correction of course. Fifteen miles to surface below. Ten. Five. At twenty-five thousand feet there were clouds, which would be particles of ice so small that they floated even so high. Then clear air, then lower clouds, and lower ones still. It was not until six thousand feet above the surface that the planet-wide cloud-level seemed to begin. From there on down it was pure opacity. Anything could exist in that dense, almost palpable grayness. There could be jagged peaks. The Nadine went down and down. At fifteen hundred feet above the unseen surface, the clouds ended. Below, there was only haze. One could see the ground, at least, but there was no horizon. There was only an end to visibility. The yacht descended as if in the center of a sphere in which one could see clearly nearby, less clearly at a little distance, and not at all beyond a quarter-mile or so. There was a shaded, shadowless twilight under the cloud-bank. The ground looked like no ground ever seen before by anyone. Off to the right a rivulet ran between improbable-seeming banks. There were a few very small hills of most unlikely appearance. It was the ground, the matter on which one would walk, which was strangest. It had color, but the color was not green. Much of it was a pallid, dirty-yellowish white. But there were patches of blue, and curious veinings of black, and here and there were other colors, all of them unlike the normal color of vegetation on a planet with a sol-type sun. Harper spoke from the direction-finder; "The signal's coming from that mound, yonder." There was a hillock of elongated shape directly in line with the Nadine's course in descent. Except for the patches of color, it was the only considerable landmark within the half-mile circle in which anything could be seen at all. The Nadine checked her downward motion. Interplanetary drive is rugged and sure, but it does not respond to fine adjustment. Burleigh used rockets, issuing great bellowings of flame, to make actual contact. The yacht hovered, and as the rocket-flames diminished slowly she sat down with practically no impact at all. But around her there was a monstrous tumult of smoke and steam. When the rockets went off, she lay in a burned-out hollow some three or four feet deep with a bottom of solid stone. The walls of the hollow were black and scorched. It seemed that at some places they quivered persistently. There was silence in the control-room save for the whining noise which now was almost deafening. Harper snapped off the switch. Then there was true silence. The space-yacht had come to rest possibly a hundred yards from the mound which was the source of the space-signal. That mound shared the peculiarity of the ground as far as they could see through the haze. It was not vegetation in any ordinary sense. Certainly it was no mineral surface! The landing-pockets had burned away three or four feet of it, and the edge of the burned area smoked noisesomely, and somehow it looked as if it would reek. And there were places where it stirred. Burleigh blinked and stared. Then he reached up and flicked on the outside microphones. Instantly there was bedlam. If the landscape was strange, here, the sounds that came from it were unbelievable. There were grunting noises. There were clickings, uncountable clickings that made a background for all the rest. There were discordant howls and honkings. From time to time some thing unknown made a cry that sounded very much like a small boy trailing a stick against a picket fence, only much louder. Something hooted, maintaining the noise for an impossibly long time. And persistently, sounding as if they came from far away, there were booming noises, unspeakably deep-bass, made by something alive. And something shrieked in lunatic fashion and something else still moaned from time to time with the volume of a steam-whistle.... "This sounds and looks like a nice place to live," said Moran with fine irony. Burleigh did not answer. He turned down the outside sound. "What's that stuff there, the ground?" he demanded. "We burned it away in landing. I've seen something like it somewhere, but never taking the place of grass!" "That," said Moran as if brightly, "that's what I'm to make a garden in. Of evenings I'll stroll among my thrifty plantings and listen to the delightful sounds of nature." Burleigh scowled. Harper flicked off the direction-finder. "The signal still comes from that hillock yonder," he said with finality. Moran said bitingly; "That ain't no hillock, that's my home!" Then, instantly he'd said it, he recognized that it could be true. The mound was not a fold in the ground. It was not an up-cropping of the ash-covered stone on which the Nadine rested. The enigmatic, dirty-yellow-dirty-red-dirty-blue-and-dirty-black ground-cover hid something. It blurred the shape it covered, very much as enormous cobwebs made solid and opaque would have done. But when one looked carefully at the mound, there was a landing-fin sticking up toward the leaden skies. It was attached to a large cylindrical object of which the fore part was crushed in. The other landing-fins could be traced. "It's a ship," said Moran curtly. "It crash-landed and its crew set up a signal to call for help. None came, or they'd have turned the beacon off. Maybe they got the lifeboats to work and got away. Maybe they lived as I'm expected to live until they died as I'm expected to die." Burleigh said angrily; "You'd do what we are doing if you were in our shoes!" "Sure," said Moran, "but a man can gripe, can't he?" "You won't have to live here," said Burleigh. "We'll take you somewhere up by the ice-cap. As Carol said, we'll give you everything we can spare. And meanwhile we'll take a look at that wreck yonder. There might be an indication in it of what solar system this is. There could be something in it of use to you, too. You'd better come along when we explore." "Aye, aye, sir," said Moran with irony. "Very kind of you, sir. You'll go armed, sir?" Burleigh growled; "Naturally!" "Then since I can't be trusted with a weapon," said Moran, "I suggest that I take a torch. We may have to burn through that loathesome stuff to get in the ship." "Right," growled Burleigh again. "Brawn and Carol, you'll keep ship. The rest of us wear suits. We don't know what that stuff is outside." Moran silently went to the space-suit rack and began to get into a suit. Modern space-suits weren't like the ancient crudities with bulging metal casings and enormous globular helmets. Non-stretch fabrics took the place of metal, and constant-volume joints were really practical nowadays. A man could move about in a late-model space-suit almost as easily as in ship-clothing. The others of the landing-party donned their special garments with the brisk absence of fumbling that these people displayed in every action. "If there's a lifeboat left," said Carol suddenly, "Moran might be able to do something with it." "Ah, yes!" said Moran. "It's very likely that the ship hit hard enough to kill everybody aboard, but not smash the boats!" "Somebody survived the crash," said Burleigh, "because they set up a beacon. I wouldn't count on a boat, Moran." "I don't!" snapped Moran. He flipped the fastener of his suit. He felt all the openings catch. He saw the others complete their equipment. They took arms. So far they had seen no moving thing outside, but arms were simple sanity on an unknown world. Moran, though, would not be permitted a weapon. He picked up a torch. They filed into the airlock. The inner door closed. The outer door opened. It was not necessary to check the air specifically. The suits would take care of that. Anyhow the ice-cap said there were no water-soluble gases in the atmosphere, and a gas can't be an active poison if it can't dissolve. They filed out of the airlock. They stood on ash-covered stone, only slightly eroded by the processes which made life possible on this planet. They looked dubiously at the scorched, indefinite substance which had been ground before the Nadine landed. Moran moved scornfully forward. He kicked at the burnt stuff. His foot went through the char. The hole exposed a cheesy mass of soft matter which seemed riddled with small holes. Something black came squirming frantically out of one of the openings. It was eight or ten inches long. It had a head, a thorax, and an abdomen. It had wing-cases. It had six legs. It toppled down to the stone on which the Nadine rested. Agitatedly, it spread its wing-covers and flew away, droning loudly. The four men heard the sound above even the monstrous cacophony of cries and boomings and grunts and squeaks which seemed to fill the air. "What the devil—." Moran kicked again. More holes. More openings. More small tunnels in the cheese-like, curd-like stuff. More black things squirming to view in obvious panic. They popped out everywhere. It was suddenly apparent that the top of the soil, here, was a thick and blanket-like sheet over the whitish stuff. The black creatures lived and thrived in tunnels under it. Carol's voice came over the helmet-phones. " They're—bugs! " she said incredulously. " They're beetles! They're twenty times the size of the beetles we humans have been carrying around the galaxy, but that's what they are! " Moran grunted. Distastefully, he saw his predicament made worse. He knew what had happened here. He could begin to guess at other things to be discovered. It had not been practical for men to move onto new planets and subsist upon the flora and fauna they found there. On some new planets life had never gotten started. On such worlds a highly complex operation was necessary before humanity could move in. A complete ecological complex had to be built up; microbes to break down the rock for soil, bacteria to fix nitrogen to make the soil fertile; plants to grow in the new-made dirt and insects to fertilize the plants so they would multiply, and animals and birds to carry the seeds planet-wide. On most planets, to be sure, there were local, aboriginal plants and animals. But still terrestrial creatures had to be introduced if a colony was to feed itself. Alien plants did not supply satisfactory food. So an elaborate adaptation job had to be done on every planet before native and terrestrial living things settled down together. It wasn't impossible that the scuttling things were truly beetles, grown large and monstrous under the conditions of a new planet. And the ground.... "This ground stuff," said Moran distastefully, "is yeast or some sort of toadstool growth. This is a seedling world. It didn't have any life on it, so somebody dumped germs and spores and bugs to make it ready for plants and animals eventually. But nobody's come back to finish up the job." Burleigh grunted a somehow surprised assent. But it wasn't surprising; not wholly so. Once one mentioned yeasts and toadstools and fungi generally, the weird landscape became less than incredible. But it remained actively unpleasant to think of being marooned on it. "Suppose we go look at the ship?" said Moran unpleasantly. "Maybe you can find out where you are, and I can find out what's ahead of me." He climbed up on the unscorched surface. It was elastic. The parchment-like top skin yielded. It was like walking on a mass of springs. "We'd better spread out," added Moran, "or else we'll break through that skin and be floundering in this mess." "I'm giving the orders, Moran!" said Burleigh shortly. "But what you say does make sense." He and the others joined Moran on the yielding surface. Their footing was uncertain, as on a trampoline. They staggered. They moved toward the hillock which was a covered-over wrecked ship. The ground was not as level as it appeared from the Nadine's control-room. There were undulations. But they could not see more than a quarter-mile in any direction. Beyond that was mist. But Burleigh, at one end of the uneven line of advancing men, suddenly halted and stood staring down at something he had not seen before. The others halted. Something moved. It came out from behind a very minor spire of whitish stuff that looked like a dirty sheet stretched over a tall stone. The thing that appeared was very peculiar indeed. It was a—worm. But it was a foot thick and ten feet long, and it had a group of stumpy legs at its fore end—where there were eyes hidden behind bristling hair-like growths—and another set of feet at its tail end. It progressed sedately by reaching forward with its fore-part, securing a foothold, and then arching its middle portion like a cat arching its back, to bring its hind part forward. Then it reached forward again. It was of a dark olive color from one end to the other. Its manner of walking was insane but somehow sedate. Moran heard muffled noises in his helmet-phone as the others tried to speak. Carol's voice came anxiously; " What's the matter? What do you see? " Moran said with savage precision; "We're looking at an inch-worm, grown up like the beetles only more so. It's not an inch-worm any longer. It's a yard-worm." Then he said harshly to the men with him; "It's not a hunting creature on worlds where it's smaller. It's not likely to have turned deadly here. Come on!" He went forward over the singularly bouncy ground. The others followed. It was to be noted that Hallet the engineer, avoided the huge harmless creature more widely than most. They reached the mound which was the ship. Moran unlimbered his torch. He said sardonically; "This ship won't do anybody any good. It's old-style. That thick belt around its middle was dropped a hundred years ago, and more." There was an abrupt thickening of the cylindrical hull at the middle. There was an equally abrupt thinning, again, toward the landing-fins. The sharpness of the change was blurred over by the revolting ground-stuff growing everywhere. "We're going to find that this wreck has been here a century at least!" Without orders, he turned on the torch. A four-foot flame of pure blue-white leaped out. He touched its tip to the fungoid soil. Steam leaped up. He used the flame like a gigantic scalpel, cutting a square a yard deep in the whitish stuff, and then cutting it across and across to destroy it. Thick fumes arose, and quiverings and shakings began. Black creatures in their labyrinths of tunnels began to panic. Off to the right the blanket-like surface ripped and they poured out. They scuttled crazily here and there. Some took to wing. By instinct the other men—the armed ones—moved back from the smoke. They wore space-helmets but they felt that there should be an intolerable smell. Moran slashed and slashed angrily with the big flame, cutting a way to the metal hull that had fallen here before his grandfather was born. Sometimes the flame cut across things that writhed, and he was sickened. But above all he raged because he was to be marooned here. He could not altogether blame the others. They couldn't land at any colonized world with him on board without his being detected as an extra member of the crew. His fate would then be sealed. But they also would be investigated. Official queries would go across this whole sector of the galaxy, naming five persons of such-and-such description and such-and-such fingerprints, voyaging in a space-yacht of such-and-such size and registration. The world they came from would claim them as fugitives. They would be returned to it. They'd be executed. Then Carol's voice came in his helmet-phone. She cried out; " Look out! It's coming! Kill it! Kill it—. " He heard blast-rifles firing. He heard Burleigh pant commands. He was on his way out of the hollow he'd carved when he heard Harper cry out horribly. He got clear of the newly burned-away stuff. There was still much smoke and stream. But he saw Harper. More, he saw the thing that had Harper. It occurred to him instantly that if Harper died, there would not be too many people on the Nadine . They need not maroon him. In fact, they wouldn't dare. A ship that came in to port with two few on board would be investigated as thoroughly as one that had too many. Perhaps more thoroughly. So if Harper were killed, Moran would be needed to take his place. He'd go on from here in the Nadine , necessarily accepted as a member of her crew. Then he rushed, the flame-torch making a roaring sound. II. They went back to the Nadine for weapons more adequate for encountering the local fauna when it was over. Blast-rifles were not effective against such creatures as these. Torches were contact weapons but they killed. Blast-rifles did not. And Harper needed to pull himself together again, too. Also, neither Moran nor any of the others wanted to go back to the still un-entered wreck while the skinny, somehow disgusting legs of the thing still kicked spasmodically—quite separate—on the whitish ground-stuff. Moran had disliked such creatures in miniature form on other worlds. Enlarged like this. It seemed insane that such creatures, even in miniature, should painstakingly be brought across light-years of space to the new worlds men settled on. But it had been found to be necessary. The ecological system in which human beings belonged had turned out to be infinitely complicated. It had turned out, in fact, to be the ecological system of Earth, and unless all parts of the complex were present, the total was subtly or glaringly wrong. So mankind distastefully ferried pests as well as useful creatures to its new worlds as they were made ready for settlement. Mosquitos throve on the inhabited globes of the Rim Stars. Roaches twitched nervous antennae on the settled planets of the Coal-sack. Dogs on Antares had fleas, and scratched their bites, and humanity spread through the galaxy with an attendant train of insects and annoyances. If they left their pests behind, the total system of checks and balances which make life practical would get lopsided. It would not maintain itself. The vagaries that could result were admirably illustrated in and on the landscape outside the Nadine . Something had been left out of the seeding of this planet. The element—which might be a bacterium or a virus or almost anything at all—the element that kept creatures at the size called "normal" was either missing or inoperable here. The results were not desirable.
|
C. vindictive
|
What statement best describes how the author feels about the magazine articles he discusses?
A. A lot of the advice is too specific to remember in the moment, even if it's well-intentioned
B. The articles vary in quality and usefulness by where they are published, but can have nuggets of wisdom
C. All of the advice suggested things that would kill the mood, which is counterproductive
D. The articles are only full of advice that no layperson can use, and aren't worth reading
|
More Bang for the Buck A friend of mine offers a theory about why Bill Clinton's poll numbers stayed so high throughout the Lewinsky scandal: The news made it possible for serious-minded people to spend lots of time--at the office and over lunch--talking about semen stains, vaginal insertions, and blow jobs. And the people were grateful. That's probably because they're not getting all that much themselves. A recent University of Chicago survey of 10,000 adults found that Americans are having considerably less sex than was generally thought. Only one American in 20 has sex three times a week. One in five didn't score at all last year. If that's true, many of us could use a little sexual self-improvement. Not me, of course. I have been happily married for 26 years, since the age of 21. Deb and I have what seems to us to be a perfectly fine amorous life, yet everywhere I turn the culture tells me--almost mocks me-- you can do better! What would happen to our sex life then, if Deb (who participated in this story because she loves me and because she has tenure) and I tried for the first time to make something happen to it? And so it was that we found ourselves for the first time ever in a sex-toy store, A Touch of Romance, located near our home in Los Angeles, across the street from a Macy's. The idea behind shops like these is to make obtaining the materials of sexual experimentation as ordinary as purchasing plumbing supplies or housewares. Which sort of works--the only sexual thrill I got from the visit was knowing that Microsoft just bought a cock ring. Choosing it wasn't easy. Most of them came in presized sets of three. I couldn't figure out which would fit right and intuited that try-ons weren't an option. So I opted instead for an adjustable circumference version, a little strip of vinyl with snaps for $11.95. Man, what a rip-off! Unless it works. It doesn't. Back home, I derived a certain depraved buzz in cinching the device on, but that was soon eclipsed. The thing works on the Roach Motel principle--your blood gets in but it can't get out. But then I got to thinking: Under battlefield conditions it doesn't get out anyway. And while I should have been paying more attention to other things, this led to thinking about the old joke with the punch line "... and right ball go POW." My wife hadn't noticed any difference at all. Overall rating, on a scale of 1 to 10: 2 toes curled. A woman I know says women's magazines are the best places in America to find sex tips. She's right--go ahead, just try to find a sewing pattern in Redbook . You're much more likely to land on "Try phone sex, dirty notes, porn videos, fantasy games and sex in new places. ... Try lingerie and no underwear. ... Try talking dirty and silk scarves. Try anything at all," or articles such as "Eight New Games for the Foreplay Challenged." An article in the April Cosmopolitan , "The Six Best Sex Positions," seemed more promising than the Redbook playbook. Each position was accompanied by a succinct write-up and a stick-figure diagram. The position we settled on was "The Butterfly," which we had to read three times to comprehend. The man stands, the woman remains supine on a bed or counter-top with her feet up on his shoulders. The whole idea is to produce a pelvic tilt for better access to the G spot. Instead, we experienced an uncomfortable pretzel feeling that stick figures must be immune to. And in general, Cosmopolitan 's exotic sex positions require the sort of body placement you can't remember in the moment of passion and even if you could, for proper alignment, you still might need mood-killing accessories such as a plumb line and a laser pen. Rating: 3 toes curled. Next we tried those "Better Sex" instructional videos advertised in the New York Times Book Review. I ordered Better Sexual Techniques , Advanced Sexual Techniques , Making Sex Fun , and Advanced Oral Sex Techniques (priced about $11.95 each, not including shipping and handling). My wife couldn't bear to watch them; I persevered but must admit it was a chore. The oral-sex tape starts with "well-known sex therapist" Diana Wiley, in her poofy hair and broad-shouldered blue power suit, looking like she was about to explain how the sales force could increase its third-quarter productivity. Instead she runs through all the euphemisms for oral sex and then the video cuts to XXX action with gratuitous commentary. Wiley's overexplanation of everything two people can do to each other with their mouths raises this question: Do you really need a five-minute video segment on whether or not to swallow? In the great tradition of hotel and travel ads, the guys tend to be markedly less attractive than the women. No way he'd be with her if this wasn't an instructional sex video! The inanity of the experts and the dubious casting make these films about as erotic as ... well, as the New York Times . You could learn more from any randomly selected porn video. Rating: 0 toes curled. Another approach is food. The notion that certain foods, such as oysters or rhino horn, are aphrodisiacs has been pretty much discounted. But it's plausible to think that cooking a meal together and then dining on it, just the two of you, could be erotic. Especially if (like me) your schedule frequently forces you to eat alone and you often find yourself standing in front of the microwave, screaming, "Come on, goddammit!" Intercourses , by Martha Hopkins and Randall Lockridge ($24.95, Terrace Publishing, 1997), preaches that for every time of day and every phase of a relationship there is a type of eating experience that will heighten sexual response. (There's also a chart showing which foods are good for eating off which body parts.) Deb and I blocked off a whole Saturday afternoon and evening for the Intercourses experiment, settling on rosemary-scented lamb over pasta (Page 87) followed by frozen coffee almond dessert (Page 31). According to the book, rosemary is sexy because of its fragrance (used in many perfumes) and because of its texture, which, so the text assured, tickles nerve endings. The dessert was mostly coffee, rum, and Kahlua, which has worked before. We shopped for the food together and cooked together, drinking wine and beer along the way. At one point while I was working on the dessert, I asked my wife how long to beat the heavy cream mixture. "Till it's stiff--it's an aphrodisiac," she said. Preparation took less than an hour, and everything came out perfectly. Eating at our dining room table for the first time ever without guests, we were having fun by candlelight. But the mood was romantic, not erotic. Overall rating: 4 toes curled. That's when we went for the Viagra ($212.50 for 10 doses, which includes a "consultation" fee). The drug was prescribed by a doctor, whom I've never met, and ordered from a pharmacy in Miami Beach, Fla., where I've never been. I completed the transaction via the Internet after filling out a cover-their-ass questionnaire in three minutes. We each decided to take one pill, clinked our glasses, and gulped. And then what? It felt awkward sitting in our bedroom, knowing that it could take up to an hour for Viagra to "work." I suggested that we play strip poker, something I'd never done. Deb had never even played poker, so I had to explain the rules. I won in about six hands, auspiciously I thought, with three aces. But we still weren't really in the mood yet. So then I got out the other purchase I'd made at A Touch of Romance--"Dirty Dice" ($4.95). One of the two pink cubes is marked with these words instead of dots: "lips," "above waist," "ear," "breast," "below waist," and "?". The other cube is labeled "kiss," "squeeze," "lick," "blow," "suck," and "eat." We took turns throwing the dice, but the activities generated seemed forced and arbitrary. Finally, as they say at NASA, there was word from the pad that the launch sequence was initiating. It was pretty much like all other sex, except for a slight lightheadedness. Deb said she noticed a remote tingling sensation. On the plus side, there was no priapism and neither of us experienced disruption of our color vision nor a fatal heart attack, which was nice. Overall rating: 5 toes curled. St. Augustine held lust to be a fitting punishment for man's disobedience to God: the body's disobeying of the mind, the will, the spirit, and even of itself. (The paradigm of this for him is the unbidden hard-on.) Jean-Paul Sartre discovered something similar, although celebrating it rather than deploring it: Essential to the erotic is the body's defiance of design and control. (The paradigm of this for him is the jiggle.) Sartre's view yields a sort of sexual Heisenberg principle: There is an inherent tension between physically abandoning yourself to another on the one hand and sexual planning on the other. The more of the one, the less of the other. And this, I discovered, is the chief obstacle to sexual self-help. Getting an erection is sexy. Making one is not. As my wife said about Viagra, "You start to have a new feeling and then you realize where it came from and then you don't have it so much. ... Anything that makes you think about it like that is just creepy." This is not to say there isn't a way out of this conflict between desire and design. With homage to our potent POTUS, there is, I think, a Third Way that's neither sexual complacency nor standard self-help. If the intrusion of consciousness is the problem, then maybe the answer is to block it out. Sure, you could do this the old-fashioned way: with alcohol and drugs. But then you have all the traditional drawbacks, including diminished physical attractiveness and degraded sexual performance. So how about this instead? Go for all the sexual self-help you can, but do it covertly . Watch a sex video (or porn flick) if you want--but by yourself, and then try to share what you learned without sharing how you learned it. Don't tell your partner you took Viagra. Or give each other standing permission to slip it into the odd after-dinner drink, saying nothing. (Of course, when you do it you'll still know, but having an unselfconsciously turned-on partner is a real compensation for that, and next time, your partner can surprise you. And yes, this requires trust. But why would you be having sex with someone you don't trust?) My main conclusion is that contrary to our blabby culture, the key to a better sex life is less communication.
|
A. A lot of the advice is too specific to remember in the moment, even if it's well-intentioned
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.