paperhash
stringlengths 40
40
| s2_corpus_id
stringlengths 3
9
| arxiv_id
stringclasses 0
values | title
stringlengths 7
324
| abstract
stringlengths 0
7.23k
⌀ | authors
sequence | summary
stringclasses 0
values | field_of_study
sequencelengths | venue
stringlengths 15
253
⌀ | publication_date
stringdate 1952-06-01 00:00:00
2019-07-01 00:00:00
| n_references
int32 0
4.92k
⌀ | n_citations
int32 0
84.2k
⌀ | n_influential_citations
int32 | introduction
stringlengths 15
173k
⌀ | background
stringlengths 2
115k
⌀ | methodology
stringlengths 40
140k
⌀ | experiments_results
stringlengths 1
142k
⌀ | conclusion
stringlengths 7
38k
⌀ | full_text
stringlengths 29
195k
| decision
bool 0
classes | decision_text
stringclasses 0
values | reviews
sequence | comments
sequence | references
sequence | hypothesis
stringlengths 105
1.27k
⌀ | month_since_publication
int32 67
872
| avg_citations_per_month
float32 0
1.24k
⌀ | mean_score
float32 | mean_confidence
float32 | mean_novelty
float32 | mean_correctness
float32 | mean_clarity
float32 | mean_impact
float32 | mean_reproducibility
float32 | openreview_submission_id
stringclasses 0
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
23355ddfdcddf4691139a7fb3aa6c841dd44a7e4 | 59822048 | null | A General System for Semantic Analysis of {E}nglish and its Use in Drawing Maps from Directions | We describe a semantic processor we are constructing which is i n t e n d e d to be of general applicability. It is designed around semantic operations which work on a s t r u c t u r e d data base of world knowledge to draw the appropriate i n f e r e n c e s and to identify the same entities i n d i f f e r e n t parts of t h e t e x t . The semantic operations capitalize on the high degree of redundancy e x h i b i t e d by all texts. Described are the operations for interpreting higher predicates, f o r d e t e c t i n g some intersententialqrelations, and in particular detail, for f i n d i n g t h e a n t e c e 6 e n t s of definite noun phrases. The processor is applied to the problem of drawing maps from d i r e c t i o n s . We describe a l a t t i c e -l i k e representation intermediate between the linguistic representation of directions and the visual representation of maps. OVERVIEW 1,2 We are trying to c o n s t r u c t a semantic processor of some 7 A This research was supported by the Research Foundation of the City University of New York under F a c u l t y G r a n t No. 11233. The author would like to express h i s indebtedness to H a r r y Elam f o r many insights i n t o the problems discussed here. | {
"name": [
"Hobbs, Jerry R."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 3 | 4 | null | null | generality. We are using as our data base a set of f a c t s involvi n g s p a t i a l terms i n English. To test t h e processor and to s t u d y the interfacing of semantic and task components, we are building a system which takes as i n p u t directions in E n g l i s h of how to get from one place to another and outputs a map, a map such as one might sketch for an unfamiliar region, hearing the directions over the phone.A typical input might be the text "Upon leaving thi,s building, turn right and follow Washington Street three blocks. Make a left, The l i b r a r y is an t h e r i g h t side of the s t r e e t before the next coxner." which refer to t h e same e p t i t y . The text, augmented and i n t e rrelated in t h i s way, is then passed over to the task component, which makes arbitrary decisions when the map requires information not given by the directions and produces the map.The kwp problems of semantic analysis are to f i n d , o u t of a p o t e n t i a l l y enormous collection of inferences, the appropriate i n f e r e n c e s , and t o f i n d them q u i c k l y . Our s o l u t i o n t o t h e first is i n our semantic o p e r a t i o n s described below. Our approach t o the second problem is in the organization of the data base.The d a t a i n the semantic coptponent is of two sorts:1. The Text: the information which is explicitly in t h e t e x t , I n the course o f semantic processing t h i s is augmented by i n f o r m a t i o n which is o n l y implicit i n the text. The text consists of the set of entities X1,X2, ..., e x p l i c i t l y and i m p l i c i t l y referred to in the text, and s t r u c t u r e s of $he form p (X1,X2) representing the statements m#de or implied about t h e s e e n t i t i e s , e . g . walk (XI) = X1 walks, building (XZ) = X is a building, 2 door ( X 3 , X2) = X is a &or of X2. | null | The World Knowledge or the Lexicon: the system's knowledge of words and the world. Words are the boundary between the Text and the LexPcon. A word is viewed a s a key indexing a l a r g e body of facts (Holzman, 1 9 7 1 ) . occurs i n t h e Text and the semantic operations determine a particular inference appropriate, its enabling conditions are checked. If they hold, t h e conclusions are instantiated by c r e a t i n g a copy of them in t h e Text with the lexical variables r e p l a c e d by Text entities.Clusters. One way td state the "frames" problem (Minsky 1974) is "How should the data base be organized to guide, confine, and make e f f i c i e n t t h e searches which the semantic o p e r a t i o n s require?" W e approach this by dividing the sets of inferences i n t o clusters according to topic and salience in the particular application. In the searches, the clusters are probed in order of their salience. In our application, the top-level cluster concerns the one-dimensional aspects of objects and actions. For example, the fact about a block that it is the distance between two intersections i s in the cluster. If "around the block" is encountered, less salient clusters will have to be accessed to f i n d i n f o r m a t i o ,~ about the two-dimensional nature of blocks, The mast important fact about an apartment building is that it is a building, to be represented by a square on the map. But if the d i r e c t i o n s take us inside the building, up the elevator, and along the hallway, the cluster of facts about the interiors of buildings must be accessed, A self-organizing list (Knath 1973) of the clusters is maintained--when a fact in a cluster i s used, it becqmes t h e toplevel cluster--on the ,assumption that t h e t e x t will continue to talk about the same thing.The ''<Truth S t a t u s " of Inferences. In natural language, unlike mathematics, one is n o t always free to draw c e r t a i n inferehces. We t a g our i n f e r e n c e s always, normally, o r sometimes. These notions are d e f i n e d o p e r a t i o n a l l y . An a l w a y s i n f e r e n c e i s one we are always f r e e t o draw, such as that a street i s a p a t h through space. A n o r m a l l y i n f e r e n c e i s o n e w e c a n draw if it is I n any p a r t i c u l a r t y p e of t e x t there are scales o r t r a n s i t i v e relations which are important enough t o deserve a more economical r e p r e d e n t a t i o n than predicate n o t a t i o n . I n this particulak task, the i m p o r t a n t scales are a distance scale, a s u b s c a l e of t h b i s indicating t h e p a t h "you" $ill travel, and a scale representing angular orientation. This is the principal information used in constructing the map. For these scales w e t r a n s l a t e i n t o a directed graph o r l a t t i c e -l i k e representation (Hobbs 1 9 7 4 ) . and ' ' p l e a s a n t " a11 a p p l y t o "walk", b u t t h e y narrow i n on d i f f e re n t aspects o f walking. T h a t i s , each demands t h a t a d i f f e r e n t inference be drawn from t h e s t a t e m e n t t h a t "X walks". "Out" and "slow" demand t h e i r arguments be motion from one place t oanother., f o r c i n g us t o infe'r f r o m " X walks'' t h a t "X goes from A "Pleasant", on the other hand, r e q u i r e s i t s argument t o be an awareness, so we must i n f e r from "X walks" t h a t "X engages i n a However this r e q u i r e s that w e t a k e very seriously m y suggestion in Hobbs (1974) t h a t the lexicon for the entire language be built, insofar as possible, along the lines of a s p a t i a l metaphor. We have n o t yet had to f a c e these problems since our only scales are p h y s i c a l -our " a t " and "on" are the locative " a t " and "on". "The walk was t i r i n g " . Here we look back for a statement whose predicate is "walk" or from which a statement involving "walkn can be i n f e r r e d . There a r e cases in which the required i n f e r e n c e is in f a c t a summary o f an entire paragraph--e.g."These actions surprised. , . "--although of course we cannot handle these cases.for consistency. Suppose X1 is the definite entity which prompted the search and its properties are and X2 is the proposed antecedent with propertiesWe must cycle through the q ' s and the r ' s to ensure they are consistent properties. Of course, to prove t w o properties q(X) and r(X) inconsistent can be an indefinitely long process with no assurance of termination. One admittedly ad hoc way we get around this is by placing into a special c l u s t e r those f a c t s we f e e l are likely to lead quickly to a contradiction. The second tool we use f o r deriving inconsistencies may t u r n out to be q u i t e significant.In the course of processing, the lattice described abave is constructed for several predicates. They c o n t a i n i n f o r m a t i o n which can be useful i n deriving a n inconsistency. Suppose we have a t e x t in which "the block" occurs explicitly several times.Toward the end of it, we encounter The search algorithm looks first for explicit mentions of "blockl" and finds them. Yet none of these entities is the one we want.Intuitively, the reason we know this is our almost visual feeling that we are already beyond those points. To do this, we appeal to the Principle of Knitting again and make the choice that will maximize the redundancy in the simplest begin with a g e n e r a l grammar and specialize it, by weeding out the rules for constructions that don't occur in the texts one is dealing with, and by adding a few rules f o r constructions and constraints peculiar to orre's application.We are trying to make a similar facility available for the most common kinds of semantic processing. Specializing the general semantic component would consist of several relatively easy steps. First the Lexicon would be organized into a cluster structure appropriate to the task. At worst, this would mean specifying the necessary knowledge in a fairly simple format.If a very large Lexicon were available, this could mean no moret h e p r o p e r t i e s p1 (X) , p Z (X), ..., are known about the d e f i n i t e entity X, the definitions o f p1,p2, ..., are probed f o r the f a c t that the entity does not normally occur in the plural. Included under this heading are proper names beginning with "the", like | null | Main paper:
.:
The World Knowledge or the Lexicon: the system's knowledge of words and the world. Words are the boundary between the Text and the LexPcon. A word is viewed a s a key indexing a l a r g e body of facts (Holzman, 1 9 7 1 ) . occurs i n t h e Text and the semantic operations determine a particular inference appropriate, its enabling conditions are checked. If they hold, t h e conclusions are instantiated by c r e a t i n g a copy of them in t h e Text with the lexical variables r e p l a c e d by Text entities.Clusters. One way td state the "frames" problem (Minsky 1974) is "How should the data base be organized to guide, confine, and make e f f i c i e n t t h e searches which the semantic o p e r a t i o n s require?" W e approach this by dividing the sets of inferences i n t o clusters according to topic and salience in the particular application. In the searches, the clusters are probed in order of their salience. In our application, the top-level cluster concerns the one-dimensional aspects of objects and actions. For example, the fact about a block that it is the distance between two intersections i s in the cluster. If "around the block" is encountered, less salient clusters will have to be accessed to f i n d i n f o r m a t i o ,~ about the two-dimensional nature of blocks, The mast important fact about an apartment building is that it is a building, to be represented by a square on the map. But if the d i r e c t i o n s take us inside the building, up the elevator, and along the hallway, the cluster of facts about the interiors of buildings must be accessed, A self-organizing list (Knath 1973) of the clusters is maintained--when a fact in a cluster i s used, it becqmes t h e toplevel cluster--on the ,assumption that t h e t e x t will continue to talk about the same thing.The ''<Truth S t a t u s " of Inferences. In natural language, unlike mathematics, one is n o t always free to draw c e r t a i n inferehces. We t a g our i n f e r e n c e s always, normally, o r sometimes. These notions are d e f i n e d o p e r a t i o n a l l y . An a l w a y s i n f e r e n c e i s one we are always f r e e t o draw, such as that a street i s a p a t h through space. A n o r m a l l y i n f e r e n c e i s o n e w e c a n draw if it is I n any p a r t i c u l a r t y p e of t e x t there are scales o r t r a n s i t i v e relations which are important enough t o deserve a more economical r e p r e d e n t a t i o n than predicate n o t a t i o n . I n this particulak task, the i m p o r t a n t scales are a distance scale, a s u b s c a l e of t h b i s indicating t h e p a t h "you" $ill travel, and a scale representing angular orientation. This is the principal information used in constructing the map. For these scales w e t r a n s l a t e i n t o a directed graph o r l a t t i c e -l i k e representation (Hobbs 1 9 7 4 ) . and ' ' p l e a s a n t " a11 a p p l y t o "walk", b u t t h e y narrow i n on d i f f e re n t aspects o f walking. T h a t i s , each demands t h a t a d i f f e r e n t inference be drawn from t h e s t a t e m e n t t h a t "X walks". "Out" and "slow" demand t h e i r arguments be motion from one place t oanother., f o r c i n g us t o infe'r f r o m " X walks'' t h a t "X goes from A "Pleasant", on the other hand, r e q u i r e s i t s argument t o be an awareness, so we must i n f e r from "X walks" t h a t "X engages i n a However this r e q u i r e s that w e t a k e very seriously m y suggestion in Hobbs (1974) t h a t the lexicon for the entire language be built, insofar as possible, along the lines of a s p a t i a l metaphor. We have n o t yet had to f a c e these problems since our only scales are p h y s i c a l -our " a t " and "on" are the locative " a t " and "on". "The walk was t i r i n g " . Here we look back for a statement whose predicate is "walk" or from which a statement involving "walkn can be i n f e r r e d . There a r e cases in which the required i n f e r e n c e is in f a c t a summary o f an entire paragraph--e.g."These actions surprised. , . "--although of course we cannot handle these cases.for consistency. Suppose X1 is the definite entity which prompted the search and its properties are and X2 is the proposed antecedent with propertiesWe must cycle through the q ' s and the r ' s to ensure they are consistent properties. Of course, to prove t w o properties q(X) and r(X) inconsistent can be an indefinitely long process with no assurance of termination. One admittedly ad hoc way we get around this is by placing into a special c l u s t e r those f a c t s we f e e l are likely to lead quickly to a contradiction. The second tool we use f o r deriving inconsistencies may t u r n out to be q u i t e significant.In the course of processing, the lattice described abave is constructed for several predicates. They c o n t a i n i n f o r m a t i o n which can be useful i n deriving a n inconsistency. Suppose we have a t e x t in which "the block" occurs explicitly several times.Toward the end of it, we encounter The search algorithm looks first for explicit mentions of "blockl" and finds them. Yet none of these entities is the one we want.Intuitively, the reason we know this is our almost visual feeling that we are already beyond those points. To do this, we appeal to the Principle of Knitting again and make the choice that will maximize the redundancy in the simplest begin with a g e n e r a l grammar and specialize it, by weeding out the rules for constructions that don't occur in the texts one is dealing with, and by adding a few rules f o r constructions and constraints peculiar to orre's application.We are trying to make a similar facility available for the most common kinds of semantic processing. Specializing the general semantic component would consist of several relatively easy steps. First the Lexicon would be organized into a cluster structure appropriate to the task. At worst, this would mean specifying the necessary knowledge in a fairly simple format.If a very large Lexicon were available, this could mean no moret h e p r o p e r t i e s p1 (X) , p Z (X), ..., are known about the d e f i n i t e entity X, the definitions o f p1,p2, ..., are probed f o r the f a c t that the entity does not normally occur in the plural. Included under this heading are proper names beginning with "the", like
:
generality. We are using as our data base a set of f a c t s involvi n g s p a t i a l terms i n English. To test t h e processor and to s t u d y the interfacing of semantic and task components, we are building a system which takes as i n p u t directions in E n g l i s h of how to get from one place to another and outputs a map, a map such as one might sketch for an unfamiliar region, hearing the directions over the phone.A typical input might be the text "Upon leaving thi,s building, turn right and follow Washington Street three blocks. Make a left, The l i b r a r y is an t h e r i g h t side of the s t r e e t before the next coxner." which refer to t h e same e p t i t y . The text, augmented and i n t e rrelated in t h i s way, is then passed over to the task component, which makes arbitrary decisions when the map requires information not given by the directions and produces the map.The kwp problems of semantic analysis are to f i n d , o u t of a p o t e n t i a l l y enormous collection of inferences, the appropriate i n f e r e n c e s , and t o f i n d them q u i c k l y . Our s o l u t i o n t o t h e first is i n our semantic o p e r a t i o n s described below. Our approach t o the second problem is in the organization of the data base.The d a t a i n the semantic coptponent is of two sorts:1. The Text: the information which is explicitly in t h e t e x t , I n the course o f semantic processing t h i s is augmented by i n f o r m a t i o n which is o n l y implicit i n the text. The text consists of the set of entities X1,X2, ..., e x p l i c i t l y and i m p l i c i t l y referred to in the text, and s t r u c t u r e s of $he form p (X1,X2) representing the statements m#de or implied about t h e s e e n t i t i e s , e . g . walk (XI) = X1 walks, building (XZ) = X is a building, 2 door ( X 3 , X2) = X is a &or of X2.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.006768 | null | null | null | null | null | null | null | null |
b53e0b8778e549955afcb864d1c6da6b74490938 | 219300730 | null | Review: \textit{ {C}omputers in the {H}umanities}, by {J}. {L}. {M}itchell, Editor | Note: The t a b l e of contents appears i n AJCZ, Microfiche 1 4 , frames 67-69, and summaries of several contributions appear on the same fiche. REVIEWED B Y EDITH J . HOLS AND DAN BURROWS U n i v e r s i t y of ~i n n e s o t a , D u l u t h 5 5 8 1 2 Computers in the Humanities is a selecti~n of papers from 115 presented at a conference on computers and the humanities, 1973, at the University o f Minnesota in Minneapolis I t s value is both as a bank of ideas and a s a cross section o f t h e very broad area a t the confluence of those two disciplines The var i e t y of interests d i s p l a y e d here is an indication of the consiarable breadth of opportunity f o r further exploration Its e d it o r hopes it will be "appropriate f o r use in such couraes in 'Computers and the Humanities' as are now found in many major American and European universities " As an i d e a book for such a Review: Computers in the Humanities course it should serve rather w e l l Certainly i t has little competition Scholars in the humanities are only beginning t o see the computer as the greatest thing since t h e invention (discovery?) of the s t y l u s , and recent extensions i n t o languages more agreeable to the soft sciences and to the arts have made more a t t r a ct i v e a continuing growth on several fronts. Concordances and indexes are growlng in number and in informatory powers, l a r g e banks of texts are being created, and new and imaginative uses of such stores are being attempted More sophisticated methods of analysis a r e being devised, and programs are being developed which perform increasingly complex tasks Much of this work needs to be done only once, s o i n order that efforts not be duplicated it is urgent that information on work completed or i n progress be p u b l i c i z e d . The Minnesota conference and the book which has came from i t share some things t h a t have been done The collection i s a mixed bag in more than one sense. Disciplines represented include music, a r t , archaeology, literary analysis, dialectology, language history, l e x i c o g r a p h y , and Roman history Papers v a r y in l e n g t h and in readability With this example, we used a multiplicative application probabilities model which was f a r more consistent w i t h the data than a non-application model, as measured by a chi-square comparison of predicted versus observed ftequencies (b Sankoff and P Rousseau, p . 7). Review Computers in the Hutnanities lyric poetry tends toward introversion and internal movement, towards travel through Shelley' s 'caverns of t h e m i n d , ' that ' t h o u g h t can with d t f f iculty v i s i t ' ( C . M a r t i n d a l e , p 57) Automating poetry is, on the whole, a f a i r l y harmless activity (R. W. B a i l e y , p 283) And, inevitably, the papers vary in importance The only serious f a u l t we find i n the collection is that e d i t o r i a l comment Fs ingufficient This is especially true in the section, 'Art and P o e t r y , ' where titles and appropriate credits are given, one o r two completed designs shown, but no textual advice on the nature of t h e programs used We have chosen to survey some of t h e papers herein from t w o p o i n t s of view, f i r s t as a list of accomplishments, and second as a source of inspiration for new accomplishments Readers outside the field, who are aware that something is going on, b u t who are not quite sure what, w i l l find that t h e computer is being used most in doing the things it can do best, that is, indexes and concordances. But i t s use ie being extended to more complicated tasks, i t s symbology is beFng extended to new alphabets These are some of the things which are being done THE COMPUTER AS WORKHORSE The computer serves best as a workhorse, doing t h i n g s t h a t are at least tedious and ttme-consuming, sometimes impossible, for the unassisted human mind P Bratley, S Lusignan, and | {
"name": [
"Hols, Edith J. and",
"Burrows, Dan"
],
"affiliation": [
null,
null
]
} | null | null | null | 1975-11-01 | 0 | 0 | null | Computers i n the Humanities Francine O u e l l e t t e , i n "JEUDEMO a text-handling system," describe j u s t such a system, remarking that "the computer 's main contribution t o l i t e r a r y endeavour i s i n the provision of concordances, word-indexes, and r a t h e r unsophisticated statistics. I IThe t e x t processing system they describe i s JEUDEMO, designed go perform "typical jobs with the Chronicle is a continuing eff~rt which is attempting to do two things: to establish a definitive text, and to make contributions to a gramar for Old English. What makes hie paper especially interesting is the full and clear account o f the way he is going about it.The obvious and inmediate task for the humanities is the assemblage of concordances and dictionaries. Two kinds of program are presented in Mitchell's book 1) the program which simply g e t s the information out and prints on order, and 2) the program which uses the information i n some kind of analysis.It is not surprising that a concordance and a dictionary of Shakespeare would be an early choice. M. Spevack, H. J. Neuhaus, and T. Finkenstaedt describe the operation of "SHAD a Shakespeare dictionary." At the time of their writing, the dictionary was being prepared with the use of an already existing concordance | null | null | null | null | Main paper:
review:
Computers i n the Humanities Francine O u e l l e t t e , i n "JEUDEMO a text-handling system," describe j u s t such a system, remarking that "the computer 's main contribution t o l i t e r a r y endeavour i s i n the provision of concordances, word-indexes, and r a t h e r unsophisticated statistics. I IThe t e x t processing system they describe i s JEUDEMO, designed go perform "typical jobs with the Chronicle is a continuing eff~rt which is attempting to do two things: to establish a definitive text, and to make contributions to a gramar for Old English. What makes hie paper especially interesting is the full and clear account o f the way he is going about it.The obvious and inmediate task for the humanities is the assemblage of concordances and dictionaries. Two kinds of program are presented in Mitchell's book 1) the program which simply g e t s the information out and prints on order, and 2) the program which uses the information i n some kind of analysis.It is not surprising that a concordance and a dictionary of Shakespeare would be an early choice. M. Spevack, H. J. Neuhaus, and T. Finkenstaedt describe the operation of "SHAD a Shakespeare dictionary." At the time of their writing, the dictionary was being prepared with the use of an already existing concordance
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0 | null | null | null | null | null | null | null | null |
52f475ef72db17343a9a956629d56575201aba2f | 219310275 | null | Interpretation and Integration of Sentences into a {C}-Net | p r o g r a m m i n g ) i n interpreting pronominal ref er~nczr,. It is shown how this t h ~o r y a c c o u n t s f o r t h e 4 i s a m b i g u a t ion a £ p r o n o m i n a l re f s r t n c e , & t h 2 determinatiiorr of f o c u s E comment, rn' ora c o a p l ~t i l y t h a n a r ~y e x i s t i n g semantic o r s y n t a c t i c t h a o r y . | {
"name": [
"Hofmann, Th. R."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 0 | 0 | null | null | null | null | T R E E S 1..explore h e r s t h c p o ; s i b i I i t y of a u t o m a t i n g t h i s a n a l y s i s to a i d i n a u t o m a t i c t r a n s l a t i o n .T r a m l a t i o n involves a n a l y s i s of cantent, w i t h o u t w h i c h it can o n l y bc a m a t c h i n g of l c~x i c d l & syntactic s t r u c t u . r e s L e t u w n l a n g u a g~s . , A n y n o n d i n t r ! g r a t i v e semantic theory is forced to cdaim that these -.shews both refer to t h e same person, or that they are has a C-net-:3 Harp (4) 4 P. 5 T e l l (4,,6,7) 6 P. predicates on any point must be %on-contradictory, t h e use of In c o n t r a s t to this, Jqba tol$wHary rfi& susi.2 c o n i n g has a C~n a t ,\ 1 / sap' Hum / Q f / \ \. L ' .&na w | null | Main paper:
:
T R E E S 1..explore h e r s t h c p o ; s i b i I i t y of a u t o m a t i n g t h i s a n a l y s i s to a i d i n a u t o m a t i c t r a n s l a t i o n .T r a m l a t i o n involves a n a l y s i s of cantent, w i t h o u t w h i c h it can o n l y bc a m a t c h i n g of l c~x i c d l & syntactic s t r u c t u . r e s L e t u w n l a n g u a g~s . , A n y n o n d i n t r ! g r a t i v e semantic theory is forced to cdaim that these -.shews both refer to t h e same person, or that they are has a C-net-:3 Harp (4) 4 P. 5 T e l l (4,,6,7) 6 P. predicates on any point must be %on-contradictory, t h e use of In c o n t r a s t to this, Jqba tol$wHary rfi& susi.2 c o n i n g has a C~n a t ,\ 1 / sap' Hum / Q f / \ \. L ' .&na w
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0 | null | null | null | null | null | null | null | null |
6ed210b58dd8c85ca2a52eff37e547b51a7d904d | 219302451 | null | A Framework for Writing Generation Grammars for Interactive Computer Programs | T h i s paper outlines t h e stucture and operation o f the 1 ingui s t i c component from a language generation system in an interactive program. The component receives messages describing what i s t o b e s a i d f o r m u l a t e d i n t h e representation of the main prograr and produces fluent English utterances appropriate t o t h e current discourse situation. T h e component is data-directed and uses a procedural grammar, organized as a set of strategies. Interactive, speclalist prograas presently under developwent will heed to produce fluent, intentional English utterances in responce to particular, complex t i t u t l o n s . This creates a requlroaont for language generating facilities that I s not faced in transformational grarapar, rochanical translation programs, or paraphrase generating programs. As a component of rn interactive, specialist program, the production of t h e English must be driven directlr by the communicative intentions of the program and by the discourse situation, We can imagine t h a t the overall program consist, § o f a number of cooperating modulesf o r parsing and interpreting what i s said t o i t , ior solving ptoblens in its domain, for managing i t s renary, and, i n particular, f o r generating u t t e r a n c e s t o c o m ~u n i c a t e w l t h I t s users* This generation component can be p r o f i t a b l y v i e w e d as having three aspects or msub-corponentsw. 1) Situation/doaain specie1 i s t s t h a t a r e activated when t h e program recognizes what situation it i s in+ They then decide what message will be produced. They will decide what e f f e c t on t h e listener is desired, and exactly what o b j e c t s and relations a r e t o be nentioned. F o t example, an a p p o i n t ~e n t scheduling program might be told to *scm~ule a group meeting for F r i d a y w and then find t h a t a critical aentber o f the group i s uncxvailable, The situation specialists in t h e scheduling prograr a r e t h e ones t o decide whether i t is more a p p r o p r i a t e t o s i m p l y say " 1 can'tR, O F w h e t h e r t o v o l u n t c p informationw I can't; Mitch won't be back u W l t Mondayn. 2) Models of t h e audience and t h e d i s c o u r s e s i t u a t i o n to u s e i n construct2ng utterances. There m u s t be a r e c o r d o f t h e p a s t conversation to gulcfe in the selection a f p r o n o u n s , A l s o , the program must have nodels of, and heuristics about what the audience a l r e a d y knows a n d t h e r e f o r e doesn't have t o be t o l d . T h i s Informtion l a y be very specific and domain dependent. Fot exarple, In chess, one can say "the white queen could take e knightn. T h e r e is no need t o say "a black k n l p h t w , because t h i s information is supplied by inferences from what one knows about c h e s sinferences that the spcarer assures the listener shares. 3) Llnpu!rtic know1 edge about how to construct understandable utterances in the English Isnpuagc. Obviously, t h i s lnformatlon vill Include a lexicon assoclatlng objects and relations from t h e min program with rtrate&i@~ for realizing them in English | {
"name": [
"McDonald, David"
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 2 | 3 | null | null | null | null | null | choice of plan is determined by the character o f the event's "actionw.The action is " < f i t person into full schedulerw, and i t will have two relevant properties in the lexicon: "plan*, and ".appingW.klan is e i t h e r the nane of a standard p l a n t o be used; or an actual p l a n , over. The p l a n is now as below.partiallyn 4 e -l (clause trans1 particle) slats fmntfngs "8aybeW subject twinston> vg node-2 (verb-group par tic1 el slots modal "canw pre-vb-adv n i l it does nothing, nothing has y e t been printed beyond the verb group, and other heuristics will be free t o a p p l y t o choose the proper position.Since ( p e r s o n being t a l k e d about, is here equal to the student, t h e person t h e prograa is talking with, i t i s realized as the pronoun "youw and t h e particle is dlsplaccd.Going irom <31-10-75,9a~-12arn> t o w t o m~r r~~ ~o r n i n g * m y be little more t h a n table lookup by a wtfme" coBposer that hat been designed to know the formats of the time expressions inside the scheduler. p l a n n f n p , i f f o r no other reason than t h a t the M e s s a g e can o n l y be | Main paper:
the t r~n s l a t i o n prooesa:
choice of plan is determined by the character o f the event's "actionw.The action is " < f i t person into full schedulerw, and i t will have two relevant properties in the lexicon: "plan*, and ".appingW.klan is e i t h e r the nane of a standard p l a n t o be used; or an actual p l a n , over. The p l a n is now as below.partiallyn 4 e -l (clause trans1 particle) slats fmntfngs "8aybeW subject twinston> vg node-2 (verb-group par tic1 el slots modal "canw pre-vb-adv n i l it does nothing, nothing has y e t been printed beyond the verb group, and other heuristics will be free t o a p p l y t o choose the proper position.Since ( p e r s o n being t a l k e d about, is here equal to the student, t h e person t h e prograa is talking with, i t i s realized as the pronoun "youw and t h e particle is dlsplaccd.Going irom <31-10-75,9a~-12arn> t o w t o m~r r~~ ~o r n i n g * m y be little more t h a n table lookup by a wtfme" coBposer that hat been designed to know the formats of the time expressions inside the scheduler. p l a n n f n p , i f f o r no other reason than t h a t the M e s s a g e can o n l y be
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.005076 | null | null | null | null | null | null | null | null |
bb0c75f36efad322f19a90ef17a0b765af49c9c4 | 219308840 | null | Computer Generation of Sentences by Systemic Grammar | The paper describes a colnnuter m d e l oC s v s t e m i c grammar. a penerative prammar f o r natural l a n p u p p e . P propram is explained which piven t h e f e a t u r e s OF an fterr, determines t h e structure of t h a t item accordinp t o a s v s t e r i c E r a m a r s ~e c i f i e c as 8ata. The vrograr thus deaonstrates the ~r i n c i p l e s of systemic grammar, a b r i e f sursarv of the vechanicg of which is a l s o i n c l u d e d . Some imvlications of the proarm for systemic grammar i t s e l f are d i s c u s s e d . In ~articular, jt is shown t h a t ~r e v i o u s d e f i n i t i o n s of t h e o n e r a t i o n o f qtructure-huildfnp rules require modification. 1. ? n t r o d u c t i o n This paver describes a computer model of s ~s t s m i c grammar, a prammar for n a t u r a l languages developed by H a l l i d a y and colleagues at U n i v e r s i t v College, London (Halliday , 1961, 197fl). Svstemic grammar has recentlv been o f i n t e r e s t t o comoutational grammarians, arimarily as a result of the imn~essive work .of Winograd. ( 1972 ) , who develoaed a natural l a n p u a ~e understandinp system one component o f which was s t r o n g l v influenced bv t h e ~r i n c i p l e s of systemic grammar. More recently, Power (1974) has a l s o investigated how svstemic pramnar can be used t o ,analvse natural language. There have, however, been no attempts to use a computer to investi.gate systemic grammar itself. As Friedman (1971) says, in introducing her computer model of transformational grammar, adequate n a t u r a l lanquage grammars are bound to be so complex t h a t some mechanical a i d in investigating their ~r o ~e r t i e s will be mandatory. The a i m s , t h e n , of d e v e l o ~i n g a computer m o d e l of svsteric prammar are threefold. First, t h e model e n a b l e s t h e Rrammar to be t e s t e d , i . e . it enables contradictions, amhip,uities and incomnletenesses i n the grammas to be found. P e c o n d l v , t h e model enables systemic grammar i t s e l f to he imnroved, slnce t h e consequences of a d f u s t i n p parameters a n d rules can be more e a s i l v followed. And, t h i r d l y , the model serves as a demonstration of how systemic grammar 'works ' | {
"name": [
"Self, John"
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 4 | 3 | null | null | null | null | This paver describes a computer model of s~s t s m i c grammar, a prammar for n a t u r a l languages developed by H a l l i d a y and colleagues at U n i v e r s i t v College, London (Halliday , 1961, 197fl) . (1) feature-realisation rules !PROCESS ) !HEAD) !STEM) -1 !BINDER) !SUBJECT) ! FINITE -) !SUBJECT) ! PINTTE) -1 ! MOOD-FOCUS TNDEPENDENT) ( + !QUEG'PION = !BINDER) DEPENDENT -1 -1 ( + !QUESTION = !MOOD-FOCUS ) INDEPENDENT) ?ALTERNATIVE (!OUESTION = !SUBJECT) WH) ( !ALTERNATIVE = ! SUBJECT) ALTERNATIVE: ! MODAL -1 -1 -1 !GOAL) ?ACTOR (NOT ACTOR- UNSPECIFIED) ) !TRANSITIVE) + ! ATTRI BUANT = !SUBJECT) ! ATTRIBUTE) ! COPULAR) ( + TACTOR = !SUBJECT)) ! INTRANS ) (!ACTOR = !SUBJECT)) (!GOAL = !NO SEQUENCE RULE 1 (.(!HOOPFOCUS OR !BINDER) = > ( IPRE-SUBJECT -> !SUBJECT) -> !POST-SUBJECT =, !PROCESS -> !POST-VERB) 2 ( !FINITE = ( ! PRE-SUBJECT = (!MODAL -> !PASSIVE)-> !PROCESS)) (dl Compatibility rule 1 means " !POST-SUBJECT must not be conflated with ! PRE-SUBJECT"( 3 1 function-realisationrules Rule 12 meas "if an item has none of the functions !SUBJECT, !GOAL, !ATTRIBUTE or !AGENT then if it has !BINDER, it has t h e feature CONJUNCTIONf'1 (!COPULAR 2 ( ! E N 3 ( !FINITE 4 ( !INTRANS 5 ( !MODAL 6 (!PASSIVE 7 (!PROCESS 8 ( !TRANSITIVE 9 ( !AGENT 10 (!ALTERNATIVE 11 ( ! ATTRIBUTIVE 12 ( !BINDER CONDITION COPULAR-VERB ) BIN-FORM) FINITE-VERB) INTRANSITIVE-VERP) MODAL-VERB) BE ) EXICAL-VERB 1 TRANSITIVE-'VERB PREPOSITIONAL) DISJUNCTIVE) (OR ADJECTIVAL NOMINAL PREPOSITIONAL) CON JUNCTION (NOT (OR ! SUBJECT ! GOAL ! ATTRIBUTE !AGENT) 1 ) ( OR NOMINAL DEPENDENT 1 1 QUESTIONING) ( OR NOMINAL DEPENDENT ) ) *Rule 10 means "an item with feature INTMISITIVE also has one of the features ATTRIBUTIVE and NO?!-ATTRIBUTIVE , and also has the features naming its supersystems, i . e . 9, 2 3 and 1, i.e. CLAUSE and ITEM". When A , B, . . are all functions, then these appear as, e . g . Thus, the structure generated is "Which of the tents were erected?"! GOAL !POST-SUBJECT !TRANSITIVE !QUESTION ! FINITE ! EN ! IjOOD-FOCUS !PASSIVE ! P R O C E S~ ! SU3JECT"Which of the tents were | ' h e obvious canelusionthat the mechanics of svstemic grammar (as described by Hudson) are s u f f i c i e n t l y well-defined to form the basis of a computer vodelis, f o r l i n g u i s t i c descriptions , a s i g n i f i c a n t one. However, the program also demonstratks that some pule-descriptions require c l a r i f ieation . Further possible extensions to the work could involve trying t o specify a lexicon so t h a t t h e generative process ends up with a structure with words as leaves, and one could also attempt to apply the rules in reverse, i . e . To start with a s t r i n g of lords and produce a etructurdl description. Both problems are, of course, very d 3 f f i c u l t ones. | Main paper:
? n t r o d u c t i o n:
This paver describes a computer model of s~s t s m i c grammar, a prammar for n a t u r a l languages developed by H a l l i d a y and colleagues at U n i v e r s i t v College, London (Halliday , 1961, 197fl) . (1) feature-realisation rules !PROCESS ) !HEAD) !STEM) -1 !BINDER) !SUBJECT) ! FINITE -) !SUBJECT) ! PINTTE) -1 ! MOOD-FOCUS TNDEPENDENT) ( + !QUEG'PION = !BINDER) DEPENDENT -1 -1 ( + !QUESTION = !MOOD-FOCUS ) INDEPENDENT) ?ALTERNATIVE (!OUESTION = !SUBJECT) WH) ( !ALTERNATIVE = ! SUBJECT) ALTERNATIVE: ! MODAL -1 -1 -1 !GOAL) ?ACTOR (NOT ACTOR- UNSPECIFIED) ) !TRANSITIVE) + ! ATTRI BUANT = !SUBJECT) ! ATTRIBUTE) ! COPULAR) ( + TACTOR = !SUBJECT)) ! INTRANS ) (!ACTOR = !SUBJECT)) (!GOAL = !NO SEQUENCE RULE 1 (.(!HOOPFOCUS OR !BINDER) = > ( IPRE-SUBJECT -> !SUBJECT) -> !POST-SUBJECT =, !PROCESS -> !POST-VERB) 2 ( !FINITE = ( ! PRE-SUBJECT = (!MODAL -> !PASSIVE)-> !PROCESS)) (dl Compatibility rule 1 means " !POST-SUBJECT must not be conflated with ! PRE-SUBJECT"( 3 1 function-realisationrules Rule 12 meas "if an item has none of the functions !SUBJECT, !GOAL, !ATTRIBUTE or !AGENT then if it has !BINDER, it has t h e feature CONJUNCTIONf'1 (!COPULAR 2 ( ! E N 3 ( !FINITE 4 ( !INTRANS 5 ( !MODAL 6 (!PASSIVE 7 (!PROCESS 8 ( !TRANSITIVE 9 ( !AGENT 10 (!ALTERNATIVE 11 ( ! ATTRIBUTIVE 12 ( !BINDER CONDITION COPULAR-VERB ) BIN-FORM) FINITE-VERB) INTRANSITIVE-VERP) MODAL-VERB) BE ) EXICAL-VERB 1 TRANSITIVE-'VERB PREPOSITIONAL) DISJUNCTIVE) (OR ADJECTIVAL NOMINAL PREPOSITIONAL) CON JUNCTION (NOT (OR ! SUBJECT ! GOAL ! ATTRIBUTE !AGENT) 1 ) ( OR NOMINAL DEPENDENT 1 1 QUESTIONING) ( OR NOMINAL DEPENDENT ) ) *Rule 10 means "an item with feature INTMISITIVE also has one of the features ATTRIBUTIVE and NO?!-ATTRIBUTIVE , and also has the features naming its supersystems, i . e . 9, 2 3 and 1, i.e. CLAUSE and ITEM". When A , B, . . are all functions, then these appear as, e . g . Thus, the structure generated is "Which of the tents were erected?"! GOAL !POST-SUBJECT !TRANSITIVE !QUESTION ! FINITE ! EN ! IjOOD-FOCUS !PASSIVE ! P R O C E S~ ! SU3JECT"Which of the tents were
conclusion:
' h e obvious canelusionthat the mechanics of svstemic grammar (as described by Hudson) are s u f f i c i e n t l y well-defined to form the basis of a computer vodelis, f o r l i n g u i s t i c descriptions , a s i g n i f i c a n t one. However, the program also demonstratks that some pule-descriptions require c l a r i f ieation . Further possible extensions to the work could involve trying t o specify a lexicon so t h a t t h e generative process ends up with a structure with words as leaves, and one could also attempt to apply the rules in reverse, i . e . To start with a s t r i n g of lords and produce a etructurdl description. Both problems are, of course, very d 3 f f i c u l t ones.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.005076 | null | null | null | null | null | null | null | null |
94a66eadac51bb471e3d239aa90e3fddd58bed95 | 219303858 | null | {PEDAGLOT} and Understanding Natural Language Processing | PEDAGLOT is a programmable parser, a 'meta-parser.' To program i t , one describes not j u s t syntax and some semantics, but also--independently--its modes of behavior. The PEDAGLOT formulation of such modes o f behavior follows a c a t e g o r i z a t i o n of parsing processes i n t o a t t e n t i o n -c o n t r o l , discovery, pred i c t i o n and c o n s t r u c t i o n . Within t h e s e o v e r a l l types o f -a c t i v i t i e s , c o n t r o l can be s p e c i f l e d covering a number of syntax-processing and semantics-processing operations. While i t i s not t h e only p o s s i b l e way of programing a metap a r s e r , t h e PEDAGLOT mode-specification technique i s suggestive i n i t s e l f of v a r i o u s new approaches t o modeling and understanding same language processing a c t i v i t i e s besides parsing, such as generation and i n f e r e n c e , 7% i s wotk was sponsored by through NIH Grant #RR643. | {
"name": [
"Fabens, William"
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 6 | 0 | null | null | I t i s well known t h a t t o process n a t u r a l language, one needs both a syntactic description of possible sentences, blended i n some way with a semantic description b f a c e r t a i n domain of discourse, and a r a t h e r d e t a i l e d description of t h e actual processes used i n hearing o r producing sentences.An augmented t r a n s i t i o n network (Woods, 1970) i s qn example of t h e blending of s y n t a c t i c and quasi-semantic descriptions, Here r e g i s t e r s would be repositories o f , o r pointers t o , semantics. When used i n conjunction with a semantic nqtwork, an ATN can be used 60 parse o r t o generate (Simmons and Slocum, 1912) sentences. The i s s u e of changing the des'cription of t h e actual processes used i n such systems has been touched on by Woods ( i n using a 'generation modet), t o some extent by Gimmons and Slo~um (usi~g decision functions t o control s t y l e of generation), and t o a l a r g e r extent by Kaplari (19751, i n h i s General Syntactic Procdssor, GSP. GSP indeed i s one example of a system i n which syntax, semantics and t o some extent processes can each be u s e f u l l y defined.If we look at syntax, semantics and processes as t h r e e describable components, these systems j u s t mentioned i l l u s t r a t e how thoroughly intertwined they can become-t o the extent t h a t t h e o r i s t s from time t o time deny the existence o r a t l e a s t t h e importance of some one of them. Ignoring t h a t dispute, I would-like t o concentrate on the question of being a b l e t o comprehensively describe one's theory of language i n terms of its syntax, semantics and processes i n a way t h a t allows f o r t h e i r necessary and extensive intertwining connections, but a t t h e sane time allows one t o describe them independently.parser which can make grammatical relaxations i f i t i s given an Ill-formed s t r i n g , so as t o arrive a t a k l o s e s t t possible parse f o r the' s t r i n g . This probl-em involved describing a k o r r e c t t grammar and then (in some way) describing a space of deviations az that night be allowed by the paxser. Thus the syntax would be fixed and the way t h e parser uses it would separately have to be described. I t was soon noticed that efficiency could be greatly enhanced i f some rudimentary notion of semantic p l a u s i b i l i t y could also be used. I t would have t o be described i n a way related t o the cbrrect syntax but s t i l l be usable by the parser. Thus, f o r my purposes, the descriptions had t o be independent of one another.One feature of a relaxation parser i s t h a t it can ' f i l l i n the gaps' of a string t h a t is missing various words. If one could, which my relaxation parser did not, specify the semantic context of a sentence, the generated sentence might be semantically rather plausible. In any case, the relaxation parser operates i n various respects l i k e an actual parser or like a generator, and it was t h i s r e l ationship between parsing and generating that became of i n t e r e s t , Out of the design of the relaxation parser, the notation (independent of syntax) which t o some extent describes various processes and choices of alternate ways of processing was developed. Thus, one may take a s e t of syntax and semantic descriptions and then through describing the processing 'modest involved, define a processor which uses the particular algorithm t h a t the individual processes together define, One may c a l l the parser that i s programmabLe i n i t s processes a meta-parser, of which various existing q a r s e r s and generators appear t o be special cases,A closer examination of t h e parser I have developed (called PEDAGLOT*) may show some such aspects of meta-parsing, especially as regards the relationship between parsing and generating. I will describe the syntactic and semantic parts o f t h e parser first: by noting i t s resemblances to t h e parser of J . Earley (1970) and the ATN system of Woods. Then I w i l l describe t h e process-type specifications t h a t are available, and the use of meta-parsers as a basis f o r defining general language behaviors. Purther detail can be found in the PEDAGMT manual (Fabens, 1972 and The fundamental operation of t h e parser i s very s i m i l a r t o t h e operation of Earleyvs parser, with augmentations f o r recording t h e r e s u l t s of parses (e,g,, their t r e s t r u c t u r e , and various of t h e i r a t t r i b u t e s , which I c a l l ftags'). It is given a grammar a s a s e t of context-free rules with various extensions, most i m p~r t a n t of which a r e t h a t LISP functions may be used as predicates instead of terminals, and thay each rule may be followed by operations t h a t are defined i n tbnns of t h e s y n t a c t i c elements a f t h e r u l e i n question, A n example of t h i s notation i s as follows: S -+ NP VP => [AGREE [REF NP] [VB VP] ] [ S U M = [REF NP] ] [OW = [REF VP] [VB = [VB VP] ] S -* NP [BE] [VPASS] BY NP => [AGREE [REF NP] [VB [BE] ]I [SUM = [REF NP I ] ] [OBJ = REF NP] ] [VB = [VB [VPASS] ] ] NP -+ [DET] [N] => [REF = [N]] W + [VINP => [VB = [V]] [REF = [REF NP]] | One can see that, except f o r the n o t a t i o n a l i n e f f i c i e n c i e s o f t h e context, free formalism (as opposed to the augmented t r a n s i t i o n network form), t h i s parser is very much like other standard parsers (especially ATN s) . I t differs i n t h a tthere is a waytof specifying how t o proceed. Currently, this system has approximately a dozen toodesr and I will present some of them here. Each mode s p e c i f i e s how t o handle a certain part of the parsing process. They can be classified i n t o four categories: attention control, prediction, discovery and construction.Since the parser operates on a chart of independent events ('parsing questions1), one must give t h e p a r s e r a method of sequencing through them.Thus, one may specify 'breadth-first1 or 'depth-first1 and t h e appropriate ~echanism will be invoked {this merely involves t h e way the processor stacks i t s jobs). A 'best-first ' option i s -under development, which, when given an evaluation function to be applied to the set of currently a c t i v e partial parses, allows the system to operate on the 'best1 problem n e x t , Experi-Bents with this mode have so far been inconclusive. with each sub-parse that p a r t i c i p a t e in, and a r e judged1 by t h e disambiguatic;~~ routines. Similar ideas are used by Lyon (1974) .As Woods (1975) as the expansion. I n t u i t i v e l y , however, people do n o t seem t o do t h i s . Instead, as i n an A'I??, they t r y one and only i f t h a t f a i l s , go I n t o t h e n e x t .I n PEDAGLOT, t h e choice of which r u l e t o t r y can be defined a s t h e r e s u l t of the c a l l t o a 'choosef function (or it can be l e f t uncontrolled] , W e have des&ned various approaches t o such p r e d i c t i o n s (e.g., a l i m i t e d key-word scan of the incoming s t r i n g , and the use of 'language s t a t i s t i c s such as t h e s e t of rules which can generate the next symbol i n t h e s t r i n g as t h e i r l e f t most symbol). what i s going t o be happening ( i . e , , a l l t a g s of a c e r t a i n name w i l l meld together i n a c e r t a i n way, unless t h e grammar s p e c i f i e s otherwise), If one, however, wants c e r t a i n foms of local behavior, one may use predicates o r functions on individuaQ r u l e s . Further, i f one wants t o change t h e order i n which predictions a r e evaluated, one can program a tchoosel function which w i l l make t h a t global change. To a large extent, the language designer may specify mch of t h e processor in broad ternas and s t i l l be able t o control local events where necessary.In a more general sense, a meta-parser allows one t o understand and build higher order theories about how people might represent and process language.For instance, while it may be true that generating i s t h e inverse of parsing, there is more than one way t o do such i n v e r t i n g . One could start from a senantic network, using the choose function along with t h e INSERT mode t o restrict means of expression consistent with the intendea message, and using AMBIG functions to weed out a l l but reasonable messages from m n g the many the parser may produce or one might simply t a k e from the semantic network a simple string o f meaningful words, and then we a less t i g h t l y programmed 'relaxation parser' t o rearrange these words to be syntactically correct. W e are now considering using a crude 'backwardsT mode which begins with the operati~n part of a r u l e and, by using predicates ( e . g . , AGREE)to yield inverses, specifies what the context-free pattern must produce. Thus there are many variations of how t o generate using a meta-parser.In the area of language inference, t o take another example of language processing,PEDAGLOT suggests various differing ways of approaching the problem. First, ofie may use it a5 a 'relaxation-parser, the 'parse tree1 can be pattern-matched against the new sentence, and hypotheses can be famed. Or, one could place a more rudimentary inference systw on the 'prediction' part of the processor i t s e l f , and using other controls, the predictions that are successful could be rewritten as a new gramar.These two learning paradigms could each be strengthened by way of t h e use of tags t o contain (in a sense) t h e meaning of t h e sentelzces t o be learned, Each of these paradips can be modeled using a meta-parser like PEDAGLM. Thus, a meta-parser can raise [and be prepared to answer) a nlrmbor of interesting questions.i n almost any order, and not just in a backtracking sense (cf. Woods, 1975). Thb efficiency i s realized here since many 'partial parses (partially recognized forns) | null | null | Main paper:
. meta-parsing modes:
One can see that, except f o r the n o t a t i o n a l i n e f f i c i e n c i e s o f t h e context, free formalism (as opposed to the augmented t r a n s i t i o n network form), t h i s parser is very much like other standard parsers (especially ATN s) . I t differs i n t h a tthere is a waytof specifying how t o proceed. Currently, this system has approximately a dozen toodesr and I will present some of them here. Each mode s p e c i f i e s how t o handle a certain part of the parsing process. They can be classified i n t o four categories: attention control, prediction, discovery and construction.Since the parser operates on a chart of independent events ('parsing questions1), one must give t h e p a r s e r a method of sequencing through them.Thus, one may specify 'breadth-first1 or 'depth-first1 and t h e appropriate ~echanism will be invoked {this merely involves t h e way the processor stacks i t s jobs). A 'best-first ' option i s -under development, which, when given an evaluation function to be applied to the set of currently a c t i v e partial parses, allows the system to operate on the 'best1 problem n e x t , Experi-Bents with this mode have so far been inconclusive. with each sub-parse that p a r t i c i p a t e in, and a r e judged1 by t h e disambiguatic;~~ routines. Similar ideas are used by Lyon (1974) .As Woods (1975) as the expansion. I n t u i t i v e l y , however, people do n o t seem t o do t h i s . Instead, as i n an A'I??, they t r y one and only i f t h a t f a i l s , go I n t o t h e n e x t .I n PEDAGLOT, t h e choice of which r u l e t o t r y can be defined a s t h e r e s u l t of the c a l l t o a 'choosef function (or it can be l e f t uncontrolled] , W e have des&ned various approaches t o such p r e d i c t i o n s (e.g., a l i m i t e d key-word scan of the incoming s t r i n g , and the use of 'language s t a t i s t i c s such as t h e s e t of rules which can generate the next symbol i n t h e s t r i n g as t h e i r l e f t most symbol). what i s going t o be happening ( i . e , , a l l t a g s of a c e r t a i n name w i l l meld together i n a c e r t a i n way, unless t h e grammar s p e c i f i e s otherwise), If one, however, wants c e r t a i n foms of local behavior, one may use predicates o r functions on individuaQ r u l e s . Further, i f one wants t o change t h e order i n which predictions a r e evaluated, one can program a tchoosel function which w i l l make t h a t global change. To a large extent, the language designer may specify mch of t h e processor in broad ternas and s t i l l be able t o control local events where necessary.In a more general sense, a meta-parser allows one t o understand and build higher order theories about how people might represent and process language.For instance, while it may be true that generating i s t h e inverse of parsing, there is more than one way t o do such i n v e r t i n g . One could start from a senantic network, using the choose function along with t h e INSERT mode t o restrict means of expression consistent with the intendea message, and using AMBIG functions to weed out a l l but reasonable messages from m n g the many the parser may produce or one might simply t a k e from the semantic network a simple string o f meaningful words, and then we a less t i g h t l y programmed 'relaxation parser' t o rearrange these words to be syntactically correct. W e are now considering using a crude 'backwardsT mode which begins with the operati~n part of a r u l e and, by using predicates ( e . g . , AGREE)to yield inverses, specifies what the context-free pattern must produce. Thus there are many variations of how t o generate using a meta-parser.In the area of language inference, t o take another example of language processing,PEDAGLOT suggests various differing ways of approaching the problem. First, ofie may use it a5 a 'relaxation-parser, the 'parse tree1 can be pattern-matched against the new sentence, and hypotheses can be famed. Or, one could place a more rudimentary inference systw on the 'prediction' part of the processor i t s e l f , and using other controls, the predictions that are successful could be rewritten as a new gramar.These two learning paradigms could each be strengthened by way of t h e use of tags t o contain (in a sense) t h e meaning of t h e sentelzces t o be learned, Each of these paradips can be modeled using a meta-parser like PEDAGLM. Thus, a meta-parser can raise [and be prepared to answer) a nlrmbor of interesting questions.i n almost any order, and not just in a backtracking sense (cf. Woods, 1975). Thb efficiency i s realized here since many 'partial parses (partially recognized forns)
:
I t i s well known t h a t t o process n a t u r a l language, one needs both a syntactic description of possible sentences, blended i n some way with a semantic description b f a c e r t a i n domain of discourse, and a r a t h e r d e t a i l e d description of t h e actual processes used i n hearing o r producing sentences.An augmented t r a n s i t i o n network (Woods, 1970) i s qn example of t h e blending of s y n t a c t i c and quasi-semantic descriptions, Here r e g i s t e r s would be repositories o f , o r pointers t o , semantics. When used i n conjunction with a semantic nqtwork, an ATN can be used 60 parse o r t o generate (Simmons and Slocum, 1912) sentences. The i s s u e of changing the des'cription of t h e actual processes used i n such systems has been touched on by Woods ( i n using a 'generation modet), t o some extent by Gimmons and Slo~um (usi~g decision functions t o control s t y l e of generation), and t o a l a r g e r extent by Kaplari (19751, i n h i s General Syntactic Procdssor, GSP. GSP indeed i s one example of a system i n which syntax, semantics and t o some extent processes can each be u s e f u l l y defined.If we look at syntax, semantics and processes as t h r e e describable components, these systems j u s t mentioned i l l u s t r a t e how thoroughly intertwined they can become-t o the extent t h a t t h e o r i s t s from time t o time deny the existence o r a t l e a s t t h e importance of some one of them. Ignoring t h a t dispute, I would-like t o concentrate on the question of being a b l e t o comprehensively describe one's theory of language i n terms of its syntax, semantics and processes i n a way t h a t allows f o r t h e i r necessary and extensive intertwining connections, but a t t h e sane time allows one t o describe them independently.parser which can make grammatical relaxations i f i t i s given an Ill-formed s t r i n g , so as t o arrive a t a k l o s e s t t possible parse f o r the' s t r i n g . This probl-em involved describing a k o r r e c t t grammar and then (in some way) describing a space of deviations az that night be allowed by the paxser. Thus the syntax would be fixed and the way t h e parser uses it would separately have to be described. I t was soon noticed that efficiency could be greatly enhanced i f some rudimentary notion of semantic p l a u s i b i l i t y could also be used. I t would have t o be described i n a way related t o the cbrrect syntax but s t i l l be usable by the parser. Thus, f o r my purposes, the descriptions had t o be independent of one another.One feature of a relaxation parser i s t h a t it can ' f i l l i n the gaps' of a string t h a t is missing various words. If one could, which my relaxation parser did not, specify the semantic context of a sentence, the generated sentence might be semantically rather plausible. In any case, the relaxation parser operates i n various respects l i k e an actual parser or like a generator, and it was t h i s r e l ationship between parsing and generating that became of i n t e r e s t , Out of the design of the relaxation parser, the notation (independent of syntax) which t o some extent describes various processes and choices of alternate ways of processing was developed. Thus, one may take a s e t of syntax and semantic descriptions and then through describing the processing 'modest involved, define a processor which uses the particular algorithm t h a t the individual processes together define, One may c a l l the parser that i s programmabLe i n i t s processes a meta-parser, of which various existing q a r s e r s and generators appear t o be special cases,A closer examination of t h e parser I have developed (called PEDAGLOT*) may show some such aspects of meta-parsing, especially as regards the relationship between parsing and generating. I will describe the syntactic and semantic parts o f t h e parser first: by noting i t s resemblances to t h e parser of J . Earley (1970) and the ATN system of Woods. Then I w i l l describe t h e process-type specifications t h a t are available, and the use of meta-parsers as a basis f o r defining general language behaviors. Purther detail can be found in the PEDAGMT manual (Fabens, 1972 and The fundamental operation of t h e parser i s very s i m i l a r t o t h e operation of Earleyvs parser, with augmentations f o r recording t h e r e s u l t s of parses (e,g,, their t r e s t r u c t u r e , and various of t h e i r a t t r i b u t e s , which I c a l l ftags'). It is given a grammar a s a s e t of context-free rules with various extensions, most i m p~r t a n t of which a r e t h a t LISP functions may be used as predicates instead of terminals, and thay each rule may be followed by operations t h a t are defined i n tbnns of t h e s y n t a c t i c elements a f t h e r u l e i n question, A n example of t h i s notation i s as follows: S -+ NP VP => [AGREE [REF NP] [VB VP] ] [ S U M = [REF NP] ] [OW = [REF VP] [VB = [VB VP] ] S -* NP [BE] [VPASS] BY NP => [AGREE [REF NP] [VB [BE] ]I [SUM = [REF NP I ] ] [OBJ = REF NP] ] [VB = [VB [VPASS] ] ] NP -+ [DET] [N] => [REF = [N]] W + [VINP => [VB = [V]] [REF = [REF NP]]
Appendix:
| null | null | null | null | {
"paperhash": [
"lyon|syntax-directed_least-errors_analysis_for_context-free_languages",
"simmons|generating_english_discourse_from_semantic_networks"
],
"title": [
"Syntax-directed least-errors analysis for context-free languages",
"Generating English discourse from semantic networks"
],
"abstract": [
"A least-errors recognizer is developed informally using the well-known recognizer of Earley, along with elements of Bellman's dynamic programming. The analyzer takes a general class of context-free grammars as drivers, and any finite string as input. Recognition consists of a least-errors count for a corrected version of the input relative to the driver grammar. The algorithm design emphasizes practical aspects which help in programming it.",
"A system is described for generating English senterraces from a form of semantic ~e{s ia rrhid~ #~e nodes are ~'ord-sense mea~dngs a~d the paths are primarily deep case relations. The grammar ~sed by {he system is ia the form of a nef~ork that imposes an ordering on a set of syntactic transfnrmations {ha{ are expressed as t J S P flmctio~ts. The generation algorithm ~ses the information in the semantic net,~ork to select appropri~, ate genera{bin paths through the grammar. The system is designed for ase as a computational tool that allahs a linguist *o develop and stmly methods for ge~erating s~rfaee strings from an undeHying semantic sm~ctt~re. Initial findings ~ith regard fo form determiners such as voice, form, tense, and mood, some n~es fk~r embedding sentences, and some attention to pronominal substitution are reported. The system is programmed in I3SP 1~5 and is avMiab|e from the authors."
],
"authors": [
{
"name": [
"G. Lyon"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. F. Simmons",
"Jonathan Slocum"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null
],
"s2_corpus_id": [
"17061193",
"12500356"
],
"intents": [
[],
[]
],
"isInfluential": [
false,
false
]
} | null | 591 | 0 | null | null | null | null | null | null | null | null |
7cb88433c5ad5c59c3224fd4b2134068cc3be0fb | 219309151 | null | A Lexical Process Model of Nominal Compounding in {E}nglish | A theoretical model for nominal compound formation in English is presented in which the rul-es are representations of lexical processes. It is argued that such rules can be g e n e r a l i z e d to account f o r many nominal compounds with similar structure and to enable new compounds to be produced and understood. It is shown that nominal compounding depends crucially on the existence of a llcharacteristic'' r e l a t i o n s h i p between a nominal and t h e vexb which occurs in a relative clause paraphrase of a compound which contains the nominal. A computer implementation of the model is presented and the problems of binding and rule selection a r e discussed. | {
"name": [
"Rhyne, James R."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 4 | 4 | null | null | Linguistic Issues.Nominal compounds are sequences of two or more nominals which have the semantic effect of noun phrases with attached relative clauses. The rightmost nominal is g e n e r a l l y i he primary referent of t h e compound the other nominals restrict the reference of the rightmost nominal i n much the same fashion t h a t a relative clause does. Tbeae are, of course, exceptions in which t h e rightmost nominal i s figurative or euphemistic (e.g. family jewels). Compounds occur frequently in English and Germanic languages, but infrequently in the Romance languages where their function i s largely performed by nominal-prepositionnominal sequences (e. g. chemin de fer , agent de change) .nominal compounds is quite simple --the three variants are NAN, N-participle-N, and N-gerund-N. In the N-N form, either 0 % the two nominals may in fact be yet another nominal compound, giving a structure like (N-N)-N or N-(N-N); the f i r s t of these forms seems to occur much more often than the second (examples of each t y p e are: t y p e w r i t e r mechanic, liquid roach poison).I assume that t h e process of nominal compounding is syntactically a process in which a r e l a t i v e clause is reduced by deleting all elements of t h e relative c l a u s e but one and preposing t h e single remaining element i n f r o n t of t h e antecedent nominal. In addition, the clause verb may be nominalized 4nd preposed. Other linguists have proposed different derivations for nominal compounds; Lees [ 3 ] , for example, derives nominal compounds from nominal-preposition-nominal sequences. There are two reasons why I feel that Lees approach i s wrong: (1) there are English compounds for which no reasonable equivalent nominal-prepos it i o nnominal paraphrase can be given ( e . g . windmill), and (2) there are subtle meaning differences between t h e nominal compounds and their nominal-preposition-nominal counterparts (county clerk vs. clerk for the county). I f nominal compounds and nominal--preposition-nominal sequences are derived from forms l i k e relative clauses, then the differences in meaning can be accounted I t happens t h a t i n E n g l i s h , whenever a, d e r i v e d nominal of a n a c t i s the r i g h t element i n a compound, t h e n t h e l e f t element is almbst always a n occupant of one of t h e c a s e s l o t s o f t h e verb. Thus the h e a r e r c a n a s s i g n steam t o t h e Means c a s e w i t h some a s s u r a n c e . verb; i n English, this is u s u a l l y indicated b y a n adverb o r an adverbial phrase. If the speaker is w i l l i n g t o a s s e r t that a boat is c h a r a c t e r i s t i c a l l y used t o catch t u r t l e s , then t h e nominal compound t u r t l e boat may be used. The hearer sill use the general r u l e t o place t u r t l e and boat i n the proper case slots, and because a compound was used b y t h e speaker, the hearer w i l l i n f e r Qhat the boat is one w h i c h is c h a r a c t e r i s t i c a l l y used to catch t u r t l e s , There are o t h e r problems which arise with the g e n e r a l i z a t i o n of rules; f o r example, compounding never produces a compound i n which the l e r t element is a proper noun, unless the proper noun ie t h e name of a process (e.g. Harkov process) o r is a Source, Performer, o r Goal of an a c t of g i v i n g . I t a l s o seems t o be t r u e t h a t compounds are not g e n e r a l l y formed when a l e x i c a l i t e m is several l e v e l s below t h e general term which appears i n the r u l e (e.g. r e p a i m i d g e t ) o r when a c r o s s -c l a s s i f i c a t o r y term is used (e.g. automobile I n d i a n as an I n d i a n who r e p a i r s automobiles).With all of the preceding discussion in mind, I would now like t o t u r n t o the model of nominal compounding which I h a v e p r e s e n t l y implemented and running.The computer model of compounding a c c e p t s r e l a t i v e c l a u s e s t r u c t u r e s as i n p u t and produces nominal compound s t r u c t u r e s a s output when t h e i n p u t is a p p r o p r i a t e . I t is w r i t t e n i n a language with many parentheses t h e language was chosen f o r its program development f a c i l i t i e s , i . e . b u i l t -i n e d i t o r , r a t h e r t h a n for its i n t e r p r e t i v e c a p a b i l i t i e s . The program which produces nominal compounds is a p a t t e r n matching i n t e r p r e t e r ; it a p p l i e s a r u l e of compound formation b y matching one side of t h e r u l e w i t h t h e input s t r u c t u r e , and i f c e r t a i n c r i t e r i a are s a t i s If this rule is t o a p p l y t o t h e r e l a t i v e c l a u s e structure glven i n Figure 1 and g e n e r a t e the compound flower m a r k e t , t h e n t h e r u l e i n t e r p r e t e r must recognize t h a t t h e r e l a t i v e c l a u s e i n Figure 1 i s a n i n s t a n c e of t h a t g i v e n i n F i g u r e 2 . The matching procedure d o e s this by d e t e r m i n i n g t h a t t h e reference s e t of the n o m i n a l flowers is a subset of the r e f e r e n c e s e t of the nominal goods. | null | null | null | Main paper:
:
Linguistic Issues.Nominal compounds are sequences of two or more nominals which have the semantic effect of noun phrases with attached relative clauses. The rightmost nominal is g e n e r a l l y i he primary referent of t h e compound the other nominals restrict the reference of the rightmost nominal i n much the same fashion t h a t a relative clause does. Tbeae are, of course, exceptions in which t h e rightmost nominal i s figurative or euphemistic (e.g. family jewels). Compounds occur frequently in English and Germanic languages, but infrequently in the Romance languages where their function i s largely performed by nominal-prepositionnominal sequences (e. g. chemin de fer , agent de change) .nominal compounds is quite simple --the three variants are NAN, N-participle-N, and N-gerund-N. In the N-N form, either 0 % the two nominals may in fact be yet another nominal compound, giving a structure like (N-N)-N or N-(N-N); the f i r s t of these forms seems to occur much more often than the second (examples of each t y p e are: t y p e w r i t e r mechanic, liquid roach poison).I assume that t h e process of nominal compounding is syntactically a process in which a r e l a t i v e clause is reduced by deleting all elements of t h e relative c l a u s e but one and preposing t h e single remaining element i n f r o n t of t h e antecedent nominal. In addition, the clause verb may be nominalized 4nd preposed. Other linguists have proposed different derivations for nominal compounds; Lees [ 3 ] , for example, derives nominal compounds from nominal-preposition-nominal sequences. There are two reasons why I feel that Lees approach i s wrong: (1) there are English compounds for which no reasonable equivalent nominal-prepos it i o nnominal paraphrase can be given ( e . g . windmill), and (2) there are subtle meaning differences between t h e nominal compounds and their nominal-preposition-nominal counterparts (county clerk vs. clerk for the county). I f nominal compounds and nominal--preposition-nominal sequences are derived from forms l i k e relative clauses, then the differences in meaning can be accounted I t happens t h a t i n E n g l i s h , whenever a, d e r i v e d nominal of a n a c t i s the r i g h t element i n a compound, t h e n t h e l e f t element is almbst always a n occupant of one of t h e c a s e s l o t s o f t h e verb. Thus the h e a r e r c a n a s s i g n steam t o t h e Means c a s e w i t h some a s s u r a n c e . verb; i n English, this is u s u a l l y indicated b y a n adverb o r an adverbial phrase. If the speaker is w i l l i n g t o a s s e r t that a boat is c h a r a c t e r i s t i c a l l y used t o catch t u r t l e s , then t h e nominal compound t u r t l e boat may be used. The hearer sill use the general r u l e t o place t u r t l e and boat i n the proper case slots, and because a compound was used b y t h e speaker, the hearer w i l l i n f e r Qhat the boat is one w h i c h is c h a r a c t e r i s t i c a l l y used to catch t u r t l e s , There are o t h e r problems which arise with the g e n e r a l i z a t i o n of rules; f o r example, compounding never produces a compound i n which the l e r t element is a proper noun, unless the proper noun ie t h e name of a process (e.g. Harkov process) o r is a Source, Performer, o r Goal of an a c t of g i v i n g . I t a l s o seems t o be t r u e t h a t compounds are not g e n e r a l l y formed when a l e x i c a l i t e m is several l e v e l s below t h e general term which appears i n the r u l e (e.g. r e p a i m i d g e t ) o r when a c r o s s -c l a s s i f i c a t o r y term is used (e.g. automobile I n d i a n as an I n d i a n who r e p a i r s automobiles).With all of the preceding discussion in mind, I would now like t o t u r n t o the model of nominal compounding which I h a v e p r e s e n t l y implemented and running.The computer model of compounding a c c e p t s r e l a t i v e c l a u s e s t r u c t u r e s as i n p u t and produces nominal compound s t r u c t u r e s a s output when t h e i n p u t is a p p r o p r i a t e . I t is w r i t t e n i n a language with many parentheses t h e language was chosen f o r its program development f a c i l i t i e s , i . e . b u i l t -i n e d i t o r , r a t h e r t h a n for its i n t e r p r e t i v e c a p a b i l i t i e s . The program which produces nominal compounds is a p a t t e r n matching i n t e r p r e t e r ; it a p p l i e s a r u l e of compound formation b y matching one side of t h e r u l e w i t h t h e input s t r u c t u r e , and i f c e r t a i n c r i t e r i a are s a t i s If this rule is t o a p p l y t o t h e r e l a t i v e c l a u s e structure glven i n Figure 1 and g e n e r a t e the compound flower m a r k e t , t h e n t h e r u l e i n t e r p r e t e r must recognize t h a t t h e r e l a t i v e c l a u s e i n Figure 1 i s a n i n s t a n c e of t h a t g i v e n i n F i g u r e 2 . The matching procedure d o e s this by d e t e r m i n i n g t h a t t h e reference s e t of the n o m i n a l flowers is a subset of the r e f e r e n c e s e t of the nominal goods.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.006768 | null | null | null | null | null | null | null | null |
ed6d11372ccd7ca3d448e498f5806280bb72bfd5 | 219300013 | null | How Does a System Know When to Stop Inferencing? | The problem of constmining the set of hfemtces added t o a set of beliefs is considered. One method, based on finding a minimal unifying structure, is frresented and discussed. The method is meant t o pnxride internal criteria f o r inference cut-off. * This work was partially suppored by NSF-Grant SCC 72-05465A01. | {
"name": [
"Rosenschein, Stan"
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 6 | 2 | null | Natural language processing systems that are sensitive t o the semantic and logical content of processed sentences and to the p~g l r a t i c s of t h e i r use generally draw inferences. A set of fonmilas representing the meaning of a sentence and the 'state of b e l i e f t of the system is augnented by other related formulas (the inferences) which are retrieved and/or constructed during the pmcessing. The problem t o be investigated here is: How can thi$ process be contmlled? Can reasonable criteria be found f o r restraining the addition of inferences?Top-down inferences fol.luwing from the meaning of lexical items (often expressed by decomposition into primitives) are clearly bounded, i f no interactions are allowed amng the generated sub-formulas. This process (which we call EXPANSION) w i l l not be discussed here. Rather, we shall be concerned with SYNlESIS, i.e., the addition of new formulas based on the zxsenceof already generated lower-level formulas, 'v~hich we shall call kliefs. In particular, we are concerned with infererces addgd because a set cf beliefs is recognized as fitting a plre-defined pattern.The question we ask is: Given an initial set of beliefs ovm a set of ;ri&ives,w h a t ' c r i t e~i o n can be us& to M t the pmcess of pattern matcljng The operatibns to be de-ibed below ate exp-mrre f u l l y in (R,75),where a desmiption of a ccquter i q l e m~n t a t h is also presented.* See also (C,75), (W,75). Second, the usual distinction between 'antecedent' and 'consequent'Ifclauses in the pattern is not h t a i n e d ; a clause in the pattern may serveas an antecedent on one occasion and a consequent on ano-Eher.Third, if 'defined1 lexical item were t o be associated w i t h tht:patterns, noting which variables are to be bound as arguments upon instantiation, then the SYN'IIESIZE function can be used to canpute sumnarizhg expressions. a*ls SYNTHESIZE remsents a possiELe formalism for lexical insertion.IV. An f5wmole of the beration SYNTHESIZE Far the sake o f illustration, l e t the primitives be: have been reversed. In Situation 2, (1) was an input and ( 2 ) was infemed , whereas in Situation 3, (2) was input wd (1) inferred. The curresponding clauses of the loan pattke2*n were serving as antecedents on one occasion and consequents on the other. This follows naturKLly fran the way SYNIPIESIZE was defined.(BENIGN ?x)In this regard the reader rnay notice that sane input belief sets might yield ' w a r r a n t e d t or 'spu~?ious' inferences--jumping to too many cmclusicms.Hawever, the incmmntal addition of new patterns corrects this anom19 i n a natural way: Patterns which formerly were 'least covers' may cease to be so i n the extended pattern set. | null | null | null | null | Main paper:
i. introduction:
Natural language processing systems that are sensitive t o the semantic and logical content of processed sentences and to the p~g l r a t i c s of t h e i r use generally draw inferences. A set of fonmilas representing the meaning of a sentence and the 'state of b e l i e f t of the system is augnented by other related formulas (the inferences) which are retrieved and/or constructed during the pmcessing. The problem t o be investigated here is: How can thi$ process be contmlled? Can reasonable criteria be found f o r restraining the addition of inferences?Top-down inferences fol.luwing from the meaning of lexical items (often expressed by decomposition into primitives) are clearly bounded, i f no interactions are allowed amng the generated sub-formulas. This process (which we call EXPANSION) w i l l not be discussed here. Rather, we shall be concerned with SYNlESIS, i.e., the addition of new formulas based on the zxsenceof already generated lower-level formulas, 'v~hich we shall call kliefs. In particular, we are concerned with infererces addgd because a set cf beliefs is recognized as fitting a plre-defined pattern.The question we ask is: Given an initial set of beliefs ovm a set of ;ri&ives,w h a t ' c r i t e~i o n can be us& to M t the pmcess of pattern matcljng The operatibns to be de-ibed below ate exp-mrre f u l l y in (R,75),where a desmiption of a ccquter i q l e m~n t a t h is also presented.* See also (C,75), (W,75). Second, the usual distinction between 'antecedent' and 'consequent'Ifclauses in the pattern is not h t a i n e d ; a clause in the pattern may serveas an antecedent on one occasion and a consequent on ano-Eher.Third, if 'defined1 lexical item were t o be associated w i t h tht:patterns, noting which variables are to be bound as arguments upon instantiation, then the SYN'IIESIZE function can be used to canpute sumnarizhg expressions. a*ls SYNTHESIZE remsents a possiELe formalism for lexical insertion.IV. An f5wmole of the beration SYNTHESIZE Far the sake o f illustration, l e t the primitives be: have been reversed. In Situation 2, (1) was an input and ( 2 ) was infemed , whereas in Situation 3, (2) was input wd (1) inferred. The curresponding clauses of the loan pattke2*n were serving as antecedents on one occasion and consequents on the other. This follows naturKLly fran the way SYNIPIESIZE was defined.(BENIGN ?x)In this regard the reader rnay notice that sane input belief sets might yield ' w a r r a n t e d t or 'spu~?ious' inferences--jumping to too many cmclusicms.Hawever, the incmmntal addition of new patterns corrects this anom19 i n a natural way: Patterns which formerly were 'least covers' may cease to be so i n the extended pattern set.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.003384 | null | null | null | null | null | null | null | null |
1f7287848add2627c75dffe5d8a756933981d528 | 219304701 | null | A Natural Language Processing Package | A set of SAIL programs has been implemented for analyzing large bodies of natural language data in which associations exist between strings and sets of strings. These programs include facilities for compiling information such as frequency of occurrence of strings (e.g. word frequencies) or substrings (e.g. consonant cluster frequencies), and describing relationships among strings (e.g. various phonological realizations af a word). Also, an associative data base may be interactively accessed on the basis of keys corresponding to different types of data elements, and a pattern matcher allows retrieval of incompletely specified elements. Applications Of this natural language processing package include analysis of phonological variation for specifying and testing phonological rules, and comparison across languages for historical reconstruction. | {
"name": [
"Brill, David and",
"Oshika, Beatrice T."
],
"affiliation": [
null,
null
]
} | null | null | null | 1975-11-01 | 0 | 2 | null | null | null | null | null | -alveolar flapping occurs under several stress conditions whidh appear to be related to noun affixes. These preliminary observations can be systematically investigated using t h e interactive query system. | Main paper:
:
-alveolar flapping occurs under several stress conditions whidh appear to be related to noun affixes. These preliminary observations can be systematically investigated using t h e interactive query system.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.003384 | null | null | null | null | null | null | null | null |
c9a47edfd737e689ffd5f92dc86f51d8a664a47e | 219302397 | null | Grammatical Compression in Notes and Records: Analysis and Computation | ~inguistic mechanisms of compression are used when making notes within a context where the objects and meanings are known. Mechanisms of compressidn in medical records for a collaborative study of breast cancer are described. The syntactic devices were mainly deletion of words having a special status in the grammar of the whole language and deletion in particular positions of word+ having a special sta&us in the sublanguage. The deIeted forms are described and sublanguage Qord classes defined. A subcorpus of the medical records was parsed by an existing computer parsing system; a component covering the deletion-forms was added to the granunar. Modifications to t,he computer grammar are discussed and the parsing results are summarized. All 1anguages"have mechanisms of compression. Sentences may be embedded within other sentenaes by means of nominalization and complementation. Various grammatical transformations involve deletion of certain parts of the sentence. In medical records, we find entries such as no evidence of metastases, which may be said to be derived Trea something like There is no evidence of metastases. Such incomplete sentences are not common in the spoken language of the medical records (i.e. dictated reports). However when physiciakrs themselves are requirbd to write-material for records, compression mechanisms are qmmonly use&. Although this paper will deal with a m i f i c corpus, similar devices would I often be used for compression in other s-ations where there is pressure to write as little as possible, Legal, educational, and scientific recordg where informal notes are kept w o u m be other examples of this class of sitqations. The original motivation for this study was to develop effective methods for storing &e information in a medical record and to be able to retrieve this information for purposes of research, medical care, or administration. Fsoan previous research, the feasibility of verbatim input of dictated narrative has been established, Computerized extraction of the information has been shown to be feaeible i~ a test system ACORN (Automated Coding of Report ~arrative). his system has been described in detail in a series of previous papers. 1 d t 3 1 I .Dm J. Bross et al. "Information in Natural Languages : A New Approach1'. | {
"name": [
"Anderson, Barbara B. and",
"Sager, Naomi"
],
"affiliation": [
null,
null
]
} | null | null | null | 1975-11-01 | 0 | 12 | null | null | null | null | null | In medical records, we find entries such as no evidence of metastases, which may be said to be derived Trea something like There is no evidence of metastases. Such incomplete sentences are not common in the spoken language of the medical records (i.e. dictated reports). However when physiciakrs themselves are requirbd to write-material for records, compression mechanisms are qmmonly use&. ha8 been reported e1sew)rere. However, the t h i t d , intermediate c l a s s of material cannot be handled by ACORN o r by TICES. Therefore, a l i n g u i s t i c a n a l y s i s of t h i s type of material has been undertaken with t h e ultimate objective of s e t t i n g up a comprehensive eomputer system t h a t can handle almost everything i n the medical records.I n the earlier effoxts t o develop n a t u r a l language technology, t h e work was f a c i l i t a t e d by t h e f a c t t h a t t h e documents involved were s t r i c t l y f o r the trans-5 mission of f a c t u a l information. Such documents a r e regarded as important both by the persons who are f i l l i n g them o u t and by t h e persons who read them. I n t h i s no-nonsense s i t u a t i o n where t h e record may be c r i t i c a l l y reviewed by t h e peers of the person who is reporting t h e information, unambiguous and informat i v e transmission of information i s a c r i t i c a l need. Some of the simplicities i n t h e present analysis may b e~e c u l i a r t o ws type of s i t u a t f o n .The existence of a subculture with shared t r a i n i n g , objectives, and experience may f a c i l i t a t e t h e note-taking process i n somewhat t h e same way t h a t a person taking notes f o r himself can somehow be more concise without ambiguity...Howeveb, r m a n y other note-taking s i t u a t i o n s would involve subculture, though not necessarily a medical one, and t h e findings here might be expected t o have sdne general a p p l i c a b i l i t y .4 1.D.J. Bross e t al. "Unobtrusive Biomedical Data-Input Systems". - Bio-Medical Computing, No. 4, 1973, pp. 219-228 .E J I.D.J.Bross, P.A. Shapiro and B.B. Anderson. "How Information Is Carried i n S c i e n t i f i c Sublangukges". Science, Vol. 176, No. 4041, 1972 , pp. 1303 -1307 .The medical&es discussed here are ffom tjhe records of t h e Surgical Adjutrant Breast Project, a nationwide collaborative study involving 36 medical i n s t i t u t i o n s . The records were f i l l e d o u t by medical and paramedical personnel a t the p a r t i c i p a t i n g i n s t i t u t i o n s and cehtralized a t Ro$glell Park Nemri&l The notes were typed v e x b a t h using An IBM Mag Card Communicator so as t o obtain simultaneously a typed paper document and a record i n computer-usable form. This device is used i n t h e data-input sgstem of T~C E S ; an existing system f o r handling completely structured records. I t would presumably be usea i n any extension of TICES which would handle medical nates. I n e i s ' a n a l y s i s the com- (1) tense and the verbbe (tb e ) ; (2) s u b j e c t , tense and t h e verbbe;(3) t h e subject; and (4) s u b j e c t , tense, and verb (V) other thanbe.A second c h a r a c t e r i s t i c of fragments which makes deleted m a t e r i a l recovera b l e is that both t h e m e t e d material and the remainders c o n s i s t of words i n easily defined subclasses, based on both d i s t r i b u t i o n a l and semantic c r i t e r i a .These subclasses are e a s i l y defined because of t h e nature of t h e sublanguage;i n general t h e vocabulary i s limited and each word has a l i m i t e d semantic range.The question on a form khich is being answered can a l s o be used a s a basis f o r r e t o r i n g deleted material.One of the most commonly deleted items i n the medical records is tbe (1 and 2 ) . Tense is perhaps t h e most important informationbe gives. The d e l e t i o n of tense i n the medical records causes no ambiguity because usually the physician describes t h e s i t u a t i o n a t the time of f i l l i n g o u t the report, O t h e r w i s e he gives the time i n a t i m e phrase: x-rays on November 2. Fragment Types I n Table 1 w e l i s t t h e fragment types, giving an example of each, but not with a l l occurring word subcl&!3ses. The types w i l l S i r s t be given according t o what material is deleted and then w i l l be f u t t h e r subclassed according t o t h e highest nodes of the t r e e s t r u c t u r e of t h e remainder. The material i n brack e t s is the word subclasses which a r e assumed .So have been deldted. 6~r o s s e t al. "Information i n Natural Languages: A New Approach," 1969. To test t h e l i n g u i s t i c analysis, a subset of the manually analyzed corpus of medical records was parsed by computer, using the NYU Linguistic String Parser. I n defining FRAGMENT, we have used parts of t h e grammar which were defined independently of t h e fragment problem. That t h i s is p o s s i b l e is i n i t s e l f a p a r t i a l v e r i f i c a t i a n of t h e conclusion from manual a n a l y s i s t h a t only limited, grammatically s p e c i f i a b l e , deletion-forms occur i n t h e fragments seen i n notes and records. For example, t h e dropping of t h e verb (type 1 of Table 1 ) can occur i n normal English when a sentence containing t h e verbbe occurs as t h e o b j e c t of a verb l i k ef i n d , e.g. W e found t h e chest c l e a r t o pekcussion and auscultation. thickening over t h e r i g h t apex. where a noun phrase (PA and'lateral c h e s t 11 -5-71) precedes an a s s e r t i o n about t h a t ngun phrase.TABLE 2. SOME WORD SUBCLASSES N -b w a $ t N-change I R-dimas ion N-disease N-exam N-locatibrl N-patient N-physician N-therapy N-time V-be-equivalent V-change V -d i s c e r V-patient-object V-patient-subject V-physician-subject V-show-Space permits only a few remarks about these d e f i n i t i o n s . Parse t r e e f o r FRAGMENT = 5-2-67 chest--no chanqe sjnce 2-7-67A summary of the parsing r e s u l t s is given i n Table 4 A second stage of processing of this type is now being applied to the parsed corpus of medical records and will be reported in a subsequent paper. A convenkent test of the adeqyacy of the parsing outputs is therefore whether they can serve as input to this second stage of processing (called forhatting). It can be seen in Table 4 that a number of "wrong" parses were still adequate as input to the formatting; the segmentation of the sentence into parts was correct even if the parts were assigned an incorrect syntactic status, e.g., object instead of adjunct. Only when the first parse was not adequate for formatting was the sentence rerun to obtain alternative analyses.The parsing times are a rough indication of the efficiency of the parsing but two points should be kept in mind. (1) The present LSP system is not a production model, but a research tool, with all that implies. (2) A bignificant fraction of the input sentences were "no data" types, e,g., None this visit.These word sequences were so limited linguistically that a literal formula could serve to reaogniae them. The experimental use of such a formula cut down parsing times on the no-data entries from about 1.817 to0.030. However, this formula was not used in the parsing summarized in Table 4 , -- | Main paper:
:
In medical records, we find entries such as no evidence of metastases, which may be said to be derived Trea something like There is no evidence of metastases. Such incomplete sentences are not common in the spoken language of the medical records (i.e. dictated reports). However when physiciakrs themselves are requirbd to write-material for records, compression mechanisms are qmmonly use&. ha8 been reported e1sew)rere. However, the t h i t d , intermediate c l a s s of material cannot be handled by ACORN o r by TICES. Therefore, a l i n g u i s t i c a n a l y s i s of t h i s type of material has been undertaken with t h e ultimate objective of s e t t i n g up a comprehensive eomputer system t h a t can handle almost everything i n the medical records.I n the earlier effoxts t o develop n a t u r a l language technology, t h e work was f a c i l i t a t e d by t h e f a c t t h a t t h e documents involved were s t r i c t l y f o r the trans-5 mission of f a c t u a l information. Such documents a r e regarded as important both by the persons who are f i l l i n g them o u t and by t h e persons who read them. I n t h i s no-nonsense s i t u a t i o n where t h e record may be c r i t i c a l l y reviewed by t h e peers of the person who is reporting t h e information, unambiguous and informat i v e transmission of information i s a c r i t i c a l need. Some of the simplicities i n t h e present analysis may b e~e c u l i a r t o ws type of s i t u a t f o n .The existence of a subculture with shared t r a i n i n g , objectives, and experience may f a c i l i t a t e t h e note-taking process i n somewhat t h e same way t h a t a person taking notes f o r himself can somehow be more concise without ambiguity...Howeveb, r m a n y other note-taking s i t u a t i o n s would involve subculture, though not necessarily a medical one, and t h e findings here might be expected t o have sdne general a p p l i c a b i l i t y .4 1.D.J. Bross e t al. "Unobtrusive Biomedical Data-Input Systems". - Bio-Medical Computing, No. 4, 1973, pp. 219-228 .E J I.D.J.Bross, P.A. Shapiro and B.B. Anderson. "How Information Is Carried i n S c i e n t i f i c Sublangukges". Science, Vol. 176, No. 4041, 1972 , pp. 1303 -1307 .The medical&es discussed here are ffom tjhe records of t h e Surgical Adjutrant Breast Project, a nationwide collaborative study involving 36 medical i n s t i t u t i o n s . The records were f i l l e d o u t by medical and paramedical personnel a t the p a r t i c i p a t i n g i n s t i t u t i o n s and cehtralized a t Ro$glell Park Nemri&l The notes were typed v e x b a t h using An IBM Mag Card Communicator so as t o obtain simultaneously a typed paper document and a record i n computer-usable form. This device is used i n t h e data-input sgstem of T~C E S ; an existing system f o r handling completely structured records. I t would presumably be usea i n any extension of TICES which would handle medical nates. I n e i s ' a n a l y s i s the com- (1) tense and the verbbe (tb e ) ; (2) s u b j e c t , tense and t h e verbbe;(3) t h e subject; and (4) s u b j e c t , tense, and verb (V) other thanbe.A second c h a r a c t e r i s t i c of fragments which makes deleted m a t e r i a l recovera b l e is that both t h e m e t e d material and the remainders c o n s i s t of words i n easily defined subclasses, based on both d i s t r i b u t i o n a l and semantic c r i t e r i a .These subclasses are e a s i l y defined because of t h e nature of t h e sublanguage;i n general t h e vocabulary i s limited and each word has a l i m i t e d semantic range.The question on a form khich is being answered can a l s o be used a s a basis f o r r e t o r i n g deleted material.One of the most commonly deleted items i n the medical records is tbe (1 and 2 ) . Tense is perhaps t h e most important informationbe gives. The d e l e t i o n of tense i n the medical records causes no ambiguity because usually the physician describes t h e s i t u a t i o n a t the time of f i l l i n g o u t the report, O t h e r w i s e he gives the time i n a t i m e phrase: x-rays on November 2. Fragment Types I n Table 1 w e l i s t t h e fragment types, giving an example of each, but not with a l l occurring word subcl&!3ses. The types w i l l S i r s t be given according t o what material is deleted and then w i l l be f u t t h e r subclassed according t o t h e highest nodes of the t r e e s t r u c t u r e of t h e remainder. The material i n brack e t s is the word subclasses which a r e assumed .So have been deldted. 6~r o s s e t al. "Information i n Natural Languages: A New Approach," 1969. To test t h e l i n g u i s t i c analysis, a subset of the manually analyzed corpus of medical records was parsed by computer, using the NYU Linguistic String Parser. I n defining FRAGMENT, we have used parts of t h e grammar which were defined independently of t h e fragment problem. That t h i s is p o s s i b l e is i n i t s e l f a p a r t i a l v e r i f i c a t i a n of t h e conclusion from manual a n a l y s i s t h a t only limited, grammatically s p e c i f i a b l e , deletion-forms occur i n t h e fragments seen i n notes and records. For example, t h e dropping of t h e verb (type 1 of Table 1 ) can occur i n normal English when a sentence containing t h e verbbe occurs as t h e o b j e c t of a verb l i k ef i n d , e.g. W e found t h e chest c l e a r t o pekcussion and auscultation. thickening over t h e r i g h t apex. where a noun phrase (PA and'lateral c h e s t 11 -5-71) precedes an a s s e r t i o n about t h a t ngun phrase.TABLE 2. SOME WORD SUBCLASSES N -b w a $ t N-change I R-dimas ion N-disease N-exam N-locatibrl N-patient N-physician N-therapy N-time V-be-equivalent V-change V -d i s c e r V-patient-object V-patient-subject V-physician-subject V-show-Space permits only a few remarks about these d e f i n i t i o n s . Parse t r e e f o r FRAGMENT = 5-2-67 chest--no chanqe sjnce 2-7-67A summary of the parsing r e s u l t s is given i n Table 4 A second stage of processing of this type is now being applied to the parsed corpus of medical records and will be reported in a subsequent paper. A convenkent test of the adeqyacy of the parsing outputs is therefore whether they can serve as input to this second stage of processing (called forhatting). It can be seen in Table 4 that a number of "wrong" parses were still adequate as input to the formatting; the segmentation of the sentence into parts was correct even if the parts were assigned an incorrect syntactic status, e.g., object instead of adjunct. Only when the first parse was not adequate for formatting was the sentence rerun to obtain alternative analyses.The parsing times are a rough indication of the efficiency of the parsing but two points should be kept in mind. (1) The present LSP system is not a production model, but a research tool, with all that implies. (2) A bignificant fraction of the input sentences were "no data" types, e,g., None this visit.These word sequences were so limited linguistically that a literal formula could serve to reaogniae them. The experimental use of such a formula cut down parsing times on the no-data entries from about 1.817 to0.030. However, this formula was not used in the parsing summarized in Table 4 , --
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.020305 | null | null | null | null | null | null | null | null |
8a819e7c232582ba1864c52b39f824ed42b62832 | 58148141 | null | A Tuneable Performance Grammar | Thlr paper d e 1 c r f b ~l a t u n a a b l e performance grrtamar) currently being d e v e l o p e d t o r SpWGh u n 4 ~t s t r n d l n g , It rhows how attrtbuter o f words a t e d r f i n a d a n d p r o p a q d t c d t o r u c c r r t i v r l y larger p h r a s e r , how other r t t r l b u t r r are required, how ' f a c t o r r ' reference them t o 1 t h e Parser ehoorc araanp c o w c t i n ? d.et1nttionr in o.rdet te i n t e r p r e t t h e utterance correctly, and haw there facterg can easily b e changed to* adapt t h e grammar to othar blseaurrrs and c O n t * x t # r Factors t h a t @iiQht b r c l u 8 i f t l d r s a8Yntcctlc* a r e amPhrriZedr b u t t h e r t t r i b u t r l t h e y rrteresace nrad not be, end 8eldam a r e , p u ~r l y ryntrctic, Thlr r a m r t c h was rupportrd by t h e Dtfense Advanced R8@8&rCh Project8 Agrncy o f t h ~ Dlprrtmanr a t D ~t e n a e and mongtoxad b y t h a U,s, Army R*r@rrch O f f l c a under Contract Ha, DkWC04-75*C-0006, 8 perfotairnce grramrr (PC) d*tihrr t h a fatm and meaning a t t h e k i n d 8 of u t t Q r a n c e r t h a t occur i n rpaatanooua dialog, When tnr detinltianr of t h e grammar R t a V i d ~ tnformrtfon t h a t h t l ~s a prtrcr ehaora these r u l e 8 ao8z likely t o l a a d t d correct int@tpretatlonr of uttrrtanCs&, t h e grrar@ar 1 8 r a i d t o be 9tunrd*, Uhan t h e tuning 1.8 crarily chrncraa wnrn t h r dornrin o t dircaurrr, ehanges, thc grasmrr 1 6 s a i d t o b r *tunrrblr#. The ability t o tune r grammar & a p a r t i c u l a r l y inportant in spreeh undrrrtanding where t h e inhatsnt Uncmrtrinty of t h e i n p u t caurer f a l r e path8 through t h e grammar t b b e multipll@d, T h i l pa)rr dercribrr a funrablr PG b a i n g d e v e l o p e d J o i n t l y by 8RI and SDC for r camputrrrnbrred 8P@@ch underrtanding eystam, 7[t# vocrbulrty and phrr,ro typesr r a l s c l t o d frsm pt'otoc44rr r r a appropriate f a t rrelng and anruering q u s 8 t i o n r about p r o p e t t i r r o t 8ubarriner. Thr PG now dafiner War 7 0 word and phraae c r t r g o r t e r , I t s reape rxtandr f a r bayand syntax, A d l # ~o u r s @ coaponrnt e n r b l ~o i f t o nand&@ rnaphora and ~l l i p i i t ~ 4s i n t *What i r t h e s u r t r c e dirpL~crsrnt o f t h e Lafeyattr?.l.. What 11 It8 arrf t ? " , and "What i n the i r n u t h o t tnr baEayrtte?...,, The Ethan i ~l e n ? ~ X r e m a n t i c 8 component definer a common meaning f a r paraphrrsar, b8 gn V h r S p e r d a t t h r Lafayettr i r $ 0 knotrH and * t h e Laf&y)tpe h.8 a rperd o f t h i r t y knatrW. ( S a r Walker r t a l . , f97St Paxtan rnd R ~b i n ~o n t 1975; nrndriw, l W 5 t D a u t l c h r 1 9 7 d r 1 Each d r @ l n L t i o n comporlng t h r PC ha8 threr p r x t r , T h e f l r s t name$ r word c r t e g o t y or a p h t r 8 e Catrgory and p r a v l d @ r r cont'rxt-ftrr production far ftr caaparltianr Tha racond p a r t # | {
"name": [
"Robinson, Jane J."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 0 | 3 | null | null | null | null | null | RUL&,DEF NP11 NP 8 ART N O M j ATTRIBUTES 'RELN FRQM NOMp Q O C U l FROM ARTl MOOD @ 'DEC, CHU s GINTER8CCT(CHU[XRT)tCHU(NOM)),THEN POOR ELSE OK, N a a t l n g | Main paper:
:
RUL&,DEF NP11 NP 8 ART N O M j ATTRIBUTES 'RELN FRQM NOMp Q O C U l FROM ARTl MOOD @ 'DEC, CHU s GINTER8CCT(CHU[XRT)tCHU(NOM)),THEN POOR ELSE OK, N a a t l n g
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.005076 | null | null | null | null | null | null | null | null |
99092ea362970d4bb867015b5a4b791c328ee1e0 | 219306786 | null | The Nature and Computational Use of a Meaning Representation for Word Concepts | V a r i c u s representations have been used to partray the ribmiqs of mrd (mtably action) mncepts. The mst pruninent of these include decanp3sition trees, 1 -repreentations such as the Predicate Calculus, and senantic netwrh. The propsition-based semntic ne-rk notation developed by S c h u b e r t (1974) is especially well suited f o r including p r a p t i c and sepantic M o mtion as part of the meaning representation of irdividual wrd mncepts. The attempt is made in t h i s paper to explore the nature of mrd concepts d s e mxmings are represented as senantic netmrks and ,W .investigate their cmpu~tional use within the fr-rk of a natural language prpcessing systan. The meaning of a cxmce1ps. is explained i n lx%rns of other concepts and thmug its relatimship to other concepts. V a r i o u s representations have been used to p r t r a y the reanbgs of corrcepts. The mst pranhen~t of these include deccmposition trees (Idcuff, 1972; Wilks, 1973), linear represenbtipns such as the Pre dicate CalcuL21~ (Sandewall , 1971) , and m t i c n e m r k s . Natural langage processing systans can comrenimtly utilize factual. knowledge represented i n the form of s a ~n t i c mmrks. The visual suggestiveness o f semantic nebmrks aids both" in the forrmrlatian and e y p s i t i o n 05 the ccmputer data structwes~ they resemble. The use of r;~xnantic mtwrks can b fourd in the wrks of m y authors writing on natural language processing ( i r ~~l o d i n g *hank Anderson ard: Bcrwer 1973; ar@ Palm! 1971) as well as other forms of understandhj (Wluding W i n s t m i 1970; and Guzman 1971). In u t i l i z i n g semantic network representations, these authDrs have made use of the f o l l ~ characteristics of s m t i c nets. First (and most important) , nsdes t h a t denote the same cmncept are m t duplimtad (in mst cases). It is | {
"name": [
"Cercone, Nick"
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 11 | 4 | null | null | null | null | The pragmatic section indludes the t a p l a t e ( s ) t h a t guides the parse of the utterance and tm lists: the f i r s t list contains p r o p o s i t h that represent t h e implications t h a t are fikely to be needed for the ccmprehensicm of subsequent text;ard the second list contains propositions representing critical implications that we expet xpect m t c h in the surface structure. In Figure 1 this first list is (P3) and the second list is (Pl,P2), The m t i c section oontains the netwrk that represents the mankg of the mrd sense. Notice that Figures 1, 3, and 6 aU have the notion of change in containment location in ccmmn. This r n r r e~n d s to a general ooncept that s u b m s not only differing senses of "drink" but also other more specific concepts as w e l l , like "eating" or "receiving an enana". This obsenmtion has led b the follow--consideration.When creating the m i n g representations (netwrks) for concepts it is desirable to amid the duplication of propositions in storage. If we extractrrore general concepts £ r a n the specific concepts that they sub-(totally or in part), we can ayoid duplicatian by associating the crmron propositions w i t h the more general concept.In a sense the mrk of both Scharak 11972) and W i l k s (1973) w r t s the amtention that the mmhg of a m p t is best representd by precatjons at the highest l e d . of ga-ierality that adequately explain the term's -.Thus we extract frcm "drinking" (and eatbg, etc.) the s t m e t y e shown in Figure 7 .W e might reasonably label the corcept expressed by this structure -"ingestT1.It is impDrtant to note, Inever, that while Schank and W i l k s might mnclude that "ingesting" is a primitive action, that I cansider it a general concept. This applies to all primitive actions prt orw wad of Schank and Wilks. l&amimtion of Figure 7 shms clearly that ingesting is -mt a primitive action but one wbse meaning is expressed in tens of causes, m t i o n , time, and other concepts. In similar fashion F i v 9 diagrams one meaning of "eating", again based on the general (mncept "iIlgest". The implications make the mst camonly used inferences part of the meaning representation of word concept. The propositions, for eample P1 and P4 are s b Figure 10 . See Cercone (1975) for sarnple l a i c a l entries, in particular the en- the word can, in tum, be represented i n an analogous muma. In particular the notiin of a "cause" seems to me to be m more "primitive1' than "drink". This met- (1975) . Associated w i t h mrd senses are templates as described above. For example, the sense *GIVE1 of the root form "give" has a t a p l a t e "X GIVE Y 2" and antry for "drink".kdalternative (ALTERN) tanplate "X GIVE Z TO Y" associated with it.The template, e.g.. "X GIVE Y Z" , is used to guide the parsing. In t h i s example X I Y and 2 are variables representing the argUru3nts of the predicate "give"that we expect to find in the surface utterance i n the given order. hbre detailed infomatian concerning the argunwtnts is obtained by exmhing the netmrk propositions, for the sense of "givett in question, that involve the arguments. Thus X, in this case, wuld represent ap AKtM?SE rvJnindl capable of "giving".This is very similar to what Shcank does when parsing in conceptual dependency theory. If the w r d s in the surface utterance do rmt satisfy the constraints for arguments of the predicate being examined, it is due t o one of four reasons.First, alternab syntactic constructions could exist. Secohd, a different sense of the qctiorr is "correct". Third, the particular action cadidate in question is not the action of the clause. Finally, sane other reason, like slang expressions might be the cause.Whenever arguments fail to satisfy predicates, a search for alternative inplication templaw begins. The result of this search is shclwn quite clearly J in Figure 1 1 of Section IV for tbe ternary predicate "give". In t h a t example "give" is u&d syntactically in two different £ o m to distinguish the indirect object, one w i t h the preposition TO and one w i t b u t . If this approach f a i l s then the list of senses for the root £om is fvther examined. Schank, R., Goldmn, N., Rieger, C., an3 Riesbeck, C. (1973) . "Margie: ?&r~@y,-- m = ( - 1 -- RmDY -- =JCMNGaVEJUDYTHF, = RED EmK. THEN, JUM! GPVE = T H E B R t Y W N B O O K r n~. - - ---1-c-t ASSOCIATED ' ACTION--WRIX&E TRIFLES +ti- FT - GIVE^ *BOOK1 Z) (*GIVE1 *JUDYl Y) ( * G N E l *JOHN1 X)) - -- - + H x D I F m - = ((N CADJ CLASF ( ( 0 0) (*rnl)) ) ) Z) - - - + t + THE: sEzmMrIcm tt3- - - w* = *JOHN1 *PROPERLYk - - X --m o o 0 1 *mn Y - - PROP0002 W K 1 PRED - PFOPOO~Z INST0003 ARG - - ~0 0 0 4 I N S T O O 0 3 AXIL; - - ~0 0 0 4 -0005 P m - PW3POOO1 INST0003 Z -- - PFOP0006 -1 PRED - - PIiL1POO06 3&$7! 0003 M G - m o o 0 1 GIVE^ PRED - - = - +.+tASSOCIATH A c q O N -v - TEUgLES +ti- - - ( ( *~m -Y j G GIVE^ w n Z ) (*em1 * m y 1 X) ) - -- - -+ H M m I F m - - ( ( M -( ( 0 0) ("B-) 1) Z ) - -- - tts. THE SEM14NTIC NE2 -ti-t - - *A!KM* w* mm* - ~P O O O~ - "JUDY1 - X - FJFDPOOO~ m PRED - - PE3P0008 ~J S T O O O~ ATI(3 - miaPOOl0 INSTOO09 IIEil: - - - r n P 0 0 1 0 'wNso011 PRED - ~~0~0 0 0 7 INSTOO09 - z - - ~~0~0 0 0 7 '!mRn. Y - - PR13POOL2 *BFmNl PRED -- - P R D P O O ,~~ INSTOO09 AFG - PNlP0007 * G N E l PRED -- ( W ) vAnalysis, kspnse G m e r a h n , and Inferme on mlishtl, ~hird-1ntepmti6nal Joint Oonference on Aftificiai ~n t e l l i g e r~l e , SRI, WO Park, California, Simmns, R. F. , an3 ~ruce, B. C. (1971) . "Scme Relations Between Predicate-Calcu- Notable systans currently in Mgue that utilize "primitives" in this way include t b s e of Wilks (1973) ap3 Schank : t al. (1973) .2 Nxds i n clauses are rrorpblogically analyzed and, based on that analysis, they are classified to determine a l l of their pssible syntactic funqtions in an utterance.In WWirogradfs (1972) w r k , "gives" is recognized as a t r a n s i t i v e action that requires tw, objects : his classification is TRANS2 . In Figure A.1, A, B , and REL are mere distinguishipg mks. They are analogous to parenthesis or ccmas in the Predicate Calculus & that they serve to relate d e m t h g terms syntactically; they are mn-demtative thenselves. Whenever possible they w i l l be cbsen to be = -,i.e. to enhance readability and be suggestive, but they amid be chosen as numeric labels as well.One adwhtage of the explicit.notation of Figure A . 1 is that it works for n -a q (1172) predicates. The sentence "John gives the h k to Mzty" involves "gives" as a three place prediate. * It is diagram& as in Figure A. 3 Figure A. 3 is appealing because of the significance w e can attach to labelsagent, object, and recipient. 3y no means is Figure A . 3 a graphical analogue of "case-skructured" grmmars. Cases are not view& as conceptually primitive binary relations as F i l h r e (1968) and res~ar~hers influenced by him, notably Schank (1972) , view than. In a case structured system the central ncde would denote a specific action or process with the property t h a t it is a "giving"and involves John, the book, and b k z y as agent, object, and Pecipient respectively. Case relations can be understood as ccmplex mnprimitive terms derived fm such causally and telmlcgically related sequences of states. The wble notion of a case derives frcm the syntactic and s a~n t i c similarities bebeen the role played by the argurrrents of many predicates. Nevertheless the mtion of an "agent" to depend in part on causal priority of a state of the supposed agent in the sequence of states mer consideration, and in part on the extent to which purpsive behaviour can be ascribed to the supposed agent in general, and in part to the extent to which the particular sequence of states which he initiated can be assum3 to be intentional on his part. See Cercone and S&ndxrt (1974) for a further discussion of cases.One f inaL notational pint by way of introduction needs to be made. The Fig. A-3 . "John gives the book t c Mary. 8t | null | Main paper:
:
The pragmatic section indludes the t a p l a t e ( s ) t h a t guides the parse of the utterance and tm lists: the f i r s t list contains p r o p o s i t h that represent t h e implications t h a t are fikely to be needed for the ccmprehensicm of subsequent text;ard the second list contains propositions representing critical implications that we expet xpect m t c h in the surface structure. In Figure 1 this first list is (P3) and the second list is (Pl,P2), The m t i c section oontains the netwrk that represents the mankg of the mrd sense. Notice that Figures 1, 3, and 6 aU have the notion of change in containment location in ccmmn. This r n r r e~n d s to a general ooncept that s u b m s not only differing senses of "drink" but also other more specific concepts as w e l l , like "eating" or "receiving an enana". This obsenmtion has led b the follow--consideration.When creating the m i n g representations (netwrks) for concepts it is desirable to amid the duplication of propositions in storage. If we extractrrore general concepts £ r a n the specific concepts that they sub-(totally or in part), we can ayoid duplicatian by associating the crmron propositions w i t h the more general concept.In a sense the mrk of both Scharak 11972) and W i l k s (1973) w r t s the amtention that the mmhg of a m p t is best representd by precatjons at the highest l e d . of ga-ierality that adequately explain the term's -.Thus we extract frcm "drinking" (and eatbg, etc.) the s t m e t y e shown in Figure 7 .W e might reasonably label the corcept expressed by this structure -"ingestT1.It is impDrtant to note, Inever, that while Schank and W i l k s might mnclude that "ingesting" is a primitive action, that I cansider it a general concept. This applies to all primitive actions prt orw wad of Schank and Wilks. l&amimtion of Figure 7 shms clearly that ingesting is -mt a primitive action but one wbse meaning is expressed in tens of causes, m t i o n , time, and other concepts. In similar fashion F i v 9 diagrams one meaning of "eating", again based on the general (mncept "iIlgest". The implications make the mst camonly used inferences part of the meaning representation of word concept. The propositions, for eample P1 and P4 are s b Figure 10 . See Cercone (1975) for sarnple l a i c a l entries, in particular the en- the word can, in tum, be represented i n an analogous muma. In particular the notiin of a "cause" seems to me to be m more "primitive1' than "drink". This met- (1975) . Associated w i t h mrd senses are templates as described above. For example, the sense *GIVE1 of the root form "give" has a t a p l a t e "X GIVE Y 2" and antry for "drink".kdalternative (ALTERN) tanplate "X GIVE Z TO Y" associated with it.The template, e.g.. "X GIVE Y Z" , is used to guide the parsing. In t h i s example X I Y and 2 are variables representing the argUru3nts of the predicate "give"that we expect to find in the surface utterance i n the given order. hbre detailed infomatian concerning the argunwtnts is obtained by exmhing the netmrk propositions, for the sense of "givett in question, that involve the arguments. Thus X, in this case, wuld represent ap AKtM?SE rvJnindl capable of "giving".This is very similar to what Shcank does when parsing in conceptual dependency theory. If the w r d s in the surface utterance do rmt satisfy the constraints for arguments of the predicate being examined, it is due t o one of four reasons.First, alternab syntactic constructions could exist. Secohd, a different sense of the qctiorr is "correct". Third, the particular action cadidate in question is not the action of the clause. Finally, sane other reason, like slang expressions might be the cause.Whenever arguments fail to satisfy predicates, a search for alternative inplication templaw begins. The result of this search is shclwn quite clearly J in Figure 1 1 of Section IV for tbe ternary predicate "give". In t h a t example "give" is u&d syntactically in two different £ o m to distinguish the indirect object, one w i t h the preposition TO and one w i t b u t . If this approach f a i l s then the list of senses for the root £om is fvther examined. Schank, R., Goldmn, N., Rieger, C., an3 Riesbeck, C. (1973) . "Margie: ?&r~@y,-- m = ( - 1 -- RmDY -- =JCMNGaVEJUDYTHF, = RED EmK. THEN, JUM! GPVE = T H E B R t Y W N B O O K r n~. - - ---1-c-t ASSOCIATED ' ACTION--WRIX&E TRIFLES +ti- FT - GIVE^ *BOOK1 Z) (*GIVE1 *JUDYl Y) ( * G N E l *JOHN1 X)) - -- - + H x D I F m - = ((N CADJ CLASF ( ( 0 0) (*rnl)) ) ) Z) - - - + t + THE: sEzmMrIcm tt3- - - w* = *JOHN1 *PROPERLYk - - X --m o o 0 1 *mn Y - - PROP0002 W K 1 PRED - PFOPOO~Z INST0003 ARG - - ~0 0 0 4 I N S T O O 0 3 AXIL; - - ~0 0 0 4 -0005 P m - PW3POOO1 INST0003 Z -- - PFOP0006 -1 PRED - - PIiL1POO06 3&$7! 0003 M G - m o o 0 1 GIVE^ PRED - - = - +.+tASSOCIATH A c q O N -v - TEUgLES +ti- - - ( ( *~m -Y j G GIVE^ w n Z ) (*em1 * m y 1 X) ) - -- - -+ H M m I F m - - ( ( M -( ( 0 0) ("B-) 1) Z ) - -- - tts. THE SEM14NTIC NE2 -ti-t - - *A!KM* w* mm* - ~P O O O~ - "JUDY1 - X - FJFDPOOO~ m PRED - - PE3P0008 ~J S T O O O~ ATI(3 - miaPOOl0 INSTOO09 IIEil: - - - r n P 0 0 1 0 'wNso011 PRED - ~~0~0 0 0 7 INSTOO09 - z - - ~~0~0 0 0 7 '!mRn. Y - - PR13POOL2 *BFmNl PRED -- - P R D P O O ,~~ INSTOO09 AFG - PNlP0007 * G N E l PRED -- ( W ) vAnalysis, kspnse G m e r a h n , and Inferme on mlishtl, ~hird-1ntepmti6nal Joint Oonference on Aftificiai ~n t e l l i g e r~l e , SRI, WO Park, California, Simmns, R. F. , an3 ~ruce, B. C. (1971) . "Scme Relations Between Predicate-Calcu- Notable systans currently in Mgue that utilize "primitives" in this way include t b s e of Wilks (1973) ap3 Schank : t al. (1973) .2 Nxds i n clauses are rrorpblogically analyzed and, based on that analysis, they are classified to determine a l l of their pssible syntactic funqtions in an utterance.In WWirogradfs (1972) w r k , "gives" is recognized as a t r a n s i t i v e action that requires tw, objects : his classification is TRANS2 . In Figure A.1, A, B , and REL are mere distinguishipg mks. They are analogous to parenthesis or ccmas in the Predicate Calculus & that they serve to relate d e m t h g terms syntactically; they are mn-demtative thenselves. Whenever possible they w i l l be cbsen to be = -,i.e. to enhance readability and be suggestive, but they amid be chosen as numeric labels as well.One adwhtage of the explicit.notation of Figure A . 1 is that it works for n -a q (1172) predicates. The sentence "John gives the h k to Mzty" involves "gives" as a three place prediate. * It is diagram& as in Figure A. 3 Figure A. 3 is appealing because of the significance w e can attach to labelsagent, object, and recipient. 3y no means is Figure A . 3 a graphical analogue of "case-skructured" grmmars. Cases are not view& as conceptually primitive binary relations as F i l h r e (1968) and res~ar~hers influenced by him, notably Schank (1972) , view than. In a case structured system the central ncde would denote a specific action or process with the property t h a t it is a "giving"and involves John, the book, and b k z y as agent, object, and Pecipient respectively. Case relations can be understood as ccmplex mnprimitive terms derived fm such causally and telmlcgically related sequences of states. The wble notion of a case derives frcm the syntactic and s a~n t i c similarities bebeen the role played by the argurrrents of many predicates. Nevertheless the mtion of an "agent" to depend in part on causal priority of a state of the supposed agent in the sequence of states mer consideration, and in part on the extent to which purpsive behaviour can be ascribed to the supposed agent in general, and in part to the extent to which the particular sequence of states which he initiated can be assum3 to be intentional on his part. See Cercone and S&ndxrt (1974) for a further discussion of cases.One f inaL notational pint by way of introduction needs to be made. The Fig. A-3 . "John gives the book t c Mary. 8t
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.006768 | null | null | null | null | null | null | null | null |
67aeaa38e224d5b460919a50b872d09f12bde719 | 219307558 | null | The Conceptual Description of Physical Activities | A system has been designed to translate connected sequences of visual images of physical activities into conceptual descriptions. The representation of such activities is based on a canonical v e r b of motion so that the conceptual description will be compatible with semantic networks i n natural language understanding systems. A c a s e structure is described which i s derived f r o m the kinds of information obtainable i n image data. A possible solution is presented to the problem of segmenting the temporal information st r e a m into linguistically and physically meaningful events. An example | {
"name": [
"Badler, Norman"
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 12 | 5 | null | null | null | null | null | If we view a motion picture such a s illustrated in Figure 1 , we a r e able to give a description of the physical activities in the scenario. This description is linguistic in the sense that the words used express our recognition of objects and movements as conceptual entities. A system for performing a sizeable part of this transformation of visual data into conceptual descriptions has been designed. It i s described in Badler (1975 );here we will present one small part of the system which is concerned with the organization of abstracted data from successive images of the scenario. W e are interested in a possible solution to the following problem: Given that a conceptual description of a scenario i s to be generated, how is it decided where one verb instance starts and another ends? In other words, we seek computational criteria which separate visual experience into discrete "chunks" o r events. By organizing the representation of an event into a case structure for a canonical motion verb, events can be described in linguistic terms. Verbs of motion have been investigated directly o r indirectly by Miller (1972) . Hendrix et aL l t 7 3a, 197 3b). Martin (1973) . and Schank (1973) ; semantic databases using variants of case structure verb representations Wllmore(1968)) include Winograd (197 Z) , Rumelhart et a1 (197 2 ) , and Simmons (197 3) . W e are concerned with physical movements of rigid o r jointed objects so that motions may be restricted to translations and rotations. Objects m a y appear o r disappear and the observer is free to move about. The resulting activities a r e combinations of the se where observer motions are factored out if at all possible. We assume that the scenarios contain recognizable objects exhibiting physically possible, and preferably natural, motions.A particular activity might consist of a single event, a sequence of events, s e t s of event sequences, o r hierarchic organizations of events. The concept of "walking" is a good example of the last. Events a r e the basic building blocks of the conceptual description, and our events indicate the motion. of objects. D O F J N ( W ) ,UP(wARD) ,NORTHWARD SOUTHWARD. EASTWARD .WESTWARD ACROSS ,AGAINST ,ALONG ,APART, AROUND , A M ,AMY -FROM, BEHIND,BY,F'ROM,IN,rnO,OFF, OW-OF,ON,ONTO,OUT,OUT-OF, OVER,THROUGH,TO ,TOGETHER, U N D mindicative of source and target between the path of an object and other (mving ) objects between an event and a previous eventTOGEmER,WITHbeyond the scope of the current system, although a semantic inference component could be included. Our descriptions consist mostly of observation of motion i n context rather than explanation of why motion occurred.The general descriptive methodology is to keep only one static relational description of the scenario, that of the current image. Changes between it and the next sequential image a r e described by storing the names of changes in event nodes i n a semantic network. In general, names of changes correspond to adverbs o r prepositions (adverbials) describing directions o r changing static relationships. Computational definitions for the set of adverbials in Table 1 appear in Badler (1975) . We a r e only concerned with the senses of the adverbials pertaining to movement. Definitions a r e l implemented a s demons: procedures which a r e activated, the executed, by the successive appearance of certain assertions in the image description o r current conceptual database. These demons a r e related to those of Charniak (1972) , although our use of them, their numbers, and their organization a r e simplified and restricted. They a r e used t o recognize o r classify properties o r changes and to generate the hierarchic descriptive structure. An essential feature of this methodology i s that the descriptions a r e continually condensed by this change abstraction process; descriptions grow in depth rather than length.The semantic information stored for each object in the scenario includes its TYPE, structural SUB-PARTS, VISIBILITY, MOBILITY, LOCATION ORIENTATION, and SIZE. Most of these properties a r e determined from the image sequence, but some a r e stored in object models (indexed by TYPE) in the semantic network,The event8 a r e also nodes in the semantic network. Each object is potentially the SUBJECT of an event node. A sequence of event nodes forms a history of movement of a n object; only the latest node i n the sequence i s active, The set of active event nodes describes the current events in the scenario seen so far. The cases of the event node along with their approximate definitions follow.SUBJECT: An object which is exhibiting movement.AGENT: A motile object which contacts the SUBJECT. INSTRUMENT: A moving object which contacts the SUBJECT. REFERENCE: A pair of object features (on a fixed object) which a r e used t o f i x absolute directions independent of the observer's position.A temporally-ordered list of adverbials and their associated objects which apply to this SUBJECT. TRAJECTORY: The spatial direction of a location change of the SUBJECT. VELOCITY: The approximate magnitude of the velocity of the SUBJECT along the TRAJECTORY; it includes a RATES list containing STARTS, STOPS and (optionally) INCREASES o r DECREASES. AXIS: The spatial direction of an axis of an orientation change (rotation) of the SUBJECT.the AXIS.NEXT: The temporal successor event node having the same SUBJECT.The time of the onset of the event.The time of the termination of the event.REPEAT-PATH: A list of event nodes which form a repeating sequence.These cases differ f r o m Miller's (1972) primarily in the lack of a "permissive" An event node is terminated when it has a non-NIL NEXT value.The function CREATE-EVENT-NODE (property pairs) creates an event node with t h e indicated case values, returning the node as a result. SHIFT manipulates property queues when they r e t p i r e updating:LAST-property: = CURRENT -property;CURRENT-property: = NEW -property; NEW-property: = $8T h e time will be abbreviated by T N and TL, F o r a particular event node E:TN: = IV3W-END-TIMII: (E); TC: = CURrnNT-END-TIME (E);Thus T N is always equal t o the present image time.Now we can present the algorithm for the demon which controls the construction of the entlre event graph. It is executed once for each image when all lower level demons have finished; it c r e a t e s , terminates, o r updates each c u r r e n t event node.A. 1. Creating event nodes. A 1 1. An event node E is created when a mobile object f i r s t becomes visible and identifiable as a n object.E: = CmATE-EVENT-NODE((SUB JECT object-node) (VELOCITY(* 0. 0. )) (ANGULAR-VELOCITY (4' 0 . 0. )) (START -TIME NIL) (END-TIME (* T N TN)) ).The NIL START-TIME h a s the interpretation that we do not know what was happening to this object p r i o r to time TN. E : = CREATE -EVENT -NODE( (SUBJECT object-part-node) (AGENT parent-object-node) ( INSTRUMENT joint-node) (REFERENCE . . . ) ( DIlsECTION . . .) (TRAJECTORY , . . ) (VELO-CITY . . .) (AXIS . . , ) (ANGULAR-VELOCITY . . .) (START -TIME T C) (END-TIME (TN TC TC)) ).This is interpreted as the parent object moving the part using the joint as 76 the "instrament". Any appfopriate attributes are placed in the N E W -property positions. The node E is then immediately terminated (A. 1.3).A. 1.3, An event node E2 is created whenever another event node E l is terminated.TC: = CURRENT-END-TIME(E 1); NEXT(E1): = CREATE-EVENT-NODE((SUBJECT.. .) (AGENT. , . ) ( INSTRUMENT.. . ) (REFERENCE.. . ) (DIIIE;CTION,. , ) (TRAJECTORY SHLFT'(TRAJECT'0RY (E 1))) (VE M C I T Y SHIFT(VELOC1TY (E 1))) (AXIS SHLFT(AXIS(E I))) (ANGULAR-VELOCITY SHIFT(ANGULAR- VELOCITY (E 1))) (START-TIME TC) (END-TIME SHIFT(END-TIME(E 1))); E2: = NEXT(E 1).SUBJECT, AGENT, INSTRUMENT, REFERENCE, and DJRECTION are those which were present at termination of the previous node, subject to any additional conditions that changes in these m a y require.A. 2. Terminating event nodes. An event node E is terminated when there a r e significant changes in its properties. All queue structures a r e deleted. Changes i n type (2) adverbials must be preceded by a change i n TRAJECTORY, but some of these changes may be too slight to cause termination from the TRAJECTORY criteria. (A. 2 . 6 . Since the type (4) adverbials a r e only indicators of current source and target, these do not change unless the path of the SUBJECT changes o r the target object moves. Therefore no terminations arise from this group.The type (5) adverbials relate paths of the SUBJECT t o other objects.They cause termination when they come into effect, and terminate their own nodes when they cease t o describe the path. What does a n event mean? This algorithm motivates a theorem that the events generated a r e the finest meaningful partition of the movements i n the image sequence into distinct activities. The hypothesis o f the assertion i e the natural environment being observed and the linguisticallybased conceptual description desired, The conclusion is that a n event node produced from this algorithm describes either the lack of motion o r else a n unimpeded, simple linear o r smoothly curving (or rotating) motion of the SUBJECT with no CONTACT changes. In addition, the orientation of the SUBJECT does not change much with respect to the trajectory. The proof of this assertion follows directly from the choice of termination conditions. W e will apply this algorithm to data obtained from each of the images in Figure 1 . The lower front edge of the house is arbitrarily chosen a s the REFERENCE feature; NORTH is toward the right of each image. We will not discuss the computation of the static relations from each image, only list i n Table 2 the changes i n the static description f r o m irnage-toimage. Trajectory and rotation data a r e omitted for simplicity, although changes of significance a r e indicated.If we "write out" the event node sequence using the canonical motion verbs MOVES and TURNS with the adverbial phrases from the RATES and DIRECTION lists, we obtain the following lengthy, but accurate. TOWARD the OBSERVER and EASTWARD, then NORTHWARD-AND-EASTWARD, then FROM the DRIVEWAY and OUT -OF the DRWEWAY, then OFF-OF the DRIVEWAY, Another condensation can be applied for the sake of less redundant output.It does not, however, permanently affect the database:The CAR MOVES TOWARD the OBSERVER, then ONTO the ROAD, while GOING FORWARD, then FROM the DRIVEWAY, then AROUND the HOUSE, then AWAY-FROM the HOUSE, then STOPS TURNING, then MOVES AWAY.Note that FROM the DRIVEWAY follows ONTQ the ROAD. This i s due to the pictorial configuration: the car is on the road before it leaves the driveway. The position of the "while GOING FORWARD" phrase could be shifted backwards in time to the beginning of the translatory motion, but this may be risky i n general. W e will leave it where it is, since this i s primarily a higher level linguistic matter.By applying demons which recognize instances of specific motion verbs t o the individual event nodes, then condensing as above, we get:The CAR APPROACHES, then MOVES ONTO the ROAD, then LEAVES the DRIVEWAY, than TURNS AROUND the HOUSE, then DRIVES AWAY -FROM the HOUSE, then STOPS TURNING, then D R I V E S AWAY.The major awkwardness with this last description is that it relates the c a r to every other object i n the scene. Normally one object o r another would be the focus of attention and statements would be made regarding i t s role. Such manipulations of the descriptions a r e yet unclear.In conclusion, we have outlined a small part of a system designed to translate sequences of images into linguistic semantic structures. Space permitted us only one example, but the method also yields descriptions for scenarios containing observer movement and jointed objects (such as walking persons). The availability of low level data has significantly shaped the definitions of the adverbials and motion verbs. Further work on these definitions, especially motion v e r b s , is anticipated. We expect that the integration of vision and language systems will benefit both domains by sharing i n the specification of representational stmctures and description processes. | Main paper:
:
If we view a motion picture such a s illustrated in Figure 1 , we a r e able to give a description of the physical activities in the scenario. This description is linguistic in the sense that the words used express our recognition of objects and movements as conceptual entities. A system for performing a sizeable part of this transformation of visual data into conceptual descriptions has been designed. It i s described in Badler (1975 );here we will present one small part of the system which is concerned with the organization of abstracted data from successive images of the scenario. W e are interested in a possible solution to the following problem: Given that a conceptual description of a scenario i s to be generated, how is it decided where one verb instance starts and another ends? In other words, we seek computational criteria which separate visual experience into discrete "chunks" o r events. By organizing the representation of an event into a case structure for a canonical motion verb, events can be described in linguistic terms. Verbs of motion have been investigated directly o r indirectly by Miller (1972) . Hendrix et aL l t 7 3a, 197 3b). Martin (1973) . and Schank (1973) ; semantic databases using variants of case structure verb representations Wllmore(1968)) include Winograd (197 Z) , Rumelhart et a1 (197 2 ) , and Simmons (197 3) . W e are concerned with physical movements of rigid o r jointed objects so that motions may be restricted to translations and rotations. Objects m a y appear o r disappear and the observer is free to move about. The resulting activities a r e combinations of the se where observer motions are factored out if at all possible. We assume that the scenarios contain recognizable objects exhibiting physically possible, and preferably natural, motions.A particular activity might consist of a single event, a sequence of events, s e t s of event sequences, o r hierarchic organizations of events. The concept of "walking" is a good example of the last. Events a r e the basic building blocks of the conceptual description, and our events indicate the motion. of objects. D O F J N ( W ) ,UP(wARD) ,NORTHWARD SOUTHWARD. EASTWARD .WESTWARD ACROSS ,AGAINST ,ALONG ,APART, AROUND , A M ,AMY -FROM, BEHIND,BY,F'ROM,IN,rnO,OFF, OW-OF,ON,ONTO,OUT,OUT-OF, OVER,THROUGH,TO ,TOGETHER, U N D mindicative of source and target between the path of an object and other (mving ) objects between an event and a previous eventTOGEmER,WITHbeyond the scope of the current system, although a semantic inference component could be included. Our descriptions consist mostly of observation of motion i n context rather than explanation of why motion occurred.The general descriptive methodology is to keep only one static relational description of the scenario, that of the current image. Changes between it and the next sequential image a r e described by storing the names of changes in event nodes i n a semantic network. In general, names of changes correspond to adverbs o r prepositions (adverbials) describing directions o r changing static relationships. Computational definitions for the set of adverbials in Table 1 appear in Badler (1975) . We a r e only concerned with the senses of the adverbials pertaining to movement. Definitions a r e l implemented a s demons: procedures which a r e activated, the executed, by the successive appearance of certain assertions in the image description o r current conceptual database. These demons a r e related to those of Charniak (1972) , although our use of them, their numbers, and their organization a r e simplified and restricted. They a r e used t o recognize o r classify properties o r changes and to generate the hierarchic descriptive structure. An essential feature of this methodology i s that the descriptions a r e continually condensed by this change abstraction process; descriptions grow in depth rather than length.The semantic information stored for each object in the scenario includes its TYPE, structural SUB-PARTS, VISIBILITY, MOBILITY, LOCATION ORIENTATION, and SIZE. Most of these properties a r e determined from the image sequence, but some a r e stored in object models (indexed by TYPE) in the semantic network,The event8 a r e also nodes in the semantic network. Each object is potentially the SUBJECT of an event node. A sequence of event nodes forms a history of movement of a n object; only the latest node i n the sequence i s active, The set of active event nodes describes the current events in the scenario seen so far. The cases of the event node along with their approximate definitions follow.SUBJECT: An object which is exhibiting movement.AGENT: A motile object which contacts the SUBJECT. INSTRUMENT: A moving object which contacts the SUBJECT. REFERENCE: A pair of object features (on a fixed object) which a r e used t o f i x absolute directions independent of the observer's position.A temporally-ordered list of adverbials and their associated objects which apply to this SUBJECT. TRAJECTORY: The spatial direction of a location change of the SUBJECT. VELOCITY: The approximate magnitude of the velocity of the SUBJECT along the TRAJECTORY; it includes a RATES list containing STARTS, STOPS and (optionally) INCREASES o r DECREASES. AXIS: The spatial direction of an axis of an orientation change (rotation) of the SUBJECT.the AXIS.NEXT: The temporal successor event node having the same SUBJECT.The time of the onset of the event.The time of the termination of the event.REPEAT-PATH: A list of event nodes which form a repeating sequence.These cases differ f r o m Miller's (1972) primarily in the lack of a "permissive" An event node is terminated when it has a non-NIL NEXT value.The function CREATE-EVENT-NODE (property pairs) creates an event node with t h e indicated case values, returning the node as a result. SHIFT manipulates property queues when they r e t p i r e updating:LAST-property: = CURRENT -property;CURRENT-property: = NEW -property; NEW-property: = $8T h e time will be abbreviated by T N and TL, F o r a particular event node E:TN: = IV3W-END-TIMII: (E); TC: = CURrnNT-END-TIME (E);Thus T N is always equal t o the present image time.Now we can present the algorithm for the demon which controls the construction of the entlre event graph. It is executed once for each image when all lower level demons have finished; it c r e a t e s , terminates, o r updates each c u r r e n t event node.A. 1. Creating event nodes. A 1 1. An event node E is created when a mobile object f i r s t becomes visible and identifiable as a n object.E: = CmATE-EVENT-NODE((SUB JECT object-node) (VELOCITY(* 0. 0. )) (ANGULAR-VELOCITY (4' 0 . 0. )) (START -TIME NIL) (END-TIME (* T N TN)) ).The NIL START-TIME h a s the interpretation that we do not know what was happening to this object p r i o r to time TN. E : = CREATE -EVENT -NODE( (SUBJECT object-part-node) (AGENT parent-object-node) ( INSTRUMENT joint-node) (REFERENCE . . . ) ( DIlsECTION . . .) (TRAJECTORY , . . ) (VELO-CITY . . .) (AXIS . . , ) (ANGULAR-VELOCITY . . .) (START -TIME T C) (END-TIME (TN TC TC)) ).This is interpreted as the parent object moving the part using the joint as 76 the "instrament". Any appfopriate attributes are placed in the N E W -property positions. The node E is then immediately terminated (A. 1.3).A. 1.3, An event node E2 is created whenever another event node E l is terminated.TC: = CURRENT-END-TIME(E 1); NEXT(E1): = CREATE-EVENT-NODE((SUBJECT.. .) (AGENT. , . ) ( INSTRUMENT.. . ) (REFERENCE.. . ) (DIIIE;CTION,. , ) (TRAJECTORY SHLFT'(TRAJECT'0RY (E 1))) (VE M C I T Y SHIFT(VELOC1TY (E 1))) (AXIS SHLFT(AXIS(E I))) (ANGULAR-VELOCITY SHIFT(ANGULAR- VELOCITY (E 1))) (START-TIME TC) (END-TIME SHIFT(END-TIME(E 1))); E2: = NEXT(E 1).SUBJECT, AGENT, INSTRUMENT, REFERENCE, and DJRECTION are those which were present at termination of the previous node, subject to any additional conditions that changes in these m a y require.A. 2. Terminating event nodes. An event node E is terminated when there a r e significant changes in its properties. All queue structures a r e deleted. Changes i n type (2) adverbials must be preceded by a change i n TRAJECTORY, but some of these changes may be too slight to cause termination from the TRAJECTORY criteria. (A. 2 . 6 . Since the type (4) adverbials a r e only indicators of current source and target, these do not change unless the path of the SUBJECT changes o r the target object moves. Therefore no terminations arise from this group.The type (5) adverbials relate paths of the SUBJECT t o other objects.They cause termination when they come into effect, and terminate their own nodes when they cease t o describe the path. What does a n event mean? This algorithm motivates a theorem that the events generated a r e the finest meaningful partition of the movements i n the image sequence into distinct activities. The hypothesis o f the assertion i e the natural environment being observed and the linguisticallybased conceptual description desired, The conclusion is that a n event node produced from this algorithm describes either the lack of motion o r else a n unimpeded, simple linear o r smoothly curving (or rotating) motion of the SUBJECT with no CONTACT changes. In addition, the orientation of the SUBJECT does not change much with respect to the trajectory. The proof of this assertion follows directly from the choice of termination conditions. W e will apply this algorithm to data obtained from each of the images in Figure 1 . The lower front edge of the house is arbitrarily chosen a s the REFERENCE feature; NORTH is toward the right of each image. We will not discuss the computation of the static relations from each image, only list i n Table 2 the changes i n the static description f r o m irnage-toimage. Trajectory and rotation data a r e omitted for simplicity, although changes of significance a r e indicated.If we "write out" the event node sequence using the canonical motion verbs MOVES and TURNS with the adverbial phrases from the RATES and DIRECTION lists, we obtain the following lengthy, but accurate. TOWARD the OBSERVER and EASTWARD, then NORTHWARD-AND-EASTWARD, then FROM the DRIVEWAY and OUT -OF the DRWEWAY, then OFF-OF the DRIVEWAY, Another condensation can be applied for the sake of less redundant output.It does not, however, permanently affect the database:The CAR MOVES TOWARD the OBSERVER, then ONTO the ROAD, while GOING FORWARD, then FROM the DRIVEWAY, then AROUND the HOUSE, then AWAY-FROM the HOUSE, then STOPS TURNING, then MOVES AWAY.Note that FROM the DRIVEWAY follows ONTQ the ROAD. This i s due to the pictorial configuration: the car is on the road before it leaves the driveway. The position of the "while GOING FORWARD" phrase could be shifted backwards in time to the beginning of the translatory motion, but this may be risky i n general. W e will leave it where it is, since this i s primarily a higher level linguistic matter.By applying demons which recognize instances of specific motion verbs t o the individual event nodes, then condensing as above, we get:The CAR APPROACHES, then MOVES ONTO the ROAD, then LEAVES the DRIVEWAY, than TURNS AROUND the HOUSE, then DRIVES AWAY -FROM the HOUSE, then STOPS TURNING, then D R I V E S AWAY.The major awkwardness with this last description is that it relates the c a r to every other object i n the scene. Normally one object o r another would be the focus of attention and statements would be made regarding i t s role. Such manipulations of the descriptions a r e yet unclear.In conclusion, we have outlined a small part of a system designed to translate sequences of images into linguistic semantic structures. Space permitted us only one example, but the method also yields descriptions for scenarios containing observer movement and jointed objects (such as walking persons). The availability of low level data has significantly shaped the definitions of the adverbials and motion verbs. Further work on these definitions, especially motion v e r b s , is anticipated. We expect that the integration of vision and language systems will benefit both domains by sharing i n the specification of representational stmctures and description processes.
Appendix:
| null | null | null | null | {
"paperhash": [
"hendrix|language_processing_via_canonical_verbs_and_semantic_models",
"schank|the_fourteen_primitive_actions_and_their_inferences.",
"charniak|toward_a_model_of_children's_story_comprehension"
],
"title": [
"Language Processing Via Canonical Verbs and Semantic Models",
"The fourteen primitive actions and their inferences.",
"Toward a model of children's story comprehension"
],
"abstract": [
"A natural language question answering system is presented. The system's parser maps semantic paraphrases into a single deep structure characterized by a canonical verb. A modeling scheme using semantic nets and STRIPS-like operators assimilates the sequence of input information. Natural language responses to questions are generated from a data base of semantic nets by \"parsing\" syntactic rules retrieved from the lexicon.",
"In order to represent the conceptual information underlying a natural language sentence, a conceptual structure has been established that uses the basic actor-action-object framework. It was the intent that these structures have only one representation for one meaning, regardless of the semantic form of the sentence being represented. Actions were reduced to their basic parts so as to effect this. It was found that only fourteen basic actions were needed as building blocks by which all verbs can be represented. Each of these actions has a set of actions or states which can be inferred when they are present.",
"Massachusetts Institute of Technology. Dept. of Electrical Engineering. Thesis. 1972. Ph.D."
],
"authors": [
{
"name": [
"G. Hendrix",
"C. Thompson",
"Jonathan Slocum"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. Schank"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Eugene Charniak"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null
],
"s2_corpus_id": [
"10422416",
"61126867",
"62620723"
],
"intents": [
[],
[],
[]
],
"isInfluential": [
false,
false,
false
]
} | null | 591 | 0.00846 | null | null | null | null | null | null | null | null |
88058cc6e4cb7cf345ed1ebeb1753d729aef0e4b | 219301590 | null | Conceptual Grammar [abstract] | In OWL, an implementation o f conceptual grammar, t h e two types o f data items are symbols and concepts and t h e two b a s i c data composition operations are specialization and restriction. A symbol is an alphanumeric s t r i n g headed by ". Symbols correspond to words, suffixes, p r e f i x e s , and word stens in Znglish and the programer can introduce them a t willm OWL concepts correspond t o t h e meanings of EEglish words and phrases. They are constructed using t h e specialization operation, comparable t o CONS i n LISP* (A B) is t h e specialization of A , a concept, by B, a concept o r symbol. OWL f o r m a branching tree under specialization, with SOMETHING a t the t o p . Concepts are given properties by restriction, which puts a concept on the reference list of another concept (compare property lists and S-expressions in LISP). A/B is the r e s t r i c t i o n of A by B. The categories in the specialization tree are semantic, but we use them also f o r the purposes usually assigned to syntactic dategories. A predication is a double specification of 2 model such as present tense or can. Examples are The pool is full of water. ((PRES-TNS (BE (FULL 94TER)) J POOL/THE) The cookie can be in t h e j a f . ((CAN (BE (IN JAR/TIIE))) COOKIE/THE) aob is the f a t h e r o f Sam. ( (PRES -TKS (BE (FATHE: SAM) ITHE) ) BOB) 3ob hits the b a l l . ((PRES-TNS (HIT BALLITHE)) Boa) Bob is hitting the b a l l . ((PRES-TNS (BE (-ING (HIT BALL/THE))))BOB) Starting from t h i s base we will discuss a number of issues buch as n~minalization incorporation, and deep vs surface cases. | {
"name": [
"Martin, William A."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 0 | 0 | null | null | null | null | The categories in the specialization tree are semantic, but we use them also f o r the purposes usually assigned to syntactic dategories.A predication is a double specification of 2 model such as present tense or can. Starting from t h i s base we will discuss a number of issues buch as n~minalization incorporation, and deep vs surface cases. | null | Main paper:
by b.:
The categories in the specialization tree are semantic, but we use them also f o r the purposes usually assigned to syntactic dategories.A predication is a double specification of 2 model such as present tense or can. Starting from t h i s base we will discuss a number of issues buch as n~minalization incorporation, and deep vs surface cases.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0 | null | null | null | null | null | null | null | null |
10c8c8ec01f4a0dd8c0d864063c0c2a1b2ca21cb | 66923707 | null | System Integration and Control in a Speech Understanding System | T w o i m p a r t a n t P r o b l a q 8 i n S p e e c h u n d e r s t a n d i q g a r c haw t o afteotlvtly integrate multiple & o u r c t s of K ~o w 1 e d g . e w i t h i n t h e | {
"name": [
"Paxton, William H. and",
"Robinson, Ann E."
],
"affiliation": [
null,
null
]
} | null | null | null | 1975-11-01 | 0 | 10 | null | null | u b t a g K~I Laak f o r a VP NP o r l o o k f a r a VERB, PriarLtLrs f a r b o t h these t a s k s are computed and t h e y are p u t on t h e t a $ k queue t o b e P r o C @ $ S a d r Tho executive t h e n r e m o v e s t h e n e x t task f r o v t h e queue and continues, In Penepalc d e c l d i n $ which t a s k to perform is o f great Importance, b c c a u r e o n l y a subset o t t h e scncduLed t a w s w i l l actUallY Prove r o be neeassarY t o U n d e r s t a n d t h e i n g u t ; t h e ethers w i l l b e * f a l s e s t e p s * leading t o d e a d a n q s , ~d e a i l y r in d c c l d i n g wh-ich task t o d o 1 t h e e x e c u t i v e w o u l d always Choose one o f t h e ntecsoary fa8ks and never take a f a l s e s t e p , The utterance would ba u n d a r r t o a d w i t h t h e unnecessary tasks s t i l l l e f t $h the Queue, Ta a P~r g e c h this i d e a l r t h c a c t u a l system must s p e n d rome nf i t s e f f o r t I n c h o d a i n g t a s k s , S u c h o f f o r g i s well spent i f i t producca a net decrease in processing t i m e , In o t h e r Wordrr t h e etticitncY of the system w i l l b~ 1rnPcoVed by deci$ionn regarding t h e order i n w h i c h t a s k s are p e r f o t m e d , i f t h e c o s t a!! t h e decirlons l a loss than the c o a t of t h e f a l s e | null | null | null | Main paper:
:
u b t a g K~I Laak f o r a VP NP o r l o o k f a r a VERB, PriarLtLrs f a r b o t h these t a s k s are computed and t h e y are p u t on t h e t a $ k queue t o b e P r o C @ $ S a d r Tho executive t h e n r e m o v e s t h e n e x t task f r o v t h e queue and continues, In Penepalc d e c l d i n $ which t a s k to perform is o f great Importance, b c c a u r e o n l y a subset o t t h e scncduLed t a w s w i l l actUallY Prove r o be neeassarY t o U n d e r s t a n d t h e i n g u t ; t h e ethers w i l l b e * f a l s e s t e p s * leading t o d e a d a n q s , ~d e a i l y r in d c c l d i n g wh-ich task t o d o 1 t h e e x e c u t i v e w o u l d always Choose one o f t h e ntecsoary fa8ks and never take a f a l s e s t e p , The utterance would ba u n d a r r t o a d w i t h t h e unnecessary tasks s t i l l l e f t $h the Queue, Ta a P~r g e c h this i d e a l r t h c a c t u a l system must s p e n d rome nf i t s e f f o r t I n c h o d a i n g t a s k s , S u c h o f f o r g i s well spent i f i t producca a net decrease in processing t i m e , In o t h e r Wordrr t h e etticitncY of the system w i l l b~ 1rnPcoVed by deci$ionn regarding t h e order i n w h i c h t a s k s are p e r f o t m e d , i f t h e c o s t a!! t h e decirlons l a loss than the c o a t of t h e f a l s e
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.01692 | null | null | null | null | null | null | null | null |
f6ded01c1cc938b792affa5813ff688d3542f3f6 | 219307945 | null | An Approach to the Organization of Mundane World Knowledge: The Generation and Management of Scripts | stories or eng ag ing i n t h e most r u d i m e n t a r y c o n v e r s a t i o n , Much of t h e knowledge that h e a r e r s u t i l i z e to establish the background or c o n t e x t o f a s t o r y a p p e a r s t o be e p i s o d i c i n h a t u r e , distilled from many e x p e r i e n c e s i n common s i t u a t i o n s like g o i n g t o restaurants, football games and supermarkets. This paper p r e s e n t s an approach t o the r e p r e s e n t a t i o n and h a n d l i n g of this type of mundane world-knowledge based upon t h e concept of a s i t u a t i o n a l script [Schank and A b e l s o n , 19753. The application ( 1 ) See, f o r example, t h e emphasis o n t h i s area i n h he ore tical Issues in Natural Language Processing", P r o c e e d i n g s of the I n t e r d i s c i p l i n a r y workshop i n Computational Linguistics, | {
"name": [
"Cullingford, R. E."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 0 | 1 | null | J o h n was hungry. H e decided t o go t o a r e s t a u r a n t . H e w e n t t o o n e , H e sat down i n a c h a i r . A w a i t e r d i d n o t g o t o t h e t a b l e . J o h n became u p s e t . He l e f t t h e r e s t a u r a n t .SAM h a s g e n e r a t e d summary and q u e s t i o n -a n s w e r i n g o u t p u t f o r t h i s s t o r y : Summary J o h n went t o a r e s t a u r a n t and he a t e a l o b s , t e r ,A :So John could ask t h e w a i t e r f o r a meal, Q:Why d i d J o h n g o t oa r e s t a u r a n t ?A :So J o h n c o u l d e a t a meal, Q:Did t h e w a i t e r g i v e J o h n a menu?A:J o h n g o t t h e menu from the h o s t e s s . Q: Why d i d J o h n l e a vHe w a i t e d a t i t a f e w m i n u t e s , H e entereda b u s . The d r i v e r g o t t h e t i c k e t from J o h n , H e went t o a s e a t . He s a t down i n i t . H e e n t e t e d a s t a t i o n . H e p u t a t o k e n i n t h e t u r n s t i l e . H e went t o t h e platform. H e w a i t e d a t i t a few m i n u t e s . H e e n t e t e d a subway c a r . [ p a r a g r a p h i n g h a s been added t o t h e computer o u t p u t f o r ease of r e a d i n g ] I n t h e s e example s t o r i e s , SAM a n a l y z e s e a c h i n p u t s e n t e n c e i n t o a C o n c e p t u a l Dependency ( C D ) r e p r e s e n t a t i o n . and carnal r e l a t k o n s among t h e s e e v e n t s [ S c h a n k , 1 9 7 3 and 19741.( 2 ) C e r t a i n a c t i o n s l i k e d r i v i n g a c a r o r p r e p a r i n g f o o d i n v o l v e complex, l e a r n e d sensory-motor s k i l l s a s w e l l a s scr i p t a l knowledge.Such a c t i o n s a r e summarized w i t h i n a s c r i p t a s a c a u s a l r e l a t i o n t e r m i n a t i n g i n t h e c h i e f s t a t e -c h a n g e e f f e For e x a m p l e , t h e main e v e n t of t h e ' o r d e r i n g ' e v e n t i n a r e s t a u r a n t i s t h e o r d e r i n g a c t i t s e l f ; a n i n i t i a l e v e n t i s r e a d i n g t h e menu; and a final e v e n t x p e c t e d t h a n would be p r e d i c t e d i f t h e script were e n t e r e d v i a a n i n i t i a l e v e n t . For example, t h e s t o r y sequence i n i t i a t e d w i t h a summary:J o h n t o o k a t r a i n t o N e w York.sounds more natural t h a n a sequence b e g i n n i n g w i t h a n i n i t i a l e v e n t :John got on a t r a i n . w h i l e l e a v i n g t h e train, h e t i p p e d t h e c o n d u c t o r .These two f u n c t i o n s of t h e summary c o n t r e s , t w i d e l y i n t h e r a n g e of p r e d i c t i o n s they invoke. However, a d d i t i o n a l i n p u t s S c h a n k , " C a u s a l i t y and R e a s o n i n g " , T e c h n i c a l . Report No.1, I n s t i t u t o per g l i s t u d i s e m a n t i c i e c o g n i t i v i , C a s t a g n o l a , S w i t z e r l a n d , 1973. S c h a n k , " U n d e r s t a n d i n g Paragraphs", T e c h n i c a l Report N o . 6 , I n s t i t u t o per g l i studi s e m a n t i c i e c o g n i t i v i , C a s t a g n o l a , Switzerland, 1 9 7 4 , S c h a n k e t a1 1 9 7 5 --R. S c h a n k and B. Nash-Webber, 1 9 7 5 .Minsky 1974 M. Minsky, "Frame-Systems", MIT A I Memorandum, 1 9 7 4 .Rieger 1 9 7 5 C.R i e g e r , "Conceptual Memory", i n Information Processing, R. Schank (ed .) , North( 4 ) The t e x t f o r t h e r e s t a u r a n t s c r i p t , p r e s e n t l y t h e largest of t h e s c r i p t s , o c c u p i e s r o u g h l y 1 0 0 blocks of PDP-10 disk storage, or about 6 4 , 0 0 0 ASCII c h a r a c t e r s . | null | null | null | null | Main paper:
:
J o h n was hungry. H e decided t o go t o a r e s t a u r a n t . H e w e n t t o o n e , H e sat down i n a c h a i r . A w a i t e r d i d n o t g o t o t h e t a b l e . J o h n became u p s e t . He l e f t t h e r e s t a u r a n t .SAM h a s g e n e r a t e d summary and q u e s t i o n -a n s w e r i n g o u t p u t f o r t h i s s t o r y : Summary J o h n went t o a r e s t a u r a n t and he a t e a l o b s , t e r ,A :So John could ask t h e w a i t e r f o r a meal, Q:Why d i d J o h n g o t oa r e s t a u r a n t ?A :So J o h n c o u l d e a t a meal, Q:Did t h e w a i t e r g i v e J o h n a menu?A:J o h n g o t t h e menu from the h o s t e s s . Q: Why d i d J o h n l e a vHe w a i t e d a t i t a f e w m i n u t e s , H e entereda b u s . The d r i v e r g o t t h e t i c k e t from J o h n , H e went t o a s e a t . He s a t down i n i t . H e e n t e t e d a s t a t i o n . H e p u t a t o k e n i n t h e t u r n s t i l e . H e went t o t h e platform. H e w a i t e d a t i t a few m i n u t e s . H e e n t e t e d a subway c a r . [ p a r a g r a p h i n g h a s been added t o t h e computer o u t p u t f o r ease of r e a d i n g ] I n t h e s e example s t o r i e s , SAM a n a l y z e s e a c h i n p u t s e n t e n c e i n t o a C o n c e p t u a l Dependency ( C D ) r e p r e s e n t a t i o n . and carnal r e l a t k o n s among t h e s e e v e n t s [ S c h a n k , 1 9 7 3 and 19741.( 2 ) C e r t a i n a c t i o n s l i k e d r i v i n g a c a r o r p r e p a r i n g f o o d i n v o l v e complex, l e a r n e d sensory-motor s k i l l s a s w e l l a s scr i p t a l knowledge.Such a c t i o n s a r e summarized w i t h i n a s c r i p t a s a c a u s a l r e l a t i o n t e r m i n a t i n g i n t h e c h i e f s t a t e -c h a n g e e f f e For e x a m p l e , t h e main e v e n t of t h e ' o r d e r i n g ' e v e n t i n a r e s t a u r a n t i s t h e o r d e r i n g a c t i t s e l f ; a n i n i t i a l e v e n t i s r e a d i n g t h e menu; and a final e v e n t x p e c t e d t h a n would be p r e d i c t e d i f t h e script were e n t e r e d v i a a n i n i t i a l e v e n t . For example, t h e s t o r y sequence i n i t i a t e d w i t h a summary:J o h n t o o k a t r a i n t o N e w York.sounds more natural t h a n a sequence b e g i n n i n g w i t h a n i n i t i a l e v e n t :John got on a t r a i n . w h i l e l e a v i n g t h e train, h e t i p p e d t h e c o n d u c t o r .These two f u n c t i o n s of t h e summary c o n t r e s , t w i d e l y i n t h e r a n g e of p r e d i c t i o n s they invoke. However, a d d i t i o n a l i n p u t s S c h a n k , " C a u s a l i t y and R e a s o n i n g " , T e c h n i c a l . Report No.1, I n s t i t u t o per g l i s t u d i s e m a n t i c i e c o g n i t i v i , C a s t a g n o l a , S w i t z e r l a n d , 1973. S c h a n k , " U n d e r s t a n d i n g Paragraphs", T e c h n i c a l Report N o . 6 , I n s t i t u t o per g l i studi s e m a n t i c i e c o g n i t i v i , C a s t a g n o l a , Switzerland, 1 9 7 4 , S c h a n k e t a1 1 9 7 5 --R. S c h a n k and B. Nash-Webber, 1 9 7 5 .Minsky 1974 M. Minsky, "Frame-Systems", MIT A I Memorandum, 1 9 7 4 .Rieger 1 9 7 5 C.R i e g e r , "Conceptual Memory", i n Information Processing, R. Schank (ed .) , North( 4 ) The t e x t f o r t h e r e s t a u r a n t s c r i p t , p r e s e n t l y t h e largest of t h e s c r i p t s , o c c u p i e s r o u g h l y 1 0 0 blocks of PDP-10 disk storage, or about 6 4 , 0 0 0 ASCII c h a r a c t e r s .
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.001692 | null | null | null | null | null | null | null | null |
3fbd873ce8ab3d89d00cf4764e7ac4c6c8a01657 | 219305028 | null | Developing a Computer System to Handle Inherently Variable Linguistic Data | t y of t h e Wes t I n d i es St. A u g u s t i h e , T r 2 n i d a d | {
"name": [
"Beckles, D. and",
"Carrington, L. and",
"Warner, G. and",
"Borely, C. and",
"Knight, H. and",
"Aquing, P. and",
"Marquez, J."
],
"affiliation": [
null,
null,
null,
null,
null,
null,
null
]
} | null | null | null | 1975-11-01 | 3 | 1 | null | null | The design and some r e s u l t s ~f the research t o which t h e computer system r e l a t e s a r e described by Carrington, Borely and Knight (1969, 1972, 1974 a + b) . Part of t h e intention of t h e p r o j e c t i s t o describe i n terms Camp (1971) and Bickerton (1973) t o r e f e r to apparently analagous s i t u a t i o n s i n Jamaica and Guyana. In addition t o Creole, English and v a r i a n t s of both, a large p a r t of t h e population i s exposed t o a l o c a l A preliminary examination of t h e d a t a shows t h a t a t t h e l e v e l of phrase-structure of utterances; t h e s t r u c t u r e s w i l l appear t o be predominantly i d e n t i c a l with English. I t is the components of the elements, t h e i r meanings and functions t h a t w i l l show t h e differences from English.Consequently, t h e analysis mst n o t e t h e l e v e l s a t which derivational t r e e s cease t o be compatible w i t h English.In view of t h e v a r i a b i l i t y inherent in t h e data, t h e analysis must As exemplified a t 1.1.1, t h e last node of each sub-part s t a t e s t h e a c t u a l l i t e r a l being described. The a c c e p t a b i l i t y of t h e i t e m as IAE i s n o t e d , O K or NOK,together with a reasonable IAE a l t e r n a t i v e . Apart from t h e obligatory i n f o r m a t i o n requifed by the procedure, the analyst may make additional comments which may be either i n keywords o r E n g l i s h . e.g. CMNT: probably idiosyncratic or CMNT: double NEG. 9.flS; 8.1 SIMP; 9.2 DEC; 8.3 S W ; 9.4 MC + TAG; 9.5 NA; 8.6 NA; 9.7 AFM ACTV 1.9 MC + SUM + PRED + DOBJ -lexical, MASC -masculine, MC -main clause, N -noun, NCO -countable noun, NEQV -no equivalent, NEUT -n e t r a l , P -phrase PADJ -possessive adjective, PATT -pattern, PL -p l u r a l , PLZR -p l u r a l i z e r PRED -predicator, PREP -preposition, PRMD -pre-head modifier, PROG -progressive, RD -t h i r d person, SG -singular, S U B J -subject, TEMP -temporal, TM -time, TRAN -t r a n s i t i v e , VBL -verbal,VTverb used t r a n s i t i v e l y *a l t e r n a t i v e parse o r meaning, @absence of. . . , [ ] enclose l i t e r a l s , end of inSonnation s e t , ,minor separator.The strucfure of t h e parse t r e e i s , i n general, quite complex and a simple ad hoc approach t o v a l i d i t y checking was quickly seen t o be inadequate. For any t r e e , each analysis s t a r t s a t the root and many of the tasks t o be described below may be regarded, i n p a r t , as a p a t t e r n matching | null | null | null | Main paper:
:
The design and some r e s u l t s ~f the research t o which t h e computer system r e l a t e s a r e described by Carrington, Borely and Knight (1969, 1972, 1974 a + b) . Part of t h e intention of t h e p r o j e c t i s t o describe i n terms Camp (1971) and Bickerton (1973) t o r e f e r to apparently analagous s i t u a t i o n s i n Jamaica and Guyana. In addition t o Creole, English and v a r i a n t s of both, a large p a r t of t h e population i s exposed t o a l o c a l A preliminary examination of t h e d a t a shows t h a t a t t h e l e v e l of phrase-structure of utterances; t h e s t r u c t u r e s w i l l appear t o be predominantly i d e n t i c a l with English. I t is the components of the elements, t h e i r meanings and functions t h a t w i l l show t h e differences from English.Consequently, t h e analysis mst n o t e t h e l e v e l s a t which derivational t r e e s cease t o be compatible w i t h English.In view of t h e v a r i a b i l i t y inherent in t h e data, t h e analysis must As exemplified a t 1.1.1, t h e last node of each sub-part s t a t e s t h e a c t u a l l i t e r a l being described. The a c c e p t a b i l i t y of t h e i t e m as IAE i s n o t e d , O K or NOK,together with a reasonable IAE a l t e r n a t i v e . Apart from t h e obligatory i n f o r m a t i o n requifed by the procedure, the analyst may make additional comments which may be either i n keywords o r E n g l i s h . e.g. CMNT: probably idiosyncratic or CMNT: double NEG. 9.flS; 8.1 SIMP; 9.2 DEC; 8.3 S W ; 9.4 MC + TAG; 9.5 NA; 8.6 NA; 9.7 AFM ACTV 1.9 MC + SUM + PRED + DOBJ -lexical, MASC -masculine, MC -main clause, N -noun, NCO -countable noun, NEQV -no equivalent, NEUT -n e t r a l , P -phrase PADJ -possessive adjective, PATT -pattern, PL -p l u r a l , PLZR -p l u r a l i z e r PRED -predicator, PREP -preposition, PRMD -pre-head modifier, PROG -progressive, RD -t h i r d person, SG -singular, S U B J -subject, TEMP -temporal, TM -time, TRAN -t r a n s i t i v e , VBL -verbal,VTverb used t r a n s i t i v e l y *a l t e r n a t i v e parse o r meaning, @absence of. . . , [ ] enclose l i t e r a l s , end of inSonnation s e t , ,minor separator.The strucfure of t h e parse t r e e i s , i n general, quite complex and a simple ad hoc approach t o v a l i d i t y checking was quickly seen t o be inadequate. For any t r e e , each analysis s t a r t s a t the root and many of the tasks t o be described below may be regarded, i n p a r t , as a p a t t e r n matching
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.001692 | null | null | null | null | null | null | null | null |
37c109dfbedd3f6b9ea9c5e73aa8e3eb3e0248b9 | 219306806 | null | {SPS}: A Formalism for Semantic Interpretation and its Use in Processing Prepositions that Reference Space | I t was originally developed t o demonstrate how English prepositions, such as "up", "down", and "through", w h i c h r e f e rence location, motion, and orientation i n space could be semantically , 33, (1 9 6 8 ) , 457-471 . 9. Woods, William A., "Progress i n Natural language Understanding -An Appl i cation t o Lunar Geology", AFIPS ConLrence Proceedings , 42, (1 973), | {
"name": [
"Sondheimer, Norman K. and",
"Perry, Doyt"
],
"affiliation": [
null,
null
]
} | null | null | null | 1975-11-01 | 10 | 3 | null | null | t o note t h a t our results t o date tend t o indicate the need for a level o f a h g t r a c t i o n somewhere b e t~e e n S i m n 'IS and Schank's semantic nets.I n developing the semantic lewl-, we are t r y i n g t o make i t t h e one where "general knowledge o f language and i t s relation t o the world" i s applied. This i s in contrast t o the pragmatic level, where situationspecific information i s used to interpret the semantic structures.In sumnary, a system employing SPS would construct syntactic t r e e s , use SPS for the production of Case structures, and employ a pragmatic processing scheme t o interpret these structures.Problems in Processing Locative Prepositions. Part of t h e problem with the semantic interpretation of locatives i s the complexity of the structures necessary t o represent them on the underlying syntactic and semantic levels. ' This section discusses these problems and introduces our semantic structure notation.The representation of locative prepositional meaning in Case structures has been problematic. The number o f cases t h a t Fillmore has postulated for them has risen t o four--Location, Source, Goal, P a t h . He a1 so features Both o f the locati* phrases w~u l d be assigned the same case -Location.Howeverj they actual ly locate different objects. Bi 1 7 ' s daughter was said to be on his lap while both o f them were said t o be in the tunnel. Similarly, the use o f an unordered s e t of cases f a i l s t o a1 l o w for the difference in meaning of the following twosentences, where the f i r s t two prepositional phrases i n each would be i n the Path case: '!He went down the h i 1 1 across the bridge t o the chapel.", and "He went across the bridge down the h i l l toI I t h e chapel. .The Case representation we are using deals w i t h these problems. This representation uses only one case for a l l spattal references. This case, the Placecase, identifies spaces which derive from the location of participants i.n i t s action, event, qr s t a t e of a f f a i r s (or event/state). Which participants and how each space relates t o them depends on the type o f event/state, The basic structure of the assertional notation can be seen by showing how a Place case wul d be represented: ( :PLACE #E/S $PO). The " : "II I I identifies a relation, the # an event/state, and the "8" objects (note that many o f these will be replaced by variables i n the actual assertions produced). The f i r s t element o f any assertion i s always a relation, which forces interpretations on t h e other elenients. W i t h the relation :PLACE, the l a s t two elements must be references t o an eventlstate and a spatial object (space), i n that order. The specific spatial objects t h a t are referred in Place assertions are c a l l ed Pl ace objects.The prepositional elements on the semantic level can relate Place objects directly. An example of this is the representation o f "She died away from where she 1 ived.", i . e . , (:PLACE #E/Sl $PO1 ) (:AWAYFROM $PO1 $PO?) (:PLF\CE iiE/SZ bP(12). here a prepositional element relates the Place object of the two event/states corresponding t o "she died" and "she 1 ived".Prepositional elements can also r e l a t e spaces derived from Place objects. (:FINAL $XI10 $X112) (:INITIAL $XI11 SX113) (:LE $XI12 $X113).The Place case proposal avoids problems 1 i ke that w i t h the Location case exarnpl e, through the representation o f certain syntactically simp1 e clauses with more t h a n one event/state. The representation o f "He held her on his lap i n the tunnel ." shows an event/state corresponding to "he held her" and one corresponding t o "she was on h i s l a p " . This complex structure solves the case problems by a1 lowing each preposition to predicate a different Place object. "On his lap" predicates afi existent i a l event/state showing where t h e female was located. "In the tunnel " can predicate the Place object of the causative event/state. The interpretation that space i s t h a t i t i s composed from the Place objects of i t s two constituent eventlstates. Hence, both peopl e w i 11 be predicated by i t.While these l a s t two devices enable us t o avoid representational problems, i t should, of course, be remembered that semantic interpretation must support these forms.* To summarize t h i s section has presented a variety of points about the semantic interpretation of locative pr'eposi tions-that they can require complex case representations, and t h a t they appear i n a variety of syntactic environments. SPS has been designed to r e l a t e the syntactic to the semantic *There are o t h e r phenohena for which the Place case proposal a1 lows. The co' mpl ete representation is descri bed el sewhere.6 What has been given here i s enough to show the difficulty of interpretation. environment of locative prepositions. How i t deals with these problems will be described a f t e r a brief over vie^ of the formal ism, SPS. The SPS formalism i s m s t closely related t o a fam'ily of semantic interpretatian schemes deriving from Woods' 1968 The close similarity t o that work 1 ies i n the basic form of rules. These rules have the form "pattern + action", where the pattern side specifies t e s t s to be made on the syntactic structures, and the action side specifies forms to be added explain how i t allows for locative prepositions we look a t a typical rule:Rul e 2-STAT-LO:((*I-S5 (1 2 . 3 4) 1 ( 4 ) *1-S7 ( ( EQ #2STAT 1-1) (COMPATIBLE 1-1 2-1) (COMPATIBLE 1-2 OBJ(I-i 1) (COMPATIBLE R(SS) SUBJ(I-1 ) ) ) I ======+ (((:PLACE R(CAUSED) !X(1 ) ) (1-1 !X(1) ! X(2)) (:PRED !X(3) $BE) (:OBJ !X(3) !l-2) (:PLACE !X(3) ! X ( 2 ) ) ) ) )This i s a rule that might be applied to interpret the prepositional phrase in variables representing some event/states or objects. The purpose of the rul e i s to r e l a t e the location of the object being held to the location o f the complement. These locations are available through event/states which identify where each of the t w o objects were. We use the predicator $BE f o r these event/states, such as i n t h e one for "his l a p " which i s produced by the rule. Hbw the correct assertions are produced from the assertional forms i s i l l u s t r a t e d in the above rule.or $ a r e inserted directly. The number-dash-number forms provide a reference to a 1 i t e r a l stoAd in a lexical entry . For prepositions t h i s 1 i t e r a l gives the physical relation t h a t the term refers to.The two Place objects are formed by the use o f a variable generation feature using the " !XIt'-number-")" form. References t o the $BE event/state are also formed i n t h i s way. The other event/state i s referenced through a register. SPS allows registers t o hold variable names as well as feature sets. The register used here must be s e t with the variable name used when the event/s tate was const~ucted.As the above example shows, the registers are used h e r e in situations where mare than one event/state results from a clause. When only one event/ s t a t e e x i s t s , a simple reference t o the major covlsti tuents o f a sentence i s necessaty. SPS allows for t h i s by automaticall v associating variables with the S and NP nodes i n trees. These a r e referenced through forms l i k e " ! I -2 " which here gets the variable associated w i t h "his lap" (presumably ! X 4 ) .This variable will also appear in the assertions describing the object; The 1 i t e r a l s on the arcs name rules that must be successfully applied be'fore a s t a t e change can occur. These nets are set u p for noun phrase and sentent i a l elements, and are used with a marking scheme such that interpretation, A formal ism f o r writing semantic interpreters, SPS, has been described, I t alfows for a semantic feature scheme that can describe the restrictions on locative prepositions. SPS also has registers t h a t can be used for these restrictions and for building up the case structures that represent the meadings of locatives. A rule-ordering scheme i s also heloful here. I t car] be said t h a t SPS i s a good vehicle for dterpreting locative prepositions, and t h a t any system for semantic %re detail on a somewhat earl i e r version of YS can be found i n Chapter VII o f [6] . | null | null | null | Main paper:
:
t o note t h a t our results t o date tend t o indicate the need for a level o f a h g t r a c t i o n somewhere b e t~e e n S i m n 'IS and Schank's semantic nets.I n developing the semantic lewl-, we are t r y i n g t o make i t t h e one where "general knowledge o f language and i t s relation t o the world" i s applied. This i s in contrast t o the pragmatic level, where situationspecific information i s used to interpret the semantic structures.In sumnary, a system employing SPS would construct syntactic t r e e s , use SPS for the production of Case structures, and employ a pragmatic processing scheme t o interpret these structures.Problems in Processing Locative Prepositions. Part of t h e problem with the semantic interpretation of locatives i s the complexity of the structures necessary t o represent them on the underlying syntactic and semantic levels. ' This section discusses these problems and introduces our semantic structure notation.The representation of locative prepositional meaning in Case structures has been problematic. The number o f cases t h a t Fillmore has postulated for them has risen t o four--Location, Source, Goal, P a t h . He a1 so features Both o f the locati* phrases w~u l d be assigned the same case -Location.Howeverj they actual ly locate different objects. Bi 1 7 ' s daughter was said to be on his lap while both o f them were said t o be in the tunnel. Similarly, the use o f an unordered s e t of cases f a i l s t o a1 l o w for the difference in meaning of the following twosentences, where the f i r s t two prepositional phrases i n each would be i n the Path case: '!He went down the h i 1 1 across the bridge t o the chapel.", and "He went across the bridge down the h i l l toI I t h e chapel. .The Case representation we are using deals w i t h these problems. This representation uses only one case for a l l spattal references. This case, the Placecase, identifies spaces which derive from the location of participants i.n i t s action, event, qr s t a t e of a f f a i r s (or event/state). Which participants and how each space relates t o them depends on the type o f event/state, The basic structure of the assertional notation can be seen by showing how a Place case wul d be represented: ( :PLACE #E/S $PO). The " : "II I I identifies a relation, the # an event/state, and the "8" objects (note that many o f these will be replaced by variables i n the actual assertions produced). The f i r s t element o f any assertion i s always a relation, which forces interpretations on t h e other elenients. W i t h the relation :PLACE, the l a s t two elements must be references t o an eventlstate and a spatial object (space), i n that order. The specific spatial objects t h a t are referred in Place assertions are c a l l ed Pl ace objects.The prepositional elements on the semantic level can relate Place objects directly. An example of this is the representation o f "She died away from where she 1 ived.", i . e . , (:PLACE #E/Sl $PO1 ) (:AWAYFROM $PO1 $PO?) (:PLF\CE iiE/SZ bP(12). here a prepositional element relates the Place object of the two event/states corresponding t o "she died" and "she 1 ived".Prepositional elements can also r e l a t e spaces derived from Place objects. (:FINAL $XI10 $X112) (:INITIAL $XI11 SX113) (:LE $XI12 $X113).The Place case proposal avoids problems 1 i ke that w i t h the Location case exarnpl e, through the representation o f certain syntactically simp1 e clauses with more t h a n one event/state. The representation o f "He held her on his lap i n the tunnel ." shows an event/state corresponding to "he held her" and one corresponding t o "she was on h i s l a p " . This complex structure solves the case problems by a1 lowing each preposition to predicate a different Place object. "On his lap" predicates afi existent i a l event/state showing where t h e female was located. "In the tunnel " can predicate the Place object of the causative event/state. The interpretation that space i s t h a t i t i s composed from the Place objects of i t s two constituent eventlstates. Hence, both peopl e w i 11 be predicated by i t.While these l a s t two devices enable us t o avoid representational problems, i t should, of course, be remembered that semantic interpretation must support these forms.* To summarize t h i s section has presented a variety of points about the semantic interpretation of locative pr'eposi tions-that they can require complex case representations, and t h a t they appear i n a variety of syntactic environments. SPS has been designed to r e l a t e the syntactic to the semantic *There are o t h e r phenohena for which the Place case proposal a1 lows. The co' mpl ete representation is descri bed el sewhere.6 What has been given here i s enough to show the difficulty of interpretation. environment of locative prepositions. How i t deals with these problems will be described a f t e r a brief over vie^ of the formal ism, SPS. The SPS formalism i s m s t closely related t o a fam'ily of semantic interpretatian schemes deriving from Woods' 1968 The close similarity t o that work 1 ies i n the basic form of rules. These rules have the form "pattern + action", where the pattern side specifies t e s t s to be made on the syntactic structures, and the action side specifies forms to be added explain how i t allows for locative prepositions we look a t a typical rule:Rul e 2-STAT-LO:((*I-S5 (1 2 . 3 4) 1 ( 4 ) *1-S7 ( ( EQ #2STAT 1-1) (COMPATIBLE 1-1 2-1) (COMPATIBLE 1-2 OBJ(I-i 1) (COMPATIBLE R(SS) SUBJ(I-1 ) ) ) I ======+ (((:PLACE R(CAUSED) !X(1 ) ) (1-1 !X(1) ! X(2)) (:PRED !X(3) $BE) (:OBJ !X(3) !l-2) (:PLACE !X(3) ! X ( 2 ) ) ) ) )This i s a rule that might be applied to interpret the prepositional phrase in variables representing some event/states or objects. The purpose of the rul e i s to r e l a t e the location of the object being held to the location o f the complement. These locations are available through event/states which identify where each of the t w o objects were. We use the predicator $BE f o r these event/states, such as i n t h e one for "his l a p " which i s produced by the rule. Hbw the correct assertions are produced from the assertional forms i s i l l u s t r a t e d in the above rule.or $ a r e inserted directly. The number-dash-number forms provide a reference to a 1 i t e r a l stoAd in a lexical entry . For prepositions t h i s 1 i t e r a l gives the physical relation t h a t the term refers to.The two Place objects are formed by the use o f a variable generation feature using the " !XIt'-number-")" form. References t o the $BE event/state are also formed i n t h i s way. The other event/state i s referenced through a register. SPS allows registers t o hold variable names as well as feature sets. The register used here must be s e t with the variable name used when the event/s tate was const~ucted.As the above example shows, the registers are used h e r e in situations where mare than one event/state results from a clause. When only one event/ s t a t e e x i s t s , a simple reference t o the major covlsti tuents o f a sentence i s necessaty. SPS allows for t h i s by automaticall v associating variables with the S and NP nodes i n trees. These a r e referenced through forms l i k e " ! I -2 " which here gets the variable associated w i t h "his lap" (presumably ! X 4 ) .This variable will also appear in the assertions describing the object; The 1 i t e r a l s on the arcs name rules that must be successfully applied be'fore a s t a t e change can occur. These nets are set u p for noun phrase and sentent i a l elements, and are used with a marking scheme such that interpretation, A formal ism f o r writing semantic interpreters, SPS, has been described, I t alfows for a semantic feature scheme that can describe the restrictions on locative prepositions. SPS also has registers t h a t can be used for these restrictions and for building up the case structures that represent the meadings of locatives. A rule-ordering scheme i s also heloful here. I t car] be said t h a t SPS i s a good vehicle for dterpreting locative prepositions, and t h a t any system for semantic %re detail on a somewhat earl i e r version of YS can be found i n Chapter VII o f [6] .
Appendix:
| null | null | null | null | {
"paperhash": [
"rumelhart|active_semantic_networks_as_a_model_of_human_memory",
"r.|two_semantic_networks_:_their_computation_and_use_for_understanding_english_sentences"
],
"title": [
"Active Semantic Networks as a Model of Human Memory",
"Two Semantic Networks : Their Computation and Use for Understanding English Sentences"
],
"abstract": [
"A g e n e r a l system to s i m u l a t e human c o g n i t i v e p r o cesses is d e s c r i b e d . The f o u r p a r t system compr ises a nodespace to s t o r e t he ne twork s t r u c t u r e ; a s u p e r v i s o r ; a t r a n s i t i o n network p a r s e r ; and an i n t e r p r e t e r . The method by wh ich noun phrases ope ra te and t h e p rocess f o r t he d e t e r m i n e r \" t h e \" i s p r e s e n t e d . A n a n a l y s i s o f ve rb s t r u c t u r e s i l l u s t r a t e s how network s t r u c t u r e s can b e c o n s t r u c t e d f rom p r i m i t i v e ve rb d e f i n i t i o n s t h a t ge t a t t h e u n d e r l y i n g s t r u c t u r e s o f p a r t i c u l a r v e r b s . The paper conc ludes w i t h an i l l u s t r a t i o n o f a p rob lem i n q u e s t i o n a s k i n g . A Model of Human Memory We have c o n s t r u c t e d a l a r g e g e n e r a l s i m u l a t i o n of human language and l o n g t e r m memory on t h e p remise t h a t t h e s t u d y o f t he i n t e r r e l a t i o n s h i p s among p s y c h o l o g i c a l p rocesses w i l l l ead t o more i n s i g h t i n t o human cog n i t i o n and memory. The g e n e r a l i m p l e m e n t a t i o n i s ba s i c a l l y c o m p l e t e , and a v a r i e t y o f use rs a r e s t a r t i n g t o s t u d y s p e c i f i c p s y c h o l o g i c a l t a s k s ( language under s t a n d i n g ; c h i l d r e n ' s development o f l anguage ; p r i m i t i v e v e r b s t r u c t u r e ; r e a d i n g ; i n f e r e n c e ; game p l a y i n g G o and Gomoku; v i s u a l r e p r e s e n t a t i o n and memory; l e a r n i n g ; and q u e s t i o n a n s w e r i n g ) . I t i s s t i l l too e a r l y t o r e p o r t o n t h e r e s u l t s o f t h e p s y c h o l o g i c a l i n v e s t i g a t i o n . . T h e r e f o r e , t h i s paper i s a p r o g r e s s r e p o r t o n t h e s y s tem and t h e u n d e r l y i n g p s y c h o l o g i c a l p r i n c i p l e s . The ma jo r g u i d e l i n e s have come f rom our a t t e m p t s to r e p r e s e n t l o n g t e r m memory s t r u c t u r e s . We know t h a t peop le r a p i d l y f o r g e t t h e d e t a i l s about t h e s u r f a c e s t r u c t u r e o f a n e x p e r i e n c e b u t r e t a i n t h e meaning o r i n t e r p r e t a t i o n o f t h a t e x p e r i e n c e i n d e f i n i t e l y . W e a l so know t h a t r e t r i e v a l o f an e x p e r i e n c e f rom memory i s u s u a l l y a r e c o n s t r u c t i o n wh ich i s h e a v i l y b i a s e d b y t h e p e r s o n ' s g e n e r a l knowledge o f t h e w o r l d . Thus , g e n e r a l w o r l d knowledge shou ld i n t e r a c t w i t h s p e c i f i c event knowledge in such a way t h a t d i s t i n c t i o n between the two i s n o t p o s s i b l e . The r e p r e s e n t a t i o n shou ld a l l o w p a r a p h r a s e . F i n a l l y , t h e l i m i t a t i o n s o f human w o r k i n g s t o r a g e ( o r s h o r t t e r m memory) p r o b a b l y compr i se a f u n damenta l p r o p e r t y o f t h e sys tem, one t h a t shou ld be v iewed as an e s s e n t i a l , p o s i t i v e component , n o t as s imp l y a per fo rmance l i m i t a t i o n .",
"In the world of computer science, networks are mathematical and computational structures composed of sets of nodes connected by directed arcs. A semantic network purports to represent concepts expressed by natural-language words and phrases as nodes connected to other such concepts by a particular set of arcs called semantic relations. Primitive concepts in this system of semantic networks are word-sense meanings. Primitive semantic relations are those that the verb of a sentence has with its subject, object, and prepositional phrase arguments in addition to those that underlie common lexical, classificational and modificational relations. A complete statement of semantic relations would include all those relations that would be required in the total classification of a natural language vocabulary. We consider the theory and model of semantic nets to be a computational theory of superficial verbal understanding in humans. W e conceive semantic nodes as representing human verbal concept structures and semantic relations connecting two such structures as representing the linguistic processes of thought that are used to combine them into natural-language descriptions of events. Some psycholin-guistic evidence supports this theory"
],
"authors": [
{
"name": [
"D. Rumelhart",
"D. Norman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"F. R."
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null
],
"s2_corpus_id": [
"2895638",
"14866811"
],
"intents": [
[],
[]
],
"isInfluential": [
false,
false
]
} | null | 591 | 0.005076 | null | null | null | null | null | null | null | null |
65fa97c765bdd2fdd5ac7db9f74a8d1336ad8c59 | 53904025 | null | Semantic-Based Parsing and a Natural-Language Interface for Interactive Data Management | We describe a natural-language recognition system having both applied and theoretical relevance. A t the applications level, the prwram w i l l give a natural ccmmunications interface facility to users of existing i n t e r a c t i v e data management systems. A t the t h e o r e t i c a l l e v e l , our work shows t h a t the useful infoxmation i n a natural-language expression (its "meaning") can be obtained by an algorithm t h a t uses no formal description of synt-. The construction of the parsing tree is c o n t r o l l e d primarily by semantics i n the form of an abstraction of the nmicxo-world" of the DMS's f u n c t i o n a l capabilities and the organizat~on and semantic relations of t h e data base content material. A prototype is c u r r e n t l y implemented in LTSP 1.5 on tho IBM 370/145 computsr at System Development Corporation. In a recent article in S c i e n t i f i c , American, Dr. Alphonse Chapanis says, "Tf t r u l y interactive computer ( ; y s t m are ever to be created, they will ~omehow have to cope w i t h the... errors and v i o l a t i o n s of format t h a t a r e the rule rather than the exception in normal human ccmmunication" [1] . A n example dialogue produced by t w a persons interacting w i t h each other by teletypewriter to solve a problem as~igned to them by experimenters showed that :not one grernaaatfcally correct sentence appears in t h e entire protocol. tl Many existing language pmcessors (woods, Kellogg , Thcmpson , e t c . ) [ 2,3,4) a r e limited to what Chapanis calls "Irmnaculate prose," that i s , "the sentences that are fed into the computer are parsed in one way or another so that the m e a n i n g of the ensemble can be inferred frm conventional rules of syntax," which are a £ 0 -d e s c r i p t i o n of the language. In effect, users are required to i n t e r a c t w i t h these s y s t e m in s m e formal language, or at l e a s t i n a language that has a formal representation i n the computer system t h a t a user's expression must conform to (we are t h i n k i n g , in t h e latter instance, of Vhampsonls REL, which has an extensible formal representation facility). In a d d i t i o n , most natural-language question-answering systems, including all referenced above, require that a user's data be restruct-wedl and reorganized acwraing t o the p a r t i c u l a r data base requirements of the natural-language system to be used. A t the level of a r t i f i c i a l i n t e l l i g e n c e research [ti ,6 ,?'I , Mere is same interest in systems that recognize meaning i n natural-language expressions by methods that dd not m i r e compiler-like syntactic analysi~ of an expression prior to asmantic interpretation. We believe it is possible, practical, and feasible, using new lingufstic processing strategies, to design a natural-language interface system that will permit flexible, intuitive coaansmicatiba w i t h information management systems and other computer programs already in existence. This interface is open-ended in that it has I. Chapanis, Alphonse. Interactive human cammunlcation, Scientific American, May, 1975. 2. Woods, W. A, Trahsition network gr-ars for natural language analysis. | {
"name": [
"Burger, John F. and",
"Leal, Antonio and",
"Shoshani, Arie"
],
"affiliation": [
null,
null,
null
]
} | null | null | null | 1975-11-01 | 3 | 5 | null | null | no prejudice about t h e user's system funckians and can be joined to almost any such system with relatively l i t t l e effort. I t i s , i n addition, able to infer t h e meaning of free-form English expressions, as they pertain to the host system, without requiring any formal description or representation of English.The syntactic inflexibiiity of existing natural-language processors limits their usefulness i n interactive man-madine tasks. O u r approach does not use a collection of syntax rules or equations as they are normally defined.Instead, we construct a dictionary in which w e define words in terms of their possible meanings with respect to the particular data base and data management system (DMS) we want to use and according to the possible relations t h a t can exist between data-base and I3MS elements ( e . g . , an averaging funct i o n on a group CKE numbers) i n the limited "micro-world" of this precisely organized data collection. Words appearing in a user's expression t h a t a r e not explicitly defined are ignored by the system i n processing the expression; an example would be t h e word "the," which is usually not meaningful in a data management environment. Wa thus avoid the expressive rigidity that formal syntactic methods hposa on tha user and the excesaivcs time and resource consumption t h a t results from the catibinatorial explosions usually produced by such rnethade.We distinguish in their d e f i n i t i o n s beween two types of words: content words m d function w o r b (or "operatore"). Content words are w a d s whoae 'meaningsw are the objects, events, and concepts that make up the s u b j e c t s being referred t o by users, More p r e c i s e l y , for data axetnagernent systems, these meanings (or "concepts") are the f i e l d names and entz'y i d e n t i f i e r s f o r *e data b-e and the names for available IHS operations such as averaging, s d n g , sorting, comparing, etc. Function words serve as connectors of content words. Their use i n natural language i s to indicate khe manner in which neighboring conltent words ar'e intended to relate to one another. In the example "the salary of the secretary ," used belaw, "salary" and "secretary," are content words, and "of" is a function word used to connect theta. To get a more i n t u i t i v e understanding of this process, suppose, again, t h a t a data base contains e n t r i e s for both secretaries and clerks w i t h salaries fox each. Suppose "Suzi&' is an instance of a secretary and om" is an instance of a clerk. We then have three words defined as follms:Suzie ( (SUZIE SECY) ) Torn ( (TOM C-LK) ) Salary ( ( sECY SECSAL) (CLK CLKSAL) )Processing me phrase "Suzie ' s salary" would i n t e r s e c t the Y i ( " (SECY) " ) from t h e d e f i n i t i o n of "Suzie" w i t h t h e Xi's ("SECY" and "CLK") from t h e definition of "salary." The intersection is nan-empty ("(SECY)") , and, i n discovering the semantic relationship the sense "SECSALI-' is assigned t o the word "salary." Similarly, "Tan's salary" assigns t h e sense "CLKSAL" t o "salary. !Ioperates for a p a r t i c u l a r DMS/data-base t a r g e t system. It contains a particular &&ionc r e a t e d for t h a t t a r g e t system. For a p a r t i c u l a r dic- In the analysis of a particular input by our system, two words i n context a r e t e~t e d using t h e "intersection" method described abave and, if they are found to be semantically r e l a t e d , they are considered candidates f o r "connection" as descrrLbed below. Two words so connected £ o m a phrase. Eunction words may connect content words in "positive," "negative ," or "peak" connections. me follming are examples of each mannax of connection:1. "Of" is a negative operator, as in " t h e salary of the SALARY 2." ' 8 " is a positive operator, as in "the s e c r e t a r y ' s salary":3 . "And" is a peak operator, as in "Atlantic and Pacific. . Woods, W. A, Trahsition network gr-ars for natural language analysis. | null | null | null | Main paper:
:
no prejudice about t h e user's system funckians and can be joined to almost any such system with relatively l i t t l e effort. I t i s , i n addition, able to infer t h e meaning of free-form English expressions, as they pertain to the host system, without requiring any formal description or representation of English.The syntactic inflexibiiity of existing natural-language processors limits their usefulness i n interactive man-madine tasks. O u r approach does not use a collection of syntax rules or equations as they are normally defined.Instead, we construct a dictionary in which w e define words in terms of their possible meanings with respect to the particular data base and data management system (DMS) we want to use and according to the possible relations t h a t can exist between data-base and I3MS elements ( e . g . , an averaging funct i o n on a group CKE numbers) i n the limited "micro-world" of this precisely organized data collection. Words appearing in a user's expression t h a t a r e not explicitly defined are ignored by the system i n processing the expression; an example would be t h e word "the," which is usually not meaningful in a data management environment. Wa thus avoid the expressive rigidity that formal syntactic methods hposa on tha user and the excesaivcs time and resource consumption t h a t results from the catibinatorial explosions usually produced by such rnethade.We distinguish in their d e f i n i t i o n s beween two types of words: content words m d function w o r b (or "operatore"). Content words are w a d s whoae 'meaningsw are the objects, events, and concepts that make up the s u b j e c t s being referred t o by users, More p r e c i s e l y , for data axetnagernent systems, these meanings (or "concepts") are the f i e l d names and entz'y i d e n t i f i e r s f o r *e data b-e and the names for available IHS operations such as averaging, s d n g , sorting, comparing, etc. Function words serve as connectors of content words. Their use i n natural language i s to indicate khe manner in which neighboring conltent words ar'e intended to relate to one another. In the example "the salary of the secretary ," used belaw, "salary" and "secretary," are content words, and "of" is a function word used to connect theta. To get a more i n t u i t i v e understanding of this process, suppose, again, t h a t a data base contains e n t r i e s for both secretaries and clerks w i t h salaries fox each. Suppose "Suzi&' is an instance of a secretary and om" is an instance of a clerk. We then have three words defined as follms:Suzie ( (SUZIE SECY) ) Torn ( (TOM C-LK) ) Salary ( ( sECY SECSAL) (CLK CLKSAL) )Processing me phrase "Suzie ' s salary" would i n t e r s e c t the Y i ( " (SECY) " ) from t h e d e f i n i t i o n of "Suzie" w i t h t h e Xi's ("SECY" and "CLK") from t h e definition of "salary." The intersection is nan-empty ("(SECY)") , and, i n discovering the semantic relationship the sense "SECSALI-' is assigned t o the word "salary." Similarly, "Tan's salary" assigns t h e sense "CLKSAL" t o "salary. !Ioperates for a p a r t i c u l a r DMS/data-base t a r g e t system. It contains a particular &&ionc r e a t e d for t h a t t a r g e t system. For a p a r t i c u l a r dic- In the analysis of a particular input by our system, two words i n context a r e t e~t e d using t h e "intersection" method described abave and, if they are found to be semantically r e l a t e d , they are considered candidates f o r "connection" as descrrLbed below. Two words so connected £ o m a phrase. Eunction words may connect content words in "positive," "negative ," or "peak" connections. me follming are examples of each mannax of connection:1. "Of" is a negative operator, as in " t h e salary of the SALARY 2." ' 8 " is a positive operator, as in "the s e c r e t a r y ' s salary":3 . "And" is a peak operator, as in "Atlantic and Pacific. . Woods, W. A, Trahsition network gr-ars for natural language analysis.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.00846 | null | null | null | null | null | null | null | null |
48910e83668103fc968fb2efaecb6347b169d169 | 219304222 | null | An Adaptive Natural Language Parser | \Jheh a user interacts with a natural language system, he may well use words and expressions which were not anticipated by the system designers. This paper describes a system which can play TIC-TAC-TOE, and discuss the game while it is in progress. I f the system encounters new words, new expressions, or inadvertent ungrammaticalities, it attempts t o understand what was meant, through contextual inference, and by asking i h t e l i i g e n t clarifying questions of the user. The system then records | {
"name": [
"Miller, Perry L."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 10 | 1 | null | null | null | The best way t o introduce the system is t o show i t in o p e r a t i a n .In the sample session that follows, user input is pteced~rd by 'U:", (1) to explore the techniques required t o achieve adaptive behavior, and ( 2 ) t o h e l p fornulate the issues which will have t o be faced when incorporating these techniques i n t o a much broader natural language system. An interesting aspect o f this approach is t h a t t h e clause-level syntax is entirely domain-independent. I t knows no thing about TIC-TAC-TOE, o r even about the words used t o talk about TIC-TAC-TOE. Tke surface frames allow semantics t o t a l k t o syntax purely in terms o f syntactic labels. As a result, one could write a single syntactic module, and t h a n insert i t unchanged i n t o many domains. | can be used when processing a sentence. replies t h a t t h e sentence follows normal order. Had the string been "verb obj pp" syntax would reply t h a t the subject had been deleted. I f the s t r i n g was @'do agent verb obj p p n , syntax would reply that subjectverb inversion had taken p l a c e . Given "gent obj verb ppn, syntax would reply that t h e object was out of position.For instance, if syntax r e p l i e s t h a t the object is out of position i n the clause, or t h a t there is incorrect agreement in number between subject and verb, t h e system may decide that t h e user has made a minor grammatical error, and allow the sentence t o be processed anyway, especially if there i s no better interpretation of the sentence. In this( 2 ) If a constituent is unknown:If an unknown constituent is p r e s e n t , then both the frame and slot information can be used to h e l p resolve its meaning. For i n s t a n c e , suppose the sentence is " I place a c r o s s in the canter squarew, and The system can then ask if "to plunk something somewherew means " t o place something somewheren, and upon getting an affirmative reply, can add t h e new frame to those associated w i t h the concept PLACE. These syntactic features, however, need not bs inflexible rules. Sentence understanding can still psocaed w e n i f tha syntactic features found by syntax do not exactly match those specified by the clausefunction frame. Thus, an inadvertent ungrammaticality cam readily be recognized as such, and processing can continue. (1) The number of concepts a v a i l a b l e t o the system a t present is very small. T h i s , in fact, is why the system's first guess is usually the correct one. I f the sentence is at a l l w i t h i n the systea's comprehension, t h e options as to its meaning a r e currently q u i t e limited.( 2 ) The range of expressive devices presently recognized is q u i t s limited as well. For instance, the system does n o t recognaze relative clauses, con junctions, o r pronouns (except f o r 1 and you).( 3 ) The system currently d e a l s only with TOTALLY UNFMILIAR words and expressions in this adaptive fashion, It w i l l not correctly handle familiar words which are used in new ways (such as a noun used eas a varb, as i n wzero the c e n t e r s q u a r e n ) .( 4 ) The system t r i e s to map the meaning o f new wards and expressiuns into i t s s p e c i f i e d s e t of underlying concepts. It then displays its hypotheses t o the user, g i v i n g him only the option of saying yas or nu. The user cann-ot say "no, not q u i t e , it meahs . . .". (Thus concepts like V h e 'northeast1 square" o r "the 'topmost' squarew would ba confusing and not correctly understood.) | null | Main paper:
sahple session:
The best way t o introduce the system is t o show i t in o p e r a t i a n .In the sample session that follows, user input is pteced~rd by 'U:", (1) to explore the techniques required t o achieve adaptive behavior, and ( 2 ) t o h e l p fornulate the issues which will have t o be faced when incorporating these techniques i n t o a much broader natural language system. An interesting aspect o f this approach is t h a t t h e clause-level syntax is entirely domain-independent. I t knows no thing about TIC-TAC-TOE, o r even about the words used t o talk about TIC-TAC-TOE. Tke surface frames allow semantics t o t a l k t o syntax purely in terms o f syntactic labels. As a result, one could write a single syntactic module, and t h a n insert i t unchanged i n t o many domains.
. 1 . 1 using t h i s information in t h i s s e c t i o n , we describe i n more d e t a i l how t h i s knowledge:
can be used when processing a sentence. replies t h a t t h e sentence follows normal order. Had the string been "verb obj pp" syntax would reply t h a t the subject had been deleted. I f the s t r i n g was @'do agent verb obj p p n , syntax would reply that subjectverb inversion had taken p l a c e . Given "gent obj verb ppn, syntax would reply that t h e object was out of position.For instance, if syntax r e p l i e s t h a t the object is out of position i n the clause, or t h a t there is incorrect agreement in number between subject and verb, t h e system may decide that t h e user has made a minor grammatical error, and allow the sentence t o be processed anyway, especially if there i s no better interpretation of the sentence. In this( 2 ) If a constituent is unknown:If an unknown constituent is p r e s e n t , then both the frame and slot information can be used to h e l p resolve its meaning. For i n s t a n c e , suppose the sentence is " I place a c r o s s in the canter squarew, and The system can then ask if "to plunk something somewherew means " t o place something somewheren, and upon getting an affirmative reply, can add t h e new frame to those associated w i t h the concept PLACE. These syntactic features, however, need not bs inflexible rules. Sentence understanding can still psocaed w e n i f tha syntactic features found by syntax do not exactly match those specified by the clausefunction frame. Thus, an inadvertent ungrammaticality cam readily be recognized as such, and processing can continue. (1) The number of concepts a v a i l a b l e t o the system a t present is very small. T h i s , in fact, is why the system's first guess is usually the correct one. I f the sentence is at a l l w i t h i n the systea's comprehension, t h e options as to its meaning a r e currently q u i t e limited.( 2 ) The range of expressive devices presently recognized is q u i t s limited as well. For instance, the system does n o t recognaze relative clauses, con junctions, o r pronouns (except f o r 1 and you).( 3 ) The system currently d e a l s only with TOTALLY UNFMILIAR words and expressions in this adaptive fashion, It w i l l not correctly handle familiar words which are used in new ways (such as a noun used eas a varb, as i n wzero the c e n t e r s q u a r e n ) .( 4 ) The system t r i e s to map the meaning o f new wards and expressiuns into i t s s p e c i f i e d s e t of underlying concepts. It then displays its hypotheses t o the user, g i v i n g him only the option of saying yas or nu. The user cann-ot say "no, not q u i t e , it meahs . . .". (Thus concepts like V h e 'northeast1 square" o r "the 'topmost' squarew would ba confusing and not correctly understood.)
Appendix:
| null | null | null | null | {
"paperhash": [
"walker|speech_understanding_through_syntactic_and_semantic_analysis"
],
"title": [
"Speech Understanding Through Syntactic and Semantic Analysis"
],
"abstract": [
"of p a r t s of spoken u t t e r a n c e s . More i m p o r t a n t l y , we l e a r n e d someth ing about t he prob lems o f speech unde r s t a n d i n g . We a l r e a d y have begun work on a second v e r S t a n f o r d Research I n s t i t u t e i s p a r t i c i p a t i n g in a s i o n t h a t w i l l use a new p a r s e r , now under deve lopment , ma jo r program o f r e s e a r c h on the a n a l y s i s o f c o n t i n u o u s and t h a t w i l l i n v o l v e a d i f f e r e n t t a s k domain. T h i s speech by computer . The goa l i s the development o f a new system i s s t i l l i n t he p rocess o f c o n s t r u c t i o n , a l speech u n d e r s t a n d i n g system capab le o f engag ing a human though the p a r s e r i s f a r enough a l o n g f o r us to p r e s e n t o p e r a t o r i n a n a t u r a l c o n v e r s a t i o n about a s p e c i f i c i t i n a companion paper (Pax ton and Rob inson , 1973 ) . p rob lem domain . The approach b e i n g t a k e n i s d l s t i n c We have chosen t o d e s c r i b e t he f i r s t v e r s i o n i n t h i s t i v e i n t he e x t e n t t o w h i c h i t depends o n s y n t a c t i c and pape r , because w e b e l i e v e t h a t t he b a s i c system concepts semant i c p r o c e s s i n g t o g u i d e the a c o u s t i c a n a l y s i s . w e a re t e s t i n g are i l l u s t r a t e d t h e r e . Moreover , t he T h i s paper p r o v i d e s a d e s c r i p t i o n o f t h e f i r s t v e r s i o n prob lems we have i d e n t i f i e d a re ones t h a t we t h i n k a re o f t h e sys tem, emphas iz ing the k i n d s o f i n f o r m a t i o n w o r t h p r e s e n t i n g t o o t h e r members o f t he a r t i f i c i a l i n t h a t need t o b e added f o r e f f e c t i v e r e s u l t s . t e l l i g e n c e community. Thoroughgo ing s o l u t i o n s t o these prob lems w i l l r e q u i r e more t h a n t h e r e s o u r c e s a v a i l a b l e w i t h i n our p r o j e c t a t SRI o r even those i n t he INTRODUCTION ARPA program as a w h o l e . The prob lems i n v o l v e d In a c o u s t i c a n a l y s i s a re no t examined h e r e ; o n l y enough i n f o r m a t i o n i s p resen ted t o c l a r i f y t he n a t u r e o f t h e S t a n f o r d Research I n s t i t u t e i s p a r t i c i p a t i n g i n a approach b e i n g taken i n the system d e s i g n . ma jo r program o f r e s e a r c h on the a n a l y s i s o f c o n t i n u o u s speech by computer (see Newel l e t a l . , 1971) b e i n g Most p r e v i o u s work on v o i c e i n p u t lo a computer i s sponsored by t he Advanced Research P r o j e c t s Agency r e f e r r e d to as \"speech r e c o g n i t i o n \" r a t h e r t han as (ARPA). The g o a l i s t h e development o f a speech u n \" speech u n d e r s t a n d i n g . \" (see H i l l , 1 9 7 1 , and L e a , 1 9 7 2 ) . d e r s t a n d i n g system capab le o f engag ing a human o p e r a t o r Research on speech r e c o g n i t i o n has aimed at p r o v i d i n g in a n a t u r a l c o n v e r s a t i o n about a s p e c i f i c t a s k domain. an o r t h o g r a p h i c t r a n s c r i p t i o n o f the sounds and words Our pa th toward t h i s goa l has been a c h a r a c t e r i s t i c a l l y c o r r e s p o n d i n g to the a c o u s t i c s i g n a l . The ma jo r empha\" a r t i f i c i a l i n t e l l i g e n c e \" app roach . We b e l i e v e t h a t s i s i n systems des igned f o r t h a t purpose has been on many o f t he c r i t i c a l prob lems i n v o l v e d cannot be a n t i c a c o u s t i c p r o c e s s i n g ; some groups have deve loped p a t t e r n i p a t e d o u t s i d e o f t h e c o n t e x t o f a f u n c t i o n i n g system. ma tch ing s t r a t e g i e s , w h i l e o t h e r s have t r i e d t o i d e n t i f y A s a r e s u l t , ou r f i r s t e f f o r t s were t o b u i l d a p r e u n i t s p h o n e t i c a l l y o r p h o n e m i c a l l y and t o aggrega te l i m i n a r y v e r s i o n , u s i n g , where p o s s i b l e , a v a i l a b l e them i n f o l a r g e r and l a r g e r u n i t s . Wh i l e t h e r e have programs a s components. Because o f t he c r i t i c a l r o l e been some r e s u l t s w i t h i s o l a t e d words f rom r e l a t i v e l y t h a t w e expec t semant i cs t o p l a y i n t h e f i n a l sys tem, sma l l v o c a b u l a r i e s , e x t r a p o l a t i o n o f these t e c h n i q u e s we chose, as a base , W i n o g r a d ' s programs f o r under to c o n t i n u o u s speech has not been s u c c e s s f u l . s t a n d i n g n a t u r a l language (Winog rad , 1971 ) . A c c o r d i n g l y , we accep ted f o r our f i r s t task domain h i s s i m u I n c o n t r a s t , r e s e a r c h on speech u n d e r s t a n d i n g seeks l a t i o n o f t he a c t i o n s o f a r o b o t t h a t knows about and to d e t e r m i n e f o r spoken u t t e r a n c e s the message i n t e n d e d can m a n i p u l a t e b l o c k s o f v a r i o u s shapes, s i z e s , and in r e l a t i o n to the accompl ishment o f some task and in c o l o r s . The i n t e n t was t o a l l o w a person speak ing t o s p i t e o f i n d e t e r m i n a c i e s and e r r o r s i n the g e n e r a t i o n , t h e computer t o ask q u e s t i o n s about t h e \" b l o c k s w o r l d , \" t r a n s m i s s i o n , and r e c e p t i o n o f an u t t e r a n c e . The p r o t o g i v e commands t h a t wou ld m o d i f y i t , and t o add c e s s i n g o f s y n t a c t i c , seman t i c , and p r a g m a t i c i n f o r i n f o r m a t i o n t h a t wou ld augment i t s s t r u c t u r e . m a t i o n i s c o n s i d e r e d e s s e n t i a l , and a q u e s t i o n a n s w e r i n g system may even be used as a ma jo r component. D u r i n g t he f i r s t year o f t h e p r o j e c t , w e comple ted a f i r s t v e r s i o n o f our system t h a t d i d a l l o w us to use There a re a v a r i e t y o f approaches to speech unde r s y n t a c t i c , s e m a n t i c , and a c o u s t i c da ta i n t h e a n a l y s i s s t a n d i n g b e i n g taken b y p a r t i c i p a n t s i n t h e ARPA P r o gram. I t wou ld be beyond the scope o f t h i s paper t o s k e t c h them o u t ; however, d e s c r i p t i o n s a re p r e s e n t e d a t The work r e p o r t e d h e r e i n was sponsored by the Advanced th is Con fe rence o f t h e work a t C a r n e g i e M e l l o n U n i v e r Research P r o j e c t s Agency o f t h e Department o f Defense s i t y (Erman e t a l . , 1973; Reddy e t a l . , 1973) and a t under C o n t r a c t DAHC04-72-C-0009 w i t h t he U.S. Army Bol t Beranek and Newman (Woods and Makhou l , 1 9 7 3 ) . Research O f f i c e . E l sewhere , t h e r e are r e p o r t s o n t h e d e s i g n o f t he System Development C o r p o r a t i o n system ( B a r n e t t , 1972) and References a r e l i s t e d a t t he end o f t he paper . o n t h e work a t L i n c o l n L a b o r a t o r y ( F o r g i e , 1972a, 1972b) . Some o f t hese e f f o r t s c o n c e n t r a t e on a c o u s t i c a n a l y s i s o f t h e speech s i g n a l , segment ing and l a b e l i n g phonemel i k e u n i t s t h a t w i l l b e grouped i n t o words—and more complex g rammat i ca l s t r u c t u r e s — a c c o r d i n g t o s y n t a c t i c and perhaps semant i c c r i t e r i a . Others accept hypo the ses f rom a number o f sou rces , f o r example, a c o u s t i c , s y n t a c t i c , and seman t i c , each of wh ich may be checked a g a i n s t \" the r e s t . A c t u a l l y , the ARPA program i s s t i l l i n i t s e a r l y s t a g e s , and none o f these sys tems—our own i n c l u d e d — c a n b e s a i d t o have e s t a b l i s h e d f i n a l des i gn s p e c i f i c a t i o n s , so s p e c i f i c c o n t r a s t s a r c hard t o draw f i r m l y . Moreover , d i f f e r e n t t ask domains would seem to respond d i f f e r e n t i a l l y t o one approach r a t h e r than a n o t h e r , a p o i n t t h a t w i l l b e cons ide red aga in l a t e r . In the system we are d e v e l o p i n g at S H I , knowledge about the t a s k domain , the grammar, and the c u r r e n t s t a t e o f the a n a l y s i s are used t o c o n s t r a i n the s e l e c t i o n of the. word or words t h a t m igh t be expected to be p r e s e n t a t a p a r t i c u l a r p l ace in the speech s t ream r e p r e s e n t i n g an u t t e r a n c e . The a c o u s t i c data f o r t h a t l o c a t i o n a re ana l yzed t o de te rm ine the degree o f c o r r e spondence w i t h each expected word by a program t h a t c h a r a c t e r i z e s i t s a c o u s t i c s t r u c t u r e . When the presence o f a word i s c o n f i r m e d , t h i s i n f o r m a t i o n , i n c o n j u n c t i o n w i t h the o t h e r sources o f knowledge in the sys tem, leads t o t he s e l e c t i o n o f ano the r word f o r t e s t i n g a t the n e x t p l ace i n t he speech s t r e a m . Success ive s teps p r o v i d e b o t h a segmen ta t i on o f the u t t e r a n c e i n t o words and a s p e c i f i c a t i o n o f i t s s y n t a c t i c and semant ic s t r u c t u r e . The d i s t i n c t i v e aspec ts o f the des ign are i t s s t r o n g dependence on syn tax and semant ics and i t s d e l i b e r a t e m i n i m i z a t i o n o f hypotheses genera ted s o l e l y on the b a s i s o f a c o u s t i c d a t a . The c a p a b i l i t i e s developed f o r t h e f i r s t v e r s i o n o f t h e SRI speech u n d e r s t a n d i n g system were rud iment a r y , bu t i t d i d p r e d i c t words and t e s i f o r t h e i r p r e s e n c e . More p r e p r o c e s s i n g o f t h e a c o u s t i c data was done than we b e l i e v e shou ld be necessary . A c o u s t i c c h a r a c t e r i z a t i o n s were p repared f o r o n l y a few words, so i t has no t been p o s s i b l e to s tep th rough a complete u t t e r a n c e . O t h e r sources o f knowle"
],
"authors": [
{
"name": [
"D. Walker"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null
],
"s2_corpus_id": [
"325508"
],
"intents": [
[]
],
"isInfluential": [
false
]
} | null | 591 | 0.001692 | null | null | null | null | null | null | null | null |
7041e9292bfafb7e8e56fb776413a5b8939bb66c | 59963340 | null | Semantic Processing for Speech Understanding | T h e semantic c o m p o n e n t O f t h c t p e c c h u n d e r s t a n d i n g s y s t g r n b e i n g Q ~v e l o p c d j b l n t l y b y and SBC r u l e s out p h r a s e combination8 t h a t a r e not m t a n l n g f u ) ~n d ~r a d u e a g semantic i n t c r g r e t a t l o n g f o r eombin&tlons t h a t area T h e s y s t e m conslstr o f a semantic network madsf a n d r o u % l f l @ $ t h a t i n t a r a c t w i t h i t , The net 1s p a r t i t i o n e d i n t o a s e t o f h l o r a r e h i c a l b y o r d e r e d s u b n e t e , f d s i l l t s t i n g t h e e n c o d i n g o f h l q h e r = ~r d e ~ p r e d l c s t e t and t h e maintenance o f m u l t i g l c p a r r i n g h y p o t h e s e s , C o m p o s i t i o n € a U t i n t t r mrnblning U t t e r a n c e earnpanant% i n t o Phrases# c o n s u l t n c t w q r k d Q c c t l p t l o n g o f p r o t o t y p e ritortiana an3 s u r f a c e -t o -d e e pc a f e m s p r , O u t p u t r f r o m t h @ r @ r o u t i n e s a r e n e t w o r k f r n q m c n t s c a n r i x t l n g o f raver.1 r u b n e t 8 t h e t I n a g g r a g a t c c a p t u r e t h e i n t c r r o l r t i a ~r h i p f b e t w e a n a p h t r 8 c e s syntax e n d r c m a n t l c s , T h i ~ r @ o c a r c h war s u p p o r t e d b y t h e Defsnsc A d v a n c e Research P r o j e c t s Agency o f t h o D c g a r t n c n t of Defcnre and mon-f5torcd b y t h e U.S. Army F 1 8 e r r c n O f f i c e u n d a r cantrrct No, DAHCOI-75-C-0006, | {
"name": [
"Hendrix, Gary G."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 0 | 1 | null | null | null | null | null | Faxton a n d Rabintanr 1 9 7 5 , a n d R e b l n s o n I 1 9 7 5 1 ,ryntactically capable a t cornblnln8 t o farm e l a r g e r P h F a r @ c | Main paper:
o u t i n e s ( s c r s ) t h a t a r t c a a r d i n a t~d w i t h t h e l a n g u a g~ d e f i n i t i o n (roughly1 t h e n g r a m m a r w f o r t h e e p t c c h u n d o r r t a n d l n g s y t t r m t g e e:
Faxton a n d Rabintanr 1 9 7 5 , a n d R e b l n s o n I 1 9 7 5 1 ,ryntactically capable a t cornblnln8 t o farm e l a r g e r P h F a r @ c
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.001692 | null | null | null | null | null | null | null | null |
eecf9246f924d236040002adc6ef21a82d8371db | 63167929 | null | Establishing Context in Task-Oriented Dialogs | Thl8 paper d t r c r l b r s part b t t h e d l 8 e ~u t ~e easpanant a t a c p ~e c h underrfrnding S Y S t l a f a r tark=0ri@ntad d i a l o g r l ~p ~c i t l ~~l l y ~ r esehinlrm f a r rrtrbllrhing a tacur a t attention t o r i d In ldsntitytng t h e referent8 O f d e r i n l t a noun Ohrrrer, In building a teprrrentatlon of t h e d l r l o g context, the dlscourre ptocssro? trkrs advantage of t h e f a c t that trrkmorlantcd d i a l o g s have a structure t h r t clareLY parallel8 t h a rtructure o f t h r trsk, The irm@ntlc nctrork of t h c system 1 8 partltlonrd into toeur rprcrr w i t h rrch foeur apace C ~n t a l n l n g o n l y thorr eonceptr p e t t i n ~n t t o t h a a i r l o g relating t o r r u b t r r k . The facur aprcer r r c link@b t a t h e i r QarraSpan4ing r u b t r r k r and a r d @ t * d In r hterarchy dotrrnlnad by t h ~ relatianr &Rang r u b t r t k o . T b i # rrrarrch war @upportad by the Deferire Advanced Rerearch Progtct8 Aprney o f the Departrant o f D ~f e n r r and raonltbrud by t h e U.8. & t 1 y Re8ewch OF t i c @ under Contrret Wa, O A H C O ~-~S -C ~O O O ~. &anouror cormunlcrtgon rntai11 thm trrnrairslon o t contcptl from t h e # p ~. k e r @ @ aodal o f t h e world t o t h r 1 i 8 t a n e r ~s . I f i s erueirl t h a t t h e rperkr? be eb1e t o comaunicrte d e r e r i P t l o n B a t concepts in h l t model i n r w r y t h i t allow8 t h e llrtrner t o pick 6uf the relevant retatad concept kn h i 8 madel, In nornri hunan | {
"name": [
"Deutsch, Barbara G."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 0 | 15 | null | null | null | null | null | A a 0 . K .A # My fingers f a t e t h e tools I rm urlng , , , IThe | Main paper:
~a bolt t h e pump t o t h e p t r t t o r m ,:
A a 0 . K .A # My fingers f a t e t h e tools I rm urlng , , , IThe
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.025381 | null | null | null | null | null | null | null | null |
7995580a134408562788a9b80e97edea79e1a609 | 219302303 | null | Using Planning Structures to Generate Stories | TALE-SPIN is s program which makes up stories by using planning structures as part of its world knowledge. Planning structures represent goals and the methods of achieving those goals. Requirements f o r aparticular method depend on static and dynamic facts about the world. TALE-SPIN changes the state of the world by creating new characters and presenting obstacles to goals. The reader / listener makes certain p l o t decisions during the telling of the story. The story is generated using the notation of Conceptual Dependency and is fed to another program which translates it i n t o English. 80 JOE BEAR THOUGHT THAT HENRY BEE WOULD GIVE THE HONEY TO W I M . JOE BEAR WALKED TO THE BEEHIVE WHERE HENRY BEE WAS. HE ASKED HENRY BEE IF HE WOULD GIVE THE HOMEY TO HIM. >> DECIDE: DOES "HENRYBEE* AGREE? *YES HENRY BEE DECIDED HE WOULD GIVE I T TO JOE BEAR. HENRY BEE GAVE IT TO JOE BEAR. BE ATE IT. HE WAS FULL. THE END. Here is a s t o r y which TALE-SPIN generates w h i c h t h e t r a n s l a t o r is n o t y e t capable of producing i n ~n g l i s h : JOE BEAR W4S HUNGRY. | {
"name": [
"Meehan, James R."
],
"affiliation": [
null
]
} | null | null | null | 1975-11-01 | 3 | 9 | null | T h e r e a d e r is asked t o make c e r t a i n decisions about the story d u r i n g t h e process of generation. Here is an example.J O E BEAR WALKED TO THE TREE WHERE I R V I N G BIRD WAS.>> DECIDE: DOES *IRVINGBIRD* AGREE? *NOHE THOUGHT THAT HE WOULD LIKE JOE BEAR TO GIVE I T TO HIM, Here is a s t o r y which TALE-SPIN generates w h i c h t h e t r a n s l a t o r is n o t y e t capable of producing i n ~n g l i s h : worry about when you're making up t h e p l a n .GROUNDJOE BEAR W4S HUNGRY. HE THOUGHT TEWJ! I R V I N G BIRDabout "runtimen preconditions u n t i l you're executing t h e plan. t h e p l a n . ) I f 1 ' m p l a n n i n g t o g e t a ----The non-procedural d a t a b a s e used by the p l a n n i n g s t r u c t u r e s i s t e a t (-6 ,-la , -4 ) . The n e x t P l a n b o x i s c a l l e d "Steal". W e ask Memory ~h e t h e r Henry is home: i f h e w e r e n ' t , J o e would simply t a k e t h e honey. C o h e r e n c y i s i m p o r t a n t : t h e r e has t o be a l o g i c a l flow from one The stories a r e more concerned w i t h r e a c t i o n t h a n i n t e r a c t i o n . 3For every p l a n , there may be a c o u n t e r -p l a n , a plan t o block t h e achievement of a goal: a p l a n f o r keeping away from somethinq or Riesbec k , Schank, R. C. and Abelson, R. P. ( 1 9 7 5 ) . Scripts, plans and knowledge, In Proceedings of t h e 4 t h I n t e r n a t i o n a l Joint C o n f e r e n c e on Artificial I n t e l l i g e n c e .d i v i d e d i n t o f i v e classes. | null | null | null | null | Main paper:
:
T h e r e a d e r is asked t o make c e r t a i n decisions about the story d u r i n g t h e process of generation. Here is an example.J O E BEAR WALKED TO THE TREE WHERE I R V I N G BIRD WAS.>> DECIDE: DOES *IRVINGBIRD* AGREE? *NOHE THOUGHT THAT HE WOULD LIKE JOE BEAR TO GIVE I T TO HIM, Here is a s t o r y which TALE-SPIN generates w h i c h t h e t r a n s l a t o r is n o t y e t capable of producing i n ~n g l i s h : worry about when you're making up t h e p l a n .GROUNDJOE BEAR W4S HUNGRY. HE THOUGHT TEWJ! I R V I N G BIRDabout "runtimen preconditions u n t i l you're executing t h e plan. t h e p l a n . ) I f 1 ' m p l a n n i n g t o g e t a ----The non-procedural d a t a b a s e used by the p l a n n i n g s t r u c t u r e s i s t e a t (-6 ,-la , -4 ) . The n e x t P l a n b o x i s c a l l e d "Steal". W e ask Memory ~h e t h e r Henry is home: i f h e w e r e n ' t , J o e would simply t a k e t h e honey. C o h e r e n c y i s i m p o r t a n t : t h e r e has t o be a l o g i c a l flow from one The stories a r e more concerned w i t h r e a c t i o n t h a n i n t e r a c t i o n . 3For every p l a n , there may be a c o u n t e r -p l a n , a plan t o block t h e achievement of a goal: a p l a n f o r keeping away from somethinq or Riesbec k , Schank, R. C. and Abelson, R. P. ( 1 9 7 5 ) . Scripts, plans and knowledge, In Proceedings of t h e 4 t h I n t e r n a t i o n a l Joint C o n f e r e n c e on Artificial I n t e l l i g e n c e .d i v i d e d i n t o f i v e classes.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 591 | 0.015228 | null | null | null | null | null | null | null | null |
8b6807492014a5e37d8159a3e9bb66a7a9873acd | 219309066 | null | Verb Paradigms for Sentence Recognition | T h i s paper d a s c r i b c s a linguisticnlly-based recognition grammar modcl, which was d c v d l o p c d as p a r t of a C o ~u p ~~t l . t c ~-r \ i J c d I n s t r u c t i n n P r o j e c t , to the t a s k s c f r e c o g n i z i n g and a n o l y z i n ~ n v a r i e t y of b a s i c s e n t e n c e t y p e s i n English. Ways of e s t e n d i n g t h e model t o t h e a n a l y s i s of c o m p l e s sentences are also suggested. The procedures and the model d e s c r i b e d herein are original; however, tjhey owe much t b i n s i g h t s found i n the w o r k of tws l i n g u i s ts , Grub,er and Fillmore. The g e n e r a l problem o f grammar r e c o g n i t i o n i s t h a t o f going f r o m a s u r f a c e s t r i n g of words rn a deep r c p r c s c n t a t i o n that p e r m i t s semantic i n t e r p r c t a 8 i o n . PIorc s p e c i f i c a l l y , our grammar r e c o g n i t i o n p r o c e d u r e depends on t h c identlfica t i o n of the ~r e c i s e function or semantic role that each noun phrase a c t a n t occurring in a given sentence exhibits with r e s p e c t t c p t h e verb of t h a t s e n t e n c e . By a s s i g n i n g verbs--or, to be more precise, verb, senses-to one o r more paradigms ( , p e r c e p t u a l l y and f u n c t i o n a l l y d e f i n e d surf ace con£ igurations) , i t becomes possible to determine algorithmically for every sentence the functional relation (e. g . , theme, causal a c t a n t , goal, source, l o c u s ) t h a t each noun phrase i n the sentence bears t o t h e verb, thereby a s s i s t i n g g r e a t l y Ln arriving at a representation of the mearring of each sentence. A nuhber of verb paradigms such ae intrans Ltive, t r a n s i t i v e and ergative Rre d e f i n e d . Verbs belonging t o the intransitive paradigm such a8 die, fall, go, etc. always have subjects that function ae t%emes. Verbs belonging to the transitive paradigm such ae k i l l , -read, -S e a t e t c . have subjects that function as causal actants -.and o b j e c t s that function as themes. The ergative paradigm, which is more complex, cons Fs t s of change-of -s t a t e verbs such a s open, melt, increase, e t c . If an ergative paradigm verb has both a subject and-an object, the subject is a causal a c t a n t and the object is a theme; however; if such a verb takes only a subject, then the subject functions as a theme. The paradigm membership of each verb sense i n the data base i s determined and is recorded a s a lexical feature of that verb. | {
"name": [
"Celce-Murcia, Marianne"
],
"affiliation": [
null
]
} | null | null | null | 1976-02-01 | 0 | 0 | null | null | The number o+ verb paradigms would ,proliferate almost ind e f i n i t e l y were it not for several devices, b u i l t into the grammar. a t i o n . For exardple, a sentence srtch as "It is raining." w i l l be analyzed as having incorporated the theme subject r a i n ' i n t o the verb w i t h the result that a abstract s t r u c t u r e resembling "Rain is f a l l i n g " g e t s rec80nstructcd and processed as an i nt r a n s i t i v e verb paradigm item.To overview the e n t i r e procedure, we s t a r t by parping the sprface structure of any given sentence. T h e major c o n s t i t u e n t s The primqry purpose of t h i s grammar was the analysis of student responges phrased in natural Englisha necessary step i n t h e answer-evaluation process. This paper describes i n a g e n e r a l way the grammar re- the other hand, make use of a s y s t e m i c -t y p e of grammar a la H a l l i d a y (1961, 1966, 1967) ; however, t h e power o f WLnograd Rather t h a n produce an imitation oC either t h e IBM Recognition Gramar or the MIT P r o j e c t --i t i s , in fact, possible t h a t neither of these approaches w i l l u l t i i a t e l y be the most useful one for recognii tion purpbses--we tr Led to reexamine recent insights , e s p e c i a l l y F i l l m o r e ' s Dteep Case Hypo thesis (1968b) , The problem we faced is, i n general terms, the f o l l o w i n g : All noun phrases occurring i n E n g l i s h sentences w i l l be viewed as a c t a n t s that bear a s p e c i f i c f u n c t i o n a l relation to t h e verb o r element of p r e d i c a t i o i n t h e sentmenee. These funct i o n a l relations (e.g., theme, causal actant, locus, e t c . ) will be dfscussed i n the n e x k s e c t i o n o f t h e p a p e r . W e v i e w the i d e n t i f i c a t i o n of the p r e c i s e f u n c t i o n s t h a t a l l noun phrase a c t a n t s i n a given s e n t e n c e e x h i b i t t o t h e i r verb as t h e b a s i c problem of sentence r e c o g n i t i o n . English, more s o than languages such a s German or Russian, tends t o give r e l a t i v e l y l i t t l e d i r e c t i n d i c a t i o n i n s u r h c e s t r u c t u r e a s t o what the f u n c t i o n of a given noup phrase in a sentence ls. This is because noun pbrases occurring ae eurface subj ec t e and eurface ollj ec ts--with the rxcaption of oome pronominal and interrogat~se o r relative dorms-are c o m p l e t e l y unmarked in English ( . they bear no i n f l e c t i o n that would exclude or eugges t a particular function) . Thus we may have a eentence with en unmarked noun pnrase as the sur face subject:Given( 1) The object deecended.Surface Subj .The same unmarked noun phrase may occur as a single unmarked eurface o b j e c t (2) or aa one of two unmarked surface objectsEQUATIONJqhn saw the object Surface Surface Sub j .Obj . ( 3) -John ga,ve Mary the g b j ec t.Surface Surf aceSubj . Obj. 1 Obj. 2Thus in sentences such as (1) through ( 3 ) , the o n l y informa t i o n we can use if we w a n t t o i d e n t i f y the function of the noun phrases is: In cases such as these, where the o n l y i n f o m a t i o n we hevs about t h e noun phrases i n a sentence has t o do. w i t h t h e i r s e r i a l order, we s a y t h e t the noun phrases a r e unmarked. -oriented h a b i t ( e . g . , wash, d r e s s , shave, e t c . ) "me" is a remote causal a c t a n t or source and n o t a remote goalWhenIn "Open the door for me. " I n this sentence "(you)" is the B c a u~a 1 a c t a n t , a theme, and a locus is the f o l l o w i n g : Camping is en j oyable . SurfEqui obi ec t-aubj ec t deletion:('lo) John t o l d B 1 --+ John t o l d B i l l t o go.I Subject-to-subject raising:John begalif t o run. A John runSub 1 ec t-to-objec t raisin&:Mary wants NP . shows how the features i n o u r r e c o g n i t i o n grammar correspond t o t h e features that might b e used i n a g e n e r a t i v e grammar. | null | null | null | Main paper:
:
The number o+ verb paradigms would ,proliferate almost ind e f i n i t e l y were it not for several devices, b u i l t into the grammar. a t i o n . For exardple, a sentence srtch as "It is raining." w i l l be analyzed as having incorporated the theme subject r a i n ' i n t o the verb w i t h the result that a abstract s t r u c t u r e resembling "Rain is f a l l i n g " g e t s rec80nstructcd and processed as an i nt r a n s i t i v e verb paradigm item.To overview the e n t i r e procedure, we s t a r t by parping the sprface structure of any given sentence. T h e major c o n s t i t u e n t s The primqry purpose of t h i s grammar was the analysis of student responges phrased in natural Englisha necessary step i n t h e answer-evaluation process. This paper describes i n a g e n e r a l way the grammar re- the other hand, make use of a s y s t e m i c -t y p e of grammar a la H a l l i d a y (1961, 1966, 1967) ; however, t h e power o f WLnograd Rather t h a n produce an imitation oC either t h e IBM Recognition Gramar or the MIT P r o j e c t --i t i s , in fact, possible t h a t neither of these approaches w i l l u l t i i a t e l y be the most useful one for recognii tion purpbses--we tr Led to reexamine recent insights , e s p e c i a l l y F i l l m o r e ' s Dteep Case Hypo thesis (1968b) , The problem we faced is, i n general terms, the f o l l o w i n g : All noun phrases occurring i n E n g l i s h sentences w i l l be viewed as a c t a n t s that bear a s p e c i f i c f u n c t i o n a l relation to t h e verb o r element of p r e d i c a t i o i n t h e sentmenee. These funct i o n a l relations (e.g., theme, causal actant, locus, e t c . ) will be dfscussed i n the n e x k s e c t i o n o f t h e p a p e r . W e v i e w the i d e n t i f i c a t i o n of the p r e c i s e f u n c t i o n s t h a t a l l noun phrase a c t a n t s i n a given s e n t e n c e e x h i b i t t o t h e i r verb as t h e b a s i c problem of sentence r e c o g n i t i o n . English, more s o than languages such a s German or Russian, tends t o give r e l a t i v e l y l i t t l e d i r e c t i n d i c a t i o n i n s u r h c e s t r u c t u r e a s t o what the f u n c t i o n of a given noup phrase in a sentence ls. This is because noun pbrases occurring ae eurface subj ec t e and eurface ollj ec ts--with the rxcaption of oome pronominal and interrogat~se o r relative dorms-are c o m p l e t e l y unmarked in English ( . they bear no i n f l e c t i o n that would exclude or eugges t a particular function) . Thus we may have a eentence with en unmarked noun pnrase as the sur face subject:Given( 1) The object deecended.Surface Subj .The same unmarked noun phrase may occur as a single unmarked eurface o b j e c t (2) or aa one of two unmarked surface objectsEQUATIONJqhn saw the object Surface Surface Sub j .Obj . ( 3) -John ga,ve Mary the g b j ec t.Surface Surf aceSubj . Obj. 1 Obj. 2Thus in sentences such as (1) through ( 3 ) , the o n l y informa t i o n we can use if we w a n t t o i d e n t i f y the function of the noun phrases is: In cases such as these, where the o n l y i n f o m a t i o n we hevs about t h e noun phrases i n a sentence has t o do. w i t h t h e i r s e r i a l order, we s a y t h e t the noun phrases a r e unmarked. -oriented h a b i t ( e . g . , wash, d r e s s , shave, e t c . ) "me" is a remote causal a c t a n t or source and n o t a remote goalWhenIn "Open the door for me. " I n this sentence "(you)" is the B c a u~a 1 a c t a n t , a theme, and a locus is the f o l l o w i n g : Camping is en j oyable . SurfEqui obi ec t-aubj ec t deletion:('lo) John t o l d B 1 --+ John t o l d B i l l t o go.I Subject-to-subject raising:John begalif t o run. A John runSub 1 ec t-to-objec t raisin&:Mary wants NP . shows how the features i n o u r r e c o g n i t i o n grammar correspond t o t h e features that might b e used i n a g e n e r a t i v e grammar.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 588 | 0 | null | null | null | null | null | null | null | null |
58bfdc091a30db78176a5f479b500822efb0b586 | 219306083 | null | Natural Language Understanding Systems within the {A}. {I}. Paradigm: A Survey and Some Comparisons | The paper s u r v e y s the najcr prcjecte on th.e understanding of natural language t h a t fall w i t h i n what nay now 5e c a l l e 3 t 5 e artificial intelligence paradigm fcr natural Language s y s t e ~s . -c4z,'2e space is devoted ta a r g u i n g that the ¶digm is new a red: i t v a n d different in significant respects from the g e n e r a b i v e paradiLm ~f present day linguistics. The caalparison's between s y s t e m s tekt3-c I 1 round questions aL.-ut UIC l c ~~l , c c ~l t r a l i t b * and ~h a n c n t -n ~1 c 7 g i i a l p l n u s i k ~i l i t y " of t h e k n ~~w l t ~~l ~. ~~ and inft-rt.nct.s t1.1.1t-must .rv.~~l,~t-'Ltto a s y s t t l m that is to uurldt*rst,~nd ~-\*t-r>*~l.~y l s ~l y u ~~g r ? . Some comparisons and contrasts Conclusion References "The CLRUSE program looks at the first wrd, to decide what w r i t the CLRUSE begins w i t h . If it sees an adverb, i t assumes the sentence begins w i t h a single-word modifier tslowly, Jack l i f t e d the book] ; iP it sres a preposition, it looks .for an i n i t i a l PREP6 Con top of the h i l l stood a tree] If it sees a BINOBR, it calls the CLAUSE program to look for a BOUND CLAUSE C~efore you get there, w e l e f t ] . In English (and possibly a l l languages) the firsb word of a construction o f t e n gives a very goad clue as to what that construction will be. fn this case, "pick" is a verb, and i n d i c a t e s that we may have an IHPERATIW CLAUSE. The program. s t a r t s the VO program w i t l a the i n i t i a l VG feature. l i s t (VG IWER), looking for a VG of this type. This must either begin w i t h some form of the verb "do" [Do not call me!] or w i t h the main verb itself [Call me!]. Since the next word is not: "do" it checks the n e x t word i n the input ( i n U~is case still the first word) : o see whether it is the i n f i n i t i v e form of a verb. If SO, it is to be attached to the parsing tree, and given the additionql feature MVB (main verb). The current structure can be diagramad as: (CLAUSE MAJOR) (VG IWER) | {
"name": [
"Wilks, Yorick"
],
"affiliation": [
null
]
} | null | null | null | 1976-02-01 | 18 | 2 | null | To euntey an enewetic field like thie one is inevitably to laavo a great deal of excellent work unextiminad, a t least if one i a going to do more than give a paragraph to each research project.I have left out of cotasideration at least seven groups of projects:(1) Early work in Artificial Intelligence and Natural Language that has been sumeyed by Wfnograd (1973) and Simmons (1970a) among others.(2) Work by graduate students of, or intellectually dependent upsn that o f , people discussed in same detail here.( 3 )Wxk that derives essentially from projects described i n detail here. This embraces several groups interested i n testing psychological hypotheses, as r e 1 1 as others constructing largescale systems for speech recognition. I have devoted no space to speech recognition as such here, for it seems to me to depend upon the quality of semantic and inferential understanding as much as anything, and so I have concentrated upon this more fundamental task.Work on language generators, as opposed to analysers and understanders.They are essential for obtaining any testable output, but are thearetically secondary. That would be wholly inappropriate i n the present s t a t e of things. A g r e a t deal o f work is being done a t the moment, and many of the p r i n c i p a l researchers change t h e i r views on very fundamental questions between one paper and t h e next without drawiw any a t t e n t i o n t o the f a c t . Cheap self-contradictions and changes of mind are a l l too easy t o f i n d , so c r i t i c i s m and smparisons are best drawn w i t h a very broad brush and a l i g h t stroke. | null | null | Winograd's systea runs a s a dialogue, i n r e a l time, between a human Winogradas parsing is top down, and depth f i r s t , with na automatic back up. The parsing progrim fur each griuxnatical c a t c q~r~~ is n functional definition in PRfXZWWR, which can be stated either a s &VP, fcr SEWPEHCE, or as a f l o w -c h a r t as below for VP:DEFIWJ pxqram VP R E m f a i l u r e Yes LHere is Winograd's own account of the start of this top-down parsing procedure for the sentence "Pick up a red block" (where t h e material i n C 1 is added explanation and not Winogradas o m ) :"The CLRUSE program looks at the first wrd, to decide what w r i t theIf it sees an adverb, i t assumes the sentence begins w i t h a single-word modifier tslowly, Jack l i f t e d the book] ; iP it sres a preposition, it looks .for an i n i t i a l PREP6 Con top of the h i l l stood a tree]If it sees a BINOBR, it calls the CLAUSE program to look for a BOUND CLAUSE Since the next word is not: "do" it checks the n e x t word i n the input ( i n U~is case still the first word) : o see whether it is the i n f i n i t i v e form of a verb.C~eforeIf SO, it is to be attached to the parsing tree, and given the additionql feature MVB (main verb). The current structure can be diagramad as:(CLAUSE MAJOR) (VG IWER) (VB MVB INF TRANS ,WRT --------------pickT W S AND VPRT cmfw fran the definition of the word "pick" when we zalled the function P M S q for a word.4'Mter this syntactic parsing, a number of "semantic specidlists'' attach cwbantic structures to specific syntactic dnes.I n the case of "a red cube", the f o l l o~i n g structure i s built up by an NP "semantic specialist"Secondly, there are two rrpes of verb definition, s e m a n t i c and inferentialWinograd does not give the semantic definition for "pick-up", but here is t h e one f o r "grasp" which is a closelb-related verb. SIZE LOCATlcW \' corn MIbWTB ---------------------- M t w (--------- C ' 5 m RED U C ) f \NITS CaEM bTAC # PifiSoB--( '~~~S~~~~------------------------ rrg& , - I N ( --I T~U FkW1Z.Pwhich says e~sentiallp that grasping is scmathinq dorle by an animate entity to a oanlpulabla one (flret line).More of the real content of such actions is found in their inferential definition.Here is the one fox '"pick-up":(CONSE TC -PICKUP I X ) (PJCKUP xr ( r n ( W S P ?X) T i i E o r n ) (C.XXAL (RAISEHAND THEOREMS)TSlis definition allows the program to actually carry out the "pick-up" caamand if it is possible to do so in the simulated world, as it would not be, for example, if -re were already a block dn top of the red one..such as GRASP and RAISEHAND each of which must be carried out in order that sameming may indeed be p i c k e d up.fesential definitions: the one given for GRASP, for example, i s somewhat differant from i t s "C1(&EANS1' definition g i v e n above, although the inferential iPefini,tiona are aim, i n sane se-e, definitions of meaning as wall as prog r a~a s for actually carrying out the associated conmands.One reason for the enormous impdct of this work was that, prior to its appearance, A 1 work was not very l i n g u i s t i c a l l y interesting, while the eystems of tho linguists had no place far the use of inference and real world knowledge.was able to breed considerable results. where it is c l e a r that the answer is both definite, and that fhding i t requires some inferential manipulation of genaralirrtiens d w u t the world.3ha reader should ask himself at this j m i n t h o w he kwwe the nsrxact refemnt of the pmnaun in that sentanca.$ m e Discussion of S W L U S~J -fax, the reaction ts Winugqad \s work has L m t k wbSrSrla ~ineri t heal. What would crAitics find to attack i f tthtzp M~S F 90 a9nde.I: F i r s t l y , that ~~i n e g r a d ' s linguistic system is highly w n s r~~? a t i w , and that UIQ distinction between 'syntax' and 'semantics' m y not L w necessary at all.a way *.at m u l d make it fnextensihle, to any genexal, real tmrld, situatioh.Suppose 'block' were allowed mean 'an obstrxxctidnl and ' a nental ir,k,i*i t i o n ' , as well as ' a cubic object'. It is dauktful whether Winograd's features and rules could express t h e ambiguity, and, nore i m~r t a n t l y ,whether the simple s t r u c t u e s he manipulated c o u l d decide correctly between t h e a l t e r n a t i v e meanings i n any given c~n t e x t of user Again, f a r more sophisticated and systematic case structures than those hg used might be needed to resolve the ambiguity of 'in1 in "He ran the mile i n five minutes and Ire ran t h e m i l t . in a pawr kwg , as we11 as 'tllu ccmbination ~, l f case w i t h w r d sense a m h i p i t y i n 'Ho p u t t h o key in t h e l~r k ' (door lock1 and 'Be threw the key in the luck1 (river leek).The b l o c k s mrld is also strongly deduct-ive and l w i c a l 1 y closed, S f g r a v i t y were introduced into i t , then anything supkwrted t h a t war pushed i n a c e r t a i n way would have, logically have, t o f a l l . B u t t h e cazmon sense wlorld, of ordinary language, i s n o t like that: in t h~ 'waaen and soldiers'example given e a r l i e r , .the pronoun ' s e v e r a l 1 can be s a i d to be resolved u s i n g same generalisation such as ' t h i n g s shot a t and hurt tend t o fall'There are no logical 'have to's1 t h e r e , even though the meaning of the ~r onoun i s perfectly d e f i n i t e .Indeed, it might be argued that, in a sense, and as r~a f i s its semantics, Winograd's system is n o t about n a t u r a l language a t a l l , but a b u t the -technical question of how goals and sub-goals are to be crganised in a problem-solving system capable of mani~ulating simple physical ehjects.If n reeqber, for example, that tfre key problem that brought b w n the emrr#rur work on mchiae translation i n the Fiftiea and Sixties, was that of the eensa d i g u i t y of nattWi.l. language wprds, then we will look in vain to SHRDLU Lor any help w i t h t h a t problem. There seem? to be only one dear exmaple of aur aLnbiguous mrd in the whole system, namely that of 'curitairat as it appears in 'The b o x contains a red block' and rhe stackAgain, i f on@ glahces back a t the definition of 'pick-up' quoted &ve, ana can see t)cdt it. i e in fact an e%pression rrE a prcrcedure f~r picking up an ob5ect i n the SHR~LU .yet=.one, understand t h e p~r f e~t l v ordinary sant;ence 'I picked up my bags £tom the plstforn, and ran for the train', let alone any sentenco n o t &out a physical action performable by the hearer. T ' k difft!zeace i s t h a t Molods t a k e s a much more logico-sanantic i n t q r e t a t i u a of t h a t s l o g a n than does Winoqrad.In partisular, fbr W s t h e meaniq of an L n p~t utterance to h i s system is t h e procedures w i t h i n t h e system that raJ+ipulata tfr t r u t h conditions of t h e utterancfe and e s t g b l i s h its t r u t h value.To p u t t h e m t t e x crudely, f o r Moods an assertion has x i meanhq if hie.system cannot e t a b l i g h its t n t h o r falsity. should show, as it were how i t is a c t u a l l y t o be applied to l a n g u a~e , b u t that is n o t the s-e as demanding that it should be w r i t t e n in a ~r~c~d u r a 3 language,line PLANNER. I shall r e t u r n to this last pifit l a t e r . The i m p r t a n t wards these are 'lt-k f e y ' , which suggest that t 3 1 r > r~ may w e l l bo c o n f i r m i n g hints to be found in the :=tory and, i f t4lerc are, than this tentative, partial, inference is cursect, and TE a c h i n of demons can 'reacht one of the passiblo xaferente i n a story then there is a suct+ass registered and the ambiguity of the corresponding pronoun is resolved.It can be seen that the information encoded in the system is of a highly specific sortin the present case it is not about containers as such, and how to g e t their contents out, but about Piggy B a n k s in particular, and everything relies on that partfcular knowledge having been put i n . Thus, the apparently English words in the PB-OUT-OF Pound b e then tied directly, or via these inference rules, to response patterns which are generated. (Hendrix et a1 l73) , the ather apprdach is adopted, w i n g a primitive action B&CI.fANGG inetead of 'transfer'.EnormousThIe implementation under c~nstruction is a front-end ,oarsex of the Woodat augmented t r a n s i t i o n network type (see Woods '701, and a ganefation system going gram the s e a~t i c networks to surface strings described8in detail i n ( S W n s and S1ocu.m ' 7 2 ) .SimPons has also given considerable an & take + book Here 'pf itldicatea past, and is the aepndendy symbol liking a PP to We ACT ( ' t a k e ' ) which i s the hub of the conceptualization, as w i t h Simon& ?'he ' 0 ' indicates the objective case, marking the dependence of the object PP on the central ACT.There is a carefully constructed syntax of linkages between the conceptual categories* that will be describpd only in part in what follows.The next stage of the notation involves an extended case notation and a set of primitive ACTS, as well as a n q e r of it:ems suoh as PHYSWNT which indicate ather stqtes, and items of a fairly simplified psychological t h e o r y (the dictionary entry for 'advise', for example, contains a subgraph t e l l i n g us that Y 'will benefit' as part of the meaning of ' X advises Y ' can consider t h i s entzy als an active 'frAxte-like~ object seeking f i l l e r itms i n any context In which it is activated. Thus, in the sentence 'John sho the girl w i t h h riflet, the variables w i l l be f i l l e d in frcm context and the case inference will be made f m the main act PROPEL, which is that its h s t r u u e n t is lkSOVEI GRASP or PFtOPRL, and so we w i l l arrive a t the whole conceatualieation: Sot in building a template for 'John drinks,wine', the whole of the above tree-formula for 'drinks' would be piaced at the central action node, another tree structure for 'John' at the agent node and so on.John PROPEL 4 - bullet <-The complexity of the system comes from the way i n which the formulas, considered as active entities, dictate how other places hn the same template should be filled.Thus, the 'drink1 formula above can be thought of as an entity that fits at a template action node, and seeks a liquid object, that is ~ say a f~rmula w i t h (FLOW STUFF) as its right-most bzanch, to put at the object noda of the same template. This seeking is preferential, in that formulas not satisfying that requirement will be accepted. but only if nothing misf a c t a n c a lXZ'-fotUEa. TIie -€Elliplate Uif ly esWIisned Tm 3 Tragm e n t of t e x t is the one in which the most formulas hive their preferences s a t i s f i e d . a general principle at work here, t h a t the r i g h t interpretation 'says the least1 in inforreation-carrying terns, T ) r h wry simple device is able to do much o f the work o f a syntax and wxdysnse wzibigukty resa1vi;tap pxagraa P n a r c u s q~e , LZ the a,mteme k d been 'John drank s whole pitcher1, the fomulr tor th. 'pitcher of klquidb w u l d hawe h e n pr.rsEerreCI to that for thar human, s f m the subf~mS;a (FLOW STUFF) could be apprtopriateAy located uithirr i t . The extracted templates express information a lready implicitly preser.t i n the text, wen though many of them are partial inferences: anes that may not necessaxily, be true. Whether or not such a, system can remain s-le with a WrutderabLe vecabulary. of say several thousand words, has y e t to be t r r t d .ft will ba ovidrnt tso any reader that Zha laat t w a systems described, Bch.nktm ud my moun, share a great deal in cccpmon. whether we ahould expect advance in understanding natural language from those tackling the problems head on, or those coroncerned to build a 'fr8ntandv. It i~ cLtt,xly the case thata n p i e c e coulL bp esrcntial to the understanding of sane story. The question is, does it follow that the epehifict.tion, organieation and formalization of that knowldge l a &a studf oP l : . a g a , because i f it is then all human enquiry f t m physics and history to medicine is a linguistic enterprise. Yet clearly, any syatQIP OP C C E P g e K ) . n senseinferences that considered such a truth when reasoning about eating would be making a mistake.One might say Ulat the phenoeenolqtcrl lrvrL of the anraly_sis was m n g even thourgh all the InF'amznces it: !!ad8 ware UueThe stme w u l d be true of any W . I . system that wade everyday inferences about physical objects by mnsiQaring their quantum structure.by uovirsg the hands to the mukh, and it might be argued that yven ulis is goisrg too far ftom the '@aaningl of eating, whataver that m y bar towsrds generally true information about ma act which, if always inferred &ut a U acts of qating, w i l l carry the systesrs nruamageably fax. But Chatniak's decoupling has the effect of completely separating these two closely related liniguistic phenomena i n what seems to me an unraallstic aanner. H i s system does inferencing to resolve pronoun ambig2 uttfes, while sense ambiguity is presumably to be done in the future by and spatial loeatioqw amow others. As wa saw earlier, tiha B1serhinatian rnvolved in actual analysis is a matter of a p c i E y i n g wry delioat. M h s Q suggests 'gunt as the default value of the instrument of %he action of shooting, but I would claim that, in an example like the earlier 'He shot her w i t h a c o l t ' , we heed to be able to see in the structure assigned whether or n o t what is offered as the apparent instrument is in fact an instrument and whether it 'is the default or riot.In other words, we need sufficient structure of application to see not only that 'shcotlng1 prefers an instrument &at is a gun, but also why it will chaose the sense of 'colt1 thatcis a gun rather than the one which i s a horse.ATtlx>ugh Schank sametinee writes of a system making 'all possible1 inferences a8 it p10ceBd8 though a textt this i e not in fact the heart o t tho dispute, since no one would want ta defend my atmng definittior oL the tom 'all poesibla infetences'.Chacniakqs argument 4s that, unless certain fornard inferences w e made during an analysis o f r say, a e -r yforward inferencest that is, that are not problem-driven; not made in rerrpcnss ta any particular problem of analy.ysia then known to the ayrtsmthan, ar a matter of empirical fact, the system will not in general be able t o solva srPbiguity or rofetence ptoblems that arise later, because it will never in fact be possible t o locata (while looking backwards a t the text, as i t were) the points Ohere those forward inferences ought to have been made.This i s r in very crude summary, Charniak's case against a purely ptoblw-driven inferencer in a natural language under-stander ,A ditficulty w i t h this.argument is the location of an axample of t e x t that c~nffrms the pofnt in a ncn-contentious manner.Chatniak has found an excerpt ftcm a book describing the l i f e of apes in which it is indeed hard to locate the reference of a particular pronoun in a given passagQ. Chamiak's case is that it is only possible to do so i f one has made eertaln inon-prublm occasioned) inferances earliez in the story.nuabet QE readers find it quite hard tb refer that particuXe pronoun anywayl which might s w e e t that, the t e x t was simply badly written. (iv) In terns of the l f n q u i s t i c and or p s y c h c l c g f c a~ plausibility of the proffered system of representation.Oversimplifying considerablyr one might say that Charniakls system akpeals mostly to (ii) and somewhat to (i) and (iv); Winogr3S1s to {iii) and scmewhat to the other three categories; Colbyls (as r e g a d s i t s natural language, rather than psychiatric, aspects) appeals almost entirely to (iii);Simmons largely to (iv) , and Sthank's and my own to dif f ereng mixtures of (ii), (iii) and (iv) .In the end, of course, only (iii) counts for enpiricistsu but there is considerable d i f f i c u l t y in getting all parties to agree to the terns of a t e s t . * A cynic might say Chat, in the end, a l l these systems analyse tho setltenres a t i t they analwe orl to put the same p o i n t a l i t t l e more W w t e t i c a l l y , these is a sense i n which systerms, those described here and tho st^ elsewhereb each define a natllcleal, languaqe, namely the one to which it applies.The difficult question is the extent to'which those mnv and mall natural lacpages resemble E n g l i~h . | The Last section ,atreased areas of cuzrsnt disagreement, but therew~u18, i f vote& mse M e n r be tmnsiderable agreement atmt~g A.X. workers on natwdl, language about where the large problems of the immediate future a : tt,e need for a g o 4 memory mdel has been stressoa by Schank (197-la) , and m y would add the need for an extended procedufbl theory of t e x t s , rather of individual example sentences, and'far a more sophisticated theory of reasons, causes, and motAFhs for use i n a thwry of understanding.Many Ptight also be pezsuaded to agree on the need to steer between the ScylLa of t r i v i a l first generation Fmplementatfons and the Charybdis of u t t e r l y fantastic ones.By the lattet, f mean projectfi that have oeen sericuly dfscussed, but never implemented for obvious reasons, that would, ray, enable a dialogue program to discuss whether or not a participant fn ufvlul o-ty '$%kt qllilt~', end if 80 why.n i e last disease has socpatbes had as a lrajor syrpptoCll an extensive use t -f the w r d 'praqmat-cs' (though this ern also indicate quite benign condf tioas in other cases) , along wdth the implicit claim that lsemant ics has been salved, t~) w e should get on w i t h the pragmatics'. It s t i l l needs repeatthat there bs rn sense whatever in which the semantics of natural language has been solved. It is still the enoxmaus barrier it has always been, even if a feu dents in its surface are beginning to appear here and What is maant by 'stock1 i s clearly the stock piece of the gun, but any preference system like mine that considers w e two sansss of ?stockt, and sees t h a t an edible, soup, sans@ of 'stock1 is the preferred object of the action ' t a s t e , w i l l infallibly opt for the w n s n g sense, Any £ g a m e or expectation s y s t m is pmaa to the same general k i d df countex-exmphe,In particular cases like this it is easy to suggest what might be done: here we might suggest a preference attached to the formula for any; thing t h a t was essentially pareof aslother thing (stock = 'part of gun' in @is casej, so that a local search was made whenevex the 'part-of1 e n t i t y was mentioned, and the satisfaction ofmt search w u l d always ' b m~p ---------- | Main paper:
winograd's understanding system:
Winograd's systea runs a s a dialogue, i n r e a l time, between a human Winogradas parsing is top down, and depth f i r s t , with na automatic back up. The parsing progrim fur each griuxnatical c a t c q~r~~ is n functional definition in PRfXZWWR, which can be stated either a s &VP, fcr SEWPEHCE, or as a f l o w -c h a r t as below for VP:DEFIWJ pxqram VP R E m f a i l u r e Yes LHere is Winograd's own account of the start of this top-down parsing procedure for the sentence "Pick up a red block" (where t h e material i n C 1 is added explanation and not Winogradas o m ) :"The CLRUSE program looks at the first wrd, to decide what w r i t theIf it sees an adverb, i t assumes the sentence begins w i t h a single-word modifier tslowly, Jack l i f t e d the book] ; iP it sres a preposition, it looks .for an i n i t i a l PREP6 Con top of the h i l l stood a tree]If it sees a BINOBR, it calls the CLAUSE program to look for a BOUND CLAUSE Since the next word is not: "do" it checks the n e x t word i n the input ( i n U~is case still the first word) : o see whether it is the i n f i n i t i v e form of a verb.C~eforeIf SO, it is to be attached to the parsing tree, and given the additionql feature MVB (main verb). The current structure can be diagramad as:(CLAUSE MAJOR) (VG IWER) (VB MVB INF TRANS ,WRT --------------pickT W S AND VPRT cmfw fran the definition of the word "pick" when we zalled the function P M S q for a word.4'Mter this syntactic parsing, a number of "semantic specidlists'' attach cwbantic structures to specific syntactic dnes.I n the case of "a red cube", the f o l l o~i n g structure i s built up by an NP "semantic specialist"Secondly, there are two rrpes of verb definition, s e m a n t i c and inferentialWinograd does not give the semantic definition for "pick-up", but here is t h e one f o r "grasp" which is a closelb-related verb. SIZE LOCATlcW \' corn MIbWTB ---------------------- M t w (--------- C ' 5 m RED U C ) f \NITS CaEM bTAC # PifiSoB--( '~~~S~~~~------------------------ rrg& , - I N ( --I T~U FkW1Z.Pwhich says e~sentiallp that grasping is scmathinq dorle by an animate entity to a oanlpulabla one (flret line).More of the real content of such actions is found in their inferential definition.Here is the one fox '"pick-up":(CONSE TC -PICKUP I X ) (PJCKUP xr ( r n ( W S P ?X) T i i E o r n ) (C.XXAL (RAISEHAND THEOREMS)TSlis definition allows the program to actually carry out the "pick-up" caamand if it is possible to do so in the simulated world, as it would not be, for example, if -re were already a block dn top of the red one..such as GRASP and RAISEHAND each of which must be carried out in order that sameming may indeed be p i c k e d up.fesential definitions: the one given for GRASP, for example, i s somewhat differant from i t s "C1(&EANS1' definition g i v e n above, although the inferential iPefini,tiona are aim, i n sane se-e, definitions of meaning as wall as prog r a~a s for actually carrying out the associated conmands.One reason for the enormous impdct of this work was that, prior to its appearance, A 1 work was not very l i n g u i s t i c a l l y interesting, while the eystems of tho linguists had no place far the use of inference and real world knowledge.was able to breed considerable results. where it is c l e a r that the answer is both definite, and that fhding i t requires some inferential manipulation of genaralirrtiens d w u t the world.3ha reader should ask himself at this j m i n t h o w he kwwe the nsrxact refemnt of the pmnaun in that sentanca.$ m e Discussion of S W L U S~J -fax, the reaction ts Winugqad \s work has L m t k wbSrSrla ~ineri t heal. What would crAitics find to attack i f tthtzp M~S F 90 a9nde.I: F i r s t l y , that ~~i n e g r a d ' s linguistic system is highly w n s r~~? a t i w , and that UIQ distinction between 'syntax' and 'semantics' m y not L w necessary at all.a way *.at m u l d make it fnextensihle, to any genexal, real tmrld, situatioh.Suppose 'block' were allowed mean 'an obstrxxctidnl and ' a nental ir,k,i*i t i o n ' , as well as ' a cubic object'. It is dauktful whether Winograd's features and rules could express t h e ambiguity, and, nore i m~r t a n t l y ,whether the simple s t r u c t u e s he manipulated c o u l d decide correctly between t h e a l t e r n a t i v e meanings i n any given c~n t e x t of user Again, f a r more sophisticated and systematic case structures than those hg used might be needed to resolve the ambiguity of 'in1 in "He ran the mile i n five minutes and Ire ran t h e m i l t . in a pawr kwg , as we11 as 'tllu ccmbination ~, l f case w i t h w r d sense a m h i p i t y i n 'Ho p u t t h o key in t h e l~r k ' (door lock1 and 'Be threw the key in the luck1 (river leek).The b l o c k s mrld is also strongly deduct-ive and l w i c a l 1 y closed, S f g r a v i t y were introduced into i t , then anything supkwrted t h a t war pushed i n a c e r t a i n way would have, logically have, t o f a l l . B u t t h e cazmon sense wlorld, of ordinary language, i s n o t like that: in t h~ 'waaen and soldiers'example given e a r l i e r , .the pronoun ' s e v e r a l 1 can be s a i d to be resolved u s i n g same generalisation such as ' t h i n g s shot a t and hurt tend t o fall'There are no logical 'have to's1 t h e r e , even though the meaning of the ~r onoun i s perfectly d e f i n i t e .Indeed, it might be argued that, in a sense, and as r~a f i s its semantics, Winograd's system is n o t about n a t u r a l language a t a l l , but a b u t the -technical question of how goals and sub-goals are to be crganised in a problem-solving system capable of mani~ulating simple physical ehjects.If n reeqber, for example, that tfre key problem that brought b w n the emrr#rur work on mchiae translation i n the Fiftiea and Sixties, was that of the eensa d i g u i t y of nattWi.l. language wprds, then we will look in vain to SHRDLU Lor any help w i t h t h a t problem. There seem? to be only one dear exmaple of aur aLnbiguous mrd in the whole system, namely that of 'curitairat as it appears in 'The b o x contains a red block' and rhe stackAgain, i f on@ glahces back a t the definition of 'pick-up' quoted &ve, ana can see t)cdt it. i e in fact an e%pression rrE a prcrcedure f~r picking up an ob5ect i n the SHR~LU .yet=.one, understand t h e p~r f e~t l v ordinary sant;ence 'I picked up my bags £tom the plstforn, and ran for the train', let alone any sentenco n o t &out a physical action performable by the hearer. T ' k difft!zeace i s t h a t Molods t a k e s a much more logico-sanantic i n t q r e t a t i u a of t h a t s l o g a n than does Winoqrad.In partisular, fbr W s t h e meaniq of an L n p~t utterance to h i s system is t h e procedures w i t h i n t h e system that raJ+ipulata tfr t r u t h conditions of t h e utterancfe and e s t g b l i s h its t r u t h value.To p u t t h e m t t e x crudely, f o r Moods an assertion has x i meanhq if hie.system cannot e t a b l i g h its t n t h o r falsity. should show, as it were how i t is a c t u a l l y t o be applied to l a n g u a~e , b u t that is n o t the s-e as demanding that it should be w r i t t e n in a ~r~c~d u r a 3 language,line PLANNER. I shall r e t u r n to this last pifit l a t e r . The i m p r t a n t wards these are 'lt-k f e y ' , which suggest that t 3 1 r > r~ may w e l l bo c o n f i r m i n g hints to be found in the :=tory and, i f t4lerc are, than this tentative, partial, inference is cursect, and TE a c h i n of demons can 'reacht one of the passiblo xaferente i n a story then there is a suct+ass registered and the ambiguity of the corresponding pronoun is resolved.
second generation svstems:
It can be seen that the information encoded in the system is of a highly specific sortin the present case it is not about containers as such, and how to g e t their contents out, but about Piggy B a n k s in particular, and everything relies on that partfcular knowledge having been put i n . Thus, the apparently English words in the PB-OUT-OF Pound b e then tied directly, or via these inference rules, to response patterns which are generated. (Hendrix et a1 l73) , the ather apprdach is adopted, w i n g a primitive action B&CI.fANGG inetead of 'transfer'.EnormousThIe implementation under c~nstruction is a front-end ,oarsex of the Woodat augmented t r a n s i t i o n network type (see Woods '701, and a ganefation system going gram the s e a~t i c networks to surface strings described8in detail i n ( S W n s and S1ocu.m ' 7 2 ) .SimPons has also given considerable an & take + book Here 'pf itldicatea past, and is the aepndendy symbol liking a PP to We ACT ( ' t a k e ' ) which i s the hub of the conceptualization, as w i t h Simon& ?'he ' 0 ' indicates the objective case, marking the dependence of the object PP on the central ACT.There is a carefully constructed syntax of linkages between the conceptual categories* that will be describpd only in part in what follows.The next stage of the notation involves an extended case notation and a set of primitive ACTS, as well as a n q e r of it:ems suoh as PHYSWNT which indicate ather stqtes, and items of a fairly simplified psychological t h e o r y (the dictionary entry for 'advise', for example, contains a subgraph t e l l i n g us that Y 'will benefit' as part of the meaning of ' X advises Y ' can consider t h i s entzy als an active 'frAxte-like~ object seeking f i l l e r itms i n any context In which it is activated. Thus, in the sentence 'John sho the girl w i t h h riflet, the variables w i l l be f i l l e d in frcm context and the case inference will be made f m the main act PROPEL, which is that its h s t r u u e n t is lkSOVEI GRASP or PFtOPRL, and so we w i l l arrive a t the whole conceatualieation: Sot in building a template for 'John drinks,wine', the whole of the above tree-formula for 'drinks' would be piaced at the central action node, another tree structure for 'John' at the agent node and so on.John PROPEL 4 - bullet <-The complexity of the system comes from the way i n which the formulas, considered as active entities, dictate how other places hn the same template should be filled.Thus, the 'drink1 formula above can be thought of as an entity that fits at a template action node, and seeks a liquid object, that is ~ say a f~rmula w i t h (FLOW STUFF) as its right-most bzanch, to put at the object noda of the same template. This seeking is preferential, in that formulas not satisfying that requirement will be accepted. but only if nothing misf a c t a n c a lXZ'-fotUEa. TIie -€Elliplate Uif ly esWIisned Tm 3 Tragm e n t of t e x t is the one in which the most formulas hive their preferences s a t i s f i e d . a general principle at work here, t h a t the r i g h t interpretation 'says the least1 in inforreation-carrying terns, T ) r h wry simple device is able to do much o f the work o f a syntax and wxdysnse wzibigukty resa1vi;tap pxagraa P n a r c u s q~e , LZ the a,mteme k d been 'John drank s whole pitcher1, the fomulr tor th. 'pitcher of klquidb w u l d hawe h e n pr.rsEerreCI to that for thar human, s f m the subf~mS;a (FLOW STUFF) could be apprtopriateAy located uithirr i t . The extracted templates express information a lready implicitly preser.t i n the text, wen though many of them are partial inferences: anes that may not necessaxily, be true. Whether or not such a, system can remain s-le with a WrutderabLe vecabulary. of say several thousand words, has y e t to be t r r t d .ft will ba ovidrnt tso any reader that Zha laat t w a systems described, Bch.nktm ud my moun, share a great deal in cccpmon. whether we ahould expect advance in understanding natural language from those tackling the problems head on, or those coroncerned to build a 'fr8ntandv. It i~ cLtt,xly the case thata n p i e c e coulL bp esrcntial to the understanding of sane story. The question is, does it follow that the epehifict.tion, organieation and formalization of that knowldge l a &a studf oP l : . a g a , because i f it is then all human enquiry f t m physics and history to medicine is a linguistic enterprise. Yet clearly, any syatQIP OP C C E P g e K ) . n senseinferences that considered such a truth when reasoning about eating would be making a mistake.One might say Ulat the phenoeenolqtcrl lrvrL of the anraly_sis was m n g even thourgh all the InF'amznces it: !!ad8 ware UueThe stme w u l d be true of any W . I . system that wade everyday inferences about physical objects by mnsiQaring their quantum structure.by uovirsg the hands to the mukh, and it might be argued that yven ulis is goisrg too far ftom the '@aaningl of eating, whataver that m y bar towsrds generally true information about ma act which, if always inferred &ut a U acts of qating, w i l l carry the systesrs nruamageably fax. But Chatniak's decoupling has the effect of completely separating these two closely related liniguistic phenomena i n what seems to me an unraallstic aanner. H i s system does inferencing to resolve pronoun ambig2 uttfes, while sense ambiguity is presumably to be done in the future by and spatial loeatioqw amow others. As wa saw earlier, tiha B1serhinatian rnvolved in actual analysis is a matter of a p c i E y i n g wry delioat. M h s Q suggests 'gunt as the default value of the instrument of %he action of shooting, but I would claim that, in an example like the earlier 'He shot her w i t h a c o l t ' , we heed to be able to see in the structure assigned whether or n o t what is offered as the apparent instrument is in fact an instrument and whether it 'is the default or riot.In other words, we need sufficient structure of application to see not only that 'shcotlng1 prefers an instrument &at is a gun, but also why it will chaose the sense of 'colt1 thatcis a gun rather than the one which i s a horse.ATtlx>ugh Schank sametinee writes of a system making 'all possible1 inferences a8 it p10ceBd8 though a textt this i e not in fact the heart o t tho dispute, since no one would want ta defend my atmng definittior oL the tom 'all poesibla infetences'.Chacniakqs argument 4s that, unless certain fornard inferences w e made during an analysis o f r say, a e -r yforward inferencest that is, that are not problem-driven; not made in rerrpcnss ta any particular problem of analy.ysia then known to the ayrtsmthan, ar a matter of empirical fact, the system will not in general be able t o solva srPbiguity or rofetence ptoblems that arise later, because it will never in fact be possible t o locata (while looking backwards a t the text, as i t were) the points Ohere those forward inferences ought to have been made.This i s r in very crude summary, Charniak's case against a purely ptoblw-driven inferencer in a natural language under-stander ,A ditficulty w i t h this.argument is the location of an axample of t e x t that c~nffrms the pofnt in a ncn-contentious manner.Chatniak has found an excerpt ftcm a book describing the l i f e of apes in which it is indeed hard to locate the reference of a particular pronoun in a given passagQ. Chamiak's case is that it is only possible to do so i f one has made eertaln inon-prublm occasioned) inferances earliez in the story.nuabet QE readers find it quite hard tb refer that particuXe pronoun anywayl which might s w e e t that, the t e x t was simply badly written. (iv) In terns of the l f n q u i s t i c and or p s y c h c l c g f c a~ plausibility of the proffered system of representation.Oversimplifying considerablyr one might say that Charniakls system akpeals mostly to (ii) and somewhat to (i) and (iv); Winogr3S1s to {iii) and scmewhat to the other three categories; Colbyls (as r e g a d s i t s natural language, rather than psychiatric, aspects) appeals almost entirely to (iii);Simmons largely to (iv) , and Sthank's and my own to dif f ereng mixtures of (ii), (iii) and (iv) .In the end, of course, only (iii) counts for enpiricistsu but there is considerable d i f f i c u l t y in getting all parties to agree to the terns of a t e s t . * A cynic might say Chat, in the end, a l l these systems analyse tho setltenres a t i t they analwe orl to put the same p o i n t a l i t t l e more W w t e t i c a l l y , these is a sense i n which systerms, those described here and tho st^ elsewhereb each define a natllcleal, languaqe, namely the one to which it applies.The difficult question is the extent to'which those mnv and mall natural lacpages resemble E n g l i~h .
cdnclueiqn:
The Last section ,atreased areas of cuzrsnt disagreement, but therew~u18, i f vote& mse M e n r be tmnsiderable agreement atmt~g A.X. workers on natwdl, language about where the large problems of the immediate future a : tt,e need for a g o 4 memory mdel has been stressoa by Schank (197-la) , and m y would add the need for an extended procedufbl theory of t e x t s , rather of individual example sentences, and'far a more sophisticated theory of reasons, causes, and motAFhs for use i n a thwry of understanding.Many Ptight also be pezsuaded to agree on the need to steer between the ScylLa of t r i v i a l first generation Fmplementatfons and the Charybdis of u t t e r l y fantastic ones.By the lattet, f mean projectfi that have oeen sericuly dfscussed, but never implemented for obvious reasons, that would, ray, enable a dialogue program to discuss whether or not a participant fn ufvlul o-ty '$%kt qllilt~', end if 80 why.n i e last disease has socpatbes had as a lrajor syrpptoCll an extensive use t -f the w r d 'praqmat-cs' (though this ern also indicate quite benign condf tioas in other cases) , along wdth the implicit claim that lsemant ics has been salved, t~) w e should get on w i t h the pragmatics'. It s t i l l needs repeatthat there bs rn sense whatever in which the semantics of natural language has been solved. It is still the enoxmaus barrier it has always been, even if a feu dents in its surface are beginning to appear here and What is maant by 'stock1 i s clearly the stock piece of the gun, but any preference system like mine that considers w e two sansss of ?stockt, and sees t h a t an edible, soup, sans@ of 'stock1 is the preferred object of the action ' t a s t e , w i l l infallibly opt for the w n s n g sense, Any £ g a m e or expectation s y s t m is pmaa to the same general k i d df countex-exmphe,In particular cases like this it is easy to suggest what might be done: here we might suggest a preference attached to the formula for any; thing t h a t was essentially pareof aslother thing (stock = 'part of gun' in @is casej, so that a local search was made whenevex the 'part-of1 e n t i t y was mentioned, and the satisfaction ofmt search w u l d always ' b m~p ----------
:
To euntey an enewetic field like thie one is inevitably to laavo a great deal of excellent work unextiminad, a t least if one i a going to do more than give a paragraph to each research project.I have left out of cotasideration at least seven groups of projects:(1) Early work in Artificial Intelligence and Natural Language that has been sumeyed by Wfnograd (1973) and Simmons (1970a) among others.(2) Work by graduate students of, or intellectually dependent upsn that o f , people discussed in same detail here.( 3 )Wxk that derives essentially from projects described i n detail here. This embraces several groups interested i n testing psychological hypotheses, as r e 1 1 as others constructing largescale systems for speech recognition. I have devoted no space to speech recognition as such here, for it seems to me to depend upon the quality of semantic and inferential understanding as much as anything, and so I have concentrated upon this more fundamental task.Work on language generators, as opposed to analysers and understanders.They are essential for obtaining any testable output, but are thearetically secondary. That would be wholly inappropriate i n the present s t a t e of things. A g r e a t deal o f work is being done a t the moment, and many of the p r i n c i p a l researchers change t h e i r views on very fundamental questions between one paper and t h e next without drawiw any a t t e n t i o n t o the f a c t . Cheap self-contradictions and changes of mind are a l l too easy t o f i n d , so c r i t i c i s m and smparisons are best drawn w i t h a very broad brush and a l i g h t stroke.
Appendix:
| null | null | null | null | {
"paperhash": [
"colby|pattern-matching_rules_for_the_recognition_of_natural_language_dialogue_expressions",
"hendrix|language_processing_via_canonical_verbs_and_semantic_models",
"schank|the_fourteen_primitive_actions_and_their_inferences.",
"russell|semantic_categories_of_nominals_for_conceptual_dependency_analysis_of_natural_language.",
"schank|primitive_concepts_underlying_verbs_of_thought.",
"simmons|some_relations_between_predicate_calculus_and_semantic_net_representations_of_discourse",
"goldman|computer_generation_of_natural_language_from_a_deep_conceptual_base",
"riesbeck|computational_understanding_:_analysis_of_sentences_and_context"
],
"title": [
"Pattern-Matching Rules for the Recognition of Natural Language Dialogue Expressions",
"Language Processing Via Canonical Verbs and Semantic Models",
"The fourteen primitive actions and their inferences.",
"semantic categories of nominals for conceptual dependency analysis of natural language.",
"Primitive concepts underlying verbs of thought.",
"Some Relations Between Predicate Calculus and Semantic Net Representations of Discourse",
"Computer generation of natural language from a deep conceptual base",
"Computational understanding : analysis of sentences and context"
],
"abstract": [
"Man-machine dialogues using everyday conversational English present problems for computer processing of natural language. Grammar-based parsers which perform a word-by-word, parts-of-speech analysis are too fragile to operate satisfactorily in real time intervieus allowing unrestricted English. In constructing a simulation of paranoid thought processes, we designed an algorithm capable of handling the linguistic expressions used by interviewers in teletyped diagnostic psychiatric interviews. The algorithm uses pattern-matching rules which attempt to characterize the input expressions by progressively transforming them into patterns uhich match, completely or fuzzily, abstract stored patterns. The power of this approach lies in its ability to ignore recognized and unrecognized words and still grasp the meaning of the message. The methods utilized are general and could serve any \"host\" system uhich takes natural language input.",
"A natural language question answering system is presented. The system's parser maps semantic paraphrases into a single deep structure characterized by a canonical verb. A modeling scheme using semantic nets and STRIPS-like operators assimilates the sequence of input information. Natural language responses to questions are generated from a data base of semantic nets by \"parsing\" syntactic rules retrieved from the lexicon.",
"In order to represent the conceptual information underlying a natural language sentence, a conceptual structure has been established that uses the basic actor-action-object framework. It was the intent that these structures have only one representation for one meaning, regardless of the semantic form of the sentence being represented. Actions were reduced to their basic parts so as to effect this. It was found that only fourteen basic actions were needed as building blocks by which all verbs can be represented. Each of these actions has a set of actions or states which can be inferred when they are present.",
"A system for the semantic categorization of conceptual objects (nominals) is provided. The system is intended to aid computer understanding of natural language. Specific implementations for \"noun-pairs\" and prepositional phrases are offered.",
"In order to create conceptual structures that will uniquely and unambiguously represent the meaning of an utterance, it is necessary to establish ''primitive'' underlying actions and states into which verbs can be mapped. This paper presents analyses of the most common mental verbs in terms of such primitive actions and states. In order to represent the way people speak about their mental processes, it was necessary to add to the usual ideas of memory structure the notion of Immediate Memory. It is then argued that there are only three primitive mental ACTs.",
"Networks can be used to represent syntactic trees of the semantic relations that hold between words in sentences. They can be alternately symbolized as association lists or conjoined sets of triples. A semantic net represents a sentence as a conjoined set of binary predicates. An algorithm is presented that converts a semantic network into predicate calculus formalism. The simpler syntax of semantic network representations in contrast of ordinary predicate logic conventions is taken as an argument for their use in computational applications. \n \nDescriptive Terms: Semantic networks, Predicate logic, Natural language, Computational linguistics, Association lists.",
"Abstract : For many tasks involving communication between humans and computers it is necessary for the machine to produce as well as understand natural language. The authors describes an implemented system which generates English sentences from Conceptual Dependency networks, which are unambiguous, language- free representations of meaning. The system is designed to be task independent and thus capable of providing the language generation mechanism for such diverse problem areas as question answering, machine translation, and interviewing.",
"Abstract : The goal of this thesis was to develop a system for the computer analysis of written natural language texts that could also serve as a theroy of human comprehension of natural language. Therefore the construction of this system was guided by four basic assumptions about natural language comprehension. First, the primary goal of comprehension is always to find meanings as soon as possible, Other tasks, such as discovering the syntactic relationships, are performed only when essential to decisions about meaning. Second, an attempt is made to understand each word as soon as it is read, to decide what it means and how it relates to the rest of the text. Third, comprehension means not only understanding what has been seen but also predicting what is likely to be seen next. Fourth, the words of a text provide the cues for finding the information necessary for comprehending that text."
],
"authors": [
{
"name": [
"K. Colby",
"R. C. Parkison",
"Bill Faught"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"G. Hendrix",
"C. Thompson",
"Jonathan Slocum"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. Schank"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"S. Russell"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. Schank",
"N. Goldman",
"C. Rieger",
"C. Riesbeck"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. F. Simmons",
"Bertram C. Bruce"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"N. Goldman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"C. Riesbeck"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null,
null,
null,
null,
null
],
"s2_corpus_id": [
"219302766",
"10422416",
"61126867",
"53560660",
"141957545",
"12191250",
"56767024",
"60975873"
],
"intents": [
[],
[],
[],
[],
[],
[],
[],
[]
],
"isInfluential": [
false,
false,
false,
false,
false,
false,
false,
false
]
} | null | 588 | 0.003401 | null | null | null | null | null | null | null | null |
e525c6d3490ae0c0d7a80787358adee893e2ff21 | 219305220 | null | A Computer Simulation of {A}merican {S}ign {L}anguage | Bowling Green S t a t e University Ohio 43403 Stokoe (1960) has d e s i g n e d a model of the structure of t h e American Sign Language (ASL) which is amenable to computer simulation. He has proposed that signs comprising the ASL lexicon are composed of three basic aspects, location (TAB), hand configuration I (DEZ) , and movement (SIG) . He has identified a finite number of each of these elements, and he has proposed that they may be combined in various ways to constitute recognizable and meaningful signs. Recent reformulations have vent u r e d some modifications ( e . g . , Stokoe, 1972), but the basic approach remains the same. Such a conceptualization of ASL implies that if a computer were furnished a set of each of these types of elements, it ought to Acknowledgments. This investigation was supported by N I H Research Grant NS-09590-05 from the National Institute of Neurological Diseases and S t r o k e . We thank the J. | {
"name": [
"Hoemann, Harry W. and",
"Florian, Vicki A. and",
"Hoemann, Shirley A."
],
"affiliation": [
null,
null,
null
]
} | null | null | null | 1976-02-01 | 1 | 2 | null | null | null | null | likely to be nonsense" in ASL. In our data, the 11 DEZ, 8 TABs, and 7 SIGs yield potentially 11 x 8 x 7 or 616 signs.Over 600 of them are nonsense. This indicates that most signs in ASL differ from one a n o t h e r i n more than one distinctive feature. Also, since many signs change DEZ and involve more than one SIG, there is a low probability of confusing one sign with another, even when the signs are presented out of c o n t e x t . | null | Main paper:
american s i g n language:
likely to be nonsense" in ASL. In our data, the 11 DEZ, 8 TABs, and 7 SIGs yield potentially 11 x 8 x 7 or 616 signs.Over 600 of them are nonsense. This indicates that most signs in ASL differ from one a n o t h e r i n more than one distinctive feature. Also, since many signs change DEZ and involve more than one SIG, there is a low probability of confusing one sign with another, even when the signs are presented out of c o n t e x t .
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 588 | 0.003401 | null | null | null | null | null | null | null | null |
90e9a3d9efe1c04a7f18c9ff4d751afae151aab7 | 219306908 | null | Analysis of {J}apanese Sentences | The system w ~u l d b e much improved i f accoarpanied by afi a p p r o p r i a t e achema f o r representing c o n t e x t . v Case g s a m ~r ' ~e n t e n c e -a n a l y s i s t h e o r i e s s u c h a s those of Fillmore (1968) and Celce-Muscia (1972) a r e b a s e d on t h e semantic r e l a t i b n s k i p s between verbs a d nouns -e v e n t s and c o n c e p t s , R. F . Simmaas (1973; 19751, n (3973), D, E. Rumelhart (1973) and s o on f o l l o w t h e s e t h e o r i e s to represent knowledge and context i n t h e i r systems. W e a l s o adopted case s and nsodifded i t to account f o r Japanese sentences. W e represent cantext fn the form of a @ @ a n t i c network. An input sentence is transformed into a c ~r x e e p o n d i n g deep saee | {
"name": [
"Nagao, Makoto and",
"Tsujii, Jun-Ichi"
],
"affiliation": [
null,
null
]
} | null | null | null | 1976-02-01 | 0 | 0 | null | null | null | null | The o r g a n r a n t i o n of a nature1 l a n g u e g~ (Japanese) p a r s e r $ 8 d~b r r In m s t a p p r~a c h e~ to t h e u n d e r s t a n d i n g of n a t u r a l barlgunpe tl~roupk a r t i f i c i a l i n t e l l i g e n c e , schemas which e n t a i l r i g i o u r c o n t e n t i o n t h a t i n t u i t i v e r e a s o n i n g 1s c o m p l e t e l y b a s e d on t h e l a n g u a g e a c t i v i t y i n the human b r a i n A s s o c i a t i v e f u n c t i o n s r e l a t i n g t o s e m a n t i c s i m i l a r i t i e s between words, s e m a n t i c d e p t h of a n i n t e r p r e t a t i o n and p r o b a b i l i t y of a s s o c i a t i v e occurrence of e v e n t s are i n h e r e n t f a c t o r s i n i n t u i t i v e u n d e r s t a n d i n n a The f i n a l a n a l y s i s produced by o u r pamor i a a srtwntic netwvtk,T h i s c o u l d h e used fox the i n t e r n a l , r a p s @ s e n t a t i s n of data i n a q u p g t t~f tanswering s y s t e m ox a s an interniwdiete e~p r~s s i u n i n '61:mchin~ t x a n s l , s t i u r t H~wu\!r, V a r i o u s syntactic p a t t e r n s i n J a p a n e s e w i t h appropriate s e w n t l c and c o n t e v t u a l c h e c k l n g f u n c t l o h s , PLATON is p r e s e n t e d i n more d e t a l l I n a n o t h e r papec b y Nagao and T s u j i l ( 1 9 7 6 ) .2 ) I n t h e dictionary a r e s The f a l l a w i n g k i s t shaws t h e relations u s e d i n t h e network:( 1 Deep C u e R e l a t i o n s . ACT', OBJ, P U C E , TIHE --A deep case r e l a t i o n connects an S-node with i t s argument C-node, S e l a a n t~c and C o n t e x t u a l f u n c t i o n s a r e propmmned i n L I S P 'Fathart ka r famili&r a pie* In order to identify a pesaon i n d i c a t e d by thc word, mi h r v t EO haw whose father he i e . In the chemical field we r a n a u i l y find auch nome ( e . 8. , 'weight' , ' t e m p e r a t u r e ' , ' color' , qnnd 'mess ' ) *E"hrdss &re e r J l c d A t t r i b u t e Wslane. Their m a n i n g e are deacrilaed i n a d i f f e r e n t way from that of ordinary n o w e , The description (t4F N-A) shows the noun belongs t o t h e group 0 f a t t r l b u t owterJl~aI ( ( SP> (AmR(STATE) (MASS) (COLOR) ( % W E ) ---------1 ) l i q u i d ( ( SP m a t e r i a l ) (ATTR (STATE *LIQVID) ( W E NIL ) ) )( WRISA ( (WP W-A) (A-ST W L W kWSS LENGTH A 1 s LZB (SP ZQKUSEZ RY8 b > )a t t r i b u t e q u a n t i t y aft Attribute noun8 a m further ~Z a s e i f i e d i n t a twa groups, quantitative and q u a l i t a t i v e A g u c * a i t a t i~g a t t r i b u t e noun cannot b e a case element a f a verb wtaech regwfres g u n t i t n t i v e nouns. Tna v e r b s FUERU [increase) and MERU (deciease) are such &xamples gf verbs. Liquid' is another r e l a t i o n a l noun. The J a p a n e s e word whirh csrreapoads t o ' l i q u l a is EKITAI. While 'liquid' in English can be e l t h e r a n o w or an adjective, ERITWI i n J a p a n e s e is c a~r g o r r z e d s y n t a~t~c a l J y as a n o m . But @emantically EKLTAF has two d i f f e~e n t meanings, one c o r r e s p o n d i n g -1 3 -( I R O ( (NF M-A) (A-ST G O L~~R ) (SP ZOKUSEL SHITSU 1))In our system t h e meaning QE a verb is d~~c s i b e d by settrirzg up k;emsal r e l a t i o n a l slots whLch will be fillad i n by nouns, JR this sen&@ the w a l i n g of e wcrb is n o t confined t o i t s e l f , b u t t w r e l r t g d to mma.We describa rhase r e l a t i o n s by u~i n g t h e cone caneept i n t r~d u r e d by CCF ( ( ACT MINGEN ) ( OB J KOTAI) ( IN EKITqI) ) human b e i n g s o l i d l i q u i d ( ( ACT NINGM ) ( OB J KOTAI) f l VST) ) human being s o l i d ( ( ACT SAN ) ODB JKINZOKU ) )acid metal -WA - The volume sf t h @ g a s i n c r w s e s -- EQUATIONCb) I O U - -MA - K I I R O I s u l f u r -- -( S U B~: , --- vv i t y , (a) IQiRE-GA M I Z U -0 - NESSLhRU h e -f S 1 M ) water -( O B J ) - h e a tHe heats t h e w a t e r . ( 1 H Z U -GA asra-pasarI _ _ - (b) TNSAN -GA A E N -0 -- T O U S U y d r o c h l o r i c a c i d -(ACT) -z i n c -(OBJ) --me1 t Hydrochloric a c i d pelts zi nc .-SUIJOUKI -MI - U Q V(a) -- -BE - - MTZU -0WSSmW.water -(P8.l) h e a t (Solpeonr) heats water J Ja gas burner. ; FACT is u s e d t o &dieere s e n t e n t i a l cotnplrnuenti~@ra. ARKWRU-MPt,t=NO(a> KOBEe -0 SMSTS~YOwOZUW-NO -- ------eNlsurrv -ro =. . . a, * et., i t -f B B J ) -tYOKO -NI - BE -b QKL'. a l c o h o l lamp s i d r -- -( P U C E ) -- beaker -\ O B J )Put ( h e n one knowa a certain event has occurred, he can a n t i c i p a t e s u c c e~s i v e events that w i l l occur and what changee t h e o b j e c t s p a r t i c i p a t i n g i n t h e evant will undergo. Thie kind of expectation plays an i m p o r t a n t role i n underetanding eentences. Various k i n d s of a s s o c i a t i o n s c l u c t e r c o n c e p t u a l l y (a) ( ADD case a-set-of-(A V)-pairs ) (b) ( DELETE case a -s e t -o f -a t t r i b u t e s ) ( c ) ( CREATE lexical-nam-of-an-objec t a-se t-of- A t y p i c a l zxample u s l n g a CON e x p r e s s i o n i s shown I n F i g u r e 2 6 t --arsund(A V j -p a i r s )r A 'j CI 4 p a d k t D 3 U 4J (Ti QJ u Qi s 0 C -4 'J; ri k 3 w 4 3 fV2 G1 C 4-4 - ul 11 0 4 3 [I) IU k CU s C, -a c a h Z 0 U w 0In t h i s expression one can see the verb TOKASU has two differetlt meanings.One cottesponds t o ' m e l t ' , and the a t h e r t o 'dissolve i n ' . When we a n a l y z e the aentense,MU -6 TQKASU .c~g p e r - We have i d e n t i f i e d s l x t e e n s e m a n t i c a l l y a c c e p t a b l e NOUN NO N O W combinations. These a r e shown i n Table 3. MhU -(F K I F A I h t e s t t u h c i n l i q u i d t ?The noun-nuun r t~m b i n a t i w , t e s t tube-SL7 in' expresses t h e 'place1 in the t e s t t u b e , ***So t h e program w i l l go back t o step (11) . T A f SWT ox\*gen (and) hvdrngcn v 0 k t~~ ( 1 S A N S 0 -WO SHXTSLRJQU -TQ S U S O -80 TAXSEKI -7'0).e4uvgenmass ( a n d ) omgen ' L olume ( 2 ) S M S Q -NO SHLTS'LrRIOU -TQ TAISEKX -(2'0)rsmgen mass ( a n d ) vnlume with Naun-h. I f found, l e t i t b e Noun-2, and go t o s t e p 3 .S t 9 1. The phrase between t h e p o s t p o s i t i o n and Noun-: are analyzed ---1 noun p h r a~e a n a l y s i s . This is now t h e second of t h e two pasallel phrases u n d e r csnshdesa t r o n ,Step -4. The phrase betore the p o s t p o s i t i o n is a n a l y z e d by t h e normal n s m phrase a n a l y e i s -i+6 -o n l y p a r t i a l knowledge about t h e context, and therefore, h i s knowledge i s not coqplete. However, he can u n d e r s t a n d the meanings of s e n t e n c e s before he reads through t h e e n t i r e s e t . This means t h a t one i s c o n t e n t w i t h (1) C o n t e x t i.; e n t e r e d a n t o the ~n t e r m d i a t e term m e m~f y ,incomplete( 2 ) Two k;lnds sf i n t e r m e d i a t e term memry a r e p r e p a r e d . (1) I n Japanese a t h e m word 1s o f t e n o m i Changes o f W S R e p i n n~n p of t h e a n n l~$ i s of S1: ((Nk IT3 Eu' 2 N1))End o f the a 1 1~1 v s I~ 51Br@nninq n f t h e ,~n a l~s i t . To solve t h i s problem we s e t up a t r a p p i n g l i s t TL, The bas i r o r g a n i z a t i o n of TL is show i n Figure 4 . 6 , h t r a p p i n g e l e m n t 'is a t r i p l e t ( t ) They check whether a n e w a t hand Bsn salve the1 problems in TL, Ciii) If it cannot do so, the system adds 1 t o N, t h e f icst element of tbe trapping element. Wen M exceeds f i v e , the t r a p p i n g element is deleted frorp TL. That i s , i t i s decided t h a t t h e problem corresponding t o the t r a p p i n g element can not be solved a t e l l . Before the deletion of a trapping elcmpenr its third element, the function F2, is eveluated. Thus far F2 has cnlJ b~a n used t o provide default values t o sllowu some intet pscrsrlon 1 1s p~n d i n p By using Pbp idea o f TL, we can t;eparate *ridas checking wchuuisari fro& the w i n program. They can be invoked crutanwtitaliu whcn o noun &ppatn i n n senten&@, The idea of TZ. msedles thee af 8. Z'harniak;'~: 'dsw4rtt (J197211,When his rryateffi sncowntars a c o c t o i n word, far cxerirlc, ' p i~ bank', i t c u o e t a~ e demon which t r i e s t o catch from the succeeding sentences any word (e.g., money) related to the key word. Ke fear that unnecessary knowledge w i l t clog the system with e 'combinatariel explosion' resulting from the p r o l i t e ration of demons, Ous trapping elemat i s g u t i n TL only temporarily In t h i e exawple there ate three nouns following the a r t i c l e which can be wcfifled by i t e y n t a c t i e a l l y , We must decide the preferable modlficat i @ pattern by using c~n t e x t u a l informention, In the analysis af a noun phrase, we scan the words one-by-one from left t o r i g h t , When we catch t h e article ' KDNLa, we put it fn the tewarasy stack. The wosd will then be checked t o aee *ether it can modify a noun in the fallowing noun phsasa.Vhea.1 we scan r h~ n o w SH31aNUN ( t s e t tube) in Fbgure 4,7, we check whet'taer the object indicated by it was elready mntioned in the preceding sentences. (2) % U S 0 -a U U ,11-KONO TAfSEKZ -0 volusle (053.51There i r s oxygen.Zn thfs c a e KONO alone desigaates the entity noun SAHSO wh:ich appears in (2) Tha Elrs t usage of KONO has thc following three veoir'ties.(i) SMSO -GA ARU. KOHO W S O -0 ---- There i a D V~P~.The ~xygstl ----The now m d i f i a d by the article is the same noun which appears i n the preceding sentence. The gas mixture ---The a r t i c l e modified a nodnalized form of the f i r s t sentence. The f i r s t sentence l g s t a n t i a t e s the case frame of the verb 'mx' FJe ewduate the N W S description of the case frame and obtain a new lnierenced object 'mixture', whose elem~ents are the oxTgen and the hydrogen. The noun XONWUKITBI modified by the a r t i c l e i s a lower concept noun oi the infesenced n o w (mixture) i n ENS.According to these three v a r i e t i e s , we provide the following three check sautines. The order of checking in shown i n (check 1) Ia there in the l i~t t h e earn noun cre t h e noun modified by KONO.. BEWFZCEHT ---1, and t h e c a s e frames of IRERU ( p u t :in) have t h e cqse PLAw-, W e can p r e d i c t t h a t t h e pronoun KORE ( t h l s , it) f i l l s the PLACE case i n t h e sentence. The semantic d e s c r i p t i o n s a y s t h a t a l o v e r concept naun of c o n t a i n e r ' o r 'Ilquid' i s p r e f e r a b l e as t h e PLACE case of the verb IRERU (put i n ) . The o b j e c t ' w a t e r ' , which i s a lower concept. man of ' l i q ui d v ,is found in NS, and is d e t e m u n e d t o b e the o b j e c t designated bt the pronounWe have some other pronouns and articles in Japanese which are analyzed in ehe same way. We provide d i f f e r e n t LISP f u n c t i o n s f o r different pronouns and put them in the d i c t i o n a r y d e f i n i t i o n 8 of these words.T, Wfn~grad treated the s a w problem in hie excellent system SkSRDLU (1971; 1972) . lbwever, the world which his eys tern can deal w i t h is very h j d t e d . Sn order t o construct a system which can treat a wider range aE ecntcncee, the gystem should be equipped with the schema representing the re$~ei-srar%h$ps between event@ and abject (an event m y iwly tl.u$ occuttrarentllc of new objects et changee in the propertie$ of sbjects). In real. world eeatencee, there e x i s t s m s e complex g h e n o e n a about anagheric expressions and a d~~i o n s of w~r d~ than those treated i n SHRDLU. We do n o t claim t h a t aut eyetern c m t r e a t suck .c&pl.ex pnensmna, but we hope that our system can The s e n t e n c e i s analyzed by the fo1law;ing steps. noun TAISEKI: 1 s a Irstver concept noun of ' a t t r i b u t e ' , which stjtisflcftzs t h e semantlc c o n d i t i o n for the case element. So thls s e n t e n c e is a n a l y z e d :in a straightfoswasd manner, However, because t h e noun TAISEKI 1s an attribute noun, we m u s t f i n d t h e corresponding entity noun. Tkqt I S , we must i d e n t i f j the o b j e c t whose volume is being referred to. As we c a n n o t f i n d such an o b j e c t i n t h e p r e c e d i n g sentences, we set up a p e n d~n g problem I n TL, I3.1 checking t h e i n f l e c t i o n of t h e verb WENUSURU (change) and n o t i n g t h a t it: is i w w d i d t ely followed by a noun, it i s r e c u g n i z e d t h a t t h e s e n t e n c e is an embedded s e n t e n c e modifying khe f o l l o w i n g noun TOKI ( t i m e , when), Fu' e t h e nbes the fdaab s t e p in the a n a l y s i s of a noun phrase, w e check whether there remain relational nauos which have no d e f i n i t e meaglng. If found, we search through NS for words whrch are suitable t o f i l l i n t h e s l o t s of the nome, The ecarchihg process 1s che same as f o r omitted words In siaple seatencea, Sometiaes t h e omitted words e x i s t i n succesding s e n t e n c e s , s o we cczn set up a problem in TL, if w e cannot f i n d an appropriate word in t h e gracedhng aentencea | null | Main paper:
:
The o r g a n r a n t i o n of a nature1 l a n g u e g~ (Japanese) p a r s e r $ 8 d~b r r In m s t a p p r~a c h e~ to t h e u n d e r s t a n d i n g of n a t u r a l barlgunpe tl~roupk a r t i f i c i a l i n t e l l i g e n c e , schemas which e n t a i l r i g i o u r c o n t e n t i o n t h a t i n t u i t i v e r e a s o n i n g 1s c o m p l e t e l y b a s e d on t h e l a n g u a g e a c t i v i t y i n the human b r a i n A s s o c i a t i v e f u n c t i o n s r e l a t i n g t o s e m a n t i c s i m i l a r i t i e s between words, s e m a n t i c d e p t h of a n i n t e r p r e t a t i o n and p r o b a b i l i t y of a s s o c i a t i v e occurrence of e v e n t s are i n h e r e n t f a c t o r s i n i n t u i t i v e u n d e r s t a n d i n n a The f i n a l a n a l y s i s produced by o u r pamor i a a srtwntic netwvtk,T h i s c o u l d h e used fox the i n t e r n a l , r a p s @ s e n t a t i s n of data i n a q u p g t t~f tanswering s y s t e m ox a s an interniwdiete e~p r~s s i u n i n '61:mchin~ t x a n s l , s t i u r t H~wu\!r, V a r i o u s syntactic p a t t e r n s i n J a p a n e s e w i t h appropriate s e w n t l c and c o n t e v t u a l c h e c k l n g f u n c t l o h s , PLATON is p r e s e n t e d i n more d e t a l l I n a n o t h e r papec b y Nagao and T s u j i l ( 1 9 7 6 ) .2 ) I n t h e dictionary a r e s The f a l l a w i n g k i s t shaws t h e relations u s e d i n t h e network:( 1 Deep C u e R e l a t i o n s . ACT', OBJ, P U C E , TIHE --A deep case r e l a t i o n connects an S-node with i t s argument C-node, S e l a a n t~c and C o n t e x t u a l f u n c t i o n s a r e propmmned i n L I S P 'Fathart ka r famili&r a pie* In order to identify a pesaon i n d i c a t e d by thc word, mi h r v t EO haw whose father he i e . In the chemical field we r a n a u i l y find auch nome ( e . 8. , 'weight' , ' t e m p e r a t u r e ' , ' color' , qnnd 'mess ' ) *E"hrdss &re e r J l c d A t t r i b u t e Wslane. Their m a n i n g e are deacrilaed i n a d i f f e r e n t way from that of ordinary n o w e , The description (t4F N-A) shows the noun belongs t o t h e group 0 f a t t r l b u t owterJl~aI ( ( SP> (AmR(STATE) (MASS) (COLOR) ( % W E ) ---------1 ) l i q u i d ( ( SP m a t e r i a l ) (ATTR (STATE *LIQVID) ( W E NIL ) ) )( WRISA ( (WP W-A) (A-ST W L W kWSS LENGTH A 1 s LZB (SP ZQKUSEZ RY8 b > )a t t r i b u t e q u a n t i t y aft Attribute noun8 a m further ~Z a s e i f i e d i n t a twa groups, quantitative and q u a l i t a t i v e A g u c * a i t a t i~g a t t r i b u t e noun cannot b e a case element a f a verb wtaech regwfres g u n t i t n t i v e nouns. Tna v e r b s FUERU [increase) and MERU (deciease) are such &xamples gf verbs. Liquid' is another r e l a t i o n a l noun. The J a p a n e s e word whirh csrreapoads t o ' l i q u l a is EKITAI. While 'liquid' in English can be e l t h e r a n o w or an adjective, ERITWI i n J a p a n e s e is c a~r g o r r z e d s y n t a~t~c a l J y as a n o m . But @emantically EKLTAF has two d i f f e~e n t meanings, one c o r r e s p o n d i n g -1 3 -( I R O ( (NF M-A) (A-ST G O L~~R ) (SP ZOKUSEL SHITSU 1))In our system t h e meaning QE a verb is d~~c s i b e d by settrirzg up k;emsal r e l a t i o n a l slots whLch will be fillad i n by nouns, JR this sen&@ the w a l i n g of e wcrb is n o t confined t o i t s e l f , b u t t w r e l r t g d to mma.We describa rhase r e l a t i o n s by u~i n g t h e cone caneept i n t r~d u r e d by CCF ( ( ACT MINGEN ) ( OB J KOTAI) ( IN EKITqI) ) human b e i n g s o l i d l i q u i d ( ( ACT NINGM ) ( OB J KOTAI) f l VST) ) human being s o l i d ( ( ACT SAN ) ODB JKINZOKU ) )acid metal -WA - The volume sf t h @ g a s i n c r w s e s -- EQUATIONCb) I O U - -MA - K I I R O I s u l f u r -- -( S U B~: , --- vv i t y , (a) IQiRE-GA M I Z U -0 - NESSLhRU h e -f S 1 M ) water -( O B J ) - h e a tHe heats t h e w a t e r . ( 1 H Z U -GA asra-pasarI _ _ - (b) TNSAN -GA A E N -0 -- T O U S U y d r o c h l o r i c a c i d -(ACT) -z i n c -(OBJ) --me1 t Hydrochloric a c i d pelts zi nc .-SUIJOUKI -MI - U Q V(a) -- -BE - - MTZU -0WSSmW.water -(P8.l) h e a t (Solpeonr) heats water J Ja gas burner. ; FACT is u s e d t o &dieere s e n t e n t i a l cotnplrnuenti~@ra. ARKWRU-MPt,t=NO(a> KOBEe -0 SMSTS~YOwOZUW-NO -- ------eNlsurrv -ro =. . . a, * et., i t -f B B J ) -tYOKO -NI - BE -b QKL'. a l c o h o l lamp s i d r -- -( P U C E ) -- beaker -\ O B J )Put ( h e n one knowa a certain event has occurred, he can a n t i c i p a t e s u c c e~s i v e events that w i l l occur and what changee t h e o b j e c t s p a r t i c i p a t i n g i n t h e evant will undergo. Thie kind of expectation plays an i m p o r t a n t role i n underetanding eentences. Various k i n d s of a s s o c i a t i o n s c l u c t e r c o n c e p t u a l l y (a) ( ADD case a-set-of-(A V)-pairs ) (b) ( DELETE case a -s e t -o f -a t t r i b u t e s ) ( c ) ( CREATE lexical-nam-of-an-objec t a-se t-of- A t y p i c a l zxample u s l n g a CON e x p r e s s i o n i s shown I n F i g u r e 2 6 t --arsund(A V j -p a i r s )r A 'j CI 4 p a d k t D 3 U 4J (Ti QJ u Qi s 0 C -4 'J; ri k 3 w 4 3 fV2 G1 C 4-4 - ul 11 0 4 3 [I) IU k CU s C, -a c a h Z 0 U w 0In t h i s expression one can see the verb TOKASU has two differetlt meanings.One cottesponds t o ' m e l t ' , and the a t h e r t o 'dissolve i n ' . When we a n a l y z e the aentense,MU -6 TQKASU .c~g p e r - We have i d e n t i f i e d s l x t e e n s e m a n t i c a l l y a c c e p t a b l e NOUN NO N O W combinations. These a r e shown i n Table 3. MhU -(F K I F A I h t e s t t u h c i n l i q u i d t ?The noun-nuun r t~m b i n a t i w , t e s t tube-SL7 in' expresses t h e 'place1 in the t e s t t u b e , ***So t h e program w i l l go back t o step (11) . T A f SWT ox\*gen (and) hvdrngcn v 0 k t~~ ( 1 S A N S 0 -WO SHXTSLRJQU -TQ S U S O -80 TAXSEKI -7'0).e4uvgenmass ( a n d ) omgen ' L olume ( 2 ) S M S Q -NO SHLTS'LrRIOU -TQ TAISEKX -(2'0)rsmgen mass ( a n d ) vnlume with Naun-h. I f found, l e t i t b e Noun-2, and go t o s t e p 3 .S t 9 1. The phrase between t h e p o s t p o s i t i o n and Noun-: are analyzed ---1 noun p h r a~e a n a l y s i s . This is now t h e second of t h e two pasallel phrases u n d e r csnshdesa t r o n ,Step -4. The phrase betore the p o s t p o s i t i o n is a n a l y z e d by t h e normal n s m phrase a n a l y e i s -i+6 -o n l y p a r t i a l knowledge about t h e context, and therefore, h i s knowledge i s not coqplete. However, he can u n d e r s t a n d the meanings of s e n t e n c e s before he reads through t h e e n t i r e s e t . This means t h a t one i s c o n t e n t w i t h (1) C o n t e x t i.; e n t e r e d a n t o the ~n t e r m d i a t e term m e m~f y ,incomplete( 2 ) Two k;lnds sf i n t e r m e d i a t e term memry a r e p r e p a r e d . (1) I n Japanese a t h e m word 1s o f t e n o m i Changes o f W S R e p i n n~n p of t h e a n n l~$ i s of S1: ((Nk IT3 Eu' 2 N1))End o f the a 1 1~1 v s I~ 51Br@nninq n f t h e ,~n a l~s i t . To solve t h i s problem we s e t up a t r a p p i n g l i s t TL, The bas i r o r g a n i z a t i o n of TL is show i n Figure 4 . 6 , h t r a p p i n g e l e m n t 'is a t r i p l e t ( t ) They check whether a n e w a t hand Bsn salve the1 problems in TL, Ciii) If it cannot do so, the system adds 1 t o N, t h e f icst element of tbe trapping element. Wen M exceeds f i v e , the t r a p p i n g element is deleted frorp TL. That i s , i t i s decided t h a t t h e problem corresponding t o the t r a p p i n g element can not be solved a t e l l . Before the deletion of a trapping elcmpenr its third element, the function F2, is eveluated. Thus far F2 has cnlJ b~a n used t o provide default values t o sllowu some intet pscrsrlon 1 1s p~n d i n p By using Pbp idea o f TL, we can t;eparate *ridas checking wchuuisari fro& the w i n program. They can be invoked crutanwtitaliu whcn o noun &ppatn i n n senten&@, The idea of TZ. msedles thee af 8. Z'harniak;'~: 'dsw4rtt (J197211,When his rryateffi sncowntars a c o c t o i n word, far cxerirlc, ' p i~ bank', i t c u o e t a~ e demon which t r i e s t o catch from the succeeding sentences any word (e.g., money) related to the key word. Ke fear that unnecessary knowledge w i l t clog the system with e 'combinatariel explosion' resulting from the p r o l i t e ration of demons, Ous trapping elemat i s g u t i n TL only temporarily In t h i e exawple there ate three nouns following the a r t i c l e which can be wcfifled by i t e y n t a c t i e a l l y , We must decide the preferable modlficat i @ pattern by using c~n t e x t u a l informention, In the analysis af a noun phrase, we scan the words one-by-one from left t o r i g h t , When we catch t h e article ' KDNLa, we put it fn the tewarasy stack. The wosd will then be checked t o aee *ether it can modify a noun in the fallowing noun phsasa.Vhea.1 we scan r h~ n o w SH31aNUN ( t s e t tube) in Fbgure 4,7, we check whet'taer the object indicated by it was elready mntioned in the preceding sentences. (2) % U S 0 -a U U ,11-KONO TAfSEKZ -0 volusle (053.51There i r s oxygen.Zn thfs c a e KONO alone desigaates the entity noun SAHSO wh:ich appears in (2) Tha Elrs t usage of KONO has thc following three veoir'ties.(i) SMSO -GA ARU. KOHO W S O -0 ---- There i a D V~P~.The ~xygstl ----The now m d i f i a d by the article is the same noun which appears i n the preceding sentence. The gas mixture ---The a r t i c l e modified a nodnalized form of the f i r s t sentence. The f i r s t sentence l g s t a n t i a t e s the case frame of the verb 'mx' FJe ewduate the N W S description of the case frame and obtain a new lnierenced object 'mixture', whose elem~ents are the oxTgen and the hydrogen. The noun XONWUKITBI modified by the a r t i c l e i s a lower concept noun oi the infesenced n o w (mixture) i n ENS.According to these three v a r i e t i e s , we provide the following three check sautines. The order of checking in shown i n (check 1) Ia there in the l i~t t h e earn noun cre t h e noun modified by KONO.. BEWFZCEHT ---1, and t h e c a s e frames of IRERU ( p u t :in) have t h e cqse PLAw-, W e can p r e d i c t t h a t t h e pronoun KORE ( t h l s , it) f i l l s the PLACE case i n t h e sentence. The semantic d e s c r i p t i o n s a y s t h a t a l o v e r concept naun of c o n t a i n e r ' o r 'Ilquid' i s p r e f e r a b l e as t h e PLACE case of the verb IRERU (put i n ) . The o b j e c t ' w a t e r ' , which i s a lower concept. man of ' l i q ui d v ,is found in NS, and is d e t e m u n e d t o b e the o b j e c t designated bt the pronounWe have some other pronouns and articles in Japanese which are analyzed in ehe same way. We provide d i f f e r e n t LISP f u n c t i o n s f o r different pronouns and put them in the d i c t i o n a r y d e f i n i t i o n 8 of these words.T, Wfn~grad treated the s a w problem in hie excellent system SkSRDLU (1971; 1972) . lbwever, the world which his eys tern can deal w i t h is very h j d t e d . Sn order t o construct a system which can treat a wider range aE ecntcncee, the gystem should be equipped with the schema representing the re$~ei-srar%h$ps between event@ and abject (an event m y iwly tl.u$ occuttrarentllc of new objects et changee in the propertie$ of sbjects). In real. world eeatencee, there e x i s t s m s e complex g h e n o e n a about anagheric expressions and a d~~i o n s of w~r d~ than those treated i n SHRDLU. We do n o t claim t h a t aut eyetern c m t r e a t suck .c&pl.ex pnensmna, but we hope that our system can The s e n t e n c e i s analyzed by the fo1law;ing steps. noun TAISEKI: 1 s a Irstver concept noun of ' a t t r i b u t e ' , which stjtisflcftzs t h e semantlc c o n d i t i o n for the case element. So thls s e n t e n c e is a n a l y z e d :in a straightfoswasd manner, However, because t h e noun TAISEKI 1s an attribute noun, we m u s t f i n d t h e corresponding entity noun. Tkqt I S , we must i d e n t i f j the o b j e c t whose volume is being referred to. As we c a n n o t f i n d such an o b j e c t i n t h e p r e c e d i n g sentences, we set up a p e n d~n g problem I n TL, I3.1 checking t h e i n f l e c t i o n of t h e verb WENUSURU (change) and n o t i n g t h a t it: is i w w d i d t ely followed by a noun, it i s r e c u g n i z e d t h a t t h e s e n t e n c e is an embedded s e n t e n c e modifying khe f o l l o w i n g noun TOKI ( t i m e , when), Fu' e t h e nbes the fdaab s t e p in the a n a l y s i s of a noun phrase, w e check whether there remain relational nauos which have no d e f i n i t e meaglng. If found, we search through NS for words whrch are suitable t o f i l l i n t h e s l o t s of the nome, The ecarchihg process 1s che same as f o r omitted words In siaple seatencea, Sometiaes t h e omitted words e x i s t i n succesding s e n t e n c e s , s o we cczn set up a problem in TL, if w e cannot f i n d an appropriate word in t h e gracedhng aentencea
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 588 | 0 | null | null | null | null | null | null | null | null |
4bc560aa37f27351c1cc82a6dcda610967ba5230 | 60330510 | null | {PLATON}--A New Programming Language for Natural Language Analysis | PLATON (Programming LAnguage for Tree OperatioN) facilities of pattern matching and flexible backtracking, language is developed t~ simplify writing analysis programs The pattern matching process has the facility t o extract subinput sentence and invoke semantic and contextual checking fo?. actions between syntactic and other components are easily obt processing r e s u l t s i n a failure, a message which expresses t : failure will be sent up. The control w i l l be modified accoru enables us t o write f a i r l y complicated non-deterministic progr manner. An example of structural analysis using PLATON is a l s -,le 1. The 1 language Erom thc Intc --. rf Ise of . Th i :- ;in:; e d " I I n t r o d u c t i o n In t h i s paper we describe a new programming language which is designed to facilitate the writing of natural language grammars. A s i m p l e structural analysis program using this language is given as an example. There are two key issues in analyzing natural language by computer: 1) how to represent knowledge (semantics, pragmatics) and t h e state (context) of the world, and 2) how to advance the programming technology appropriate for syntacticsemantic, syntactic-contextual interface. The point in designing a programming language i s to make this kind of programming less painful. | {
"name": [
"Nagao, Makoto and",
"Tsujii, Jun-Ichi"
],
"affiliation": [
null,
null
]
} | null | null | null | 1976-02-01 | 0 | 1 | null | null | of s e m a n t i c and p r a g m a t i c c o n d i t i o n s on the b r a n c h e s between s t a t e s . By d o i n g this, c l o s e i n t e r a c t i o n s between the s y n t a c t i c and other components can b e e a s i l y accomplished.However ATN h a s the f o l l o w i n g s h o r t c o m i n g s , e s p e c i a l l y when we a p p l y i t to the p a r s i n g of Japanese s e n t e n c e s :1. It scans words one-by-one from the leftmost end of an input s e n t e n c e , checks t h e a p p l i c a b i l i t y of a r u l e , and makes t h e t r a n s i t i o n from one s t a t e t o a n o t h e r . This method may b e w e l l s u i t e d for E n g l i s h s e n t e n c e s ,but b e c a u s e t h e o r d e r of words and p h r a s e s i n Japanese s e n t e n c e s is r e l a t i v e l y free, i t is preferable t o check t h e a p p l i c a b i l i t y o f a r u l e by a flexible p a t t e r n -m a t c h i n g method. I n a d d i t i o n , w i t h o u t a p a t t e rn-matching mechanism, a s i n g l e rewriting r u l e of an o r d i n a r y grammar i s often to b e expressed by several rules belonging to d i f f e r e n t s t a t e s i n Woods ATN parser.2. An ATN model e s s e n t i a l l y performs a k i n d of top-down analysls of s e n t e n c e s . T h e r e f o r e recovery f ronl f a i l u r e s in p r e d i c t f . o n is most d1.f f i c u l t . | null | null | The ATN model has the following additional merits : .x p r e s s i o n e q u i v a l e n t t o t r a n s f o r m a t i o n a l grammars 2.I t mai11tai.n~ much of the r e n d a b i l i t y of context-free grammars.3 . R u l e s of n grammar can b e changed easily, so we can improve them through a t r i a l -a n d -e r r o r process whi1.e w r i t i n g t h e grammar.A grammar, whether g e n e r a t i v e o r a n a l y t i c a l . i s r e p r e s e n t e d as a (2) a c t i o n s which must be executed, i f the r u l e is a p p l i c a b l e(3) a s t r u c t u r a l p a t t e r n i n t o which the i n p u t s t r u c t u r e s h o u is found, the i n p u t s t r u c t u r e is transformed i n t o a n o t h e r s t r u c t u r e s p e c i f i e d by t h e r u l e and the control makes the s t a t e t r a n s i t i o n . , when i t i s a p p l i e d . > i s ( ( A ( R1 (# C D E B ) ) ( R2 ( G ( R3 H ) ) ) ) F G ) .I f the r u l e is a POP-type one, then this s t r u c t u r e w i l l be returned t o t h e h i g h e r level processing. If i t is NEXTor NEXTB-type, then the c o n t r o l w i l l The v a r i a b l e :I1 is bound to tius i n t e g e r . This i n t e g e r i s added t o t h e sum of the i n t e g e r s , :N, i f the t o t a l does n o t exceed t e n (SUMUP -I-, con). | Main paper:
it p r o v i d e s power of e:
x p r e s s i o n e q u i v a l e n t t o t r a n s f o r m a t i o n a l grammars 2.I t mai11tai.n~ much of the r e n d a b i l i t y of context-free grammars.3 . R u l e s of n grammar can b e changed easily, so we can improve them through a t r i a l -a n d -e r r o r process whi1.e w r i t i n g t h e grammar.
i t i s p o s s i b l e t o impose various types:
of s e m a n t i c and p r a g m a t i c c o n d i t i o n s on the b r a n c h e s between s t a t e s . By d o i n g this, c l o s e i n t e r a c t i o n s between the s y n t a c t i c and other components can b e e a s i l y accomplished.However ATN h a s the f o l l o w i n g s h o r t c o m i n g s , e s p e c i a l l y when we a p p l y i t to the p a r s i n g of Japanese s e n t e n c e s :1. It scans words one-by-one from the leftmost end of an input s e n t e n c e , checks t h e a p p l i c a b i l i t y of a r u l e , and makes t h e t r a n s i t i o n from one s t a t e t o a n o t h e r . This method may b e w e l l s u i t e d for E n g l i s h s e n t e n c e s ,but b e c a u s e t h e o r d e r of words and p h r a s e s i n Japanese s e n t e n c e s is r e l a t i v e l y free, i t is preferable t o check t h e a p p l i c a b i l i t y o f a r u l e by a flexible p a t t e r n -m a t c h i n g method. I n a d d i t i o n , w i t h o u t a p a t t e rn-matching mechanism, a s i n g l e rewriting r u l e of an o r d i n a r y grammar i s often to b e expressed by several rules belonging to d i f f e r e n t s t a t e s i n Woods ATN parser.2. An ATN model e s s e n t i a l l y performs a k i n d of top-down analysls of s e n t e n c e s . T h e r e f o r e recovery f ronl f a i l u r e s in p r e d i c t f . o n is most d1.f f i c u l t .
basic o p e r a t i o n s of platon:
A grammar, whether g e n e r a t i v e o r a n a l y t i c a l . i s r e p r e s e n t e d as a (2) a c t i o n s which must be executed, i f the r u l e is a p p l i c a b l e(3) a s t r u c t u r a l p a t t e r n i n t o which the i n p u t s t r u c t u r e s h o u is found, the i n p u t s t r u c t u r e is transformed i n t o a n o t h e r s t r u c t u r e s p e c i f i e d by t h e r u l e and the control makes the s t a t e t r a n s i t i o n . , when i t i s a p p l i e d . > i s ( ( A ( R1 (# C D E B ) ) ( R2 ( G ( R3 H ) ) ) ) F G ) .I f the r u l e is a POP-type one, then this s t r u c t u r e w i l l be returned t o t h e h i g h e r level processing. If i t is NEXTor NEXTB-type, then the c o n t r o l w i l l The v a r i a b l e :I1 is bound to tius i n t e g e r . This i n t e g e r i s added t o t h e sum of the i n t e g e r s , :N, i f the t o t a l does n o t exceed t e n (SUMUP -I-, con).
thc augmcn t c d t r i 1 1 1~ i l i o n ilc twork (a1l'n) proposccl by w woods (19 70) from our p o i n t of view gives an e s p e c i a l l y good framework for natural xa~~guage a n a l y s i s systems. one of-t h e most a t t r a c t i v e features is t h e clear dtscrimination between grammatical rules and t h e control mechanism. t h i s enables us t o develop t h e model by adding various f a c i l i t i e s t o its control mechanism.:
The ATN model has the following additional merits : .
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 588 | 0.001701 | null | null | null | null | null | null | null | null |
6fbf780d194c39cf715487411dd3d6db223e6a0f | 219303792 | null | Artificial Intelligence and Language Processing: A Directory of Research Personnel | In t h e s u m m c r o f 1 9 7 4 ~ S P ~ war r r k a d b? h *tianal I n g t i t v t a a t Elucatlon t o camp118 4 a i r P C t o r y o f p e r r a n 8 e n g a q r d in rc8rrrch on a t t i b i ~i 1 1 i n t @ l l i g ~n c e ~ Who r intltrsfited ~p a c i f i c r l l y in lrngurg8 PraOesrrr, ~h i r t a s k was on# o f s e v e r a l u n l a r t a k @ n t o h e l p WSlg i t r r l u s t r t h e p o t c l n t i r l canttibution b f such r c t l v i t i ~r t o t h e r r p l a n r t i ~n o f t h e p r o c r r o r ~ Ldv6lv.d in Canprch4ndlnq 8pQlUln and p r i n t a d nmssawr. The r e q u i r e m e n t ran i p d c i f i e d a# follew#t A directory w h i c h i d e n t i f i e s prrsoflr r h o a r e r ~g u l ~r l y c ~n t r j b u t l n g t@ t h 4 literlturet ARTITLCXAL IWTCLLIGEYCC AND LANGUAGE PROCESSING A Directory o f Research Perronnel SUMMARY OF XNTBRPSTS 8Patrm B u l l d i n g r e m . * . m o m * e a m m m . e l a r . ! 6 3 Q u a ~t l ~n A n f i w Q r l ~q o . . . m . ~m o , m m , , e ~e ~ 91 S p e d c h Und4rrtanbin9 . w m m . o m , m m . . , . . . 36 Compt*hen8lon ~r ~m ~~~~m m ~ 8 1 ffiatructton , , l . m l . ~m l . m . . . . m m . . . ~w ~m 2 2 w o r l d Nodallng m . . . e . * . . m r . * m , m a , m . e o 7Q B85lrf N 0 d a l l n ~ e m ~r m r . m w ~r ~w o . ~m ~~~. | {
"name": [
"Walker, Donald E."
],
"affiliation": [
null
]
} | null | null | null | 1976-02-01 | 0 | 1 | null | null | null | null | Parkt California 9 4 9 2 5 . 1Dr. Naglb A , Badre I B M Watlon Research Cerkter P,QI Box 2 1 8 Yotktown HeiQhtl, New York 1 0 5 9 0 a 14-chrnnrl vocqder; t h e S t r i n g Oontrins many @ r r 6 r a 0 ] This e r r a r t u l StrLng is p a r s e d w i t h r grn8ralited t y p o 0 grammar, A Directory o f Pe8eWch P 8 t r o n n e l w i n g the pr4grrmr&np languqge PROLOG, T h w r r r r t w o r r l 0 t r u l e r ! p h a n o l a g i c~l r u l e s t h r t taka i n t o r c e a u a t t h e e r r o r s o t S@qm*ntrtlon r n d rlcoQnition o f P h o n a m e l , and rYntacCltal r u l b r for p a r r i n g t h e $ @ n t a n e @ , T h a n e t v b p l t t l a r e a r n b 8~d e d i n one parser u r i n q r top-down r t r a t a g y w i t h brekttrcking, T h e o u t p u t o f t h p a t m e r 1 ; I d q a p rtructure, A p a r t o f t h e p r e g t a l a t r i m Them8 include leXic81 rnemor~, i e a a n t i c p r i r n i t l v c~t prrgmrtlc8, I t r u c t u r e r f a r m a i n t a i n i n g r a m a n t i c knawladOer a n d ~e m r n t l c tales, IHDA Directory o f Pal&arch PrrronnelMajor R e n t a r c h Interertsr a u t~m a t l c thearcrm*~ P r~v i n g a n d induction8@ntOnc+, C u r r e n t l y , I am p u r a u l n g r r l r g r c h i n r e m a n t i c s o f L a n g u a g e i n t h e domain o f t h e p r r 5 0 n r l assistant p r o j e c t , w i t h t h e p l a n o f o r g a n l t i n g Dr. H a r r y C , B u n tArtificial Iht&lll~@nca G r o u p p h i l i p s RCIerrch LIboratory W B 3 E i n d h o v b n , T h e Netherlands 1.l a t u r a l language quest$onaans*ering. system optr(t1onc. Dr. E u~a n e C h a t n i a k I s t J t u r o p e r q l i S t u d i semantic4 e Cognltivi i 7 Rue Candoll* 1 2 4 0 Geneva, Switzerland A D l r @ c t~r y a t Rererrch Parsannrl Mr. Horace E n a r C o m p U t r r S c j , @ n c 8 D * p a r t m a n t S t a n f o r d U n i~r r s~i t Y Stanford, California 9 4 3 0 5L a r g e dictionary c o n l t r u c t i a n 3 , Th@$rurur construction aclectmd w i t h v i e w tO f n8W k n a w l 4 d g e . Department ad Computar Selenct Carnrgli-Nollon U n l vA D i r e c t o r y of R e S t U C h P e r a o n n r l Towards C o n p u t e r L~x i c o m e t r y ; Dr, Neil M e Coldman USC Infor-matian GciencrrMarina d e l , Ray, California 9 0 2 9 1 1 Knowledge t e p r e 6 t n t a t i~n and organization a . o p r e r a n t r t i a n c l o s e t o t h a t r e q u i r e d b y the n r n a n t c o m b i n e d r y @ t~m a compiler d r l v a n b r h a~i e t a l l i m u l l t i o n l 8 n g U a g~f a u t~r n a t f c g a n~t r t f v e s8nrntic g r r m m r t i c r 1 ,Ann h r b a r , M i c h i g a n 1 8 1 0 1 1 , @d, A fuzzy p r r r 8 r intrrprrts the Z t u d a n t r r q u a r t s a n d b u i l d 8 r L18P f u n c t i o n w h i c h O P @ r 4 t 8 S on a semantic network t o d o t a r m i n e i t s r 8 r p o n r e . A sacand e f f o r t h a 8 f o c u r t d o n t h e s t u d y o f b g e n a r r i m o d a l t The N e t h e r l a n d s w h a t llnpuirtic teatores a r e a 8 S o c i a t t d w i t h C o n t r X t stenotypy t r a n r c r l~t i o n~ a n d n a t u r a l l a n g u a g e translation. To s~c c a 8 s f u l l y U n d e r a t a n d h u m r n l a n g u a g e b e h a v i o r t and t o e m b o d y t h i s u n d e x 8 t a n a l n g in u a r f u l c o m p u t a t i o n a l f l b d e l s , i t w i l l b e r o o t e d in coherent d o t c r i p t i o n S o f t h e p r & g m a t l C a n u n d e r s t a n d i n g , lrnquage, f h e main concern is t o axarnina w h y v a r i o u s construct8 a v a i l a b l e in l a n g u a g e a r e used -- A Directory o f Rerearch P Q r r o n n 8 l p r o b l e m o f d e v e l o p i n g a n " d a p t i v 8 ' n a t u r a l l a n g u a g e r y r t e m w h i c h , i f c o n f r o n t e d with a S 8 n t~n c e c o n t a i n i n g a w o r d o r c x p r e r r S o n it d i d not u n d e r s t a n d , c o u l d a s k I n t e l l i g r n t l Y tor c l a r i f S c a t S o n , 2 , HIT A 1 L a b o r a t o r y Dr, K r n n t t h I . w h i c h a c c e p t s string, t r e e s and lists a n d t r a n t f o r m s t h~m In a r b i t r a r y way#, h a v e been d c v t l a p e d , P L A T O N 1 s b a s s d on t h e auqmented t c r n s l t i g n network m b d r l ~f W e Wood), and h a s uariouw a d d i t i o n a l caprbilitler o t r e c e n t A I l a n g u a g e s , t h a t i s r p a t t e r n -m a t c h i n g a n d f l e x i b l e back-tracking mechanisms, B y u s i n g t h e w , we censtructed a p a r s i n t r i e v a l S y S t @ m s w h i a h ern c a r r y o u t i n t t l l l p e n t ~0 n V e r S a t i o n s w i t h p r o p 1 8 t n r a u g h n a t u r a l e c t i o n of t h o a b o v e , w h e r e p r o d~c t l o n s y s t e m s a r e U n d e t r t a n d i n q 1 n a p t c i , L l~ t h e M~V P~~O r e~l r e d t o i d e n t i f y ~~t h o a r w i t h e x t e r n a l situatian8, CMSfUJN) 4 , Protocol a n a l y a i r o f v e r b a l r e p o r t # r a r p c c l r 1 X y t h e r u t o m a t i z a t~o n of Such analysis, ( P A S ? )1, ThoThe modal h a $ been s i m u l a t e d on a l a r g e d i g i t a l C Q W P U~~~ a n d lncludea working i n r a r p t e t e r rnd r u q n e n t e d -t r a n s i t i o n n e t k o r k P a r g e t , Phone --512~47l-l52dt 426-2800, e x t , 4 9 2 1 4 4 4 -0 1 W e now have a n o r k i n g @ystarv which c a n h a S u i t a b l e f o r m a n w m a c h i n e c a n r n u n i c a t i @ n in n a t u r a l langur0c. R e p t c a c n t a t l o n a n d utilization o f complar knowl@dQs d t r u~t u ~8 r f~r n a n e r o f a l l b u t t h e ~o s t l e r r n a d (habitual1 o f rctionr, I am j n v t t t i g a t i n g rnalog!erd rearonlnp in t h i n . ?ma.A Directory of Research P@rronn@l W a f , S t u a r t C, S h a p l r o C o m p u t e r S e i~n c e P@pilr'fm(rnf Indiana U n t v a r s l t~ 1 0 1 L i n d l r y Hall B l~o m i n g t a n , Indiana 4 1 4 0 1 h t a r a c t i v e s y s t e m s b a s e d on semantic ( c o n c e p t u a l ) networks, S f i 3 l n t f C 8 and 8 e h r n k 0 s inference molseulen a 8 a p e c i a 1 c a s e s , For the p a s t s a v , T h e o r i~r about n a t u r a l Language undttrtandinQ, a n d s y s t e m s e m b o d y i n g 6uCh t h e o r i e s , 3 , Theor!.*$ ~b n c r r n l n g k n o w l e d g e r e p r e n t t n t r t i o n r 4 n d methods f o r P u t t i n g t h e $ # t o c g f e c t i v l U S e ., Modal logics and I l n g U i l t l c C~ computer i n v d r t l g r t i o n r of k o n t @ w u c '~ t r e a t m e n t o f a t r r g m c n t of L n g l i r h , R a n u l t n i n c l u d e g e n e r a t i o n a n d p a r b i n g r o u t i n r r t ~t U d i 8 # o f t h e t r r n s l r t i~n t o m o d a l l o g i c a r e In p r o g r e s s ,4 ,S p e e c h u n d e r l t a n d i n g r y r t e m s t m y a t e r r r s p e c t~ o f t h e nodelt how t h a v a r i o u s s o u r c a r o f k n a W l 8 d g 4 intaract,t~I c 8 1 Q C C O~P O S~Z~O~ and d i l o~u r a t r n r l y~i s~c , m l r n l n g s of prlncival t e t m i n o l o q y o f s p a c e -t i m e .8 3 4 5 V a n t u r r 'BOulrVardTatzrnr, C a l i f o r n i a 9 1 3 5 64 5 T e c h n o l o g y Squar8, R m 8 2 4 Cambridge, M a s s a c h u s r t t s O2i39. t h e d c v c l o p m e n t o f c o m P u t a t i~n a 1 a l q o r l t h m s f o r D a r l i n g ~t n t t n c t~ w i t h r e s p e c t t o such ~r a m m r r~~ and3 , s x p e r i r r e n t r l application# involving t h e ernWeyrnent o f t r a n~f o r m a t l o n a l l y -d e f f n e d , machine-understandable sobsots o f n a t u r a l English f o r Interaction w i t h computers, t.9.r in q u~s t l o n~a n r w e r f n g en f o r m a t t e d d a t a b a s t s ,, C o m~u t a t i o n a l and n a t h l m r t~c r l t h e o r i t r o f n a t u r a l language s y n t a x and semrntic8, 3 , S~% t @ m~w b u l l d i n g a i d 8 f a r L i n p U l r t S , The f i r s t o f t h e s e is aimed a t t h~ c a n~t r u c t i o n o f s Y n t a e t l c a n d s e m a n t i c 8 y o t e m s r us;nQ d t y p e o f g e n e r a t i v e s m I n t i c~ model, A t t h e time o f t h i s writing, r s y d t c m i s running in SITBOL on t h e PDPI10, but n o | null | Main paper:
,:
L a r g e dictionary c o n l t r u c t i a n 3 , Th@$rurur construction aclectmd w i t h v i e w tO f n8W k n a w l 4 d g e . Department ad Computar Selenct Carnrgli-Nollon U n l vA D i r e c t o r y of R e S t U C h P e r a o n n r l Towards C o n p u t e r L~x i c o m e t r y ; Dr, Neil M e Coldman USC Infor-matian GciencrrMarina d e l , Ray, California 9 0 2 9 1 1 Knowledge t e p r e 6 t n t a t i~n and organization a . o p r e r a n t r t i a n c l o s e t o t h a t r e q u i r e d b y the n r n a n t c o m b i n e d r y @ t~m a compiler d r l v a n b r h a~i e t a l l i m u l l t i o n l 8 n g U a g~f a u t~r n a t f c g a n~t r t f v e s8nrntic g r r m m r t i c r 1 ,Ann h r b a r , M i c h i g a n 1 8 1 0 1 1 , @d, A fuzzy p r r r 8 r intrrprrts the Z t u d a n t r r q u a r t s a n d b u i l d 8 r L18P f u n c t i o n w h i c h O P @ r 4 t 8 S on a semantic network t o d o t a r m i n e i t s r 8 r p o n r e . A sacand e f f o r t h a 8 f o c u r t d o n t h e s t u d y o f b g e n a r r i m o d a l t The N e t h e r l a n d s w h a t llnpuirtic teatores a r e a 8 S o c i a t t d w i t h C o n t r X t stenotypy t r a n r c r l~t i o n~ a n d n a t u r a l l a n g u a g e translation. To s~c c a 8 s f u l l y U n d e r a t a n d h u m r n l a n g u a g e b e h a v i o r t and t o e m b o d y t h i s u n d e x 8 t a n a l n g in u a r f u l c o m p u t a t i o n a l f l b d e l s , i t w i l l b e r o o t e d in coherent d o t c r i p t i o n S o f t h e p r & g m a t l C a n u n d e r s t a n d i n g , lrnquage, f h e main concern is t o axarnina w h y v a r i o u s construct8 a v a i l a b l e in l a n g u a g e a r e used -- A Directory o f Rerearch P Q r r o n n 8 l p r o b l e m o f d e v e l o p i n g a n " d a p t i v 8 ' n a t u r a l l a n g u a g e r y r t e m w h i c h , i f c o n f r o n t e d with a S 8 n t~n c e c o n t a i n i n g a w o r d o r c x p r e r r S o n it d i d not u n d e r s t a n d , c o u l d a s k I n t e l l i g r n t l Y tor c l a r i f S c a t S o n , 2 , HIT A 1 L a b o r a t o r y Dr, K r n n t t h I . w h i c h a c c e p t s string, t r e e s and lists a n d t r a n t f o r m s t h~m In a r b i t r a r y way#, h a v e been d c v t l a p e d , P L A T O N 1 s b a s s d on t h e auqmented t c r n s l t i g n network m b d r l ~f W e Wood), and h a s uariouw a d d i t i o n a l caprbilitler o t r e c e n t A I l a n g u a g e s , t h a t i s r p a t t e r n -m a t c h i n g a n d f l e x i b l e back-tracking mechanisms, B y u s i n g t h e w , we censtructed a p a r s i n t r i e v a l S y S t @ m s w h i a h ern c a r r y o u t i n t t l l l p e n t ~0 n V e r S a t i o n s w i t h p r o p 1 8 t n r a u g h n a t u r a l e c t i o n of t h o a b o v e , w h e r e p r o d~c t l o n s y s t e m s a r e U n d e t r t a n d i n q 1 n a p t c i , L l~ t h e M~V P~~O r e~l r e d t o i d e n t i f y ~~t h o a r w i t h e x t e r n a l situatian8, CMSfUJN) 4 , Protocol a n a l y a i r o f v e r b a l r e p o r t # r a r p c c l r 1 X y t h e r u t o m a t i z a t~o n of Such analysis, ( P A S ? )1, ThoThe modal h a $ been s i m u l a t e d on a l a r g e d i g i t a l C Q W P U~~~ a n d lncludea working i n r a r p t e t e r rnd r u q n e n t e d -t r a n s i t i o n n e t k o r k P a r g e t , Phone --512~47l-l52dt 426-2800, e x t , 4 9 2 1 4 4 4 -0 1 W e now have a n o r k i n g @ystarv which c a n h a S u i t a b l e f o r m a n w m a c h i n e c a n r n u n i c a t i @ n in n a t u r a l langur0c. R e p t c a c n t a t l o n a n d utilization o f complar knowl@dQs d t r u~t u ~8 r f~r n a n e r o f a l l b u t t h e ~o s t l e r r n a d (habitual1 o f rctionr, I am j n v t t t i g a t i n g rnalog!erd rearonlnp in t h i n . ?ma.A Directory of Research P@rronn@l W a f , S t u a r t C, S h a p l r o C o m p u t e r S e i~n c e P@pilr'fm(rnf Indiana U n t v a r s l t~ 1 0 1 L i n d l r y Hall B l~o m i n g t a n , Indiana 4 1 4 0 1 h t a r a c t i v e s y s t e m s b a s e d on semantic ( c o n c e p t u a l ) networks, S f i 3 l n t f C 8 and 8 e h r n k 0 s inference molseulen a 8 a p e c i a 1 c a s e s , For the p a s t s a v , T h e o r i~r about n a t u r a l Language undttrtandinQ, a n d s y s t e m s e m b o d y i n g 6uCh t h e o r i e s , 3 , Theor!.*$ ~b n c r r n l n g k n o w l e d g e r e p r e n t t n t r t i o n r 4 n d methods f o r P u t t i n g t h e $ # t o c g f e c t i v l U S e ., Modal logics and I l n g U i l t l c C~ computer i n v d r t l g r t i o n r of k o n t @ w u c '~ t r e a t m e n t o f a t r r g m c n t of L n g l i r h , R a n u l t n i n c l u d e g e n e r a t i o n a n d p a r b i n g r o u t i n r r t ~t U d i 8 # o f t h e t r r n s l r t i~n t o m o d a l l o g i c a r e In p r o g r e s s ,4 ,S p e e c h u n d e r l t a n d i n g r y r t e m s t m y a t e r r r s p e c t~ o f t h e nodelt how t h a v a r i o u s s o u r c a r o f k n a W l 8 d g 4 intaract,t~I c 8 1 Q C C O~P O S~Z~O~ and d i l o~u r a t r n r l y~i s~c , m l r n l n g s of prlncival t e t m i n o l o q y o f s p a c e -t i m e .8 3 4 5 V a n t u r r 'BOulrVardTatzrnr, C a l i f o r n i a 9 1 3 5 64 5 T e c h n o l o g y Squar8, R m 8 2 4 Cambridge, M a s s a c h u s r t t s O2i39. t h e d c v c l o p m e n t o f c o m P u t a t i~n a 1 a l q o r l t h m s f o r D a r l i n g ~t n t t n c t~ w i t h r e s p e c t t o such ~r a m m r r~~ and3 , s x p e r i r r e n t r l application# involving t h e ernWeyrnent o f t r a n~f o r m a t l o n a l l y -d e f f n e d , machine-understandable sobsots o f n a t u r a l English f o r Interaction w i t h computers, t.9.r in q u~s t l o n~a n r w e r f n g en f o r m a t t e d d a t a b a s t s ,, C o m~u t a t i o n a l and n a t h l m r t~c r l t h e o r i t r o f n a t u r a l language s y n t a x and semrntic8, 3 , S~% t @ m~w b u l l d i n g a i d 8 f a r L i n p U l r t S , The f i r s t o f t h e s e is aimed a t t h~ c a n~t r u c t i o n o f s Y n t a e t l c a n d s e m a n t i c 8 y o t e m s r us;nQ d t y p e o f g e n e r a t i v e s m I n t i c~ model, A t t h e time o f t h i s writing, r s y d t c m i s running in SITBOL on t h e PDPI10, but n o
list y o u r moat i m p e r t a n t p~b l i c a t i o n & &nd r r p o r t r in t h e area a f 8rtifieirl 1 n t e l l i~e n c e and b&aguag) processing an t h a back o f t h i s page (or a t t a c h such r listing), [return t o d, e n w & l k s r , stanford r a l e a r c h i n l t i t~t 8 r henlo:
Parkt California 9 4 9 2 5 . 1Dr. Naglb A , Badre I B M Watlon Research Cerkter P,QI Box 2 1 8 Yotktown HeiQhtl, New York 1 0 5 9 0 a 14-chrnnrl vocqder; t h e S t r i n g Oontrins many @ r r 6 r a 0 ] This e r r a r t u l StrLng is p a r s e d w i t h r grn8ralited t y p o 0 grammar, A Directory o f Pe8eWch P 8 t r o n n e l w i n g the pr4grrmr&np languqge PROLOG, T h w r r r r t w o r r l 0 t r u l e r ! p h a n o l a g i c~l r u l e s t h r t taka i n t o r c e a u a t t h e e r r o r s o t S@qm*ntrtlon r n d rlcoQnition o f P h o n a m e l , and rYntacCltal r u l b r for p a r r i n g t h e $ @ n t a n e @ , T h a n e t v b p l t t l a r e a r n b 8~d e d i n one parser u r i n q r top-down r t r a t a g y w i t h brekttrcking, T h e o u t p u t o f t h p a t m e r 1 ; I d q a p rtructure, A p a r t o f t h e p r e g t a l a t r i m Them8 include leXic81 rnemor~, i e a a n t i c p r i r n i t l v c~t prrgmrtlc8, I t r u c t u r e r f a r m a i n t a i n i n g r a m a n t i c knawladOer a n d ~e m r n t l c tales, IHDA Directory o f Pal&arch PrrronnelMajor R e n t a r c h Interertsr a u t~m a t l c thearcrm*~ P r~v i n g a n d induction8@ntOnc+, C u r r e n t l y , I am p u r a u l n g r r l r g r c h i n r e m a n t i c s o f L a n g u a g e i n t h e domain o f t h e p r r 5 0 n r l assistant p r o j e c t , w i t h t h e p l a n o f o r g a n l t i n g Dr. H a r r y C , B u n tArtificial Iht&lll~@nca G r o u p p h i l i p s RCIerrch LIboratory W B 3 E i n d h o v b n , T h e Netherlands 1.l a t u r a l language quest$onaans*ering. system optr(t1onc. Dr. E u~a n e C h a t n i a k I s t J t u r o p e r q l i S t u d i semantic4 e Cognltivi i 7 Rue Candoll* 1 2 4 0 Geneva, Switzerland A D l r @ c t~r y a t Rererrch Parsannrl Mr. Horace E n a r C o m p U t r r S c j , @ n c 8 D * p a r t m a n t S t a n f o r d U n i~r r s~i t Y Stanford, California 9 4 3 0 5
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 588 | 0.001701 | null | null | null | null | null | null | null | null |
6648e0bf1767bec5c4cdc6c2bc9efc901b8a9a69 | 245118146 | null | {C}hinese-{E}nglish machine translation project on linguistic analysis, {U}niversity of {C}alifornia, {B}erkeley | rkclcy) during the period 1967 to 1975. During the cnrly part of the effort, Systcln I \\'as | {
"name": [
"Wang, William S-Y."
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 0 | null | dcvcloped which includes: a) CIIIDIC: A Chinese to English machine dictionary of about 80,000 entrics (GO percent physics, 30 percent Biod emistry, and 10 pcrcent gcnclral), and b) Monolithic grammar of about 4, 000 rules (context-3, phrasc-s t r u c turc rules). In 1973, two factors caused redesign of the approach t~\~n r c I the development of Systcln II. One, the gram tnar had beconle so culxtbersome and ad hoc that its effectiveness as well as fts pebntial for ip~provement were curtailed. Second, the sponsor requested coxlrel;sion of the system from CDC t~~n c h i~~l c s to IBM maaines. In response t o these factors, System I1 is designed along the lllles of f t s t r u c t u r e d program m i n g r t (i. e. , it is built on self-contained ppograrn modules). It is also designed to pe machineindependent, sc that it can be implem ented at different computer installations,Efforts in research and development have been aimed at a n operational system. We have e?r~~erimrnted w iU1 nutllcrous trial sentences as well as several 'liveu texts (from articles of 3, 000 characters in k~i g t h ) a11d have accumulated machine t e s t s of over 560,000 characters. System II i s incomplete, lacking especially the machine-editing of output to conform to those morphological features absent in Clli~lese but required in English.IJang received his Ph.D in Linguistics at the University of IIicIligan in 1960, and was appointed Professor of Linguistics at the Universf ty of C a l i f o r n i a (Berkeley) in 1967. Re is interested in the structure and function of language, including the processcs whereby one l a n g u a g e i s translated into a n o t h e r , Some of h i s work have been on system simulation o f linguistic processes humans do easily, such as speech recognitiorr and machine t e x t analysis. He is the e d i t o r of a b i l i n g u a l journal, Journal of Chinese Linguistics. | null | null | null | null | Main paper:
:
dcvcloped which includes: a) CIIIDIC: A Chinese to English machine dictionary of about 80,000 entrics (GO percent physics, 30 percent Biod emistry, and 10 pcrcent gcnclral), and b) Monolithic grammar of about 4, 000 rules (context-3, phrasc-s t r u c turc rules). In 1973, two factors caused redesign of the approach t~\~n r c I the development of Systcln II. One, the gram tnar had beconle so culxtbersome and ad hoc that its effectiveness as well as fts pebntial for ip~provement were curtailed. Second, the sponsor requested coxlrel;sion of the system from CDC t~~n c h i~~l c s to IBM maaines. In response t o these factors, System I1 is designed along the lllles of f t s t r u c t u r e d program m i n g r t (i. e. , it is built on self-contained ppograrn modules). It is also designed to pe machineindependent, sc that it can be implem ented at different computer installations,Efforts in research and development have been aimed at a n operational system. We have e?r~~erimrnted w iU1 nutllcrous trial sentences as well as several 'liveu texts (from articles of 3, 000 characters in k~i g t h ) a11d have accumulated machine t e s t s of over 560,000 characters. System II i s incomplete, lacking especially the machine-editing of output to conform to those morphological features absent in Clli~lese but required in English.IJang received his Ph.D in Linguistics at the University of IIicIligan in 1960, and was appointed Professor of Linguistics at the Universf ty of C a l i f o r n i a (Berkeley) in 1967. Re is interested in the structure and function of language, including the processcs whereby one l a n g u a g e i s translated into a n o t h e r , Some of h i s work have been on system simulation o f linguistic processes humans do easily, such as speech recognitiorr and machine t e x t analysis. He is the e d i t o r of a b i l i n g u a l journal, Journal of Chinese Linguistics.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0 | null | null | null | null | null | null | null | null |
c694b043669325196fcd395518a0655e6fcec2ba | 245118101 | null | Foundations of machine translation: operations | Compare translation with transportation: Hnnnibal could n o t have conquered Rome if he had waited f o r development of j e t aircraft. Do what *you can d o ; MT is t h e one t h i n g we c n n n o ~ do with present kno~gledge. Consider a s y g t e m , one o f 100 t h a t might bc built. have a problem, can do something about it; b u t choose o n l y & a t we know can be done. First, a d i s p l a y , keyboard, and p o i n t e r . Next; an e d i t o r (program) and d i c t i o n a r y lookup Then, morphological a n a l y s i s , which is linguistically easy. | {
"name": [
"Kay, Martin"
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 0 | null | null | null | null | null | null | Main paper:
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0 | null | null | null | null | null | null | null | null |
b95a58b01b7f27c55d6cc505fdfe8d04738733b3 | 245118149 | null | {CULT}, {C}hinese {U}niversity Language Translator | Research into machine translation at the Cllinese University takes a different approach than the others in that the Chinese University of IIong Kong places a heavy etnphasis on prc-editing the source text instead of post-editing the target text. It i s the only group taking this approach of computer-preeditor partnership. All the other grodps, who realized the FAIIQT is not really attaii~able in the near future, have adopted a tendency to compromise in finding some computer-posteditor partnership. A fixed set of me-editing rules nlust be formulated to enable inexperienced and even mono-lingual people to transform quickly tbe input into machinetranslatable form. With this ar rangemcnt, post-editing can be kept to a minimum, if not all together eliminated. Given time and better programming techniques, these pre-editing rules will gradually be reduced so that the computer will eventually take up this routine work. Pre-editing can therefore solve many of the present linguistic problems that are otherwise dependent on further research in natural language, computational linguistics, and transformation mathetnatics. In the present stage of development, very complex sentences can be translated with the aid of pre-editing, * CULT (Chinese University Language Translator) was developed based on the urinciple mentioned above and has been rigorously examined and tested. Since the beginning of 1975, the CULT-System has been used on regular basis to translate two Chinese scientific journals, ACTA Mathematica Sinica *An average of 5% of text is pre-edited by computer or editor. | {
"name": [
"Loh, Shiu-Chang"
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 1 | null | At present, the semantic a~lalyzer is able to offer only lim ited facilities, and the problem of semantic ambiguities is esse~ltially 'esolved by: 1) a dictionary with specialized subject matter and 2) by pre-editing. | null | The function of the output module is simply to rearrange word-order of the output sentence structure appropriate to the target language. | null | and ACTA Physica Sinica, which are published by the Peking Academy o-fScience. This accomplishment indicates the correctness of our approach an4 the potential capability of CULT. The basic dictionary look-up algorithm employs the 'largest matchTT principle, designed lor Chinese input (i, e. , five digits numbers) and can readily be used for other non-alphabe tic language input. However, anadditional procedure for languages with alphabets (f. e. , English, Malay, etc. )may be required to convert the alphabetic characters into numerical form by forming a t%ash" before performing the look-up.The main function of the syntactic analyzer i s to determine precisely the role that the individual words play in the sentence (i. e . , to which parts of speech the words belong, whether noun, verb, e t c . ) . The process is acco~nplished by means of a rather sophisticated true-false table.While working on the machine trallslntion of Ule Chinese scientific journals, a number of interesting linguistic diCficult ics experienced have been identified and defined. Previously, such strucluyes would have to be pre-edited or post-edited in order to obtain the correct translation, but now they can be readily translated without any pre-editing. | Main paper:
syntactic analyzer module:
The main function of the syntactic analyzer i s to determine precisely the role that the individual words play in the sentence (i. e . , to which parts of speech the words belong, whether noun, verb, e t c . ) . The process is acco~nplished by means of a rather sophisticated true-false table.While working on the machine trallslntion of Ule Chinese scientific journals, a number of interesting linguistic diCficult ics experienced have been identified and defined. Previously, such strucluyes would have to be pre-edited or post-edited in order to obtain the correct translation, but now they can be readily translated without any pre-editing.
semantic analyzer module:
At present, the semantic a~lalyzer is able to offer only lim ited facilities, and the problem of semantic ambiguities is esse~ltially 'esolved by: 1) a dictionary with specialized subject matter and 2) by pre-editing.
output module:
The function of the output module is simply to rearrange word-order of the output sentence structure appropriate to the target language.
:
and ACTA Physica Sinica, which are published by the Peking Academy o-fScience. This accomplishment indicates the correctness of our approach an4 the potential capability of CULT. The basic dictionary look-up algorithm employs the 'largest matchTT principle, designed lor Chinese input (i, e. , five digits numbers) and can readily be used for other non-alphabe tic language input. However, anadditional procedure for languages with alphabets (f. e. , English, Malay, etc. )may be required to convert the alphabetic characters into numerical form by forming a t%ash" before performing the look-up.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0.001704 | null | null | null | null | null | null | null | null |
8d23b5aba9c919c06d12e6b53d51deae97513163 | 245118124 | null | Summary remarks for machine translation conference | An issue whlch emerged early in the conference and recurred either e x p l i c i t l y or, more o f ten, i m p l i c i t l y during subsequent scsslnns concerned the relative values of pragmatic solutions and more basic research. An additdona1 factor w a s the often presumed r e l a t i o n s h i p between more basYc research and s c i e n c e , and between t h e pragmatic and fts synonym, 'ad !~ocness. 1 7 so on.. Although 1've never been a p r o f e s s i o n a l t r a n s l a t o r , I assume that they have analogous requirements. Therefore, I ' d u r g e t h a t e system to be used i n machine translation either provide larger screens or keep a kind of running summary which could be used to alert the translator through underlining, a warning message, or wl~atever , t h a t , for example, a g i v e n word or phrase was being used roo o f t e n . As you see, I am again speaking o t the i s s u e of r e a d a b i l i t y for, i n s o f a r , a s possible, translations should be readable. | {
"name": [
"Sedelow, Sally Yeates"
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 0 | null | between the prggmatics (for example, w r i t i n g a poem) and basic reseach. I would argue, for example, that much research on and criticism about a poem i s , simply, in effect allother poem or s e t of poems, even | null | null | null | null | Main paper:
:
between the prggmatics (for example, w r i t i n g a poem) and basic reseach. I would argue, for example, that much research on and criticism about a poem i s , simply, in effect allother poem or s e t of poems, even
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0 | null | null | null | null | null | null | null | null |
c0853fbc083097b7286fe692ab4f8a236edff22a | 245118125 | null | {C}hinese-{E}nglish Translation Assistance group | The U. S. based, intergovernme~~t/academic CETA (Chinese-English Translation Assistance) Group is building a machine-readable dictio~lary Cilc for use in on-line retrieval and for dcvclopmcl~t of dictioi~arics and indexes for use of human translators. The cspcrimcntnl on-line retrieva! system can store an unlinlited number of entries, The current file of 640,000 machine-readable entries is divided into approsimately 110,000 gcncral entries; 10, 000 colloquial entries; and 500,000 scientific and technical Chi icse-English entries. The exyer imental system designed for an IBNI 360 illustrates the facility of computer storage, retrieval, and display of Chinese characters and Rotn an alphabet as well a s other scripts. It also illustrates the facility of c o t ~~p u t c ~ techniques for i~ldclti~lg Chi.11~ sc characters and special adaptability for spntllesizix~g Chinese queries to search telecodc-sorted files. | {
"name": [
"Mathias, J."
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 1 | null | null | null | null | null | null | Main paper:
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0.001704 | null | null | null | null | null | null | null | null |
8921c3bc0cb04c38a88d863efbbb5ab11bae2251 | 245118106 | null | Semantics and world knowledge in {MT} | I presented very simple and straightforward paragraphs from recent newspapers to show that even t h e most congenial real t e x t s require, for their translation, sume notions of Inference, knowledge, and what 1 call "preferenrules", | {
"name": [
"Wilks, Yorick"
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 1 | null | null | null | null | null | I argued that the MT problem has not been.solved in any sense even though there have been r e a l improvements in t h e performance of large commercial systems, y e t , contrary to some impzessions given at the seminar, we are by nb means exactly where we were twenty years ago and about t o go through the same agonizing cycle of optimism and disillusion again. I then presented a sketch of a small research English- i m p o r t a n t r o l e in advancing MT, by occupying a space, as it were, between t h r e q better-known positions: (i) that we can j u s t go on a s before with "brute force" systems (ii) t h a t we can only g e t advance by devoting ourselves here and now to purely theoretical A 1 systems t h a t "represent a l l knowledge"and (iii) t h a t we should make do with techniques t h a t are simple but f u l l y understood, such as on-line editors. | Main paper:
i argued f i n a l l y that systems of t h i s s o r t can p l a y an:
i m p o r t a n t r o l e in advancing MT, by occupying a space, as it were, between t h r e q better-known positions: (i) that we can j u s t go on a s before with "brute force" systems (ii) t h a t we can only g e t advance by devoting ourselves here and now to purely theoretical A 1 systems t h a t "represent a l l knowledge"and (iii) t h a t we should make do with techniques t h a t are simple but f u l l y understood, such as on-line editors.
:
I argued that the MT problem has not been.solved in any sense even though there have been r e a l improvements in t h e performance of large commercial systems, y e t , contrary to some impzessions given at the seminar, we are by nb means exactly where we were twenty years ago and about t o go through the same agonizing cycle of optimism and disillusion again. I then presented a sketch of a small research English-
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0.001704 | null | null | null | null | null | null | null | null |
78b3da527be6c97aec736a71ce09aca8f6a336a6 | 245118115 | null | Foundations of machine translation: linguistics | There i s presently a theoreltical opening in linguistics. Computers have been unfashionable; t h e p a r t y l i n e has been against them, except in phonetics. Linguistics has s u f f e r e d a r e a l l a g in n1anipulat:ing l a r g e amounts of data. 1 , i n g u i s t b consider MT an impo~sible dream: The dreamer doos not know what kind of thing a language i s . Devices for machine-aided translation do not define a basic aqea for the linguist; the real interest is in sinulating the processes of a human translator. Framework for MT: Surface structure (what 2 s directly represented) vs. deep s t r u c t u r e : a m b i g u i t y , i d i o m s . Translation via conceptual r e p t e s e n t a t i o l n , which may or may not b e the same i n all languages. Nature of the conceptual rcpresentation isthe b a s i c question for many fields. Two views: L o g i c a l n e t , e a s y t o compute, a g r e a t d i s c o v e r y i f correct; analogic form, not easy to compute, a mental image. At what point does one make the image-language conversion? Different p l a n s i n d i f f e r e n t languages: i n Southeast A s i a , the image i s more s p a t i a l than temporal. Japanese does not open a discourse with a summary of what is to follow. Years of hard work and c r e a t i v e insight are needed f o r PIT. Real MT takes such deep knowledge it is utopian. Intermediate goals: Stepwise simulation. (Notes by DGH) | {
"name": [
"Chafe, Wallace L."
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 0 | null | null | A i r Force d e a l i n g w i t h the semantic prerequisites to lhachine t r a n s l a t k o n . | null | null | null | Main paper:
:
A i r Force d e a l i n g w i t h the semantic prerequisites to lhachine t r a n s l a t k o n .
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0 | null | null | null | null | null | null | null | null |
bade638e2018ccc6cf61d3e7b9fcc6a7b30e6d71 | 245118098 | null | {R}ussian-{E}nglish system, {G}eorgetown {U}niversity | Michael Zarechnak The Gcorl;ctown Univereity Russian-Zn~lish System is r u n n i n g on IRM 360/70 .CPU time for ZOO0 words bJ 9 acconds* The t e x t s translated i n c l u d e o b i e n t i f i c , t c c ~n o l o ~a l , a n d economic mnterinls. M.2arcchna.k in c l o s e cooporation w i t h t h e l i n ~u i s t i c research otaff. The linguistic etatcmcnts are codad in symbolic l n n ~u n g e d c s i g n e d by Dr. A .Brown ( ' SLCv-Programming Language ) . Inpu t/outpu t is in Assembler language. A dictionary entry containa a s p l i t or u n s p l i t Ruosian ntem, grammatic a l codinc, lexical number, and E n g l i ~h part. The c l u ~t c r c d entries arc r c c o ~n i z c d through s p e c i a l l o c a l operations when t h e calling ~i g n a l r occur within the sentence under processing. Syntactic a n a l y s i s i s partly based on merphosyntactic markings and partly on s c ~a n t i c coding. | {
"name": [
"Zarechnak, Michael"
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 1 | null | null | The undditcd translation ie used p r i m a r i l y for i n f o r m a t i a n purposce, although i n a few instances, the t r a n s l a t i o n e w e r e post-cditcd when t h e user requested it.The quality o f the p r e~e n t translation i s t h e G a m e ac it was in 1964.The senantic level will be added. S i z e o f the d i c t i o n a r y r 50,000 otoma. | null | null | null | Main paper:
:
The undditcd translation ie used p r i m a r i l y for i n f o r m a t i a n purposce, although i n a few instances, the t r a n s l a t i o n e w e r e post-cditcd when t h e user requested it.The quality o f the p r e~e n t translation i s t h e G a m e ac it was in 1964.The senantic level will be added. S i z e o f the d i c t i o n a r y r 50,000 otoma.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0.001704 | null | null | null | null | null | null | null | null |
5ccd0e2ff4ef306ee013e181fbce82318a7def25 | 245118111 | null | Machine translation: generalities and guide to action | Golem t h e a u t o m a t a is the symbol of man's horror of t h e t h i n g t h a t straddles t h e line bctwccn spirit and flesh. T ! I ~ crumbling tower symbolizes ethnocentricity and sunophobia. Combiaed, these i r r a t i o n a l feelings can influence national palicy and r e t a r d progress toward i m p o r t a n t g o a l s . To hove too fast ienas much an error as n o t to move at all. The principles of t h e f i r s t s e c t i o n summarize my reactLon to the contributions presented at the conference; the guides of the second s e c t i o n express my opinion about t h e making of decisions in a f a i r l y broad area. GENERALITIES 1. Almost everyone h a t e s computers, i n c l u d i n g most computer scientists. In '"Information I I a n d l i n g " ( C u r r e n t Trends in Linguistics, ed. T. A . Sebcnk et al., volume 12, pp. 2719-2740), I noted t h a t professors w h o g i v e their s t udents clever tricks for skimming technical a r t i c l e s r e f u s e to permit their computer programs to use the same t r i c k s ; the computer must work t h e hard way, in accordance w i t h general theories of the s t r u c t u r e of information. A friend suggests t h a t hatred of the machine must be responsible. Anyone who h a t e s computers is likely to d e s i g n cumbersome systems. | {
"name": [
"Hays, David G."
],
"affiliation": [
null
]
} | null | null | Foreign Broadcast Information Service Seminar on Machine Translation | 1976-03-01 | 0 | 0 | null | The more programmers there are, the lower their average skill. In the e a r l y days of computation, t h e few programmers were brilliant; ss the number has increased, the number of brilliant programmers has gone up, but the number of adequate or inadequate programmers has gone up f a s t e r .The buyer of a system must ask which kind will make it.The best in computing is v a s t l y b e t t e r than ever before, but almost everything is worse. T a s k s t h a t required senior professionals long hours ten years ago can now be accomplished by students in courses, because the software is that a news story i s about a certain frame ( d e t e n t e ) , and that the source is Sadat. than t o translate t h e whole; and the summary ("Sadat endorses d e t e n t e " ) might be more helpful to the user than the translation would be.A prima facie case has been made for gradual i n t r oduction of language-processing capacity i n t o intelligence facilities.T h e design should take into account as 'fully as possible the needs.of users of translations. | System design and c o s t analysis remain t h e essential prerequisites to procurement,is from character processing ( e d i t i n g systems) to word processing ( d i c t i o n a r i e s ) . | null | No adequate reason for selecting a single system and excluding the rest has come to l i g h t thus far. | null | Main paper:
.:
System design and c o s t analysis remain t h e essential prerequisites to procurement,
.:
The best in computing is v a s t l y b e t t e r than ever before, but almost everything is worse. T a s k s t h a t required senior professionals long hours ten years ago can now be accomplished by students in courses, because the software is that a news story i s about a certain frame ( d e t e n t e ) , and that the source is Sadat. than t o translate t h e whole; and the summary ("Sadat endorses d e t e n t e " ) might be more helpful to the user than the translation would be.A prima facie case has been made for gradual i n t r oduction of language-processing capacity i n t o intelligence facilities.T h e design should take into account as 'fully as possible the needs.of users of translations.
.:
No adequate reason for selecting a single system and excluding the rest has come to l i g h t thus far.
the main developmental track for a f e w years ahead:
is from character processing ( e d i t i n g systems) to word processing ( d i c t i o n a r i e s ) .
2.:
The more programmers there are, the lower their average skill. In the e a r l y days of computation, t h e few programmers were brilliant; ss the number has increased, the number of brilliant programmers has gone up, but the number of adequate or inadequate programmers has gone up f a s t e r .The buyer of a system must ask which kind will make it.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 587 | 0 | null | null | null | null | null | null | null | null |
01f5532b10c64c59e1938c2db42eb1da89568095 | 219306619 | null | {CLAM}: A computer language model | This paper describes a program which translates Engllsh into French, It is difficult to delineate the subset which a program can deal with, so sample sentences are given. | {
"name": [
"Dobree, Nicholas J. S."
],
"affiliation": [
null
]
} | null | null | null | 1976-05-01 | 0 | 0 | null | There are, I hope three reasons why CLAM will be of interest to computational linguists.( 1 )Ir; 1s a working model. This is not a "paperf1 containing ltideas". Lt is a description of a model which works,To be more specific, it is a description of a large program, written in FORTRAN, which runs on a 960-40.English $ext, carrieg out a syntactic and semantic analysis of! it, stores the result, and translates it into good French.(2) The subset of English which it is capable of analysing is, by present standards, extremely large, The vocabulary is about 1300 words, many of which have a variety of meanings. More important that the size of the vocabulary is, of course, the range of syntactic structure's and, perhaps most significantly, the degree of complexity of senkences which can be dealt with,Increasing the length and complexltg of sentences does not bring likelihood of combinat orial explosion. The amount of working store and computing time required to analyse a sentence is of the order of the number of words in a sentence, although of course it varies according to the number of meanings of the words and the types of syntactic structure involved.(3) The program is continuously extensible. This extensi'bility applies first t~ the subset of language which can be analysed, secondly to the target languages into which translations can be generated, and thirdly bo the uses to which the analysis of the rext can be put. Tn other words, I believe that the program embodies 3 sound method of syntactic and semantic analysis such as must be the basis of a computer l a w a g e model.Extension of the subset of language which can be m a l y s e d is a matter of addition and refinement. It can be stated with confidence that such extension can be achieved because nothingfundamentally different from what has already been achieved is involved. New syntactic structures, well formed or otherwise, can be incorporated, by addition partly to the files and partly to Ohe program. Continual refinements can be made to the method of finding pronoun antecedents. This problem, which seems to be generally accepted as the most difficult single problem in analysis, will never be solved by one simple algorithm, and the fact that a particular program at any given stage of its development gives the wrong answer in a particular case, so far from invalidating the prograp, rather points the way to further refinements (cf. Wilks, June 1975) . What is important is that the program should provide the tools which enable the refinement to be made, and CLAM does this.Extension of the target languages involves applying to other languages the same method which is used to generate French.This can be done, and indeed part of the actual program used for French would be generally applic3ble. It will be interesting to attack a language outside the Indo-European group, and Arabic is the first one I have in mind, although how soon this can be done is a question of time and priocities,The obvious use to which the analysis can be put other than 6 translation is a questionLanswer system, and work on this is at present in hanb. A question-answer system must be based on an effective malyser, and it is beljeved that CLAM can provide this. However, I do not maintain that the analyser S b~u l d be independent of9the memory and inferencihg part of the sy$te~n.Obviously it should not be independent of the memory, since an analyser nu st create and use its own memory, and although it would be theoretically possible for the analyser to have one type of memory and the latter part of the program to have another, this would be a ludicrous arrangement. The same argument applies tp inferencing, which again has to be performed.by an analyser. Therefore it seems that a question-answer system should be more integrated than rnany A.1. reseqrcher,~ appear to allow. On this score, I support the view of Wilks vis-a-vis Charniak.To create a q u e~t i o n~a n s w e r system, and, indeed, to improve the translation program, -t;he memory and the semantics of the present program have to be developed. I use the word "developed" advisedly because I believe that the ekisting memory and semantics form a sound basis upon which a more comprehensive system can be built.It is normal practice when describing a language model to leave d;iscussion of the achievedents of the model until the end. Having said this, I am immediately conYronted by the problem adrunbrated by Woods of how a reader can assess the range and scope of a particular model, and by implication, o f how the programmer can honestly present it. There are two standard methods of prepentation. One is by rather sweeping general statements such as "the program can cope with noun clauses.adjectival clauses, conjunction, questions" etc., according to what claims are being made, Such generalisations are inevitably suspect and rightly so, since no reader will believe that he could not find, for example, adjectival clauses which the program could not cope with. The alternative method of presentation is to give sample sentences which the program has coped with, and hope that the reader will make for himself the type of generalisation which the programmer has scrupulously avoided. If the first method is adopted, the programmer may justifiably be branded as a charlatan. If the second, he runs the risk of having his sent-encesdismissed as "a few examples",The problem is real, and the solution far from obvious: how to define a subset of language. Supposing that we were conCerned only with single sentences and not longer texts; and supposing that it were possible, which evidently it is not, $0 list all the sentences of the subset: the11 how can we find a definition which ,wou&d include a l l the sentences which we have listed and exc1ud.e any which we have not listed? 3. there is no word between the noun and the relative clause unless it is part of a supervening relative clause;4, the noun 1s not part of a subsidiary clause unless the subsidiary clause is itself a relatlve clause provided that (a) the noun is not the object of the clause and (b) the noun is not a 'time1 noun.All of the above provisos are of a type which could well be applicable at any particular stage in the development of a program, although some may be more likely than others. The programmerls difficulty is that until he has tried an appropriate type of sentence, he probably will not realise the existence of a particular limitation. The first indication of it is that the sentence doesn't work, and h e then has to rack his brain to findrout w h y not, and alter the program to eliminate the limitation, thereby enlarging the subset i n that particular direction (hoping that he is not at the same time being so stupid as to reduce it' i n another). can arlalyse both of these sentences:1. The man who came to dinner stole the silver.The man was hit by a bus.Does it follow that it can analyse this?The man who was hit by a bus stole the silver. After -k h i s r a t h w prolix introduction, 1 eorrle eventually to my own "Pist of ssntenc\@st1 ellat have been successftrlly annlyssd by CLAM. They fall into two categqries: those which have been tranglat~d into French, in which case the Frexic$l translation is given; and those which have sirtiply been a n a l y s e d s y l t actically,and semantically and reduced to a base form. Tliis 1 s because durfng the last year 1 have not been working on the French generator but concentrating on certain asyects of the ana1yser;and so in order to save computer time, the French seneration has been omitted, Thus the sentenceswithout translation have been processed last.The question arises of what exactly is meant by "analwed syntactically and semantically" and "reduced to base form". T h s will be more fully explained in the subsequent test. At this, stage a"t is sufficient to say that a syntactic tree has been formed and serntxntic ambiguities resolved, and that semantic relations betr3een words in the tree have been determined ( e . g . a syntactic subject of a passive verb is recorded as the semantic object).Singre word meanings are retained as basic units. There is no Schankian-type resolutiun lnto semantic primitives, e x c e p t insofar as this is implicit in the classification system. This is the base form from which the French has b e e n generated. It has no% so far proved necessary to go any baser. Development, as will be explained later, is envisaged along the lines of extending the network rather than breaking down the units.The follow3.ng are samples of sentences which have been correctly anaW8ed by the program. They are given, together with the French translations where these have been produced by the program, and with comments on points .of interest in the s8rTtences.The shirt which you sold is dirty.La chemise que vous avez vendue est sale.Relative clause.The man and woman doctors saw have eaten the bread, L'homme et la femme que les medecins on* vu ont mange l e pain.Contact clause (relative clause with relative pronoun missing).No article i n English but article required in French.I want the king to read the book.Je veux que le roi lise le livre.AccusativQ and infinitive.I thought she would eat, J'ai pense qufelle rnangerait, Object clause with "that" missing,He hurt some donkeys last month. D i f f e r e n t meanings of "worku.Quand e s -c e que vous avez ouvert l a p o r t e ? Q u e s t i o n . | null | null | La femme a 1 1 a i r 0MUye apd drprime.Semantic r e s o l u t i o n of "looks" l k n u y e ' and ldeprirno should be f ominine. M y ignorarlce. 25. The k i n g i s a s l a r g e a s a cow. Men can understand which book is b e s t .Before going into some detail about the method used to achieve these results, I would like to say something about the danger of over-sophistication on the part of the reader, There is a natural tendency for researchers, on reading something new, to look for points of broad similarity with something, anything, %*hat they have read, before; and, having found it, to sit back with relief and feel absolved -from reading any further. In a field in which vast amounts are being written, it is a proper self-defence on the part of the reader, but in A.1. in particular, it has its special dangers.When one passes from the realm of pure ideas to the-,hard We are now left with a code range for the first 'in1 containing three codes, In such eases, it is the first code of the range which 1 s selected. So 'in' has been dlsambfguated to 621 in the first case, and tg 6311 in the second.Of course, it is all very well for a program to be able to digest, She walked in fields in May.But can it also cope with this?The farmers we were talking about grew, and the greengrocers, thieves and liars, sold those apples.In other words, the program must be capable of being expanded to deal with the myriad complexities and exceptions of natural No one who has struggled with German case endings is ignorant of this, In English, w e have the concord between subject and verb in the third perSbn of the present, patently unnecessary since it exists only in this one instance, There are many sentences i n which the semantics alone a r e clearly sufficient.In the sentence, "The man ate the steak with a fork,", the words could appear in any sequence and the meaning would be decipherable, although it might t a k e longer to decipher. The interesting question is what features of the syntax can be consistently ignored, without occasional sentences cropping up which can only be deciphered with the help of these features.There now follows a description of the treatment of three notoriotisly awkward problems--relative clauses, pronouns, and conjunction.Relative Clauses Six cases are distinguished:1. The man who met you.The man you met.. T h e man who(m) you met.6. The man you gave i t to.3. The rnan who (m) you gave it to.4, The man to whom you gave it.After the lead of a noun EP, In the c a s e s of 941, 942 and 943, the only EP which i s open for the next word is the relative clause EP. For 941 the next necessary word in the EP is the lead verb, while for 942 and 943 the next necessary word is the subject. In practice, one or m o r e of these EPs is usually eliminated on the next word.When a contact noun is recognised, it is marked i n the noun EP as being in reality a relative pronoun. Then the procedure for 942 and 943 above is followed; but in addition, the contact noun is entered as the subject of the relative clause EP. Pronouns For either a translation or a questio~l-a~isweri~lg program, the noun which the pronoun replaces, called here the replacement noun, has to be identified. In a questionallawering program, the reasons are obvious enough. I n a translation progrml, it is necsssary for sertiantic ~natchlng and also because in many target languages the gender of the prohoun varies with that of the replacement noun.The replacement noun might; be i n the same sentence as the pronoun, or in a previous sentence. Therefore, in dealing with pronouns, the program must be able to refer to preceding sentences. So after ENDR, the essential information for the sentence just processed is extracted from the first chain of The man went into the shop where h e had seen the raincoat.H e bought a hat an4 took it away. This "formula of prioritiesv, which is only applied if there i s no semantic preference for one noun, i s probably at 3 9 the moment .a rather blunt instrumeat. It is concerned with two factors -which noun occurred in a 1-ater clause, and which noun has the same function as the pronoun; subject, objept, preposition object, or object of the same preposition.In the majority of cases it produces the correct answer, but it is possible to think up examples in which it doesn't.With experience of use, the formula will be refined.A complication is addea ~y rhe possibility that, when a subdect, 'itt may be impersonal. This sense is treated essentially as one possible replacement noun.There is still work to be done in developing the for,mula of priorities. CLAM extracts the information required to solve the pronoun problem. The question is, how to use it.Conjunction No m t of the program is more complex than that dealing with conjunct'fon. The principles are clear,Qeven Semantic restrictions on syntactically associated pairs of words khich give preference to one meaning. Example:"I killed t h s man with a gun." Here, there is a synta~tic as well as a semantic ambiguity. It is less straightforward than the previous example because the ambiguous word is 'with', whlch might be ~JI instrument preposition attached to the verb k i l l 1 , or a possessicon preposition attached to the noun Vrnanl. The semantic relationships which determine t h e choice, however, ~n l y involve 'witht i n d l ectly. They are between 'kill1and 'gun1 in one case, and between 'man1 and 'gun1 in the other, NormalLy tlXle preference would be for the instrunlent;interpretation because l g~p l ' is more strongly associated with 'killt as an instrument than with 'man1 as a possession. CLISrcI shall see in a moment, it is not always correct to do so, 5, Remoter contextual environment. Sometimes the factors enabling a choice to be made are more remote from the word in question than In the examples given above, In order to find these factors, a longer journey has to be made into the environment of the word, Examples: (i) "The mayor hit the alderman so hard that he fell down." The normal rules for selection of pronotin antecedents would prefer 'mayor' as the antecedent of 'he' because it i s the subjec-tt, but in the environment of hitting, it i s much more likely to be the person hit who falls down rather than t b hitter, so 'alderman' must be preferred.(ii) "Two men came in. One had a gun and tho other had a knife. I killed the man with a gun." Here 'withT is obviously not an instrument preposition attached to lkilll, but a possession preposition attached to 'menv. This is so because the definite article 'thet attached to vmpn' implies that 'man1 has already been defined, But i n fact two men have already been defined, and more information is needed to determine which of them is referred to. The only possible additional information which could satisfy this requirement is 'with a gunr, whioh does suffice to distinguish one of the previously determined men. Therefore this phrase must be attached to 'manv, At present, CLAM could not resolve either sf these ambiguities. In order to do so it would need, in the fix"st case, more information about the environment of 'hitt than is contained in the semantic restrictions now at its disposal, and i n the second case, both a better memory and a routine for dealing with definition of nouns. Work is in progress on these vital additions. They will involve adding to the type and range of the semantic relationships between pairs of words referred to in the definition of the purpose of analysis given at the beginning of this summary.At present, CLAM only holds semantic relationships be-tween words which are syntactically related. This is not enough.Adding to the types of relationships held, and extending them to pairs of words which are syntactically remote, will greatly increase the scope of the model.As shown in the flowchart, the sentence is operated o n sequentially by four subroutines Should it not be done avring the reduction of the E n g l i s h sentence to base fo'rm? The answer is that logically it should, and it will sooner or later be transferred, probably to ENDR. But at present it doesn't matter. The part of the program described in the section on pronouns which stores the base form of the last sentences i s i n fact performed after the French translation has been generated, and therefore, after the verbs have been welded.ITRN takes each word in the sentence in turn. I;t finds the code number in FRILE, the French dictionary file, and extracts the French word(s). Sometimes of course there is more than one, Sometimes there is zero because the English word does not have to be translated. Any particular French word may not have the same function in the sentence as the English word. In such cases, the French word entry in FRILE is followed by a code which specifies the word's function inx'glation to the English word being translated. For example, if 2.32237 isl the code for 'potato', -t;hen'FR1LE entry will be 212237 POMME F DE 6 TERRE 6x2. The F after POMME shows that it is feminine. The 6 after DE shows that its x'unc$ion is as a plcepqsition in the EP of which POMME is the lead. The 6x2after T E W shows that it is the object in the EP of which DE is the lead,Sometimes it isaecessary to go up Che tree. Fop exampleY1x5 means a n adverb (5) in the verb EP (1) of which the English word is a subsidiary ( Y ) .It is thus possible to generate a French sentdnce of a radically different shape from English.ITRY also finds a French sequence code for each word. At present the processing takes about 15 seconds per word o n average, of which READ takes 4076, the semantic and syntactic analysis about 20%, and the French generation 40%.No serims attempt has y e t been made to optimise the program and this time could certainly be peduced. But the reduction would be offset by the eventual need to keep JSP on disk. So as a practical proposition for translating texts, it would be necessary for the processing time to be reduced by a factor of about 10. Presumably this will come sooner or later with improvement in hardware.There are certain improvements which would have -bo be made to the p o g r a m before it could be used, apart from the extension of the vocabulary. Most obvious: " A s N comparative.Bstored. It may be noted in passing that the system of coding contains the elements of both syntactic and semantic classification. The distinction between the two is at tlmes tenuous.Further explanation of the coding is given in the appendix. | Buvez p l u s rapidement l e l a i t .10 Peel the potatoes for youp mother.Epluchez les pornmes de terre pour votre mere.Multiple-word noun.Teachers write pla$s in March in some countnies.Les instituteurs ecrivent des pieces en Mars dans des campagnes.Semantic resolution of "in".lDes campagnest should be tcertnJrlrpays14. H e stood up to put the fire off.I1 stest leve pour eteindre le chauffage.Two-word verbs.15. That waiter, fat and stupid, was breaking the plates.Ce serv'eur gros et stupide cassait les assiettes, Appositional adjectives between commas.Continuous tense.The man who drank the wine does not laugh.L1hornrne qui a bu le vin ne rit pas.Negative .17. You frightened the man whose pen you stoCLe,Vous avez effraye lthomme dont vous avee vole la plume."Whose"--difficult constsuction,18. The woman who you swam with is happy.La femme avec qui vois avez nage est contente.Floating preposition at end of relative clause. | Main paper:
,:
The man and woman doctors saw have eaten the bread, L'homme et la femme que les medecins on* vu ont mange l e pain.Contact clause (relative clause with relative pronoun missing).No article i n English but article required in French.I want the king to read the book.Je veux que le roi lise le livre.AccusativQ and infinitive.I thought she would eat, J'ai pense qufelle rnangerait, Object clause with "that" missing,He hurt some donkeys last month. D i f f e r e n t meanings of "worku.
.:
The man who drank the wine does not laugh.L1hornrne qui a bu le vin ne rit pas.Negative .17. You frightened the man whose pen you stoCLe,Vous avez effraye lthomme dont vous avee vole la plume."Whose"--difficult constsuction,18. The woman who you swam with is happy.La femme avec qui vois avez nage est contente.Floating preposition at end of relative clause.
whgn d i d you open the door?:
Quand e s -c e que vous avez ouvert l a p o r t e ? Q u e s t i o n .
drink t h e m i l k f a s t e r .:
Buvez p l u s rapidement l e l a i t .10 Peel the potatoes for youp mother.Epluchez les pornmes de terre pour votre mere.Multiple-word noun.Teachers write pla$s in March in some countnies.Les instituteurs ecrivent des pieces en Mars dans des campagnes.Semantic resolution of "in".lDes campagnest should be tcertnJrlrpays14. H e stood up to put the fire off.I1 stest leve pour eteindre le chauffage.Two-word verbs.15. That waiter, fat and stupid, was breaking the plates.Ce serv'eur gros et stupide cassait les assiettes, Appositional adjectives between commas.Continuous tense.
the woman l o o k s d e p r e s s e d and bored.:
La femme a 1 1 a i r 0MUye apd drprime.Semantic r e s o l u t i o n of "looks" l k n u y e ' and ldeprirno should be f ominine. M y ignorarlce. 25. The k i n g i s a s l a r g e a s a cow. Men can understand which book is b e s t .Before going into some detail about the method used to achieve these results, I would like to say something about the danger of over-sophistication on the part of the reader, There is a natural tendency for researchers, on reading something new, to look for points of broad similarity with something, anything, %*hat they have read, before; and, having found it, to sit back with relief and feel absolved -from reading any further. In a field in which vast amounts are being written, it is a proper self-defence on the part of the reader, but in A.1. in particular, it has its special dangers.When one passes from the realm of pure ideas to the-,hard We are now left with a code range for the first 'in1 containing three codes, In such eases, it is the first code of the range which 1 s selected. So 'in' has been dlsambfguated to 621 in the first case, and tg 6311 in the second.Of course, it is all very well for a program to be able to digest, She walked in fields in May.But can it also cope with this?The farmers we were talking about grew, and the greengrocers, thieves and liars, sold those apples.In other words, the program must be capable of being expanded to deal with the myriad complexities and exceptions of natural No one who has struggled with German case endings is ignorant of this, In English, w e have the concord between subject and verb in the third perSbn of the present, patently unnecessary since it exists only in this one instance, There are many sentences i n which the semantics alone a r e clearly sufficient.In the sentence, "The man ate the steak with a fork,", the words could appear in any sequence and the meaning would be decipherable, although it might t a k e longer to decipher. The interesting question is what features of the syntax can be consistently ignored, without occasional sentences cropping up which can only be deciphered with the help of these features.There now follows a description of the treatment of three notoriotisly awkward problems--relative clauses, pronouns, and conjunction.Relative Clauses Six cases are distinguished:1. The man who met you.The man you met.. T h e man who(m) you met.6. The man you gave i t to.3. The rnan who (m) you gave it to.4, The man to whom you gave it.After the lead of a noun EP, In the c a s e s of 941, 942 and 943, the only EP which i s open for the next word is the relative clause EP. For 941 the next necessary word in the EP is the lead verb, while for 942 and 943 the next necessary word is the subject. In practice, one or m o r e of these EPs is usually eliminated on the next word.When a contact noun is recognised, it is marked i n the noun EP as being in reality a relative pronoun. Then the procedure for 942 and 943 above is followed; but in addition, the contact noun is entered as the subject of the relative clause EP. Pronouns For either a translation or a questio~l-a~isweri~lg program, the noun which the pronoun replaces, called here the replacement noun, has to be identified. In a questionallawering program, the reasons are obvious enough. I n a translation progrml, it is necsssary for sertiantic ~natchlng and also because in many target languages the gender of the prohoun varies with that of the replacement noun.The replacement noun might; be i n the same sentence as the pronoun, or in a previous sentence. Therefore, in dealing with pronouns, the program must be able to refer to preceding sentences. So after ENDR, the essential information for the sentence just processed is extracted from the first chain of The man went into the shop where h e had seen the raincoat.H e bought a hat an4 took it away. This "formula of prioritiesv, which is only applied if there i s no semantic preference for one noun, i s probably at 3 9 the moment .a rather blunt instrumeat. It is concerned with two factors -which noun occurred in a 1-ater clause, and which noun has the same function as the pronoun; subject, objept, preposition object, or object of the same preposition.In the majority of cases it produces the correct answer, but it is possible to think up examples in which it doesn't.With experience of use, the formula will be refined.A complication is addea ~y rhe possibility that, when a subdect, 'itt may be impersonal. This sense is treated essentially as one possible replacement noun.There is still work to be done in developing the for,mula of priorities. CLAM extracts the information required to solve the pronoun problem. The question is, how to use it.Conjunction No m t of the program is more complex than that dealing with conjunct'fon. The principles are clear,Qeven Semantic restrictions on syntactically associated pairs of words khich give preference to one meaning. Example:"I killed t h s man with a gun." Here, there is a synta~tic as well as a semantic ambiguity. It is less straightforward than the previous example because the ambiguous word is 'with', whlch might be ~JI instrument preposition attached to the verb k i l l 1 , or a possessicon preposition attached to the noun Vrnanl. The semantic relationships which determine t h e choice, however, ~n l y involve 'witht i n d l ectly. They are between 'kill1and 'gun1 in one case, and between 'man1 and 'gun1 in the other, NormalLy tlXle preference would be for the instrunlent;interpretation because l g~p l ' is more strongly associated with 'killt as an instrument than with 'man1 as a possession. CLISrcI shall see in a moment, it is not always correct to do so, 5, Remoter contextual environment. Sometimes the factors enabling a choice to be made are more remote from the word in question than In the examples given above, In order to find these factors, a longer journey has to be made into the environment of the word, Examples: (i) "The mayor hit the alderman so hard that he fell down." The normal rules for selection of pronotin antecedents would prefer 'mayor' as the antecedent of 'he' because it i s the subjec-tt, but in the environment of hitting, it i s much more likely to be the person hit who falls down rather than t b hitter, so 'alderman' must be preferred.(ii) "Two men came in. One had a gun and tho other had a knife. I killed the man with a gun." Here 'withT is obviously not an instrument preposition attached to lkilll, but a possession preposition attached to 'menv. This is so because the definite article 'thet attached to vmpn' implies that 'man1 has already been defined, But i n fact two men have already been defined, and more information is needed to determine which of them is referred to. The only possible additional information which could satisfy this requirement is 'with a gunr, whioh does suffice to distinguish one of the previously determined men. Therefore this phrase must be attached to 'manv, At present, CLAM could not resolve either sf these ambiguities. In order to do so it would need, in the fix"st case, more information about the environment of 'hitt than is contained in the semantic restrictions now at its disposal, and i n the second case, both a better memory and a routine for dealing with definition of nouns. Work is in progress on these vital additions. They will involve adding to the type and range of the semantic relationships between pairs of words referred to in the definition of the purpose of analysis given at the beginning of this summary.At present, CLAM only holds semantic relationships be-tween words which are syntactically related. This is not enough.Adding to the types of relationships held, and extending them to pairs of words which are syntactically remote, will greatly increase the scope of the model.As shown in the flowchart, the sentence is operated o n sequentially by four subroutines Should it not be done avring the reduction of the E n g l i s h sentence to base fo'rm? The answer is that logically it should, and it will sooner or later be transferred, probably to ENDR. But at present it doesn't matter. The part of the program described in the section on pronouns which stores the base form of the last sentences i s i n fact performed after the French translation has been generated, and therefore, after the verbs have been welded.ITRN takes each word in the sentence in turn. I;t finds the code number in FRILE, the French dictionary file, and extracts the French word(s). Sometimes of course there is more than one, Sometimes there is zero because the English word does not have to be translated. Any particular French word may not have the same function in the sentence as the English word. In such cases, the French word entry in FRILE is followed by a code which specifies the word's function inx'glation to the English word being translated. For example, if 2.32237 isl the code for 'potato', -t;hen'FR1LE entry will be 212237 POMME F DE 6 TERRE 6x2. The F after POMME shows that it is feminine. The 6 after DE shows that its x'unc$ion is as a plcepqsition in the EP of which POMME is the lead. The 6x2after T E W shows that it is the object in the EP of which DE is the lead,Sometimes it isaecessary to go up Che tree. Fop exampleY1x5 means a n adverb (5) in the verb EP (1) of which the English word is a subsidiary ( Y ) .It is thus possible to generate a French sentdnce of a radically different shape from English.ITRY also finds a French sequence code for each word. At present the processing takes about 15 seconds per word o n average, of which READ takes 4076, the semantic and syntactic analysis about 20%, and the French generation 40%.No serims attempt has y e t been made to optimise the program and this time could certainly be peduced. But the reduction would be offset by the eventual need to keep JSP on disk. So as a practical proposition for translating texts, it would be necessary for the processing time to be reduced by a factor of about 10. Presumably this will come sooner or later with improvement in hardware.There are certain improvements which would have -bo be made to the p o g r a m before it could be used, apart from the extension of the vocabulary. Most obvious: " A s N comparative.Bstored. It may be noted in passing that the system of coding contains the elements of both syntactic and semantic classification. The distinction between the two is at tlmes tenuous.Further explanation of the coding is given in the appendix.
intfroduction:
There are, I hope three reasons why CLAM will be of interest to computational linguists.( 1 )Ir; 1s a working model. This is not a "paperf1 containing ltideas". Lt is a description of a model which works,To be more specific, it is a description of a large program, written in FORTRAN, which runs on a 960-40.English $ext, carrieg out a syntactic and semantic analysis of! it, stores the result, and translates it into good French.(2) The subset of English which it is capable of analysing is, by present standards, extremely large, The vocabulary is about 1300 words, many of which have a variety of meanings. More important that the size of the vocabulary is, of course, the range of syntactic structure's and, perhaps most significantly, the degree of complexity of senkences which can be dealt with,Increasing the length and complexltg of sentences does not bring likelihood of combinat orial explosion. The amount of working store and computing time required to analyse a sentence is of the order of the number of words in a sentence, although of course it varies according to the number of meanings of the words and the types of syntactic structure involved.(3) The program is continuously extensible. This extensi'bility applies first t~ the subset of language which can be analysed, secondly to the target languages into which translations can be generated, and thirdly bo the uses to which the analysis of the rext can be put. Tn other words, I believe that the program embodies 3 sound method of syntactic and semantic analysis such as must be the basis of a computer l a w a g e model.Extension of the subset of language which can be m a l y s e d is a matter of addition and refinement. It can be stated with confidence that such extension can be achieved because nothingfundamentally different from what has already been achieved is involved. New syntactic structures, well formed or otherwise, can be incorporated, by addition partly to the files and partly to Ohe program. Continual refinements can be made to the method of finding pronoun antecedents. This problem, which seems to be generally accepted as the most difficult single problem in analysis, will never be solved by one simple algorithm, and the fact that a particular program at any given stage of its development gives the wrong answer in a particular case, so far from invalidating the prograp, rather points the way to further refinements (cf. Wilks, June 1975) . What is important is that the program should provide the tools which enable the refinement to be made, and CLAM does this.Extension of the target languages involves applying to other languages the same method which is used to generate French.This can be done, and indeed part of the actual program used for French would be generally applic3ble. It will be interesting to attack a language outside the Indo-European group, and Arabic is the first one I have in mind, although how soon this can be done is a question of time and priocities,The obvious use to which the analysis can be put other than 6 translation is a questionLanswer system, and work on this is at present in hanb. A question-answer system must be based on an effective malyser, and it is beljeved that CLAM can provide this. However, I do not maintain that the analyser S b~u l d be independent of9the memory and inferencihg part of the sy$te~n.Obviously it should not be independent of the memory, since an analyser nu st create and use its own memory, and although it would be theoretically possible for the analyser to have one type of memory and the latter part of the program to have another, this would be a ludicrous arrangement. The same argument applies tp inferencing, which again has to be performed.by an analyser. Therefore it seems that a question-answer system should be more integrated than rnany A.1. reseqrcher,~ appear to allow. On this score, I support the view of Wilks vis-a-vis Charniak.To create a q u e~t i o n~a n s w e r system, and, indeed, to improve the translation program, -t;he memory and the semantics of the present program have to be developed. I use the word "developed" advisedly because I believe that the ekisting memory and semantics form a sound basis upon which a more comprehensive system can be built.It is normal practice when describing a language model to leave d;iscussion of the achievedents of the model until the end. Having said this, I am immediately conYronted by the problem adrunbrated by Woods of how a reader can assess the range and scope of a particular model, and by implication, o f how the programmer can honestly present it. There are two standard methods of prepentation. One is by rather sweeping general statements such as "the program can cope with noun clauses.adjectival clauses, conjunction, questions" etc., according to what claims are being made, Such generalisations are inevitably suspect and rightly so, since no reader will believe that he could not find, for example, adjectival clauses which the program could not cope with. The alternative method of presentation is to give sample sentences which the program has coped with, and hope that the reader will make for himself the type of generalisation which the programmer has scrupulously avoided. If the first method is adopted, the programmer may justifiably be branded as a charlatan. If the second, he runs the risk of having his sent-encesdismissed as "a few examples",The problem is real, and the solution far from obvious: how to define a subset of language. Supposing that we were conCerned only with single sentences and not longer texts; and supposing that it were possible, which evidently it is not, $0 list all the sentences of the subset: the11 how can we find a definition which ,wou&d include a l l the sentences which we have listed and exc1ud.e any which we have not listed? 3. there is no word between the noun and the relative clause unless it is part of a supervening relative clause;4, the noun 1s not part of a subsidiary clause unless the subsidiary clause is itself a relatlve clause provided that (a) the noun is not the object of the clause and (b) the noun is not a 'time1 noun.All of the above provisos are of a type which could well be applicable at any particular stage in the development of a program, although some may be more likely than others. The programmerls difficulty is that until he has tried an appropriate type of sentence, he probably will not realise the existence of a particular limitation. The first indication of it is that the sentence doesn't work, and h e then has to rack his brain to findrout w h y not, and alter the program to eliminate the limitation, thereby enlarging the subset i n that particular direction (hoping that he is not at the same time being so stupid as to reduce it' i n another). can arlalyse both of these sentences:1. The man who came to dinner stole the silver.The man was hit by a bus.Does it follow that it can analyse this?The man who was hit by a bus stole the silver. After -k h i s r a t h w prolix introduction, 1 eorrle eventually to my own "Pist of ssntenc\@st1 ellat have been successftrlly annlyssd by CLAM. They fall into two categqries: those which have been tranglat~d into French, in which case the Frexic$l translation is given; and those which have sirtiply been a n a l y s e d s y l t actically,and semantically and reduced to a base form. Tliis 1 s because durfng the last year 1 have not been working on the French generator but concentrating on certain asyects of the ana1yser;and so in order to save computer time, the French seneration has been omitted, Thus the sentenceswithout translation have been processed last.The question arises of what exactly is meant by "analwed syntactically and semantically" and "reduced to base form". T h s will be more fully explained in the subsequent test. At this, stage a"t is sufficient to say that a syntactic tree has been formed and serntxntic ambiguities resolved, and that semantic relations betr3een words in the tree have been determined ( e . g . a syntactic subject of a passive verb is recorded as the semantic object).Singre word meanings are retained as basic units. There is no Schankian-type resolutiun lnto semantic primitives, e x c e p t insofar as this is implicit in the classification system. This is the base form from which the French has b e e n generated. It has no% so far proved necessary to go any baser. Development, as will be explained later, is envisaged along the lines of extending the network rather than breaking down the units.The follow3.ng are samples of sentences which have been correctly anaW8ed by the program. They are given, together with the French translations where these have been produced by the program, and with comments on points .of interest in the s8rTtences.The shirt which you sold is dirty.La chemise que vous avez vendue est sale.Relative clause.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 585 | 0 | null | null | null | null | null | null | null | null |
57e652771cd4cf47d14c08c8ba5a44cfc81f29a5 | 219306605 | null | A Survey of Syntactic Analysis Procedures for Natural Language | FYris survey was prepared under contract No. N00014-67A-0467-0032 w i t h the Office of N a v a l Research, and was o r i g i n a l l y i s s u e d as Report No. NSO-8 of the Courant Institute of Mathematical Sciences, New York ~niversity. | {
"name": [
"Grishman, Ralph"
],
"affiliation": [
null
]
} | null | null | null | 1976-05-01 | 13 | 0 | null | A Parser for U n r e s t r i c t e d R e w r i t i n g Rule Grammars.. 54 d e p e n d i n g on t h e a p p l i c a t i o n , t h e o u t p u t may b e a command i n a n information retrieval l a n g u a g e , a s t r u c t u r e based on some set of semantic " p r i m i t i v e s " , o r a t a b u l a r s t r u c t u r e s u i t a b l e as a d a t a base; this s t a g e should r e c o g n i z e some o f the p a r a p h r a s e s due t o a l t e r n a t i v e choices of words. ( 3 ) pracjmatic: i n t e r p r e t s the t e x t based on p a r t i c u l a r c o n t e x t (problem s i t u a t i o n or data base) ; this stage should r e c o g n i z e s e n t e n c e s which are e q u i v a l e n t i n e f f e c t , (such as " T h r~w that switch." and "Turn on the light.") .The reader w i l l note t h a t these stages are very-.vaguely charact e r i zed. Current language p r o c e s s i n g s y s tems differ very g r e a t l y i n t h e i r s t r u c t u r e and n o t oven these general d i v i s i o n s can be identified in a l l s y s t e m s . Ihe pragmatic s t a g e is the most heterogeneous and t h e common t h r e a d s which do appear are based more on g e n e r a l problem-solving methods t h a n on specifically linguistic t e c h n i q u e s . Since the s e m h n t i c s t a g e maps into the n o t a t i o n required by the p r a g m a t i c s , T h e four t y p e s are : (0) Develops a s i n g l e parse tree a t a t i m e ; a t any i n s t a n t t h e store holds a set o f nodes c o r r e s p o n d i n g t o the nodes of an incomplete potential parse tree f t ) The s t o r e holds a s e t of n o d e s , each of which represents t h e fact t h a t some substring of t h e s e n t e n c e , from word f to word R , can be a n a l y z~d as some symbol N. <ao, a l r a 2 , .. ,a, The v a r i a b l e NODES holds a c o u n t of the number of npdes i n t h e p a r s e t r e e ; this is also t h e number af t h e node most r q c e n t l y added to t h e tree. WORD h o l d s the number of t h e n e x t word in fie s e n t e n c e to be matched.The heart of the parser is the r e c u r s i v e p r o c e d~r e EXPAND. E X P L t D i s passed one a r g u m e n t , the number of a node i n t h e parse tree. I Y EXPAND has n o t been called for this node before, i t w i l l try t o expand the node, fbe. , build a p a r s e t r e e below the node which matches p a r t of t h e r e m a i n d e r of the s e n t e n c e . If EXPAND h a s a l r e a d y been c a l l e d once for this node -s o t h a t a tree a l r e a d y e x i s t s below this node --EXPAND tries to f i n d an a l t e r n a t e tree below t h e node w h i c h w i l l match u p with p a r t of the remainder of the s e n t e n c e .If EXPAND i s successful -an ( a l t e r n a t e ) tree below the node was found -it r e t u r n s the value true; i f it is u n s u e e e s s f u l , ,/* if t r e e s p a n s e n t i r e s e n t e n c e , add to set */ This algorithm i s sometimes called t h e "Immediate Co~stituent A n a l y s i s " (IcA) alqorithm, because i t was u s e d q u i t e early i n parsing n a t u r a l langqage with ICA grammars, 1%-c o n s t r u c t s a l l nodes in a single l e f t -t o -r i g h t p a s s over the s e n t e n c e . A s each word i s scanned, the p a r s e r builds a l l nodes w h i c h subsume a p o r t i o n of t h e s e n t e n c e ending a t that word. The nodes ( " s p a n s " ) are accumualted i n a two-dimensional a r r a y S P A N , whose f i r s t subscript s p e c i f i e s the number of the span and whose second subscript selects a component of the span, a s follows: SPAN (n, 'NAME ' ) = name of s p a n n SPAN (n, 'FW' ) = number of f i r s t sentence word subsumed by span n SPAN (n, 'LW+ll) = (number of last s e n t e n c e word subsumed by span n) + 1 SPAN ( n , 'D'AUGHTERS') = t u p l e of numbers of d a u g h t e r s p a n s o f s p a n n. (1 <= \tI <= SPANS I (SP.W(I;'NAMG') aq ROOT) and (SPAN (1, ' FW' ) eq 1) and (SPAN(I,'LW+lt) eq SE SENTENCE)+^))) NODES = 1; TREE = nl;T R E E C~) = S P A N I I I ; TREE: (1, 'DAUGHTERS ' ) = E X P A N D (TREE (I, ' DAUGEITERS ' ) , 1) ; r e t u r n s a s e t , each e l e m e n t of w h i c h is an ( n t l ) -t u p l e , whose t a i l is one of the n -t u p l eof s p a n s and whose head i s the r l u n h e r of t h e first word s p a n n e d by the n -t u p l e of s p a n s * / if E L I S T cq n u 9 t then r e t u r n ( ( E N D w D P~~~) ; e l s e r e t u r n [U: i f EXPANDED (TREE (X, 'NAME ' ) , TREE (X, ' FW' ) ) eq true t h e n /* the expansions f o r this symbol have been computed before */ /* if t h i s is a new node, get its WFS e n t r i e s */ The type 2 bottom-up parallel algorithm a l s o saw e a r l y use i n 5% n a t u r a l language processing. The parser d e s i g n e d by Cocke for the Rand system w a s a special version of t h i s a l g o r i t h m for Chomsky normal form grammars. A thorough survey of t h e d i f f e r e n t o r d e r i n g strategies p o s s i b l e w i t h t h i s algorithm was given by Hays [1967] . This algorithm was subsequently developed by Cocke (among others) into a t y p e 1 bottom-up paralleL algorithm named "nodal spans" and s u b s e q u e n t l y i n t o a type .1 top-down p a r a l l e l algorithm called "improved nodal spans" (see Cocke [1970] f The advantage of type 1 over t y p e 2 algorithms depends on t h e degree of ambiguity of the grammar. How f r e q u e n t l y can a p o r t i o n of i+ s e n t e n c e be a n a l y z e d as a particular symbol i n s e v e r a l ways?1 <= 1 <= NODESJ (if ( S P A N (I, ' N A M E ' ) eq ELXST ( # E L I S T ) ) and (SPAN (I, 'LFJ+ll ) eq E N D W D P 1 ) then CNTUP + <I>, NTUP E bylTCH ( E L I S T (1: ( $ E L I S T ) ,SPAN (1, 'FlY' ) ) 1if TREE ( x , 'ALTERNATE WFS ' ) eq Q then TREE(X,'ALTERNATE WFS') = IS, 1 -< S -< WFSS I (WFS (S , 'NAME ) eq TREE a, 'For unaugmented c o n t e x t -f r e e grammars t h e answer in general has been very f r e q u e n t l y -t h i s was one of t h e problems of t h e context-free systems. For such grammars, type 1 algorithms would be much more e f f i c i e n t . When r e s t r i c t i o n s are added, however, t h e y discriminate some of the analyses from o t h e r s . A number of n a t u r a l language systems, such as REL (at t h e C a l i f o r n i a Institute of Technology) a n d the Q-system (at t h e U n i v e r s i t y of Montreal) have used u n r e s t r i c t e d phrase s t r u c t u r e grammars. In such grammars, each rule specifies t h a t some sequence of symbols be r e w r i ' t t e n as some o t h e r sequence of symbols. The parsing algorithm used i n these s y s t e m was described by Kay in 1967 ("Experiments w i t h a Powerful. P a r s e r , " Martin Kay, ih 26me C o n f e r e n c e I n t e r n a t i o n a l e sur le Traitmcnt Automatique des Langues , Grenoble) . however, we s h a l l n o t be concerned w i t h these a d d i t i o n a l f e a t u r e s ; o n l y t h e b a s i c p a r s i n g procedure will be described below.The parser to be p r e s e n t e d r e p r e s e n t s o n l y a small modification to the context-free parser B (the "immediate c o n s t i t u e n t a n a l y z e r " ) g i v e n e a r l i e r . To u n d e r s t a n d t h i s modification, consider t h e f o l l o w i n g example. We a r e given a context-Free grammar which includes t h e p T h e parscr would then a p p l y x -c y a i n r c v c r s e : g e t t i n g a s p a n which subsumes the e n t i r e sentence. Now c o n s i d e r a n a l y z i n g t h e same s e n t e n c e w i t h t h e u n r e s t r i c t e d p h r a s e s t r u c t u r e ! grammar z + y ax + z bWe b e g i n by ilsing t h e f i r s t production tc r e d u c e the s e~~t c n c e to y a b . T h i s raises the problem of how to l a b e l tthc no~ic between arcs a and b. ~l t h o u g h a and b t o g e t h e r s u b s~~a the l a s t t h r e e words of the s e n t e n c e , no f r a c t i o n of-this C~I I be a s s i g n e d i n d i v i d u a l l y to a or to b; h e n c e we c a n n o t l a b e l this new node with t h e number of a sentence r e t u r n ; end ADDSPANS;T h e u n r e s t r i c t e d r e w r i t i n g r u l e p a r s e r has t h e power of a 1966 ,1973 ,1975 Keyser 1967 ; P l a t h 1974a, 1974bl.Interestingly enough, these modifications have b r o u g h t Petrick's parser much closer t o the o r i g i n a l MITRE design.Since the structure of trans fom'iational grammar h a s v a r i e d in t i m e and b e t w e e n d i f f e r e n t schools of linguistic t h e o r y , the notion of a transformational parser is not we11 defined. In order to p r e s e n t a p a r s i n g algorithm, we have selected a particularly simple granular formulation. This E o r i n u l a t i o n corresponds a~p r o x i r n a t e~l y to the early work o fChomsky ( e . g . , Syntactic S t r u c t z t r e s ) a n d If s c i j is an integer,between 1 and n, t h e new node is t h e node matched to the s c i j -t h e l e m e n t of t h e s t r u c t u r a l i n d e x ; i f SCij is a t e r m i n a l symbol, the new node is a t e r m i n a l node w i t h t h a t name.Because the v a l u e of sci may be t h e n u l l t u p l e < >, i t is p o s s i b l e f o r a node in the tree to be left w i t h n o successors. We t h e r e f o r e "clean u p " the tree a f t e r a p p l y i n g the t r a n s f o r m a t i o n by d e l e t i n g any n o n t e m i n a l node not dominating at l e a s t one t e r m i n a l node.The p r e s c r i p t i o n just g i v e n is i n a d e q u a t e f o r components ~f the structural i n d e x e q u a l to " X " , since these may match zero or moxe t h a n one node. ide s h a l l cons t r a i n t h e t r a n s f o r n a - The r u l e COMP -+ # S # , which maes the base conponcnt r e c u r s i v e , a l s o p l a y s a s p e c i a l r o l e in t h e t r a n s f o r m a t i o n s .appears in the parse t r c e , w e call the tree dominated by t h a t S a c ? 3 ) i s t i t l t d n t (or embedded) s e n t e n c e , and t h e tsce &~n i n a t e d by the n e x t S above COMP t h e rnadl?,ix s e n t e n c e parse t r e e . . > o t e n t i a l surface structure parse t r e e s for a s e n t e n c e and $hen, by a p p l y i n g the t r a n s f o r m a t i o n s i n r e v e r s e , try t o obtain a v a l i d deep s t r u c t u r e from each of these. We shall deal w i t h these t w o steps i n t u r n .The surface s t r u c t u r e p a r s e tree will, i n general, c o n t a i n many s t r u c t u r e s which could n o t be directly g e n e r a t e d by t h e base component. Because t h e l a n g u a g e d e f i n e d by the augmented base component i s larger t h a n t h a t d e f i n e d by t h e t r a n x E o r m a t i o n a 1 graminar, by some early results o b t a i n e d by the MITRZ group. The MITRE system did not have a procedure for automatically augmenting the base component; t h e i r s was assembled manually. Using a small g r m a r . , one of their 12-word test s e n t e n c e s obtained 48 surface analyses, almost a l l of thorn s p u r i o u s . P e t r i c k had s i m i l a r experience: he found that t h e covering grammars produced by his procedure-were too broad, p r o d u c i n g too many surface p a r s e s . He has i n s t e a d , l i k e t h e MITRE group, produced his surface grammars manually, by a n a l y z i n g o o n s t r u c t i o n s which appear in the s u r f ace s t r u c t u r e of i n p u t sentences to determine w h i c h productions are r e q u i r e d . In How can this s h u f f l e be reversed? W e begin by c r e a t i n g a n(1 <= j <= r) according to. j i s i ( j ) = if r s c ( j ) is an i n t e g e r t h e n s i ( r s c ( j ) o r i g i n a l string s j s ( j ) = if isc( j ) i s an i n t e g e r t h e n s ' ( k c ( j ) ) else iscl j) I f there are several matches to t h e isi, the t r a n s f o r m a t i o n must be a p p l i e d t o a l l ; w e can o n l y be s u r e t h a t one of t h e r e s u l t i n g strings w i l l be s . I f the forward t r a n s f o r m a t i o n 1 is a recoverable deletion i n v o l v i 3 s i d e n t i t y conditions, t h e formulas g i v e n above are somewhat more complicated.Given a s e t of reverse t r a n s f o r m a t i o n s , we m u s t f i n a l l y specify t h e s e q u e n c i n g among them. The reverse transformations should be c o n s i d e r e d i n precisely t h e reverse o r d e r fro11 t h a t of the c o r r e s p o n d i n g f o r w a r d t r a n s f o r m a t i o n s . The sequericihg is a g a i n cyclic, with each i t e r a t i o n now c r e a t i n g an embedded sentence.Even i f a reverse t r a n s f o r m a t i o n matches t h e s e n t e n c e being decomposed, one cannot be s u r e t h a t the corresponding forward transformation was involved in the generation of the sentence. Uridoing the t r a n s f o r m a t i o n may lead t o a dead end [ P e t r i c k 1965 ] S t q l e y R e P e t r i c k , A R e c o g n i t i o n Proceddre f o r T r a n s formational Grammars. Doctoral Disseztatf on .[ p e t r i c k 19661 Stanley R. F e t r i c k , A i r o g r a m for Transformational S y n t a c t i c Analysis, Air Force Cambridge Research Laboratories, AFCRL--66-698.[ P e t r i c k 19731 S t a n l e y R . P e t r i c k , T r a n s f o r m a t i o n a l Analysis, I I I n N a t u r a l Language Processing, ed. R. R u s t i n , Algorithmics P r e s s , N. Y.[ P e t r i c k 19 75 I Stanley R . P e t r i c k , "Design of the Underlying S t r u c t u r e for a D a t a Base Retrieval. S y s t e m . " In D i r e c t i o n s in A r t i f i c i a l I n t e l l i g e n c e : Natural Language P r o c e s s i n g , ed. R. Gsishman C o u r a n t 9 7 2 , 19731.The augmented t r a n s i t i o n network, and i n p a r t icular the formalism developed bv Woods, has proven t a be ant h e s u r f a c e a n a l y s i s .T h i s w o l l d s e e m to be d i s a d v a n t a g e o u s from the p o i n t of view of e f f i c i c n c y , s i n c e e r r o n e o u s p a r s e s which m i g h t be aborted a t the ' b e g i n n i n g of t h e s u x f a c e a n a l y s i s m u s t be followed t h r o u g h t h e e n t i r e s u r f a c e a n a l y s i s an d p a r t of the transformational d e c o n p s s i t i o n . Second, the t r a n s format i o n s are n o t a s s o c i a t e d w i . m p a r t i c u l a r productions of the s u r f a c e grarxnar, b u t r a t h e r w i t h p a r t i c u l a r p a t t e r n s in t h e tree ( " s t r u c t u r a l descriptions") , so p a t t e r n match; n g o p e r at i o n s a r e r e q u i r e d to d e t e r m i n e 1;rhich t r a n s f o r m a t i o n s t o a p p l y . These d i f f e r e n c e s r e f l e c t P e t r i c k ' s c'iesire to remain as close as is p r a c t i c a l to the f o r m a l i s m of t r a n s f o r m a t i o n a l l i n g u i s t i c s .T h e primary d i s t i n c t i o n o f t h e Woods system is that the deep structure tree is built during the surface analysis. Consequently, his " t r a n s f o r m a t i . o n a 1 " procedures c o n s i s t of t r e e b u i l d i n g r a t h e r t h a n tree t r a n s f o r m i n g o p e r a t i o n s . The t r a d e o f f s between this approach and the two-stage a n a l y z e r s of p e t r i c kt h e augmented base component i s c a l l e d a c o d a 1 9 i n g gramrzar. S i n c e each s p u r i o u s s u r f a c e a n a l y s i s will have to undergo a len'gthy reverse t r a~s f o r m a t i o n a l process before i t !S r e c o g n i z e d as i n v a l i d , i t i s important t o minimize the number of such parses. The s e r i o u s n e s s of t h i s problem is i n d i c a t e d -, A N 7 , 1473 E D I T t O N O F 1 NOV 6 5 I S OBSOLETE UNCLASSIFIEDS E C U R I T Y C L A S S I F I C A T I O N 3 F THIS P A G E (WhenData Entered,'The s y s t e m s w e s h a l l be d e s c r i b i n g are a l l m o t i v a t e d by particular applications r e q u i r i n g natural language input, rather t h a n by p u r e l y l i n g u i s t i c considerations. C o n s e q u e n t l y , t h e pjarsing of a t e x t (determining i t s s t r u c t u r e ) w -i l l be viewed as ari essential s t e p p r e l i m i n a r y to processing t h e i n f o r m a t i o n i n t h e text, r a t h e r than as an end in i t s e l f .T h e r e are a wide variety of a p p l i c a t i o n s i n v o l v i n g natural l a n g u a g e i n p u t , such as machine t r a n s l a t i o n , i n f o r m a t i o n r e t r i e v a l , q u e s t i o n answering, conunand s y s t e m s , and d a t a collection. I t may t h e r e f o r e s e e m a t f i r s t that there w o u l d be l i t t l e t e x t processing which w a id' be generally u s e f u , i beyond the determination of a s t r u c t u r j a l d e s c r i p t i o n (e. g . a par'se tree) f o r each sentence. There are, however, a numbet of o p e r a t i o n s whfch can r e q u l a r i z e s e n t e n c e s t r u c t u r e , and thereby s i m p l i f y -t h e subsequen vA a p p l ic a t i o n -s p e c i f i c p r o c e s s i n g . For example, Some m a t e r i a l i n sentences (enclosed i n brackets i n t h e ekamples below) can be o m i t t e d or " z e r o e d " :John a t e cake and Mary [ a t e ] cookies.. . t h e r e i s h a r d l y a field of science or e n q i n e e r i n g which i s clearly deIineaked from i t s n e i g h b o r s .The last few years have seen most work in l a n g u a g e processing devoted to t h e development of i n t e g r a t e d s y s terns, combining syntactic, semantic, pragmatic, and generative components. T h i s was a h e a l t h y a n d p~d i c t h l e r e a c t i o n t o t h e e a r l i e r r e s e a r c h , which had l a r g e l y approached s y n t a c t i c p r o c e s s i n g i n i s o l a t i o n froin these other a r e a s . P t produced some systcms whose modest successes d i s p e l l e d t h e s k e p t i c i s m t h a t n a t u r a l l a n g u a g e proccssors would ever be a b l e todo a n y t h i n g . These systcms i n d i c a t e d how s y n t a c t i c , semantic, and praglnn t i c i n f o r m L~t ion I I I U S~ i n t e r a c t to s e l e c t t h e c o r r e c t sentence a n a l y s i s .It is now generally understood t h a t s y n t a c t i c p r o c e s s i n g by itself i s i n a d e q u a t e to select t h e i n t e n d e d a n a l y s i s of a s e n t e n c e . e s h o u l d n o t conclude from t h i s , however, that i t is impossible to s t u d y the processes of s y n t a x a n a l y s i s s e p a r a t e l y from the o t h e r components. Rather, i t means t h a t s y n t a x a n a l y s i s m u s t be s t u d i e d w i t h an u n d a r s t a n d i n g of i t s r o l e i n a l a r g e r system and t h e i n f o -r m a t i o n i t s h o u l d be able to c a l l upcjn from o t h e r components ( i e . , t h e p r o c e s s i n g which t h e s u b s e q u e n t con~ponen t s must do to select among t h e a n a l y s e s produced by t h e syntactic component) .While r e c o g n i z i n g the i m p o r t a~~c e of t o t a l systems i n i n s u r i n g t h a t none of t h e problems h a s f a l l e n i n t h e gaps between s t a g e s and been f o r g o t t e n , it s t i l l seers t h a t more s p e c i a l i z e d research projects are e s s e n t i a l if t h e field-of n a t u r a l l a l~y u a g e process i n g is to mature. The development of a n o t h e r t o t a l system will n o t advance the f i e l d unless it endeavors t o p e r f o r m some p a r t i c u l a r p r o c e s s i n g task b e t t e r t h a n i t s predecessors; t h e Some researchers have a s s e r t e d recently that h a t u r a l language p r o c e s s i n g can be done w i t h o u t s y n t a x a n a l y s i s . I t seems to US that s u c h c l a i m s are e x a g g e r a t e d , b u t t h e y do arise o u t of some o b s e r v a t i o n s t h a t are n o t w i t h o u t validity:(1) For the r e k i t i v e l y s i m p l e , s e n t e n c e s whose s l e m a n t i c s i s w i t h i n t h e scope of current a r t i f i c i a l d n t~l l i g e n c e s y s t e m s , s o p h i s t i c a t e d s y n t a c t i c p r o c e s s i n g i s u n n e c e s s a r y .T h i s Was c e r t a i n l y t r u e of some e a r l y q u e s t i p n -a n s w e r i n g s y s t e q s , I n any case, i t i s h a r d t o imagine how s e n t e n c e s of t h e complexity t y p i c a l i n technical w r i t i n g c~u l d be u n d e t s t o o d w i t h o u t u t i l i z i n g s y n t a c t i c ( a s w e l l as s e m a n t i c ) r e s t r i c t i o $ s t o s e l e c t t h e correct a n a l y s i s .( 2 ) S y n t a c t i c analysis may appear in g u i s k s o t h e r t h a n the t r a d i t i o n a l p a r s i n g p r o c e d u r e s ; i t can b e i n t e r w o v e n with o t h e r components of the system and cqh be embedded into t h e a n a l y s i s programs t h e m s e l v e s . T h i s w i l l often i n c r e a s e the p a r s i n g speed c o n s i d e x a b l y .The "grammar i n program" approach which c h a r a c t e r i z e d many of the early machine t r a n s l a t i o n efforts i s s t i l l employed i n some (3) Syntax analvsis can be d r i v e n by s e m a n t i c a n a l y s i s ( i n s t e a dof b e i n g a separate, e a r l i e r s t a g e ) , and, i n p a r t i c u l a r , can be done by l o g k i n g f o r s e m a n t i c p a t t e r n s i n t h e s e n t e n c e . Syntax a n a l y s i s i s done s e p a r a t e l y because there a r e r u l e s of sentence formation and %ransformation which can be s t a t e d in terms of t h e relatively broad s y n t a c t i c c a t e g o r i e s ( t e n s e d verb, c o u n t n o u n , e t c . ) . If t h e semantic classes are subcategoriz a t i o n s of t h e syntactic o n e s t h e n clearly t h e tr2i.p~ f o r~n n t i o n s c o u l d be s t a t e d i n terms of sequepces o f s e m a n t i c c l a s s e s . For those t r n n s f o r~n a t i o n s w h i c h are p r o p e r l y syntactic, however, w e would find that s e v e r a l t r a n s f o r m a t i o n s a t the s e m a n t i c s t a g e w o u l d be r e q u i r e d i n p l a c e of one a t the syntactic s t a g e ; certain u s e f u l g e n e r a l i z a t i o n s would bc l a s t .The strongest axgument of those advocating a s c m~n t i s s -d r i . \ . c n s y n t a x i s the a b i l i t y of p e o p l e to i n t e r p r e t s e n t c n c o s from s e m a n t i c c l u e s i n the f a c e of s y n t a c t i c errors or m i s s i n g i n f o r -11 mation ("I want t oxx t o the inovies tonight. ) . T h i s a r g u m e n t w o r k s both ways, however -p e o p l e can also u s e s y n t a c t i c r u l e s when s e m a n t i c s i s lackirlg; S o x e s a m p l e , t o understand the function of a word i n a sentence without k n o w i n g i t s meaning ("Isn't t h a t man wearing a very frimple c o a t ? " ) . U l t i m a t e l y , w e want an analyzer w h i c h c a n work from p a r t i a l information of e i t h e r k i n d , and r e s e a r c h i n t h a t d i r e c t i~n i s t o be welcomed ( s o n e work o n p a r s i n g i n t h e f a c e of u n c e r t a i n t y has been done by s p e e c h -u n d e rs t a n d i n g g r o u p s ) . A t the same time, s i n c e s c z c e s s f u l p r o c e s s i n g of "perfect" s e n t e n c e s i s p r e s u m a b l y a p r e r e q u i s i t e f o r p r o c e s s i n g i m p e r f e c t sentences, i t seems r e a s o n a b l e t o c o n t i n u e d e v o t i n g s u b s t a n t i a l effort t o t h e rrmsibrable pl-oblems which r e m a i n in a n a l y z i n g p e r f e c t s e n t e n c e s .C o m p u t a t i o n a J and T h e o r e t i c a l L i n g u i s t i c s In p a r t i 6 u l a r , t h e y a r e concerned trri th language u n i v e r s a l s -p r i n c i p l e s o f grammar which a p p l y t o a l l n a t u r a l l a n g u a g e s .Computational l i n g u i s t s , i n contrast, a r e usually d e l i g h t e d i f t h e y can manage t o h a n d l e one language (two, i f t h e y ' r e t r a n s l a t i n g ) . Their primary c o n c e r n lies i h t r a n s f o r m i n g sentences -often assumed t o be gramrnatioal -i n t o a form acceptable t o some p a r t i c u l a r a p p l i c a t i o n s y s tern. They are concerned w i t h t h e efficiency of such processing, whereas t h e o r e t i c a l linguists g e n e r a l l y don ' t worry a b o u t t h e r e c o g n i t i o n problem at a l l .I.Ionetheless, t h e two s p e c i a l t i e s should have many common areas of interest. Q u e s t i o n s of g r a m m a t i c a l i t yare i m p o r t a n t , b e c a u s e e x u e r i e n c e has shown t h a t a g r a m m a t i c a l c o n s t r a i n t which+ one c a s e d e t e r m i n e s i f h sentence i s o r i s n o t a c c e p t a b l e w i l l i n other cases be needed t o choose between correct and i n c o r r e c t a n a l y s e s o f a sentence. The r e l a t i o n s between s e t s of s e n t e n c e s , which are a prime f o c u s of t r a n s f o rm a t i o n a l grammar, p a r t i c u l a r l y i n t h e H a r r i s i a n f famework, a r e c r u c i a l t o t h e success of s y n t a c t i c a n a l y s i s p r o c e d u r e s , s i n c e t h e y enable a l a r g e v a r i e t y of s e n t e n c e s t o be r e d u c e d t o a r e l a t i v e l y s m a l l number o f s t r u c t u ' r e s .More g e n e r a l l y , both s p e c i a l t i e s seek t o understand a p a r t i c u l a r mode'of communication. T r a d i t i o n a l l i n g u i s t s axe interested i n a mode whioh has e v o l v e d as an efficient means of communicating ideas between people; u l t i m a t e l y , w e may hope that t h e y will u n d e r s t a n d n o t o n l y the principles of language structure, b u t also some of tlle reasons why language has developed i n this way. Computational l i n g u i s t s , in s t u d y i n g how language c a n be wed for man-machine communication, are r e a l l y a s k i n g much the same questions. They want t o develop a mode of communication for which people are n a t u r a l l y s u i t e d and t h e y want t o understand the p r i n c i p l e s f o r d e s i g n i n g languages which are e f f i c i e n t for c o m n~m i c a t i n g i d e a s .W e can impose s e v e r a l rough groupings on t h e s e t of parsers in order t o structure t h e f o l l a w i n g survey. T o b e g i n withr we may try to separate those s y s tcms davclobed w i t h solme r e f e r e n c e to t r a n s f o r m a t i o n a l theory from the n o n t -r a n s f o r l n a t i o n a l s y s terns. T h i s t u r n s o u t s l s o to be an approximate h i s t o r i c a l d i + v i s i o n , s i n c e most s y s t e m s written since 1 9 6 5 have made soir~c connection w i t h transformational theory, even though their ihcthods o f a n a l y s i s nmy ke only d i % t z a n t l y r e l a t e d t o t r a n s f o l -m a t i o n a l mc ch an i s IIIS .Z'hc t r~~n s f o r m a t i a n a l s y s t c m s may i n t u r n be d i v i d e d intoi n part a r e s u l t of o u r i n a d e q u a t e t h e o r e t i c a l understanding of t r a n ' s f o r m a t i o n a l grammars, and may b e reduced by some r e c e n t t h e o r e tical work on t r a n s f o r m a t i o n a l grammar's.E a r l y Systems : C o n t e x t L F r e e . and C o n t e x t -S e n s i t i v e Parsers The p r c t r a n s f o r m a t i o n a l systems, developed mostly between 1 9 5 9 a n d 1 9 6 5 , were, w i t h a few e s c c p t i o n s , p a r s e r s f o r c o n t e x tf r e e l a n g u a g e s , a l t h o u g h c l o a k e d in a n u r~b e r of d i f f e r e n t g u i s e s .These s y s terns were based on immediate c o n s t i t u e n t a n a l y s i s , The l a r g e s t a n d probably the most i m p o r t a n t o f t h e s e early projects was t h e H a r v a r d P r e d i c t i v e Analyzer [Runo 1 9 6 2 1 . A predictive a n a l y z e r i s a top-down p a r s e r for c o n t e s t -f r e e g r a m m a r s written in Greibach normal form; t h i s formulation of t h e grammar w a s adopted from e a r l i e r work by I d a Rhodes for h e r R u s s i a n -E n g l i s h t r a n s l a t i o n p r o j e c t .The size of t h e grammar was staggering: a 1 9 6 3 r e p o r t [Kuno 19631 q u o t e s f i g u r e s of 1 3 3 Uord c l a s s e s 'and about 2100 p r o d u c t i o n s . Even w i t h a grammar of this size, the s y s t e m did not i n c o r p o r a t e simple agreement restrictions o f E n g l i s h syntax Since t h e program was designed t o produce p a r s e s f o r s e n t e n c e s which were presumed to be grammatical ( a n d n o t t o d i f f e r e n t i a t e between grammatical, and nongrammatica1 septences) , it was a t first hoped that i t could operate without these r e s t r i ctions. I t was soon discoveredr however, t h a t these r e s t r i ctions number agreement would cause a l a r g e increase i n an a l r e a d y very l a r g e grammar, the Harvarq g r o u p chose i n s t e a d t o include a s p e c i a l mechanism i n the parsing program t o p e r f o r m a rudimentary check on number agreement. Thus the Harvard P r e d i ctive A n a l y z e r , though p r o b a b l y the most successful of t h e c o n t e x t 4 ree a n a l y z e r s , c l e a r l y i n d i c a t e d the inadequacy of a context-free f o r m u l a t i o n of natural languqge grammar.were required t o eliminate i n vThe ~a r v a r d P r e d i c t i v e Analyzer p a r s i n g a l g o r i t h m progressed through several stages. The f i r s t version of t h e p r e d i c t i v e a n a l y z e r produced only one a n a l y s i s of a s e n t e n c e . The n e x t version i n t r o d t x e d an automatic backup mechanism in order t o produce a l l a n a l y s e s of a sentence. This i s an e x p o n e n t i a l t i m e algorithm, h e n c e very slow f o r l o n g sentences; a 1 9 6 2 r e p o r t gives t y p i c a l times as 1 m i n u t e for an 18 word s e n t e n c e and 12 m i n u t e s for a 35 word sentence. A n improvement o f more t h a n an 0 -of magnitude was obtained i n t h e f i n a l v e r s i o n of t h e program by using a b i t matrix for a p a t h -e l i m i n a t i o n technique [Kuno 19651. When an attempt was made to match a nonterminal symbol t o t h e sentence b e g i n n i n g a t a p a r t i c u l a r word and no match was found; t h e c o r r e s p o n d i n g bit was turned on; if the same symbol came up a g a i n l a t e r i n the ~a r s i n g a t t h e same p o i n t i n the sentence, the program would n o t have t o try t o match it again.Another important e a r l y p a r s e r was t h e immediate c o n s t i t u e n t anal-yzer u s e d at RAND. U n i v e r s i t y o f P e n n s y l v a n i a [Harris 1 9 6 5 1 . T h i s proctxlure, called a cycling c a n c e l l i n g . automaton, makes n s t v i e s rule then returned a fail-ure s i g n a l t o the parser, i n d i c a t i n g that the analysis was semantically anomalous, and t h i s a n a l y s i s was aborted. W o o d s has noted [Woods 197Ql t h a t t h e p a r s e r used i n t h e D m m 4 pro jecf may produce redundant parses, and has g i v e n a parsing algorithm for c o n t e x t -s e n s i t i v e languages which repedies this deficiency .When the theory of transformational grammar was elaborated in the early l . 9 6 0 1 s there was c o n s i d e r a b l e i n t e r e s t in finding a corresponding recognition procedure. Because t h e grammar is stated i n a generative form, however, t h i s i s n o s i m p l e matter. A &hornsky) tree transformational grammar c o n s i s t s of a set of context-sensitive phrase s t r u c t u r e rules, which g e n e r a t e a set of base treesa, and a set of t r a n s f o r m a t i o n s , which a c t on base trees to produce the s u r f ace trees. A (Harris) s t r i n g t r a n s f o~t i o n a l grammar consists of a finite set of sequences of word categories, c a l l e d kernel s e n t e n c e s , and a set of transformations which combine and modify t h e s e kernel s e n t e n c e s t6 e the other sentences of t h e language. There are a t least three b a s i c problems i n reversing the g e n e r a t i v e process:(1) for a tree t r a n s f o r m a t i o n a l grammar, a s s i g n i n g t o a given sentence a set of p a r s e trees which includes a l l the s u r f a c e trees which would be assigned by the trans f o r m a t i o n a l grammar( 2 ) giveq a t r e e n o t i n t h e b a s e , d e t e r m i n i n g w h i c h sequences of transf o r m a t i g n s might have applied t o generate this t r e e( 3 ) h a v i n g d e c i d e d on a transformation whose r e s u l t may be the p r e s e n t t r e e , undoing this t r a n s f o r m a t i o n If we a t t a c k each of these p r o b l e m s i n t h e most s t r a i g h t f o r w a r d manner, w e are likely t o try many false p a t h s w h i c h w i l l not l e a d t o an analysis. F o r the f i r s t problem, w e could u s e a c o n t e x t -£ ree gramrnar which will g i v e all t h e s u r f a c e t r e e s assigned by the t r a n s f o r m a t i o n a l granunar, a n d probably l o t s more The s u p e r a b u n d a n c e of ''false" s u r f a c e trees is a g g r a v a t e d by t h e f a c t that most E~~g l i s h words have more t h a n one word category ( p l a y more t h a n o n e s y n t a c t i c role), a l t h o u g h n o r m a l l y o n l y o n e is u s e d in any g i v e n sen'tcnce. (1) a n a l y s i s of t h e s e n t e n c e u s i n g t h e c o n t e x t -f r e e c o v e r i n g grammar (with a bottom-up p a r s e r )( 2 ) a p p l i c a t i o n of t h e revexse trans r o r r n a t i o n a l rules 'i3) for each candidate base tree produced by steps (1) and ( 2 ) , a check whether it can i n f a c t be generated by the base component( 4 ) f o r each base tree and sequence of transformations which passes the t e s t in s t e p ( 3 ) , the (forward) t r a n s - The covering grammar produced a large number of spurious surface analyses which the parser must process. The 1 9 6 5 r e p o r t f o r example, cites a 1 2 word sentence which produced 4 8 parses with t h e covering grammar; each must be followed t h r o u g h s t e p s ( 2 ) and ( 3 ) before most can be e l i m i n a t e d . The system was t h e r e f o r e very s l o w ; 36 minutes were r e q u i r e d to a n a l y z e one 11 word s e n t e n c e . r u l e s had a s i g n i f i c : a n t e f f e c t on parsing t i m e sthe 11 word 1 s e n t e n c e w h i c h t o o k 36 m i n u t e s hefore now took o n l y 6The system developed by P e t r i c k [ P e t r i c k 1965, 1966; K e y s e r 19671 is similar in o u t l i n e : a p p l y i n g a s e r i e s oE reverse t r a n s f o r m a t i o n s , c h e c k i n g if the r e s u l t i n g tree can be generated by the base component, and the: verifying the a n a l y s i s by applying the forward t r a n s f o~m a t i o n s ta the base The price f o r g e n e r a l i t y was p a i d in e f f i c i e n c y . Petrick's problems w e r e more severe t h a n M I T R Z ' s f o r t w o reasons. ~i r s t , the ~b s e n c e of a s e n t e n c e tree during t h e application of the reverse t r a n s formational r u l e s meant t h a t many sequences of re-rse transformations were t r i e d which did n o t correspond to any sequence of tree transformations and hence would eventually be rejected. Second, i f several rever se transformations Could apply a t some point in the a n a l y s i s , the procedure c o u l d not tell i n advance which would l e a d t o a v a l i d deep s t r u c t u r e . C o n s e q u e n t l y , each one had to be t r i e d and the r e s u l t i n g s t r u cture followed t o a d e e p structure of a "dead end" (where no more transformations a p p l y ) . T h i s produces a growth i n the number of a n a l y s i s p a t h s which is exponential in the number of r e v e r s e transformations applied. This explssion ntn be avoided o n l y if t h e reverse transformations include t e s t s of the c u r r e n t a n a l y s i s t r e e to d6terrnine which transformations a p p l i e d to generate .this tree.Such t e s t s were included in the manually p r e p a r e d reverse t r a n sformations of t h e MITRE g r o u p , b u t i t would have b e e n fax t o o complicated f o r Pe trick t o produce s u c h tests automatlcally when i n v e r t i n g the trans formations. Potrick's system has been s i g n i f i c a n t l y revised over the past decade [ P e t r i c k 1 9 7 3 , Plath 1974aI. The r e s u l t i n g system is fast enough to be used i n an i n f o r m a t i o n r e t r i e v a l system with a grammar of moderate s i z e ; most r e q u e s t s are processed in less than one minute.Transformational A n a l y z e r s : Subsequent Developmi!ints One result of the early t r a n s f o r r n a t i~n a l s y s t e m s w a s a r e c o g n i t i o n of t h e importance sf f i n d i n g an e f f i c i e n t p a r s i n g procedure i f traiisformationa.l a n a l y s i s was ever t o be a u s e f u l t e c h n i~u e . As the systems i n d i c a t e d , there are two main obstacles to an e f f i c i e n t procedure. First, there is the problem of r e f i n i ' n g the s u r f a c e analysis, s o Chat each s e n t e n c e p r o d u c e s fewer troes f o r which transformational decomposition must be a t t e n~p t e d . This h a s g e n e r a l l y been approached by u s i n g a inore The first s y s t e m u s i n g such a n e t w o r k was developed by Thorne, B r a t l e y , and D e w a r a t E d i n b u r g h [Thorne 1968, D e w a r 1 9 6 9 1 . They s t a r t e d w i t h a regular base grammar, i -e . , a transition network. The i m p o r t a k e of using a r e g u l a r base l i e s i n their claim that some transformations are equAvalent i n e f f e c t to changing the base t o a recursive transition network. Transform a t i o n s which could n o t be handled i n this fasion, such as conjunction, were incorporated in^^ the p a r s i n g program. P a r s i n g a sentence with t h i s surface grammar s The recutsive t r a n s i t i o n network was developed i n t o an augmented recursive t r a n s i t i o n network grammar i n the system of Bobrow and Frasex m w 9An ausmented network i s one in which an a r b i t r a r y predicate, w r i t t e n i n some g e n e r a l purpose language ( i n this case, L I S P ) . may be associated u i t h each a r c i n the network. A t r a n s i t i o n i n the n e t w~r k is n o t allowed if t h e predicate associated w i t h t h e arc f a i l s . These predicates perform t w o functions i n t h e grammar. F i r s t , they are used t o i n c o r p o r a t e r e s t r i c t i o n s i n the language which would be difficult or impossible to s t a t e w i t h i n the a m t e x t -f r e e mechanisms of t h e recursive n e t w o r k , e. g. , agreement r e s t r i c t i o n s .sentence is being parsed.The augmented t r a n s i t i o n network was further developed by Woods a t B o l t B e r a n e k and Newman.In order to r e g u l a r i z e the predicates, he i n t r o d u c e d a standard set of operations for b u i l d i n g and t e s t i n g the deep structure lWoods 1970bl. He considerably e n l a~g e d t h e scope of the grammar and added a semantic component f b r t r a n s l a t i n g t h e deep structure into information r e t r i e v a l commands. With that, from a n a n a l y s i s of the sentence into linguistic s t r i n g s , one could directly d e t e r m i n e the transformations w h i c h acted to produce t h e s e n t e n c e , w i t h o u t h a v i n g to t r y many s e q u e n c e s sf reverse transformations. T h e i r proposed sys tern therefore c o n s i s t e d of a procedure f o r l i n g u i s t i c string analysis ( a context-free p a r s i n g problem at the level ,of s i m p l i f i c a t i o n of t h e i r o r i g i n a l proposal) and a s e t of r u l e s w h i c h constructed from each s t r i n g a corresponding k e r n e l -l i k e sentence.T h e i r o r i g i n a l proposal was a s i m p l i f i e d scheme which accounted for only a l i m i t e d s e t of trans.formations. not specifically oriented towards r e c o g n i t i o n , to d k t e r r n i n e the features of a sentence which i n d i c a t e t h a t a particu1a.c t r a n s f o rmation a p p l i e d i n generating i t , and hence t o produce an e f f icient analysis procedure*Another group which has used l i n g u i s t i c s t r i n g a n a l y s i s is t h e Linguistic S t r i n q Project a t New York University, led by Sager [Sager 1967 [Sager , 1973 Grishman 1973a, 1973131. T h e i r s y s t e m , which has gone through s s v e r a l versions since 1965, is based on a contextfree grammar augmented w i t h r e s t r i c t i o n s . Because they were conce ned with processing s c i e n t i f i c text, rather t h a n commands or queries, t h e y were l e d to develop a grammar of particularly broad coverage. The p r e s e n t gramrr~ar has about 250 context-free rules and about 2 0 0 r e s t r i c k i o n s ; although not as swift as some of the smaller s y s t e m s , t h e parser is able to analyze most sentences in less than one minute. Because of the l a r g e size of their grammar, t h i s group *has been p a r t i c u l a r l y concerned w i t h techniques for organizing and s p e c i f y i n g the grammar which w i l l f a c i l i t a t e f u r t h e r development. In p a r t i c u l a r , the most recent implementation of t h e i r s y s t e m has added a s p e c i a l language designed for t h e economical and perspicuous s t a e m e n t of the r e s t r i c t i o n s [Sager 19751.One of t h e e a r l i e r versions of t h i s s y s t e m , w i t h a much more restricted grammar, was used as the f r o n t end for an information r e t r i e v a l s y s t e m d e v e l o p e d by C a u t i n at t h e U n i v e r s i t y of P e n n s y l v a n i a [ C a u t i n 1369 ] .The ~i n g u i s t i c S t r i n g P r o j e c t s y s t e m has r e c e n t l y been e x t e n d e d to include a transformational decomposition p h a s e ; t h i s phase follows t h e l i n g u i s t i c ? s t r i n g a n a l y s i s [IIobbs 1 9 7 5 1 . should semantic a n a l y s i s be done c o n c u r r e n t l y with Syntactic a n a l y s i s . P a r a l l e l processing is p r e f e r r e d if t h e added time r e q u i r e d by the deeper a n a l y s i s i s outweighed by t h e f r a c t i o n of i n c o r r e c t analyses which can be e l i m i n a t e d early i n the parsing erocess. In the case of s ?mantic analysis, it clearly depends on t h e r e l a t i v e complexity of t h e s y n t a c t i c and semantic components. I n t h e case of transformational a n a l y s i s , it depends on the f r a c t i o n of grammatical and s e l e c t i~n a l c o n s t r a i n t s which can be e x p r e s s e d at: t h e surface l e v e l (if most of these can only be realized through transformational allalysis , concurrent trans formational a n a l y s i s i s probably more e f f i c i e n t ) . This may depend i n Lurn on the t y p e of surface a n a l y s i s ; for example, the r e l a t i o n s h i p s exhibited by l i n g u i s t i c string analysis axe s u i t a b l e for expressing many of these c o n s t r a i n t s , so there i s less motivation in the Linguistic S t r i n g Colmerauer and de Chastellier [de C h a s t e l l i e r 196 9 ] have also investigated the possibility of using Wijngaarden grammars (as were developed f o r s p e c i f y i n g ALGOL 6 8 ) f o r :tr.ansformational decomposition and machine t r a n s l a t i o n . L i k e u n r e s t r i c t e d r e w r i t i n g rules, W-grammars can d e f i n e every r e c u r s i v e l y enumerable l a q u a g e , and s o can perform t h e functions of t h e syrface and reverse transformational components. They show how p o r t i o n s of transformational grammars of English and French may be rewritten as W-grammarsr w i t h t h e pseudo-rules in t h e W-grammar t a k i n g the place of the t r a n s f o l ' m a t i o n s~In all t h e systems dqscribed above, a s h a r p line was drawn between correct and incomect parses: a t e m i n a l node e i t h e r did or d i d n o t m a t c h t h e n e x t word in t h e s e n t e n c e : an a n a l y s i s of a phrase was e i t h e r acceptable o r unacceptable. There are circumstances under which we would want to r e l a x these requirements. For one thing, i n analyzing connected speech, We segmentation and i d e n t i f i c a t i o n of words can never be done with complete c e r t a i n t y . A t b e s t , one can say that a c e r t a i n sound has some p r o b a b i l i t y of being one phoneme and son= o t h e r p r o b a b i J i ty of being another phoneme; some e x p e c t e d phonemes may be l o s t e n t i r e l y in the sound received. c o n s e q u e n t l y , one w i l l a s s o c i a t e some n u f i e r w i t h each t e r m i n a l node, i n d ic a t i n g t h e p r o b a b i l i t y or q u a l i t y of match; noflcrminal nodes will be a s s i g n e d some v a l u e based on t h e v a l u e s of t h e t e r m i n a l nodes ) x n e a t h . Another circuinstance a r i s e s in n a t u r a l l a n g u p a r s e t r e e u n t i l i t g e t s s t u c k (generates a t e r m i n a l n o d e which does n o t match the next s e n t e n c e word) ; it t h e n "backs up" to t r y another a l t e r n a t i v e .IleL A measure is a s s o c i a t e d with each a l t e r n a t i v e path, fadicating the l i k e l i h o o d t h a t this a n a l y s i s matches the sentence p w s s e d so far and t h a t it can be extended to a let@ s@xtence analysis. At each moment, t h e path w i t h the higbsst likelihood is extended; if its measure f a l l s below that otner path, the parser s h i f t s its a t t e n t i o n to t h a t allowed patterns o f occwl-rence of con j o i n i n g s i n a sentence are q u i t e r e g u l a r . Loosely speaking, a s e q u e n c e of e nts in the s e n t e n c e tree may be followed by a con j u n c t i o n aad by same or all of the elements immediately preceding the jrmction. For example, allowed p a t t e r n s of con joining subject-verb-ob ject-and-sub ject-verb-ob ject ( I drank d w and nary ate cake. ) , sub ject-verb-ob ject-and-verb-object a milk and a t e cake. ) and sub ject-verb-ob ject-and-ob ject For i n s t a n c e , i n t h e s e n t e n c e ) "I . T h i s may be done either during t h e s y n t a c t i c a n a l y s i s [woods 1 9 7 3 , Simmons 1 9 7 5 1 o r after the syntax phase is complete [~o r g i d a 1 9 7 5 , Hobbs 19751. C o n t e x t -f r e e grammars p l a y e d a major role i the e a r l y s t a g e s of automatic n a t u r a l l a n g u a g e a n a l y s i s . A1 t h o u g h t h e y have now g e n e r a l l y been superceded by more c o n~p l e x and powerful grammars, many of these grammars are based o n or have as one of t h e i r components a context-f ree grammar. The s e l e c t i o n of an e f f i c i e n t c o n t e x t -f r e e p a r s e r t h e r e f ore r e m a i n s an i~n p o r t a n t considerat ion i n n a t u r a l Language analysis. T h e o t h e r p a r s e r s g r a b a l l y accumulate d a t a from which a l l parses of a s e n t e n c e can be e x t r a c t e d ; t y p e s 1, 2 and 3 s t o r e t h i s data i n decreasingly compact r e p r e s e n t a t i o n s . | null | null | null | I t a l s o c o n t a ' i n s p r e c i s e . s p e c i f i c a t i o n s i n t h e programming l a n g u a g e S E T L o f a n u m b e r o f p a r s i n g a l g o r i t h m s , i n c l u d i n q s e v e r a l c o n t e x t -f r e e p a r s e r s , a u n r e s t r i c t e d r e w r i t i n g r u l e P a r s e r , a n d a transformational p a r s e r . ......Parsing A l g o r i t h m s for C o n t e x t -F r e e Grammars. | Main paper:
the role of s y n t a c t i c a n a l y s i s:
The s y s t e m s w e s h a l l be d e s c r i b i n g are a l l m o t i v a t e d by particular applications r e q u i r i n g natural language input, rather t h a n by p u r e l y l i n g u i s t i c considerations. C o n s e q u e n t l y , t h e pjarsing of a t e x t (determining i t s s t r u c t u r e ) w -i l l be viewed as ari essential s t e p p r e l i m i n a r y to processing t h e i n f o r m a t i o n i n t h e text, r a t h e r than as an end in i t s e l f .T h e r e are a wide variety of a p p l i c a t i o n s i n v o l v i n g natural l a n g u a g e i n p u t , such as machine t r a n s l a t i o n , i n f o r m a t i o n r e t r i e v a l , q u e s t i o n answering, conunand s y s t e m s , and d a t a collection. I t may t h e r e f o r e s e e m a t f i r s t that there w o u l d be l i t t l e t e x t processing which w a id' be generally u s e f u , i beyond the determination of a s t r u c t u r j a l d e s c r i p t i o n (e. g . a par'se tree) f o r each sentence. There are, however, a numbet of o p e r a t i o n s whfch can r e q u l a r i z e s e n t e n c e s t r u c t u r e , and thereby s i m p l i f y -t h e subsequen vA a p p l ic a t i o n -s p e c i f i c p r o c e s s i n g . For example, Some m a t e r i a l i n sentences (enclosed i n brackets i n t h e ekamples below) can be o m i t t e d or " z e r o e d " :John a t e cake and Mary [ a t e ] cookies.. . t h e r e i s h a r d l y a field of science or e n q i n e e r i n g which i s clearly deIineaked from i t s n e i g h b o r s .The last few years have seen most work in l a n g u a g e processing devoted to t h e development of i n t e g r a t e d s y s terns, combining syntactic, semantic, pragmatic, and generative components. T h i s was a h e a l t h y a n d p~d i c t h l e r e a c t i o n t o t h e e a r l i e r r e s e a r c h , which had l a r g e l y approached s y n t a c t i c p r o c e s s i n g i n i s o l a t i o n froin these other a r e a s . P t produced some systcms whose modest successes d i s p e l l e d t h e s k e p t i c i s m t h a t n a t u r a l l a n g u a g e proccssors would ever be a b l e todo a n y t h i n g . These systcms i n d i c a t e d how s y n t a c t i c , semantic, and praglnn t i c i n f o r m L~t ion I I I U S~ i n t e r a c t to s e l e c t t h e c o r r e c t sentence a n a l y s i s .It is now generally understood t h a t s y n t a c t i c p r o c e s s i n g by itself i s i n a d e q u a t e to select t h e i n t e n d e d a n a l y s i s of a s e n t e n c e . e s h o u l d n o t conclude from t h i s , however, that i t is impossible to s t u d y the processes of s y n t a x a n a l y s i s s e p a r a t e l y from the o t h e r components. Rather, i t means t h a t s y n t a x a n a l y s i s m u s t be s t u d i e d w i t h an u n d a r s t a n d i n g of i t s r o l e i n a l a r g e r system and t h e i n f o -r m a t i o n i t s h o u l d be able to c a l l upcjn from o t h e r components ( i e . , t h e p r o c e s s i n g which t h e s u b s e q u e n t con~ponen t s must do to select among t h e a n a l y s e s produced by t h e syntactic component) .While r e c o g n i z i n g the i m p o r t a~~c e of t o t a l systems i n i n s u r i n g t h a t none of t h e problems h a s f a l l e n i n t h e gaps between s t a g e s and been f o r g o t t e n , it s t i l l seers t h a t more s p e c i a l i z e d research projects are e s s e n t i a l if t h e field-of n a t u r a l l a l~y u a g e process i n g is to mature. The development of a n o t h e r t o t a l system will n o t advance the f i e l d unless it endeavors t o p e r f o r m some p a r t i c u l a r p r o c e s s i n g task b e t t e r t h a n i t s predecessors; t h e Some researchers have a s s e r t e d recently that h a t u r a l language p r o c e s s i n g can be done w i t h o u t s y n t a x a n a l y s i s . I t seems to US that s u c h c l a i m s are e x a g g e r a t e d , b u t t h e y do arise o u t of some o b s e r v a t i o n s t h a t are n o t w i t h o u t validity:(1) For the r e k i t i v e l y s i m p l e , s e n t e n c e s whose s l e m a n t i c s i s w i t h i n t h e scope of current a r t i f i c i a l d n t~l l i g e n c e s y s t e m s , s o p h i s t i c a t e d s y n t a c t i c p r o c e s s i n g i s u n n e c e s s a r y .T h i s Was c e r t a i n l y t r u e of some e a r l y q u e s t i p n -a n s w e r i n g s y s t e q s , I n any case, i t i s h a r d t o imagine how s e n t e n c e s of t h e complexity t y p i c a l i n technical w r i t i n g c~u l d be u n d e t s t o o d w i t h o u t u t i l i z i n g s y n t a c t i c ( a s w e l l as s e m a n t i c ) r e s t r i c t i o $ s t o s e l e c t t h e correct a n a l y s i s .( 2 ) S y n t a c t i c analysis may appear in g u i s k s o t h e r t h a n the t r a d i t i o n a l p a r s i n g p r o c e d u r e s ; i t can b e i n t e r w o v e n with o t h e r components of the system and cqh be embedded into t h e a n a l y s i s programs t h e m s e l v e s . T h i s w i l l often i n c r e a s e the p a r s i n g speed c o n s i d e x a b l y .The "grammar i n program" approach which c h a r a c t e r i z e d many of the early machine t r a n s l a t i o n efforts i s s t i l l employed i n some (3) Syntax analvsis can be d r i v e n by s e m a n t i c a n a l y s i s ( i n s t e a dof b e i n g a separate, e a r l i e r s t a g e ) , and, i n p a r t i c u l a r , can be done by l o g k i n g f o r s e m a n t i c p a t t e r n s i n t h e s e n t e n c e . Syntax a n a l y s i s i s done s e p a r a t e l y because there a r e r u l e s of sentence formation and %ransformation which can be s t a t e d in terms of t h e relatively broad s y n t a c t i c c a t e g o r i e s ( t e n s e d verb, c o u n t n o u n , e t c . ) . If t h e semantic classes are subcategoriz a t i o n s of t h e syntactic o n e s t h e n clearly t h e tr2i.p~ f o r~n n t i o n s c o u l d be s t a t e d i n terms of sequepces o f s e m a n t i c c l a s s e s . For those t r n n s f o r~n a t i o n s w h i c h are p r o p e r l y syntactic, however, w e would find that s e v e r a l t r a n s f o r m a t i o n s a t the s e m a n t i c s t a g e w o u l d be r e q u i r e d i n p l a c e of one a t the syntactic s t a g e ; certain u s e f u l g e n e r a l i z a t i o n s would bc l a s t .The strongest axgument of those advocating a s c m~n t i s s -d r i . \ . c n s y n t a x i s the a b i l i t y of p e o p l e to i n t e r p r e t s e n t c n c o s from s e m a n t i c c l u e s i n the f a c e of s y n t a c t i c errors or m i s s i n g i n f o r -11 mation ("I want t oxx t o the inovies tonight. ) . T h i s a r g u m e n t w o r k s both ways, however -p e o p l e can also u s e s y n t a c t i c r u l e s when s e m a n t i c s i s lackirlg; S o x e s a m p l e , t o understand the function of a word i n a sentence without k n o w i n g i t s meaning ("Isn't t h a t man wearing a very frimple c o a t ? " ) . U l t i m a t e l y , w e want an analyzer w h i c h c a n work from p a r t i a l information of e i t h e r k i n d , and r e s e a r c h i n t h a t d i r e c t i~n i s t o be welcomed ( s o n e work o n p a r s i n g i n t h e f a c e of u n c e r t a i n t y has been done by s p e e c h -u n d e rs t a n d i n g g r o u p s ) . A t the same time, s i n c e s c z c e s s f u l p r o c e s s i n g of "perfect" s e n t e n c e s i s p r e s u m a b l y a p r e r e q u i s i t e f o r p r o c e s s i n g i m p e r f e c t sentences, i t seems r e a s o n a b l e t o c o n t i n u e d e v o t i n g s u b s t a n t i a l effort t o t h e rrmsibrable pl-oblems which r e m a i n in a n a l y z i n g p e r f e c t s e n t e n c e s .C o m p u t a t i o n a J and T h e o r e t i c a l L i n g u i s t i c s In p a r t i 6 u l a r , t h e y a r e concerned trri th language u n i v e r s a l s -p r i n c i p l e s o f grammar which a p p l y t o a l l n a t u r a l l a n g u a g e s .Computational l i n g u i s t s , i n contrast, a r e usually d e l i g h t e d i f t h e y can manage t o h a n d l e one language (two, i f t h e y ' r e t r a n s l a t i n g ) . Their primary c o n c e r n lies i h t r a n s f o r m i n g sentences -often assumed t o be gramrnatioal -i n t o a form acceptable t o some p a r t i c u l a r a p p l i c a t i o n s y s tern. They are concerned w i t h t h e efficiency of such processing, whereas t h e o r e t i c a l linguists g e n e r a l l y don ' t worry a b o u t t h e r e c o g n i t i o n problem at a l l .I.Ionetheless, t h e two s p e c i a l t i e s should have many common areas of interest. Q u e s t i o n s of g r a m m a t i c a l i t yare i m p o r t a n t , b e c a u s e e x u e r i e n c e has shown t h a t a g r a m m a t i c a l c o n s t r a i n t which+ one c a s e d e t e r m i n e s i f h sentence i s o r i s n o t a c c e p t a b l e w i l l i n other cases be needed t o choose between correct and i n c o r r e c t a n a l y s e s o f a sentence. The r e l a t i o n s between s e t s of s e n t e n c e s , which are a prime f o c u s of t r a n s f o rm a t i o n a l grammar, p a r t i c u l a r l y i n t h e H a r r i s i a n f famework, a r e c r u c i a l t o t h e success of s y n t a c t i c a n a l y s i s p r o c e d u r e s , s i n c e t h e y enable a l a r g e v a r i e t y of s e n t e n c e s t o be r e d u c e d t o a r e l a t i v e l y s m a l l number o f s t r u c t u ' r e s .More g e n e r a l l y , both s p e c i a l t i e s seek t o understand a p a r t i c u l a r mode'of communication. T r a d i t i o n a l l i n g u i s t s axe interested i n a mode whioh has e v o l v e d as an efficient means of communicating ideas between people; u l t i m a t e l y , w e may hope that t h e y will u n d e r s t a n d n o t o n l y the principles of language structure, b u t also some of tlle reasons why language has developed i n this way. Computational l i n g u i s t s , in s t u d y i n g how language c a n be wed for man-machine communication, are r e a l l y a s k i n g much the same questions. They want t o develop a mode of communication for which people are n a t u r a l l y s u i t e d and t h e y want t o understand the p r i n c i p l e s f o r d e s i g n i n g languages which are e f f i c i e n t for c o m n~m i c a t i n g i d e a s .W e can impose s e v e r a l rough groupings on t h e s e t of parsers in order t o structure t h e f o l l a w i n g survey. T o b e g i n withr we may try to separate those s y s tcms davclobed w i t h solme r e f e r e n c e to t r a n s f o r m a t i o n a l theory from the n o n t -r a n s f o r l n a t i o n a l s y s terns. T h i s t u r n s o u t s l s o to be an approximate h i s t o r i c a l d i + v i s i o n , s i n c e most s y s t e m s written since 1 9 6 5 have made soir~c connection w i t h transformational theory, even though their ihcthods o f a n a l y s i s nmy ke only d i % t z a n t l y r e l a t e d t o t r a n s f o l -m a t i o n a l mc ch an i s IIIS .Z'hc t r~~n s f o r m a t i a n a l s y s t c m s may i n t u r n be d i v i d e d intoi n part a r e s u l t of o u r i n a d e q u a t e t h e o r e t i c a l understanding of t r a n ' s f o r m a t i o n a l grammars, and may b e reduced by some r e c e n t t h e o r e tical work on t r a n s f o r m a t i o n a l grammar's.
. 1:
E a r l y Systems : C o n t e x t L F r e e . and C o n t e x t -S e n s i t i v e Parsers The p r c t r a n s f o r m a t i o n a l systems, developed mostly between 1 9 5 9 a n d 1 9 6 5 , were, w i t h a few e s c c p t i o n s , p a r s e r s f o r c o n t e x tf r e e l a n g u a g e s , a l t h o u g h c l o a k e d in a n u r~b e r of d i f f e r e n t g u i s e s .These s y s terns were based on immediate c o n s t i t u e n t a n a l y s i s , The l a r g e s t a n d probably the most i m p o r t a n t o f t h e s e early projects was t h e H a r v a r d P r e d i c t i v e Analyzer [Runo 1 9 6 2 1 . A predictive a n a l y z e r i s a top-down p a r s e r for c o n t e s t -f r e e g r a m m a r s written in Greibach normal form; t h i s formulation of t h e grammar w a s adopted from e a r l i e r work by I d a Rhodes for h e r R u s s i a n -E n g l i s h t r a n s l a t i o n p r o j e c t .The size of t h e grammar was staggering: a 1 9 6 3 r e p o r t [Kuno 19631 q u o t e s f i g u r e s of 1 3 3 Uord c l a s s e s 'and about 2100 p r o d u c t i o n s . Even w i t h a grammar of this size, the s y s t e m did not i n c o r p o r a t e simple agreement restrictions o f E n g l i s h syntax Since t h e program was designed t o produce p a r s e s f o r s e n t e n c e s which were presumed to be grammatical ( a n d n o t t o d i f f e r e n t i a t e between grammatical, and nongrammatica1 septences) , it was a t first hoped that i t could operate without these r e s t r i ctions. I t was soon discoveredr however, t h a t these r e s t r i ctions number agreement would cause a l a r g e increase i n an a l r e a d y very l a r g e grammar, the Harvarq g r o u p chose i n s t e a d t o include a s p e c i a l mechanism i n the parsing program t o p e r f o r m a rudimentary check on number agreement. Thus the Harvard P r e d i ctive A n a l y z e r , though p r o b a b l y the most successful of t h e c o n t e x t 4 ree a n a l y z e r s , c l e a r l y i n d i c a t e d the inadequacy of a context-free f o r m u l a t i o n of natural languqge grammar.were required t o eliminate i n vThe ~a r v a r d P r e d i c t i v e Analyzer p a r s i n g a l g o r i t h m progressed through several stages. The f i r s t version of t h e p r e d i c t i v e a n a l y z e r produced only one a n a l y s i s of a s e n t e n c e . The n e x t version i n t r o d t x e d an automatic backup mechanism in order t o produce a l l a n a l y s e s of a sentence. This i s an e x p o n e n t i a l t i m e algorithm, h e n c e very slow f o r l o n g sentences; a 1 9 6 2 r e p o r t gives t y p i c a l times as 1 m i n u t e for an 18 word s e n t e n c e and 12 m i n u t e s for a 35 word sentence. A n improvement o f more t h a n an 0 -of magnitude was obtained i n t h e f i n a l v e r s i o n of t h e program by using a b i t matrix for a p a t h -e l i m i n a t i o n technique [Kuno 19651. When an attempt was made to match a nonterminal symbol t o t h e sentence b e g i n n i n g a t a p a r t i c u l a r word and no match was found; t h e c o r r e s p o n d i n g bit was turned on; if the same symbol came up a g a i n l a t e r i n the ~a r s i n g a t t h e same p o i n t i n the sentence, the program would n o t have t o try t o match it again.Another important e a r l y p a r s e r was t h e immediate c o n s t i t u e n t anal-yzer u s e d at RAND. U n i v e r s i t y o f P e n n s y l v a n i a [Harris 1 9 6 5 1 . T h i s proctxlure, called a cycling c a n c e l l i n g . automaton, makes n s t v i e s rule then returned a fail-ure s i g n a l t o the parser, i n d i c a t i n g that the analysis was semantically anomalous, and t h i s a n a l y s i s was aborted. W o o d s has noted [Woods 197Ql t h a t t h e p a r s e r used i n t h e D m m 4 pro jecf may produce redundant parses, and has g i v e n a parsing algorithm for c o n t e x t -s e n s i t i v e languages which repedies this deficiency .When the theory of transformational grammar was elaborated in the early l . 9 6 0 1 s there was c o n s i d e r a b l e i n t e r e s t in finding a corresponding recognition procedure. Because t h e grammar is stated i n a generative form, however, t h i s i s n o s i m p l e matter. A &hornsky) tree transformational grammar c o n s i s t s of a set of context-sensitive phrase s t r u c t u r e rules, which g e n e r a t e a set of base treesa, and a set of t r a n s f o r m a t i o n s , which a c t on base trees to produce the s u r f ace trees. A (Harris) s t r i n g t r a n s f o~t i o n a l grammar consists of a finite set of sequences of word categories, c a l l e d kernel s e n t e n c e s , and a set of transformations which combine and modify t h e s e kernel s e n t e n c e s t6 e the other sentences of t h e language. There are a t least three b a s i c problems i n reversing the g e n e r a t i v e process:(1) for a tree t r a n s f o r m a t i o n a l grammar, a s s i g n i n g t o a given sentence a set of p a r s e trees which includes a l l the s u r f a c e trees which would be assigned by the trans f o r m a t i o n a l grammar( 2 ) giveq a t r e e n o t i n t h e b a s e , d e t e r m i n i n g w h i c h sequences of transf o r m a t i g n s might have applied t o generate this t r e e( 3 ) h a v i n g d e c i d e d on a transformation whose r e s u l t may be the p r e s e n t t r e e , undoing this t r a n s f o r m a t i o n If we a t t a c k each of these p r o b l e m s i n t h e most s t r a i g h t f o r w a r d manner, w e are likely t o try many false p a t h s w h i c h w i l l not l e a d t o an analysis. F o r the f i r s t problem, w e could u s e a c o n t e x t -£ ree gramrnar which will g i v e all t h e s u r f a c e t r e e s assigned by the t r a n s f o r m a t i o n a l granunar, a n d probably l o t s more The s u p e r a b u n d a n c e of ''false" s u r f a c e trees is a g g r a v a t e d by t h e f a c t that most E~~g l i s h words have more t h a n one word category ( p l a y more t h a n o n e s y n t a c t i c role), a l t h o u g h n o r m a l l y o n l y o n e is u s e d in any g i v e n sen'tcnce. (1) a n a l y s i s of t h e s e n t e n c e u s i n g t h e c o n t e x t -f r e e c o v e r i n g grammar (with a bottom-up p a r s e r )( 2 ) a p p l i c a t i o n of t h e revexse trans r o r r n a t i o n a l rules 'i3) for each candidate base tree produced by steps (1) and ( 2 ) , a check whether it can i n f a c t be generated by the base component( 4 ) f o r each base tree and sequence of transformations which passes the t e s t in s t e p ( 3 ) , the (forward) t r a n s - The covering grammar produced a large number of spurious surface analyses which the parser must process. The 1 9 6 5 r e p o r t f o r example, cites a 1 2 word sentence which produced 4 8 parses with t h e covering grammar; each must be followed t h r o u g h s t e p s ( 2 ) and ( 3 ) before most can be e l i m i n a t e d . The system was t h e r e f o r e very s l o w ; 36 minutes were r e q u i r e d to a n a l y z e one 11 word s e n t e n c e . r u l e s had a s i g n i f i c : a n t e f f e c t on parsing t i m e sthe 11 word 1 s e n t e n c e w h i c h t o o k 36 m i n u t e s hefore now took o n l y 6The system developed by P e t r i c k [ P e t r i c k 1965, 1966; K e y s e r 19671 is similar in o u t l i n e : a p p l y i n g a s e r i e s oE reverse t r a n s f o r m a t i o n s , c h e c k i n g if the r e s u l t i n g tree can be generated by the base component, and the: verifying the a n a l y s i s by applying the forward t r a n s f o~m a t i o n s ta the base The price f o r g e n e r a l i t y was p a i d in e f f i c i e n c y . Petrick's problems w e r e more severe t h a n M I T R Z ' s f o r t w o reasons. ~i r s t , the ~b s e n c e of a s e n t e n c e tree during t h e application of the reverse t r a n s formational r u l e s meant t h a t many sequences of re-rse transformations were t r i e d which did n o t correspond to any sequence of tree transformations and hence would eventually be rejected. Second, i f several rever se transformations Could apply a t some point in the a n a l y s i s , the procedure c o u l d not tell i n advance which would l e a d t o a v a l i d deep s t r u c t u r e . C o n s e q u e n t l y , each one had to be t r i e d and the r e s u l t i n g s t r u cture followed t o a d e e p structure of a "dead end" (where no more transformations a p p l y ) . T h i s produces a growth i n the number of a n a l y s i s p a t h s which is exponential in the number of r e v e r s e transformations applied. This explssion ntn be avoided o n l y if t h e reverse transformations include t e s t s of the c u r r e n t a n a l y s i s t r e e to d6terrnine which transformations a p p l i e d to generate .this tree.Such t e s t s were included in the manually p r e p a r e d reverse t r a n sformations of t h e MITRE g r o u p , b u t i t would have b e e n fax t o o complicated f o r Pe trick t o produce s u c h tests automatlcally when i n v e r t i n g the trans formations. Potrick's system has been s i g n i f i c a n t l y revised over the past decade [ P e t r i c k 1 9 7 3 , Plath 1974aI. The r e s u l t i n g system is fast enough to be used i n an i n f o r m a t i o n r e t r i e v a l system with a grammar of moderate s i z e ; most r e q u e s t s are processed in less than one minute.Transformational A n a l y z e r s : Subsequent Developmi!ints One result of the early t r a n s f o r r n a t i~n a l s y s t e m s w a s a r e c o g n i t i o n of t h e importance sf f i n d i n g an e f f i c i e n t p a r s i n g procedure i f traiisformationa.l a n a l y s i s was ever t o be a u s e f u l t e c h n i~u e . As the systems i n d i c a t e d , there are two main obstacles to an e f f i c i e n t procedure. First, there is the problem of r e f i n i ' n g the s u r f a c e analysis, s o Chat each s e n t e n c e p r o d u c e s fewer troes f o r which transformational decomposition must be a t t e n~p t e d . This h a s g e n e r a l l y been approached by u s i n g a inore The first s y s t e m u s i n g such a n e t w o r k was developed by Thorne, B r a t l e y , and D e w a r a t E d i n b u r g h [Thorne 1968, D e w a r 1 9 6 9 1 . They s t a r t e d w i t h a regular base grammar, i -e . , a transition network. The i m p o r t a k e of using a r e g u l a r base l i e s i n their claim that some transformations are equAvalent i n e f f e c t to changing the base t o a recursive transition network. Transform a t i o n s which could n o t be handled i n this fasion, such as conjunction, were incorporated in^^ the p a r s i n g program. P a r s i n g a sentence with t h i s surface grammar s The recutsive t r a n s i t i o n network was developed i n t o an augmented recursive t r a n s i t i o n network grammar i n the system of Bobrow and Frasex m w 9An ausmented network i s one in which an a r b i t r a r y predicate, w r i t t e n i n some g e n e r a l purpose language ( i n this case, L I S P ) . may be associated u i t h each a r c i n the network. A t r a n s i t i o n i n the n e t w~r k is n o t allowed if t h e predicate associated w i t h t h e arc f a i l s . These predicates perform t w o functions i n t h e grammar. F i r s t , they are used t o i n c o r p o r a t e r e s t r i c t i o n s i n the language which would be difficult or impossible to s t a t e w i t h i n the a m t e x t -f r e e mechanisms of t h e recursive n e t w o r k , e. g. , agreement r e s t r i c t i o n s .sentence is being parsed.The augmented t r a n s i t i o n network was further developed by Woods a t B o l t B e r a n e k and Newman.In order to r e g u l a r i z e the predicates, he i n t r o d u c e d a standard set of operations for b u i l d i n g and t e s t i n g the deep structure lWoods 1970bl. He considerably e n l a~g e d t h e scope of the grammar and added a semantic component f b r t r a n s l a t i n g t h e deep structure into information r e t r i e v a l commands. With that, from a n a n a l y s i s of the sentence into linguistic s t r i n g s , one could directly d e t e r m i n e the transformations w h i c h acted to produce t h e s e n t e n c e , w i t h o u t h a v i n g to t r y many s e q u e n c e s sf reverse transformations. T h e i r proposed sys tern therefore c o n s i s t e d of a procedure f o r l i n g u i s t i c string analysis ( a context-free p a r s i n g problem at the level ,of s i m p l i f i c a t i o n of t h e i r o r i g i n a l proposal) and a s e t of r u l e s w h i c h constructed from each s t r i n g a corresponding k e r n e l -l i k e sentence.T h e i r o r i g i n a l proposal was a s i m p l i f i e d scheme which accounted for only a l i m i t e d s e t of trans.formations. not specifically oriented towards r e c o g n i t i o n , to d k t e r r n i n e the features of a sentence which i n d i c a t e t h a t a particu1a.c t r a n s f o rmation a p p l i e d i n generating i t , and hence t o produce an e f f icient analysis procedure*Another group which has used l i n g u i s t i c s t r i n g a n a l y s i s is t h e Linguistic S t r i n q Project a t New York University, led by Sager [Sager 1967 [Sager , 1973 Grishman 1973a, 1973131. T h e i r s y s t e m , which has gone through s s v e r a l versions since 1965, is based on a contextfree grammar augmented w i t h r e s t r i c t i o n s . Because they were conce ned with processing s c i e n t i f i c text, rather t h a n commands or queries, t h e y were l e d to develop a grammar of particularly broad coverage. The p r e s e n t gramrr~ar has about 250 context-free rules and about 2 0 0 r e s t r i c k i o n s ; although not as swift as some of the smaller s y s t e m s , t h e parser is able to analyze most sentences in less than one minute. Because of the l a r g e size of their grammar, t h i s group *has been p a r t i c u l a r l y concerned w i t h techniques for organizing and s p e c i f y i n g the grammar which w i l l f a c i l i t a t e f u r t h e r development. In p a r t i c u l a r , the most recent implementation of t h e i r s y s t e m has added a s p e c i a l language designed for t h e economical and perspicuous s t a e m e n t of the r e s t r i c t i o n s [Sager 19751.One of t h e e a r l i e r versions of t h i s s y s t e m , w i t h a much more restricted grammar, was used as the f r o n t end for an information r e t r i e v a l s y s t e m d e v e l o p e d by C a u t i n at t h e U n i v e r s i t y of P e n n s y l v a n i a [ C a u t i n 1369 ] .The ~i n g u i s t i c S t r i n g P r o j e c t s y s t e m has r e c e n t l y been e x t e n d e d to include a transformational decomposition p h a s e ; t h i s phase follows t h e l i n g u i s t i c ? s t r i n g a n a l y s i s [IIobbs 1 9 7 5 1 . should semantic a n a l y s i s be done c o n c u r r e n t l y with Syntactic a n a l y s i s . P a r a l l e l processing is p r e f e r r e d if t h e added time r e q u i r e d by the deeper a n a l y s i s i s outweighed by t h e f r a c t i o n of i n c o r r e c t analyses which can be e l i m i n a t e d early i n the parsing erocess. In the case of s ?mantic analysis, it clearly depends on t h e r e l a t i v e complexity of t h e s y n t a c t i c and semantic components. I n t h e case of transformational a n a l y s i s , it depends on the f r a c t i o n of grammatical and s e l e c t i~n a l c o n s t r a i n t s which can be e x p r e s s e d at: t h e surface l e v e l (if most of these can only be realized through transformational allalysis , concurrent trans formational a n a l y s i s i s probably more e f f i c i e n t ) . This may depend i n Lurn on the t y p e of surface a n a l y s i s ; for example, the r e l a t i o n s h i p s exhibited by l i n g u i s t i c string analysis axe s u i t a b l e for expressing many of these c o n s t r a i n t s , so there i s less motivation in the Linguistic S t r i n g Colmerauer and de Chastellier [de C h a s t e l l i e r 196 9 ] have also investigated the possibility of using Wijngaarden grammars (as were developed f o r s p e c i f y i n g ALGOL 6 8 ) f o r :tr.ansformational decomposition and machine t r a n s l a t i o n . L i k e u n r e s t r i c t e d r e w r i t i n g rules, W-grammars can d e f i n e every r e c u r s i v e l y enumerable l a q u a g e , and s o can perform t h e functions of t h e syrface and reverse transformational components. They show how p o r t i o n s of transformational grammars of English and French may be rewritten as W-grammarsr w i t h t h e pseudo-rules in t h e W-grammar t a k i n g the place of the t r a n s f o l ' m a t i o n s~In all t h e systems dqscribed above, a s h a r p line was drawn between correct and incomect parses: a t e m i n a l node e i t h e r did or d i d n o t m a t c h t h e n e x t word in t h e s e n t e n c e : an a n a l y s i s of a phrase was e i t h e r acceptable o r unacceptable. There are circumstances under which we would want to r e l a x these requirements. For one thing, i n analyzing connected speech, We segmentation and i d e n t i f i c a t i o n of words can never be done with complete c e r t a i n t y . A t b e s t , one can say that a c e r t a i n sound has some p r o b a b i l i t y of being one phoneme and son= o t h e r p r o b a b i J i ty of being another phoneme; some e x p e c t e d phonemes may be l o s t e n t i r e l y in the sound received. c o n s e q u e n t l y , one w i l l a s s o c i a t e some n u f i e r w i t h each t e r m i n a l node, i n d ic a t i n g t h e p r o b a b i l i t y or q u a l i t y of match; noflcrminal nodes will be a s s i g n e d some v a l u e based on t h e v a l u e s of t h e t e r m i n a l nodes ) x n e a t h . Another circuinstance a r i s e s in n a t u r a l l a n g u p a r s e t r e e u n t i l i t g e t s s t u c k (generates a t e r m i n a l n o d e which does n o t match the next s e n t e n c e word) ; it t h e n "backs up" to t r y another a l t e r n a t i v e .IleL A measure is a s s o c i a t e d with each a l t e r n a t i v e path, fadicating the l i k e l i h o o d t h a t this a n a l y s i s matches the sentence p w s s e d so far and t h a t it can be extended to a let@ s@xtence analysis. At each moment, t h e path w i t h the higbsst likelihood is extended; if its measure f a l l s below that otner path, the parser s h i f t s its a t t e n t i o n to t h a t allowed patterns o f occwl-rence of con j o i n i n g s i n a sentence are q u i t e r e g u l a r . Loosely speaking, a s e q u e n c e of e nts in the s e n t e n c e tree may be followed by a con j u n c t i o n aad by same or all of the elements immediately preceding the jrmction. For example, allowed p a t t e r n s of con joining subject-verb-ob ject-and-sub ject-verb-ob ject ( I drank d w and nary ate cake. ) , sub ject-verb-ob ject-and-verb-object a milk and a t e cake. ) and sub ject-verb-ob ject-and-ob ject For i n s t a n c e , i n t h e s e n t e n c e ) "I . T h i s may be done either during t h e s y n t a c t i c a n a l y s i s [woods 1 9 7 3 , Simmons 1 9 7 5 1 o r after the syntax phase is complete [~o r g i d a 1 9 7 5 , Hobbs 19751. C o n t e x t -f r e e grammars p l a y e d a major role i the e a r l y s t a g e s of automatic n a t u r a l l a n g u a g e a n a l y s i s . A1 t h o u g h t h e y have now g e n e r a l l y been superceded by more c o n~p l e x and powerful grammars, many of these grammars are based o n or have as one of t h e i r components a context-f ree grammar. The s e l e c t i o n of an e f f i c i e n t c o n t e x t -f r e e p a r s e r t h e r e f ore r e m a i n s an i~n p o r t a n t considerat ion i n n a t u r a l Language analysis. T h e o t h e r p a r s e r s g r a b a l l y accumulate d a t a from which a l l parses of a s e n t e n c e can be e x t r a c t e d ; t y p e s 1, 2 and 3 s t o r e t h i s data i n decreasingly compact r e p r e s e n t a t i o n s .
1:
A Parser for U n r e s t r i c t e d R e w r i t i n g Rule Grammars.. 54 d e p e n d i n g on t h e a p p l i c a t i o n , t h e o u t p u t may b e a command i n a n information retrieval l a n g u a g e , a s t r u c t u r e based on some set of semantic " p r i m i t i v e s " , o r a t a b u l a r s t r u c t u r e s u i t a b l e as a d a t a base; this s t a g e should r e c o g n i z e some o f the p a r a p h r a s e s due t o a l t e r n a t i v e choices of words. ( 3 ) pracjmatic: i n t e r p r e t s the t e x t based on p a r t i c u l a r c o n t e x t (problem s i t u a t i o n or data base) ; this stage should r e c o g n i z e s e n t e n c e s which are e q u i v a l e n t i n e f f e c t , (such as " T h r~w that switch." and "Turn on the light.") .The reader w i l l note t h a t these stages are very-.vaguely charact e r i zed. Current language p r o c e s s i n g s y s tems differ very g r e a t l y i n t h e i r s t r u c t u r e and n o t oven these general d i v i s i o n s can be identified in a l l s y s t e m s . Ihe pragmatic s t a g e is the most heterogeneous and t h e common t h r e a d s which do appear are based more on g e n e r a l problem-solving methods t h a n on specifically linguistic t e c h n i q u e s . Since the s e m h n t i c s t a g e maps into the n o t a t i o n required by the p r a g m a t i c s , T h e four t y p e s are : (0) Develops a s i n g l e parse tree a t a t i m e ; a t any i n s t a n t t h e store holds a set o f nodes c o r r e s p o n d i n g t o the nodes of an incomplete potential parse tree f t ) The s t o r e holds a s e t of n o d e s , each of which represents t h e fact t h a t some substring of t h e s e n t e n c e , from word f to word R , can be a n a l y z~d as some symbol N. <ao, a l r a 2 , .. ,a, The v a r i a b l e NODES holds a c o u n t of the number of npdes i n t h e p a r s e t r e e ; this is also t h e number af t h e node most r q c e n t l y added to t h e tree. WORD h o l d s the number of t h e n e x t word in fie s e n t e n c e to be matched.The heart of the parser is the r e c u r s i v e p r o c e d~r e EXPAND. E X P L t D i s passed one a r g u m e n t , the number of a node i n t h e parse tree. I Y EXPAND has n o t been called for this node before, i t w i l l try t o expand the node, fbe. , build a p a r s e t r e e below the node which matches p a r t of t h e r e m a i n d e r of the s e n t e n c e . If EXPAND h a s a l r e a d y been c a l l e d once for this node -s o t h a t a tree a l r e a d y e x i s t s below this node --EXPAND tries to f i n d an a l t e r n a t e tree below t h e node w h i c h w i l l match u p with p a r t of the remainder of the s e n t e n c e .If EXPAND i s successful -an ( a l t e r n a t e ) tree below the node was found -it r e t u r n s the value true; i f it is u n s u e e e s s f u l , ,/* if t r e e s p a n s e n t i r e s e n t e n c e , add to set */ This algorithm i s sometimes called t h e "Immediate Co~stituent A n a l y s i s " (IcA) alqorithm, because i t was u s e d q u i t e early i n parsing n a t u r a l langqage with ICA grammars, 1%-c o n s t r u c t s a l l nodes in a single l e f t -t o -r i g h t p a s s over the s e n t e n c e . A s each word i s scanned, the p a r s e r builds a l l nodes w h i c h subsume a p o r t i o n of t h e s e n t e n c e ending a t that word. The nodes ( " s p a n s " ) are accumualted i n a two-dimensional a r r a y S P A N , whose f i r s t subscript s p e c i f i e s the number of the span and whose second subscript selects a component of the span, a s follows: SPAN (n, 'NAME ' ) = name of s p a n n SPAN (n, 'FW' ) = number of f i r s t sentence word subsumed by span n SPAN (n, 'LW+ll) = (number of last s e n t e n c e word subsumed by span n) + 1 SPAN ( n , 'D'AUGHTERS') = t u p l e of numbers of d a u g h t e r s p a n s o f s p a n n. (1 <= \tI <= SPANS I (SP.W(I;'NAMG') aq ROOT) and (SPAN (1, ' FW' ) eq 1) and (SPAN(I,'LW+lt) eq SE SENTENCE)+^))) NODES = 1; TREE = nl;T R E E C~) = S P A N I I I ; TREE: (1, 'DAUGHTERS ' ) = E X P A N D (TREE (I, ' DAUGEITERS ' ) , 1) ; r e t u r n s a s e t , each e l e m e n t of w h i c h is an ( n t l ) -t u p l e , whose t a i l is one of the n -t u p l eof s p a n s and whose head i s the r l u n h e r of t h e first word s p a n n e d by the n -t u p l e of s p a n s * / if E L I S T cq n u 9 t then r e t u r n ( ( E N D w D P~~~) ; e l s e r e t u r n [U: i f EXPANDED (TREE (X, 'NAME ' ) , TREE (X, ' FW' ) ) eq true t h e n /* the expansions f o r this symbol have been computed before */ /* if t h i s is a new node, get its WFS e n t r i e s */ The type 2 bottom-up parallel algorithm a l s o saw e a r l y use i n 5% n a t u r a l language processing. The parser d e s i g n e d by Cocke for the Rand system w a s a special version of t h i s a l g o r i t h m for Chomsky normal form grammars. A thorough survey of t h e d i f f e r e n t o r d e r i n g strategies p o s s i b l e w i t h t h i s algorithm was given by Hays [1967] . This algorithm was subsequently developed by Cocke (among others) into a t y p e 1 bottom-up paralleL algorithm named "nodal spans" and s u b s e q u e n t l y i n t o a type .1 top-down p a r a l l e l algorithm called "improved nodal spans" (see Cocke [1970] f The advantage of type 1 over t y p e 2 algorithms depends on t h e degree of ambiguity of the grammar. How f r e q u e n t l y can a p o r t i o n of i+ s e n t e n c e be a n a l y z e d as a particular symbol i n s e v e r a l ways?1 <= 1 <= NODESJ (if ( S P A N (I, ' N A M E ' ) eq ELXST ( # E L I S T ) ) and (SPAN (I, 'LFJ+ll ) eq E N D W D P 1 ) then CNTUP + <I>, NTUP E bylTCH ( E L I S T (1: ( $ E L I S T ) ,SPAN (1, 'FlY' ) ) 1if TREE ( x , 'ALTERNATE WFS ' ) eq Q then TREE(X,'ALTERNATE WFS') = IS, 1 -< S -< WFSS I (WFS (S , 'NAME ) eq TREE a, 'For unaugmented c o n t e x t -f r e e grammars t h e answer in general has been very f r e q u e n t l y -t h i s was one of t h e problems of t h e context-free systems. For such grammars, type 1 algorithms would be much more e f f i c i e n t . When r e s t r i c t i o n s are added, however, t h e y discriminate some of the analyses from o t h e r s . A number of n a t u r a l language systems, such as REL (at t h e C a l i f o r n i a Institute of Technology) a n d the Q-system (at t h e U n i v e r s i t y of Montreal) have used u n r e s t r i c t e d phrase s t r u c t u r e grammars. In such grammars, each rule specifies t h a t some sequence of symbols be r e w r i ' t t e n as some o t h e r sequence of symbols. The parsing algorithm used i n these s y s t e m was described by Kay in 1967 ("Experiments w i t h a Powerful. P a r s e r , " Martin Kay, ih 26me C o n f e r e n c e I n t e r n a t i o n a l e sur le Traitmcnt Automatique des Langues , Grenoble) . however, we s h a l l n o t be concerned w i t h these a d d i t i o n a l f e a t u r e s ; o n l y t h e b a s i c p a r s i n g procedure will be described below.The parser to be p r e s e n t e d r e p r e s e n t s o n l y a small modification to the context-free parser B (the "immediate c o n s t i t u e n t a n a l y z e r " ) g i v e n e a r l i e r . To u n d e r s t a n d t h i s modification, consider t h e f o l l o w i n g example. We a r e given a context-Free grammar which includes t h e p T h e parscr would then a p p l y x -c y a i n r c v c r s e : g e t t i n g a s p a n which subsumes the e n t i r e sentence. Now c o n s i d e r a n a l y z i n g t h e same s e n t e n c e w i t h t h e u n r e s t r i c t e d p h r a s e s t r u c t u r e ! grammar z + y ax + z bWe b e g i n by ilsing t h e f i r s t production tc r e d u c e the s e~~t c n c e to y a b . T h i s raises the problem of how to l a b e l tthc no~ic between arcs a and b. ~l t h o u g h a and b t o g e t h e r s u b s~~a the l a s t t h r e e words of the s e n t e n c e , no f r a c t i o n of-this C~I I be a s s i g n e d i n d i v i d u a l l y to a or to b; h e n c e we c a n n o t l a b e l this new node with t h e number of a sentence r e t u r n ; end ADDSPANS;T h e u n r e s t r i c t e d r e w r i t i n g r u l e p a r s e r has t h e power of a 1966 ,1973 ,1975 Keyser 1967 ; P l a t h 1974a, 1974bl.Interestingly enough, these modifications have b r o u g h t Petrick's parser much closer t o the o r i g i n a l MITRE design.Since the structure of trans fom'iational grammar h a s v a r i e d in t i m e and b e t w e e n d i f f e r e n t schools of linguistic t h e o r y , the notion of a transformational parser is not we11 defined. In order to p r e s e n t a p a r s i n g algorithm, we have selected a particularly simple granular formulation. This E o r i n u l a t i o n corresponds a~p r o x i r n a t e~l y to the early work o fChomsky ( e . g . , Syntactic S t r u c t z t r e s ) a n d If s c i j is an integer,between 1 and n, t h e new node is t h e node matched to the s c i j -t h e l e m e n t of t h e s t r u c t u r a l i n d e x ; i f SCij is a t e r m i n a l symbol, the new node is a t e r m i n a l node w i t h t h a t name.Because the v a l u e of sci may be t h e n u l l t u p l e < >, i t is p o s s i b l e f o r a node in the tree to be left w i t h n o successors. We t h e r e f o r e "clean u p " the tree a f t e r a p p l y i n g the t r a n s f o r m a t i o n by d e l e t i n g any n o n t e m i n a l node not dominating at l e a s t one t e r m i n a l node.The p r e s c r i p t i o n just g i v e n is i n a d e q u a t e f o r components ~f the structural i n d e x e q u a l to " X " , since these may match zero or moxe t h a n one node. ide s h a l l cons t r a i n t h e t r a n s f o r n a - The r u l e COMP -+ # S # , which maes the base conponcnt r e c u r s i v e , a l s o p l a y s a s p e c i a l r o l e in t h e t r a n s f o r m a t i o n s .appears in the parse t r c e , w e call the tree dominated by t h a t S a c ? 3 ) i s t i t l t d n t (or embedded) s e n t e n c e , and t h e tsce &~n i n a t e d by the n e x t S above COMP t h e rnadl?,ix s e n t e n c e parse t r e e . . > o t e n t i a l surface structure parse t r e e s for a s e n t e n c e and $hen, by a p p l y i n g the t r a n s f o r m a t i o n s i n r e v e r s e , try t o obtain a v a l i d deep s t r u c t u r e from each of these. We shall deal w i t h these t w o steps i n t u r n .The surface s t r u c t u r e p a r s e tree will, i n general, c o n t a i n many s t r u c t u r e s which could n o t be directly g e n e r a t e d by t h e base component. Because t h e l a n g u a g e d e f i n e d by the augmented base component i s larger t h a n t h a t d e f i n e d by t h e t r a n x E o r m a t i o n a 1 graminar, by some early results o b t a i n e d by the MITRZ group. The MITRE system did not have a procedure for automatically augmenting the base component; t h e i r s was assembled manually. Using a small g r m a r . , one of their 12-word test s e n t e n c e s obtained 48 surface analyses, almost a l l of thorn s p u r i o u s . P e t r i c k had s i m i l a r experience: he found that t h e covering grammars produced by his procedure-were too broad, p r o d u c i n g too many surface p a r s e s . He has i n s t e a d , l i k e t h e MITRE group, produced his surface grammars manually, by a n a l y z i n g o o n s t r u c t i o n s which appear in the s u r f ace s t r u c t u r e of i n p u t sentences to determine w h i c h productions are r e q u i r e d . In How can this s h u f f l e be reversed? W e begin by c r e a t i n g a n(1 <= j <= r) according to. j i s i ( j ) = if r s c ( j ) is an i n t e g e r t h e n s i ( r s c ( j ) o r i g i n a l string s j s ( j ) = if isc( j ) i s an i n t e g e r t h e n s ' ( k c ( j ) ) else iscl j) I f there are several matches to t h e isi, the t r a n s f o r m a t i o n must be a p p l i e d t o a l l ; w e can o n l y be s u r e t h a t one of t h e r e s u l t i n g strings w i l l be s . I f the forward t r a n s f o r m a t i o n 1 is a recoverable deletion i n v o l v i 3 s i d e n t i t y conditions, t h e formulas g i v e n above are somewhat more complicated.Given a s e t of reverse t r a n s f o r m a t i o n s , we m u s t f i n a l l y specify t h e s e q u e n c i n g among them. The reverse transformations should be c o n s i d e r e d i n precisely t h e reverse o r d e r fro11 t h a t of the c o r r e s p o n d i n g f o r w a r d t r a n s f o r m a t i o n s . The sequericihg is a g a i n cyclic, with each i t e r a t i o n now c r e a t i n g an embedded sentence.Even i f a reverse t r a n s f o r m a t i o n matches t h e s e n t e n c e being decomposed, one cannot be s u r e t h a t the corresponding forward transformation was involved in the generation of the sentence. Uridoing the t r a n s f o r m a t i o n may lead t o a dead end [ P e t r i c k 1965 ] S t q l e y R e P e t r i c k , A R e c o g n i t i o n Proceddre f o r T r a n s formational Grammars. Doctoral Disseztatf on .[ p e t r i c k 19661 Stanley R. F e t r i c k , A i r o g r a m for Transformational S y n t a c t i c Analysis, Air Force Cambridge Research Laboratories, AFCRL--66-698.[ P e t r i c k 19731 S t a n l e y R . P e t r i c k , T r a n s f o r m a t i o n a l Analysis, I I I n N a t u r a l Language Processing, ed. R. R u s t i n , Algorithmics P r e s s , N. Y.[ P e t r i c k 19 75 I Stanley R . P e t r i c k , "Design of the Underlying S t r u c t u r e for a D a t a Base Retrieval. S y s t e m . " In D i r e c t i o n s in A r t i f i c i a l I n t e l l i g e n c e : Natural Language P r o c e s s i n g , ed. R. Gsishman C o u r a n t 9 7 2 , 19731.The augmented t r a n s i t i o n network, and i n p a r t icular the formalism developed bv Woods, has proven t a be ant h e s u r f a c e a n a l y s i s .T h i s w o l l d s e e m to be d i s a d v a n t a g e o u s from the p o i n t of view of e f f i c i c n c y , s i n c e e r r o n e o u s p a r s e s which m i g h t be aborted a t the ' b e g i n n i n g of t h e s u x f a c e a n a l y s i s m u s t be followed t h r o u g h t h e e n t i r e s u r f a c e a n a l y s i s an d p a r t of the transformational d e c o n p s s i t i o n . Second, the t r a n s format i o n s are n o t a s s o c i a t e d w i . m p a r t i c u l a r productions of the s u r f a c e grarxnar, b u t r a t h e r w i t h p a r t i c u l a r p a t t e r n s in t h e tree ( " s t r u c t u r a l descriptions") , so p a t t e r n match; n g o p e r at i o n s a r e r e q u i r e d to d e t e r m i n e 1;rhich t r a n s f o r m a t i o n s t o a p p l y . These d i f f e r e n c e s r e f l e c t P e t r i c k ' s c'iesire to remain as close as is p r a c t i c a l to the f o r m a l i s m of t r a n s f o r m a t i o n a l l i n g u i s t i c s .T h e primary d i s t i n c t i o n o f t h e Woods system is that the deep structure tree is built during the surface analysis. Consequently, his " t r a n s f o r m a t i . o n a 1 " procedures c o n s i s t of t r e e b u i l d i n g r a t h e r t h a n tree t r a n s f o r m i n g o p e r a t i o n s . The t r a d e o f f s between this approach and the two-stage a n a l y z e r s of p e t r i c kt h e augmented base component i s c a l l e d a c o d a 1 9 i n g gramrzar. S i n c e each s p u r i o u s s u r f a c e a n a l y s i s will have to undergo a len'gthy reverse t r a~s f o r m a t i o n a l process before i t !S r e c o g n i z e d as i n v a l i d , i t i s important t o minimize the number of such parses. The s e r i o u s n e s s of t h i s problem is i n d i c a t e d -, A N 7 , 1473 E D I T t O N O F 1 NOV 6 5 I S OBSOLETE UNCLASSIFIEDS E C U R I T Y C L A S S I F I C A T I O N 3 F THIS P A G E (WhenData Entered,'
:
I t a l s o c o n t a ' i n s p r e c i s e . s p e c i f i c a t i o n s i n t h e programming l a n g u a g e S E T L o f a n u m b e r o f p a r s i n g a l g o r i t h m s , i n c l u d i n q s e v e r a l c o n t e x t -f r e e p a r s e r s , a u n r e s t r i c t e d r e w r i t i n g r u l e P a r s e r , a n d a transformational p a r s e r . ......Parsing A l g o r i t h m s for C o n t e x t -F r e e Grammars.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 585 | 0 | null | null | null | null | null | null | null | null |
cd23da0154f7973685238eb8f3056258e9d7ea78 | 59766402 | null | Computer Understanding of Metaphorically Used Verbs | A major problem conrronting computer pro:lram:. driven by nnturol-1 ngu ge i n p u t consists of thc interprctntion of 1 i n g u h . l ic cxprossiobs f o r which the int endcd literal meaning i s not e : q ~l i c i t l y given by the l c ~i o i ~l coml>onents of the expresL;ion. ~'tn ccxi~mple is thc "extended useu of the verb ' l e a p t in 'the country l e a p t to prosperity'. buch extended usages--whether c u n s i -a s inilntcd or original-czn be considered metaphorical t o t h e cktcnt t h -, t t h e y are based on analogies. T h i s paper establishes a framework f o r interpretin? netaphoric,ll expres ionk by analysis o f underl y i n g cbstrgct cor:~ponents --such 9 s "trnns tio on" and "inten-' i ' h ~s i s i n c o n t r c s t t o p r esity" for t h e nbovc example. viouc: epproaches which rely on 3 a ~m b c r ~f word senses intended to represent m e t n p oricL21 usi ges d i r e c t l y . An eexperimentcl r r o )ram f i n d , literal inter7rctations f o r inyut rcprc~snting a simyle scnt~l ce m \.hit!. t h e "verb21 concepttf ( a c t i o n , s t a t e or c t t r i b u t c ) is used metaphorically. This in ut has t h c gcner 1 c o n A g u r a t i o n 1 ~u B JCfl Vc,& OBJECT ;~UXCE/GO.IL or 2 UBJLCT PiLDICA TE-ADJECTIVEf. The i n t e r p r e t a t i o n a r e ~i v e ? i n the f o m of primitive ;;nglisFI pcraphrases . The-e parsphrases , which a r e intended merely to illustrate the informcrtisn which czn be extrrcted from metr.?:~orical inmt , d r e b L sed on scmr.ntic representations which are convertible to structures specified by SchL.nkt s conceptual deaendency theory. The interpretation of mctaphori~nlly used verbs thus represents a particul..r case o f the general tusks of disambiguation and interpretation encountered by the conce-tual dependency parser, The h p p r o x i m ~t i o n to the l i t e r a l mc, ning of n metaphorical verb i s achieved through r e f e r e n c e t o s e m a n t i c d e s c r i ptions based p r i m a r i l y on n srn.111 number of. c o n c e p t u a l features and absrrrct structures. THese descriptors a r e specif i c d Lor classes of those concepts which ore c:qrcosed in d n g l i s h by nouns, verbs, a d j e c t i v e s .:nd p r e p o s i t i o n a l phrases. The complete set of values L-or the c l e s c r i p t o r s o r verbal concepts is represented as a multi-dimensional matrix containing the defined conceptd. This m a t r i x , which is only p ~r t i a l l y described in this paper, exhibits relatlonshipb and analogies wldch underlie metaphorically used verbs. The relative independence and c b u t r c c t chnracter of the b a s i c sernuntic d e s c r i p t o r s render rhe system easily extensible to f u r t h c r c a p a b i l i t i e s , such a s more conclusive interpretations or the tre tment or' more chnllenging expressions. The emphasis on systematic uescriptions ,nd p r i m i t i v e concepts t o produce s i m ~l e p raphrases is viewed a s r e i l e c t i n g h w n understanding or novel linguistic expressions an& providing a model t o explore q u e s t i o n s related to such under-tanding. Contents 1 . Approach 1.1 h a l o g i e s 1.2 Conceptual dependency interpretations 2. Characterization of Verbal Concepts 2.1 Levels 2.2 States 2.3 Structures 2,4 Features 3, Characterization of NOMLNALs 3.1 Features 3.2 Function descriptors 4. Nethod of Interpretation 4 . 1 Conditions on metaphorical extension 4.2 Operational context 4.3 General procedure 4.4 Operation of r o u t i n e 4.5 Tests and c r i t e r i a 5 . Examples 5.1 L e v e l s h i f t 5.2 Category shift 5 . 3 R-O switch 5 . 4 Intra-level feature shift 5.4,L Actor-feature s h i f t 5.4.2 Object-feature shift 5.5 ~o u n compounds 6. Conclusion Metaphorical usages have often been regarded n s "special ccscou to vhich the particular language analysis mcthod under discussitm did not apply. This pnpcr prcsdnts n mcthod for comsutez* undcistnnding of a class of phrases in which the verb i s used 9 n e t n p h o r i c o l l y t t , but which ignorer: the distinction between rrextcndedlv and " a s s i ~n i l t e d t r usages. This approach provides flexibility i n handling previously unseen usages, The assumjtion underlying t h i s approach is t h a t analogies ore involved i n language understanding t o a g r e a t e r extent than speakers consciously realize. 1.1. -Analogies Analogies arc t h e means by which we substitute, extend or borrow concepts. In the use of an analogy, a word is borro?+erl Jrom its usual context t o exprcs3 some component of meaning sh'rrcd by t h e concept underlying the borrowed word in its l i t e r a l sense and the concept which t h e borrowed word i s t o reprclsent. T h i s results i n an extendeu o r netaphorical use or the word. The system t o be described i s intended to show the analogy comprehension necessary for the interpretation of metaphorical usages of verbs. The problem of determining the meanins of a metaphorical expression is one of knowing the c r i t i c a l similarities and differences which a borrowed sense of a word has w i t h respect to the orilginol sensc. In somc case.; an es::cntic,lly meto-?horic,lL usage ccoses to be t l ~o u g h t of a s b o r r o r ~c d , nnd a c q u i r e s on idiomatic scnsc of its own. r!owcvcr, i r thc sirr,il.;:rit.ies and dir'r'erencu which ontcr i n t o : r t n p h o r i c . ; l usnses c.In be i d c n t i f i c d , vr? c:?n still ht ndlc ~u c h nn exprcssian as wc uo those exnressions which :.rc p:encrally viewed as iact n ~~h o r i c n l . Concid~r t hc L xk3l.~ples L) L'he ilousc 1;illeck t h e bill 2) I scc w h ~t you mean Here the f i r s t example oprlears to be ncraphorical, the second not. A langu,~ge nnaLy..er prepared to handle only non-meta-pl10ri.c~ 1 input might achieve the c o r r e c t interpret2 t i o n o f '1 see1 i n the sensc of '1 un~crst'nd I . :io,:evcr, it would succcecl only if lsectswere listed' in t h e dictionall) as enuiva l e n t to tunaerut..i~d' in one sensc. Such n s o l u t i o n ignores the cc:~pahilities which humrns hhve for correctly interpreting such sentences w i t h o u t h ving lecrned t h i s s:.nonymity. | {
"name": [
"Russell, Sylvia Weber"
],
"affiliation": [
null
]
} | null | null | null | 1976-05-01 | 0 | 9 | null | The idea of growing their own radishes was born , Each verb w i l l thus be classified, not i n t e r m s of a single category, but in terms of two type9 of variables having values according to these two criteria. Thus levels and structure-concepts must be determined which can b e used as a basic form of description of verbs in the dictionary. | null | null | null | In addition to such verb descriptions, which serve the 'forget ' and other 'verbs of mental transition. It is these primitive concept8 rather than :my snecif ic lexical verb which w i l l be retrieved from the matrix as output of an operational metaphor routine.The four levels postulated for vcrbnl concepts are: Each of these levels has a few sublevels (e.g. SENSORY:eye, ear) which are sometimes s p e c i f i c a l l y referenced in metaphorical extensions. These are described in (5).Given the matrix format of the verb descriptors, the specified levels (row components ) can best be clsrified : byI--* b) start TRiNSITION (leave STA'IE) (TR-L) ' I I-- C ) f i n i s h TILINSITIUNstructure of a verb from its ordir.sry select ional restrictions.prosperity' in t h a t 'prosperity ' rather than 'countryt sppeors to be the goal :>nd is thus i n i t i a l l y oadigned role R r. ther t h~n 0 . This kind or metaphor may a c t u a l l y includc B category s h i f t (which itself may include n levcl s h i f t ) ,and is unad to express a change of s t a t e (of 'countryt) ns a transitioh (of lcountry' ).Looking Thi.: example points out the acco~aplishments and limits of the ny;tcm in defining components :signif icnnt to metaphor. ) R fulfills condition tc-l*;t h e v~r i o u s metaphor mech:tnisms which have been identified. The (;uestion arises as to the extent to which such mechanisms hold for any metaphorical use of a verbal or a t t r i b u t i v e | Main paper:
conceptual dependency interpretations:
In addition to such verb descriptions, which serve the 'forget ' and other 'verbs of mental transition. It is these primitive concept8 rather than :my snecif ic lexical verb which w i l l be retrieved from the matrix as output of an operational metaphor routine.The four levels postulated for vcrbnl concepts are: Each of these levels has a few sublevels (e.g. SENSORY:eye, ear) which are sometimes s p e c i f i c a l l y referenced in metaphorical extensions. These are described in (5).Given the matrix format of the verb descriptors, the specified levels (row components ) can best be clsrified : byI--* b) start TRiNSITION (leave STA'IE) (TR-L) ' I I-- C ) f i n i s h TILINSITIUNstructure of a verb from its ordir.sry select ional restrictions.
r-0 switch b-0 switch isexemplif i e d by 't l e country l e a p t t o:
prosperity' in t h a t 'prosperity ' rather than 'countryt sppeors to be the goal :>nd is thus i n i t i a l l y oadigned role R r. ther t h~n 0 . This kind or metaphor may a c t u a l l y includc B category s h i f t (which itself may include n levcl s h i f t ) ,and is unad to express a change of s t a t e (of 'countryt) ns a transitioh (of lcountry' ).Looking Thi.: example points out the acco~aplishments and limits of the ny;tcm in defining components :signif icnnt to metaphor. ) R fulfills condition tc-l*;t h e v~r i o u s metaphor mech:tnisms which have been identified. The (;uestion arises as to the extent to which such mechanisms hold for any metaphorical use of a verbal or a t t r i b u t i v e
1):
The idea of growing their own radishes was born , Each verb w i l l thus be classified, not i n t e r m s of a single category, but in terms of two type9 of variables having values according to these two criteria. Thus levels and structure-concepts must be determined which can b e used as a basic form of description of verbs in the dictionary.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 585 | 0.015385 | null | null | null | null | null | null | null | null |
9f700f8decabe48b71c892ef7116e51ca0f5eca8 | 219301440 | null | A Computational Treatment of Coordinate Conjunctions | The present paper describes a computational solution to the problem of locating words that are zeroed under conjunction. In this solution, which is based on general properties 3f conjunctional constructions, a mechanism locates zeroed elements in the conjunction strings and cross-references them d t h respect to elements in the h e a d construction. Constraints can then be a p p l i e d to elided conjuncts a s though t h e y w e r e expanded, and transformational expansion, which reconstructs the complete sentences underlying conjunctional occurrences, can be carried out straightforwardly by following the pointers which have been set up. The main innovation is called "stacking". It is a nond.eterministic programming device which causes restrictions ( i . . , subprograms applying detailed constraints) to be reexecuted on conjoined segments whenever the restriction is inovked on an element which has a conjunct. | {
"name": [
"Raze, Carol"
],
"affiliation": [
null
]
} | null | null | null | 1976-09-01 | null | null | null | The present paper describes a computational solution to the problem of locating words that are zeroed under conjunction. In this solution, which is based on general properties 3f conjunctional constructions, a mechanism locates zeroed elements in the conjunction strings and cross-references them d t h respect to elements in the h e a d construction. Constraints can then be a p p l i e d to elided conjuncts a s though t h e y w e r e expanded, and transformational expansion, which reconstructs the complete sentences underlying conjunctional occurrences, can be carried out straightforwardly by following the pointers which have been set up.The main innovation is called "stacking". It is a nond.eterministic programming device which causes restrictions ( i . . , subprograms applying detailed constraints) to be reexecuted on conjoined segments whenever the restriction is inovked on an element which has a conjunct. One particularly intricate problem in the computer parsing of natural language texts is the complexity introduced into the parsing system by conjunctions. This complexity is due to the richness of conjunctional constructions and to the material implicit in sentences containing conjunctions, The computafional problem can be divided into three parts: (I) generating parse trees which cover the occurrences of conjunction-headed strings in sentences;(11) locating words in the sentences which are repeated implicitly in a "zeroed," i.e., elided, form in particular positions in conjunction strings;and (III) reconstructing the complete sentences underlying conjunctional occurrences when the application requires such an expansion.A generalized algorithmic solution to problem (I) was provided over a decade ago by the New York University Linguistic String Project (LSP) in the framework of linguistic string analysis (Sager e.t al. 1966 , Sager 1967 , Raze 1 9 6 7 ) , and similar devices have been described more recently, e.g.,by Woods in the framework of augmented transition networks (Woods 1973) . The present paper describes a compukational solution to problem (II), locating the words that are zeroed under conjunction. In this solution, which is based on general properties of conjunctional constructions, a mechanism locates zeroed elements in the conjunction strings and cross-references them with respect to elements in the head construction. Constraints can then be applied to elided conjuncts as though they were expanded, and transformational expansion, which was problem (111) above, can be carried out straightforwardLy by following the pointer's which have been set up.The main innovation is called "stacking." It is a nondeterministic programming device which causes restrictions e . , subprograms applying d e t a i l e d c o n s t r a i n t s ) t o be re-executed on conjoined segments wheqever t h e r e s t r i c t i o n i s invoked on a n element which h a s a c o n j u n c t . S i n c e t h e cons t r a i n t u s u a l l y a p p l i e s t o a combination o f words i n t h e main s t r i n g and i n t h e c o n j u n c t , t h e d e v i c e must e n s u r e t h a t t h e p r o p e r grammatical r e l a t i o n h o l d s between t h e s e wards. That i s , it t r e a t s t h e s e s e p a r a t e d words as a s i n g l e l i n g u i s t i c e n t i t y , o b t a i n i n g t h e e f f e c t of a f u l l expansion o f t h e conj u n c t i o n a l o c c u r r e n c e w i t h o u t c a r r y i n g o u t t h e p h y s i c a l rearrangement and copying or t h e p a r s e t r e e . The s t r a t e g y of t r e a t i n g e l l i p s i s i n t w o steps ( l o c a t i n g d e l e t e d elements and l a t e r c a r r y i n g o u t t h e p h y s i c a l e x p a n s i o n ) is See Appendix f o r a d e f i n i t i o n of t h e term " s t r i n g " as used formally i n t h e LSP grammar. t h e elements NVAR and RN (rumors h a s t i l y p r i n t e d ) . The d i f f e r e n c e i n t h e s e two t r e e s shows t h e a m b i g u i t v . i n t h e g i v e n s e n t e n c e .Another parse t r e e of noun p h r a s e :Hearsay and rumors h a s t i l y p r i n t e dLNR I AND Q-CONJ L-- --% 1Hearsay and l a r g e s t p a r t o f t h e grammar, and w i t h o u t them, t e x t p a r s i n g 6 i s o u t of t h e q u e s t i o n . I n a d d i t i o n , w e have found t h a t roughly one t h i r d of a l l t e x t sent e n c e s c o n t a i n c o o r d i n a t e o r comparative c o n j u n c t i o n s , many t i m e s i n complicated i n t e r r e l a t i o n . I t is t h e r e f o r e e s s e n t i a l t h a t t h e r e be a means f o r e x e c u t i n g r e s t r i c t i o n s on s e n t e n c e s c o n t a i n i n g c o n j u n c t i o n s . c e s of code which are used a g a i n and a g a i n . Then t h e s t a c k i n g mechanism can be implemented e f f -i c i e n t l y by changing t h e r o u t i n e s r a t h e r t h a n t h e more numerous r e s t r i c t i o n s . I n t h e LSP cage, t h e r o u t i n e s which w e r e modified are !WAR TI N ------7 I *N J ' V E N P A S S 4 rumors ILVSA LVENR * ----. SA PASSOBJ RV SA t . A _---_I_.-l a ------. . l --A -. . ] *D p r i n t e d h a s t i l y 2. R E S T R I C T I O N S UNDER C O N J U N Z T I O N S ; S T A C K I N G .SA TENSE $8 VERB S A OBJECT ANDSTG RV SA L ------ l -L -L -$ -----A - -. . - --7- -I--- _I They spread rubrs i Y l 8 LAND SA Q-CONJ s SUBJECT S A TENSE SA- VERB SA OBJECT L-.--.- --. -----------A they* p r i n t hearsayFor the sentence shown i n Fig. 5 , They may spread b u t n o t p r i n t the rumors the execution of WSELl i s d i f f e r e n t . The OBJECT ( t h e rumors) h a s no ' conjunct b u t t h e VERB does. I n t h i s c a s e t h e COELEMl$NT(VEZ.B) r o u t i n e goes t o t h e first VERB (spread) from OBJECT and saves t h e second VERB ( p i n t ) or. looking a t t h e second VERB i n ) . Therefore, t h e well-formedness of p i n t rumors is also checked. sequences: heard f a c t s , heard rumors, b u t follow t h e t e n s e p o s i t i o n ( d i d ) .T h i s c o n j u n c t i o n a l occurrence i s covered i n t h e ESP grammar by a node c a l l e d NULLC.The p r e t e r w i l l be " l o o k i n g a t " E2.$STACK-CONJUNCTS = VERIFY ITERATE $STACK-X.$STACK-X = DO $POSTCONJ; STACK.$STACK-CONJUNCTS locates a l l t h e c o n j u n c t s o f t h e node by i t e r a t i n g $STACK-X.I t t h e n r e t u r n s t o t h e s t a r t i n g node. $STACK-X goes t o e a c h conj u n c t by f i r s t e x e c u t i n g $POSTCONJ and t h e n c a l l i n g STACK, t h e o p e r a t o r which p u t s the c o n j u n c t o n t h e r e -e x e c u t i o n s t a c k . I n F i g . 4 , s t a r t i n g a t t h e f i r s t OBJECT, $STACK-TEST w i l l c a l l STACK f o r t h e second OBJECT. i s assumed t o be a t X.I t i s a l s o assumed t h a t t h e r o u t i n e which c a l l e d $STACK-FOR-LEFT-TO-X s t a r t e d a t some node Y , saved Y i n r e g i s t e r X200 and we& from Y l e f t one o r more nodes t o a r r i v e B t X. For i n s t a n c e , t h i s To produce a s y n t a c t i c a n a l y s i s of n a t u r a l language sentences the computer programusestwocomponents: a word dictionary ( F i t z p a t r i c k and Sager 1974) andan English grammar (Sager The p a r s e r a n a l y z e s a s e n t e n c e by b u i l d i n g a p a r s e t r e e f o r the s e n t e n c e . , ETTHER OR , BOTH AND , which permit the logical combina- -- -- -- tion of= ONE OF $AT-ATOM, $DESCEND-TO-ATOM, $DESCEND-TO-STRING.$AT-ATOM = TEST FOR ATOM.= DESCEND TO ATOM NOT PASSING THROUGH ADJSET1.$DESCEND-TO-STRING = DESCEND TO STRING NOT PASSING THROUGH ADJSET1.The CQRE routine locates the sentence word corresponding to a higher level grammatical element E by descending to a terminal node ("atom") from E. This is done by $DESCEND-TO-ATOM. When C O W descends from E it does n o t look at structures which are adjuncts, i . e . , on list ADJSET1. Thus for One rumor hastily printed can ruin careers shown in Fig. A 2 above, the routine CORE, starting a t SUBJECT, w i l l n o t s e a r c h below t h e l e f t -a d j u n c t node LN ( a r r i v i n g m i s t a k e n l y a t o n e ) and w i l l . arrive a t N (the noun rumor). Sometimes t h e s t a r t i n g l o c a t i o n of CORE w i l l be a n a t o m i c node. T h i s is provided f o r by $AT-ATOM, which t e s t s whether t h e c u r r e n t node is an atomic node, i . e . , o n l i s t ATOM. Sometimes a s t r i n g occurs i n a p a r t i c u l a r s e n t e n c e i n place of a noun.I n H i s p r i n t i n g rumors r u i n e d careers, shown i n F i g . Thus CORE s t a r t i n g a t SUBJECT i n OR DO RIGHTR(X) .= ITERATE GO LEFT UNTIL TEST FOR X SUCCEEDS.= ITERATE GO RIGHT UNTIL TEST FOR X SUCCEEDS. G i v e n t h a t X and Y are e l e m e n t s of some s t r i n g , COELEMENT starts a t Y and goes t o X. COELEMENT uses several o t h e r basic r o u t i n e s : ROUTINE LEFTR(X) goes.l e f t from Y u n t i l i t locates X; ROUTINE RIGHTR(X) s e a r c h e s t o t h e r i g h t o f Y to locate X; and combining t h e two, ROtjTINE COELl.(X) s e a r c h e s b o t h s i d e s of Y t o f i n d X. I n F i g .~4 COELEMENT(SUBJECT), s t a r t i n g a t VERB i n ASSERTION, locates SUBJECT by e x e c u t i n g $SAME-LEVEL. ROUTINE LEFTR(SUBJECT) successf u l l y locates SUBJECT, which is to the l e f t o f VERB; t h i s s a t i s f i e s COELl(SUBJECT), which s a t i s f i e s $SAME-LEEL. I f X is i n a s t r i n g segment,COELEMENT w i l l locate it by e x e c u t i n g $X-IN-SEGMENT. I n Fig. A 4 , COELEMENT(OBJECT), s t a r t i n g a t TPOS, f i r s t locates VINGO by executing COELl(STGSEG). I t t h e n locates OBJECT by c a l l i n g r o u t i n e ELEMENT(0BJECT).A different s i t u a t i o n occurs when COEUMENT(TP0S) starts a t OBJECT. T h i s by e x e c u t i n g $ASCNT and t h e n goes t o RN by e x e c u t i n g t h e r o u t i n e RIGHTR(RADJSET). RIGHT-ADJUNCT goes t o VENPRSS by e x e c u t i n g t h e CORE r o u t i n e .Thus t h e p a s s i v e s t r i n g VENPASS ( h a s t i l y p r i n t RV may immediately f o l l o w t h e verb as i n H e r a n q u i c k l y , shown i n ~i g . A S S E R T I O N parse t r e e of H e r a n q u i c k l y . *See Appendix f o r a d e t a i l e d d e s c r i p t i o n of t h e routines'.T h i s w a s done because it i s f a s t e r t o e x e c u t e a n a d d r e s s t h a n a r o u t i n e . | null | null | null | null | Main paper:
summary:
The present paper describes a computational solution to the problem of locating words that are zeroed under conjunction. In this solution, which is based on general properties 3f conjunctional constructions, a mechanism locates zeroed elements in the conjunction strings and cross-references them d t h respect to elements in the h e a d construction. Constraints can then be a p p l i e d to elided conjuncts a s though t h e y w e r e expanded, and transformational expansion, which reconstructs the complete sentences underlying conjunctional occurrences, can be carried out straightforwardly by following the pointers which have been set up.The main innovation is called "stacking". It is a nond.eterministic programming device which causes restrictions ( i . . , subprograms applying detailed constraints) to be reexecuted on conjoined segments whenever the restriction is inovked on an element which has a conjunct. One particularly intricate problem in the computer parsing of natural language texts is the complexity introduced into the parsing system by conjunctions. This complexity is due to the richness of conjunctional constructions and to the material implicit in sentences containing conjunctions, The computafional problem can be divided into three parts: (I) generating parse trees which cover the occurrences of conjunction-headed strings in sentences;(11) locating words in the sentences which are repeated implicitly in a "zeroed," i.e., elided, form in particular positions in conjunction strings;and (III) reconstructing the complete sentences underlying conjunctional occurrences when the application requires such an expansion.A generalized algorithmic solution to problem (I) was provided over a decade ago by the New York University Linguistic String Project (LSP) in the framework of linguistic string analysis (Sager e.t al. 1966 , Sager 1967 , Raze 1 9 6 7 ) , and similar devices have been described more recently, e.g.,by Woods in the framework of augmented transition networks (Woods 1973) . The present paper describes a compukational solution to problem (II), locating the words that are zeroed under conjunction. In this solution, which is based on general properties of conjunctional constructions, a mechanism locates zeroed elements in the conjunction strings and cross-references them with respect to elements in the head construction. Constraints can then be applied to elided conjuncts as though they were expanded, and transformational expansion, which was problem (111) above, can be carried out straightforwardLy by following the pointer's which have been set up.The main innovation is called "stacking." It is a nondeterministic programming device which causes restrictions e . , subprograms applying d e t a i l e d c o n s t r a i n t s ) t o be re-executed on conjoined segments wheqever t h e r e s t r i c t i o n i s invoked on a n element which h a s a c o n j u n c t . S i n c e t h e cons t r a i n t u s u a l l y a p p l i e s t o a combination o f words i n t h e main s t r i n g and i n t h e c o n j u n c t , t h e d e v i c e must e n s u r e t h a t t h e p r o p e r grammatical r e l a t i o n h o l d s between t h e s e wards. That i s , it t r e a t s t h e s e s e p a r a t e d words as a s i n g l e l i n g u i s t i c e n t i t y , o b t a i n i n g t h e e f f e c t of a f u l l expansion o f t h e conj u n c t i o n a l o c c u r r e n c e w i t h o u t c a r r y i n g o u t t h e p h y s i c a l rearrangement and copying or t h e p a r s e t r e e . The s t r a t e g y of t r e a t i n g e l l i p s i s i n t w o steps ( l o c a t i n g d e l e t e d elements and l a t e r c a r r y i n g o u t t h e p h y s i c a l e x p a n s i o n ) is See Appendix f o r a d e f i n i t i o n of t h e term " s t r i n g " as used formally i n t h e LSP grammar. t h e elements NVAR and RN (rumors h a s t i l y p r i n t e d ) . The d i f f e r e n c e i n t h e s e two t r e e s shows t h e a m b i g u i t v . i n t h e g i v e n s e n t e n c e .Another parse t r e e of noun p h r a s e :Hearsay and rumors h a s t i l y p r i n t e dLNR I AND Q-CONJ L-- --% 1Hearsay and l a r g e s t p a r t o f t h e grammar, and w i t h o u t them, t e x t p a r s i n g 6 i s o u t of t h e q u e s t i o n . I n a d d i t i o n , w e have found t h a t roughly one t h i r d of a l l t e x t sent e n c e s c o n t a i n c o o r d i n a t e o r comparative c o n j u n c t i o n s , many t i m e s i n complicated i n t e r r e l a t i o n . I t is t h e r e f o r e e s s e n t i a l t h a t t h e r e be a means f o r e x e c u t i n g r e s t r i c t i o n s on s e n t e n c e s c o n t a i n i n g c o n j u n c t i o n s . c e s of code which are used a g a i n and a g a i n . Then t h e s t a c k i n g mechanism can be implemented e f f -i c i e n t l y by changing t h e r o u t i n e s r a t h e r t h a n t h e more numerous r e s t r i c t i o n s . I n t h e LSP cage, t h e r o u t i n e s which w e r e modified are !WAR TI N ------7 I *N J ' V E N P A S S 4 rumors ILVSA LVENR * ----. SA PASSOBJ RV SA t . A _---_I_.-l a ------. . l --A -. . ] *D p r i n t e d h a s t i l y 2. R E S T R I C T I O N S UNDER C O N J U N Z T I O N S ; S T A C K I N G .SA TENSE $8 VERB S A OBJECT ANDSTG RV SA L ------ l -L -L -$ -----A - -. . - --7- -I--- _I They spread rubrs i Y l 8 LAND SA Q-CONJ s SUBJECT S A TENSE SA- VERB SA OBJECT L-.--.- --. -----------A they* p r i n t hearsayFor the sentence shown i n Fig. 5 , They may spread b u t n o t p r i n t the rumors the execution of WSELl i s d i f f e r e n t . The OBJECT ( t h e rumors) h a s no ' conjunct b u t t h e VERB does. I n t h i s c a s e t h e COELEMl$NT(VEZ.B) r o u t i n e goes t o t h e first VERB (spread) from OBJECT and saves t h e second VERB ( p i n t ) or. looking a t t h e second VERB i n ) . Therefore, t h e well-formedness of p i n t rumors is also checked. sequences: heard f a c t s , heard rumors, b u t follow t h e t e n s e p o s i t i o n ( d i d ) .T h i s c o n j u n c t i o n a l occurrence i s covered i n t h e ESP grammar by a node c a l l e d NULLC.The p r e t e r w i l l be " l o o k i n g a t " E2.$STACK-CONJUNCTS = VERIFY ITERATE $STACK-X.$STACK-X = DO $POSTCONJ; STACK.$STACK-CONJUNCTS locates a l l t h e c o n j u n c t s o f t h e node by i t e r a t i n g $STACK-X.I t t h e n r e t u r n s t o t h e s t a r t i n g node. $STACK-X goes t o e a c h conj u n c t by f i r s t e x e c u t i n g $POSTCONJ and t h e n c a l l i n g STACK, t h e o p e r a t o r which p u t s the c o n j u n c t o n t h e r e -e x e c u t i o n s t a c k . I n F i g . 4 , s t a r t i n g a t t h e f i r s t OBJECT, $STACK-TEST w i l l c a l l STACK f o r t h e second OBJECT. i s assumed t o be a t X.I t i s a l s o assumed t h a t t h e r o u t i n e which c a l l e d $STACK-FOR-LEFT-TO-X s t a r t e d a t some node Y , saved Y i n r e g i s t e r X200 and we& from Y l e f t one o r more nodes t o a r r i v e B t X. For i n s t a n c e , t h i s To produce a s y n t a c t i c a n a l y s i s of n a t u r a l language sentences the computer programusestwocomponents: a word dictionary ( F i t z p a t r i c k and Sager 1974) andan English grammar (Sager The p a r s e r a n a l y z e s a s e n t e n c e by b u i l d i n g a p a r s e t r e e f o r the s e n t e n c e . , ETTHER OR , BOTH AND , which permit the logical combina- -- -- -- tion of= ONE OF $AT-ATOM, $DESCEND-TO-ATOM, $DESCEND-TO-STRING.$AT-ATOM = TEST FOR ATOM.= DESCEND TO ATOM NOT PASSING THROUGH ADJSET1.$DESCEND-TO-STRING = DESCEND TO STRING NOT PASSING THROUGH ADJSET1.The CQRE routine locates the sentence word corresponding to a higher level grammatical element E by descending to a terminal node ("atom") from E. This is done by $DESCEND-TO-ATOM. When C O W descends from E it does n o t look at structures which are adjuncts, i . e . , on list ADJSET1. Thus for One rumor hastily printed can ruin careers shown in Fig. A 2 above, the routine CORE, starting a t SUBJECT, w i l l n o t s e a r c h below t h e l e f t -a d j u n c t node LN ( a r r i v i n g m i s t a k e n l y a t o n e ) and w i l l . arrive a t N (the noun rumor). Sometimes t h e s t a r t i n g l o c a t i o n of CORE w i l l be a n a t o m i c node. T h i s is provided f o r by $AT-ATOM, which t e s t s whether t h e c u r r e n t node is an atomic node, i . e . , o n l i s t ATOM. Sometimes a s t r i n g occurs i n a p a r t i c u l a r s e n t e n c e i n place of a noun.I n H i s p r i n t i n g rumors r u i n e d careers, shown i n F i g . Thus CORE s t a r t i n g a t SUBJECT i n OR DO RIGHTR(X) .= ITERATE GO LEFT UNTIL TEST FOR X SUCCEEDS.= ITERATE GO RIGHT UNTIL TEST FOR X SUCCEEDS. G i v e n t h a t X and Y are e l e m e n t s of some s t r i n g , COELEMENT starts a t Y and goes t o X. COELEMENT uses several o t h e r basic r o u t i n e s : ROUTINE LEFTR(X) goes.l e f t from Y u n t i l i t locates X; ROUTINE RIGHTR(X) s e a r c h e s t o t h e r i g h t o f Y to locate X; and combining t h e two, ROtjTINE COELl.(X) s e a r c h e s b o t h s i d e s of Y t o f i n d X. I n F i g .~4 COELEMENT(SUBJECT), s t a r t i n g a t VERB i n ASSERTION, locates SUBJECT by e x e c u t i n g $SAME-LEVEL. ROUTINE LEFTR(SUBJECT) successf u l l y locates SUBJECT, which is to the l e f t o f VERB; t h i s s a t i s f i e s COELl(SUBJECT), which s a t i s f i e s $SAME-LEEL. I f X is i n a s t r i n g segment,COELEMENT w i l l locate it by e x e c u t i n g $X-IN-SEGMENT. I n Fig. A 4 , COELEMENT(OBJECT), s t a r t i n g a t TPOS, f i r s t locates VINGO by executing COELl(STGSEG). I t t h e n locates OBJECT by c a l l i n g r o u t i n e ELEMENT(0BJECT).A different s i t u a t i o n occurs when COEUMENT(TP0S) starts a t OBJECT. T h i s by e x e c u t i n g $ASCNT and t h e n goes t o RN by e x e c u t i n g t h e r o u t i n e RIGHTR(RADJSET). RIGHT-ADJUNCT goes t o VENPRSS by e x e c u t i n g t h e CORE r o u t i n e .Thus t h e p a s s i v e s t r i n g VENPASS ( h a s t i l y p r i n t RV may immediately f o l l o w t h e verb as i n H e r a n q u i c k l y , shown i n ~i g . A S S E R T I O N parse t r e e of H e r a n q u i c k l y . *See Appendix f o r a d e t a i l e d d e s c r i p t i o n of t h e routines'.T h i s w a s done because it i s f a s t e r t o e x e c u t e a n a d d r e s s t h a n a r o u t i n e .
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 581 | null | null | null | null | null | null | null | null | null |
ad287935905a5155558b2501c368516617ef2905 | 219309098 | null | Multiple Environments Approach to Natural Language | p r o v i d i n g ~1 o n h i s t i c a t e d f a c i l i t i e n fcr-mani p u l a t i n g p o s s i b l e world'' d e s c r i n t i o n s should be one of t h e main concerns i n designing a n a t u r a l language understanding system. The l o g i c a l \\ & notion o f possible world" h a s a c l o s e c o u n t e -r p a ~t i n the computer s c i e n c e n n t i o n o f t h e environment o f expression e v a l u a t i o n . The i?.ea of treating utterance^ a s programs 1 s g e n e r a l i z e d bv a l l o w i n g enui*onmt! h t s~aritohing d u r i n g t h a evalu a t i o n of a n u t t e r a n c e . A model o f natural language, based on m u l t i p l e envi-ronments i n t h e sense j u s t mentioned, i s o u t l i n e d i n terms o f computer science. A rough c l a s s i f i c a t i o n o f environment t y o e s i s given. | {
"name": [
"Bien, Janusz Stanis{\\l}aw"
],
"affiliation": [
null
]
} | null | null | null | 1976-09-01 | 6 | 0 | null | null | Most of t h e examples i n t h e paper a r e d i r e c t q u o t a t i o n s f r o m the referenced literature; herein, some are employed somewhat differently than was their original intent.2. Discourses a _ proarams. | null | 1% is now obvious that the human ability to use language i o re- Longuet-Higgins (1972) ~vho s t a t e s t h a t n a t u r a l l a n g~a g e u tt erances a r e j u s t programs t o be r u n i n sur brains.Some i n t e r e s t i n g analogies between lahguage ~mderstsnding and rumsing a POP-2 program have been shown e. g. i n ( 3 a v i e s , I n a r d 1 9 7 2 ) . I pursue t h i s approach i n a n o t h e r d i r e c t i o n , c h a r o c t e r i s e d by an i n t e n s i v e use o f t h e n o t i o n o f environment, I n t h e e a r l i e r s t a g e o f t h e i n q u i r y , r e p r e s e n t e d by (Bien 1 9 7 5 )~ It should be noted t h a t c o n s i d e r i n g a l l u t t e r a n c e s as a kind of imperative is not a new i d e a f o r l i n g u i s t s ; i t can be found ( 1 ) Be c a r e f u l , he might b i t e you. ( 3 ) France is an interesting c o -m t r~ to study in.where t h e knowledge that t h e Sorbonne is a French u n i v e r s r t y has t o s u p p l y t h e missing link. I n g e n e r a l , a t e x t is s i t u ationally coherent only relative to a given domain of knowledge. by Karttunen (1974:191) who gave the f o l l o w i n g examples:( 4 ) 1 would! like t o i n t r o d u c e you t o my wife. ~vlother had t o C e l l h e r t h a t she had n o t been invlted. ]sac the lo\*;: t p r i m l t -\ C ? i T f 1' V t T , T: c a l l t h e eft'icislwy r u l c . . YCL:; 1: f i c n J t,.kduine, c a n be found in A:dukiewicc ( ) f i l i s i v l~ '-tnSc;7$ h a t if we kave t o choose between t w c i n t e r p r c t n t i o r . . cf R :eW*=--> ;t e c c e , \:F c ' l : z~~e the c t h e r i n t e r p r e t a t L o :~, ( 2 2 . I ) Jack and Bill were outside flying R kite.( 2 2 . 2 ) A s t r o n g wind came by and the string broke.( 2 2 . 3 ) ~a n e t and lice W~T C ' '~u t~i c i + ?t h e house.( 2 2 . 4 ) J a n e t loaked up and s a i d : ( K i p a r s k y 1971 : 3 4 5 ) :( 2 3 ) John r e g r e t s t h a t i t i s raining.( 2 4 ) John thinks ttst it is r a i n i n g . it is r a i n i n g , while tile addre::-ee ' 3 environmer?t r emainx u~ichanged . .lo has been pointed out byIdo~gan (1"\.3)it is not t r u e t h a t non-factual e e n t e n c e a have no p r e s u p p ositions. I t can be e a s i l y seen i n t h e f o l l o w i n g text: ( 2 7 . 2 ) A s t r o n g wind came by and t h e string broke.( 2 7 . 3 ) Janet and A l i c e were o u t s i d e the house.(27.4) When Janet looked up she s a w a k i t e .a k i t e of (27.1) and (27.4) in r e a l i t y r e f e r s t o t h e same ob-j e c t , but in (27.4) it io r e f e r r e d again by means of t h e i n d e fi n i t c noun p h r a c e t o mark J a n e t ' s ignorance about i t , h'e lnriy e a o i l y account for it by e v a l u a t i n g thio noun phrase only i n t h o environment ol, J a n e t , Distinguishing t h e pretense environment from t h e knowledge one is necegsary to handle the cases of lying, e. g.( 3 I ) F r e d is l y i n g when he says he l i l c e s Stanleyc u book.we interprete as meaning In the c a z e o f t h e t e x t (45.2) She p u t t h e p l a t e on the k i t c h e n t a b l e and w e n t . i n t o living room." t h e p l a~e ' ' of (45.2) by i t s e l f refers t o e v e r y p l a t e of the world. ( 5 0 . I ) Jack wanted a k i t t e n . ( 5 0 . 4 ) George was w i l l i n g t o t r a d e s o Jack g o t his k i t t e n .It is i n t e r e a t i n g that t h e SI1RDLU program ( Y / i i n o g r a d~l~7 2 ) t r e a t Obviously, ' the interpretations (57) a n d (58) differ in the way \\ t h e p h r a s e "the friend of Powalskf's brother is split be tween the environments of the knowledge of t h e sender and the ih?owledge of Smith.Mnce we allow designators to switA e n v i r o n m~n t s , we have no problem with so called nouns vvi tll erlpty denotdtion ; they are " ~o h n " cannot i n f l u e a c e its v a l u e . The s t r o n g f e e l i n g t h a t t h e v a l u e o f "he" should I am anxious t o s e e a sentence of this type i n an a u t h e n t i c E n g l i s h t e x t , n o t a s a n example of a r e f e r e n c e problem, because Laboratory ..,,?ma :lIZy:-2'7 3 ,(Yiinograd 1 972) . T e r r y :;'inogred. Undcre tanding Xs t a r a l I o c r~s g e , Edinburgh: Univeroity Press.('Jinogred 1074). Terry Yinograd. F i v e Lectures sn i t i f i c i a lIntelligence Laboratory Xemo XI?: 80.246. S t a n f o r d ' J n l~e~s i t y .(Ziff 1972). Paul Z i f r " . '?/+"at is said. In ( D av i d s o z , Iiarnan1972) pp 709-723.r 3 i )~ ~I J $ * $ Q P E . x c t | null | Main paper:
utterances an anoarams,:
1% is now obvious that the human ability to use language i o re- Longuet-Higgins (1972) ~vho s t a t e s t h a t n a t u r a l l a n g~a g e u tt erances a r e j u s t programs t o be r u n i n sur brains.Some i n t e r e s t i n g analogies between lahguage ~mderstsnding and rumsing a POP-2 program have been shown e. g. i n ( 3 a v i e s , I n a r d 1 9 7 2 ) . I pursue t h i s approach i n a n o t h e r d i r e c t i o n , c h a r o c t e r i s e d by an i n t e n s i v e use o f t h e n o t i o n o f environment, I n t h e e a r l i e r s t a g e o f t h e i n q u i r y , r e p r e s e n t e d by (Bien 1 9 7 5 )~ It should be noted t h a t c o n s i d e r i n g a l l u t t e r a n c e s as a kind of imperative is not a new i d e a f o r l i n g u i s t s ; i t can be found ( 1 ) Be c a r e f u l , he might b i t e you. ( 3 ) France is an interesting c o -m t r~ to study in.where t h e knowledge that t h e Sorbonne is a French u n i v e r s r t y has t o s u p p l y t h e missing link. I n g e n e r a l , a t e x t is s i t u ationally coherent only relative to a given domain of knowledge. by Karttunen (1974:191) who gave the f o l l o w i n g examples:( 4 ) 1 would! like t o i n t r o d u c e you t o my wife. ~vlother had t o C e l l h e r t h a t she had n o t been invlted. ]sac the lo\*;: t p r i m l t -\ C ? i T f 1' V t T , T: c a l l t h e eft'icislwy r u l c . . YCL:; 1: f i c n J t,.kduine, c a n be found in A:dukiewicc ( ) f i l i s i v l~ '-tnSc;7$ h a t if we kave t o choose between t w c i n t e r p r c t n t i o r . . cf R :eW*=--> ;t e c c e , \:F c ' l : z~~e the c t h e r i n t e r p r e t a t L o :~, ( 2 2 . I ) Jack and Bill were outside flying R kite.( 2 2 . 2 ) A s t r o n g wind came by and the string broke.( 2 2 . 3 ) ~a n e t and lice W~T C ' '~u t~i c i + ?t h e house.( 2 2 . 4 ) J a n e t loaked up and s a i d : ( K i p a r s k y 1971 : 3 4 5 ) :( 2 3 ) John r e g r e t s t h a t i t i s raining.( 2 4 ) John thinks ttst it is r a i n i n g . it is r a i n i n g , while tile addre::-ee ' 3 environmer?t r emainx u~ichanged . .lo has been pointed out byIdo~gan (1"\.3)it is not t r u e t h a t non-factual e e n t e n c e a have no p r e s u p p ositions. I t can be e a s i l y seen i n t h e f o l l o w i n g text: ( 2 7 . 2 ) A s t r o n g wind came by and t h e string broke.( 2 7 . 3 ) Janet and A l i c e were o u t s i d e the house.(27.4) When Janet looked up she s a w a k i t e .a k i t e of (27.1) and (27.4) in r e a l i t y r e f e r s t o t h e same ob-j e c t , but in (27.4) it io r e f e r r e d again by means of t h e i n d e fi n i t c noun p h r a c e t o mark J a n e t ' s ignorance about i t , h'e lnriy e a o i l y account for it by e v a l u a t i n g thio noun phrase only i n t h o environment ol, J a n e t , Distinguishing t h e pretense environment from t h e knowledge one is necegsary to handle the cases of lying, e. g.( 3 I ) F r e d is l y i n g when he says he l i l c e s Stanleyc u book.we interprete as meaning In the c a z e o f t h e t e x t (45.2) She p u t t h e p l a t e on the k i t c h e n t a b l e and w e n t . i n t o living room." t h e p l a~e ' ' of (45.2) by i t s e l f refers t o e v e r y p l a t e of the world. ( 5 0 . I ) Jack wanted a k i t t e n . ( 5 0 . 4 ) George was w i l l i n g t o t r a d e s o Jack g o t his k i t t e n .It is i n t e r e a t i n g that t h e SI1RDLU program ( Y / i i n o g r a d~l~7 2 ) t r e a t Obviously, ' the interpretations (57) a n d (58) differ in the way \\ t h e p h r a s e "the friend of Powalskf's brother is split be tween the environments of the knowledge of t h e sender and the ih?owledge of Smith.Mnce we allow designators to switA e n v i r o n m~n t s , we have no problem with so called nouns vvi tll erlpty denotdtion ; they are " ~o h n " cannot i n f l u e a c e its v a l u e . The s t r o n g f e e l i n g t h a t t h e v a l u e o f "he" should I am anxious t o s e e a sentence of this type i n an a u t h e n t i c E n g l i s h t e x t , n o t a s a n example of a r e f e r e n c e problem, because Laboratory ..,,?ma :lIZy:-2'7 3 ,(Yiinograd 1 972) . T e r r y :;'inogred. Undcre tanding Xs t a r a l I o c r~s g e , Edinburgh: Univeroity Press.('Jinogred 1074). Terry Yinograd. F i v e Lectures sn i t i f i c i a lIntelligence Laboratory Xemo XI?: 80.246. S t a n f o r d ' J n l~e~s i t y .(Ziff 1972). Paul Z i f r " . '?/+"at is said. In ( D av i d s o z , Iiarnan1972) pp 709-723.r 3 i )~ ~I J $ * $ Q P E . x c t
:
Most of t h e examples i n t h e paper a r e d i r e c t q u o t a t i o n s f r o m the referenced literature; herein, some are employed somewhat differently than was their original intent.2. Discourses a _ proarams.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 581 | 0 | null | null | null | null | null | null | null | null |
8c28bab501d45d92ab4e73f6eb39d7516b66c517 | 219302192 | null | Book Review: \textit{ {E}ssays on {L}exical {S}emantics}, Vol. {I}, Edited by {V}. {J}u. {R}ozencvejg | BOOK REV1 EW Sprakforlaget Skriptox 8-104 65 Stockholm iv -k 404 pages; SKZ 60 | {
"name": [
"Gumb, Raymond D."
],
"affiliation": [
null
]
} | null | null | null | 1976-09-01 | 0 | 0 | null | null | null | null | t h e i n f ; e r l l n~l l a , t h e s o u r r p l a n n~a~e s r a t e l l c e a n dt~r~r t langparre seoterce have +he snwe * e~r i v = . The ( R P R~~S~ r ) Aict i o r a q entry f o r a word, whlch is taltcr. to he either the ram of an indlvidllal or a one-o r two-pMae aredicate (thro~-p7ere nredicates beinv defined, by arrpeel l n~ t o what amo~mt s to the r o t i o n o f 'causatlvityf, ? n teres of t w o -p l a c~ ~r~c q i c a t r s ) , covsistd of a i t s t of srssnt1c fact~rs. (Rcwrvrr, t.be rV1oaonhical problem of d e f i h i n p lhypnry:~al 9f a c e r f a i r , c l n s s , snecif ical4 y -nredicates a n r l y i r p t o t h e determirates of a d etermlaehle, whlch leads Fodor and Patz t o i r l t r o +~l c~ t h e v a c n w s I n o t j or of demantic d i s t t * m i i shersl , i s ivtractable u s i v~ a 'list o r colrjunctian of semantic factt3rs. ) Ififormat on abovt cases and ren nos it ions e n c~d e d in t h e dlc~ionarv d e t e r~i~e~ t h e nlace L --.-f o r an armment of a ~rsdicate. On a~a l o p g with de~endehcy prammars in syntax, the u e a n i n p of a source l a r p a p e exnrbssinn' i s initially b u i l t UD as a t p~e of reorznt jr f ' n c t q r P and evoctuelly hwomes, ! l -~~t q k a~r q l~ 7 4 &.' -TI,?,! 2h ~0 r r : B i p t ! | null | Main paper:
:
t h e i n f ; e r l l n~l l a , t h e s o u r r p l a n n~a~e s r a t e l l c e a n dt~r~r t langparre seoterce have +he snwe * e~r i v = . The ( R P R~~S~ r ) Aict i o r a q entry f o r a word, whlch is taltcr. to he either the ram of an indlvidllal or a one-o r two-pMae aredicate (thro~-p7ere nredicates beinv defined, by arrpeel l n~ t o what amo~mt s to the r o t i o n o f 'causatlvityf, ? n teres of t w o -p l a c~ ~r~c q i c a t r s ) , covsistd of a i t s t of srssnt1c fact~rs. (Rcwrvrr, t.be rV1oaonhical problem of d e f i h i n p lhypnry:~al 9f a c e r f a i r , c l n s s , snecif ical4 y -nredicates a n r l y i r p t o t h e determirates of a d etermlaehle, whlch leads Fodor and Patz t o i r l t r o +~l c~ t h e v a c n w s I n o t j or of demantic d i s t t * m i i shersl , i s ivtractable u s i v~ a 'list o r colrjunctian of semantic factt3rs. ) Ififormat on abovt cases and ren nos it ions e n c~d e d in t h e dlc~ionarv d e t e r~i~e~ t h e nlace L --.-f o r an armment of a ~rsdicate. On a~a l o p g with de~endehcy prammars in syntax, the u e a n i n p of a source l a r p a p e exnrbssinn' i s initially b u i l t UD as a t p~e of reorznt jr f ' n c t q r P and evoctuelly hwomes, ! l -~~t q k a~r q l~ 7 4 &.' -TI,?,! 2h ~0 r r : B i p t !
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 581 | 0 | null | null | null | null | null | null | null | null |
d2b7a10bc9d781d73d4069cb4b92f90fcdb9c235 | 219304825 | null | Review: \textit{ {D}escription {G}rammaticale du {P}arler de l'{I}le-aux-{C}oudres, {Q}uebec}, by {E}mile {S}eutin | f Kansas Lawrence 66045 S e u t i n has succeeded in c o n d e n s i n g h i s 600-page d i s s e r t ation into a useful grammatical description o f t h e F r e n c h spoken on t h e Ile-aux-C0udre.s (IAC) in the St: Lawrence. Structurali s t and rigidly descriptive., it l i s ~s forms, frequenqies of occurrence, and variants, with few and isnlated e x p l a n a t o r y comments. Desplee a discussion in the introduction ro the second part o f the need for a d i f f e r e n t approach to syntax, t h e second like tne f i x s t p~rt--consists of a straighc presenta-. t f o n . o f d a t a i n n a structuralist~framework The ~n l y harm in this is t h a t S e u t i n seems t o be t r y i n g t o do something d i f f er e n t . There is no need. the data gathered are significant in themselves and are presented in suc-h a way as to be useful to | {
"name": [
"Dinneen, David A."
],
"affiliation": [
null
]
} | null | null | null | 1976-12-01 | 0 | 0 | null | null | null | null | researcners in d i a l e c t study,.histbrical and comparative ling'uisrirs, sociolinguistic8 and other subdisciplines.seutin not only d e s c r k b e b 'the, m~rphology~ and syntax. of IACbu$ h e cofipaoes the usage in the' isLqnd with that of standard French as, described in F r a n o a i s Fondamental (-FF) He recogn k e s and agcounks for the diff-erences between each corpus and avoids making generalizatidns when the two cannot be reasonably compared. Hb d o e s , however, seem to forget from time to time that FF is not as "colloquial" as his corpus, *even though he mentions more than once the need for a current description of "familiar" French.The work i s a good e x a m p l e of the use of the computer f o r recording and scoring grammatical and lexical data and, more interesting, for searching and analyzing the corpus and comparing data from other sources. Little description is given of the program but the results indicate that Seutin and his colleagues were able to handle a very large corpus and extract from it the data in which they were interested. It would appear that most of the syntactic analysis was done by hand, but once forms were encoded, the program could f i n d all examples of each structure being studied and prirlt them out in a usable format. Since, as in most concordances, forms are (can be)printed out in conzext, including an i n d i c a t o r of the speakerr, the same data could and should be used in the f~t u r r for many different kinds of studies. | null | Main paper:
:
researcners in d i a l e c t study,.histbrical and comparative ling'uisrirs, sociolinguistic8 and other subdisciplines.seutin not only d e s c r k b e b 'the, m~rphology~ and syntax. of IACbu$ h e cofipaoes the usage in the' isLqnd with that of standard French as, described in F r a n o a i s Fondamental (-FF) He recogn k e s and agcounks for the diff-erences between each corpus and avoids making generalizatidns when the two cannot be reasonably compared. Hb d o e s , however, seem to forget from time to time that FF is not as "colloquial" as his corpus, *even though he mentions more than once the need for a current description of "familiar" French.The work i s a good e x a m p l e of the use of the computer f o r recording and scoring grammatical and lexical data and, more interesting, for searching and analyzing the corpus and comparing data from other sources. Little description is given of the program but the results indicate that Seutin and his colleagues were able to handle a very large corpus and extract from it the data in which they were interested. It would appear that most of the syntactic analysis was done by hand, but once forms were encoded, the program could f i n d all examples of each structure being studied and prirlt them out in a usable format. Since, as in most concordances, forms are (can be)printed out in conzext, including an i n d i c a t o r of the speakerr, the same data could and should be used in the f~t u r r for many different kinds of studies.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 578 | 0 | null | null | null | null | null | null | null | null |
52e3ff1960f026007bb131b711ce4f095c2b54eb | 219303847 | null | Review: \textit{ {R}epresentation and {U}nderstanding: {S}tudies in {C}omputer {S}cience}, Edited by {D}aniel {G}. {B}obrow and {A}llan {C}ollins | A n ~a j o r g o a l o f A ) -t i f i c l a l ' T n t c l l i g c n c c r c s r a l -c h t o d a y is t o d c s i g n systcms t h a t " u n d c r s t n n d " p hocly o f h n o v l c d g c , i . c . | {
"name": [
"Mylopoulos, John"
],
"affiliation": [
null
]
} | null | null | null | 1976-12-01 | 1 | 0 | null | null | null | null | null | c o l l c c t i v c l y r a t h e r tllan i n .d i v i d u a l l y , w c i~r r i v c d a t a slightly L .& J P w +-r 'La c L G C k s w .r , L . G f m ... - 5 -I '5 L - 5 G rc. . . . *f f C .- t . C - , , E: * d s C* * yr r , C, c L 0 C C * $a *F F.1 7 . S c h a n k , R . ''Ilsjng Knowlcdgc t o I J n d e r~t a n d~~ TINLAP rrocccd i n g s pp. 1 1 7 -1 2 1 , J u n e 3 9 7 5 . | Main paper:
in a t t e m p t i n g t o rcvicw t h c pilpcrs t h a t nppcnr i n this book:
c o l l c c t i v c l y r a t h e r tllan i n .d i v i d u a l l y , w c i~r r i v c d a t a slightly L .& J P w +-r 'La c L G C k s w .r , L . G f m ... - 5 -I '5 L - 5 G rc. . . . *f f C .- t . C - , , E: * d s C* * yr r , C, c L 0 C C * $a *F F.1 7 . S c h a n k , R . ''Ilsjng Knowlcdgc t o I J n d e r~t a n d~~ TINLAP rrocccd i n g s pp. 1 1 7 -1 2 1 , J u n e 3 9 7 5 .
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 578 | 0 | null | null | null | null | null | null | null | null |
79e1cc39acfc3936e666dc8fd3a712dd0cdb6872 | 219303767 | null | {SNOPAR}: A Grammar Testing System | A grammar testing program has been developed which permits modeling augmented transition network grammars as a series of SNOBOL4 functions. SNOPAR is designed for lknguistics teaching and research. Emphasis is placed or1 the development of small to medium grammars in a variety of languages. The system has been used so far to develop a grammar of English for use transformational grammar course and develop small grammars of a Nigerian and an American Indian language. Intended applications of SNOPAR are in fi.' & linguistics and grammar model testing. The main part of the program is the routine PARSER. When PARSER is c'allbd with a lexicon and grammar, input* strings are parsed according to the model grammar. The PARSER functions available for grammar developmerlt are CAT, PARSE, SETR, GETR, RESET, TESTR, GETF, GETCL, TO, BACK, FINDWRD, and BUILDS. The function operations and descriptions of their argumerits are given in Table 1. After a parsing, PARSER returns' control to the user permitting examination of stacks and registers at all | {
"name": [
"Kehler, T. P."
],
"affiliation": [
null
]
} | null | null | null | 1976-12-01 | 0 | 2 | null | null | W A ) 1 E N G L I S H : DCl3 M Y B I G A P L U R D O Y O U W A N T TO E X A M I N E T E E R E G I S T E R S ? NO I N P U T S T R U C T U R E TO B E P A R S E D AS F A N A B I J I Y C E M W A AS F A N A B L J I M C E KWA S T A T E ? O SC O M P L E M E N T S T R I N G : C E B U I L D S T R U C T U R E B J J I H S T A T E P L T C O M P L E M E N T S T R I N G : B u r m S T~H J C T Y R Z ~W A S T A T E N P C O M P L E M E N T S T R I N G : D I D N O T P A R S E B U I L D S T R U C T U R E M'dA D O Y O U W A N T TO E X A M I N E T H E R E G I S T E R S ? N O 1 N P U T . S T R U C T U R E T O BE P A R S E D AS WdA A S MWA S T A T E P O S C O M P L E . M E~T I S T R I N G : M W A B U I L D S T R C~C T U R E ' A S S T A T E K T COMPLEMENT S T R e I N G : M W A B U I L D S T R U C T U a E S T A T E A D J C O N P L E M E N T STRING: MrdA B U X L C S T R U C T U R E S T A T E K O M C O M P L E M E f i T S T R I N G : pl !A' B U I L D S T R U C T U R E S T A T E DET G O X P L E M G N T S T R I N G : M W A B U I L D , S T R U C T U R E S T A T E P L C O M P L E Y E N T S T S I N G : M W A B U I L D S T R U C T U R E S T A T E F L T C O M P L E Y E N T S T R I N G : B U I L D S T R U C T U R E M W A S T A T E P O P U P C O M P L E X E Y T S T R I N G : B U I L D S T F U C T U f i E Y W A ST A'-?' E Y C O X P L E 3 , F N T S T R I N G : B U I L D . " s T R i l C W T U F E P C AS) ( P L M X A ) ) E N G L I S H : , D O G P L U R D O Y O U W A N T T O E X A N Y E T H E R E G I S T E R S ? u n Example 3 F L~v C T I O N DEFINITIONS G,RAP C E F I N E ( " s t ] N v ) C E F I K F ( ' E S O p ) CEFJhW('NP[)PINw) CEFIREtPPP()') G F I F I W [ p ' V P ( ) N * )A XQ f i PP f i P S E ( V F ( 3 )PO P s t ; E = B U I L D S ( * / r J P / !~P P / P O S S . / @ 4 tETo(,NTPP)3 CPT("AD3'): ( I S~T E S T~ T TYPE^, ~D C L . ) P S S E T ( @ T Y P E~, . T F ? P A~~~ - 1 1 T'r'PE L . 1 I , UE 1 I.I~F-1-PF 0 1 1 1 I I FFED 1 $ ' 1 T i d PF'E' 4 1 I *; T l~lRt{T-& - I I { z I T '~' P E '~I C L I I -l-IE,_I I~~P I~F '~J I I 1 I IEI'~:'F 1$'Ft1t,! 13n)Ff?E.Z *I .#*. DGLJ 4:KE:J I I I F '~F ' F ' O I t 1 . 1 tF'PED f V P 1 T I i I P H f T l r v T IHIIII tnBJ ~.~{P;PPD 41'UtJl 1 1PF.EP IIIITHI~{F'I, FFIO HEPl 1 1 i 1 1 1 1 1 1 ) 1 1 r t w r STRUCTURE T O BE PARSED , J 0 F l i l m S 1 :~L -I E V I N G THAT tIfIR'I'.iOl-It4 " S E:IZLIEt?II.IG THAT MARY I S GCIII4G 'TO THE VILL-AGE JS M Y S TEriIOUS SIf'lTL: S CL7i rl"I, Eflr3i.4T ST133:PICi : M Y S T F r i 1 OUS 1IU:l LEI .I; TI\LJCTUI'<E :( SL f YPC rrcl, i ! C i~~~ ( I F (E.lrK LICII-li\,! S ( I='OSs ( VF'( TfiS F.'r;CS ( VT r : E L I c v E ) ( l3E.i' (NI'" ( COril ' ( 3 i ! f C.1, i ( SLJD J, ( (141 '( NI''~.' Mr;TiY ) j ( F'FiED ( VI-' ( k U X I{E ) 1 I V G 0 9 i l F t F i E f ' I I V I L L , ,^. G 1 1 ) ) ) ) ) 1 ) ) $ ) ) ) ( P R E D ( VI'' ( V EE i ( T N S T'FiE.2 i ChLI,J M' I' S T'ERIOUG) ) 1 ) DO YCSU WAIdT 7 0 IIYkMTNE TI-IE REGISTERS ? NO IPIF'UT STRUCTUI7T: TO BIZ F'ARSED T l i A J tIE, EFiOliE tlER 11 751-1 IS SERIOUS T l ' l~l I--IE I I I 1'5 SEIXIUUS STATE S Ct3MF'L.t:Ml11.I T S f R T N G : S E R I O U S L~UILTr S T R U C T U R E t (S(TYT'E DCI,) (SbKtJ* (NF'(CO45F' (S(TYF'E KlCL) ( S U E J (NF1(I-"TiO H E ) ) ) ( F R E D ( VT-' t T14S P h S T ) ( u f T:I?EI^II.\ ) C 0E.J ( T413 ( P R O I !El7 1 i N, L I I S I I) 1 ) 1 ) ) z FRED ( VF'( v f i~ r~i s FIRES i ~,LI.J S E R~O U S 1 I 1 DO Y O U WANT T O E X A h I N E Ti4E REGISTERS ? ?\I0 - II!F'UT STRUCTURE TO BF F'ARSED TI-IE EO'i' ?!L'E(I~ I t.IZ ' T I-If.: f;L(.52S I S MLJLLIGAN Ti-IE F:OY I:HECII,I~.IG TI-IE G L A S S 1S M U L L I G A N STATE S CO~-~F'LEH~ZP~T STR J N G : MULLIGAN EU-I LK1 $7 r.:UCTURE t ( 5 ( 7 YIZ'C DCI, 1 ( SUBJ C i t T I t EjDY 1 ( EMP ((IF' ( T N S F'F'RG ( VT EREAIi) (,OBJ u w ; ( Z s C r I ~I -A S S ) ) ;~~~) ( F ' K E D (LIF'(V BEHTNS FRES) ( flTNT' i PJ!:# ( N r l R M l l L 1-1 GAf4 j ) 1 1 ) 1 D O YOU w n t u T Q Eitnrm<c I REGISTERS ? NO TP!F'UT STRUCTURE T O h E PARSED JCltlrdD S IiEIt!G T t i I N ' I S N I C E JUIIPJQ4; K:EING TI-IIIJ I S N I C E STATE ' 3 COP~F'LEI.II:I~T S T R I N G : N I C E H U l L L I 43-1 RUCT'UliE : (E:<TYFIF, DCI,itSUI!J !f4F1(tJF'R J O I i N m S i * ( P O S S i L I F ' ( U -E E ) ( T N S F'F'KG) ( A B J 7 H l l l ) j ) ) :~a ( F ' I ; ' C L I i l : E ) I T N S I-'li'CS)(kLIJ P I I C E ) ) ) )110 YOU W A N T T U EXAMINE THE I?'I,GISTEI?S '3 NQ J NFUT STF:LJCTLIr*:C: 7 CI CL":'*AKSTD COrlPLEMENT STRING: JOHN BUILrl S TI?UCTUI7E *' (S(TYF'E UCI-1 (SUGJ OEP(DET TI-IE) (id D o Y ) (EMB (Ul='iU RUN) (TNS F'F'RGI (UFp (I"RI,r-' T CI (Ef1'-C DE r THE ) ( N I-IUUSE > > > )'> j ) > i P I E D ( VF' ( U BE) V PE ) J I j J r I ( I t I t h j ( E4 MAFI 1 i 1 j 1 ) 1 i ) ( l'*l<L I1 ( VI-' ( V BE ) | null | null | looks up the word class of t h e c u r r e n t first word.in the.input string. If t h e word i s not in t h e lexicon an add r o u t i n e is called which permits additions. If CAT succeeds by matching the current word class with its argument, the word is removed f r a m the input string and pushed ont-o a staclc (SAVEW) . sets the values of registers. It has three arguments level, register name, and value. Each cavil ~f SETR causes the r e g i s t e r name s p e c i f i e d to be placed on -3 list for the s p e c i f i e d level. SETR e n r r i e s are treated as stacks, providing automatic saves for recursi-ve calls.returns the contents of the register name specified by its argument, and pops it o f k the stack saving the last value.looks at the value of the register name specified by its argument without popping it off the stack.changes the vaLue of a register without changing s t a c k Levels.looks up the f e a t u r e v a l u e for a f e a t u r e speciFiied by i t s argurnebt of the current value of the word. register. Any word can be s p e c i f -t e d by giving a second aqpment. Tf GETF f a i l s for the word it looks at the root form of the h~ic.ri c a n I n d i a n 1 nn):u;ij:cs) I and has trace capabilities fsr grammar debugging. S = BUILDS(S) s { R E T U R N ) C A T ( ' D E T ' ).t (-TO(.P O P N P ) ) S E T R ( .NS, ' D E -T * , Q ) C A T ( * A D J * ) a .S(TO(.TRYPP)) N P = B U I : L D S ( N P ) $ ( R E T U R N ) CAT( -P R E P ) t F ( @ R E T U R N )C A T ( 'V * ) t F~F ' R E T U R N ) SETRC .VP, ' vV P = BUILDS(VP) [ R E T U R N ) TY L E X E N G . 1.C A N = ( A U X ) ( T N S P ' R E S ' ) . C O U L D = (EOR!/l ' C A : : ) . W I L L = ( A U X ) ( T t l S F U T ) .I.T H A T = ( C L I ! ! l ) ) . B O Y = (tl)(!JUP : ; L . t i f ; ) . B O Y S = ( N ) ( N b S P L ) . . G I R L = (N)(tl9!{ L I N G ) . G L R L S -(E'0ItE.I I ) 1 P L ) . M A N = ( N ) ( t . l n r i S L l i G ) . M E N = ( N ) ( t ! U R P L ) ..W O H E N = (N) (1132 P I , ) . T A B L E = ( N ) (!:!3R S f !lC) . )ID Y O U W A L K T O T H E V I L L A G E D L D Y O 2 W A L K T O T H E V I L L A G E S T A T E C U E S COF?PLEuZZdT STRING:' YOU U A L K TO T H E V I L L A G E B U I L D S T S U C T 3 R E DID S T A T E P F O C O M P L E Y E Y T S T R I N G a W A L K ~d T H E V f L L A C E B U I L D S T R U C T U F E Y O U STATE P O P N P COMPLEEE!;T S T R i N C ; W A L K T O THE V I L L A G E B U l L P S T R U C T U R E Y O U S T A T E QEiP C O M P L E X E A T S T R i ? : G ; b X A L X T O T H E V I L L A G E BUILD S T R U C T U R E ( N F ( P R O YOU)^ S T A T E T R Y V P COMPLEE'.ENT. S T R L N G : W A L K T O THE VC O M P L E X E K T S T R i N G t V I L L A G E B U I L D S T R U C T U R E STATE P O P N P C O M P L E 3 S S T S T R I N G : B U I L D S T R K T U R E V I L L A G ET H E ) ( N V I L L A G E ) ) ) ) STATE P O P V P C O M P L E M E A T S T R I N G :T ; i E ) ( N VILLAGE)))) S T A T E P O P S C O M P L E Y E N T S,Ti?I N G : B U L L D S S R U C T U R E : ( V P ( VSTATE S COMPLEME?iT S T R I N G 8 B U I L D S T R U C T U R E r (S(TYPE Q 3 E S T I I : O ) ) A A X D D D D J T E E I S E P A S T ) ( S ' J & J ( N P ( P R 0 Y O U ) ) ) ( P R E D ( V -P ( V WALK) ( P P ( P a P ( P R E P T O ) ( P R E P K P ( N P ( D E T T H E ) ( N V I L L A G E ) ) ) ) ) ) ) ) DO Y-OU WANT T O E X A K I N E T H E R E G I S T E R S ? YES IIE P \ P \ O F P C P s L<'RIZTO> = ' ( A D J ) ( P L -P L l ) ( E N C G O O D ) ' L < * R~~' J ' -F f I J i T O > = ' ( A D J ) (P'L P L ) (EMG G O O D ) ' L < * B I J I M * > = [ A D J ) ( P L -PL) ( E N G N I f 3 ) ' e < '~~' t j -W A N *, = *(ADJ,) ( P L P L ) ( E~. I S P I C > * L < ' G A K . > ( J E N O N E ) * L < ' U A P . $ = ( I t i T W O ) L < ' N Y I I * > = . ( I ) F T ) ( I . , N C ' I ' I l J~~)~ L < '~J A ' > = * ( D l~T ) ( b , t l~~ T t ! b ; ) - t c ' c~; ' , = ' ( D E T ) (E!;'; A ) ' L < * M W A -3 = ' 1 ) ( t i E J L~J P ) * L < ' R u L U . ' ; * ? = ( 1 1 ( i "AMRES T A T E 3 . P | Main paper:
c o y p l e e e n t s t r i n g : b u i l d s t r u c t u r e ( m p ( n o l t n a s . ) ( p o s p r q f a n a h a d j n a n -? : a~j ) ( d e t c e h p l:
W A ) 1 E N G L I S H : DCl3 M Y B I G A P L U R D O Y O U W A N T TO E X A M I N E T E E R E G I S T E R S ? NO I N P U T S T R U C T U R E TO B E P A R S E D AS F A N A B I J I Y C E M W A AS F A N A B L J I M C E KWA S T A T E ? O SC O M P L E M E N T S T R I N G : C E B U I L D S T R U C T U R E B J J I H S T A T E P L T C O M P L E M E N T S T R I N G : B u r m S T~H J C T Y R Z ~W A S T A T E N P C O M P L E M E N T S T R I N G : D I D N O T P A R S E B U I L D S T R U C T U R E M'dA D O Y O U W A N T TO E X A M I N E T H E R E G I S T E R S ? N O 1 N P U T . S T R U C T U R E T O BE P A R S E D AS WdA A S MWA S T A T E P O S C O M P L E . M E~T I S T R I N G : M W A B U I L D S T R C~C T U R E ' A S S T A T E K T COMPLEMENT S T R e I N G : M W A B U I L D S T R U C T U a E S T A T E A D J C O N P L E M E N T STRING: MrdA B U X L C S T R U C T U R E S T A T E K O M C O M P L E M E f i T S T R I N G : pl !A' B U I L D S T R U C T U R E S T A T E DET G O X P L E M G N T S T R I N G : M W A B U I L D , S T R U C T U R E S T A T E P L C O M P L E Y E N T S T S I N G : M W A B U I L D S T R U C T U R E S T A T E F L T C O M P L E Y E N T S T R I N G : B U I L D S T R U C T U R E M W A S T A T E P O P U P C O M P L E X E Y T S T R I N G : B U I L D S T F U C T U f i E Y W A ST A'-?' E Y C O X P L E 3 , F N T S T R I N G : B U I L D . " s T R i l C W T U F E P C AS) ( P L M X A ) ) E N G L I S H : , D O G P L U R D O Y O U W A N T T O E X A N Y E T H E R E G I S T E R S ? u n Example 3 F L~v C T I O N DEFINITIONS G,RAP C E F I N E ( " s t ] N v ) C E F I K F ( ' E S O p ) CEFJhW('NP[)PINw) CEFIREtPPP()') G F I F I W [ p ' V P ( ) N * )A XQ f i PP f i P S E ( V F ( 3 )PO P s t ; E = B U I L D S ( * / r J P / !~P P / P O S S . / @ 4 tETo(,NTPP)3 CPT("AD3'): ( I S~T E S T~ T TYPE^, ~D C L . ) P S S E T ( @ T Y P E~, . T F ? P A~~~ - 1 1 T'r'PE L . 1 I , UE 1 I.I~F-1-PF 0 1 1 1 I I FFED 1 $ ' 1 T i d PF'E' 4 1 I *; T l~lRt{T-& - I I { z I T '~' P E '~I C L I I -l-IE,_I I~~P I~F '~J I I 1 I IEI'~:'F 1$'Ft1t,! 13n)Ff?E.Z *I .#*. DGLJ 4:KE:J I I I F '~F ' F ' O I t 1 . 1 tF'PED f V P 1 T I i I P H f T l r v T IHIIII tnBJ ~.~{P;PPD 41'UtJl 1 1PF.EP IIIITHI~{F'I, FFIO HEPl 1 1 i 1 1 1 1 1 1 ) 1 1 r t w r STRUCTURE T O BE PARSED , J 0 F l i l m S 1 :~L -I E V I N G THAT tIfIR'I'.iOl-It4 " S E:IZLIEt?II.IG THAT MARY I S GCIII4G 'TO THE VILL-AGE JS M Y S TEriIOUS SIf'lTL: S CL7i rl"I, Eflr3i.4T ST133:PICi : M Y S T F r i 1 OUS 1IU:l LEI .I; TI\LJCTUI'<E :( SL f YPC rrcl, i ! C i~~~ ( I F (E.lrK LICII-li\,! S ( I='OSs ( VF'( TfiS F.'r;CS ( VT r : E L I c v E ) ( l3E.i' (NI'" ( COril ' ( 3 i ! f C.1, i ( SLJD J, ( (141 '( NI''~.' Mr;TiY ) j ( F'FiED ( VI-' ( k U X I{E ) 1 I V G 0 9 i l F t F i E f ' I I V I L L , ,^. G 1 1 ) ) ) ) ) 1 ) ) $ ) ) ) ( P R E D ( VI'' ( V EE i ( T N S T'FiE.2 i ChLI,J M' I' S T'ERIOUG) ) 1 ) DO YCSU WAIdT 7 0 IIYkMTNE TI-IE REGISTERS ? NO IPIF'UT STRUCTUI7T: TO BIZ F'ARSED T l i A J tIE, EFiOliE tlER 11 751-1 IS SERIOUS T l ' l~l I--IE I I I 1'5 SEIXIUUS STATE S Ct3MF'L.t:Ml11.I T S f R T N G : S E R I O U S L~UILTr S T R U C T U R E t (S(TYT'E DCI,) (SbKtJ* (NF'(CO45F' (S(TYF'E KlCL) ( S U E J (NF1(I-"TiO H E ) ) ) ( F R E D ( VT-' t T14S P h S T ) ( u f T:I?EI^II.\ ) C 0E.J ( T413 ( P R O I !El7 1 i N, L I I S I I) 1 ) 1 ) ) z FRED ( VF'( v f i~ r~i s FIRES i ~,LI.J S E R~O U S 1 I 1 DO Y O U WANT T O E X A h I N E Ti4E REGISTERS ? ?\I0 - II!F'UT STRUCTURE TO BF F'ARSED TI-IE EO'i' ?!L'E(I~ I t.IZ ' T I-If.: f;L(.52S I S MLJLLIGAN Ti-IE F:OY I:HECII,I~.IG TI-IE G L A S S 1S M U L L I G A N STATE S CO~-~F'LEH~ZP~T STR J N G : MULLIGAN EU-I LK1 $7 r.:UCTURE t ( 5 ( 7 YIZ'C DCI, 1 ( SUBJ C i t T I t EjDY 1 ( EMP ((IF' ( T N S F'F'RG ( VT EREAIi) (,OBJ u w ; ( Z s C r I ~I -A S S ) ) ;~~~) ( F ' K E D (LIF'(V BEHTNS FRES) ( flTNT' i PJ!:# ( N r l R M l l L 1-1 GAf4 j ) 1 1 ) 1 D O YOU w n t u T Q Eitnrm<c I REGISTERS ? NO TP!F'UT STRUCTURE T O h E PARSED JCltlrdD S IiEIt!G T t i I N ' I S N I C E JUIIPJQ4; K:EING TI-IIIJ I S N I C E STATE ' 3 COP~F'LEI.II:I~T S T R I N G : N I C E H U l L L I 43-1 RUCT'UliE : (E:<TYFIF, DCI,itSUI!J !f4F1(tJF'R J O I i N m S i * ( P O S S i L I F ' ( U -E E ) ( T N S F'F'KG) ( A B J 7 H l l l ) j ) ) :~a ( F ' I ; ' C L I i l : E ) I T N S I-'li'CS)(kLIJ P I I C E ) ) ) )110 YOU W A N T T U EXAMINE THE I?'I,GISTEI?S '3 NQ J NFUT STF:LJCTLIr*:C: 7 CI CL":'*AKSTD COrlPLEMENT STRING: JOHN BUILrl S TI?UCTUI7E *' (S(TYF'E UCI-1 (SUGJ OEP(DET TI-IE) (id D o Y ) (EMB (Ul='iU RUN) (TNS F'F'RGI (UFp (I"RI,r-' T CI (Ef1'-C DE r THE ) ( N I-IUUSE > > > )'> j ) > i P I E D ( VF' ( U BE) V PE ) J I j J r I ( I t I t h j ( E4 MAFI 1 i 1 j 1 ) 1 i ) ( l'*l<L I1 ( VI-' ( V BE )
back:
looks up the word class of t h e c u r r e n t first word.in the.input string. If t h e word i s not in t h e lexicon an add r o u t i n e is called which permits additions. If CAT succeeds by matching the current word class with its argument, the word is removed f r a m the input string and pushed ont-o a staclc (SAVEW) . sets the values of registers. It has three arguments level, register name, and value. Each cavil ~f SETR causes the r e g i s t e r name s p e c i f i e d to be placed on -3 list for the s p e c i f i e d level. SETR e n r r i e s are treated as stacks, providing automatic saves for recursi-ve calls.returns the contents of the register name specified by its argument, and pops it o f k the stack saving the last value.looks at the value of the register name specified by its argument without popping it off the stack.changes the vaLue of a register without changing s t a c k Levels.looks up the f e a t u r e v a l u e for a f e a t u r e speciFiied by i t s argurnebt of the current value of the word. register. Any word can be s p e c i f -t e d by giving a second aqpment. Tf GETF f a i l s for the word it looks at the root form of the h~ic.ri c a n I n d i a n 1 nn):u;ij:cs) I and has trace capabilities fsr grammar debugging. S = BUILDS(S) s { R E T U R N ) C A T ( ' D E T ' ).t (-TO(.P O P N P ) ) S E T R ( .NS, ' D E -T * , Q ) C A T ( * A D J * ) a .S(TO(.TRYPP)) N P = B U I : L D S ( N P ) $ ( R E T U R N ) CAT( -P R E P ) t F ( @ R E T U R N )C A T ( 'V * ) t F~F ' R E T U R N ) SETRC .VP, ' vV P = BUILDS(VP) [ R E T U R N ) TY L E X E N G . 1.C A N = ( A U X ) ( T N S P ' R E S ' ) . C O U L D = (EOR!/l ' C A : : ) . W I L L = ( A U X ) ( T t l S F U T ) .I.T H A T = ( C L I ! ! l ) ) . B O Y = (tl)(!JUP : ; L . t i f ; ) . B O Y S = ( N ) ( N b S P L ) . . G I R L = (N)(tl9!{ L I N G ) . G L R L S -(E'0ItE.I I ) 1 P L ) . M A N = ( N ) ( t . l n r i S L l i G ) . M E N = ( N ) ( t ! U R P L ) ..W O H E N = (N) (1132 P I , ) . T A B L E = ( N ) (!:!3R S f !lC) . )ID Y O U W A L K T O T H E V I L L A G E D L D Y O 2 W A L K T O T H E V I L L A G E S T A T E C U E S COF?PLEuZZdT STRING:' YOU U A L K TO T H E V I L L A G E B U I L D S T S U C T 3 R E DID S T A T E P F O C O M P L E Y E Y T S T R I N G a W A L K ~d T H E V f L L A C E B U I L D S T R U C T U F E Y O U STATE P O P N P COMPLEEE!;T S T R i N C ; W A L K T O THE V I L L A G E B U l L P S T R U C T U R E Y O U S T A T E QEiP C O M P L E X E A T S T R i ? : G ; b X A L X T O T H E V I L L A G E BUILD S T R U C T U R E ( N F ( P R O YOU)^ S T A T E T R Y V P COMPLEE'.ENT. S T R L N G : W A L K T O THE VC O M P L E X E K T S T R i N G t V I L L A G E B U I L D S T R U C T U R E STATE P O P N P C O M P L E 3 S S T S T R I N G : B U I L D S T R K T U R E V I L L A G ET H E ) ( N V I L L A G E ) ) ) ) STATE P O P V P C O M P L E M E A T S T R I N G :T ; i E ) ( N VILLAGE)))) S T A T E P O P S C O M P L E Y E N T S,Ti?I N G : B U L L D S S R U C T U R E : ( V P ( VSTATE S COMPLEME?iT S T R I N G 8 B U I L D S T R U C T U R E r (S(TYPE Q 3 E S T I I : O ) ) A A X D D D D J T E E I S E P A S T ) ( S ' J & J ( N P ( P R 0 Y O U ) ) ) ( P R E D ( V -P ( V WALK) ( P P ( P a P ( P R E P T O ) ( P R E P K P ( N P ( D E T T H E ) ( N V I L L A G E ) ) ) ) ) ) ) ) DO Y-OU WANT T O E X A K I N E T H E R E G I S T E R S ? YES IIE P \ P \ O F P C P s L<'RIZTO> = ' ( A D J ) ( P L -P L l ) ( E N C G O O D ) ' L < * R~~' J ' -F f I J i T O > = ' ( A D J ) (P'L P L ) (EMG G O O D ) ' L < * B I J I M * > = [ A D J ) ( P L -PL) ( E N G N I f 3 ) ' e < '~~' t j -W A N *, = *(ADJ,) ( P L P L ) ( E~. I S P I C > * L < ' G A K . > ( J E N O N E ) * L < ' U A P . $ = ( I t i T W O ) L < ' N Y I I * > = . ( I ) F T ) ( I . , N C ' I ' I l J~~)~ L < '~J A ' > = * ( D l~T ) ( b , t l~~ T t ! b ; ) - t c ' c~; ' , = ' ( D E T ) (E!;'; A ) ' L < * M W A -3 = ' 1 ) ( t i E J L~J P ) * L < ' R u L U . ' ; * ? = ( 1 1 ( i "AMRES T A T E 3 . P
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 578 | 0.00346 | null | null | null | null | null | null | null | null |
bc4f3ead5745b375566f7c870da53bf78d329fdc | 219309279 | null | Review: \textit{ {T}he {R}ole of {S}peech in {L}anguage}, Edited by {J}ames {F}. {K}avanagh and {J}ames {E}. {C}utting | The book under review contains the proceedings of a small conference ( 2 2 p a r t i c i p a n t s | {
"name": [
"Nooteboom, Sieb"
],
"affiliation": [
null
]
} | null | null | null | 1976-12-01 | 4 | 0 | null | Stockholm, who s y s t e m a t i c a l l y explores t h e explanatory value of q u a n t i t a t i v e models of speech production and perception i n phonology, e . g . Lindblom 1 9 7 2 , 1 9 7 5 ) . The o r g a n i z e r s of the conference, Kavanagh and Liberman, have taken c a r e t o s e l e c t well-known r e s e a r c h e r s with d i f f e r e n t backgrounds and d i f f e r e n t i n t e r e s t s t o d i s c u s s t h e v a r i o u s problems which may be derived from t h e c e n t r a l q u e s t i o n : "do we i n c r e a s e our understanding of language when w e t a k e i n t o account t h a t i t i s spoken?"T h e r e s u l t i n g t e x t s make i n t e r e s t i n g r e a d i n g , although one w i l l look i n v a i n f o r a convincing answer t o the i n i t i a l q u e s t i o n . D i f f e r e n t i n v e s t i g a t o r s have d i f f e r e n t opinions and the p r e s e n t s t a t e of knowledge does n o t seem t o make i tThe R o J~I of Speech in Language p o s s i b l e t o settle the m a t t e r . In most papers specialist knowledge i s freely intermixed w i t h s p e c u l a t i o n , and i t i s not always e a s y t o t e l l the one f r o m the o t h e r . The discussions g e n e r a l l y serve more to con-tinrle speculation than t o criticize i n d e t a i l each other's t h i n k i n g . These remarks a r e not meant a s a criticism of the conference and i t s proceedings. They (Cutting, Rosner and Foard 1976) . Furthermore, to my knowledge, nobody has yet seriously discussed the. difficulties for a theory of "wired-in" feature detectors stemming from perceptual normalization experiments in which it is shown that response distributions in phoneme identification tasks may shift systematically due to the immediate environm e n t of the test segment (e .g . Fourcin 1972 ) .T h e volume under review is not only remarkable for the many interesting and stimulating papers it contains but also for -what it does not con&ain. In a collection of papers with the title "The r o l e of speech in language" one w o~l d have expected to find at least one contribution seriously discussing the relation between speech prosody and linguistic structure. It is ironical that the only paper in which intonational contrast is given more ateention than obligatory lip service is Stokoe's contribution "The shape of soundles~ language", dealing with Stokoe's treatment of intonation and its kinesic correlate in sign language seems to make explicit why so many speech researchers do not pay attention to speech prosody. He suggests that intonational contrasts "are not necessarily linguistic and have more affinity with other systems that signal affect than with phonemic contrasts. There remain then only phonemic contrasts between consonant and consonant, vowel and vowel, and tone and tone (when so used) as the ihdisputably linguistic, basic features of language". One may fear that this undue overemphasis on phonemic contrast in speech perception research will persist until speech scientists turn away from the study of isolated CV-syllables and start wondering about the perception of normal spontaneous connected speech. | null | null | null | null | Main paper:
:
Stockholm, who s y s t e m a t i c a l l y explores t h e explanatory value of q u a n t i t a t i v e models of speech production and perception i n phonology, e . g . Lindblom 1 9 7 2 , 1 9 7 5 ) . The o r g a n i z e r s of the conference, Kavanagh and Liberman, have taken c a r e t o s e l e c t well-known r e s e a r c h e r s with d i f f e r e n t backgrounds and d i f f e r e n t i n t e r e s t s t o d i s c u s s t h e v a r i o u s problems which may be derived from t h e c e n t r a l q u e s t i o n : "do we i n c r e a s e our understanding of language when w e t a k e i n t o account t h a t i t i s spoken?"T h e r e s u l t i n g t e x t s make i n t e r e s t i n g r e a d i n g , although one w i l l look i n v a i n f o r a convincing answer t o the i n i t i a l q u e s t i o n . D i f f e r e n t i n v e s t i g a t o r s have d i f f e r e n t opinions and the p r e s e n t s t a t e of knowledge does n o t seem t o make i tThe R o J~I of Speech in Language p o s s i b l e t o settle the m a t t e r . In most papers specialist knowledge i s freely intermixed w i t h s p e c u l a t i o n , and i t i s not always e a s y t o t e l l the one f r o m the o t h e r . The discussions g e n e r a l l y serve more to con-tinrle speculation than t o criticize i n d e t a i l each other's t h i n k i n g . These remarks a r e not meant a s a criticism of the conference and i t s proceedings. They (Cutting, Rosner and Foard 1976) . Furthermore, to my knowledge, nobody has yet seriously discussed the. difficulties for a theory of "wired-in" feature detectors stemming from perceptual normalization experiments in which it is shown that response distributions in phoneme identification tasks may shift systematically due to the immediate environm e n t of the test segment (e .g . Fourcin 1972 ) .T h e volume under review is not only remarkable for the many interesting and stimulating papers it contains but also for -what it does not con&ain. In a collection of papers with the title "The r o l e of speech in language" one w o~l d have expected to find at least one contribution seriously discussing the relation between speech prosody and linguistic structure. It is ironical that the only paper in which intonational contrast is given more ateention than obligatory lip service is Stokoe's contribution "The shape of soundles~ language", dealing with Stokoe's treatment of intonation and its kinesic correlate in sign language seems to make explicit why so many speech researchers do not pay attention to speech prosody. He suggests that intonational contrasts "are not necessarily linguistic and have more affinity with other systems that signal affect than with phonemic contrasts. There remain then only phonemic contrasts between consonant and consonant, vowel and vowel, and tone and tone (when so used) as the ihdisputably linguistic, basic features of language". One may fear that this undue overemphasis on phonemic contrast in speech perception research will persist until speech scientists turn away from the study of isolated CV-syllables and start wondering about the perception of normal spontaneous connected speech.
Appendix:
| null | null | null | null | {
"paperhash": [
"cutting|perceptual_categories_for_musiclike_sounds:_implications_for_theories_of_speech_perception",
"fourcin|perceptual_mechanisms_at_the_first_level_of_speech_processing",
"lindblom|phonetics_and_the_description_of_language"
],
"title": [
"Perceptual Categories for Musiclike Sounds: Implications for Theories of Speech Perception",
"Perceptual Mechanisms at the First Level of Speech Processing",
"Phonetics and the Description of Language"
],
"abstract": [
"Sawtooth acoustic stimuli of different rise times are identified as coming from a plucked string instrument (pluck) or a bowed one (bow). Like stop consonants, these sounds are perceived categorically—discrimination is poor for stimuli identified as belonging to a single class but good for those identified as members of different classes. Varying the interval between two successive musiclike stimuli hardly alters discrimination. Sawtooth stimuli lasting 750 ms are clearly perceived categorically; those lasting 250 ms are not. Prolonged exposure to a pluck or bow stimulus can shift the rise-time boundary between categories. Shifts due to such selective adaptation decrease as adapting and test stimuli share fewer characteristics. Adaptation of postulated “feature detectors” therefore may occur in input systems prior to the detectors themselves. Our findings contradict previous claims that categorical perception and selective adaptation are manifestations of psychological processes unique to speech perception.",
"Three sets of experiments will be briefly described. The first deals with a particular aspect of the perception of pitch and is intended to make a contribution to the understanding of the low level processing of intonation in speech. The second set of experiments deals with a feature of the perception of tone in Cantonese and is particularly concerned with the way in which the phonetic value of a tone is determined by its relation to a reference pattern. The last experiments involved the use of single formant whispered stimuli which are now interpreted in relation to a reference pattern which is a function of an inferred vocal tract. \"file results of the three lots of experiments are interpreted as showing that although reference to the mechanism of speech production may be of great value, very powerful perceptual mechanisms are necessarily available for the perceptual processing of speech in ways which are not related to its formation in the speaker’s vocal tract, and, indeed, that our ability to produce speech may be in large measure dependent on our auditory ability to perceive its patterns and partly controlled by a process of auditory pattern feedback.",
"In the scientific quest for insight, theory construction is an important tool. Involved in the evaluation of a certain theory is the notion of explanation. Explanatory goals have been formulated for many fields of inquiry including the study of language. For segmental phonology -— with which we shall be concerned in the present study — concrete proposals as to how such goals might be attained have been explored notably by Chomsky and Halle (1968). What should be meant by an explanatory theory of phonology? It would be unwise to try to give a comprehensive answer that most investigators concerned would find acceptable. Nevertheless, adopting a sufficiently general point of view it seems clear that such a theory would deal with ‘classical’ problems of phonology such as SOUND CHANGE (Passy 1890, Martinet 1968) and with the principles underlying for example, SYSTEMS (vowel and consonant inventories) (Jakobson 1968) and CONTRASTS (Jakob— son et al. 1952, Ladefoged, 1967a, Wang, 1967, Chomsky and Halle 1968, Chapter 7), SYLLABLE STRUCTURES (phonotactic patterns of segments) (Greenberg 1965, Sigurd 1968), or the origins of ‘rules’, ‘rule ordering’, ‘natural classes’, and ‘features’. Is such a phonology at all possible at the present stage? An optimistic answer may be questioned on a number of grounds. Linguists might express doubts because of the enormous task of accounting for the wealth of facts involved, or more importantly, because of limitations of present research strategies. Let us briefly examine the latter motivation, with particular regard to the role that phonetics plays within the influential framework of recent generative phonology (Chomsky and Halle 1968)"
],
"authors": [
{
"name": [
"J. Cutting",
"B. Rosner",
"C. F. Foard"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"A. Fourcin",
"A. Rigault",
"R. Charbonneau"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"B. Lindblom",
"A. Rigault",
"R. Charbonneau"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null
],
"s2_corpus_id": [
"33693340",
"64532343",
"65075010"
],
"intents": [
[],
[],
[]
],
"isInfluential": [
false,
false,
false
]
} | null | 578 | 0 | null | null | null | null | null | null | null | null |
8238eaf4af71cc03c340ba353898756d6e41fbe7 | 219308473 | null | Review: \textit{ {D}ynamic {I}nformation and {L}ibrary {P}rocessing}, by {G}erard {S}alton | Austin 78712 First, an overall characterization of the book It is an outstanding work. Time may well establish it as a masterpiece Salton has succeeded in combining. (1) the presentation of an interesting and, more importantly, a challenging concept--the "dynamic library" --toward which he believes libraries and information agencies ought to direct their research, developmental, and organizational efforts, (2) extensive guides to the relevant literature in several fields, through late 1974, (3) a textbook Eor at least two semesterlength courses, for which my suggested titles would be "Language Processing for Information Storage and Retrieval" and "Library Systems Analysisf', plus a good part of a third semester on "Library Automation", and ( 4 ) an invaluable reference work for computational Wnguists, information scientists, and librarians. REVIEW* DYNAMIC INFORMATION AND LIBRARY PROCESSING Now to the details. Since this review is being prepared for the American Journal of Computational Linguistics, it will be presented in a somewhat unusual format. Instead of beginning at the beginning of the book, I shall start by discussing what seem to me to be the book's highlights for the computational linguist Only after that discussion shall I deal with the gerleral plan of the book and with other specific parts of it The book's ten chapters are intended to be capable of being read independently of one another, although most readers will want to peruse Chapter 1 ahead of any other in order to understand Salton's underlying theme for the book Of the ten chapters, those most immediately relevant to computational linguistics are undoubtedly the last two, plus Chapter 3. The last two are part of a section called "Dynamic Information Processing," in which Salton connects basic concepts in file organization and language processing with their potential applications in the dynamic library (about which more is said below) At the heart of computational linguistics, Chapter 9, "Language Processing," condenses into 49 pages a frank evaluative review of the state-of-the-art in this field Salton links the research in the field with its potential for applications to information systems by saying: A content analysis system going beyond the identification of individual terms . . . requires a t l e a s t three parts: a d e s c r l p t l o n of an area of 'discourse i n terms of basic e n t i t i e s , or concepts, of importance in t h i s area, including also the main logical-semantic relationships that must be identifled between these e n t i t i e s ; a l l n g u i s t l c | {
"name": [
"Wyllys, Ronald E."
],
"affiliation": [
null
]
} | null | null | null | 1977-02-01 | 0 | 0 | null | Now to the details. Since this review is being prepared for the American Journal of Computational Linguistics, it will be presented in a somewhat unusual format. Instead of beginning at the beginning of the book, I shall start by discussing what seem to me to be the book's highlights for the computational linguist Only after that discussion shall I deal with the gerleral plan of the book and with other specific parts of it The book's ten chapters are intended to be capable of being read independently of one another, although most readers will want to peruse Chapter 1 ahead of any other in order to understand Salton's underlying theme for the book Of the ten chapters, those most immediately relevant to computational linguistics are undoubtedly the last two, plus Chapter 3. The last two are part of a section called "Dynamic Information Processing," in which Salton connects basic concepts in file organization and language processing with their potential applications in the dynamic library (about which more is said below) At the heart of computational linguistics, Chapter 9, "Language Processing," condenses into 49 pages a frank evaluative review of the state-of-the-art in this field Salton links the research in the field with its potential for applications to information systems by saying: ships that must be identifled between these e n t i t i e s ; a l l n g u i s t l c The trouble with these arguments i s t h a t a c o r r e c t premise--that most automatic indexing products a r e imperfect--leads wrongly to t h e conclusion t h a t t h e automatic product is n e c e s s a r i l y i n f e r i o r t o one obtained i n t e l l e c t u a l l y by human experts.He concludes that although "it is hazardous to extrapolate test results obtained in a laboratory environment t o operational situations involving possibly hundreds of thousands of items", nevertheless, a number of different, independent tests--several of which he discusses--have shown that "relatively s i m p l e automatic text analysis systems do not produce in a document retrieval environment search results inferior to" those of conventional manual indexing As befits a final chapter, Chapter 10, entitled (like its superordinate) "Dynamic Information Processing", shows how the theories and techniques developed earlier in the book can be applied to the book's main theme, the dynamic llbrary. As Salton puts it Having dealt with the chapters that I suggest will be of primary interest to computational linguists, we can now examine the book as a whole Salton states that his overall purpose in the book is to brjdge the gap between computer science and information science by intrdducing a new environment, called the dynamic library, and a set of dynamic information processing tasks to operate in that environment. The idea is to carry out most processing tasks, such as content analysis, classification, information searcb, and retrieval, interactively under user control, while simultaneously accommodating the file updating and maintenance procedures that are inherent in a changing data processing situation.The key to achieving the goals of the dynamic library is the This important concept of the clustered file i-s discussed in detail in Chapter 8, "Automatic Document and Query Classification"The use of clustered files makes it practical to "maintain the library system in a continuous state of flux"--i e , to make it a dynamic library--by facilitating query processing in whlch both query vectors and document vectors are continually subjected to small changes As its vector changes accumulate, a document's "classification", i e , its cluster, may change As document changes accwnl~late, a cluster's centroid may change.The book as a whole, then is devoted to expounding the them6 of the dynamic library and to explicating the necessary detalls 2 , 3 , 4 , 5 , 5 , ( 9 ) A course in "Library Systems Analysis"As part of a course in "Library Automation' 1, 2 , 4Every reviewer finds a few nits to pick I wish Salton had not used the abbreviation "log" for "natural logarithm" instead of the now starldard "ln", or at least that he had explicitly stated his usage A few of the tables contaln minor numerical errors, none that I noticed affeots the conclusions belng drawn, and at least one ("18" instead of "22" In Table 1 -2) when corrected strengthens the argument I wlsh Salton had dealt less curtly with systems analysis, but, after all, the book contains 537well-filled pages as it standsIn conclusion, I t h b k it likely t h a t this book w i l l come to be viewed as a master contribution to the professional and pedagogical Uterature in natural-language analysis., information science, and library science | null | null | null | null | Main paper:
review* dynamic information and library processing:
Now to the details. Since this review is being prepared for the American Journal of Computational Linguistics, it will be presented in a somewhat unusual format. Instead of beginning at the beginning of the book, I shall start by discussing what seem to me to be the book's highlights for the computational linguist Only after that discussion shall I deal with the gerleral plan of the book and with other specific parts of it The book's ten chapters are intended to be capable of being read independently of one another, although most readers will want to peruse Chapter 1 ahead of any other in order to understand Salton's underlying theme for the book Of the ten chapters, those most immediately relevant to computational linguistics are undoubtedly the last two, plus Chapter 3. The last two are part of a section called "Dynamic Information Processing," in which Salton connects basic concepts in file organization and language processing with their potential applications in the dynamic library (about which more is said below) At the heart of computational linguistics, Chapter 9, "Language Processing," condenses into 49 pages a frank evaluative review of the state-of-the-art in this field Salton links the research in the field with its potential for applications to information systems by saying: ships that must be identifled between these e n t i t i e s ; a l l n g u i s t l c The trouble with these arguments i s t h a t a c o r r e c t premise--that most automatic indexing products a r e imperfect--leads wrongly to t h e conclusion t h a t t h e automatic product is n e c e s s a r i l y i n f e r i o r t o one obtained i n t e l l e c t u a l l y by human experts.He concludes that although "it is hazardous to extrapolate test results obtained in a laboratory environment t o operational situations involving possibly hundreds of thousands of items", nevertheless, a number of different, independent tests--several of which he discusses--have shown that "relatively s i m p l e automatic text analysis systems do not produce in a document retrieval environment search results inferior to" those of conventional manual indexing As befits a final chapter, Chapter 10, entitled (like its superordinate) "Dynamic Information Processing", shows how the theories and techniques developed earlier in the book can be applied to the book's main theme, the dynamic llbrary. As Salton puts it Having dealt with the chapters that I suggest will be of primary interest to computational linguists, we can now examine the book as a whole Salton states that his overall purpose in the book is to brjdge the gap between computer science and information science by intrdducing a new environment, called the dynamic library, and a set of dynamic information processing tasks to operate in that environment. The idea is to carry out most processing tasks, such as content analysis, classification, information searcb, and retrieval, interactively under user control, while simultaneously accommodating the file updating and maintenance procedures that are inherent in a changing data processing situation.The key to achieving the goals of the dynamic library is the This important concept of the clustered file i-s discussed in detail in Chapter 8, "Automatic Document and Query Classification"The use of clustered files makes it practical to "maintain the library system in a continuous state of flux"--i e , to make it a dynamic library--by facilitating query processing in whlch both query vectors and document vectors are continually subjected to small changes As its vector changes accumulate, a document's "classification", i e , its cluster, may change As document changes accwnl~late, a cluster's centroid may change.The book as a whole, then is devoted to expounding the them6 of the dynamic library and to explicating the necessary detalls 2 , 3 , 4 , 5 , 5 , ( 9 ) A course in "Library Systems Analysis"As part of a course in "Library Automation' 1, 2 , 4Every reviewer finds a few nits to pick I wish Salton had not used the abbreviation "log" for "natural logarithm" instead of the now starldard "ln", or at least that he had explicitly stated his usage A few of the tables contaln minor numerical errors, none that I noticed affeots the conclusions belng drawn, and at least one ("18" instead of "22" In Table 1 -2) when corrected strengthens the argument I wlsh Salton had dealt less curtly with systems analysis, but, after all, the book contains 537well-filled pages as it standsIn conclusion, I t h b k it likely t h a t this book w i l l come to be viewed as a master contribution to the professional and pedagogical Uterature in natural-language analysis., information science, and library science
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 576 | 0 | null | null | null | null | null | null | null | null |
a36cf03433746388bb2ca852f98185ac0f5678ed | 219301765 | null | Computation of a Subclass of Inferences: Presupposition and Entailment | as a t e s t rather than as a rule in a f o m l system. One discovers a p i r i c a l l y whether St is an entailment of S by trying to construct a context in which S is true, but in which St is false. Entailment is not the same as material implication. For instance, let S by "John managed to kiss Mary,' which entails sentence S f , "John kissed Mary. l1 Givon (1973) argues that even if N S T is true, we would not want to say that 'Wohn did not m a g e to kiss Mary. l1 The reason is that "managerf seems to presume an attempt. Hence, if John did not kiss Mary, we cannot conclude that John did not manage to kiss Mary, for he may not have attempted to kiss Mary. Though S entails S ' , it is not the case that S S t , since that would require N S ' S N S . We have shuwn that entailments may be s u b f d a -d e r ~v e d , that is, that they may be computed by structural means. As an example, consider the sentence S below; one could represent its rrreaning representatbn as L. S entails S f , with meaning representation Lf . S. John forced us to leave. L. (IN-m-PAST (force John (EVENT ( IN-THE-PAST (leave we ) ) ) 1 1 S f , We left. L f . (IN=-PAST (leave we)) F r o m the meaning representation selected it is easy to see the appropriate s u b f d and the identity tree t r a n s f o r m t i o ~ which demonstrate that this is a subformula-derived entailment. (This is, of course, a t r i v i a l tree .h.ansformation. A nontrivial example appears in Section 1.4, for pnsupposition. ) Many ewmples of entailment axe given in Secticn 2. | {
"name": [
"Joshi, Aravind K. and",
"Weischedel, Ralph"
],
"affiliation": [
null,
null
]
} | null | null | null | 1977-02-01 | 0 | 0 | null | The Athletics, who won the World Series last year, play today. m l y that there is soone referent which must have that quality. On the other hand, nonrestrictive r e l a t i v e clauses, such as ( 5 ) p s u p p o s e that the particular object named also has in addition the quality mehtioned in the relative clause. Sentence (5a) might be taken as a parapkase of "The Athletics play today, and the Athletics won the World Series last year." W e can easily imagine (6a) being spoken at the beginning of a press conference to inform the news agency of the t r u t h of (6b). , itagain", "other", and "anotherr1, carry the maning of sanething being repeated. These words have presuppositions that the i t e m occurred at Seast o n e hfo* 9. a. B did not play again today.Hmeverb. B did not play at least once before.Note that these words include various syntactic categories. ttAlso" , "too" 3 "again" , are adverbial elements (adjuncts ) . In section 5.1, the role of preeuppcsitbn and entailments as inferences is pinpointed. In section 5.2, the use of e m t i c pr-tiws is considered.The tm ninfemncelf has been used recently to refer to any conjecture made, given a text in scme natwal language. Chamiak (1973, 19721, Schank (19731, Schank and Rieger (19731, Schank, et. dl. (19751, and Wilks 1975 the r e~~t in their definition that they be independent of the 6 i t u a t h ( a l l context not =presented s t r u c U l y ) is strong. For instance, fkm sentence S Wow, one might feel that S t should be entailed; yet, it is not.John saw Jim i n the h a l l , and Mary saw Jim in his office.S : John and Mary s w J h in different places.By appropriately chosen previous t e x t s , S t need not be true whenever S is. . Because S t is presupposed by S, SVt becrmes a presuppositim of S, nat merely an entailment.Who prevented John f h n leaving?Someone *vented John f k m leaving.John did not leave. | We use context to refer to the situation in which a sentence m y occur. Thus, it would include all discourse prior to the sentence under consideration, beliefs of the interpreter, i. e. , in shwt thestate of the i n t q r e t e r . We use p m t i c s t o describe bll phenomena (and computations mdelling them) that reflect the effect of context.A sentence S entails a sentence S t if and onlv if in everv context which S is m e , S t is also true. We may say then that St is an entaibmt of S. This definition is used within linguistics -8 r as a t e s t rather than as a rule in a f o m l system. One discovers a p i r i c a l l y whether St is an entailment of S by trying to construct a context in which S is true, but in which St is false.Entailment is not the same as material implication. For instance, let S by "John managed to kiss Mary,' which entails sentence S f , "John kissed Mary. l1 Givon (1973) We have shuwn that entailments may be s u b f d a -d e r~v e d , that is, that they may be computed by structural means. As an example, consider the sentence S below; one could represent its rrreaning representatbn as L.S entails S f , with meaning representation Lf . Notice that it is questionakde whether one understands sentence S or t h e word Ivforce" if he d e s not knaw t h a t S t is true whenever S is. In this sense, entailment is certainly necessary knowledge ( though not sufficient) for understan-natural* language. We w i l l see this again for presupposition. presuppoges that there is a greatest prime n -.The fact that there is none explaine why the sentence is anamlous. As an example of a subfonmila derived presupposition consider sentences S1 and S1' below. It i s easy to see that whether S1 is true or false, S1' is assumed to be true.John stopped beating Mary.LJ: (IN-LTKE-PAST (stop (EVENT (beat John Mary) ) ) ) S1' : John had been beating Mary.Ll and L1' are semantic representations for S1 and S1' respectively. The w e l lf o m d subformula in this case i s all af L3.. The tree transformation from W. to L1' offers a n o n~v i a l e-le of a subfonmila-derived presupposition.Notice that one might wonder whether sentence S1 and the meaning of "stopt1 w e r e understood i f one did not. h u w t h a t Sly rust be true whether John stopped o r not. In this sense, presupposition is necessary (but not sufficient) knowledge for understanding natural language.We have s h m that presuppositions (as we have defined them above) m y be subfonmila-d-ved. Henceforth, we w i l l use "entailment" t o mean an entai3.nm-t whjch is not also a presupposition.-11- | As long as the assumption of context-free syntax for semantic representations is satisfied, the same algorithms and data structures of our system can be used regardless of choice of semantic primitives or type of semantic representation.Let S and S f be sentences w i t h meaning representations L and Lr respectively. If there is a well-formed subforuila P of L and sane tree trwmformation F such that Lf = F(P), then we say S t may be subformula-derived h S. The type of -tree transfornations that are acceptable for F have been formalized and studied extensively i n ccmnputat ional linguistics as f inite-state tree transformat ions.The main point of this work is that the presuppositions and entailments of a sentence may be subfomula-derived. We have built a system by which we m y specify subformulas P and tree tmnsformations F. The system then automitically generates presuppositions and e n t a i h m t s from an input sentence S. | M s section is divided into two subsections, Section 2.1 deals with presuppositions, section 2.2 with entailments. All example sentences arentmimred. An (a) sentence has as presuppsition or entailment the comsponding (b) sentence.Presuppositions arise fm two different structural sources: syntactic constructs (the syntactic or relational strzlctur?e) and lexical items (semantic structure 1 .Perhaps the mst intriguing cases of presupposition are those that arise f h m syntactic constructs, for these demnstrate c w l e x interaction between semantics and syntax.A construction bown as the cleft sentence gives rise to presuppositions for the corresponding surface sentences. Consider that if someone says (1) to you, you m i & t respond with (2a).1. I am sure one of the players won the game for us yesterday, but I do not knm who did.2. a. It is B who won the game.b. Scaneone won the game.The form of the cleft sentence is the word "it" followed by a tensed form of the word "be1', follawed by a noun phrase or p~p o s i t i o n a l phrase, followed by a relative clause.Note particularly that the presupposition (2b) did not arise frcw any of the individual words. Rather, the pempposition, which is clearly senwtic since it i s part of the tmth qmditions of the sentence, arose fram the syntactic constmct. Thus, t h e syntactic (or relational) s t r u m of the sentence can carry important samntic information.C l e f t sentences i l l u s t~t e one important use of presuppositions: mreference. C l e f t sentences asert the identity of one individual with anothw fidividual referred to peviously i n the dialogue. 3. a. John's brother plays for the Phillies.b, John has a brother.4. a. The team that fhe Phillies play today has won three games in a m w .b. The Phillies play a team today.A t least five distinct semantic classes of words having entailments have been identified by Karttunen (1970) . In the following exmples, the (b) sentence is entailed by the (a) sentence.Redibates such as l % e be a positicplt, "have the oppmtmityu, and ' %e &Left, are called "~nly-ifv verbs be~xiuse the embedded satence is entailed only if the predicate is in the negative. Far instance, (1Oa) entails (lob), but (11) has no entailment.10. a. The P h i l H e s w e r e not in a position to w i n the pennant.b. The Phillies did not w i n the pennan-tr.11, a. The Phillies were i n a position to w i n the pennant.V e r b s such as ttforce", "causetr, and are "if" verbs, for the embedded sentence is entailed if they are in t h e positive.In tha folluwing sections, we w i l l consider the effect on presupposition 2of embeddk.lg (1) mdes v a r h predicates taking enbedded sentences.-19-In this section, the fc3llcrwix-g question is considered: Suppose that a sentence S has a set of e n t a i m t s and a set of presuppositions. Suppose further, that S is &ded in another sentence S t . Are the e n t a h t s and presuppositions of S also entailments and presuppositions of S t as a whole?This has been referred to as the pmjection problem for entailmints As an example sentence, consider (11, which presupposes (2).1. Jack regretted that John left.Many predicates taking embedded sentences could be called holes because they let presuppositions of embedded sentences t h r o m to betrone presuppositions of the compound sentence. "harett is such a predicate;( 3) presupposes (2 1 , 3. Mary is aware that Jack regretted that John left.A l l prwdmtes taking embedded sentences, except for the v@bs of saying, the predicates of p p o s i t i o n a l attitude, and the connectives appear to be holes.The For instance, (4) presupposes (5), not (2).Mary asked whether Jack regretted that John left.Mary claimed John left.Analysis of predicates af propositional attitude is very similar to that of speech acts. SOE predicates of popositicnal attitude are "believe", The effect of connectives is rather complex, as (8) and (9) demrmsmte.Sentence (8) presupposes (21, but ( 9 ) clearly does not.In the examples, we w i l l embed (15) under various predicates, to see hm the c~t a i h n t (16) of (15) is affected.Fred prevented Mary fmPn leaving.Mary did not leave.Corrresponding to the class of holes for presuppositions, two cases arise for entaihmts. 21. John said that bed prevented Mary f m leaving.Srraby (1975) analyses these predicates i n the same way as the speech acts. "Believet' , "think" , and "suspect" are examples of a subclass analogous t o ''if predicatesv1 or t o "say", "declare1', and %ffhvV. v~u b t t tis an example of a second subclass analogous to "negative two-way Implicative predicates" such as "failv. 26. John believed that Mary did not leqve.For "if A then Bn , the entailments are of the form "if A then Cfr , where C is an entailment of B, For ''A and B" , t h e entailments are the union of the entailments of A and of the entailments of B, since both A and B are entailed by "A and B". For "A or B", t h m do not seem to be any useful entailments. Then, we may lassociate with that parti-path the txee .trensforsmtion yielding the presupposition of that syntactic construct. For instance, cleft sentences are syntactically marked as the ward I ' i t U , followed by a tensed form of l1beH 9 follwed by either a noun phnase or a prwpositianal Just before popping to a higher sententid level a projection function i s applied, which is merely a CASE statement for the four? cases described in We have only outlined hew to q u t e presupposition and e n t a i h n t . Jack was there, then Jack regretted t h a t John left.9. If John left, t%en Jack regretted that John left. Let A and B be the antecedent and consequent respectively of the conpow1dsentence "if A -then B".The examples of (8) and (9) are complex, f &r they seem to demonstrate that the context set up by the antecedent A must be part of the canputation. FW the examples given then, (8 > psupposes (10) , and (9 ) presupposes (11) which is a tautology.10. If Jack was there, then John left. A second subclass of verbs includes "deny". They are analogous to "negative if verbs1'. When "denyu is i n the negative, embedded entailments are blocked. Hawever, when lldenyN is positive the entailnmts of the | Bench forced the game t o go into extra innings.The game went i n t o e x t m innings.Johnny Bench did not force the game to go into extra innings.Note that (12a) entails (12b), but (13) has no such entailment.A "negative-if" verb entails the negative of the embedded sentence when the verb is positive. If the entailment is positive, we m y call these "positive tm-way implicative" verbs. There are also "negative two-way implicative1t vefbs. Cansider 18and (19). W e nuw turn our attention to various factors that must be accounted f w in ccaxputing pres~positions and e n t a i h m t s of canpound sentences.The five cases disussed above outline a solution to the projection pmblem fop presuppositions.The limitations of the system are of two kinds: those that could be handled witbin the f h m w o k of the system but are not because of limitstims of man-hours, and those that could not be handled wi* the present flxamwrk.The system i s currently limited i n four ways, each of which could be removed, given time. One set of restrictions results froen the fact that our pragrmn represents only a small part of a camplete natural langauge processing system. Only the syntactic component is included (though these inferences, which are semantic, are canputed while parsing). fQ a consequence, no a n b i g u i q is resolved except that which is syntactically resolvable.Second, though a trmsforsm.tional output component is included to facilitate reading the output, it has a very limited range of constructions.The principles used i n designing the component are sound though.A third aspect is caputation time. Since our main interest was a new type of computation for a syn-tactic ccxnponent, we have not stressed efficiency in time nor storage; rather, we have concentrated on writing the system f a i r l y rapidly. Considering the nLnnber of conceptually simple, efficiency meesures t h a t we sacrificed for speed i n implementing the system, we are quite pleased that the average CPU time to canpute the presupposition and e n t a i h n t s of a sentence is twenty seconds on the DEC -10. eneies muld have be devised in order i~ encode these depehdencies. | Main paper:
fmgmt ics and context:
We use context to refer to the situation in which a sentence m y occur. Thus, it would include all discourse prior to the sentence under consideration, beliefs of the interpreter, i. e. , in shwt thestate of the i n t q r e t e r . We use p m t i c s t o describe bll phenomena (and computations mdelling them) that reflect the effect of context.A sentence S entails a sentence S t if and onlv if in everv context which S is m e , S t is also true. We may say then that St is an entaibmt of S. This definition is used within linguistics -8 r as a t e s t rather than as a rule in a f o m l system. One discovers a p i r i c a l l y whether St is an entailment of S by trying to construct a context in which S is true, but in which St is false.Entailment is not the same as material implication. For instance, let S by "John managed to kiss Mary,' which entails sentence S f , "John kissed Mary. l1 Givon (1973) We have shuwn that entailments may be s u b f d a -d e r~v e d , that is, that they may be computed by structural means. As an example, consider the sentence S below; one could represent its rrreaning representatbn as L.S entails S f , with meaning representation Lf . Notice that it is questionakde whether one understands sentence S or t h e word Ivforce" if he d e s not knaw t h a t S t is true whenever S is. In this sense, entailment is certainly necessary knowledge ( though not sufficient) for understan-natural* language. We w i l l see this again for presupposition. presuppoges that there is a greatest prime n -.The fact that there is none explaine why the sentence is anamlous. As an example of a subfonmila derived presupposition consider sentences S1 and S1' below. It i s easy to see that whether S1 is true or false, S1' is assumed to be true.John stopped beating Mary.LJ: (IN-LTKE-PAST (stop (EVENT (beat John Mary) ) ) ) S1' : John had been beating Mary.Ll and L1' are semantic representations for S1 and S1' respectively. The w e l lf o m d subformula in this case i s all af L3.. The tree transformation from W. to L1' offers a n o n~v i a l e-le of a subfonmila-derived presupposition.Notice that one might wonder whether sentence S1 and the meaning of "stopt1 w e r e understood i f one did not. h u w t h a t Sly rust be true whether John stopped o r not. In this sense, presupposition is necessary (but not sufficient) knowledge for understanding natural language.We have s h m that presuppositions (as we have defined them above) m y be subfonmila-d-ved. Henceforth, we w i l l use "entailment" t o mean an entai3.nm-t whjch is not also a presupposition.-11-
elementary examples:
M s section is divided into two subsections, Section 2.1 deals with presuppositions, section 2.2 with entailments. All example sentences arentmimred. An (a) sentence has as presuppsition or entailment the comsponding (b) sentence.Presuppositions arise fm two different structural sources: syntactic constructs (the syntactic or relational strzlctur?e) and lexical items (semantic structure 1 .Perhaps the mst intriguing cases of presupposition are those that arise f h m syntactic constructs, for these demnstrate c w l e x interaction between semantics and syntax.A construction bown as the cleft sentence gives rise to presuppositions for the corresponding surface sentences. Consider that if someone says (1) to you, you m i & t respond with (2a).1. I am sure one of the players won the game for us yesterday, but I do not knm who did.2. a. It is B who won the game.b. Scaneone won the game.The form of the cleft sentence is the word "it" followed by a tensed form of the word "be1', follawed by a noun phrase or p~p o s i t i o n a l phrase, followed by a relative clause.Note particularly that the presupposition (2b) did not arise frcw any of the individual words. Rather, the pempposition, which is clearly senwtic since it i s part of the tmth qmditions of the sentence, arose fram the syntactic constmct. Thus, t h e syntactic (or relational) s t r u m of the sentence can carry important samntic information.C l e f t sentences i l l u s t~t e one important use of presuppositions: mreference. C l e f t sentences asert the identity of one individual with anothw fidividual referred to peviously i n the dialogue. 3. a. John's brother plays for the Phillies.b, John has a brother.4. a. The team that fhe Phillies play today has won three games in a m w .b. The Phillies play a team today.A t least five distinct semantic classes of words having entailments have been identified by Karttunen (1970) . In the following exmples, the (b) sentence is entailed by the (a) sentence.Redibates such as l % e be a positicplt, "have the oppmtmityu, and ' %e &Left, are called "~nly-ifv verbs be~xiuse the embedded satence is entailed only if the predicate is in the negative. Far instance, (1Oa) entails (lob), but (11) has no entailment.10. a. The P h i l H e s w e r e not in a position to w i n the pennant.b. The Phillies did not w i n the pennan-tr.11, a. The Phillies were i n a position to w i n the pennant.V e r b s such as ttforce", "causetr, and are "if" verbs, for the embedded sentence is entailed if they are in t h e positive.In tha folluwing sections, we w i l l consider the effect on presupposition 2of embeddk.lg (1) mdes v a r h predicates taking enbedded sentences.-19-
catrplex iscamples: m d e d ehtailments and hsuppositions:
In this section, the fc3llcrwix-g question is considered: Suppose that a sentence S has a set of e n t a i m t s and a set of presuppositions. Suppose further, that S is &ded in another sentence S t . Are the e n t a h t s and presuppositions of S also entailments and presuppositions of S t as a whole?This has been referred to as the pmjection problem for entailmints As an example sentence, consider (11, which presupposes (2).1. Jack regretted that John left.Many predicates taking embedded sentences could be called holes because they let presuppositions of embedded sentences t h r o m to betrone presuppositions of the compound sentence. "harett is such a predicate;( 3) presupposes (2 1 , 3. Mary is aware that Jack regretted that John left.A l l prwdmtes taking embedded sentences, except for the v@bs of saying, the predicates of p p o s i t i o n a l attitude, and the connectives appear to be holes.The For instance, (4) presupposes (5), not (2).Mary asked whether Jack regretted that John left.Mary claimed John left.Analysis of predicates af propositional attitude is very similar to that of speech acts. SOE predicates of popositicnal attitude are "believe", The effect of connectives is rather complex, as (8) and (9) demrmsmte.Sentence (8) presupposes (21, but ( 9 ) clearly does not.In the examples, we w i l l embed (15) under various predicates, to see hm the c~t a i h n t (16) of (15) is affected.Fred prevented Mary fmPn leaving.Mary did not leave.Corrresponding to the class of holes for presuppositions, two cases arise for entaihmts. 21. John said that bed prevented Mary f m leaving.Srraby (1975) analyses these predicates i n the same way as the speech acts. "Believet' , "think" , and "suspect" are examples of a subclass analogous t o ''if predicatesv1 or t o "say", "declare1', and %ffhvV. v~u b t t tis an example of a second subclass analogous to "negative two-way Implicative predicates" such as "failv. 26. John believed that Mary did not leqve.For "if A then Bn , the entailments are of the form "if A then Cfr , where C is an entailment of B, For ''A and B" , t h e entailments are the union of the entailments of A and of the entailments of B, since both A and B are entailed by "A and B". For "A or B", t h m do not seem to be any useful entailments. Then, we may lassociate with that parti-path the txee .trensforsmtion yielding the presupposition of that syntactic construct. For instance, cleft sentences are syntactically marked as the ward I ' i t U , followed by a tensed form of l1beH 9 follwed by either a noun phnase or a prwpositianal Just before popping to a higher sententid level a projection function i s applied, which is merely a CASE statement for the four? cases described in We have only outlined hew to q u t e presupposition and e n t a i h n t .
what the systemdoes not do:
The limitations of the system are of two kinds: those that could be handled witbin the f h m w o k of the system but are not because of limitstims of man-hours, and those that could not be handled wi* the present flxamwrk.The system i s currently limited i n four ways, each of which could be removed, given time. One set of restrictions results froen the fact that our pragrmn represents only a small part of a camplete natural langauge processing system. Only the syntactic component is included (though these inferences, which are semantic, are canputed while parsing). fQ a consequence, no a n b i g u i q is resolved except that which is syntactically resolvable.Second, though a trmsforsm.tional output component is included to facilitate reading the output, it has a very limited range of constructions.The principles used i n designing the component are sound though.A third aspect is caputation time. Since our main interest was a new type of computation for a syn-tactic ccxnponent, we have not stressed efficiency in time nor storage; rather, we have concentrated on writing the system f a i r l y rapidly. Considering the nLnnber of conceptually simple, efficiency meesures t h a t we sacrificed for speed i n implementing the system, we are quite pleased that the average CPU time to canpute the presupposition and e n t a i h n t s of a sentence is twenty seconds on the DEC -10. eneies muld have be devised in order i~ encode these depehdencies.
a.:
The Athletics, who won the World Series last year, play today. m l y that there is soone referent which must have that quality. On the other hand, nonrestrictive r e l a t i v e clauses, such as ( 5 ) p s u p p o s e that the particular object named also has in addition the quality mehtioned in the relative clause. Sentence (5a) might be taken as a parapkase of "The Athletics play today, and the Athletics won the World Series last year." W e can easily imagine (6a) being spoken at the beginning of a press conference to inform the news agency of the t r u t h of (6b). , itagain", "other", and "anotherr1, carry the maning of sanething being repeated. These words have presuppositions that the i t e m occurred at Seast o n e hfo* 9. a. B did not play again today.Hmeverb. B did not play at least once before.Note that these words include various syntactic categories. ttAlso" , "too" 3 "again" , are adverbial elements (adjuncts ) . In section 5.1, the role of preeuppcsitbn and entailments as inferences is pinpointed. In section 5.2, the use of e m t i c pr-tiws is considered.The tm ninfemncelf has been used recently to refer to any conjecture made, given a text in scme natwal language. Chamiak (1973, 19721, Schank (19731, Schank and Rieger (19731, Schank, et. dl. (19751, and Wilks 1975 the r e~~t in their definition that they be independent of the 6 i t u a t h ( a l l context not =presented s t r u c U l y ) is strong. For instance, fkm sentence S Wow, one might feel that S t should be entailed; yet, it is not.John saw Jim i n the h a l l , and Mary saw Jim in his office.S : John and Mary s w J h in different places.By appropriately chosen previous t e x t s , S t need not be true whenever S is. . Because S t is presupposed by S, SVt becrmes a presuppositim of S, nat merely an entailment.Who prevented John f h n leaving?Someone *vented John f k m leaving.John did not leave.
if:
Jack was there, then Jack regretted t h a t John left.9. If John left, t%en Jack regretted that John left. Let A and B be the antecedent and consequent respectively of the conpow1dsentence "if A -then B".The examples of (8) and (9) are complex, f &r they seem to demonstrate that the context set up by the antecedent A must be part of the canputation. FW the examples given then, (8 > psupposes (10) , and (9 ) presupposes (11) which is a tautology.10. If Jack was there, then John left.
mary l e e .:
The five cases disussed above outline a solution to the projection pmblem fop presuppositions.
job clximd that mary aid not leave.:
A second subclass of verbs includes "deny". They are analogous to "negative if verbs1'. When "denyu is i n the negative, embedded entailments are blocked. Hawever, when lldenyN is positive the entailnmts of the
a. johnny:
Bench forced the game t o go into extra innings.The game went i n t o e x t m innings.Johnny Bench did not force the game to go into extra innings.Note that (12a) entails (12b), but (13) has no such entailment.A "negative-if" verb entails the negative of the embedded sentence when the verb is positive. If the entailment is positive, we m y call these "positive tm-way implicative" verbs. There are also "negative two-way implicative1t vefbs. Cansider 18and (19). W e nuw turn our attention to various factors that must be accounted f w in ccaxputing pres~positions and e n t a i h m t s of canpound sentences.
:
As long as the assumption of context-free syntax for semantic representations is satisfied, the same algorithms and data structures of our system can be used regardless of choice of semantic primitives or type of semantic representation.Let S and S f be sentences w i t h meaning representations L and Lr respectively. If there is a well-formed subforuila P of L and sane tree trwmformation F such that Lf = F(P), then we say S t may be subformula-derived h S. The type of -tree transfornations that are acceptable for F have been formalized and studied extensively i n ccmnputat ional linguistics as f inite-state tree transformat ions.The main point of this work is that the presuppositions and entailments of a sentence may be subfomula-derived. We have built a system by which we m y specify subformulas P and tree tmnsformations F. The system then automitically generates presuppositions and e n t a i h m t s from an input sentence S.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 576 | 0 | null | null | null | null | null | null | null | null |
e2769d2a459f87712a0bf94e3346d31e7a4cfd5e | 219305200 | null | Towards a {``}Natural{''} Language Question-Answering Facility | Maxwell is also associated with the Computation Center -Tuggle is also associated with the School of Business Copyright @ 1977 Associa t l o n f p r Computational L i n g u i s t i c s This study describes the structure, implementation and potential of a simple computer program that understands and answers questions in a humanoid mannet. An emphasis bas been placed on the creation of an extendible memory structure--one capable of supposting conversation in normal, unrestricted ErIgli6h on a variety of topics. An attempt has also been made to find procedures that can easily and accurately determine the meaning of input text-A parser using a combination of syntax and semantics has been developed for understanding YES-NO questions, in particular, DO-type (DO, DID, DOES, etc.) questions. A third and major emphasis has been on the development of procedures to allow the program to converse, easily and "naturally1' with a human. This general gaql has been met by developing procedures that generate answers to DO-questions in a manner similar to the way a person might answer them. | {
"name": [
"Maxwell, Bill D. and",
"Tuggle, Francis D."
],
"affiliation": [
null,
null
]
} | null | null | null | 1977-02-01 | 16 | 0 | null | There a r e a number of e x t a n t computer programs which i n t e r a c t i n t e l l i g e n t l y w i t h human i n t e r r o g a t o r s , b u t a l l do s o i n a way w e c h a r a c t e r i e e a s "unnatural". By "unnatural," we mean t h a t they conv e r s e i n a way s i g n i f i c a n t l y u n l i k e two normal a d u l t humans do. For example, Winograd's SHRBLU [19] seems c h i l d -l i k e , and i t s h a r e s w i t h t h e woods' moonrocks system [20] t h e problem of being task-const rained.Weizenbaum's ELIZA [16] As an example of a "natural" and a n "unnatural" d i a l o g u e , imagine two computer programs (CPs) conversing w i t h a human (H) i n which each CP a l r e a d y knows "THE BOY O~JNS A BALL." One CP is "natural" (NCP); t h e o t h e r i s "unnatural" (uCP) .The boy l i v e s on Main S t r e e t . UCP:. By ' t h e boy', I assume you mean t h e one who owns t h e r e d b a l l . NCP: O.K.llhete does t h e boy p l a y with h i s b a l l ? UCP : I don' t krtow. NCP: I don't know, but I assume near h i s residence on Main S t r e e t .A house is on Main S t r e e t . UCP: O.K. The b a l l broke a window.UCP: By ' t h e Ball, ' 1 assume you mean t h e red one the boy owns. BY 'a window,' I assume i r i s a p a r t of t h e house i n Main S t r e e t .Was t h e window expensive t o r e p a i r ? UCP: I don't know. NCP: S i~c e t h e house was expensive t o c o n s t r u c t , I assume t h a t i t s windows a r e expensive t o repair. To g e t a f e e l i n g f o r t h e types of responses t h e currentJy Implemented program CBn produce, t h e following s h o r t dialogue i s presented. WThe dialogue i s given i n upper c a s e w i t h commentary i n lower case.The j: and p: ( i d e n t i f y i n g JZMMY3 and person) were added liere f o r c l a r i t y .JIMMY3 must f i n d o u t who i t is t a l k i n g with i n orderr t o t r a n s l a t e t h e pronoun "I" t o f a c i l i t a t e memory searching. p : BdANDT MAXWELL.BR@DT@MAXWELL i s rtcognized as a l e g i t i m a t e name. The @ i n BRANDT@EWLL is used t o i n d i c a t e t h e combining o f two o r more words t o form a Separate erttity. I n t h e above example, t h e names BRANDT and W E L L a r e recognized as f i r s t and l a s t names, respectively, t h a t , when appearing together, a r e i n t e r p r e t e d a s t h e f u l l name, [email protected], YOU.Parsing is accomplished by matching i n p u t t o templates c o n s i s t i n g of (ACTOR ACT OBJECT) describing t h e d i f f e r e n t meaning senses f o r main verbs. T h i s question is parsed using t h e (PERSON KNOW PERSON) d e f i n i t i o n of KNOW. On i n p u t , YOU is t r a n s l a t e d t o JIEIMY3 (which has ISA PERSON a s one of i t s properties).Memory is matched a g a i n s t t h e (JIMMY3 KNOW PERSON) p a t t e r n which y i e l d s t h e match (JIEIMY3 KNOWS BRANDT@MAxwELL)' . Output procedures then cdnvert t h e match i n t o -t h e given response.NO.The search using (JIMMY3 KNOW BILLeMAXWELL) A t t h e t i m e t h i s dialogue was produced, t h e only meani n g of KNOW contained i n memory was (PERSON KNOW PERSON). Therefore, t h e reasonable p a r s e of (PERSON KNOW THING) was not found. DO u n t i l person i s through t a l k i n g :(1) . Request u s e r i n p u t and t r a n s l a t e English words i n t o i n t e r n a l. codes (?EMORY node numbers).(2) . P a r s e i n p u t t o c r e a t e t h e "best" p a r s e network(s).(3) . Match each p a r s e network with s t r u c t u r e s i n memory t o produce. t h e "best" match(es) .. Produce a response based on t h e memory match(es).To make t h e p r o c e s s i n g of t e x t more e f f i c i e h t , English words and p u n c t u a t i o n a r e t r a n s l a t e d i n s t e p (1) i n t o node p o i n t e r s . Undefined | The data s t r u c t u r e used t o drive the parser is the c r i p l e (ACTION GENERIC node) which s p e c i f i e s the semantics f o r the major components of the parse. By applying the t r i p l e a s a template t o the input, the ACTOR,ACT and OBJECT can be identified.As the input is parsed, i t s meaning is converted' i n t o a parse network and a network "score" is calculated Usually there a r e several parse networks constructed from a s i n g l e input representing d i f f e r e n t meanings of that input. The b e s t parse is t h a t one which has t h e highest score from i t s This is very important l a t e r during the matching of input to MEMORY where compatibility between the two is necessary.The parser has been developed t o correctly handle restricted forms of W;questions and declarative sentences. The question mark and period a r e the only terminating punctuation symbols currently allowed. A l l of these components a r e expanded i n Table 2 i standardize" it and b) t h e determination of t h e type of i n p u t recelved so t h e proper p a r s i n g technique can be selected.-37-The f i r s t operation perfonaed on t h e input is the t r a n s l which an item can occur plus its replacement form.This p a r t of the parsing algorithm 2s where the kznd of input, i.e., DO-question, IS-question, Wh-question, declarative statement, etc., is deter mined. The Find r i g h t modification (prepositional phrase) f o r the ACTOR. The input ACTOR is the name of a set f 6 r which t h e MEMORY ACTOR is a member, i. e. , MEMORY ACTOR (ISA) input ACTOR. 8. (-I-0) ACTOR missing from e i t h e r input o r MEMORY. . . (6) . . S e l e c t an ACTOR t o match on.. . (7) ..-• (8).. . . 2. Use a l l nodes above the ACTOR i n its hierarchy.OBJECT b u t is a n element from t h e same s e t as t h e i n p u t OBJECT. T h i s w i l l be t h e c a s e when matching o b j e c t s such as GFEEN I300K w i t h YELLOW BOOK o r OLD BOOK w i t h NEW BOOK. t h e ACTOR o r ACT does n o t match, t h e n p r i n t the-whole MEMORY node given by t h e memory match s t r u c t u r e .ACTOR o r OBJECT a r e generated by t h e following procedure.1. Add t h e m n e r i c node f o r t h e ACTOR (OBJECT) t o t h e end of t h e o u t p u t production list. S e t i t s f u n c t i o n type t o SUBJECT (OBJECT). 1. JIbIMY3 and t h e person's name g e t t r a n s l a t e d t o the pronouns "I" ind "YOU". A t t h i s s t a g e , t h e form of t h e pronoun may b e wrong.The input ACTOR is a member of the s e t named by the MEMORY ACTOR, i. e., input ACTOR (ISA) MEMORY ACTOR.4. Use a l l synonyms of the input ACTOR. I f any a r e found, r e v e r s e t h e s e t t i n g of t h e mode. For example, t h e i n p u t "DID LIRANDT PLAY OUTSIDE?" which would match t h e MEMORY node, {' BRANDT PLAYED INSIDE." would have a n e g a t i v e mode since INSIDE and OUTSIDE a r e mutually e x c l u s i v e . | ACT (main verb). 1. uon't use an a r t i c l e i f t h e word modified is a pronoun or a proper name.Following the completion of t h i s operation t h e output is printed. 3. I f t h e memory search f a i l s and t h e r e was no a c t o r i n memory, t h i s should b e reported e x p l i c i t l y r a t h e r than saying "I IXN'T KMW." For example, consider t h e question "DOES RILL MAXWELL HAVE A SON?" I f BILL@MN(WI?LL is n o t p r e s e n t i n MEMORY, r e p o r t "I DON' T KNOtJ BILL MAXJELL .I1 However, i f t h e memory match procedures were s u c c e s s f u l i n f i n d i n g g e n e r i c information such as "bEN HAVE SONS" then t h e phrase "BUT IIE COULD HAVE A SON." could a l s o be generated. Why do you want to know? NCP: I fear I may be responsible for the debt.I thought that might be the casq. NCP: You mean I do owe you money?No, it's just that I regret making you feel uncomfortable. NCP: Then why did you ask if the window was expensive to repair?To test your power of deduction. NCP: You really don't seem to understand me.Computer analysis of such complex interchanges is dependent upon the existence of psychological models of both parties. Successful realization is probably many years off. The first problem concerns the appropriate response to the question:"What is Charles Dickens' phone number?" The normal human response of "That's a stupid question." or "Phones weren't invented then. requires the action of a plausibility check on the question before an attempt is made to find an answer. = ((1,s) where G is an arbitrary phrase structure grammar and S is a formal semantics (defined in the paper). S may be either context free or context sensitive. S models the following notion of meaning: the meaningful units where w i s a sentence of G and m is a meaning assigned to w by S.We prove the following results: The set of phrase structure languages is just the set of products of r.e. sets. Every phrase structure language has a description using a regular grammar and a context free semantics. For every description D with an qnrestricted grammar and context sensitive semantics there is a description D' using a context free grammar and context free semantics such that L(D) = L(D').Furthermore, D and Q are "strongly equivalent" in the sense that the phrase trees assigned by Dl to each sentence are just the skeleton trees of the phrase structures assigned by D to the sentence. The notions of "weak" and "strong equivalence" are extended to semantics (if two descriptions are strongly equivalent in a semantic sense, then the structure of their semantic functions is identical -in a programming sense, the same programs can be used to compute the meanings of the same ~"entencesj. In this sense, D and D' are not strongly equivalent. However, if D has a context: free semantics, then D and D' are semantically strongly equivalent. Also, we prove that there is a description Dm' for L(D) using a context sensitive semantics which is strongly equivalent to D in both the syntactic and semantic senses, Next we define translation on phrase structure languages and consider a particularly appealing strategy for translation, which we call "syntax-controlled" translation. (I have avoided the term "syntax-directed" because it has had differing uses in t11e li~erature.) We prove the following results: Eveiy computable translation is definable as a syntax-controlled translation. For two arbitrary descriptions D and D', it is undecidable whether any syntax-controlled translation from L(D) to L(D1) exists. We give an algorithm which, given two arbitrary descriptions D and D', will halt and produce the definition The AuRnented F i n i t e S t a t e E-lachine - added procedural s t a t e n e n t s . I n t h i s system procedural statements have been added t o a much simpler mechanism, the f i n i t e s t a t e nachine, The advantame of t h e AFSIf approach is t h a t the phonetic s t r i n g may be processed i n a s i n g l e l i n e a r pass proposition stated i n the complement as derived from the f a c t i v i t y of the verb (in these s # :ies the reader must infer that everything that mother says i s true!). Lexical entries for main verbs t h a t take predicate complements contain pointers to the implication class. These relations can then be expanded t o give the proper axiom schemes.The use of lexical relations allows us to,express both syntactic and semantic information i n a form that i s compact, easy to retrieve, and that provides effective .input to both parsing and deductive processes. Thus given (11, a l a t e r expectancy of (4) would be s e t up by (3) (4) John was not hungry. But systemic knowledge contains generalizations, not inviolable t r u t h s , and t h e inference may not be valid. This can be marked by the use of an adversative connective: Our basic approach has been to combine careful structural distinctions with a mixed recognition strategy. The central distinction is in the way that utterances can be related to the methods in the dialogue model. First, an utterance (called an initiator) may Introduce a method that corresponds to one of the standard activities in an environment (for example, asking a question at an information desk or requesting help from a consultant). I 1 imprecision" and "ambiguity" of natural language usage in procedural donlains Potentially, such an investigation could result in an alternative to formal programming languages for the linguistic man-machine interface -e, g., NaturalLanguage Procedure Specification.We report on our progress tp date in the analysis of a corpus of recipes from ' I' he JOY of Cooking. Our present understanding of the communication process in recipes is that the imperative verb is a call to some procedure which returns a case-frame into which are mapped the remaining object-group and verb-qualifier elements of the surface text. We present statistics concerning case frequencies, syntactic structures, and word usage, and we detail our approach for the automatic comprehension and symbolic modelling of the activities invblved in recipe execution (we are using Heidornts NLP LISP system). Otliez examples i l l u s t r a t i r~g the diskinctic-n be.tween "in" and I1onft and tlie meaning of that elusive adverb "even" will be presented. . MEMORY STRUCTURE(ACMR = +3, OBJECT = +3, A= + 1, RED = +1, and BOOK = -1) .x & a p a r t can a l s o be expressed compactly u s i n g t h e T r e l a n o n . Wea f s o need t o make deductions from main v e r b s i n p r e d i c a t e complement: cons t r u c t i o n s ; deduction6 such a s ,the speaker's view of t h e t r u t h of t h eSecond, an utterance may correspond to a step in a standard path In a method already underway; here, a standard path is a normally expected succession of activity steps. Third, m utterance may be part of recovery discussion, which Is generated when when some violation of standard expectations occurs, necessitating clarification, correction, etc. Flnally, an utterance may belong to mdadiscussion, a relatively constrained class whose function is to lay out the context for other utterances so that these may be identified w~th the appropriate method step. Given the static model of dialogue embodied in the methods, the problem is to find the correct method step that relates to a particular input. We handle this problem by deflning a set of special structures to aid in matchag, by using the methods to generate expectations dynamtcally, and by differentiating overall matching strategies according to the four utterance classes described.T h e ideas presented here have been implemented In a prototype system called Susie Software, whtch is embedded in OWL-IThe OWL system is currently under development in the Knowledge-Bases Systems Group at the M.I.T. Laboratory for Computer Science. This research was supported by the Advanced Research P r o F u Agency of the Department of Defense and was monitored by the Office of Naval Research under Contract Number N00014-75CMj61. | Problems t o be solved before "natural" sounding output can be produced) 1. I f input was not parsed, respond with "1 DON' T UNDERSTANDTHAT." Then p r i n t a l l undefined words i n the form: I DON' T KNOW IdliAT word1 (OR trord2 OR w r d 3 . . . ) IIISANS . " For example, the input "DOES BRANDT RIDE A BICYCLE?" would generate t h eresponse "I DON' T UUDERSTAND THAT. I DON'T KNOW WHAT RIDE OR BICYCLE I W J S ."2. I f t h e input i s parsed, but t h e r e was a poor, o r no, memory match, then check t h e ' GENERIC node f o r t h e ACT t o s e e i f we respond with "YES", "EIO", "I DON' T KNOW" o r some "canned" response. What is searched f o r is a n a t t r i b u t e which is sometimes present i n ACT nodes and i s t h e answer t o be given when no match i s found with kCMORY. An example is the. ACT "KNOW". I f t h e program i s asked "DO YOU KNOW x?" and x i s n o t i n MEtlORY, then the response w i l l be "NO." r a t h e r than "I DON' T KNOW.I f the i n p u t is parsed and t h e r e is a good memory match then t h e following s t e p s w i l t be executed. | null | Main paper:
introduction:
There a r e a number of e x t a n t computer programs which i n t e r a c t i n t e l l i g e n t l y w i t h human i n t e r r o g a t o r s , b u t a l l do s o i n a way w e c h a r a c t e r i e e a s "unnatural". By "unnatural," we mean t h a t they conv e r s e i n a way s i g n i f i c a n t l y u n l i k e two normal a d u l t humans do. For example, Winograd's SHRBLU [19] seems c h i l d -l i k e , and i t s h a r e s w i t h t h e woods' moonrocks system [20] t h e problem of being task-const rained.Weizenbaum's ELIZA [16] As an example of a "natural" and a n "unnatural" d i a l o g u e , imagine two computer programs (CPs) conversing w i t h a human (H) i n which each CP a l r e a d y knows "THE BOY O~JNS A BALL." One CP is "natural" (NCP); t h e o t h e r i s "unnatural" (uCP) .The boy l i v e s on Main S t r e e t . UCP:. By ' t h e boy', I assume you mean t h e one who owns t h e r e d b a l l . NCP: O.K.llhete does t h e boy p l a y with h i s b a l l ? UCP : I don' t krtow. NCP: I don't know, but I assume near h i s residence on Main S t r e e t .A house is on Main S t r e e t . UCP: O.K. The b a l l broke a window.UCP: By ' t h e Ball, ' 1 assume you mean t h e red one the boy owns. BY 'a window,' I assume i r i s a p a r t of t h e house i n Main S t r e e t .Was t h e window expensive t o r e p a i r ? UCP: I don't know. NCP: S i~c e t h e house was expensive t o c o n s t r u c t , I assume t h a t i t s windows a r e expensive t o repair. To g e t a f e e l i n g f o r t h e types of responses t h e currentJy Implemented program CBn produce, t h e following s h o r t dialogue i s presented. WThe dialogue i s given i n upper c a s e w i t h commentary i n lower case.The j: and p: ( i d e n t i f y i n g JZMMY3 and person) were added liere f o r c l a r i t y .JIMMY3 must f i n d o u t who i t is t a l k i n g with i n orderr t o t r a n s l a t e t h e pronoun "I" t o f a c i l i t a t e memory searching. p : BdANDT MAXWELL.BR@DT@MAXWELL i s rtcognized as a l e g i t i m a t e name. The @ i n BRANDT@EWLL is used t o i n d i c a t e t h e combining o f two o r more words t o form a Separate erttity. I n t h e above example, t h e names BRANDT and W E L L a r e recognized as f i r s t and l a s t names, respectively, t h a t , when appearing together, a r e i n t e r p r e t e d a s t h e f u l l name, [email protected], YOU.Parsing is accomplished by matching i n p u t t o templates c o n s i s t i n g of (ACTOR ACT OBJECT) describing t h e d i f f e r e n t meaning senses f o r main verbs. T h i s question is parsed using t h e (PERSON KNOW PERSON) d e f i n i t i o n of KNOW. On i n p u t , YOU is t r a n s l a t e d t o JIEIMY3 (which has ISA PERSON a s one of i t s properties).Memory is matched a g a i n s t t h e (JIMMY3 KNOW PERSON) p a t t e r n which y i e l d s t h e match (JIEIMY3 KNOWS BRANDT@MAxwELL)' . Output procedures then cdnvert t h e match i n t o -t h e given response.NO.The search using (JIMMY3 KNOW BILLeMAXWELL) A t t h e t i m e t h i s dialogue was produced, t h e only meani n g of KNOW contained i n memory was (PERSON KNOW PERSON). Therefore, t h e reasonable p a r s e of (PERSON KNOW THING) was not found. DO u n t i l person i s through t a l k i n g :(1) . Request u s e r i n p u t and t r a n s l a t e English words i n t o i n t e r n a l. codes (?EMORY node numbers).(2) . P a r s e i n p u t t o c r e a t e t h e "best" p a r s e network(s).(3) . Match each p a r s e network with s t r u c t u r e s i n memory t o produce. t h e "best" match(es) .. Produce a response based on t h e memory match(es).To make t h e p r o c e s s i n g of t e x t more e f f i c i e h t , English words and p u n c t u a t i o n a r e t r a n s l a t e d i n s t e p (1) i n t o node p o i n t e r s . Undefined
g e t person ahd number of t h e pronouns t o a g r e e w i t h t h e:
ACT (main verb). 1. uon't use an a r t i c l e i f t h e word modified is a pronoun or a proper name.Following the completion of t h i s operation t h e output is printed. 3. I f t h e memory search f a i l s and t h e r e was no a c t o r i n memory, t h i s should b e reported e x p l i c i t l y r a t h e r than saying "I IXN'T KMW." For example, consider t h e question "DOES RILL MAXWELL HAVE A SON?" I f BILL@MN(WI?LL is n o t p r e s e n t i n MEMORY, r e p o r t "I DON' T KNOtJ BILL MAXJELL .I1 However, i f t h e memory match procedures were s u c c e s s f u l i n f i n d i n g g e n e r i c information such as "bEN HAVE SONS" then t h e phrase "BUT IIE COULD HAVE A SON." could a l s o be generated. Why do you want to know? NCP: I fear I may be responsible for the debt.I thought that might be the casq. NCP: You mean I do owe you money?No, it's just that I regret making you feel uncomfortable. NCP: Then why did you ask if the window was expensive to repair?To test your power of deduction. NCP: You really don't seem to understand me.Computer analysis of such complex interchanges is dependent upon the existence of psychological models of both parties. Successful realization is probably many years off. The first problem concerns the appropriate response to the question:"What is Charles Dickens' phone number?" The normal human response of "That's a stupid question." or "Phones weren't invented then. requires the action of a plausibility check on the question before an attempt is made to find an answer. = ((1,s) where G is an arbitrary phrase structure grammar and S is a formal semantics (defined in the paper). S may be either context free or context sensitive. S models the following notion of meaning: the meaningful units where w i s a sentence of G and m is a meaning assigned to w by S.We prove the following results: The set of phrase structure languages is just the set of products of r.e. sets. Every phrase structure language has a description using a regular grammar and a context free semantics. For every description D with an qnrestricted grammar and context sensitive semantics there is a description D' using a context free grammar and context free semantics such that L(D) = L(D').Furthermore, D and Q are "strongly equivalent" in the sense that the phrase trees assigned by Dl to each sentence are just the skeleton trees of the phrase structures assigned by D to the sentence. The notions of "weak" and "strong equivalence" are extended to semantics (if two descriptions are strongly equivalent in a semantic sense, then the structure of their semantic functions is identical -in a programming sense, the same programs can be used to compute the meanings of the same ~"entencesj. In this sense, D and D' are not strongly equivalent. However, if D has a context: free semantics, then D and D' are semantically strongly equivalent. Also, we prove that there is a description Dm' for L(D) using a context sensitive semantics which is strongly equivalent to D in both the syntactic and semantic senses, Next we define translation on phrase structure languages and consider a particularly appealing strategy for translation, which we call "syntax-controlled" translation. (I have avoided the term "syntax-directed" because it has had differing uses in t11e li~erature.) We prove the following results: Eveiy computable translation is definable as a syntax-controlled translation. For two arbitrary descriptions D and D', it is undecidable whether any syntax-controlled translation from L(D) to L(D1) exists. We give an algorithm which, given two arbitrary descriptions D and D', will halt and produce the definition The AuRnented F i n i t e S t a t e E-lachine - added procedural s t a t e n e n t s . I n t h i s system procedural statements have been added t o a much simpler mechanism, the f i n i t e s t a t e nachine, The advantame of t h e AFSIf approach is t h a t the phonetic s t r i n g may be processed i n a s i n g l e l i n e a r pass proposition stated i n the complement as derived from the f a c t i v i t y of the verb (in these s # :ies the reader must infer that everything that mother says i s true!). Lexical entries for main verbs t h a t take predicate complements contain pointers to the implication class. These relations can then be expanded t o give the proper axiom schemes.The use of lexical relations allows us to,express both syntactic and semantic information i n a form that i s compact, easy to retrieve, and that provides effective .input to both parsing and deductive processes. Thus given (11, a l a t e r expectancy of (4) would be s e t up by (3) (4) John was not hungry. But systemic knowledge contains generalizations, not inviolable t r u t h s , and t h e inference may not be valid. This can be marked by the use of an adversative connective: Our basic approach has been to combine careful structural distinctions with a mixed recognition strategy. The central distinction is in the way that utterances can be related to the methods in the dialogue model. First, an utterance (called an initiator) may Introduce a method that corresponds to one of the standard activities in an environment (for example, asking a question at an information desk or requesting help from a consultant). I 1 imprecision" and "ambiguity" of natural language usage in procedural donlains Potentially, such an investigation could result in an alternative to formal programming languages for the linguistic man-machine interface -e, g., NaturalLanguage Procedure Specification.We report on our progress tp date in the analysis of a corpus of recipes from ' I' he JOY of Cooking. Our present understanding of the communication process in recipes is that the imperative verb is a call to some procedure which returns a case-frame into which are mapped the remaining object-group and verb-qualifier elements of the surface text. We present statistics concerning case frequencies, syntactic structures, and word usage, and we detail our approach for the automatic comprehension and symbolic modelling of the activities invblved in recipe execution (we are using Heidornts NLP LISP system). Otliez examples i l l u s t r a t i r~g the diskinctic-n be.tween "in" and I1onft and tlie meaning of that elusive adverb "even" will be presented. . MEMORY STRUCTURE(ACMR = +3, OBJECT = +3, A= + 1, RED = +1, and BOOK = -1) .x & a p a r t can a l s o be expressed compactly u s i n g t h e T r e l a n o n . Wea f s o need t o make deductions from main v e r b s i n p r e d i c a t e complement: cons t r u c t i o n s ; deduction6 such a s ,the speaker's view of t h e t r u t h of t h eSecond, an utterance may correspond to a step in a standard path In a method already underway; here, a standard path is a normally expected succession of activity steps. Third, m utterance may be part of recovery discussion, which Is generated when when some violation of standard expectations occurs, necessitating clarification, correction, etc. Flnally, an utterance may belong to mdadiscussion, a relatively constrained class whose function is to lay out the context for other utterances so that these may be identified w~th the appropriate method step. Given the static model of dialogue embodied in the methods, the problem is to find the correct method step that relates to a particular input. We handle this problem by deflning a set of special structures to aid in matchag, by using the methods to generate expectations dynamtcally, and by differentiating overall matching strategies according to the four utterance classes described.T h e ideas presented here have been implemented In a prototype system called Susie Software, whtch is embedded in OWL-IThe OWL system is currently under development in the Knowledge-Bases Systems Group at the M.I.T. Laboratory for Computer Science. This research was supported by the Advanced Research P r o F u Agency of the Department of Defense and was monitored by the Office of Naval Research under Contract Number N00014-75CMj61.
(+4):
The input ACTOR is a member of the s e t named by the MEMORY ACTOR, i. e., input ACTOR (ISA) MEMORY ACTOR.4. Use a l l synonyms of the input ACTOR. I f any a r e found, r e v e r s e t h e s e t t i n g of t h e mode. For example, t h e i n p u t "DID LIRANDT PLAY OUTSIDE?" which would match t h e MEMORY node, {' BRANDT PLAYED INSIDE." would have a n e g a t i v e mode since INSIDE and OUTSIDE a r e mutually e x c l u s i v e .
parsing strategy:
The data s t r u c t u r e used t o drive the parser is the c r i p l e (ACTION GENERIC node) which s p e c i f i e s the semantics f o r the major components of the parse. By applying the t r i p l e a s a template t o the input, the ACTOR,ACT and OBJECT can be identified.As the input is parsed, i t s meaning is converted' i n t o a parse network and a network "score" is calculated Usually there a r e several parse networks constructed from a s i n g l e input representing d i f f e r e n t meanings of that input. The b e s t parse is t h a t one which has t h e highest score from i t s This is very important l a t e r during the matching of input to MEMORY where compatibility between the two is necessary.The parser has been developed t o correctly handle restricted forms of W;questions and declarative sentences. The question mark and period a r e the only terminating punctuation symbols currently allowed. A l l of these components a r e expanded i n Table 2 i standardize" it and b) t h e determination of t h e type of i n p u t recelved so t h e proper p a r s i n g technique can be selected.-37-The f i r s t operation perfonaed on t h e input is the t r a n s l which an item can occur plus its replacement form.This p a r t of the parsing algorithm 2s where the kznd of input, i.e., DO-question, IS-question, Wh-question, declarative statement, etc., is deter mined. The Find r i g h t modification (prepositional phrase) f o r the ACTOR. The input ACTOR is the name of a set f 6 r which t h e MEMORY ACTOR is a member, i. e. , MEMORY ACTOR (ISA) input ACTOR. 8. (-I-0) ACTOR missing from e i t h e r input o r MEMORY. . . (6) . . S e l e c t an ACTOR t o match on.. . (7) ..-• (8).. . . 2. Use a l l nodes above the ACTOR i n its hierarchy.OBJECT b u t is a n element from t h e same s e t as t h e i n p u t OBJECT. T h i s w i l l be t h e c a s e when matching o b j e c t s such as GFEEN I300K w i t h YELLOW BOOK o r OLD BOOK w i t h NEW BOOK. t h e ACTOR o r ACT does n o t match, t h e n p r i n t the-whole MEMORY node given by t h e memory match s t r u c t u r e .ACTOR o r OBJECT a r e generated by t h e following procedure.1. Add t h e m n e r i c node f o r t h e ACTOR (OBJECT) t o t h e end of t h e o u t p u t production list. S e t i t s f u n c t i o n type t o SUBJECT (OBJECT). 1. JIbIMY3 and t h e person's name g e t t r a n s l a t e d t o the pronouns "I" ind "YOU". A t t h i s s t a g e , t h e form of t h e pronoun may b e wrong.
overview:
Problems t o be solved before "natural" sounding output can be produced) 1. I f input was not parsed, respond with "1 DON' T UNDERSTANDTHAT." Then p r i n t a l l undefined words i n the form: I DON' T KNOW IdliAT word1 (OR trord2 OR w r d 3 . . . ) IIISANS . " For example, the input "DOES BRANDT RIDE A BICYCLE?" would generate t h eresponse "I DON' T UUDERSTAND THAT. I DON'T KNOW WHAT RIDE OR BICYCLE I W J S ."2. I f t h e input i s parsed, but t h e r e was a poor, o r no, memory match, then check t h e ' GENERIC node f o r t h e ACT t o s e e i f we respond with "YES", "EIO", "I DON' T KNOW" o r some "canned" response. What is searched f o r is a n a t t r i b u t e which is sometimes present i n ACT nodes and i s t h e answer t o be given when no match i s found with kCMORY. An example is the. ACT "KNOW". I f t h e program i s asked "DO YOU KNOW x?" and x i s n o t i n MEtlORY, then the response w i l l be "NO." r a t h e r than "I DON' T KNOW.I f the i n p u t is parsed and t h e r e is a good memory match then t h e following s t e p s w i l t be executed.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 576 | 0 | null | null | null | null | null | null | null | null |
c0a56a924e6ac8ffd3733960f8c9973259b0250f | 219308387 | null | Explanation Capabilities of Production-Based Consultation Systems | A computer program t h a t models an expert in a g i v e n domain f a more l i k e l y to be accepted by experts i n t h a t domain, and by non-experts seeking i t 9 advice, i f the system c a n explain i t s actions. An explanation c a p a b i l i t y n o t o n l y a d d s to t h e system s c r e d i b i l i t y , but also enables t h e non-expert user t o l e a r n from i t . Furthermore, clear e x p l a n a t i o n s a l l o w an e x p e r t t o cbeck the system's " r e a s o n i n q f l , p o s s i b l y d i s c o v e r i n g the need f o r r e f i n e s e n t s and additions t o t h e system s knowledge b a s e u I n a d e v e l o p i n q system, a n e x p l a n a t i o n c a p a b i l i t v can be u s e d as a d e b u q ~l n q aid t o verify | {
"name": [
"Scott, A. Carlisle and",
"Clancey, William J. and",
"Davis, Randall and",
"Shortliffe, Edward H."
],
"affiliation": [
null,
null,
null,
null
]
} | null | null | null | 1977-02-01 | null | null | null | null | null | null | A computer program t h a t models an expert in a g i v e n domain f a more l i k e l y to be accepted by experts i n t h a t domain, and by non-experts seeking i t 9 advice, i f the system c a n explain i t s actions.An Scope of MYCIN'S E x p l a n a t i o n Capability . . rn . . program-generated e x~l a n a t i o n s . The purpose of an explanation c a~a b i l i t v (KC) i s t o qive the user access t o a s much of the system s knowledge as posbible. I d e a l l y , i t should be easy for a user t o get a complete, understandable answer t o anv sort of question about t h e system's knowledge and pera at ion -both i n general, and w i t h reference t o a particular consultation. This l l p l i e s three major goals i n t h e development of an explanation capability. What decision ~t made about aome subproblem Why i t didn t use a certain plece of information Why i t failed t o ma!ce a certain decision Why it required a certain piece o f infornatlon Why it didn t require a certain ~i e c e of information How i t w i l l find out a certain plece of Information [while tRe consultatinn i s i n oroqress] What the system i s currently doing? [while t h e consultation is i n ~r o a r e s s ] The m e c i f i c s e t of explanation types which are chosen as basics, however, w i l l depend on the particular system.2) To enable t h e user t o get an explanation which answers t h e question completely and comprehensively.3) To make the EC easy t o use. A novice should be able t o use t h e EC without f i r s t spending a large amount of time lehrninq how t o request explanations. r e a s o n i n g -s t a t u s checker, a n d i t s g e n e r a l . q u e s t i o n -a n s w e r e r w i l l o n l y be a c c e s s i b l e a t t h e t e r m i n a t i o n of the c o n s u l t a t i a n .An EC must know what is i n t h e system's knowledge base, and how it r u l e s Also may be necessary t o t h e extent t h a t t h e c o n t e n t of a r u l e may e x p l a i n why it was necessary to u s e t h i s rule, or may affect which futupe rules w i l l be t r l e d .A g e n e r a l question-answerer will need more ~n f o m a t l o n about t h e system since t h e scope o f i t s e x p l a n a t i o n s is much broader. i t s t a s k is t o answer g e n e r a l questions about Lhe system's knowledge base. To do this, i t must know how the system stores knowledge about i t s area of e x p e r t i s e ( t h e static kdowledge with which it starts each c o n s u l t a t i o n ) and how i t stores facts gathered d u r i n g a p a r t i c u l a r c o n s u l t a t i o n (its dynamlc knowledge).These two types o f information will a l l o w a GQA t o answer g u e s t i o n s about t h e substance and e x t e n t of the ~r o d u c t i o n system's c u r r e n t knowledge.I f a n e x p l a n a t i o n c a p a b i l i t y also i s t a p The 6 , 31, 351, 39, 4 1 , 42, 44, 347, 49 The following r u l e s could have b?en used to determine t h a t t h e i d e n t i t y of ORGANISM4 is s t r e p t o c o c c u s : RULEO33. However, none of these succeeded i n t h e context o f ORGANISM-1. I f you would l i k e an e x p l a n a t i o n f o r why any of these rules failed, please enter their numbers:*# 33 Clause 2 of rule033 ["the morpholoqy of the orqanasm i s coccusw]was already known t o be false for ORGANISM-1, so t h e rule was never tried.It is definite t h a t the morphology of ORGAVISY-1 i s rod. mentioned t h e r e l e v a n t c o n t e x t .There are i n t e r a c t i o n s among t h e d i f f e r e n t types of questions which weren't mentioned e a r l i e r for t h e sake of c l a r i t y . Sometiaes a q u e s t i o n r a y i n d i c a t e a n incorrect assumption about the s a t e of MYCIN'S knowledge. For examole a q u e s t i o n could a s k "Whv don't you think that ORGANISM-1 is E.coli?" when, i n fact, the system has concluded t h a t the organism is E.coli. To answer t h i s q u e s t i o n , t h e e x p l a n a t i o n system would explain how i t | The f i n a l t y p e o f knowledge t h a t some g e n e r a l q u e s t~o n -a n s w e r i n g t h e v a l u e is known w i t h c e r t a i n t y o r there are no r u l e s l e f t t o use. t h e n bo compare t h e answers t o these questions.The ourpose of t h e explanation system is .[preceded by the first 14 q u e s t i o n s in t h e c o n s u l t a t i o n ] . The f o l l o w i n g were used: [ 3.1 1 RULE027 n d iWHAT Each word in the dictionary has a synonym p o i n t e r t o i t s t e r m i n a l word (terminal words p o i n t t o themselbes). F o r the purpose o f analyzing t h e q u e s t i o n , a non-terminal word is considered t o be equivalent t o its | Main paper:
other domain-independent knowledge:
The f i n a l t y p e o f knowledge t h a t some g e n e r a l q u e s t~o n -a n s w e r i n g t h e v a l u e is known w i t h c e r t a i n t y o r there are no r u l e s l e f t t o use. t h e n bo compare t h e answers t o these questions.The ourpose of t h e explanation system is .[preceded by the first 14 q u e s t i o n s in t h e c o n s u l t a t i o n ] . The f o l l o w i n g were used: [ 3.1 1 RULE027 n d iWHAT Each word in the dictionary has a synonym p o i n t e r t o i t s t e r m i n a l word (terminal words p o i n t t o themselbes). F o r the purpose o f analyzing t h e q u e s t i o n , a non-terminal word is considered t o be equivalent t o its
determining what pieces of knowledge are relevant:
The 6 , 31, 351, 39, 4 1 , 42, 44, 347, 49 The following r u l e s could have b?en used to determine t h a t t h e i d e n t i t y of ORGANISM4 is s t r e p t o c o c c u s : RULEO33. However, none of these succeeded i n t h e context o f ORGANISM-1. I f you would l i k e an e x p l a n a t i o n f o r why any of these rules failed, please enter their numbers:*# 33 Clause 2 of rule033 ["the morpholoqy of the orqanasm i s coccusw]was already known t o be false for ORGANISM-1, so t h e rule was never tried.It is definite t h a t the morphology of ORGAVISY-1 i s rod. mentioned t h e r e l e v a n t c o n t e x t .There are i n t e r a c t i o n s among t h e d i f f e r e n t types of questions which weren't mentioned e a r l i e r for t h e sake of c l a r i t y . Sometiaes a q u e s t i o n r a y i n d i c a t e a n incorrect assumption about the s a t e of MYCIN'S knowledge. For examole a q u e s t i o n could a s k "Whv don't you think that ORGANISM-1 is E.coli?" when, i n fact, the system has concluded t h a t the organism is E.coli. To answer t h i s q u e s t i o n , t h e e x p l a n a t i o n system would explain how i t
:
A computer program t h a t models an expert in a g i v e n domain f a more l i k e l y to be accepted by experts i n t h a t domain, and by non-experts seeking i t 9 advice, i f the system c a n explain i t s actions.An Scope of MYCIN'S E x p l a n a t i o n Capability . . rn . . program-generated e x~l a n a t i o n s . The purpose of an explanation c a~a b i l i t v (KC) i s t o qive the user access t o a s much of the system s knowledge as posbible. I d e a l l y , i t should be easy for a user t o get a complete, understandable answer t o anv sort of question about t h e system's knowledge and pera at ion -both i n general, and w i t h reference t o a particular consultation. This l l p l i e s three major goals i n t h e development of an explanation capability. What decision ~t made about aome subproblem Why i t didn t use a certain plece of information Why i t failed t o ma!ce a certain decision Why it required a certain piece o f infornatlon Why it didn t require a certain ~i e c e of information How i t w i l l find out a certain plece of Information [while tRe consultatinn i s i n oroqress] What the system i s currently doing? [while t h e consultation is i n ~r o a r e s s ] The m e c i f i c s e t of explanation types which are chosen as basics, however, w i l l depend on the particular system.2) To enable t h e user t o get an explanation which answers t h e question completely and comprehensively.3) To make the EC easy t o use. A novice should be able t o use t h e EC without f i r s t spending a large amount of time lehrninq how t o request explanations. r e a s o n i n g -s t a t u s checker, a n d i t s g e n e r a l . q u e s t i o n -a n s w e r e r w i l l o n l y be a c c e s s i b l e a t t h e t e r m i n a t i o n of the c o n s u l t a t i a n .An EC must know what is i n t h e system's knowledge base, and how it r u l e s Also may be necessary t o t h e extent t h a t t h e c o n t e n t of a r u l e may e x p l a i n why it was necessary to u s e t h i s rule, or may affect which futupe rules w i l l be t r l e d .A g e n e r a l question-answerer will need more ~n f o m a t l o n about t h e system since t h e scope o f i t s e x p l a n a t i o n s is much broader. i t s t a s k is t o answer g e n e r a l questions about Lhe system's knowledge base. To do this, i t must know how the system stores knowledge about i t s area of e x p e r t i s e ( t h e static kdowledge with which it starts each c o n s u l t a t i o n ) and how i t stores facts gathered d u r i n g a p a r t i c u l a r c o n s u l t a t i o n (its dynamlc knowledge).These two types o f information will a l l o w a GQA t o answer g u e s t i o n s about t h e substance and e x t e n t of the ~r o d u c t i o n system's c u r r e n t knowledge.I f a n e x p l a n a t i o n c a p a b i l i t y also i s t a p
Appendix:
| null | null | null | null | {
"paperhash": [
"davis|an_overview_of_production_systems",
"colby|pattern-matching_rules_for_the_recognition_of_natural_language_dialogue_expressions"
],
"title": [
"An overview of production systems",
"Pattern-Matching Rules For The Recognition Of Natural Language Dialogue Expressions"
],
"abstract": [
"Abstract : Since production systems were first proposed in 1943 as a general computational mechanism, the methodology has seen a great deal of development and has been applied to a diverse collection of problems. Despite the wide scope of goals and perspectives demonstrated by the various systems, there appear to be many recurrent themes. This paper is an attempt to provide an analysis and overview of those themes, as well as a conceptual framework by which many of the seemingly disparate efforts can be viewed, both in relation to each other, and to other methodologies. Accordingly, the authors use the term 'production system' in a broad sense, and attempt to show how most systems which have used the term can be fit into the framework. The comparison to other methodologies is intended to provide a view of PS characteristics in a broader context, with primary reference to procedurally-based techniques, but with reference also to some of the current developments in programming and the organization of data and knowledge bases.",
"Gramhrar-heeocl p u r h e r s whi c h par form a uord-byword , p a r t so f-s p c o ~ t l p s y c h i a t r i c i n t e r v i e w s. The a l g o r i t h m uoeo p a t t e r n-m a t c h i n g r u l e s u h i c h a t t e m p t t o c h a r a c t e r i z e ths i n p u t e x p r e s s i o n s by p r o g r e s e i v e t y t r a n s f o r n ~ l n g them i n t o p a t t e r n s u h l c h match, c o n l p l e t e l y o r f u z z i I y , a b s t r a c t s t o r e d p a t t e r n s. The methods u t i l i z e d a r e general and c o u l d s e r v e any \" h o s t \" s y s t e m w h i c h take3 n a t u r a l language i n p u t. Summary o f a tncthod f o r t r a n s f o r m i n g n a t i ~ r a l i6hgq~9r! i n p u t expressions u n t i l a ~ ~ c r t t e r n i s o b t a i n e d uhich completely o r f u z z i l y matches a more a b s t r a c t s t o r e d p a t t o r n ."
],
"authors": [
{
"name": [
"Randall Davis",
"Jonathan J. King"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Kenneth Mark Colby",
"R. C. Parkison",
"Bill Fought"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null
],
"s2_corpus_id": [
"53531425",
"62309250"
],
"intents": [
[],
[]
],
"isInfluential": [
false,
false
]
} | null | 576 | null | null | null | null | null | null | null | null | null |
50a001da61d508d7cfa61c084d3d44c8d68671ee | 219300712 | null | A Goal Oriented Model of Human Dialogue | This section describes our Dialogue-game Model at i t s current state of dcvclopmcnt. I t starts w i t h a brdef overview of dialogue and how it i s structured, t h e n describes the dominant knowledge structures w h i c h guide the model, and finally dcscribcs a set of processes which apply these knowledge structures t o text to comprehend i t Within the mb.dcl., each participant in a dialogue i s simply pursuing his own goals of t h c moment. The t w o participants interact smoothly because the conventions of communication coordinate their goals and give them continuihg reasons t o speak and listen. These goals have a number of attributes which are not necessarily consequences of c i t h c r human activity in general, or communication in particula'r; but which are nonetheless characteristic of human communication i n the form of dialogue: 1. Goals are cooperatively esta5lished. Bidding and acceptance activities serve to intfoduce goals. 2. Goa/s.aremufua//yknown. E a c h p a r t y a s s u m e s o r c o m e s t p k n o w goals of the othcr, and each interprets the entire dialogue relative to currently known goals. 3. Goalsareconfieu~edbyconvenfion. Setsofgoalsforusein dialogue (and othcr lwguage use as well) are tacitly'known and employed by all competent spe;l&rs o f t h e language. 4. Goa/s are bilateral. Each dialogue participant assumes goals complementary to those of his partner. Gas/ssreubiguilous. A h o a r e r v i e w s t h c s p e ~k e r a s a l w a y s having goals hc i s pursuing by speaking. Furthermore, the hearer recognizes and uses thcsc goals as part of his understanding of the utterance. An ~ninlerrupted dialogue goes through three phases: establishing goals, pcir sui ng eobl s, dccommitting from goals. A Madcl of Dialogue Rcsults: RO-1 i s an activation of the loft half of Rulel. Case/l corrcsponds to (L asks about a proposition) Case12 = (How does L get Runoff working) corresponds to the proposition. An activation of Rule1 i s created in the WS. | {
"name": [
"Moore, James A. and",
"Levin, James A. and",
"Mann, William C."
],
"affiliation": [
null,
null,
null
]
} | null | null | null | 1977-09-01 | 14 | 0 | null | by providing a framework for integrating the comprehcnsion of an utterance w i t h that of i t s prcdcccssors. Recently, we have propased (Leuin & Moore, 1976 : 1977 Rr Lcvin, 1977) multi-scntential knowledge units that are specified primarily by the speaker's and hcarcr's goals.Thcsc goal-oriented units, which w e call Dialogue-gsmcs [l] , specify the kinds of language interactions in w h i c h people engage, r a t h c r than the spccific content of thcsc intcractions. Pcoplc use l a n g u a~c primarily t o comrnunicatc with other pcoplc l o achieve their own goals. Thc Dialoguc-game mu1 ti-scntontial structures wcrc dcvcloped to represent this knowledge about language and how i t can be uscd to achicve goals.- [I] Thc term "Oialoguc-game" was adopted by analogy from Wittgcnstcin's term (Searle, 1969) . The direct comprehension of these sentences fails to d e r i v e the main communicative~effcct. For example, declarative scntenccs can be used to seek information ("1 nced to know your Social Security number."): questions can be u-scd to convey information ("Did you know that John and Harriet got married?") or t o request an action ("Could you pass the salt?''). These kinds of utterances, w h i c h have b c c n extensively analyzed b y philosophers of language (Austin, 1962 : Searle, 196 9, 1975 : Grice, 1975 , are not handled satisfactorily b y any of the current theories of t h e d i r c c t comprchcnsion of language. However, these indirect language usages ara widespread in naturally occurring language--even two-year-old children can comprehend indirect requests for action almost as well as dircct requests (Shatz, 1975) .O n e theory proposcd to account for these indirect uses of language i s based on the concept of "convcrsotional postulates" (Grice, 1975 : Gordon 1 975formalized and tested this model, and found that people's rasponse times tend to support a three-stage model (deriving the l i t e r a l mcaning, check its plausibility and, i f implausible, dcriving the "intended" meaning" from convcrsational rules). I n general, this approach to i n d i r e d speech acts i s infc~ence-bascd, depending on t h e application of conversational rules to infer the indirect meaning from the d i r c c t mcaning and the context. A different approach has been proposcd b y L s b o v~f i F a n s~c l (1 974) and by Levin & Moore (1976 : 1977 . Multi-sentential knowledge, organizing a scgmcnt of language interaction, can form the basis for deriving the indikect effect of u t t c r~n c c w i t h i n the segment.For example, a multi-sentential structure for an information-seeking interaction can sypply the appropriate context for interpreting the subscqucnt utterances to s~c k and t-hen supply information. The infcrcncc-bascd approach rcquircs one set of convcrs~tional rulc-, for information requests, a dif fcrcnt 9ct of r u l c s for answers to these rcquc:ts, and a way t o tic thcnc t w o rulc sets together. The Dialogue-game model postulates a single k n o w l c d~e struclurc for this kind of interaction, w i t h coopcrating proccssc; for: (1) rccognizinp; when this kind of interaction i s proposcd,(2) using this knowlcdgc to comprchcnd uttcranccn within its scope, and (3) identifying w h e n the interaction i s to be terminated, A M o d c l of DialogueOur t h c o r y of human language use has bccn strongly influenced by w o r k in human p r o b l e m solving (Ncwcll XI Simon, 1972) in which the bchavior of a human i s modeled as an information. processing system, having goals to pursue and selecting actions w h i c h tend t o schicvc thcsc goals. Wc view humans as engaging in linguistic bchavior in order t o advance the state of certain of thcir eoals. Thcy dccide to use language, they sclcct ( o r accept) t h c other participant for a dialogue, they choose the details of linguistic c x p r c s s i o n -all with the expectation that some of their desired state specifications can thcrcby be rcalizcd,In this thcory of lancuagc, a participant i n a linguistic exchange views the other as an indcpcndcnt information-processing system, w i t h separate knowledge, goals, abilities and acccss l o the world. A spcsker h a s a range of potcntial changes he can cffcct i n h i s l i~t c n c r , a corresponding collection of linguistic actions which may result in each such chance, and some notion of the conscqucnccs of performing each of these. T h e spcokcr may view the hcarcr as a resource for information, a potential actor, or as an object to bc moldcd into sorrrc dcsircd state.A dialogue involves t w o speakers, who altcrnatc as hearers. I n choosing t o initiate o r conlinuc tho cxchany,~, a participant attcmpts to satisfy his own goals: in intcrprcting on uttcrancc of his partner, each participant attcmpts to find the way in w h i c h that utterance serves the goals of his partner. Thus a dialoguo continues because the participants continue to scc it as furthering thcir own goals. Likewise, w h e n the dialoguc no l o n~o r serves the goals of one of the participants, i t i s redirected to new goals o r tcrminatcd.this rrlcchanism of joint interaction, via cxchange of uttcranccs, i n pursuit of d c s i r c d t.itcs, i s uscful for ochiovihg ccrtain relatcd pairs +of participanls' ~o a l s ( c .~. , I t v~r n i rlr./tcact~inc, buyinc/sc\ling, gctting h c i p /~i v i n g hclp, ...). Many of thcsc paired sets of I correspond to hichly structured collections of knowlcdgc, shorcd by thc rncrnbcrr, o f thc langunpc community. Thcsc ~ollcctions specify such things as: 1) what chnractcri:tics an individual must h a v c to cngagc in a dialogue of this sort, 2) how t h i s dialocuc i s initiated, pursued and tcrminatcd, 3) what ranee .of infarmation can bo comrnunicotcd imp1 icitl y , and 4) undcr what Circumstances tho dialoguo will "succeed" ( s c r v c tho function for which it was initiated) and how this &ill bo cxhibitcd in the participants7 bck~uvior.VJo h~~o allr:mptc:d to rrproscnt those collr:.ctiong_of knowlcdgc and tho wily in w t~i c h t h c y arc u:cd to f a c i l i t~t o tho cornprchcnsion of a diala~ua, in tha O i n l o e~~o t;nrno Mod(-? I.This section describes our Dialogue-game Model at i t s current state of dcvclopmcnt. I t starts w i t h a brdef overview of dialogue and how it i s structured, t h e n describes the dominant knowledge structures w h i c h guide the model, and finally dcscribcs a set of processes which apply these knowledge structures t o text to comprehend i t Within the mb.dcl., each participant in a dialogue i s simply pursuing his own goals of t h c moment. The t w o participants interact smoothly because the conventions of communication coordinate their goals and give them continuihg reasons t o speak and listen. These goals have a number of attributes which are not necessarily consequences of c i t h c r human activity in general, or communication in particula'r; but which are nonetheless characteristic of human communication i n the form of dialogue:Goals are cooperatively esta5lished. Bidding and acceptance activities serve to intfoduce goals.Goa/s.aremufua//yknown. E a c h p a r t y a s s u m e s o r c o m e s t p k n o w goals of the othcr, and each interprets the entire dialogue relative to currently known goals.Goalsareconfieu~edbyconvenfion. Setsofgoalsforusein dialogue (and othcr lwguage use as well) are tacitly'known and employed by all competent spe;l&rs o f t h e language.Goa/s are bilateral. Each dialogue participant assumes goals complementary to those of his partner. | null | having goals hc i s pursuing by speaking. Furthermore, the hearer recognizes and uses thcsc goals as part of his understanding of the utterance.An ~ninlerrupted dialogue goes through three phases: establishing goals, pcir sui ng eobl s, dccommitting from goals.Typically this sequcncc i s repeated several times over the coursc of a few rninutcs.We havc crcotcd knowlcdse structurcs to rcprescnt these convcntions,, and proccsscs to apply the conventions to actual dialo~ucs to comprehend them; Since the knowledsc structures dominatc all of the activity, they are described first. The assimilation of an uttcranco in the dialogue i s rcprcscntcd in this model by a sequence of modifica\ions of a "Work~pacc"[2] which rcprcscnfs the attention or awareness af the listening party. Tho modificN~tions arc roughly cyclic:1. A ncw item of text f i s brought into attention through the "Par scr." [-21 2. Interpretive conscqucnces~of T are developed in the Workspace by a variety of proccsscs.3. An exprcssian E appears in thc Wor'kspace w h i c h specifics the relation between i and the imputed goals of the spcaker of T.This final cxprcssion i s of coursc a formal expression in the knowledge representation of the modcl. E rcprcsents the proposition (held by the hcarer) that in uttering T, the spcaker was performing an act in pursuit of G, a-spbaker's goal known to thc hcarer. Sucrcssful comprchcnsion i s cquatcd with relating tcxt to salisf action of spcakcr's goals.To makc an explicit account of dialoguc in this way, wc now describc the knowledge structures that rcprcscnt those c~~nvcgtions which supply tho goals for the participants to pursue. I n particular, wc will anewcr thc following thrco questions: Thc dialogue types w c have represented so far as Dialogue-games have each r e q u i r e d only thrcc Parameters: the two participants involved (called "Roles"), and the subjcct of the dialogue (called "Topic").Onc of t h e major aspects distinguishing various types of d'ialogucs i s the set of goals hcld by t h e participants. Another such aspect i s the set of kno'wledgc states of the participants. We have found that each type of dialogue has a char$cteristic set of eaal and knowledge states of the participants, vis-a-vis each other and the subject. W i t h i n the formalism of the Dialogue-game, these are called the Parameter Spccifications, and a r e r c p r p s c n t e d b y a collection of predicates on the Parameters.These Spccifications are known to the participants of t h e dialogue, and the requirement that they be satisfied during the conduct of a game i s used by thp participants t o signal what Dialogue-games they wish to conduct, t o recognize what game i s b e i n g bid, t o dccide how t o respond to a bid, to conduct the game once t h e bid i s accepted, and t o terminate t h e garno when appropriate. These Spccifications also provide the means with w h i c h t o explain the implicit, but clearly succcssful, communication which accompanics any natural d i a l o g~c . Examples and discussions of these Specifications will accompany the f o l l o w i n g description of !he Helping-game.W h i l e the Paramctsr Spccificstions represent those aspects of a dialogue t y p e that r c m a i n constant throughout the course of a dialogue of that type, we h a v e also found t h a t certain aspects change i n systematic ways. Thcse are reprcsented i n Dialogue-games a5Components. I n the Dialogue-games we have developed so far, the Components a r e r c p r c s c n t c d 9s a set of participants' subgoals, partially ordered i n time.Eiddinp, and Acccptancc arc entry operations which people use t o c n t c r Dialo~uc-games. Bidding 1. identifies thc game, 2. indicates the bidder s interest i n pursuing t'hc game, 3. idcn tifies the Psramctcr configuration intcnded.Bidding i s performed many diffcrcnt ways, often very bricfly. I t i s typically t h e source of a great deal of implicit comrnunicotion, since a bricf bid can cornmunicatc all of the Pararncters and thcir Specifications f o r j h e Dialogue-game being bid.Acceptance i s one of tho typical responses to a Bid, and leads to p u r w i t of t h c game. Acccptsncc cxhibi t : .1. acknov~lcdy,rncnt that 3 bid t~s s hccn rnndc, Once a eamc has bccn bid and acccptcd, the two participants each pursue t h e subgoals spccificd for their role by the Components of this game. Thcse subgoals are mutually complcn~cntary, each set facilitating the other. Furthermore, by the time t h e tcrrninati on stage has been rcachcd, pursuit of the Component-specified subgoals will have assurcd satisfaction of the higher, initial goals of the participants, for which the Came was initiated in thc first place.I n this scction, wc c.xhibit a specific Diolocuc-game: the Helpinggn/ne. This game i s prcscntcd in an informal rcprcsentation, in order to emphasize the informational content, rather than the representational power of our formalism. Later in this report we will prcscnt thc formal analocue of this same game. I n what follows, the b o l d face indicates the information contained in tho representation of this particular Dialogue-game: the tcxt in regular type i s explanatory commentary.The (annotated) Helping-game. - - - - - - - - 1 - - - 1 1 - - - - - - - - - - - - - - - -A Modcl of Dialogue Thcse Spccifications not only constrain who would qualify as filling the rolc of HELPEE, but also provide reliable information about t h e HELPEE, given that this individual i s believed to be engaged i n the Helping-game. This prohibits someone from asking for help on a problem he did not want solved. Similarly, i f one r c c~i v c s what he judges to be a sincere request for help to do some task, the helper normally as:umes that the requester has the necessary authority to do the task, i f only he knew how.HELPER: able to provide help.So, in ordcr to be a HELPER, an individual must be willing and able t o provide the needed assistance. Since this Dialogue-game rcprcscnts shared knowledge, the HELPER knows t h e s e Spccifications, and therefore will not bid the Helping-game to someone who i s not likely to meet them. And similarly, no one who fails to meet these Specifications (and knows he fails) w i l l accept a bid for the Helping-game with himself as HELPER.Thcre are three components: the first two constitute the "Diagnosis" phase to communicate what the problem is.. HELPEE wants HEfPFR to know about a sef of unexcepfiona/, acfuuii/ events.'The HELPEE sets up a context by describing a situation where everything, so far, i s going well. Since the HELPEE assumes that the TASK i s understood by the HELPER, he also assumes that the HELPER shares his expectations for r~bsequent activity. This pattern of a Helping-game i s sufficiently well known to the participants, that the HELPEE almost never needs t o actually ask a question at this point. By simply exhibiting a failure of expectation, the HELPEE has communicated that this acts as a block to his successfully pursuing the TASK. The HELPER i s expected to explain w h y the failure occurred and how HELPEE canavoid it or otherwise continue in the TASK.The third componcnt specifies the 'Treatment" phase where the HELPER communicates an explanation for the perceived failure.3 HELPER wants HELPEE fo know about an action which w i l l avoid the undesired event or cause the desired one.The context description enables the HELPEE to identify a collection of activities which he understands, and in which the HELPEE i s attempting to participate.The violation-of-expectation description points out just where the HELPEE's image of the activities differs from the correct image. I t i s from this area of difference that the HELPER selects an action for the HELPEE. 1) nomination, 2) recognition, 3) instantiation, 4) conduct, 5) termination.Our description of the model should be viewed as representing the changing coy,nitive state of one of the participants, throunhout the course of the dialogue. That is, two models are involved, one for each participant. Since the same processing occurs for both, we w i l l describe only one.T h c Dialogue-Game Modcl consists of a Long-Term Memory (LTM), a Workspacc (WS), and a set of proccsscs that modify the contents of WS, contingent upon the contents of LTM and WS. LTM conbins a rcprescntation of the knowledge that the partigular di ologuc participant bl ines to the dialogue b c f o~e it starts. This includcs knowlcdgc about t h e world, relevant objects, processes, concepts, the cognitive statc of h i s partner in dialogue, rules of inference and evidence, as well as linguistic knowlcdp;e (words and t h c i r semantic rcprcscntation, case frames for verbs and predicates and the multi-turn language s'trbctures, the Dialogue-games).WS i s the volatile short-term rncmory of thc modcl, containing all the partial and temporary rcsults of processing. The contcnte of WS at any momcnt rcprc.;cnt thc madel's state of comprchcnsion and focus at that point. Tho processes arc autonomous specialists, opcrl~tiny: indcpcndcntly and in parallel, to modify thqentitics in WS (callcd "activations"). Thcsc proccsscs a r c also influcnccd by the contents of WS, as well a. ; b y thc knowlcdgc in LTM. Thus, WS i s the place in which thcso concurrcnt~ly operating proccsscs interact with each othcr. This anarchistic control structure rcscrnblcs that Di aloguc-game suggests that Dialogue-game as a possi bilily for initiation.The Dialogue-Game Modcl has two ways in which these nominations of n c w Dialo~uc-games occur. One of the processes of the modcl i s a "spreading activation" p r o c c s s call& Protcus (Lcvin, 1976) . Protcus gcncratcs n e w activations i n WS on thc basic of cclations in LTM, from concepts (nodes i n the semantic network) that are already i n WS. Protcus brings into focus concepts somehow related to those already thcrc. A collection of concepts in WS leads to focusing on some aspect of a particular Dialocuc-game, in this sense "nominating" it as a possible new Dialoeue-game.MATCH and DEDUCE are two of thc modcl s processes which operatc in .conjunction to ccncratc ncw activations from existing ones, means of finding and ap,prp@rulc-like transformations, Thcy operate through partial match and plausible i nfcrencetcchniques, and i f thcy activate Pardrnetcrs, thcn the Dialogue-gsrnc that contains those ~a r a r n d t c r s bccomcs nomina,tcd as.a candidate Dialogue-game. Match and Deduce operate to,gether as a kind of production system (Newell, 1973) . "I tried to send a message to <person> at <computer-site3 and it~didn't go." the following t w o scqucnccs of associations and inferences result: ( l a ) I trjcd toX. (25) 1 wpntcd to X. (3a) 1'want to X. (4a) HELPEE wants to do TASK.( I b) It didn't go. (2b) What I tried to do didn't work. (3b) X didn't work. (4b) I can't X.(58) 1 don't know ha to X. (6b) HEL.PEE\doc$n7t know how to do TASK.A Model of Dialogue (Where: I = HELPEE and X = do TASK = send a message to <person> at <computer-site>.)A t t h i s point, (45) and (6b), since they are both Parameter Specifications for the Helping-game, cause the model to focus on this Dialogue-game, in effect nominating i t as an organizing structure for the dialogue being initiated.Thc proccsscs described so far are reasonably unselective and may activate a number of possible Dialogue-eamcs, some of which may be mutually incompatible or othcrwisc inappropriate.The Dialogue-garnc Manager investigates each of the nomi natcd Dial.oguc-games, verifying infcrcnccs based on the Parameter Specifications, and eliminating< those Dialogue-gamcs for which one or more Specifications are contraclictcd.A second rncchanism (part of Protcus) identifies those activations which are incornpati blc and scts about accumulating evidence in support of a decision to accept one and dclctc the rest from the WS.Fdr cxarnplc, suppose the question "How do I get RUNOFF to work?" lcads to the nomination of two games:Info-scck-game (pcrson asking question wants to know answer) andInfo-probe-game (pcrson asking q u c d i~n wants to k n~w if other knows answcr)Thcso I w o Dialocuc-~nrncs have a l o t in common hut differ in ono crucial aspect,: I n the Info-scck-gamc, thc qucctioncr docs not know the answcr to thc question, whilc i n thc Info-probc-game hc doc:. Thcsc two prcdicatcs arc rcprcscntcd in the Parameter Spc.cifications of t hc two Di al.ogue-games, and upon thcir joint nomination are discovarcd to bc contradictory. Prolcus rcprcsent: this d i~c o v e r y with a structure which ha5 the c f f c c t of climinatinp, thc conflicting Dialogue-came with the least supporting evidence. Such support might be, for cxamplc, cither the knowledge that the speaker i s t h~ hearcr's tcachcr or that hc i s a novice prograrnrncr (which would icnd support f6r tho choicc of t h e Info-prabc-errme or Info-seek-garnc, respectively).A Model of, DialogueT h r o u g h these proccsscs, the number of candidate Dialogue-games i s reduced until those remaining are rompatible w i t h each other and with the knowledge currently in WS and in LTM. To illustrate this, suppose that the following come to be represented i n WS (i.e., known) in t h e course of assimilating an utterance: SPEAKER does not know how to do a TASK. S P E A~E R wants to know how t o do that TASK. SPEAKER wants t-o do the TASK* Thcso a r e adequate to nominate the Helping-game. I n the process of instantiating t h i s Dialogue-game, the following predicates are added to WS: Thc model predicts that jhcsc predicates w i l l bo implicitly communicated by an utterance which r;uccecds in instantiating thc Helping-game. This corresponds t o a dialogue in which "&I can't gct this thing to work" i o taken to eommuhicate thot'thc speaker wnnts to "get this thing to work" (even,though, on the surface, i t i s only a simple declarative of the speaker9$ abi'lity). what he i s next to say: for thc hcarcr, these provide expectations for the functions to be served by the speaker s subscqucnt utterances.SPEAKERThcse "tactical" goals are central to our theory of language: an utterance i s not dccmcd ta be comprehended until some direct consequence of i t i s seen as serving a goal imputed to the spcakcr Furtherfiore, although the goals of the ~o m p o n c n i s arc active only w i t h i n the conduct ef a particular game, their pursuit leads to the satisfaction of t h e goals described in the Parameter §pccifications, which were held by the participants p r i o r t o the evocation cf ihe Dialogue-game.I n the case of the Helping-game, the goals i n tho "diagnostic" phase arc that thc tIELPEE dcscribc a scquoncc of related, uncxceptional cvcnte leading up to a failure of h i s cxpectotions. Thcse goals model the state cyf th6 HELPER as he assimilates this initial part of t h c dialogue, both in that h e knows how tho H E~P E E i s attempting l o dcscribc h i s problcrn, and also that thc HELPER knows whcn this phase i s past, and thc time has come (Ihc "trcatrncnt" phase) for him to provide the help which has been implicitly rcquesteel.The processes described above perform thc identification and pursuit of Dialo~ue-games. How, then, arc DGs terminated? Thc Parameter Specifications r c p r e s c n t thosc a:pects of dialogues that arc constant over that particular t y p e of dialogue. The Oinloguc-Game Modcl pushas this a step further in specifying that the Dislo~uc-,game continues only sr long a r the Parameter Specifications continuc t o hold.Whcnevcr any predicate i n the Specificelion ceases to hold, then tho model prcdicls tho i rnpcr~ding tcrrnin;jtion of this Dio1oguc;garnc.For r_.xnrnplct, if the IIIIPEF' na longer wants to pcrfarm thn TASK (r:~lhrlr b y nccornplid--~ir~f: it or b y nb:rndonirl~ that goal), hv 'indicl~trr.i fhia with nn irttcsriinct. wlrich hd; f o r tr:rrnin:il~on. The: lit,lp~ng game ihr:n tnrrnln:\tc~; this corrc;pr~nrl; lo lhq :imultsrir.ou~ tcrrr~instion of thc h o l p~n g interact~on. If the HELPER bccomc; unwilling l o give hclp, or discovtrrs that hc 1s unelble, thc:n Ihr! Eirlptng-game also terminiltcs. Again, we haddo one simplc rule ,lh?it corcr; h di.dcr:;!y of casos--a rule for tormination that captures tho variety uf w a y s that the dialogue? wa havo ~tudied end.The Dialogue -,name ProcessesIn this section we describe the major process elements of the Dialogue-Game Model. All the major parts and their connectivity are shown in Fieure 1. Thcse parts (two rncmorics and six Proccsscs) will each be described separately. Thc appendix contains an extensive, detailed trace of the model as i t analyzes (via hand simulation) a naturally occurring dialogue fragment. Finally, we will summarize our experience with the model to date.The Long-Term Memory i s the rnodcl's representation of a participant's knowledge of the external world. It contains the initial knowledge states of the participants: the grammatical case frames, the semantic structures for word-senses, tho knowledge of t h e s~b j c c t , m a t t e r of the dialopu~, the various ways in which dialoeues are structured, ctc.LTMis a semantic network, containing a set of nodes (also called concepts) and t h e relations that hold between them at the Iowost ievel. This information i s stored i n t h e form of triples:<node-1 relation node-2>Wc have this machinery encoded and working--a 6 1 1 complement of read and write primitives for this representation. However, it has proven awkward for us to specify knowlcdge at this level, so we have implemented further machinery (named SIM) t o tran.olatc n-ary predicates into these triples. Thuq far a predicate, P, having arguments A 1, A2, and A3, SIM can be given the input:PI: (Alpha P Beta Gamma)[ m c a~i n g that P1 i s defined to be an instance of P (the predicate always goes iR s w d position) w i t h arguments Alpha for A l , Beta for A2 and Gamma-?or A3. The Workspace is the model's representation for that information which the participant i s activcly using, This memory corresponds roughly ta a model of the participant's focus of attention.While the L'TM i s static during the operation of the model (we are not attempting to simulate learning), the WS i s extremely volatile, with elements (activations) coming into and out of focus c~t i n u o u s l y . All incoming sensations (i.e., utterances) appear in the WS, as do all augmentations of the participant's knowledge and goal statc. The representational format of the WS is the same as in LTM. Each node in the WS isa token (copy) of some node in LTM. Whenever some process determines that the model's attention (WS) should include a token of a specific node (C) from LTM, a new node (A) i s created by copying C and this new node is added to the WS. A i s referred to as an a r t l r , n t l n n n ( r a n r l i h r , a i w -This rcprescntatim providcs the associative links between an object i n attention, and the body of knowledge assbciated with it, but not yet broucH into attention.This module produces activations representing each successive utterance to be processed. These rcprcscntations are generated from the surface string using a standard ATN Grammar similar to those developed by Woods (19701 and Norman, Rumelhart, fir the LNR Research Group (1975) . We use a case grammar represeniation, w i t h each utterance spccificd as a main predicate with a set of parameters. Bccausc this module i s a conventional parser whose implementation i s well understood, we hove so farproduced hand parses of the input utterances, following an ATN grammar.This i s a sprcodine activation mechanism, which modifies thc activation of conccpts spccificd as rclatcd in LTM whenever a givcn c o n c~p t bccomcs active. This mcchanism provides a way to intcgratc top-down and bottom-up processing within a uniform framework (Lcvin, 1976) . The Dial ogue-Game iilodel uses Protcus to activate a conccpt, given tha a number of closcly relatcd conccpts (Componcnts, fcaturcs, instances, etc.) arc active.Protcus opcratcs on all c u r r w t activations to modify their salience", a numbcr associated with each activation that generally represents the importance or rclcvancc of t h e conccpt. Two kinds of influence relations can exist bctwccn conccpts: c x c i t c or inhibit.I f an excite r e l a t i~n exists, then Protcus increases the salience af the activation of that concept in proportion to the saliencc of the influencing conccpt. The higher the salience of an activation, the larger i t s influence on directly r e l a t e d conccpts. I f an inhibit relation i s spccificd. then Process decreases the salience of the activation of the neighboring conccpt.TI-is P r o c c x i idkntifics conccpts in LTM that arc congruent to cxistinc activationr. The Diologuc-Game Modcl conthins a numbcr of cquivalencc-like relations, which Mal'ch uses to idcntify a conccpt in LTM as r c p r c s c n t i n~ thc same thing as an activation of somo ~ccrningly different concept. Once this equivalent conccpt is found, i t i s activated. Dcpcnding on how this conccpt i o dcfincd in LTM, i t s activation may havo cffects. on othcr processes (for cxamplo, i f thc canccpt i s part of a rulc, Dcducc may bo invoked).in LTM which correspond, accordintta some set of critcrThe basic tactic i . , to ottrrrnpt A Model of Dialogue t o find a form of cquivalencc relationship between A and C, without delving into t h e i r structure at all. Only if this fails arc their respective substructures examined. I n t h s sccond case, the s a m e match which was attempted. at t h e top l e v e l i s t r i e d b c t w c c n corresponding subparts of A and C. Match proceeds in five steps:1. I s i t alrcady known that A i s an activation of C? I f so, the match ferminates with a positive conclusion, 2. Is there any other activation (A7j, and/or conccpt (C') such that A"is k n o w n t o bc a view of A, C 1s known ta bc a kind of C' , and A' i s known (by step 1) to bc an actlvation of C'? The relations (i.. i s a v i e w of ...) gnd (... isba kind of ...) r c p r e s c n t stored relations between pairs of activations 1 and' concepts, rcspcctivcly. One concept "is a kind of" another conccpt r e p ,~~~, l t s , a s~p c r c l a s s inclusion, t r u a for all timcand cdntexts. '(Whdever else h e might be, J o h n i s a k i n d o f huma'n being:) On the other hand, one activation may be "a view of" another only under certain circumstances--a conditional, or tactical relationship. Undcr diffcrent.conditions, if i s appropriate to v i e w John as a Husband, Father, Child; Hcl p-seeker, Advice-giver, e tc,A l i s t of matched pairs of activations and concepts reprcscnt corrcspondcnccs found el scwhmc, w i t h w h i c h match must be consi stcnt. (N.B.: t h i s Match, as we w i l l see later, may be i n service of anothcr Match galled' on siructuros containing the current A and C.) I f thc p a r [A,C] i s a matchcd pair, then these t y o have been previously found t o match,.so we may hcrc concl'udo the same thing and Match exits, 4. On the other hand, i f there i s either an X or a Y such that [A,X] (or [Y,C]) i s a matchcd pair, then replace this match with an attempt to match C and X (or A and Y). Clearly, for structures of significant complexity, Match may eventually call itc,clf rccursivcly, t o an arbitrory depth. However, since each subordinate call i s on a strictly smaller unit, this process must coitvcrge.Our experience has shown us that this type of mechanism plus a collection of rewrite rules enable us to eventually map a wide variety of input parsing structirrcs to pre-stored, abstract knowledge structures, in a way that a significant aspect o f their intended meaning has been assimilated in the process.This opcratcs to carry out a rule when that rule has become active. Rules are of the form (Condition)->(Action), and Dcduce scnscs thc activity of a rule and applies the rule by activating the concept for the action. Whatcver corresponocnces were evolved in the coursc.of cccating the activation of the condition (left) half of i h e rule are carried over into thc activation of the action (right) half. gives us the capability of a production syctin the WS. Dcducc attempts to match the left half of this rule with some other activation i n the WS. (This has ty-pically already been done by match.) Assuming this i s accomplished, Dcduce c r e s t e t an ac,tivalion of the right half of the rule, substi luting in the activation f o r all subparts for which thcke are correspondences with the i c f t half.Once a Dialogue-game has been activated (by Protcu.,) as possibly the comrnunrcation f o r m being bid for a dialogue, the Dialogue-game Manager uses i t to guide t h c assimilation of successive utterances of the dialogue, through four stages:1. establish the Parameter values and verify that no Specification i s contradictcd, 2. cstabli ;h olhcrwi sc unsupported Specifications as assumptions, 3. cstrrblish the Components as goal$ of thc participanfs, 4. dctcct the circumstancc~ which indicate that the Dialogue-game is.terminating and represent thc conscquenccs of this.Thc first two of thesc phaces hsppcn in parallel. Whcn the Manager accesses e a c h of thc Para~nctcrs, they arc found either to have activations in thc WS or not. I f they do, the cgrrespondcnccs bclwccn activation and Psrsmetcr are established in the WS. This correspond: l o a::ignint a value to thc Paramcicr for this particular evocation of thc Dialogue-game. Any Parameter that has no activation i c put on a list w h i c h i s A Model of Dialogue periodically chcckcd i n the hope that later activity by the Manager will lead t o the creation of appropriate activations.F o r each of the Specifitations, a check i s made to determine i f it already has an activation in WS. (In most cases, the activation of some of these Specifications will have led t o the activity of the Dialogue-game itself.) The Specifications having activations n e e d no further attention.For all remaining Spcc,ifications, activations are created substituting for t h e Parametcrs as determined above. At this stage, the Dialogue-game Manager calls Protcus to determine the stability of thcso new activations. Any new activation w h i c h contradicts existing activations will have its level of activity sharply reduced by Proteus. I f this happens, the Dialogue-game Manager concludes that some of the necessary preconditions for the game db not hold (are i n conflict with current understanding) and that this particular game should be abandoned. Otherwise, the new activaticns stand as new knowledge, following from the hypothesis that the chosen game i s appropriate, Finally, the Manager detects that one of the Specifications no longer appears to hold. This signals the impending termination of the Dialogue-game. I n fact, t h e utterance whikh contai'ns this information i s a bid to terminate. At this point, i f t h e partici pants7 initial goals are satisfied (thus contradicting the Specification which calls f o r t h e prcsence of those goals) the interaction ends "successfully".Otherwise, t h e Dialogue-game i s terminated for some other reason (e.g., one participant's unwillingness o r inability to continue) and would generally be regarded as a "failure". These consequences are infcrred by the Manager and added to the WS. When a Dialogue-game has terminated, i t s salience goes to zero and i t i s removed from the WS.The Dialogue-Gamo Model contains a set of Pronoun Processes, including an I-Process, a You-Process, and an It-Process. Each of those i s invoked whenever the /' associated surface w o r d appears in an input utterance,, and operates to identify some preexisting activationthat can be seen as a view of the same object.Each of these Processes search the curreot context, as represented by the current sct of actirations in the WS, using tho katures specified there ro identify a set of possible co-rcfctcntial expressions. When there is more than one possibility, the one with a hi ghcr salience i s sclectcd.With the understanding we new have of the multi-sentential aspects of human communication, it i s easy to see why man-machine cornrnun'ietion appears so alien, highly restrictive, uncomprehending and awkward. This i s because major regulation and interpret4 f ion structures are missing.In Table 1 , we compare human dialogue and typical man-machine com.munication with respect to some of these features, The table designates a "sender" ancfa "receiver" which should be identified with the person and the computer, respeclively, in the man-machine communication case. The ideal interface, arid the sort toward which this research i s direded, would be continuously askihg itself: "Why did he say that?". From answers to this, the interface would infcr just what the human was expecting as a response. This would con~titute a major s l c p toward the enabling t h o intcrface to servo the actual (rather than the poorly cxprcsscd) needs of the user. Finally, such an intorface would require much l c o~ adaptation on thc parf of the user, and so, by our original hypotheois, would significantly enhance the cffcctivencss of the man-machine partnership.This paper has described a research effort into the modeling of human dialogue. The purpose of this research has been to uncover and describe i n process models, reculari tics that occur in dialogue. I t i s hopcd that the enhqnced understanding of human communication which rcsul ts, will facilitate the development of more natural (and thus more effective) man-machine interfatcs.Thc principal regularity w'e have discovered i s a collection of knowledge and goal structures, called Dialoeue-games, which seem to be crucial in understanding the structure of naturally-occurring dialogues. According to the theory we have proposed, one or more of these Dialogue-games serve as the major organizing influence on e v e r y human didlogue.Each Dialogue-game specifies what knowledge each person must have to ehgage in such a dialogue, and what goals of the participants might be served by that interchange. A Dialogue-game also spccifies, as a sequence of "tactical" goals, the manner in which the dialogue i s conducted.The Diatogua-game Model i s a collection of cooperative processes which continuausly updated a representation of each participant's attention state in a Workspaco. The model recognizes when a particular Dialogue-game i s being bid, accepted, pursued and terminated, and represents these states appropriately in the Workspace. A particular Dialogue-game, the Helping-game, was described in some dctajl. A simulation of the evocation and use of the Helping-game on a segment of natural dialogue i s contained in the Appendix.Our experience so far with the Dialogue-game Model has reinforced our hypothcses that an understanding of the goal-serving aspects of dialogue i s a powerfdl . In this appendix we describe an extensive simulation of the a r r e n t state of the Dialogue-game Model. We make use of a particular version of the Helping-game and alsc explore another structure, an Execution Scene, which describes the customary events surrounding the successful execution of a particular program (Runoff).We start by describing this more detailed version of the Helping-game, introducing names f o r the various aspects, to be used later. Next we show a short, naturally occurring dialogue between a computer operator and a user. 'Then we describe t h e operation of the Dialoeuc-garnc'Model as if assimilates this dialogue, up to the point at w h i c h i t concludes that thc Helping-game i s an appropriate structure t h r o u~h which to understand the subr;cqucnt utterances.Once this hypothesis for the form of thedialogue has been chosen, we continue the simulation to examine how Jhc model dccidcs that a particular Execution Scene i s appropriate for assimilating the content of the dialogue. 'Next, we see how this choice of occnes cnhances the set of goals imputed to the speaker, thus facilitating the cornprehcnsion of what he i s saying. Finally, we summarize our experience with the Dialogue-game Model so far.What fol4ows i s the substance of the communication structure we have namcd the Hclping-game. In the interests of clarity af presentation, the formal structureo of the definition have been expressed in prose. However, the elements of the following description correspond one-to-one to those in the actual Helping-game used in ?the simulation.Thc parameters are two roles (HELPER and HELPEE) and a topic (TASK/HG).Parameter specifications:The HELPER and HELPEE are each a kind of person. H1 = A goal of t h e HELPEE is that h e perform TASK/HG. H2 = I t i s not true that HELPEE i s able to perform this TASKIHG. H5 = The HELPEE wants to be ablc to pcrform the TASKIHG. H I 1 = T h c HELPEE wants the HELPER to enable him to perform the TASKIHG.(bcing enabled to perform the task.is6a subgoal of performing the task) Game components:HGX 1 = The HELPEE knows of a particular execuiion scene, XS/HE.[note: a n execution scene i s a flowchart-like description of thc use af a particular process; more details below] (much more detail, below) whereupon the next segment is parsed and input for processing.HGX2 =H o w docs thc model know to evoke the Helping-game? To exhibit answers to t h i~ a n d subscqucnt questions, we lead tho reader through a simulation of the model as i t p r o c c s s c s the beginning of dialogue OC117. We indulge in fhe samc use of prorie f o r formalism as aboQe, again w i t h the same assuranaes of correspondcnccs with tho actual sirnulati on.Thc simulation proceeds in cycles: i n each cycle, we exhibit the operation of a sinzlc processor, performing one iteration of i t s function. We do not address h c r c the is:uc; of h o w the model would select w.hich processor to call next. In fact, our dcsign calls f o r these processors to be mgxirnally autonomous and parallel i n their operation, operating w h c n c v c r circumstances are ripe for their function and dormant otherwise. Thc format of this sirnuistion i s as follow;: Thc cycle number i s first, in the form: :cy,mcnt nurnbcr9--cycle number in this scgrncnt,.Next i s tho name of the p r a c c w o r operating in this cyclc. Aftcr that i s EI description of the nature of the pracossiny. donc d u P l n b that cyclc. Finally, tharo i s a l i s t of tha rcsults for this cycle, that is, ;dl tho irnportljnt c h a n g w i n WS, Cycle 1-1 --Parse.The parser reads one utterance/segment of input and translates it into the formalism f o r activations in the workspace. No claim 'is made that this translation retains all the content of the original text, only that i t is adequately faithful to the level of detail we a r e simulating.Results: Case/9 (= (0 perceives that L asks (how do I get Runoff working?))) is activated.Certain words (e,g. pronouns, determiners) are taken to be signals that a reference i s being made t o conccpts introduced elsewhere. Sne presence of a concept i n t h e workspoce corresponding to one of these words lcads to the calling of the. process-specialist which attempts t o resolve the implied reference. Thus, the presence of "I" i n the text leads to the calling of the I-process, whose sole function i s to determine t h e referent of the .'I" and modify the stored concept to reflect this. This process judges that i f L i s asking a question which contains "I" as its subject, then this constitutes adequate evidence to hypothesize that "I" is being used to refer to L.Results: 0 perceives that L asks (how does L get Runoff working?)Cycle 1-3 --Match Match i s always on the lookout for pairs of nodes, one in the WS and the other i n the LTM, such that the activation (node in WS) matches the concept (node i n LTM). T h i s i s taken to be evidence that the activation is also to be t i k e n as an activation of the matched concept. I t should be understood that we areaexamining only some of the succewful matches which occurred, Starting in this cycle, we see a pattern which recurs regularly, and which accounts fcr a significant piece of the action, as the model assimilates the dialogue. Match dcterrnincs that a particular activation matches the left half (condition side, i f part, etc.) of a production-like rule srorcd in LTM. This successful match leads t o the identification of the corrcspondcnces between the aspects of the activation and those of the left half of the rule, ae well as creating an activation of the rule itself. The activation of a rule leads to calling the Deduce processor in thenext cycle, which applies the activated rub t o the node i n the WS responsible for the rule's activation. This application of a rule (which also results in thc removal of the rule's activation from the WS) creates a new activation structure in the WS. Cycle 1-22 --ProteusAs a result of the numerous rcfarcncar to Runoff and XS/HE, tho activationo for thcsc two conccpts are "highly active". ~o n s c~u c n t l~, when Protcus io called, tho eonccpt XSjRO (the execution sccno of tho Runoff proccos) bocarnoo active and, duo to its similarity to XSfHE, i s taken to be equivalent to it. Sinca XS/RO is more dctailad (contein~ more information) than XS/HE, XS/RO is used in place i f XS/HE in all of the expressions introduced in Cycle 1-21.Something wo pnsscd ovcr in thc earlier examples was tha issuo of vyhcn tho modcl i s willine to stop processing a given piece of tcxl and eo on la the nexf onc. It scorns inappropriate l o demand that tho rncdol wring all possiblo information end deductions out of each utterance. Yet there must bo soms demands mada on tho assimilation. A n altcrnatc form of tho question is: what ncedr of his own does the hearer see the incoming text as potentially satisfyinc? We have taken the position that a hearer (tentatively) understands an utterance, when he successfully views i t as serving some goal imputed to the spcskcr. That is, to a first approximation, the hearer has assimilated an utterance if hc fisures out why thc spcakcr said it.Thc modcl has already established (HCX9 and HGX10, above) that L wants to dcscribc (implicitly, to 0) certain action; in XS/RO b a t L expected to perceive, and in s o n w csscs, did, Thus, in thc following uttcrsnces, we see the modol matching the parsed input structure with one of thcsc two goals, thus it i s sccn as bcing in service of a goal of thc spcakcr, and need bc examined no further (for tho time being).In thc subscqucnl example, we use two ncw rules: RS (Satisfaction) and RQ (Quicsccncc). RS dctcrrnincs when an uttcranco i s sccn to satisfy a speaker's goal and R Q , rcscts to this dcfectcd satisfaction by marking the utterance quicsccnt.(Opcrationolly, this means that in tho next cycle, thc Parser i s called to input the next scgmcnt af text.)Wc resume the example at the point where the first segment has been marked quicsccnt, and the Parser is called.Results: Casc9a = 0 pcrceivcs that L declares (I executed it).Cycle 2-2 --I-processoronto cithcr the line printer or another file. The execution scene of Runoff, as stored in our model, i s similar t o figure A-1. | null | null | Main paper:
gas/ssreubiguilous. a h o a r e r v i e w s t h c s p e~k e r a s a l w a y s:
having goals hc i s pursuing by speaking. Furthermore, the hearer recognizes and uses thcsc goals as part of his understanding of the utterance.An ~ninlerrupted dialogue goes through three phases: establishing goals, pcir sui ng eobl s, dccommitting from goals.Typically this sequcncc i s repeated several times over the coursc of a few rninutcs.We havc crcotcd knowlcdse structurcs to rcprescnt these convcntions,, and proccsscs to apply the conventions to actual dialo~ucs to comprehend them; Since the knowledsc structures dominatc all of the activity, they are described first. The assimilation of an uttcranco in the dialogue i s rcprcscntcd in this model by a sequence of modifica\ions of a "Work~pacc"[2] which rcprcscnfs the attention or awareness af the listening party. Tho modificN~tions arc roughly cyclic:1. A ncw item of text f i s brought into attention through the "Par scr." [-21 2. Interpretive conscqucnces~of T are developed in the Workspace by a variety of proccsscs.3. An exprcssian E appears in thc Wor'kspace w h i c h specifics the relation between i and the imputed goals of the spcaker of T.This final cxprcssion i s of coursc a formal expression in the knowledge representation of the modcl. E rcprcsents the proposition (held by the hcarer) that in uttering T, the spcaker was performing an act in pursuit of G, a-spbaker's goal known to thc hcarer. Sucrcssful comprchcnsion i s cquatcd with relating tcxt to salisf action of spcakcr's goals.To makc an explicit account of dialoguc in this way, wc now describc the knowledge structures that rcprcscnt those c~~nvcgtions which supply tho goals for the participants to pursue. I n particular, wc will anewcr thc following thrco questions: Thc dialogue types w c have represented so far as Dialogue-games have each r e q u i r e d only thrcc Parameters: the two participants involved (called "Roles"), and the subjcct of the dialogue (called "Topic").Onc of t h e major aspects distinguishing various types of d'ialogucs i s the set of goals hcld by t h e participants. Another such aspect i s the set of kno'wledgc states of the participants. We have found that each type of dialogue has a char$cteristic set of eaal and knowledge states of the participants, vis-a-vis each other and the subject. W i t h i n the formalism of the Dialogue-game, these are called the Parameter Spccifications, and a r e r c p r p s c n t e d b y a collection of predicates on the Parameters.These Spccifications are known to the participants of t h e dialogue, and the requirement that they be satisfied during the conduct of a game i s used by thp participants t o signal what Dialogue-games they wish to conduct, t o recognize what game i s b e i n g bid, t o dccide how t o respond to a bid, to conduct the game once t h e bid i s accepted, and t o terminate t h e garno when appropriate. These Spccifications also provide the means with w h i c h t o explain the implicit, but clearly succcssful, communication which accompanics any natural d i a l o g~c . Examples and discussions of these Specifications will accompany the f o l l o w i n g description of !he Helping-game.W h i l e the Paramctsr Spccificstions represent those aspects of a dialogue t y p e that r c m a i n constant throughout the course of a dialogue of that type, we h a v e also found t h a t certain aspects change i n systematic ways. Thcse are reprcsented i n Dialogue-games a5Components. I n the Dialogue-games we have developed so far, the Components a r e r c p r c s c n t c d 9s a set of participants' subgoals, partially ordered i n time.Eiddinp, and Acccptancc arc entry operations which people use t o c n t c r Dialo~uc-games. Bidding 1. identifies thc game, 2. indicates the bidder s interest i n pursuing t'hc game, 3. idcn tifies the Psramctcr configuration intcnded.Bidding i s performed many diffcrcnt ways, often very bricfly. I t i s typically t h e source of a great deal of implicit comrnunicotion, since a bricf bid can cornmunicatc all of the Pararncters and thcir Specifications f o r j h e Dialogue-game being bid.Acceptance i s one of tho typical responses to a Bid, and leads to p u r w i t of t h c game. Acccptsncc cxhibi t : .1. acknov~lcdy,rncnt that 3 bid t~s s hccn rnndc, Once a eamc has bccn bid and acccptcd, the two participants each pursue t h e subgoals spccificd for their role by the Components of this game. Thcse subgoals are mutually complcn~cntary, each set facilitating the other. Furthermore, by the time t h e tcrrninati on stage has been rcachcd, pursuit of the Component-specified subgoals will have assurcd satisfaction of the higher, initial goals of the participants, for which the Came was initiated in thc first place.I n this scction, wc c.xhibit a specific Diolocuc-game: the Helpinggn/ne. This game i s prcscntcd in an informal rcprcsentation, in order to emphasize the informational content, rather than the representational power of our formalism. Later in this report we will prcscnt thc formal analocue of this same game. I n what follows, the b o l d face indicates the information contained in tho representation of this particular Dialogue-game: the tcxt in regular type i s explanatory commentary.The (annotated) Helping-game. - - - - - - - - 1 - - - 1 1 - - - - - - - - - - - - - - - -A Modcl of Dialogue Thcse Spccifications not only constrain who would qualify as filling the rolc of HELPEE, but also provide reliable information about t h e HELPEE, given that this individual i s believed to be engaged i n the Helping-game. This prohibits someone from asking for help on a problem he did not want solved. Similarly, i f one r c c~i v c s what he judges to be a sincere request for help to do some task, the helper normally as:umes that the requester has the necessary authority to do the task, i f only he knew how.HELPER: able to provide help.So, in ordcr to be a HELPER, an individual must be willing and able t o provide the needed assistance. Since this Dialogue-game rcprcscnts shared knowledge, the HELPER knows t h e s e Spccifications, and therefore will not bid the Helping-game to someone who i s not likely to meet them. And similarly, no one who fails to meet these Specifications (and knows he fails) w i l l accept a bid for the Helping-game with himself as HELPER.Thcre are three components: the first two constitute the "Diagnosis" phase to communicate what the problem is.. HELPEE wants HEfPFR to know about a sef of unexcepfiona/, acfuuii/ events.'The HELPEE sets up a context by describing a situation where everything, so far, i s going well. Since the HELPEE assumes that the TASK i s understood by the HELPER, he also assumes that the HELPER shares his expectations for r~bsequent activity. This pattern of a Helping-game i s sufficiently well known to the participants, that the HELPEE almost never needs t o actually ask a question at this point. By simply exhibiting a failure of expectation, the HELPEE has communicated that this acts as a block to his successfully pursuing the TASK. The HELPER i s expected to explain w h y the failure occurred and how HELPEE canavoid it or otherwise continue in the TASK.The third componcnt specifies the 'Treatment" phase where the HELPER communicates an explanation for the perceived failure.3 HELPER wants HELPEE fo know about an action which w i l l avoid the undesired event or cause the desired one.The context description enables the HELPEE to identify a collection of activities which he understands, and in which the HELPEE i s attempting to participate.The violation-of-expectation description points out just where the HELPEE's image of the activities differs from the correct image. I t i s from this area of difference that the HELPER selects an action for the HELPEE. 1) nomination, 2) recognition, 3) instantiation, 4) conduct, 5) termination.Our description of the model should be viewed as representing the changing coy,nitive state of one of the participants, throunhout the course of the dialogue. That is, two models are involved, one for each participant. Since the same processing occurs for both, we w i l l describe only one.T h c Dialogue-Game Modcl consists of a Long-Term Memory (LTM), a Workspacc (WS), and a set of proccsscs that modify the contents of WS, contingent upon the contents of LTM and WS. LTM conbins a rcprescntation of the knowledge that the partigular di ologuc participant bl ines to the dialogue b c f o~e it starts. This includcs knowlcdgc about t h e world, relevant objects, processes, concepts, the cognitive statc of h i s partner in dialogue, rules of inference and evidence, as well as linguistic knowlcdp;e (words and t h c i r semantic rcprcscntation, case frames for verbs and predicates and the multi-turn language s'trbctures, the Dialogue-games).WS i s the volatile short-term rncmory of thc modcl, containing all the partial and temporary rcsults of processing. The contcnte of WS at any momcnt rcprc.;cnt thc madel's state of comprchcnsion and focus at that point. Tho processes arc autonomous specialists, opcrl~tiny: indcpcndcntly and in parallel, to modify thqentitics in WS (callcd "activations"). Thcsc proccsscs a r c also influcnccd by the contents of WS, as well a. ; b y thc knowlcdgc in LTM. Thus, WS i s the place in which thcso concurrcnt~ly operating proccsscs interact with each othcr. This anarchistic control structure rcscrnblcs that Di aloguc-game suggests that Dialogue-game as a possi bilily for initiation.The Dialogue-Game Modcl has two ways in which these nominations of n c w Dialo~uc-games occur. One of the processes of the modcl i s a "spreading activation" p r o c c s s call& Protcus (Lcvin, 1976) . Protcus gcncratcs n e w activations i n WS on thc basic of cclations in LTM, from concepts (nodes i n the semantic network) that are already i n WS. Protcus brings into focus concepts somehow related to those already thcrc. A collection of concepts in WS leads to focusing on some aspect of a particular Dialocuc-game, in this sense "nominating" it as a possible new Dialoeue-game.MATCH and DEDUCE are two of thc modcl s processes which operatc in .conjunction to ccncratc ncw activations from existing ones, means of finding and ap,prp@rulc-like transformations, Thcy operate through partial match and plausible i nfcrencetcchniques, and i f thcy activate Pardrnetcrs, thcn the Dialogue-gsrnc that contains those ~a r a r n d t c r s bccomcs nomina,tcd as.a candidate Dialogue-game. Match and Deduce operate to,gether as a kind of production system (Newell, 1973) . "I tried to send a message to <person> at <computer-site3 and it~didn't go." the following t w o scqucnccs of associations and inferences result: ( l a ) I trjcd toX. (25) 1 wpntcd to X. (3a) 1'want to X. (4a) HELPEE wants to do TASK.( I b) It didn't go. (2b) What I tried to do didn't work. (3b) X didn't work. (4b) I can't X.(58) 1 don't know ha to X. (6b) HEL.PEE\doc$n7t know how to do TASK.A Model of Dialogue (Where: I = HELPEE and X = do TASK = send a message to <person> at <computer-site>.)A t t h i s point, (45) and (6b), since they are both Parameter Specifications for the Helping-game, cause the model to focus on this Dialogue-game, in effect nominating i t as an organizing structure for the dialogue being initiated.Thc proccsscs described so far are reasonably unselective and may activate a number of possible Dialogue-eamcs, some of which may be mutually incompatible or othcrwisc inappropriate.The Dialogue-garnc Manager investigates each of the nomi natcd Dial.oguc-games, verifying infcrcnccs based on the Parameter Specifications, and eliminating< those Dialogue-gamcs for which one or more Specifications are contraclictcd.A second rncchanism (part of Protcus) identifies those activations which are incornpati blc and scts about accumulating evidence in support of a decision to accept one and dclctc the rest from the WS.Fdr cxarnplc, suppose the question "How do I get RUNOFF to work?" lcads to the nomination of two games:Info-scck-game (pcrson asking question wants to know answer) andInfo-probe-game (pcrson asking q u c d i~n wants to k n~w if other knows answcr)Thcso I w o Dialocuc-~nrncs have a l o t in common hut differ in ono crucial aspect,: I n the Info-scck-gamc, thc qucctioncr docs not know the answcr to thc question, whilc i n thc Info-probc-game hc doc:. Thcsc two prcdicatcs arc rcprcscntcd in the Parameter Spc.cifications of t hc two Di al.ogue-games, and upon thcir joint nomination are discovarcd to bc contradictory. Prolcus rcprcsent: this d i~c o v e r y with a structure which ha5 the c f f c c t of climinatinp, thc conflicting Dialogue-came with the least supporting evidence. Such support might be, for cxamplc, cither the knowledge that the speaker i s t h~ hearcr's tcachcr or that hc i s a novice prograrnrncr (which would icnd support f6r tho choicc of t h e Info-prabc-errme or Info-seek-garnc, respectively).A Model of, DialogueT h r o u g h these proccsscs, the number of candidate Dialogue-games i s reduced until those remaining are rompatible w i t h each other and with the knowledge currently in WS and in LTM. To illustrate this, suppose that the following come to be represented i n WS (i.e., known) in t h e course of assimilating an utterance: SPEAKER does not know how to do a TASK. S P E A~E R wants to know how t o do that TASK. SPEAKER wants t-o do the TASK* Thcso a r e adequate to nominate the Helping-game. I n the process of instantiating t h i s Dialogue-game, the following predicates are added to WS: Thc model predicts that jhcsc predicates w i l l bo implicitly communicated by an utterance which r;uccecds in instantiating thc Helping-game. This corresponds t o a dialogue in which "&I can't gct this thing to work" i o taken to eommuhicate thot'thc speaker wnnts to "get this thing to work" (even,though, on the surface, i t i s only a simple declarative of the speaker9$ abi'lity). what he i s next to say: for thc hcarcr, these provide expectations for the functions to be served by the speaker s subscqucnt utterances.SPEAKERThcse "tactical" goals are central to our theory of language: an utterance i s not dccmcd ta be comprehended until some direct consequence of i t i s seen as serving a goal imputed to the spcakcr Furtherfiore, although the goals of the ~o m p o n c n i s arc active only w i t h i n the conduct ef a particular game, their pursuit leads to the satisfaction of t h e goals described in the Parameter §pccifications, which were held by the participants p r i o r t o the evocation cf ihe Dialogue-game.I n the case of the Helping-game, the goals i n tho "diagnostic" phase arc that thc tIELPEE dcscribc a scquoncc of related, uncxceptional cvcnte leading up to a failure of h i s cxpectotions. Thcse goals model the state cyf th6 HELPER as he assimilates this initial part of t h c dialogue, both in that h e knows how tho H E~P E E i s attempting l o dcscribc h i s problcrn, and also that thc HELPER knows whcn this phase i s past, and thc time has come (Ihc "trcatrncnt" phase) for him to provide the help which has been implicitly rcquesteel.The processes described above perform thc identification and pursuit of Dialo~ue-games. How, then, arc DGs terminated? Thc Parameter Specifications r c p r e s c n t thosc a:pects of dialogues that arc constant over that particular t y p e of dialogue. The Oinloguc-Game Modcl pushas this a step further in specifying that the Dislo~uc-,game continues only sr long a r the Parameter Specifications continuc t o hold.Whcnevcr any predicate i n the Specificelion ceases to hold, then tho model prcdicls tho i rnpcr~ding tcrrnin;jtion of this Dio1oguc;garnc.For r_.xnrnplct, if the IIIIPEF' na longer wants to pcrfarm thn TASK (r:~lhrlr b y nccornplid--~ir~f: it or b y nb:rndonirl~ that goal), hv 'indicl~trr.i fhia with nn irttcsriinct. wlrich hd; f o r tr:rrnin:il~on. The: lit,lp~ng game ihr:n tnrrnln:\tc~; this corrc;pr~nrl; lo lhq :imultsrir.ou~ tcrrr~instion of thc h o l p~n g interact~on. If the HELPER bccomc; unwilling l o give hclp, or discovtrrs that hc 1s unelble, thc:n Ihr! Eirlptng-game also terminiltcs. Again, we haddo one simplc rule ,lh?it corcr; h di.dcr:;!y of casos--a rule for tormination that captures tho variety uf w a y s that the dialogue? wa havo ~tudied end.The Dialogue -,name ProcessesIn this section we describe the major process elements of the Dialogue-Game Model. All the major parts and their connectivity are shown in Fieure 1. Thcse parts (two rncmorics and six Proccsscs) will each be described separately. Thc appendix contains an extensive, detailed trace of the model as i t analyzes (via hand simulation) a naturally occurring dialogue fragment. Finally, we will summarize our experience with the model to date.The Long-Term Memory i s the rnodcl's representation of a participant's knowledge of the external world. It contains the initial knowledge states of the participants: the grammatical case frames, the semantic structures for word-senses, tho knowledge of t h e s~b j c c t , m a t t e r of the dialopu~, the various ways in which dialoeues are structured, ctc.LTMis a semantic network, containing a set of nodes (also called concepts) and t h e relations that hold between them at the Iowost ievel. This information i s stored i n t h e form of triples:<node-1 relation node-2>Wc have this machinery encoded and working--a 6 1 1 complement of read and write primitives for this representation. However, it has proven awkward for us to specify knowlcdge at this level, so we have implemented further machinery (named SIM) t o tran.olatc n-ary predicates into these triples. Thuq far a predicate, P, having arguments A 1, A2, and A3, SIM can be given the input:PI: (Alpha P Beta Gamma)[ m c a~i n g that P1 i s defined to be an instance of P (the predicate always goes iR s w d position) w i t h arguments Alpha for A l , Beta for A2 and Gamma-?or A3. The Workspace is the model's representation for that information which the participant i s activcly using, This memory corresponds roughly ta a model of the participant's focus of attention.While the L'TM i s static during the operation of the model (we are not attempting to simulate learning), the WS i s extremely volatile, with elements (activations) coming into and out of focus c~t i n u o u s l y . All incoming sensations (i.e., utterances) appear in the WS, as do all augmentations of the participant's knowledge and goal statc. The representational format of the WS is the same as in LTM. Each node in the WS isa token (copy) of some node in LTM. Whenever some process determines that the model's attention (WS) should include a token of a specific node (C) from LTM, a new node (A) i s created by copying C and this new node is added to the WS. A i s referred to as an a r t l r , n t l n n n ( r a n r l i h r , a i w -This rcprescntatim providcs the associative links between an object i n attention, and the body of knowledge assbciated with it, but not yet broucH into attention.This module produces activations representing each successive utterance to be processed. These rcprcscntations are generated from the surface string using a standard ATN Grammar similar to those developed by Woods (19701 and Norman, Rumelhart, fir the LNR Research Group (1975) . We use a case grammar represeniation, w i t h each utterance spccificd as a main predicate with a set of parameters. Bccausc this module i s a conventional parser whose implementation i s well understood, we hove so farproduced hand parses of the input utterances, following an ATN grammar.This i s a sprcodine activation mechanism, which modifies thc activation of conccpts spccificd as rclatcd in LTM whenever a givcn c o n c~p t bccomcs active. This mcchanism provides a way to intcgratc top-down and bottom-up processing within a uniform framework (Lcvin, 1976) . The Dial ogue-Game iilodel uses Protcus to activate a conccpt, given tha a number of closcly relatcd conccpts (Componcnts, fcaturcs, instances, etc.) arc active.Protcus opcratcs on all c u r r w t activations to modify their salience", a numbcr associated with each activation that generally represents the importance or rclcvancc of t h e conccpt. Two kinds of influence relations can exist bctwccn conccpts: c x c i t c or inhibit.I f an excite r e l a t i~n exists, then Protcus increases the salience af the activation of that concept in proportion to the saliencc of the influencing conccpt. The higher the salience of an activation, the larger i t s influence on directly r e l a t e d conccpts. I f an inhibit relation i s spccificd. then Process decreases the salience of the activation of the neighboring conccpt.TI-is P r o c c x i idkntifics conccpts in LTM that arc congruent to cxistinc activationr. The Diologuc-Game Modcl conthins a numbcr of cquivalencc-like relations, which Mal'ch uses to idcntify a conccpt in LTM as r c p r c s c n t i n~ thc same thing as an activation of somo ~ccrningly different concept. Once this equivalent conccpt is found, i t i s activated. Dcpcnding on how this conccpt i o dcfincd in LTM, i t s activation may havo cffects. on othcr processes (for cxamplo, i f thc canccpt i s part of a rulc, Dcducc may bo invoked).in LTM which correspond, accordintta some set of critcrThe basic tactic i . , to ottrrrnpt A Model of Dialogue t o find a form of cquivalencc relationship between A and C, without delving into t h e i r structure at all. Only if this fails arc their respective substructures examined. I n t h s sccond case, the s a m e match which was attempted. at t h e top l e v e l i s t r i e d b c t w c c n corresponding subparts of A and C. Match proceeds in five steps:1. I s i t alrcady known that A i s an activation of C? I f so, the match ferminates with a positive conclusion, 2. Is there any other activation (A7j, and/or conccpt (C') such that A"is k n o w n t o bc a view of A, C 1s known ta bc a kind of C' , and A' i s known (by step 1) to bc an actlvation of C'? The relations (i.. i s a v i e w of ...) gnd (... isba kind of ...) r c p r e s c n t stored relations between pairs of activations 1 and' concepts, rcspcctivcly. One concept "is a kind of" another conccpt r e p ,~~~, l t s , a s~p c r c l a s s inclusion, t r u a for all timcand cdntexts. '(Whdever else h e might be, J o h n i s a k i n d o f huma'n being:) On the other hand, one activation may be "a view of" another only under certain circumstances--a conditional, or tactical relationship. Undcr diffcrent.conditions, if i s appropriate to v i e w John as a Husband, Father, Child; Hcl p-seeker, Advice-giver, e tc,A l i s t of matched pairs of activations and concepts reprcscnt corrcspondcnccs found el scwhmc, w i t h w h i c h match must be consi stcnt. (N.B.: t h i s Match, as we w i l l see later, may be i n service of anothcr Match galled' on siructuros containing the current A and C.) I f thc p a r [A,C] i s a matchcd pair, then these t y o have been previously found t o match,.so we may hcrc concl'udo the same thing and Match exits, 4. On the other hand, i f there i s either an X or a Y such that [A,X] (or [Y,C]) i s a matchcd pair, then replace this match with an attempt to match C and X (or A and Y). Clearly, for structures of significant complexity, Match may eventually call itc,clf rccursivcly, t o an arbitrory depth. However, since each subordinate call i s on a strictly smaller unit, this process must coitvcrge.Our experience has shown us that this type of mechanism plus a collection of rewrite rules enable us to eventually map a wide variety of input parsing structirrcs to pre-stored, abstract knowledge structures, in a way that a significant aspect o f their intended meaning has been assimilated in the process.This opcratcs to carry out a rule when that rule has become active. Rules are of the form (Condition)->(Action), and Dcduce scnscs thc activity of a rule and applies the rule by activating the concept for the action. Whatcver corresponocnces were evolved in the coursc.of cccating the activation of the condition (left) half of i h e rule are carried over into thc activation of the action (right) half. gives us the capability of a production syctin the WS. Dcducc attempts to match the left half of this rule with some other activation i n the WS. (This has ty-pically already been done by match.) Assuming this i s accomplished, Dcduce c r e s t e t an ac,tivalion of the right half of the rule, substi luting in the activation f o r all subparts for which thcke are correspondences with the i c f t half.Once a Dialogue-game has been activated (by Protcu.,) as possibly the comrnunrcation f o r m being bid for a dialogue, the Dialogue-game Manager uses i t to guide t h c assimilation of successive utterances of the dialogue, through four stages:1. establish the Parameter values and verify that no Specification i s contradictcd, 2. cstabli ;h olhcrwi sc unsupported Specifications as assumptions, 3. cstrrblish the Components as goal$ of thc participanfs, 4. dctcct the circumstancc~ which indicate that the Dialogue-game is.terminating and represent thc conscquenccs of this.Thc first two of thesc phaces hsppcn in parallel. Whcn the Manager accesses e a c h of thc Para~nctcrs, they arc found either to have activations in thc WS or not. I f they do, the cgrrespondcnccs bclwccn activation and Psrsmetcr are established in the WS. This correspond: l o a::ignint a value to thc Paramcicr for this particular evocation of thc Dialogue-game. Any Parameter that has no activation i c put on a list w h i c h i s A Model of Dialogue periodically chcckcd i n the hope that later activity by the Manager will lead t o the creation of appropriate activations.F o r each of the Specifitations, a check i s made to determine i f it already has an activation in WS. (In most cases, the activation of some of these Specifications will have led t o the activity of the Dialogue-game itself.) The Specifications having activations n e e d no further attention.For all remaining Spcc,ifications, activations are created substituting for t h e Parametcrs as determined above. At this stage, the Dialogue-game Manager calls Protcus to determine the stability of thcso new activations. Any new activation w h i c h contradicts existing activations will have its level of activity sharply reduced by Proteus. I f this happens, the Dialogue-game Manager concludes that some of the necessary preconditions for the game db not hold (are i n conflict with current understanding) and that this particular game should be abandoned. Otherwise, the new activaticns stand as new knowledge, following from the hypothesis that the chosen game i s appropriate, Finally, the Manager detects that one of the Specifications no longer appears to hold. This signals the impending termination of the Dialogue-game. I n fact, t h e utterance whikh contai'ns this information i s a bid to terminate. At this point, i f t h e partici pants7 initial goals are satisfied (thus contradicting the Specification which calls f o r t h e prcsence of those goals) the interaction ends "successfully".Otherwise, t h e Dialogue-game i s terminated for some other reason (e.g., one participant's unwillingness o r inability to continue) and would generally be regarded as a "failure". These consequences are infcrred by the Manager and added to the WS. When a Dialogue-game has terminated, i t s salience goes to zero and i t i s removed from the WS.The Dialogue-Gamo Model contains a set of Pronoun Processes, including an I-Process, a You-Process, and an It-Process. Each of those i s invoked whenever the /' associated surface w o r d appears in an input utterance,, and operates to identify some preexisting activationthat can be seen as a view of the same object.Each of these Processes search the curreot context, as represented by the current sct of actirations in the WS, using tho katures specified there ro identify a set of possible co-rcfctcntial expressions. When there is more than one possibility, the one with a hi ghcr salience i s sclectcd.With the understanding we new have of the multi-sentential aspects of human communication, it i s easy to see why man-machine cornrnun'ietion appears so alien, highly restrictive, uncomprehending and awkward. This i s because major regulation and interpret4 f ion structures are missing.In Table 1 , we compare human dialogue and typical man-machine com.munication with respect to some of these features, The table designates a "sender" ancfa "receiver" which should be identified with the person and the computer, respeclively, in the man-machine communication case. The ideal interface, arid the sort toward which this research i s direded, would be continuously askihg itself: "Why did he say that?". From answers to this, the interface would infcr just what the human was expecting as a response. This would con~titute a major s l c p toward the enabling t h o intcrface to servo the actual (rather than the poorly cxprcsscd) needs of the user. Finally, such an intorface would require much l c o~ adaptation on thc parf of the user, and so, by our original hypotheois, would significantly enhance the cffcctivencss of the man-machine partnership.This paper has described a research effort into the modeling of human dialogue. The purpose of this research has been to uncover and describe i n process models, reculari tics that occur in dialogue. I t i s hopcd that the enhqnced understanding of human communication which rcsul ts, will facilitate the development of more natural (and thus more effective) man-machine interfatcs.Thc principal regularity w'e have discovered i s a collection of knowledge and goal structures, called Dialoeue-games, which seem to be crucial in understanding the structure of naturally-occurring dialogues. According to the theory we have proposed, one or more of these Dialogue-games serve as the major organizing influence on e v e r y human didlogue.Each Dialogue-game specifies what knowledge each person must have to ehgage in such a dialogue, and what goals of the participants might be served by that interchange. A Dialogue-game also spccifies, as a sequence of "tactical" goals, the manner in which the dialogue i s conducted.The Diatogua-game Model i s a collection of cooperative processes which continuausly updated a representation of each participant's attention state in a Workspaco. The model recognizes when a particular Dialogue-game i s being bid, accepted, pursued and terminated, and represents these states appropriately in the Workspace. A particular Dialogue-game, the Helping-game, was described in some dctajl. A simulation of the evocation and use of the Helping-game on a segment of natural dialogue i s contained in the Appendix.Our experience so far with the Dialogue-game Model has reinforced our hypothcses that an understanding of the goal-serving aspects of dialogue i s a powerfdl . In this appendix we describe an extensive simulation of the a r r e n t state of the Dialogue-game Model. We make use of a particular version of the Helping-game and alsc explore another structure, an Execution Scene, which describes the customary events surrounding the successful execution of a particular program (Runoff).We start by describing this more detailed version of the Helping-game, introducing names f o r the various aspects, to be used later. Next we show a short, naturally occurring dialogue between a computer operator and a user. 'Then we describe t h e operation of the Dialoeuc-garnc'Model as if assimilates this dialogue, up to the point at w h i c h i t concludes that thc Helping-game i s an appropriate structure t h r o u~h which to understand the subr;cqucnt utterances.Once this hypothesis for the form of thedialogue has been chosen, we continue the simulation to examine how Jhc model dccidcs that a particular Execution Scene i s appropriate for assimilating the content of the dialogue. 'Next, we see how this choice of occnes cnhances the set of goals imputed to the speaker, thus facilitating the cornprehcnsion of what he i s saying. Finally, we summarize our experience with the Dialogue-game Model so far.What fol4ows i s the substance of the communication structure we have namcd the Hclping-game. In the interests of clarity af presentation, the formal structureo of the definition have been expressed in prose. However, the elements of the following description correspond one-to-one to those in the actual Helping-game used in ?the simulation.Thc parameters are two roles (HELPER and HELPEE) and a topic (TASK/HG).Parameter specifications:The HELPER and HELPEE are each a kind of person. H1 = A goal of t h e HELPEE is that h e perform TASK/HG. H2 = I t i s not true that HELPEE i s able to perform this TASKIHG. H5 = The HELPEE wants to be ablc to pcrform the TASKIHG. H I 1 = T h c HELPEE wants the HELPER to enable him to perform the TASKIHG.(bcing enabled to perform the task.is6a subgoal of performing the task) Game components:HGX 1 = The HELPEE knows of a particular execuiion scene, XS/HE.[note: a n execution scene i s a flowchart-like description of thc use af a particular process; more details below] (much more detail, below) whereupon the next segment is parsed and input for processing.HGX2 =H o w docs thc model know to evoke the Helping-game? To exhibit answers to t h i~ a n d subscqucnt questions, we lead tho reader through a simulation of the model as i t p r o c c s s c s the beginning of dialogue OC117. We indulge in fhe samc use of prorie f o r formalism as aboQe, again w i t h the same assuranaes of correspondcnccs with tho actual sirnulati on.Thc simulation proceeds in cycles: i n each cycle, we exhibit the operation of a sinzlc processor, performing one iteration of i t s function. We do not address h c r c the is:uc; of h o w the model would select w.hich processor to call next. In fact, our dcsign calls f o r these processors to be mgxirnally autonomous and parallel i n their operation, operating w h c n c v c r circumstances are ripe for their function and dormant otherwise. Thc format of this sirnuistion i s as follow;: Thc cycle number i s first, in the form: :cy,mcnt nurnbcr9--cycle number in this scgrncnt,.Next i s tho name of the p r a c c w o r operating in this cyclc. Aftcr that i s EI description of the nature of the pracossiny. donc d u P l n b that cyclc. Finally, tharo i s a l i s t of tha rcsults for this cycle, that is, ;dl tho irnportljnt c h a n g w i n WS, Cycle 1-1 --Parse.The parser reads one utterance/segment of input and translates it into the formalism f o r activations in the workspace. No claim 'is made that this translation retains all the content of the original text, only that i t is adequately faithful to the level of detail we a r e simulating.Results: Case/9 (= (0 perceives that L asks (how do I get Runoff working?))) is activated.Certain words (e,g. pronouns, determiners) are taken to be signals that a reference i s being made t o conccpts introduced elsewhere. Sne presence of a concept i n t h e workspoce corresponding to one of these words lcads to the calling of the. process-specialist which attempts t o resolve the implied reference. Thus, the presence of "I" i n the text leads to the calling of the I-process, whose sole function i s to determine t h e referent of the .'I" and modify the stored concept to reflect this. This process judges that i f L i s asking a question which contains "I" as its subject, then this constitutes adequate evidence to hypothesize that "I" is being used to refer to L.Results: 0 perceives that L asks (how does L get Runoff working?)Cycle 1-3 --Match Match i s always on the lookout for pairs of nodes, one in the WS and the other i n the LTM, such that the activation (node in WS) matches the concept (node i n LTM). T h i s i s taken to be evidence that the activation is also to be t i k e n as an activation of the matched concept. I t should be understood that we areaexamining only some of the succewful matches which occurred, Starting in this cycle, we see a pattern which recurs regularly, and which accounts fcr a significant piece of the action, as the model assimilates the dialogue. Match dcterrnincs that a particular activation matches the left half (condition side, i f part, etc.) of a production-like rule srorcd in LTM. This successful match leads t o the identification of the corrcspondcnces between the aspects of the activation and those of the left half of the rule, ae well as creating an activation of the rule itself. The activation of a rule leads to calling the Deduce processor in thenext cycle, which applies the activated rub t o the node i n the WS responsible for the rule's activation. This application of a rule (which also results in thc removal of the rule's activation from the WS) creates a new activation structure in the WS. Cycle 1-22 --ProteusAs a result of the numerous rcfarcncar to Runoff and XS/HE, tho activationo for thcsc two conccpts are "highly active". ~o n s c~u c n t l~, when Protcus io called, tho eonccpt XSjRO (the execution sccno of tho Runoff proccos) bocarnoo active and, duo to its similarity to XSfHE, i s taken to be equivalent to it. Sinca XS/RO is more dctailad (contein~ more information) than XS/HE, XS/RO is used in place i f XS/HE in all of the expressions introduced in Cycle 1-21.Something wo pnsscd ovcr in thc earlier examples was tha issuo of vyhcn tho modcl i s willine to stop processing a given piece of tcxl and eo on la the nexf onc. It scorns inappropriate l o demand that tho rncdol wring all possiblo information end deductions out of each utterance. Yet there must bo soms demands mada on tho assimilation. A n altcrnatc form of tho question is: what ncedr of his own does the hearer see the incoming text as potentially satisfyinc? We have taken the position that a hearer (tentatively) understands an utterance, when he successfully views i t as serving some goal imputed to the spcskcr. That is, to a first approximation, the hearer has assimilated an utterance if hc fisures out why thc spcakcr said it.Thc modcl has already established (HCX9 and HGX10, above) that L wants to dcscribc (implicitly, to 0) certain action; in XS/RO b a t L expected to perceive, and in s o n w csscs, did, Thus, in thc following uttcrsnces, we see the modol matching the parsed input structure with one of thcsc two goals, thus it i s sccn as bcing in service of a goal of thc spcakcr, and need bc examined no further (for tho time being).In thc subscqucnl example, we use two ncw rules: RS (Satisfaction) and RQ (Quicsccncc). RS dctcrrnincs when an uttcranco i s sccn to satisfy a speaker's goal and R Q , rcscts to this dcfectcd satisfaction by marking the utterance quicsccnt.(Opcrationolly, this means that in tho next cycle, thc Parser i s called to input the next scgmcnt af text.)Wc resume the example at the point where the first segment has been marked quicsccnt, and the Parser is called.Results: Casc9a = 0 pcrceivcs that L declares (I executed it).Cycle 2-2 --I-processoronto cithcr the line printer or another file. The execution scene of Runoff, as stored in our model, i s similar t o figure A-1.
:
by providing a framework for integrating the comprehcnsion of an utterance w i t h that of i t s prcdcccssors. Recently, we have propased (Leuin & Moore, 1976 : 1977 Rr Lcvin, 1977) multi-scntential knowledge units that are specified primarily by the speaker's and hcarcr's goals.Thcsc goal-oriented units, which w e call Dialogue-gsmcs [l] , specify the kinds of language interactions in w h i c h people engage, r a t h c r than the spccific content of thcsc intcractions. Pcoplc use l a n g u a~c primarily t o comrnunicatc with other pcoplc l o achieve their own goals. Thc Dialoguc-game mu1 ti-scntontial structures wcrc dcvcloped to represent this knowledge about language and how i t can be uscd to achicve goals.- [I] Thc term "Oialoguc-game" was adopted by analogy from Wittgcnstcin's term (Searle, 1969) . The direct comprehension of these sentences fails to d e r i v e the main communicative~effcct. For example, declarative scntenccs can be used to seek information ("1 nced to know your Social Security number."): questions can be u-scd to convey information ("Did you know that John and Harriet got married?") or t o request an action ("Could you pass the salt?''). These kinds of utterances, w h i c h have b c c n extensively analyzed b y philosophers of language (Austin, 1962 : Searle, 196 9, 1975 : Grice, 1975 , are not handled satisfactorily b y any of the current theories of t h e d i r c c t comprchcnsion of language. However, these indirect language usages ara widespread in naturally occurring language--even two-year-old children can comprehend indirect requests for action almost as well as dircct requests (Shatz, 1975) .O n e theory proposcd to account for these indirect uses of language i s based on the concept of "convcrsotional postulates" (Grice, 1975 : Gordon 1 975formalized and tested this model, and found that people's rasponse times tend to support a three-stage model (deriving the l i t e r a l mcaning, check its plausibility and, i f implausible, dcriving the "intended" meaning" from convcrsational rules). I n general, this approach to i n d i r e d speech acts i s infc~ence-bascd, depending on t h e application of conversational rules to infer the indirect meaning from the d i r c c t mcaning and the context. A different approach has been proposcd b y L s b o v~f i F a n s~c l (1 974) and by Levin & Moore (1976 : 1977 . Multi-sentential knowledge, organizing a scgmcnt of language interaction, can form the basis for deriving the indikect effect of u t t c r~n c c w i t h i n the segment.For example, a multi-sentential structure for an information-seeking interaction can sypply the appropriate context for interpreting the subscqucnt utterances to s~c k and t-hen supply information. The infcrcncc-bascd approach rcquircs one set of convcrs~tional rulc-, for information requests, a dif fcrcnt 9ct of r u l c s for answers to these rcquc:ts, and a way t o tic thcnc t w o rulc sets together. The Dialogue-game model postulates a single k n o w l c d~e struclurc for this kind of interaction, w i t h coopcrating proccssc; for: (1) rccognizinp; when this kind of interaction i s proposcd,(2) using this knowlcdgc to comprchcnd uttcranccn within its scope, and (3) identifying w h e n the interaction i s to be terminated, A M o d c l of DialogueOur t h c o r y of human language use has bccn strongly influenced by w o r k in human p r o b l e m solving (Ncwcll XI Simon, 1972) in which the bchavior of a human i s modeled as an information. processing system, having goals to pursue and selecting actions w h i c h tend t o schicvc thcsc goals. Wc view humans as engaging in linguistic bchavior in order t o advance the state of certain of thcir eoals. Thcy dccide to use language, they sclcct ( o r accept) t h c other participant for a dialogue, they choose the details of linguistic c x p r c s s i o n -all with the expectation that some of their desired state specifications can thcrcby be rcalizcd,In this thcory of lancuagc, a participant i n a linguistic exchange views the other as an indcpcndcnt information-processing system, w i t h separate knowledge, goals, abilities and acccss l o the world. A spcsker h a s a range of potcntial changes he can cffcct i n h i s l i~t c n c r , a corresponding collection of linguistic actions which may result in each such chance, and some notion of the conscqucnccs of performing each of these. T h e spcokcr may view the hcarcr as a resource for information, a potential actor, or as an object to bc moldcd into sorrrc dcsircd state.A dialogue involves t w o speakers, who altcrnatc as hearers. I n choosing t o initiate o r conlinuc tho cxchany,~, a participant attcmpts to satisfy his own goals: in intcrprcting on uttcrancc of his partner, each participant attcmpts to find the way in w h i c h that utterance serves the goals of his partner. Thus a dialoguo continues because the participants continue to scc it as furthering thcir own goals. Likewise, w h e n the dialoguc no l o n~o r serves the goals of one of the participants, i t i s redirected to new goals o r tcrminatcd.this rrlcchanism of joint interaction, via cxchange of uttcranccs, i n pursuit of d c s i r c d t.itcs, i s uscful for ochiovihg ccrtain relatcd pairs +of participanls' ~o a l s ( c .~. , I t v~r n i rlr./tcact~inc, buyinc/sc\ling, gctting h c i p /~i v i n g hclp, ...). Many of thcsc paired sets of I correspond to hichly structured collections of knowlcdgc, shorcd by thc rncrnbcrr, o f thc langunpc community. Thcsc ~ollcctions specify such things as: 1) what chnractcri:tics an individual must h a v c to cngagc in a dialogue of this sort, 2) how t h i s dialocuc i s initiated, pursued and tcrminatcd, 3) what ranee .of infarmation can bo comrnunicotcd imp1 icitl y , and 4) undcr what Circumstances tho dialoguo will "succeed" ( s c r v c tho function for which it was initiated) and how this &ill bo cxhibitcd in the participants7 bck~uvior.VJo h~~o allr:mptc:d to rrproscnt those collr:.ctiong_of knowlcdgc and tho wily in w t~i c h t h c y arc u:cd to f a c i l i t~t o tho cornprchcnsion of a diala~ua, in tha O i n l o e~~o t;nrno Mod(-? I.This section describes our Dialogue-game Model at i t s current state of dcvclopmcnt. I t starts w i t h a brdef overview of dialogue and how it i s structured, t h e n describes the dominant knowledge structures w h i c h guide the model, and finally dcscribcs a set of processes which apply these knowledge structures t o text to comprehend i t Within the mb.dcl., each participant in a dialogue i s simply pursuing his own goals of t h c moment. The t w o participants interact smoothly because the conventions of communication coordinate their goals and give them continuihg reasons t o speak and listen. These goals have a number of attributes which are not necessarily consequences of c i t h c r human activity in general, or communication in particula'r; but which are nonetheless characteristic of human communication i n the form of dialogue:Goals are cooperatively esta5lished. Bidding and acceptance activities serve to intfoduce goals.Goa/s.aremufua//yknown. E a c h p a r t y a s s u m e s o r c o m e s t p k n o w goals of the othcr, and each interprets the entire dialogue relative to currently known goals.Goalsareconfieu~edbyconvenfion. Setsofgoalsforusein dialogue (and othcr lwguage use as well) are tacitly'known and employed by all competent spe;l&rs o f t h e language.Goa/s are bilateral. Each dialogue participant assumes goals complementary to those of his partner.
Appendix:
| null | null | null | null | {
"paperhash": [
"rieger|the_commonsense_algorithm_as_a_basis_for_comfuter_models_of_human_memory,_inference,_belief_and_contextual_language_comprehension",
"charniak|organization_and_inference_in_a_frame-like_system_of_common_sense_knowledge",
"schank|scripts,_plans_and_knowledge"
],
"title": [
"The Commonsense Algorithm as a Basis for Comfuter Models of Human Memory, Inference, Belief and Contextual Language Comprehension",
"Organization and Inference in a Frame-Like System of Common Sense Knowledge",
"Scripts, Plans and Knowledge"
],
"abstract": [
"The notion of a commonsense algorithm is presented as a basic data structure for modeling human cognition. This data structure unifies many current ideas about human memory and information processing. The structure is defined by specifying a set of proposed cognitive primitive links which, when used to build up large structures of actions, states, statechanges and tendencies, provide an adequate formalism for expressing human plans and activities, as well as general mechanisms and computer algorithms. The commonsense algorithm is a type of framework (as Minsky has defined the term) for representing algorithmic processes, hopefully the way humans do.",
"My goals have not changed since (Charniak 72). I am still interested in the construction of a computer program which will answer questions about simple narration (e.g. children's stories). More exactly, if one makes the somewhat unrealistic division of the problem into (a) going from natural language to a convenient internal representation, and (b) being able to \"reason\" about the information in the story in order to answer questions, my interests are clearly in the latter section. I will take it as given that such reasoning requires large amounts of \"common sense knowledge\" about the topics mentioned in the text, so I will not demonstrate this point. (However it should come out incidentally from the examples used to demonstrate other points.) To reason with this knowledge requires that it be organized, by which I simply mean it must be structured so that the system can get at necessary knowledge when it is needed, but that unnecessary knowledge will not clog the system with the all too familiar \"combinatorial explosion\". I will start with my current thoughts on organization.",
"\"Of what a strange nature is knowledge! It clings to the mind, when it has once seized on it, like a lichen on the rock,\" Abstract We describe a theoretical system intended to facilitate the use of knowledge In an understand ing system. The notion of script is introduced to account for knowledge about mundane situations. A program, SAM, is capable of using scripts to under stand. The notion of plans is introduced to ac count for general knowledge about novel situa tions. I. Preface In an attempt to provide theory where there have been mostly unrelated systems, Minsky (1974) recently described the as fitting into the notion of \"frames.\" Minsky at tempted to relate this work, in what is essentially language processing, to areas of vision research that conform to the same notion. Mlnsky's frames paper has created quite a stir in AI and some immediate spinoff research along the lines of developing frames manipulators (e.g. Bobrow, 1975; Winograd, 1975). We find that we agree with much of what Minsky said about frames and with his characterization of our own work. The frames idea is so general, however, that It does not lend itself to applications without further specialization. This paper is an attempt to devel op further the lines of thought set out in Schank (1975a) and Abelson (1973; 1975a). The ideas pre sented here can be viewed as a specialization of the frame idea. We shall refer to our central constructs as \"scripts.\" II. The Problem Researchers in natural language understanding have felt for some time that the eventual limit on the solution of our problem will be our ability to characterize world knowledge. Various researchers have approached world knowledge in various ways. Winograd (1972) dealt with the problem by severely restricting the world. This approach had the po sitive effect of producing a working system and the negative effect of producing one that was only minimally extendable. Charniak (1972) approached the problem from the other end entirely and has made some interesting first steps, but because his work is not grounded in any representational sys tem or any working computational system the res triction of world knowledge need not critically concern him. Our feeling is that an effective characteri zation of knowledge can result in a real under standing system in the not too distant future. We expect that programs based on the theory we out …"
],
"authors": [
{
"name": [
"C. Rieger"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Eugene Charniak"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. Schank",
"R. Abelson"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null
],
"s2_corpus_id": [
"3906011",
"10595802",
"18113275"
],
"intents": [
[],
[],
[]
],
"isInfluential": [
false,
false,
false
]
} | Problem: The paper describes the development of a Dialogue-game Model and its application in understanding human communication patterns.
Solution: The hypothesis investigated in this paper is that the Dialogue-game Model can effectively represent the structure of naturally occurring dialogues by identifying knowledge structures and goals of participants in the dialogue. | 569 | 0 | null | null | null | null | null | null | null | null |
0e6b308f424d8cb33fd2c018856ce3a861d5c334 | 33730973 | null | {BETA}-systemet: En sammanfattning (The {BETA} system: A summary) [In {S}wedish] | BETA-ayatemet år ett programmeringsspråk uppbyggt helt och h ånet kring subatitutionsgrammatikens principer. Substltutionsregler har långe snvånts inom lingvistiken får att beskriva vissa fenomen (både syntaktiska och fonologiaka), Chomaky år vå! den aom gjort metoden mest kand# men Aven andra har förespråkat aubatitutlonareg* lernas viktiga roll i lingvistiken oberoende av Chomaky? t,ex, Hocket och Harris. I logiken och matematiken har substitutlonsmetoden varit långe kånd; den "uppfanns" av norrmannen Thue omkr 1918, och logikerna E, Post och A,H. Turing undersökte dess teoretiska aspekter under 30-och 4B-ta!en, | {
"name": [
"Brodda, Benny"
],
"affiliation": [
null
]
} | null | null | Proceedings of the 1st Nordic Conference of Computational Linguistics ({NODALIDA} 1977) | 1977-10-01 | 1 | 0 | null | Default för MD år i 6t GENERELLA 3TYRPAR! bverst på varje regellista skall upp till 3 generella parametrar skrivas. Dessa kallar vi RECDEL, HL och PROCVAR, RECOEL (Reoeldelimitern) år A,3CII*koden för det tecken aom användes för att definie ra vad som menas med att vånsterledet och hönerledet i en substitution ar "slut", Default år 32 (mellanslag), men om man måste skriva en regel omfattande mellanslaget måste man vålja ett annat tecken (som dock aldrig får förekomma i en regel).HL förklaras i avsnittet om regelordning nedan, PROGVAR kan ha vården I eller 3 och att man har lite olika prooramvarianter, I Ar defeultvårdet och Innebår precis det system aom vi förutsett hår# PROCVAR = 3 innebår att man har inga statusparametrar i reglerna. Regelformatet blir då:X Y LC RC MV MO 9! REGELORDNING!Reglerna utvårderas i princip i den ordning de låsts in. Den första regeln uppifrån som är tillåmpbar tillåmpas. En sådan regelordning kallas för dlsjunktiv, Att reglerna utvårderas disjunktivt Ar dock en sanning med modifikation, det år i stort sått sant, men inte alldeles. Hur utvärderingen i detalj utföres bestämmes av parameter nr 2, HL# i de generella styroarametrarna. Denna parameter lägger upp reglerna i s k heah-klaaser enligt vänsterleden i aubstftutlonsreolerna. Om HL t.ex, år =2 (defeultvårdet) betraktas alla regler med minst två identiska tecken i vånsterledet ("X") i substltutlonsreglerna 6om hörande till aamma haehklaas; och dessa utvårderas före de som innehåller endast ett tecken i vänsterledet. Om HL sattes = 3 utvårderas de regler som innehåller minst tre tecken före de som innehåller två, vilka i aln tur utvårderas före de som innehåller endast ett tecken. Inom varje haah*klasa utvårde ras reglerna strikt disjunktivt (detta är viktigt att komma ihåg, s3 att dot inte "nyper" fel regel tidigare). Huvudprincipen för regelskrlvandet skall alltså vara att regler med starkare villkor (dvs längre eller havande "hårdare" kontext-villkor) måste skrivas tidigare 1 listan. Något krav p3 att reglerna skall skrivas 1 bokstavsordning eller dyl föreligger inte, Tvärtom, det rekommen deras att regler som "looiskt" hör ihop sammanföres. Sådana logiskt hopförda regler kommer då att fungera som subrutiner i vanliga programmeringsspråk. Konventionen med hash-klassernas utvårderlngsordnlng underlättar att man kan skriva reglerna på detta sått.26Proceedings of NODALIDA 1977 | BETA-ayatemet år ett programmeringsspråk uppbyggt helt och h ånet kring subatitutionsgrammatikens principer. Substltutionsregler har långe snvånts inom lingvistiken får att beskriva vissa fenomen (både syntaktiska och fonologiaka), Chomaky år vå! den aom gjort metoden mest kand# men Aven andra har förespråkat aubatitutlonareg* lernas viktiga roll i lingvistiken oberoende av Chomaky? t,ex, Hocket och Harris. I logiken och matematiken har substitutlonsmetoden varit långe kånd; den "uppfanns" av norrmannen Thue omkr 1918, och logikerna E, Post och A,H. Turing undersökte dess teoretiska aspekter under 30-och 4B-ta!en, Det aktuella systemet har regler som innebår en mild "generalise ring" av regier av Turlng-typ utformade som "productiona" av Post typ; generaliseringen år gjord med syfte att åstadkomma regier aom &r "iagomt" bekvåma f&r morfologisk analys. (Som bekant &r Turingreglerna -trots sin enkelhet * * kraftfulla nog att åstadkomma allt som överhuvud taget år möjligt att åstadkomma i dan h&r branschen. Ev. generaliseringar adderar i sak ingenting annat *n möjligen bekv&mllghet.) i! SUBSTITUTI0N3CRAMMATIK: En substitutlon Innebår att man i en strång W (mer om detta begrepp strax) ersätter en delstr&ng med X med en annan delatr&ng Y* Han erhåller så en ny strång på vilken man sen kan utföra ytterligare avbtltutioner etc. De substitutloner man får utföra beståms av vilka substltutionsregler man hart dessa sammantagna utgör en subst i tut 1onsgrammat1 k.En grammatik 1 den h&r meningen akall nu anvåndas på följande sått! Man tar en strång av ett visst slag som input till grammatiken, Oenna knådar så om atr&ngen sa länge det finna regler som passar# och nar inga regler långre år tillämpliga definieras den nu kanske rått kraftigt omknådade strången som grammatikens output.Vad som får vara input och vad aom blir outout år nu inte en gåno får alla givet; det beror Ju på vilken sorts grammatik man skrivit oeh för vilket syfte. Om t.ex. input år vanliga meningar och output syntaktiska analyser av densamma kallar man grammatiken en analysgrammatik. Om input utgörea av en enda abstrakt meningasymbol S och outtut av en faktisk mening har vi en syntesgrammatik, vanligen kallad en generativ grammatik. Många andra slag finnes och det generella begreppet utgår transduktor (transducer), BETA-ayatemet år ett sätt att göra varje dator till en transduktor, Fdr att datorn akall förstå hur den skal! tillåmpa reglerna måste det vara ordnino och reda på såttet att skriva reglerna. Reglerna 20 BETA-systemet: En sammanfattning Benny Brodda Proceedings of NODALIDA 1977, pages 20-26 b&r ha något aå når fixt format (utseende), och det måste vara våldeflnjerat vad som skall handa når man tillåmpar regeln. Har man vål fixerat regel-formatet och dess tolkning ger det sig nästan ajålv hur man skall få datorn att uppföra sig på det förväntade sattet; de programmeringstekniska detaljerna låmnsr vi helt darhån f&r ögonblicket -i någon mening Ar det Ju ointressant h u r datorn b&r sig åt att göra det man ber den och kanske mer Intres sant att veta v a d man kan be den göra. Har f&ljar nu an kort sammanfattning av det senare.Hed en sträng menas en aemmanh&noande sekvens av tecken (karaktårer); dvs bokatSver* siffror, typooraffaka tecken o dy!, Kort sagt slit det man kan skriva ned på en skrivmaskin (inkl, sådana tecken som normalt inte "syns"# och darfAr brukar kallas "vita" tecken! vagnretur# radframmatning; mellanslag, tabbar o dyl), I det aktuel la SETA-aystemet sntages den interna representationen vara ASCIIalfabetet, d v a den representation man erhåller om materialet stansas på en ak Teletype-meakin. I och får sig år BtTA-systemet helt generellt och behöver inte alla förutsåttas vara ASCII* orienterat? men får att förstå de exempel aom ges är det bra att veta vad det Innebar# F&r att kunna benåmna tecknen anvåndes då och då tecknets decimala kodnummer. Se separat bilaga, (Varav framgår att t.ex, mellanslag har kodnr 32# har kod 33, "0" har kod 38# "a" har kod 65 etc, Med en delstrång menas helt enkelt en likaledes sammanh&ngande strång som inoår som del av den större; cd år alltså en delatr&ng av beds# men ca &r inte en delstrång. Om vi har en regel som tillåter utbyte av cd mot era kan v! alltså få strangen beraa.Om man vil! databehandla en hel text, kan det kanske vara oprak tiskt (Ja, omöjligt) att 1 maskinen handskas med hela texten på en gång. Om det år syntaktiska egenskaper o3 meningsnivå man vill syssla med; vil! man som input ge en mening i taget. Om det ar morfologisk analys på ordrivå är det ju oroet som är det menings fulla objektet. Han mgste alltså kunna tala o?" för datorn hur stora sjok som skall in åt sången. Varje tecken skall i bETA*syatemet 3sättas ett typvårde aom anvåndes för att styra input, Allt som typeta som typ 1 (under rubriken DEFTYP i exemplen) betraktas som nostavskiljare, eller om m^n 3 & v 1 1 1 , allt mellan två i i o r b l i r dst datorn får som input. F&r att ocks3 styra outDut (så att man fé lasbart format) behandlas allt som ar typet som en 2:a som en potentiell radavslutare. On m m vill jobba på meninnsnivå bör man alltså typa punkt och frågetecken som l!nr, mellanslag och komma som 2:or och allt annat som hög e. Om bokct&ver t m a s som H!a får man vid inlåsningen samnanf^ring av ord avhuggna vid rudslLt med blnJestrecu, Typilnr-n måste obligatoriskt vara med i en BETA-reeeluppsRttning (rrn^s under rubriken PbrTYP), Kjnventione*) Jon, att de tecken rerr t-åkna* upp ti l höger åsåttes drt t y o^. =8okståverAnm,! "0"Anvåndes som "stand in" får radslut. Om "å" typaa som 2 men och "?" tllr post=mening, Om mellanslag (32) dessutom types som I blir post=ord.BETA-ayatemet arbetar på följande sått! NAr en post kommer in (vilket sker automatiskt! det finns ingen sårakild låalnstruktion# ej heller någon tryck*d:o, när en post år f&rdiobehandlad åker den ut och nästa kommer in? når man trycker på "START" kommer första in) kan man tanka sig att en liten tomte, dot tkallad, ställer sig långst til! vänster i strängen. Sen kutar han fram och tillbaka, tjenstevillig sor bara den? och substituerer och atår i. Vid varje ögonblick når han inte håller på och substituerer "år" han någons tans i strängen uch det han dår gör år att kolla om nåoon regel &r tillämplig Just d a r. Om så icke år fallet tar han ett steg till höger och upprepar processen. Når han ramlat utanför atrången til! höger år "Jobbet" klart och man erhåller resultatet aom output. Det andra fallet; dvs han upptäcker a t t en reoel år tillämplig år förstås det intressantaste: En regel innehåller tre huvuddelar A) vilken substitution aom skal! utföraa B) villkoren för att aubstitutlonen skall få utföras C) vad göra sen, ("Action") MV a MOVE, order om vart dot skall stålla sio hårnåst HO s HODE, uppgift om av. alternativa substitutloner skall utfåras (för att ta hand om ambiguåss strängar).För att regeln i fråga åver huvud taget skall tillåmpas skall tre villkor vara uppfyllda, nämligen på vånsterkontext, på hdgerkontest och på rådande "tillstånd". Om nåqot av dessa villkor icke Ar uppfyllt tillämpas regeln icke (och dot prövar en annan regel eller -om ingen sådan finns -knallar ett steg åt höger). Dessa tre villkor utvårderas helt enligt aamma princip. Men innan Jag beskri ver den principen kanske vi skulle tala något om begreppet "rådande tillstånd", Vid varje Ögonblick antagea systemet befinnas vara i ett slags tillstånd 8, som dot måste kolla av för att se om han dverhuvvdtaoet får tillåmpa regeln. Han kan t&nka sig det hela som att det h&nger kulörta lyktor lite runt omkring, och i varje Ögonblick lyser en av dessa (vid starten lyser den "neutrala" vita), I en viss regel kan det ingå Instruktioner att ändra detta rådande tillstånd, dva slåcka den aktuella lyktan och tända en ny (detta år innebörden av SR, resulterande tillstånd). Detta år ett sått att minnas vad som hånt tidigare, I systemet anges tillstånden med tal liggande i intervallet 1-127 (SJålva talvärdet betyder ingenting i alg, det år bara ett namn).Parametern 3C (som också år ett tal i samma Intervall men kan också vara negerat) Innebar nu ett villkor på det rådande tills tåndet? och detta år viktigt att komms ihåg. Om det rådande tills tåndet år '12' och villkoret i regeln såger '17' kan det mycket vå! intråffa at '12' uppfyller villkoret '17', Alltså! parametern SC år ett villkor på rådande tillstånd och inte ett namn på oet tillstånd som akal! råds. Oet ar lått att tånka fe! hår# men Just denna subtilitet gör systemet mycket flexibelt. ilur vet man nu att rådande tillstånd 8 uppfyller villkoret SC? Ja# ta exemplet ovant '12" uppfyller '17' om *12' finns uppråknad bland de tillstånd aom år angivna til! höger om rubriken 17! under OEFSET. OEfSET har samma konvention som OEFTYP men tillåter "croasclasaiflcation", Ext DEFSET n i 2 3 2* 3 17t 3 1 2 1 5 1 7 3 2 3 5 , .?Tillstånd '1' uppfyller enbart villkor '1*, likaså '2' (som alltså ej uppfyller '2')* '3' uppfyller både '1', '2' och "17' och som sagt vad, +12' uppfyller '17*. Villkoret *-17' vilket innebår villkoret '17' negerat# dvs rådande tillstånd får inte ingå 1 '17' uppfyllea bara av '1' och '2'.Når det gåller vånster-och hdgerkontext tittar dot bara på t a c k n e e n nårmaat till vånster och till håger om den strång X (som 23 23 Proceedings of NODALIDA 1977 ev, skall substitueres). Vill man Jobba med långre kontexter får man använda sig av tillstånd som kjåttrar uop och nedy BETAsystemet år narmaat tänkt för fonologiskt/motologiska tillämpningar och dar gåller 1 förbluffande höo orad att man endast behöver narmaate grannkontext). Parametrarna LC och HC år nu v i 1 1 k o r på dessa nårorannary ocn villkoren år av samma art som tillståndsvdlkoren. Anger man villkoret '17' som LC betyder det att tecknet omedelbart till vånster om X skall finnas uppräknat till höger om 17: uncer OEFSET, I exemplet ovan år tecknen mellanslag (32); punkt; komma, frågetecken uppräknade vid '17'# dvs typiska ordavskiljare; LC = 17 skulle alltså Innebåra att regeln endast får tillämpas om tecknet till vånster år en ordavakiljara, eller m a o enbart i ordb&rjan.LC# RC eller SC s B innebår "intet villkor" (noll defaultvårdet) 6! ACTION:Cm dot nu konstaterat ett regeln f å r tillåmpas# vad gör då dot? Ja# först och fråmat utföres su&atltutlonen# men vad mer? Först och fråmst skal! rådande tillstånd åndras till SR resulteran de tillstånd, men endast under förutsättning att 3R år s B, i annat fall b i b eh å 1 1 e s rådande tillståno, Nåsta parameter MV (MOVE) aåger var dot skal! gå hårnåat, och man har i stort sett 6 standardpositioner att oå till, och dessa år inte fixa utan relaterade till den nyas utförda aubatltutfonen. Får att förklara dessa framståller vi det hela schematiskt! Fig I, år hur det ser ut Just innan subatitutionen utförts. Varje ruta symboliserar ett tecken, en avlång låda en sträng. Den stråno som skall bytas ut år "mittlådan". De tre lådorna tillsammans utgör dan aktuella strången,S -. . . . . .*..L R...........e* Figl.(Oot; dvs "**, finns inte med i själva strången utan kutar a a e ovanpå)Vi antager nu att dot kollat att R ingår i RC# L ingår i LC och 3 ingår i SC, Oot plockar då bort X och pluggar in Y i atållet. Just vid sjålva substitutionsögonblicket kan man tånka sio att hela strången man arbetar med har tre delar: De!otrån9 en til! vånster om Y, Y självt och delstrången till höoer om Y, detta symboliserat med de tre lådorna nedan i fig 2. Varje sådan delstrång har (kan ha) ett första tecken och ett sista tecken, vi har alltså 6 våldeflnierade punkter i strången. Vi numrerar dessa från vånster til! h6per 1-6, och får då de atan-d^rdpositioner till vilka man kan dirigera dot, Skriver man 4 som MV parameter atåller sia dot alldeles t v om det sista tecknet i V (Om Y år tom, d v s X deleterats blir förstås positionerna 3, 4 och 5 desamma), Ut&ver dessa sex standardpositioner kan man dirigera dot tl!!*några andra positioner utanför den aktuella strången, 24 24 Proceedings of NODALIDA 1977 0$ Hela strången deleteras 7: Strången betrektas som fårdlgbehandlsd och ges som outPut (ut 1) 8! Strången ges som output på separat fil (ut 2) 9! Strängen dirigeras till radskrivaren (Lpr) Genom möjligheterna 8 och 9 kan man anvånda BETA som att mycket avancerat excerperinoaprogram. 0-Ning kan tillämpas om flera alternativ bearbetas samtidigt och något av dem visade sig vara dödfött) Jfr MD-parsmetern nedsn) Positionerna B, 1# 6 och 7 finns faktiskt som en del av strången som BETA arbetar med och innehåller alla tecknet ' " ("brådgård"), Strången i utrymmet mellan I och 6 år den egna strången man jobbar med. Startposition år 1, "halkar" man ut till höger år man fåroig, "halkar" man ut. till vånster dör strången.Default för MV år 5.Om man håller på med syntaktisk analys år det bekant att man Ibland kan erhålla atrångar (konstruktioner) som år analyserbara på mer ån ett sått. Problemet år att kunna ange samtliga möjliga analyser. Detta år ganska lätt att åstadkomma i BETA, HMINSTONE 0n HAN KAN SKRIVA REGLERNA Så ATT MAN SER ATT EN VALSITUATION AR FOR HAUOEN, Detta anges 1 BETA-reglerna med den sk M0DE*(MD) parametern. Denna parameter kan ha 3 värden 1,2 och 3 (samt också "ekvivalenta" vården 5, 6 och 7? Om dessa senare användes "frågar" dot om regeln 1 fråga får anvåndes, Oenna facllltet förutsätter att man kör BETA interaktivt och sitter vid terminalen och besvarar dessa frågor).Reglerna i BETA tolkas normalt diajunktivt. Oet betyder att den första regel uppifrån som år tillåmpbar tillämpas och om MO år a I# vilket år det normala (Jfr dock avsnittet om regelordning), så år det dArmed inget mer tal om den saken. Om MO år satt till 2, lågges det nyss erhållna resultatet en stund 3t sidan, och dot söker efter ytterligare en regel# och om en sådan återfinnes utföres åven denna substitutlon, år åven denna re9el mårkt som en 2:a slaskas åven det nya resultatet undan och nåsta tlllåmpbara regel tlllåmpas etc.De uppståndna alternativen långes sedan 1 en kö, och denna kö avbetas först innan någon ny post inhåmtas utifrån, I denna kö lagras det "halvfabrikat" som hitintills uppstått inklusive det aktuella tillståndet. Når detta halvfabrikat återhämtas från kön fortsätter man alltså behandlingen precis dår man avbröt den (för att fortsåtta med andra alternativ).MO s 3 år ett specialfall av 2. Substitutlonen utföres och resulta tet av densamma lagras 1 kön precis aom 1 fallet MO t 2, Någon ny rega! uppsökas inte och dot fortsåtter att Jobba med "originalet" aom ingenting hade hånt#! man kan alltså både ta och alåppa en regel. | null | null | null | Main paper:
proceedings of nodalida 1977:
Default för MD år i 6t GENERELLA 3TYRPAR! bverst på varje regellista skall upp till 3 generella parametrar skrivas. Dessa kallar vi RECDEL, HL och PROCVAR, RECOEL (Reoeldelimitern) år A,3CII*koden för det tecken aom användes för att definie ra vad som menas med att vånsterledet och hönerledet i en substitution ar "slut", Default år 32 (mellanslag), men om man måste skriva en regel omfattande mellanslaget måste man vålja ett annat tecken (som dock aldrig får förekomma i en regel).HL förklaras i avsnittet om regelordning nedan, PROGVAR kan ha vården I eller 3 och att man har lite olika prooramvarianter, I Ar defeultvårdet och Innebår precis det system aom vi förutsett hår# PROCVAR = 3 innebår att man har inga statusparametrar i reglerna. Regelformatet blir då:X Y LC RC MV MO 9! REGELORDNING!Reglerna utvårderas i princip i den ordning de låsts in. Den första regeln uppifrån som är tillåmpbar tillåmpas. En sådan regelordning kallas för dlsjunktiv, Att reglerna utvårderas disjunktivt Ar dock en sanning med modifikation, det år i stort sått sant, men inte alldeles. Hur utvärderingen i detalj utföres bestämmes av parameter nr 2, HL# i de generella styroarametrarna. Denna parameter lägger upp reglerna i s k heah-klaaser enligt vänsterleden i aubstftutlonsreolerna. Om HL t.ex, år =2 (defeultvårdet) betraktas alla regler med minst två identiska tecken i vånsterledet ("X") i substltutlonsreglerna 6om hörande till aamma haehklaas; och dessa utvårderas före de som innehåller endast ett tecken i vänsterledet. Om HL sattes = 3 utvårderas de regler som innehåller minst tre tecken före de som innehåller två, vilka i aln tur utvårderas före de som innehåller endast ett tecken. Inom varje haah*klasa utvårde ras reglerna strikt disjunktivt (detta är viktigt att komma ihåg, s3 att dot inte "nyper" fel regel tidigare). Huvudprincipen för regelskrlvandet skall alltså vara att regler med starkare villkor (dvs längre eller havande "hårdare" kontext-villkor) måste skrivas tidigare 1 listan. Något krav p3 att reglerna skall skrivas 1 bokstavsordning eller dyl föreligger inte, Tvärtom, det rekommen deras att regler som "looiskt" hör ihop sammanföres. Sådana logiskt hopförda regler kommer då att fungera som subrutiner i vanliga programmeringsspråk. Konventionen med hash-klassernas utvårderlngsordnlng underlättar att man kan skriva reglerna på detta sått.26Proceedings of NODALIDA 1977
:
BETA-ayatemet år ett programmeringsspråk uppbyggt helt och h ånet kring subatitutionsgrammatikens principer. Substltutionsregler har långe snvånts inom lingvistiken får att beskriva vissa fenomen (både syntaktiska och fonologiaka), Chomaky år vå! den aom gjort metoden mest kand# men Aven andra har förespråkat aubatitutlonareg* lernas viktiga roll i lingvistiken oberoende av Chomaky? t,ex, Hocket och Harris. I logiken och matematiken har substitutlonsmetoden varit långe kånd; den "uppfanns" av norrmannen Thue omkr 1918, och logikerna E, Post och A,H. Turing undersökte dess teoretiska aspekter under 30-och 4B-ta!en, Det aktuella systemet har regler som innebår en mild "generalise ring" av regier av Turlng-typ utformade som "productiona" av Post typ; generaliseringen år gjord med syfte att åstadkomma regier aom &r "iagomt" bekvåma f&r morfologisk analys. (Som bekant &r Turingreglerna -trots sin enkelhet * * kraftfulla nog att åstadkomma allt som överhuvud taget år möjligt att åstadkomma i dan h&r branschen. Ev. generaliseringar adderar i sak ingenting annat *n möjligen bekv&mllghet.) i! SUBSTITUTI0N3CRAMMATIK: En substitutlon Innebår att man i en strång W (mer om detta begrepp strax) ersätter en delstr&ng med X med en annan delatr&ng Y* Han erhåller så en ny strång på vilken man sen kan utföra ytterligare avbtltutioner etc. De substitutloner man får utföra beståms av vilka substltutionsregler man hart dessa sammantagna utgör en subst i tut 1onsgrammat1 k.En grammatik 1 den h&r meningen akall nu anvåndas på följande sått! Man tar en strång av ett visst slag som input till grammatiken, Oenna knådar så om atr&ngen sa länge det finna regler som passar# och nar inga regler långre år tillämpliga definieras den nu kanske rått kraftigt omknådade strången som grammatikens output.Vad som får vara input och vad aom blir outout år nu inte en gåno får alla givet; det beror Ju på vilken sorts grammatik man skrivit oeh för vilket syfte. Om t.ex. input år vanliga meningar och output syntaktiska analyser av densamma kallar man grammatiken en analysgrammatik. Om input utgörea av en enda abstrakt meningasymbol S och outtut av en faktisk mening har vi en syntesgrammatik, vanligen kallad en generativ grammatik. Många andra slag finnes och det generella begreppet utgår transduktor (transducer), BETA-ayatemet år ett sätt att göra varje dator till en transduktor, Fdr att datorn akall förstå hur den skal! tillåmpa reglerna måste det vara ordnino och reda på såttet att skriva reglerna. Reglerna 20 BETA-systemet: En sammanfattning Benny Brodda Proceedings of NODALIDA 1977, pages 20-26 b&r ha något aå når fixt format (utseende), och det måste vara våldeflnjerat vad som skall handa når man tillåmpar regeln. Har man vål fixerat regel-formatet och dess tolkning ger det sig nästan ajålv hur man skall få datorn att uppföra sig på det förväntade sattet; de programmeringstekniska detaljerna låmnsr vi helt darhån f&r ögonblicket -i någon mening Ar det Ju ointressant h u r datorn b&r sig åt att göra det man ber den och kanske mer Intres sant att veta v a d man kan be den göra. Har f&ljar nu an kort sammanfattning av det senare.Hed en sträng menas en aemmanh&noande sekvens av tecken (karaktårer); dvs bokatSver* siffror, typooraffaka tecken o dy!, Kort sagt slit det man kan skriva ned på en skrivmaskin (inkl, sådana tecken som normalt inte "syns"# och darfAr brukar kallas "vita" tecken! vagnretur# radframmatning; mellanslag, tabbar o dyl), I det aktuel la SETA-aystemet sntages den interna representationen vara ASCIIalfabetet, d v a den representation man erhåller om materialet stansas på en ak Teletype-meakin. I och får sig år BtTA-systemet helt generellt och behöver inte alla förutsåttas vara ASCII* orienterat? men får att förstå de exempel aom ges är det bra att veta vad det Innebar# F&r att kunna benåmna tecknen anvåndes då och då tecknets decimala kodnummer. Se separat bilaga, (Varav framgår att t.ex, mellanslag har kodnr 32# har kod 33, "0" har kod 38# "a" har kod 65 etc, Med en delstrång menas helt enkelt en likaledes sammanh&ngande strång som inoår som del av den större; cd år alltså en delatr&ng av beds# men ca &r inte en delstrång. Om vi har en regel som tillåter utbyte av cd mot era kan v! alltså få strangen beraa.Om man vil! databehandla en hel text, kan det kanske vara oprak tiskt (Ja, omöjligt) att 1 maskinen handskas med hela texten på en gång. Om det år syntaktiska egenskaper o3 meningsnivå man vill syssla med; vil! man som input ge en mening i taget. Om det ar morfologisk analys på ordrivå är det ju oroet som är det menings fulla objektet. Han mgste alltså kunna tala o?" för datorn hur stora sjok som skall in åt sången. Varje tecken skall i bETA*syatemet 3sättas ett typvårde aom anvåndes för att styra input, Allt som typeta som typ 1 (under rubriken DEFTYP i exemplen) betraktas som nostavskiljare, eller om m^n 3 & v 1 1 1 , allt mellan två i i o r b l i r dst datorn får som input. F&r att ocks3 styra outDut (så att man fé lasbart format) behandlas allt som ar typet som en 2:a som en potentiell radavslutare. On m m vill jobba på meninnsnivå bör man alltså typa punkt och frågetecken som l!nr, mellanslag och komma som 2:or och allt annat som hög e. Om bokct&ver t m a s som H!a får man vid inlåsningen samnanf^ring av ord avhuggna vid rudslLt med blnJestrecu, Typilnr-n måste obligatoriskt vara med i en BETA-reeeluppsRttning (rrn^s under rubriken PbrTYP), Kjnventione*) Jon, att de tecken rerr t-åkna* upp ti l höger åsåttes drt t y o^. =8okståverAnm,! "0"Anvåndes som "stand in" får radslut. Om "å" typaa som 2 men och "?" tllr post=mening, Om mellanslag (32) dessutom types som I blir post=ord.BETA-ayatemet arbetar på följande sått! NAr en post kommer in (vilket sker automatiskt! det finns ingen sårakild låalnstruktion# ej heller någon tryck*d:o, när en post år f&rdiobehandlad åker den ut och nästa kommer in? når man trycker på "START" kommer första in) kan man tanka sig att en liten tomte, dot tkallad, ställer sig långst til! vänster i strängen. Sen kutar han fram och tillbaka, tjenstevillig sor bara den? och substituerer och atår i. Vid varje ögonblick når han inte håller på och substituerer "år" han någons tans i strängen uch det han dår gör år att kolla om nåoon regel &r tillämplig Just d a r. Om så icke år fallet tar han ett steg till höger och upprepar processen. Når han ramlat utanför atrången til! höger år "Jobbet" klart och man erhåller resultatet aom output. Det andra fallet; dvs han upptäcker a t t en reoel år tillämplig år förstås det intressantaste: En regel innehåller tre huvuddelar A) vilken substitution aom skal! utföraa B) villkoren för att aubstitutlonen skall få utföras C) vad göra sen, ("Action") MV a MOVE, order om vart dot skall stålla sio hårnåst HO s HODE, uppgift om av. alternativa substitutloner skall utfåras (för att ta hand om ambiguåss strängar).För att regeln i fråga åver huvud taget skall tillåmpas skall tre villkor vara uppfyllda, nämligen på vånsterkontext, på hdgerkontest och på rådande "tillstånd". Om nåqot av dessa villkor icke Ar uppfyllt tillämpas regeln icke (och dot prövar en annan regel eller -om ingen sådan finns -knallar ett steg åt höger). Dessa tre villkor utvårderas helt enligt aamma princip. Men innan Jag beskri ver den principen kanske vi skulle tala något om begreppet "rådande tillstånd", Vid varje Ögonblick antagea systemet befinnas vara i ett slags tillstånd 8, som dot måste kolla av för att se om han dverhuvvdtaoet får tillåmpa regeln. Han kan t&nka sig det hela som att det h&nger kulörta lyktor lite runt omkring, och i varje Ögonblick lyser en av dessa (vid starten lyser den "neutrala" vita), I en viss regel kan det ingå Instruktioner att ändra detta rådande tillstånd, dva slåcka den aktuella lyktan och tända en ny (detta år innebörden av SR, resulterande tillstånd). Detta år ett sått att minnas vad som hånt tidigare, I systemet anges tillstånden med tal liggande i intervallet 1-127 (SJålva talvärdet betyder ingenting i alg, det år bara ett namn).Parametern 3C (som också år ett tal i samma Intervall men kan också vara negerat) Innebar nu ett villkor på det rådande tills tåndet? och detta år viktigt att komms ihåg. Om det rådande tills tåndet år '12' och villkoret i regeln såger '17' kan det mycket vå! intråffa at '12' uppfyller villkoret '17', Alltså! parametern SC år ett villkor på rådande tillstånd och inte ett namn på oet tillstånd som akal! råds. Oet ar lått att tånka fe! hår# men Just denna subtilitet gör systemet mycket flexibelt. ilur vet man nu att rådande tillstånd 8 uppfyller villkoret SC? Ja# ta exemplet ovant '12" uppfyller '17' om *12' finns uppråknad bland de tillstånd aom år angivna til! höger om rubriken 17! under OEFSET. OEfSET har samma konvention som OEFTYP men tillåter "croasclasaiflcation", Ext DEFSET n i 2 3 2* 3 17t 3 1 2 1 5 1 7 3 2 3 5 , .?Tillstånd '1' uppfyller enbart villkor '1*, likaså '2' (som alltså ej uppfyller '2')* '3' uppfyller både '1', '2' och "17' och som sagt vad, +12' uppfyller '17*. Villkoret *-17' vilket innebår villkoret '17' negerat# dvs rådande tillstånd får inte ingå 1 '17' uppfyllea bara av '1' och '2'.Når det gåller vånster-och hdgerkontext tittar dot bara på t a c k n e e n nårmaat till vånster och till håger om den strång X (som 23 23 Proceedings of NODALIDA 1977 ev, skall substitueres). Vill man Jobba med långre kontexter får man använda sig av tillstånd som kjåttrar uop och nedy BETAsystemet år narmaat tänkt för fonologiskt/motologiska tillämpningar och dar gåller 1 förbluffande höo orad att man endast behöver narmaate grannkontext). Parametrarna LC och HC år nu v i 1 1 k o r på dessa nårorannary ocn villkoren år av samma art som tillståndsvdlkoren. Anger man villkoret '17' som LC betyder det att tecknet omedelbart till vånster om X skall finnas uppräknat till höger om 17: uncer OEFSET, I exemplet ovan år tecknen mellanslag (32); punkt; komma, frågetecken uppräknade vid '17'# dvs typiska ordavskiljare; LC = 17 skulle alltså Innebåra att regeln endast får tillämpas om tecknet till vånster år en ordavakiljara, eller m a o enbart i ordb&rjan.LC# RC eller SC s B innebår "intet villkor" (noll defaultvårdet) 6! ACTION:Cm dot nu konstaterat ett regeln f å r tillåmpas# vad gör då dot? Ja# först och fråmat utföres su&atltutlonen# men vad mer? Först och fråmst skal! rådande tillstånd åndras till SR resulteran de tillstånd, men endast under förutsättning att 3R år s B, i annat fall b i b eh å 1 1 e s rådande tillståno, Nåsta parameter MV (MOVE) aåger var dot skal! gå hårnåat, och man har i stort sett 6 standardpositioner att oå till, och dessa år inte fixa utan relaterade till den nyas utförda aubatltutfonen. Får att förklara dessa framståller vi det hela schematiskt! Fig I, år hur det ser ut Just innan subatitutionen utförts. Varje ruta symboliserar ett tecken, en avlång låda en sträng. Den stråno som skall bytas ut år "mittlådan". De tre lådorna tillsammans utgör dan aktuella strången,S -. . . . . .*..L R...........e* Figl.(Oot; dvs "**, finns inte med i själva strången utan kutar a a e ovanpå)Vi antager nu att dot kollat att R ingår i RC# L ingår i LC och 3 ingår i SC, Oot plockar då bort X och pluggar in Y i atållet. Just vid sjålva substitutionsögonblicket kan man tånka sio att hela strången man arbetar med har tre delar: De!otrån9 en til! vånster om Y, Y självt och delstrången till höoer om Y, detta symboliserat med de tre lådorna nedan i fig 2. Varje sådan delstrång har (kan ha) ett första tecken och ett sista tecken, vi har alltså 6 våldeflnierade punkter i strången. Vi numrerar dessa från vånster til! h6per 1-6, och får då de atan-d^rdpositioner till vilka man kan dirigera dot, Skriver man 4 som MV parameter atåller sia dot alldeles t v om det sista tecknet i V (Om Y år tom, d v s X deleterats blir förstås positionerna 3, 4 och 5 desamma), Ut&ver dessa sex standardpositioner kan man dirigera dot tl!!*några andra positioner utanför den aktuella strången, 24 24 Proceedings of NODALIDA 1977 0$ Hela strången deleteras 7: Strången betrektas som fårdlgbehandlsd och ges som outPut (ut 1) 8! Strången ges som output på separat fil (ut 2) 9! Strängen dirigeras till radskrivaren (Lpr) Genom möjligheterna 8 och 9 kan man anvånda BETA som att mycket avancerat excerperinoaprogram. 0-Ning kan tillämpas om flera alternativ bearbetas samtidigt och något av dem visade sig vara dödfött) Jfr MD-parsmetern nedsn) Positionerna B, 1# 6 och 7 finns faktiskt som en del av strången som BETA arbetar med och innehåller alla tecknet ' " ("brådgård"), Strången i utrymmet mellan I och 6 år den egna strången man jobbar med. Startposition år 1, "halkar" man ut till höger år man fåroig, "halkar" man ut. till vånster dör strången.Default för MV år 5.Om man håller på med syntaktisk analys år det bekant att man Ibland kan erhålla atrångar (konstruktioner) som år analyserbara på mer ån ett sått. Problemet år att kunna ange samtliga möjliga analyser. Detta år ganska lätt att åstadkomma i BETA, HMINSTONE 0n HAN KAN SKRIVA REGLERNA Så ATT MAN SER ATT EN VALSITUATION AR FOR HAUOEN, Detta anges 1 BETA-reglerna med den sk M0DE*(MD) parametern. Denna parameter kan ha 3 värden 1,2 och 3 (samt också "ekvivalenta" vården 5, 6 och 7? Om dessa senare användes "frågar" dot om regeln 1 fråga får anvåndes, Oenna facllltet förutsätter att man kör BETA interaktivt och sitter vid terminalen och besvarar dessa frågor).Reglerna i BETA tolkas normalt diajunktivt. Oet betyder att den första regel uppifrån som år tillåmpbar tillämpas och om MO år a I# vilket år det normala (Jfr dock avsnittet om regelordning), så år det dArmed inget mer tal om den saken. Om MO år satt till 2, lågges det nyss erhållna resultatet en stund 3t sidan, och dot söker efter ytterligare en regel# och om en sådan återfinnes utföres åven denna substitutlon, år åven denna re9el mårkt som en 2:a slaskas åven det nya resultatet undan och nåsta tlllåmpbara regel tlllåmpas etc.De uppståndna alternativen långes sedan 1 en kö, och denna kö avbetas först innan någon ny post inhåmtas utifrån, I denna kö lagras det "halvfabrikat" som hitintills uppstått inklusive det aktuella tillståndet. Når detta halvfabrikat återhämtas från kön fortsätter man alltså behandlingen precis dår man avbröt den (för att fortsåtta med andra alternativ).MO s 3 år ett specialfall av 2. Substitutlonen utföres och resulta tet av densamma lagras 1 kön precis aom 1 fallet MO t 2, Någon ny rega! uppsökas inte och dot fortsåtter att Jobba med "originalet" aom ingenting hade hånt#! man kan alltså både ta och alåppa en regel.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 568 | 0 | null | null | null | null | null | null | null | null |
fff93231fb63651b0afdc6cd4e53475a34c5c7cb | 394923 | null | Experiments with odd languages | Asian languages may seem exotic, with the special difficulties of strange cultural vocabulary, script, and the few opportunities of communicating in them. However, Chinese, Arabic, Indonesian, Hindi, and Persian are not only languages with long cultural histories, but also very influential linguae francae in their political and cultural spheres today. The problems in Chinese seem to be the reverse of those studied by Europeans. One begins with basic roots, neatly separated, and a "machine translation* must m show how they are put together. The Chinese written language, being composed of unit areas, has always been much better organised grammatically, phonetically, and bibliographically. The line-plotter has been used for Chinese since 1968, at least, and now we have the data screen, the matrix printer, jet plotters, and, in Japan, the line-printer. There is teaching in the University of Illinois. Coding in four-digit numbers is well established, though the usual goal, to input all available texts, would be too much for foreign students to attempt, but could be easily achieved by Chinese students. I am interested in: (1) Design of special vocabularies to be put into a small computer, in compact Chinese and English. (2) A database for finding quotations, built on the existing concordances and indexes. Two rather rare characters could well establish a unique original source. This is an attempt to bypass the dictionary, and go to the original text, for which there is often a standard English or French translation. (3) Index to the second character of a twocharacter compound. This is to speed up technical translation by reducing l o o k -uptime. Alphabetic scripts present few problems. Golfballs are available for most languages already. Here I am interested in inputting the basic dictionary, and finding some algorithm that will enable the text to be compacted as much as possible for use in small desk computers. | {
"name": [
"Grinstead, Eric"
],
"affiliation": [
null
]
} | null | null | Proceedings of the 1st Nordic Conference of Computational Linguistics ({NODALIDA} 1977) | 1977-10-01 | 1 | 0 | null | null | null | null | null | null | Main paper:
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 568 | 0 | null | null | null | null | null | null | null | null |
bfe72d3915ff0b2c7da0623da4d04f2d8eb6823c | 2653530 | null | {CHITAB} {--} a {``}poor man{'}s{''} shortcut to computer processing of linguistic data | CHIIAB -a "poor nan's" shortcut to computer processing of linguistic data 1. Background. The primary purpose of the computer programme CHIIAB is modest: we hope it to be of use mainly to individual linguists or small projects with limited resources, who want to cope with sizable corpuses involving delicate classifications. In such a case processing the data manually will soon become laborious. To be more adaptable to linguis tic data processing, the present programme introduces some improvements over similar programmes already existing in various programme libraries (e.g., in HYLPS, in the Univac 1108 system of the University of Helsinki, and in the SPSS, for Dec 10). These improvements include the possibility for alpha-numeric coding, the use of up to ten "filter variables" to ex tract from the corpus precisely the desired variables for cross-tabulation, and the possibility to pick, out of the classifications of any variable, only the frequencies of any individual classificatory principles {e.g., codes 1,3,8,A,C, out of the total range from 1-F). These improvements mean a more economic use of the card space, and a more versatile use of the com puter. The basic idea in the system is that, instead of processing both the text and the coded symbols, only the symbols are fed into the computer. This solution naturally excludes certain kinds of research, such as vocabu lary frequency studies, but it is adequate for frequency counts and cross tabulations of, e.g., various semantic, syntactic and textual features. It is thus adaptable for a variety of research purposes. An important advan tage of the system is that it is remarkably cheap: the punching of the cards is fairly quick and straightforward, one card can accommodate a large num ber of classifications, and the computer processing is also quick. For the benefit of the individual researcher, who is frequently un sophisticated in computer technology, we have a l s o attempted to make the actual use of the computer as simple as possible. Thus, in order to have the computer carry out the desired cross-tabulations, the user only needs 82 CHITAB -a "poor man's" shortcut to computer processing of linguistic data | {
"name": [
"Salmela, Jussi and",
"Kohonen, Viljo"
],
"affiliation": [
null,
null
]
} | null | null | Proceedings of the 1st Nordic Conference of Computational Linguistics ({NODALIDA} 1977) | 1977-10-01 | 0 | 1 | null | The primary purpose of the computer programme CHIIAB is modest: we hope it to be of use mainly to individual linguists or small projects with limited resources, who want to cope with sizable corpuses involving delicate classifications. In such a case processing the data manually will soon become laborious. To be more adaptable to linguis tic data processing, the present programme introduces some improvements over similar programmes already existing in various programme libraries (e.g., in HYLPS, in the Univac 1108 system of the University of Helsinki, and in the SPSS, for Dec 10). These improvements include the possibility for alpha-numeric coding, the use of up to ten "filter variables" to ex tract from the corpus precisely the desired variables for cross-tabulation, and the possibility to pick, out of the classifications of any variable, only the frequencies of any individual classificatory principles {e.g., codes 1,3,8,A,C, out of the total range from 1-F). These improvements mean a more economic use of the card space, and a more versatile use of the com puter.The basic idea in the system is that, instead of processing both the text and the coded symbols, only the symbols are fed into the computer. This solution naturally excludes certain kinds of research, such as vocabu lary frequency studies, but it is adequate for frequency counts and cross tabulations of, e.g., various semantic, syntactic and textual features. It is thus adaptable for a variety of research purposes. An important advan tage of the system is that it is remarkably cheap: the punching of the cards is fairly quick and straightforward, one card can accommodate a large num ber of classifications, and the computer processing is also quick.For the benefit of the individual researcher, who is frequently un sophisticated in computer technology, we have a l s o attempted to make the actual use of the computer as simple as possible. Thus, in order to have the computer carry out the desired cross-tabulations, the user only needs to punch one or two cards: one card for the specification of the variables to be cross-tabulated (and giving the title of the table if desired), and another card to give the "filter" variables (if these are needed; cf. below).2. Coding of the data. Mhet. analysing the primary material (text), the researcher transforms the features chosen for the study directly into a series of coding symbols. To do this he must have an idea of what he is looking for from the text, in the form of hypotheses to be tested or ques tions to which he wants to get quantitative answers. The codings are enter ed manually into primary matrices, according to a specific coding plan. The development of such a plan frequently involves pilot studies with experi mental data. For computer processing, the data are punched onto cards.In the present programme, one variable can be given a field of max. 5 colimns, and a record can comprise a maximum of 5 cards ( i.e ., 400 onecolumn variables); with the alpha-numeric coding, one column can accommo date some 40 different symbols. In most cases, however, some 15-20 subclassifications will be enough, and one column will thus suffice. The first few columns will be two-or-more column variables, for an exact identifica tion of the data (e.g., text/page/line, or consecutive numbering). M e have not adopted the mnemonic coding symbols used, e.g., in M AM BA, because the one-digit alpha-numeric symbols are more economical. In practice, we have noticed that the researcher can (and does) memorize quite a detailed coding plan with "decontextualized" symbols within the first few days (or weeks) of intensive coding work. This relieves him from constant checking of the symbols and thus speeds up the coding process. In the presentation of the results one naturally has to refer to the original text for relevant exam ples.3. Checking and correcting of the data. In addition to the veri fying punching, the programme provides four possibilities for the checking of the data. (1) The checking of the min. and max. limits of codings in each variable (e.g., from 1 to F; anything above F must be an error) reveals er rors that are typically due to punching (these can be eliminated by veri fying punching, though this is laborious in a large corpus). (2) The prog ramme also gives the totals of coded and lacking symbols in each variable.These are useful in cases in which the researcher knows that certain vari ables should have the same total frequencies (cf. check 3), and in extreme cases in which the researcher is reasonably sure that given variables should contain coding in all or very few records. (3) With interrelated variables, the programme can be used to check simultaneous presence/absence of coding in any combination of variables, for example, if the coding plan has four columns for different properties of the subjects of sentences (such as length, structure, givenness and position), all of these must have some coding or be blank. The computer prints out the records showing lack of agreement. (4) further, "impossible" cross-tabulations can also be used to spot errors. Thus, for example, if passive sentences are entered in column X with codings 3,4,6, and their subjects are analysed in a subsequent column Y,l-8, a tabu lation of the identification variables with X-3,4,6 and Y-blank/p ( i.e ., no coding) as the filte r variables will give a lis t of records in which the subject properties had not been entered into the matrix. Errors in (3) and (4) are thus frequently due to the researcher forgetting to enter the rele vant information in all the places; such errors cannot be revealed by veri fying punching. With (4) one can even get inside the total frequencies of the variables, by specifying parts of them to be checked for agreement, f i nally, the actual data cross-tabulations will, of course, serve as further checks in cases of impossible or meaningless contingencies. The programme can be used to print out the dubious records for checking and eventual cor rection. The erroneous records will be replaced on the magnetic tape by the corrected ones. 4. The cross-tabulation programme can be utilized for the following purposes: (1) extraction of data according to the desired specifications or their combinations ("filters"), (2) cross-tabulation of any two variables against each other, (3) calculation of the Chi square test and the contin gency coefficient, and (4) calculation of the relative frequencies by the totals of the rows and columns, and the sum total.(1) The maximum of the filte r variables is 10. These are given on a separate card (or more than one card), after the card specifying the varia bles to be cross-tabulated. An example of a table request thus looks as fol* lows:16, 28 -1 first Table 5 -1^4,5,9,A 6 -3,4 25 * 1-8,f This means that variables 16 and 28 will be cross-tabulated against each other, and the data is extracted from that part of the corpus which is spec ified by the above values of variables 5, 6 and 25. If only portions of 84 the variables are desired for the tabulation, these must be given in the filte r variables (e.g., giving 16 = 1,3,5,A on the second card in the above example). To save computer time, the programme has a "peeping device i.e ., if precisely the same filters are used in the following table, an auxiliary file is formed of the relevant records. This file is kept until a new combination of the filte r variables is read. To make use of this pos sib ility to save time, the researcher should group his table requests in such a way that the tables to be run from precisely the same sub-part of the corpus are in a consecutive order. This is not necessary, however, if it is more desirable to group the runs in some other way. A consequence of this limitation is that the tables to be run without any filte r variables ( i.e ., from the whole corpus) must be run first.(2) For the cross-tabulations, the printout format provides 25 col umns and 155 rows, if needed. These limitations should be taken into account when grouping the row and column specifications. The variable given first in the table request is always interpreted as the row variable, and the second as the column variable. The frequency table is automatically supple mented by the percentages of the row and column totals out of the sum total, for quick reviews. By way of speeding up the programme, the testing of the filte r variables and the up-dating of the frequency tables are both done as a uniform binary search. The programming language is FO RTRAN 4.(3) The calculation of the Chi square test is optional. It is done according to the formulas given in Siegel (1956) .^Before applying the for mulas, the programme checks that the conditions for the calculation of the Chi are satisfied by the contingency table. In the negative case, the prog ramme prints out the reason for the inapplicability of the test. As this module also accepts contingency table data from the cards, it can be used as an independent unit for calculating the test, after possible re-groupings of the tabulated frequency data. When the Chi test is applicable, the prog ramme also calculates the value of the contingency coefficient (C), as a measure of the degree of association between the two variables. These are given immediately below the frequency tables, with the df value, (4) The calculation of the percentages by the totals of the rows, columns and the sum total is also optional, and it is possible to choose any of these. The members of the Text Linguistics Re search Group are working on syntactic data collected from Finnish, Swedish and English, to be processed on the CHIIAB. Kohonen has collected a corpus of some 4,000 clauses from Old and Early Middle English (between ca. 1000 and 1200) for a study of the development of 0E word order. The results will be published in the research reports of the group in 1978. A documentation of the progranme and the coding plans developed for Finnish and Swedish, with some preliminary findings, will appear in Erik Andersson (ed.), Work ing Papers on Computer Processing of Syntactic Data, Abo 1978.Our future plans for the development of the programme include a module that can be used for matching a dependent clause to its matrix clause and forming a separate file of the matrix clauses, for analyses of desired features in them. This module could also be used in contrastive studies, when comparing translations with the originals. Another Improve ment could be added to the tables, where the decontextualized symbols could be automatically replaced, if desired, by mnemonic 4-character title s, to be fed into the computer separately. This would s till preserve the ease and economy of the input, while making the interpretation of the tables more convenient. A further improvement could also be added to the "peeping de vice": instead of taking the full records into the auxiliary file (when pre cisely the same part of the data is going to be used for several consecu tive tables), only the relevant variables, i.e ., those actually needed for the tables requested, would be extracted from the data. This would again speed up the programme.The programme is now available in the Dec 10 system of the Univer sity of Turku. W e are planning to get it into the Univac 1108 system of the University of Helsinki, with terminals all over FlnlahdJ 1 The designing of the operations carried out by the programme has been done jointly by the two authors, while all the technical programming work has been done by Jussi Salmela. If somebody is interested in getting a copy of the programme he should contact Jussi Salmela. Proceedings of NODALIDA 1977 | null | null | null | null | Main paper:
background.:
The primary purpose of the computer programme CHIIAB is modest: we hope it to be of use mainly to individual linguists or small projects with limited resources, who want to cope with sizable corpuses involving delicate classifications. In such a case processing the data manually will soon become laborious. To be more adaptable to linguis tic data processing, the present programme introduces some improvements over similar programmes already existing in various programme libraries (e.g., in HYLPS, in the Univac 1108 system of the University of Helsinki, and in the SPSS, for Dec 10). These improvements include the possibility for alpha-numeric coding, the use of up to ten "filter variables" to ex tract from the corpus precisely the desired variables for cross-tabulation, and the possibility to pick, out of the classifications of any variable, only the frequencies of any individual classificatory principles {e.g., codes 1,3,8,A,C, out of the total range from 1-F). These improvements mean a more economic use of the card space, and a more versatile use of the com puter.The basic idea in the system is that, instead of processing both the text and the coded symbols, only the symbols are fed into the computer. This solution naturally excludes certain kinds of research, such as vocabu lary frequency studies, but it is adequate for frequency counts and cross tabulations of, e.g., various semantic, syntactic and textual features. It is thus adaptable for a variety of research purposes. An important advan tage of the system is that it is remarkably cheap: the punching of the cards is fairly quick and straightforward, one card can accommodate a large num ber of classifications, and the computer processing is also quick.For the benefit of the individual researcher, who is frequently un sophisticated in computer technology, we have a l s o attempted to make the actual use of the computer as simple as possible. Thus, in order to have the computer carry out the desired cross-tabulations, the user only needs to punch one or two cards: one card for the specification of the variables to be cross-tabulated (and giving the title of the table if desired), and another card to give the "filter" variables (if these are needed; cf. below).2. Coding of the data. Mhet. analysing the primary material (text), the researcher transforms the features chosen for the study directly into a series of coding symbols. To do this he must have an idea of what he is looking for from the text, in the form of hypotheses to be tested or ques tions to which he wants to get quantitative answers. The codings are enter ed manually into primary matrices, according to a specific coding plan. The development of such a plan frequently involves pilot studies with experi mental data. For computer processing, the data are punched onto cards.In the present programme, one variable can be given a field of max. 5 colimns, and a record can comprise a maximum of 5 cards ( i.e ., 400 onecolumn variables); with the alpha-numeric coding, one column can accommo date some 40 different symbols. In most cases, however, some 15-20 subclassifications will be enough, and one column will thus suffice. The first few columns will be two-or-more column variables, for an exact identifica tion of the data (e.g., text/page/line, or consecutive numbering). M e have not adopted the mnemonic coding symbols used, e.g., in M AM BA, because the one-digit alpha-numeric symbols are more economical. In practice, we have noticed that the researcher can (and does) memorize quite a detailed coding plan with "decontextualized" symbols within the first few days (or weeks) of intensive coding work. This relieves him from constant checking of the symbols and thus speeds up the coding process. In the presentation of the results one naturally has to refer to the original text for relevant exam ples.3. Checking and correcting of the data. In addition to the veri fying punching, the programme provides four possibilities for the checking of the data. (1) The checking of the min. and max. limits of codings in each variable (e.g., from 1 to F; anything above F must be an error) reveals er rors that are typically due to punching (these can be eliminated by veri fying punching, though this is laborious in a large corpus). (2) The prog ramme also gives the totals of coded and lacking symbols in each variable.These are useful in cases in which the researcher knows that certain vari ables should have the same total frequencies (cf. check 3), and in extreme cases in which the researcher is reasonably sure that given variables should contain coding in all or very few records. (3) With interrelated variables, the programme can be used to check simultaneous presence/absence of coding in any combination of variables, for example, if the coding plan has four columns for different properties of the subjects of sentences (such as length, structure, givenness and position), all of these must have some coding or be blank. The computer prints out the records showing lack of agreement. (4) further, "impossible" cross-tabulations can also be used to spot errors. Thus, for example, if passive sentences are entered in column X with codings 3,4,6, and their subjects are analysed in a subsequent column Y,l-8, a tabu lation of the identification variables with X-3,4,6 and Y-blank/p ( i.e ., no coding) as the filte r variables will give a lis t of records in which the subject properties had not been entered into the matrix. Errors in (3) and (4) are thus frequently due to the researcher forgetting to enter the rele vant information in all the places; such errors cannot be revealed by veri fying punching. With (4) one can even get inside the total frequencies of the variables, by specifying parts of them to be checked for agreement, f i nally, the actual data cross-tabulations will, of course, serve as further checks in cases of impossible or meaningless contingencies. The programme can be used to print out the dubious records for checking and eventual cor rection. The erroneous records will be replaced on the magnetic tape by the corrected ones. 4. The cross-tabulation programme can be utilized for the following purposes: (1) extraction of data according to the desired specifications or their combinations ("filters"), (2) cross-tabulation of any two variables against each other, (3) calculation of the Chi square test and the contin gency coefficient, and (4) calculation of the relative frequencies by the totals of the rows and columns, and the sum total.(1) The maximum of the filte r variables is 10. These are given on a separate card (or more than one card), after the card specifying the varia bles to be cross-tabulated. An example of a table request thus looks as fol* lows:16, 28 -1 first Table 5 -1^4,5,9,A 6 -3,4 25 * 1-8,f This means that variables 16 and 28 will be cross-tabulated against each other, and the data is extracted from that part of the corpus which is spec ified by the above values of variables 5, 6 and 25. If only portions of 84 the variables are desired for the tabulation, these must be given in the filte r variables (e.g., giving 16 = 1,3,5,A on the second card in the above example). To save computer time, the programme has a "peeping device i.e ., if precisely the same filters are used in the following table, an auxiliary file is formed of the relevant records. This file is kept until a new combination of the filte r variables is read. To make use of this pos sib ility to save time, the researcher should group his table requests in such a way that the tables to be run from precisely the same sub-part of the corpus are in a consecutive order. This is not necessary, however, if it is more desirable to group the runs in some other way. A consequence of this limitation is that the tables to be run without any filte r variables ( i.e ., from the whole corpus) must be run first.(2) For the cross-tabulations, the printout format provides 25 col umns and 155 rows, if needed. These limitations should be taken into account when grouping the row and column specifications. The variable given first in the table request is always interpreted as the row variable, and the second as the column variable. The frequency table is automatically supple mented by the percentages of the row and column totals out of the sum total, for quick reviews. By way of speeding up the programme, the testing of the filte r variables and the up-dating of the frequency tables are both done as a uniform binary search. The programming language is FO RTRAN 4.(3) The calculation of the Chi square test is optional. It is done according to the formulas given in Siegel (1956) .^Before applying the for mulas, the programme checks that the conditions for the calculation of the Chi are satisfied by the contingency table. In the negative case, the prog ramme prints out the reason for the inapplicability of the test. As this module also accepts contingency table data from the cards, it can be used as an independent unit for calculating the test, after possible re-groupings of the tabulated frequency data. When the Chi test is applicable, the prog ramme also calculates the value of the contingency coefficient (C), as a measure of the degree of association between the two variables. These are given immediately below the frequency tables, with the df value, (4) The calculation of the percentages by the totals of the rows, columns and the sum total is also optional, and it is possible to choose any of these. The members of the Text Linguistics Re search Group are working on syntactic data collected from Finnish, Swedish and English, to be processed on the CHIIAB. Kohonen has collected a corpus of some 4,000 clauses from Old and Early Middle English (between ca. 1000 and 1200) for a study of the development of 0E word order. The results will be published in the research reports of the group in 1978. A documentation of the progranme and the coding plans developed for Finnish and Swedish, with some preliminary findings, will appear in Erik Andersson (ed.), Work ing Papers on Computer Processing of Syntactic Data, Abo 1978.Our future plans for the development of the programme include a module that can be used for matching a dependent clause to its matrix clause and forming a separate file of the matrix clauses, for analyses of desired features in them. This module could also be used in contrastive studies, when comparing translations with the originals. Another Improve ment could be added to the tables, where the decontextualized symbols could be automatically replaced, if desired, by mnemonic 4-character title s, to be fed into the computer separately. This would s till preserve the ease and economy of the input, while making the interpretation of the tables more convenient. A further improvement could also be added to the "peeping de vice": instead of taking the full records into the auxiliary file (when pre cisely the same part of the data is going to be used for several consecu tive tables), only the relevant variables, i.e ., those actually needed for the tables requested, would be extracted from the data. This would again speed up the programme.The programme is now available in the Dec 10 system of the Univer sity of Turku. W e are planning to get it into the Univac 1108 system of the University of Helsinki, with terminals all over FlnlahdJ 1 The designing of the operations carried out by the programme has been done jointly by the two authors, while all the technical programming work has been done by Jussi Salmela. If somebody is interested in getting a copy of the programme he should contact Jussi Salmela. Proceedings of NODALIDA 1977
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 568 | 0.001761 | null | null | null | null | null | null | null | null |
a657c7ff6d9472633b3759632b3188a464a03e41 | 43772316 | null | Function words in H{\'a}konar saga [In {N}orwegian] | noe mer utførlige definisjon av begrepet function words i bakhode kommer jeg i første omgang til & kon sentrere meg om de fire av Daniel Steible anførte kategorier: Preposisjoner, konjunksjoner, korrelativer, negasjoner. Man får da likevel mer enn nok å gjøre. Det materiale man finner i en norrøn tekst, lar seg nemlig ikke behendig putte inn i de fire esker vi blir tilbudt. En systematisk behandling av dette utvalget vanskeliggjøres ved at det foreligger formal kongruens av noen høyfrekvente preposisjoner med konjunksjoner, til dels også med andre ordklasser. Går man igjennom en KWIC-index for & kunne skille homograf-komponentene fra hverandre, finner man f.eks. at man ved til f&r med ikke mindre enn fem komponenter å gjøre. | {
"name": [
"Mundt, Marina"
],
"affiliation": [
null
]
} | null | null | Proceedings of the 1st Nordic Conference of Computational Linguistics ({NODALIDA} 1977) | 1977-10-01 | 0 | 0 | null | Ved samtlige preposisjoner og en del konjunksjoner må vi regne med tvillingsformer som er adverbier. Tar vi med alle negasjonene, ser vi at det i tre av våre fire esker finnes noe som henviser oss til gruppen adverb, rettere sagt, til gruppen "rene" adverb. Jeg anser det som rimelig at dette faktum får konsekvenser for det arbeid jeg holder.på med.Håkonar saga Håkonarsonar hører inn under genren kongesaga. Den ble til 1263-6$. En kritisk utgave av eldre datum ble lagt til grunn, da jeg fikk punchet teksten, sammen med en del andre sagatekster, for vel ti år siden. Jeg hadde i mellomtiden rikelig anledning til å komme til bake til H^konar saga <g dens saerpreg, under arbeidet med en diplomatarisk utgave av en bestemt versjon av samme saga. Etter at jeg således hadde vaert opptatt av Håkonar sagas språkdrakt i lengre tid og under vekslende synsvinkler,kom min interesse, så langt som den angår ordforrådet, til å bli konsentrert om gruppen function words.Function words -det er stort sett de samme som i norsk grammatikk går under navnet "lukkede ordklasser".Det gjelder også for function words som i første omgang faller inn under de fire åpne ordklasser (verb, subst., adj., adv.). Dette kommer tydelig fram i Olav Naes' fremstilling, hvor han anfører de modale hjelpeverb som "lukket" underklasse til verbene eller de nektende adverbier som lukket under klasse til adverbiene. For ikke å vanskeliggjøre kommuni kasjonen med diskusjonsdeltakere fra nabolandene unødig, tror jeg likevel det er lurest i det følgende å unngå betegnelsen "lukkede ordklasser".Mine hittil tentative overveielser i tilknytning til gruppen function words i norrønt har sitt utspring i interessen for to problemstillinger. Den ene vedrører attribusjonsspørsmålet. I den utstrekning det arbeides med ordfrekvenser på dette området, har utviklingen i den senere tid gått i retning av å bruke stadig flere function words som diskriminatorer enn meningsbaerende gloser/ordkombinasjoner.^ Den andre problemstillingen jeg er opptatt av, er målbarheten av stilistiske meritte* eller unoter. Jeg unngår her bevisst "evaluering",for jeg har litun tro på at spørsmålet god/dårlig stil kan be svares *^d hjelp av frekvensundersøkelser, bortsett fra at dc h;'lt elendige tekster kanskje ville skille seg ut: Til det er det for stort spillerom mellom reseptorene 2 ) n.h.t. hva de anser som "godt".Dessuten er det mange måter å vaere god på. Men det vil som oftest vaere mulig å oppnå enighet om en teksts karakterisering som tørr eller spennende, stiv eller tilnaermet hverdagstale, om det er Selv om jeg ville sette pris på å få kjennskap til eventuelle försök med formordene i de moderne skandinaviske språk, må det sies med en gang, at man ikke kunne vonte seg å treffe liknende forhold i en norrdn tekst. Det skyldes hovedsakelig to grunner. Preposisjonene har etter hvert fått et langt st%rre virkeområde enn f%r, fordi de i dag brukes til å indisere en rekke forhold som tidligere ble uttrykt ved obllkve kasus alene. En annen grunn er at man finner en annen gruppering av de adverbiale konjunksjoner i en norrøn tekst enn hva som er tilfelle i nåtidens skan dinaviske språk.Det ligger 1 sakens natur at function words er ord som ikke gjør mye av seg. Slike "fargeløse" ord, spesielt de unnvaerlige, er dessverre mindre konsistent enn andre,hvor man har med håndskrifter fra middelalderen å gjøre. For Hdkonar saga -som for de fleste norrøne tekster -betyr det: Det antall belegg vi finner i en (annenhånds-)avskrift eller i en kritisk utgave som er laget på grunnlag av to/tre slike avskrifter, kan ikke tas for god fisk bestandig. Så her gjelder i høyeste grad: No variable is entirely safe. Investigate. -Det kanskje beste eksempel jeg kan anføre i den forbindelse, er "ok", ok topper ranglisten ved alle islendingesagaer som jeg har tall for. Den relative frekvens ligger noe høyere i Håkonar saga enn i Knytlinga, men den stemmer godt overens med det en finner i Heimskringla.Så den er det lite å gjøre med. Med anvendelsen av "ok" i teksten, forekomster soa de vurderte soa overflødige. Son illustrasjon av skrivernes behandling av gjc tok jeg vare p& et eks. fra Sth.perg.no 8, bl.42v20: ....ut. ok Vdru beir becar drepnir. Mistanken om at ikke var med i den opprinnelige teksten, blir bekreftet, når man ser på andre avskrifter av samme saga.For ikke bare å oppholde meg ved negative resultater: Når man omsider kommer fram til negasjonene, vil man der bl.a. finne noe, hvor Håkonar saga tydelig skiller seg ut blant (konge-)sagaer som det er naturlig å sammenlikne med. Vi kommer da i siste avdeling til aldri? aldriel. Heimskringla har 109 belegg på 228 000 ord, Knytlinga har 29 på 48 ?00 ord. I forhold til det er belegg-antallet i Håkonar saga påfallende lite:25 på 99 000 ord Hvorfor er det slik? Cår forfatteren av veien for å bruke sterke ord? En slik tanke har nok i første omgang ikke mer for seg enn den gjengse oppfatning at utstrakt bruk av polysyndese gjør en tekst kjedelig. Men spørsmålet må jo bli, hvor mange andre avvik kan vi konstatere som fører til samme indikasjon, og -minst like viktig -om vi greier å finne noe som peker i motsatt retning.Proceedings of NODALIDA 1977Function words in Hákonar saga Marina Mundt Proceedings of NODALIDA 1977, pages 75-78Proceedings of NODALIDA 1977 | null | null | null | null | Main paper:
:
Ved samtlige preposisjoner og en del konjunksjoner må vi regne med tvillingsformer som er adverbier. Tar vi med alle negasjonene, ser vi at det i tre av våre fire esker finnes noe som henviser oss til gruppen adverb, rettere sagt, til gruppen "rene" adverb. Jeg anser det som rimelig at dette faktum får konsekvenser for det arbeid jeg holder.på med.Håkonar saga Håkonarsonar hører inn under genren kongesaga. Den ble til 1263-6$. En kritisk utgave av eldre datum ble lagt til grunn, da jeg fikk punchet teksten, sammen med en del andre sagatekster, for vel ti år siden. Jeg hadde i mellomtiden rikelig anledning til å komme til bake til H^konar saga <g dens saerpreg, under arbeidet med en diplomatarisk utgave av en bestemt versjon av samme saga. Etter at jeg således hadde vaert opptatt av Håkonar sagas språkdrakt i lengre tid og under vekslende synsvinkler,kom min interesse, så langt som den angår ordforrådet, til å bli konsentrert om gruppen function words.Function words -det er stort sett de samme som i norsk grammatikk går under navnet "lukkede ordklasser".Det gjelder også for function words som i første omgang faller inn under de fire åpne ordklasser (verb, subst., adj., adv.). Dette kommer tydelig fram i Olav Naes' fremstilling, hvor han anfører de modale hjelpeverb som "lukket" underklasse til verbene eller de nektende adverbier som lukket under klasse til adverbiene. For ikke å vanskeliggjøre kommuni kasjonen med diskusjonsdeltakere fra nabolandene unødig, tror jeg likevel det er lurest i det følgende å unngå betegnelsen "lukkede ordklasser".Mine hittil tentative overveielser i tilknytning til gruppen function words i norrønt har sitt utspring i interessen for to problemstillinger. Den ene vedrører attribusjonsspørsmålet. I den utstrekning det arbeides med ordfrekvenser på dette området, har utviklingen i den senere tid gått i retning av å bruke stadig flere function words som diskriminatorer enn meningsbaerende gloser/ordkombinasjoner.^ Den andre problemstillingen jeg er opptatt av, er målbarheten av stilistiske meritte* eller unoter. Jeg unngår her bevisst "evaluering",for jeg har litun tro på at spørsmålet god/dårlig stil kan be svares *^d hjelp av frekvensundersøkelser, bortsett fra at dc h;'lt elendige tekster kanskje ville skille seg ut: Til det er det for stort spillerom mellom reseptorene 2 ) n.h.t. hva de anser som "godt".Dessuten er det mange måter å vaere god på. Men det vil som oftest vaere mulig å oppnå enighet om en teksts karakterisering som tørr eller spennende, stiv eller tilnaermet hverdagstale, om det er Selv om jeg ville sette pris på å få kjennskap til eventuelle försök med formordene i de moderne skandinaviske språk, må det sies med en gang, at man ikke kunne vonte seg å treffe liknende forhold i en norrdn tekst. Det skyldes hovedsakelig to grunner. Preposisjonene har etter hvert fått et langt st%rre virkeområde enn f%r, fordi de i dag brukes til å indisere en rekke forhold som tidligere ble uttrykt ved obllkve kasus alene. En annen grunn er at man finner en annen gruppering av de adverbiale konjunksjoner i en norrøn tekst enn hva som er tilfelle i nåtidens skan dinaviske språk.Det ligger 1 sakens natur at function words er ord som ikke gjør mye av seg. Slike "fargeløse" ord, spesielt de unnvaerlige, er dessverre mindre konsistent enn andre,hvor man har med håndskrifter fra middelalderen å gjøre. For Hdkonar saga -som for de fleste norrøne tekster -betyr det: Det antall belegg vi finner i en (annenhånds-)avskrift eller i en kritisk utgave som er laget på grunnlag av to/tre slike avskrifter, kan ikke tas for god fisk bestandig. Så her gjelder i høyeste grad: No variable is entirely safe. Investigate. -Det kanskje beste eksempel jeg kan anføre i den forbindelse, er "ok", ok topper ranglisten ved alle islendingesagaer som jeg har tall for. Den relative frekvens ligger noe høyere i Håkonar saga enn i Knytlinga, men den stemmer godt overens med det en finner i Heimskringla.Så den er det lite å gjøre med. Med anvendelsen av "ok" i teksten, forekomster soa de vurderte soa overflødige. Son illustrasjon av skrivernes behandling av gjc tok jeg vare p& et eks. fra Sth.perg.no 8, bl.42v20: ....ut. ok Vdru beir becar drepnir. Mistanken om at ikke var med i den opprinnelige teksten, blir bekreftet, når man ser på andre avskrifter av samme saga.For ikke bare å oppholde meg ved negative resultater: Når man omsider kommer fram til negasjonene, vil man der bl.a. finne noe, hvor Håkonar saga tydelig skiller seg ut blant (konge-)sagaer som det er naturlig å sammenlikne med. Vi kommer da i siste avdeling til aldri? aldriel. Heimskringla har 109 belegg på 228 000 ord, Knytlinga har 29 på 48 ?00 ord. I forhold til det er belegg-antallet i Håkonar saga påfallende lite:25 på 99 000 ord Hvorfor er det slik? Cår forfatteren av veien for å bruke sterke ord? En slik tanke har nok i første omgang ikke mer for seg enn den gjengse oppfatning at utstrakt bruk av polysyndese gjør en tekst kjedelig. Men spørsmålet må jo bli, hvor mange andre avvik kan vi konstatere som fører til samme indikasjon, og -minst like viktig -om vi greier å finne noe som peker i motsatt retning.Proceedings of NODALIDA 1977Function words in Hákonar saga Marina Mundt Proceedings of NODALIDA 1977, pages 75-78Proceedings of NODALIDA 1977
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 568 | 0 | null | null | null | null | null | null | null | null |
aa4e433df205b51ce066b946591db8071aafa707 | 219303239 | null | Spatial Reference and Semantic Nets | This paper presents an analysis in a semantic net formalism of the semantic structure of English sentences containing references to spatial-location. Spatial reference, hereafter -SR, provides either static location or motional information John is at home, Fred ran across the street to the store. .The task for the semantic analysis of sentences with SR's is to,make clear what is being positioned. THis has been difficult to do. Previous proposals have left unanalyzed many phenomena including important motional references. This paperv* main conclusion is that a much improved analysis can be obtained by representing the SR's as positioning abstract events and states of affairs. The analysis in semantic nets has the location of an event or state of affairs represented as a node which is linked to the node showing the event or state by arcs: indicating its staus as the spatial attribute. A few SR's are shown as naming these locational entities, which we call place ,object. These SR' s involve examples with "where", "here", and "there" However, most SRts are represented as relating place objects to the position of objects in the manner of prepositional phrases. This primacy ok prepositions is argued for in the paper. Motional references are allowed for by functions represented in the nets which If we hear example 2.2 t h e n more than John i s known t o be i n the basement. His c a r d s a r e , for. example. | {
"name": [
"Sondheimer, Norman K."
],
"affiliation": [
null
]
} | null | null | null | 1977-12-01 | 0 | 0 | null | The paradigmatic phenomenon for tte analyses that claim physical objects as the referents of S R 1 s is the noun phrase modifier: This figure shows "the man" being located (LOCI at a time, indicated by the T -link, and at a location which was in "the car". This style of analysis seems simple and direct. It appeals to the intuition that only physical objects take up space. It promises t~ be easy to apply, FIGURE 2 . I "The man i n t h e c a r l e f t " i n t h e st-yle o f Schubert (1936) . s i n c e a l l t h a t i s required i s t o a s s o c i a t e SR's with t h e s e n t e n t i a l elements which a r e modified which r e f e r e n c e physical o b j e c t s . Unfortunately, t h e r e are problems.Often t h e r e i s more t o an event than. j u s t i t s partiripahts' l o c a t i o n s :2.2 John i s playing s o l i t a i r e i n the basement.If we hear example 2.2 t h e n more than John i s known t o be i n the basement. His c a r d s a r e , for. example. F u r t h e r , t h e l o c a t i o n of t h e a c t i o n i s more than t h e i n s t a n t a n e o u s p o s i t i o n of John and his c a r d s . For example, space where t h e cards may, p o t e n t i a l l y be plaoed m u s t be included. S i m i l a r l y , t h e following does not Here, successive changes seem to be appropriate. However, one class of references to motion seems to defeat this entire approach:2.8 The man walked across the puddle. 2.9 The man walked around the puddle.The man walked through the puddle.Examples like the above involve duration in a key way and can not be shown with reference to one position. For example, at no time was the man "across" the puddle like ~aleigh's cloak was across it. Similarly, two points showing the man's change of position are inadequate since the same initial and final positions are a'cceptable in all three cases. Finally, adding an intermediate point will not be adequate, since the man might reach that point while on a path that otherwise holds a different relation to the puddle. As shall be seen, the lesson to be learned from these examples is that in allowing for motion, it is the entire path that must be considered and not selected positions of objects. 2.12 John walked from his car across the yard to the ilouse.How the event of 2.12 can be "from", "across", and "to" simultaneously and also have these aspects temporally ordered is nowhere explained in these analyses.Finally, even if SR's are associated with events and states of affairs, the fact that something is often learned about participants' location must be explained. The t h i r d s t y l e of SR a n a l y s i s i s nonuniform i n nature. These e i t h e r mix the two uniform analyses or e l a b o r a t e on the simple event o r s t a t e a n a l y s i s .Mixed analyses claim t h a t some SR's l o c a t e concrete o b j e c t s while some l o c a t e events or s t a t e s of a f f a i r s (see for example, Winograd, 1972, and Schank, 1973) By s a c r i f i c i n g the s i m p l i c i t y t h a t comes from uniformity , these analyses avoid the uniform analyses' complementary problems. However, the mutual problems, e s p e c i a l l y motion, are l e f t unsolved.The nonuniform analyses t h a t e l a b o r a t e on the nature of events and s t a t e s of a f f a i r s a r e best represented by Case analyses, see Bruce (1975) . . I n terms of events and s t a t e s of a f f a i r s , t h e f i r s t case can e i t h e r be used f o r overall event o r s t a t e l o c a t i o n o r i t may be used t o l o c a t e an a s p e c t of t h e event. The f i n a l t h r e e c a s e s a l l r e l a t e t o d i f f e r e n t a s p e c t s of a motional event. This allows f o r examples l i k e 2.12, with i n h e r e n t temporal o r d e r i n g among t h e cases allowing f o r t h e o r d e r i n g of t h e SR's.The Case a n a l y s e s s t i l l has problems The underlined phrases r e f e r t o motion ordered i n time, e . g . , he walked t h e h i l l before t h e b r i d g e . However, Case a n a l y s i s g i v e s no way t o order i n s t a n c e s of the same c a s e . Gruber (1965) p o i n t s out t h e same problem with t h e Goal case:2.14 I walked t o New York t o my mother's.F i n a l l y , t h e Case proposal rnuat be given some physical i n t e r p r e t a t i o n . Any r e p r e s e n t a t i o n of meaning must a t some p o i n t be r e l a t e d t o a m~d e l of t h e world.I n t h i s i n s t a n c e t h e i d e a of a source, g o a l , and path must be somehow r e l a t e d t o models of motion.This paper p r e s e n t s a proposal f o r an a n a l y s i s t h a t i s nonuniform i n t h e same way t h e Case a n a l y s i s i s . A uniform source for l o c a t i o n s modified by SR's i sgiven, but t h e p r e d i c a t i o n of t h e s e spaces by SR's i s shown t o be much more complex than previously thought. F u t t h e r , sentences are not seen as being as simple with respect to SR's as previously supposed.Before presenting the analysis, two sections will be devoted to preliminary topics: our semantic net formalism and the eyntactic status o f the phenomena considered. (19751, Shapiro (,1971), Simmons (19731, and Woods (1975) All this information is essential to any artificially intelligent entity, just as the model is essential'to any analysis in the predicate calculur. However, for showing the semntic relations in which we are mainly interested, an abbreviation is sufficient just as only the formulas are sufficient in most studies using symbolic logic. Hence a special abbreviation will be used in all sections except fX where the definitional level wili be discwsed. | null | state by arcs: indicating its staus as the spatial attribute. A few SR's are shown as naming these locational entities, which we call place ,object. These SR' s involve examples with "where", "here", and "there" However, most SRts are represented as relating place objects to the position of objects in the manner of prepositional phrases. This primacy ok prepositions is argued for in the paper. Motional references are allowed for by functions represented in the nets which produce parts of place objects which are then positioned by prepositional f c m s . The necessary ordering that'comes with motional references is allowed for by associating temporal elements with the functions.While the positioned elements are simple, the overall semantic structure of the sentences containing SR's is often complicated by the involvement of more than one event or state of affairs. The paper includes a survey of the sentential semantic structures necessary to deal with SR's.A similar complexity is necessary to deal with the informakion on the location of objects which is gained from sentences with SR's. The paper suggests-the use of inference rules to allow for this.The most surprising of the paper's oonclusions is that a strong tie exists between referehces to space and temporal information. In fact, the locations of all events and states of affairs placed by SR's are argued to be locations in both space and time. The effect of this conclusion is most clearly seen in a formalized definition of the primitives of the semantic seructures , which is also presented in semantic nets. There, as ane possible interpretation of the place object, it is shown as a set of pairs of volumes in space and points in time. . . . . This paper presents an analpis in a semantic net formalism of the semantic structure of English sentence8 containing references to spatial location.Spatial referefnce, hereafter -SR, provides either'static location or motional information :1.1 John is at home 1.2 Fred ran across the street to the store.The task for the semantic analysis of sentences with SR's is to make clear what is being positioned. This has been difficult to do. Previous proposals have left unanalyzed many phenomena including important motional references. This paper's main conclusion is that a much.improved analysis can be obtained by representing the SR's as positioning ab~tract events and states of affairs.The analysis in semantic nets has the location of an event or state ~f affairs represented as a node which is linked to the node showing the event or state by arcs indicating its status as the spatial attribute. A few SR's are shown as naming these locational entities, which we call pLaee object. These SR's involve examples with "where", "here", and "therei'. However, most SR' s are represented as relating place objects to the position of objects in the manner of prepositional phrases. This primacy of prepositions is argued for in the paper.Motional references are allowed for by functions represented in the nets which produce parts of place objects which are then positioned by prepositional forms.The necessary ordering that comes with motional references is allowed for by associating temporal elements with the functions.While the positioned elements are simple, the overall semantic structure of the sentences containing SR's is often complicated by the involvement of more than one event or state of affairs. The paper includes a survey of the sentential There is available a discussion in greater detail of a preliminary analysis to the one given here (~ondheimer, 1975) . There is also available for comparison an analy is by this authbr of the same meaning phenomena, in the competing paradigm of model-theoretic semantics (Sondheimer , 1978) The current.paper is distinguishable by its better developed semantic net formalism nnd i t n emphasis 3n producing computationally j u s t i f i e d structures.Ihe pazt has seen many studies of SR phenomena. There ha8 been interest in connecting. language and scenes, e. g. , Coles ( 19681, Kochen ( 19691, Winograd (1972) are claimed to locate. In some cases, the SR'B apply to only physical objects..In athers, they apply to only abstract forme identifying events and states o affairs. A broad third type of analysis shows different sorts of entities being modified. Each has.its limitations.Central to our abbreviation will be nodes that collapse types and tokens.These will identigy the verbal concepts that characterize the events and states of affairs. We will call them "event/state" nodes. They will be circled and capital letters will be used for abstract types, such as CAUSING. Nonabstract forms will be shown with names that suggest the interpretation, e.g., Sleeping will suggest the sleeping state. When a node represent8 a physical object, identifying information will be included in quotes, e.g., "the bus". Names placed on ascs will abbreviate and suggest the functional roles of attributes.For example, -ANTE for antecedent and -CONS for consequence will be used with CAUSING. Case names will be used with many event and state of affairs types.T for "Time" showing the time an event occurred or state held.-A for "Agent" showing the instigator of an event or state.-0 for "Object", the neutral case (as Fillmore (1971) explains it "the wastebasket") .Restrictions on types of entities which will be necessary will be shown by nonoval shapes for nodes. For example, time instances will be shown in parentheses and time intervals in square brackets. Finally, because it is not essential for our purposes, specification of time will often be left out of most semantic structures. Similarly, ire will consider only declarative statements. Some concepts that act as functions will also be used. Each of these will look like a relation associating parameters with a value. The value will be identifiied by a -VALUE arc. Inference rules will be presented in the form ofLoving \ \ (I I MOryl l ) [ 1.11 ] FIGURE 3 . 1 "John loved Mary all last year." "subnetll' 2 "subnet2", where on seeing slibnet 1' subnet* is to 'be added to the semantic net. These rules will include variables within nodes, where the variables are to be bound on matching and referenced on inferencing. These variables will be in the form of capital letters, e.g., X.To summarize, our semantic net formalism uses concept names, descriptions of objects, mnemonic arc names, and mnemonic shapes for nodes to abbreviate the two levels in a semantic net. Also used are functions and inference ruLes. This will be enough to represent the semantic relations involving reference to space that are being considered. Unfortunately, it is one more unique formalism. However, it sdds no new structures, only abbreviating others. We leave as an unproven claim that it will fit in with any formalism which shows identifiable event and state of affairs nodes such as Norman add Rumelhart (1975) and Schank (1973) . In this sectian, the syntax function of prepositional phraees will be considered and arguments for their primacy will be presented.Our main interest in syntax is in structuring our didcussion of semantics.However, the problems of parsing and generation make the syntax of' SR's independently imprtant. These are not our topics here. However, in an earlier issue of this journal we presented a parsing scheme that produces semantic from syntactic structure and applied the scheme to current clags of phenomena (Sondheimer and Perry, 1975). There is some controversy on the distinction between these two types. We can present two syntactic and one semantic classification procedures. First, adjuncts are never required for granrmaticality, while locative objects can be: These are the four primary uses of locative prepositions. We claim that the semantic structure of other SK'B can be represented through these forms. We will now show this. In general, this will be done by observing the SR's structure or by paraphrase arguments.Some spatial terms can have syntactic and semantic functions similar to prepositions in that they directly serve to relate two forms: | These examples can immediately be given prepositional-like semantic structures.In other sentences, these terms appear as nouns and adjectives: A diverse variety of non-prepositional locative adverbs can be handled with Bennett (1975) points out, the "across1' and "from" phrases combine in euch a way that we understand that it is the way that must be travelled in starting from home and going to John that is "across the streetB'.* This can be allowed for in semantic nets with a function, -WAY, producing a path through space joining two points identified by INIT for Initial and FIN for Final links, see In our semantic net model, the locations of events and states of affairs will be shown as attributes of e v e n t h a t e nodes through arcs leading from the nodes to locational entities. For each event/state node involved with an SR there will be only one such arc and locational entity. Theee atcs will be labelled -P to suggest a spatial attribute or "Place" case. The locational entities will be referred to as place objects. They are the basis of our analysis. These place objects can be taken for the time being as volumes in space. The sort of ~o l u m e they are will be ela6orated upon. Place objects will be identified by boxes. We are finally far away from New York.Finally, direct analysis can be given to qualifier usages which either apply with a static sense to nouns describing physical objects (5.8) or act like static adjuncts with respect to verbal noun8 ( 5 . 9 ) :5.8 T k b man i n the car l e f t .5.9 swimming i n the lake i s fun.These qualifier usages can be contrasted with those that ahow motian (5.10), act l i k e locative objects to verbal nouns ( 5 . 1 1 ) or @how extent (5.12):5.10 The bus t o Chicago l e f t . 5.11 ~w i m i n g i n to a cave i s fun.The bridge from Ohio to Weet Virginia i e old.One t e s t for adjuncts i n the last section was to see i f it located the e n t i r e t y of the event or s t a t e discussed. The s t a t i c adjuncts are i d e n t i f i a b l e in t h i s way.Since the place object shows the location of that e n t i r e t y , static adjuncts can therefore be directly a p p l i e d t o them. The ice bag is to John's head, but both the ice bag and John are in the car. The first SR involves a locative object, the second an adjunct. With a simple approach to e v e n t h a t e location, they would not be differentiated. There is a similar problem in some adjunct references to the location of only part of an event or state of affairs. For instance, in 6.2, only the boy isplaced which the hawk is definitely physically present:6.2In an open field, a boy watched a hawk.In the latter case, although not the former, the use of inference rules might be suggested. However, a better answer can be found.The I heard through t h e door.On these occasiow, an assumed e n t i t y can be added t o the semantic s t r u c t u r e : I heard ( sane thing) through the door.Inference rules play an important part in these analyses. For example, the positioning of John and the ice bag must be derivable from the structure of There are a number of other classes of verbs that take static locative objects, see Table 1 . We will survey their analysis in the remainder of the section and close with a comment on several related forms. : grab, hit, kick, kiss, kneel, punch, slap, slug , touch. : break, chop, cook, cut, fry, shatter, spill, 6 .4.FIGGRE 6.4 He is j u s t now in the house.Problems with motion arise in every analysis of SR considered in Section 11.In Only "take" and "bring" show the same pattern. For this reason, an abstract concept of pure motion, called -GOING, will be used in our analysis. Figure 7.1 shows the sentential structure into which most movement SR's will t . The structures for "take" and "bring" will have Going and Coming, respectively, in place of the abstract form. For "go" and "come" themselves, the semantic structures will match the motional event/state shown wj th the other verbs with the exception of the type of eventhtate. The place objects of all the motional events can be considered the same, as can the way SR's apply to the different types of motion. We can also think of motional qualifiers as analyzable with the same structure. Because of this, the structure of movement predication will be considered in general and isolated from other forms.Thinking About Motion r A s was pointed o u t i n Section 11, one reason t h a t motional SR1s are d i f f i c u l t i s the m u l t i p l e predications of different types which must be orderable i n t i m e . These problems can be overcome with a p p r o p r i a t e consideration of the motion and t h e place objects of motional events.The i n e i g h t for a b e t t e r a n a l y s i s comes from considering answers t o questions of where motion occurs.Marathon was run. It i s probably some thing l i k e "in Greece" o r "from Marathon t o ~t h e n s " . These tend t o place the e n t i r e t y of motion. It i s u n l i k e l y t o be j u s t "from ~a r a t h o n " or "to Athens". These j u s t place, p a r t of t h e motioa. People tend I t t o locate motion as i f i t were a s i n g l e thing, a motion" so t o speak. This is 7.9 He walked through t h e puddle.how7.10 H e walked across t h e puddle. 7.11 He walked around the puddle. 7.12 He walked over the.puddle.As was pointed out i n Section 11, t h e aboue r e q u i r e a r e p r e s e n t a t i o n t h a t considers every i n s t a n c e of movement. The t r a c e i d e a does t h i s i n such a sidewal k FIGURE 7.2 A ball rolling across a sidewalk to a porch.way that the SR's can be shown applying to the trace directly. Further, it does it in a way that allows the basic static use of the preposition to be used in the representation:7.13 The bridges across the Mississippi are closed.This was pointed out in Section IV to be the same sense that applied in the I 1across-from" form:7.14 The man stopped across the street from here.Hence three usages collapse into one with this representation.This concept can be extended to allow for differentiating "up" and "down" by considering the solid traces to have an inherent ordering basedoon the direction of motion:7.16 He walked up the hill. 7.17 He walked down the hill.Hence, the traces in 7.16 and 7.17 could be exactly the same except for the ordering and the preposition could be sensitive to this. This ordering s e n s i t i v i t y shows up with other uses of the prepositions and other prepositions:7.18The c a r o t i d arteries extend up the neck t o the head. 7.19 A woman stood a t the f r o n t of the l i n e while a man stood a t the rear. 7.21 He walked out of the house.W i t h the above we can not say t h a t the overall path of motion wao either "into"the house or "out of" the house i n the s t a t i c sense of these prepositions Hoerever, there i s a way we could use the s t a t i c sense. I f we could r e f e r t o p o d t i o n e achieved by the moving object a s i t followed the path, we could say t h a t there were positions where the object f i r s t got t o be "into t h e corner" and "out of the house". This would be l i k e allowing reference discussed, then they too can be compared. For the phrase j u s t mentioned, a p a r t of t h e motional object t h a t~w a e across the yard could be compared t o a p a r t t h a t vas up the s t a i r s a s being l e e s f u r t h e r along it. The same could be done t o c a p a t e the parte involved with instantaneous reference.To sunuuarize3 t h e idea i s t o think of movement as a t r a c e of the event over t b , vbich has an inherent o r i e n t a t i o n and which cah be predicated i n p a r t . W e can now almost present our representation. W e w i l l f i r s t present a s l i g h t l y i n c a p l e k e proposal and then r w l s e i t .Tentatively, we propose two different functions to produce parts from complete place objects. These are called SEGMENT and UNIT. - The temporal ordering of the partial traces is the one tentative part of the analysis. To have it be sensible, some scale of comparison must exist. The appropriate choice appears to be the temporal scale. -When the locations were achieved is, of course, what is being ordered. There must also be conventions on application of the comparison. This is because there must be a way to force the comparison on only the appropriate end points of segments. We might develop a way of making these conventions inherent, but I propose to make them explicit. Our final proposal for the structure of motional SR's is to include time parameters with the functions. In this way, both the end points of the segments and the temporal scale can be identified. For the SEGMENT function, two linkr, T1 and T2, will identify the times that initial and final points were occupied.TheyFor the UNIT function, one link, T, will identify the time the final position war achieved. These structures are shown in Figure 7 .4 and 7.5.FIGURE 7.4 The motional elements in "John walked acreas the yard." An interesting aspect of the semantic structure of Figure 7 .6 is the static representation of "from". It is to be understood as showing that up to some point in the journey the moving object was not away from the house, but that it eventually got to be away from it. "Out of" and "off of" are analyzed similarly.Besides the ,durational and instantaneoue predications of motion, there can be overall predications of moving objects. These come in two forms. Adjuncts in movement sentences place the entirety of motion:7.22 In Chicago, he walked around the downtown. of whole place objects. Therefore, we must show the over411 predication applying to different forms. these must be the complete place objects representing the entirety of motion. This is consistant with our other anlayees, as will shortly be seen in more detail. It .will also simplify the inference rules that bring down overall spatial predications from higher levels to the motional place objects.This analysis is seen in Figure 7 .7 which essentially summarizes this section.We have introduced twd new functions and types of place objects. These have allowed for movement locative objects. We must, however, realize that there are other uses-for this analysis. We will see why in the next section. The last several sections presented the "syntax" of our semantic analysis.The term syntax is appropriate since the form of the analysis was presented. The semantics or interpretation to be given the proposed structures was only informally discussed, as when the trace or path analogy for motion was introducld.We noted in Section I11 that semantic nets do not just allow for the syntactic In this case, the concept has two defining attributes since a placelet must have a space and a time. The two attributes are shown accordingly with the one in the role restricted to be a SPACE and one in the role calledt being a -TIME. The concepts of SPACE and TIME will be treated as primitive, here.aspectThe structure we defined for place objects will be referenced whenevel place objects are used. One reference will be in eventlstatea where place objects are involved with the case P. Hence, with the conceptual definition of every type of event/state that has a location, there will be a definingattribute with role P and value-restriction Place Object.It is also the case that with each event/state, there will be a way to ehaw how the place object fits in with the .definition of the event/state. Thig will include the way in which the place object will be related to the participants in The everitlstate also has a structure identified by a special 5-C link, tor structural conditions, which is used to identify how the event/stete is struc- The above eqample a s s e r t s t h a t a t each i n s t a n c e during the walking, the walker was "in" with r@spect t o the p o s i t i o n of h i s shoes. Such examples r e q u i r e t h a t t h e o b j e c t ' s p o s i t i o n a t each i n s t a n t during t h e e v e n t l s t a t e be compared t o t h e l o c a t i o n of t h e event a t t h a t i n s t a n t .Ulowing for the preporititma]. concepts applying to place objects produced by the UNIT function must be done differently: A s*le interpretation of the prepositional concept in the above may be problematic. Since your arm could be in motion, a stationary observer would include sare w t i o n attributable to your arm in the ant's path. Further, even if r wanted to take the position of the arm at erne one instant it is unclear wbich t o take. These problems, however, dieappear with. the realization that the motion refereaced is not with respect t o an arbitrary observer but to one on the am.For him, the SR can be treated as involving not a moving arm but one essentially static in space. This can be allowed for by requiring the conceptual definition of the prepositions to project the referenced objects' positions shown in the place object in. the P case onto the base object shown in the G case. This will be like taking the base as a static ground and the referenced object as a figure seen against it.*Since the change into the final state m y be gradual and not dramatic, fuzzy relatioucl (~a d e h , 1973) might be used. For instance, the degree of "into"' ednees could be quantified with the analysis showing that a certain degree was reached. proieqtion function which takes the place object in the F role and the object in the G role and produces the projection, shown by an arc of that name, and an abstract space to compare it to, shown by the Abstraction arc.To summarize, the section has shown how conceptual level interpretation can be given the semantic structures proposed earlier in the paper. Any system that uses the semantic structures can also use the interpretations. Of course, the interpretations are baaed on one way of structuring plade objects. Since there are other ways, other interpretations are possible.There are definite limits to the claims we wish to make. In this concluding section, we point out several half-solved and unsolved problems, one area where we could conceivably expand our claims, and then end with a summary and final defense.Metaphorical usages are i~lportant but difficult subjects for semantic representation. Things like "climbing the ladder of success" are far enough away from spatial reference to be ignorable. However, some SR phenomena appear to be metaphors: 20.1 John yelled his greetings to John.In the above, an imaginary object, "his greetings", seems to be sent through space. In the.following, a hypothetical journey is referenced: Others seem t o coordinate w i t h SR's:10.5 Go s t r a i g h t i n t o t h e house.Here " s t r a i g h t " c a n be shown a a p r e d i c a t i n g t h e p a r t of t h e journey up t o t h e t i m e t h e house was e n t e r e d . However, I do n o t know how many o t h e r terms remain t o b u t t h e l a t t e r s a y s t h a t a c e r t a i n type of e v e n t must occur when t h e person i s i n -New York. I have s o l u t i o n s f o r n e i t h e r problem. W e can o n l y a p p e a l t o t h e f a c t t h a t t h e s e phenomena do p r e s e n t problems i n many o t h e r a r e a s of s e m a n t i c s . e v e n t s w i t h space-time zones" i n s t e a d of times and s p a c e s . How t h i s would be done remains t o be s e e n . However, i f w e have n o t a l r e a d y met our gdal of p u t t i n g s p a c e on a par with t i e , t h a t would c e r t a i n l y do i t .beIn suampry, this paper has ahown how the .emantic rtructure of 8pati.l reference8 can be shown as locating events and states of affairs. within a semantic net, this has the form of showing a location ae an attribute of event/state nodes. In line with this, the concept of a place object, showingwhere events and states of affairs held at instances of time, was developed.Several functions were developed for use in predicating locations. Inferencing of spatial facts, the use of prepositional-like concepts for showing spatial relatibnships, and the overall semantic structure of utterances was also discussed.Throughout the paper, the main jus tif ication has been that the analysis handles phenomena that other analyses do not. However, there are other j u r t i f ications. Only one source for space simplifies the modeling of spatial phenomena.Using only static forms simplifies the interpretation of spatial terms. Also, the use of static forms fits in with proposals for state-based semantic representations (~ercone and Schubert, 1975) . Finally, we can see that the analysis of )I semantic structures, in general, fits in with deeper" analyses of semantic structure such as Schank (1973) and Rumelhart (1975) . In sum, there appears to be a strong case for the analysis. | null | Main paper:
san francisco is north of los angeles. 4.31 the car is to the left of the building.:
These examples can immediately be given prepositional-like semantic structures.In other sentences, these terms appear as nouns and adjectives: A diverse variety of non-prepositional locative adverbs can be handled with Bennett (1975) points out, the "across1' and "from" phrases combine in euch a way that we understand that it is the way that must be travelled in starting from home and going to John that is "across the streetB'.* This can be allowed for in semantic nets with a function, -WAY, producing a path through space joining two points identified by INIT for Initial and FIN for Final links, see In our semantic net model, the locations of events and states of affairs will be shown as attributes of e v e n t h a t e nodes through arcs leading from the nodes to locational entities. For each event/state node involved with an SR there will be only one such arc and locational entity. Theee atcs will be labelled -P to suggest a spatial attribute or "Place" case. The locational entities will be referred to as place objects. They are the basis of our analysis. These place objects can be taken for the time being as volumes in space. The sort of ~o l u m e they are will be ela6orated upon. Place objects will be identified by boxes. We are finally far away from New York.Finally, direct analysis can be given to qualifier usages which either apply with a static sense to nouns describing physical objects (5.8) or act like static adjuncts with respect to verbal noun8 ( 5 . 9 ) :5.8 T k b man i n the car l e f t .5.9 swimming i n the lake i s fun.These qualifier usages can be contrasted with those that ahow motian (5.10), act l i k e locative objects to verbal nouns ( 5 . 1 1 ) or @how extent (5.12):5.10 The bus t o Chicago l e f t . 5.11 ~w i m i n g i n to a cave i s fun.The bridge from Ohio to Weet Virginia i e old.One t e s t for adjuncts i n the last section was to see i f it located the e n t i r e t y of the event or s t a t e discussed. The s t a t i c adjuncts are i d e n t i f i a b l e in t h i s way.Since the place object shows the location of that e n t i r e t y , static adjuncts can therefore be directly a p p l i e d t o them. The ice bag is to John's head, but both the ice bag and John are in the car. The first SR involves a locative object, the second an adjunct. With a simple approach to e v e n t h a t e location, they would not be differentiated. There is a similar problem in some adjunct references to the location of only part of an event or state of affairs. For instance, in 6.2, only the boy isplaced which the hawk is definitely physically present:6.2In an open field, a boy watched a hawk.In the latter case, although not the former, the use of inference rules might be suggested. However, a better answer can be found.The I heard through t h e door.On these occasiow, an assumed e n t i t y can be added t o the semantic s t r u c t u r e : I heard ( sane thing) through the door.Inference rules play an important part in these analyses. For example, the positioning of John and the ice bag must be derivable from the structure of There are a number of other classes of verbs that take static locative objects, see Table 1 . We will survey their analysis in the remainder of the section and close with a comment on several related forms. : grab, hit, kick, kiss, kneel, punch, slap, slug , touch. : break, chop, cook, cut, fry, shatter, spill, 6 .4.FIGGRE 6.4 He is j u s t now in the house.Problems with motion arise in every analysis of SR considered in Section 11.In Only "take" and "bring" show the same pattern. For this reason, an abstract concept of pure motion, called -GOING, will be used in our analysis. Figure 7.1 shows the sentential structure into which most movement SR's will t . The structures for "take" and "bring" will have Going and Coming, respectively, in place of the abstract form. For "go" and "come" themselves, the semantic structures will match the motional event/state shown wj th the other verbs with the exception of the type of eventhtate. The place objects of all the motional events can be considered the same, as can the way SR's apply to the different types of motion. We can also think of motional qualifiers as analyzable with the same structure. Because of this, the structure of movement predication will be considered in general and isolated from other forms.Thinking About Motion r A s was pointed o u t i n Section 11, one reason t h a t motional SR1s are d i f f i c u l t i s the m u l t i p l e predications of different types which must be orderable i n t i m e . These problems can be overcome with a p p r o p r i a t e consideration of the motion and t h e place objects of motional events.The i n e i g h t for a b e t t e r a n a l y s i s comes from considering answers t o questions of where motion occurs.Marathon was run. It i s probably some thing l i k e "in Greece" o r "from Marathon t o ~t h e n s " . These tend t o place the e n t i r e t y of motion. It i s u n l i k e l y t o be j u s t "from ~a r a t h o n " or "to Athens". These j u s t place, p a r t of t h e motioa. People tend I t t o locate motion as i f i t were a s i n g l e thing, a motion" so t o speak. This is 7.9 He walked through t h e puddle.how7.10 H e walked across t h e puddle. 7.11 He walked around the puddle. 7.12 He walked over the.puddle.As was pointed out i n Section 11, t h e aboue r e q u i r e a r e p r e s e n t a t i o n t h a t considers every i n s t a n c e of movement. The t r a c e i d e a does t h i s i n such a sidewal k FIGURE 7.2 A ball rolling across a sidewalk to a porch.way that the SR's can be shown applying to the trace directly. Further, it does it in a way that allows the basic static use of the preposition to be used in the representation:7.13 The bridges across the Mississippi are closed.This was pointed out in Section IV to be the same sense that applied in the I 1across-from" form:7.14 The man stopped across the street from here.Hence three usages collapse into one with this representation.This concept can be extended to allow for differentiating "up" and "down" by considering the solid traces to have an inherent ordering basedoon the direction of motion:7.16 He walked up the hill. 7.17 He walked down the hill.Hence, the traces in 7.16 and 7.17 could be exactly the same except for the ordering and the preposition could be sensitive to this. This ordering s e n s i t i v i t y shows up with other uses of the prepositions and other prepositions:7.18The c a r o t i d arteries extend up the neck t o the head. 7.19 A woman stood a t the f r o n t of the l i n e while a man stood a t the rear. 7.21 He walked out of the house.W i t h the above we can not say t h a t the overall path of motion wao either "into"the house or "out of" the house i n the s t a t i c sense of these prepositions Hoerever, there i s a way we could use the s t a t i c sense. I f we could r e f e r t o p o d t i o n e achieved by the moving object a s i t followed the path, we could say t h a t there were positions where the object f i r s t got t o be "into t h e corner" and "out of the house". This would be l i k e allowing reference discussed, then they too can be compared. For the phrase j u s t mentioned, a p a r t of t h e motional object t h a t~w a e across the yard could be compared t o a p a r t t h a t vas up the s t a i r s a s being l e e s f u r t h e r along it. The same could be done t o c a p a t e the parte involved with instantaneous reference.To sunuuarize3 t h e idea i s t o think of movement as a t r a c e of the event over t b , vbich has an inherent o r i e n t a t i o n and which cah be predicated i n p a r t . W e can now almost present our representation. W e w i l l f i r s t present a s l i g h t l y i n c a p l e k e proposal and then r w l s e i t .Tentatively, we propose two different functions to produce parts from complete place objects. These are called SEGMENT and UNIT. - The temporal ordering of the partial traces is the one tentative part of the analysis. To have it be sensible, some scale of comparison must exist. The appropriate choice appears to be the temporal scale. -When the locations were achieved is, of course, what is being ordered. There must also be conventions on application of the comparison. This is because there must be a way to force the comparison on only the appropriate end points of segments. We might develop a way of making these conventions inherent, but I propose to make them explicit. Our final proposal for the structure of motional SR's is to include time parameters with the functions. In this way, both the end points of the segments and the temporal scale can be identified. For the SEGMENT function, two linkr, T1 and T2, will identify the times that initial and final points were occupied.TheyFor the UNIT function, one link, T, will identify the time the final position war achieved. These structures are shown in Figure 7 .4 and 7.5.FIGURE 7.4 The motional elements in "John walked acreas the yard." An interesting aspect of the semantic structure of Figure 7 .6 is the static representation of "from". It is to be understood as showing that up to some point in the journey the moving object was not away from the house, but that it eventually got to be away from it. "Out of" and "off of" are analyzed similarly.Besides the ,durational and instantaneoue predications of motion, there can be overall predications of moving objects. These come in two forms. Adjuncts in movement sentences place the entirety of motion:7.22 In Chicago, he walked around the downtown. of whole place objects. Therefore, we must show the over411 predication applying to different forms. these must be the complete place objects representing the entirety of motion. This is consistant with our other anlayees, as will shortly be seen in more detail. It .will also simplify the inference rules that bring down overall spatial predications from higher levels to the motional place objects.This analysis is seen in Figure 7 .7 which essentially summarizes this section.We have introduced twd new functions and types of place objects. These have allowed for movement locative objects. We must, however, realize that there are other uses-for this analysis. We will see why in the next section. The last several sections presented the "syntax" of our semantic analysis.The term syntax is appropriate since the form of the analysis was presented. The semantics or interpretation to be given the proposed structures was only informally discussed, as when the trace or path analogy for motion was introducld.We noted in Section I11 that semantic nets do not just allow for the syntactic In this case, the concept has two defining attributes since a placelet must have a space and a time. The two attributes are shown accordingly with the one in the role restricted to be a SPACE and one in the role calledt being a -TIME. The concepts of SPACE and TIME will be treated as primitive, here.aspectThe structure we defined for place objects will be referenced whenevel place objects are used. One reference will be in eventlstatea where place objects are involved with the case P. Hence, with the conceptual definition of every type of event/state that has a location, there will be a definingattribute with role P and value-restriction Place Object.It is also the case that with each event/state, there will be a way to ehaw how the place object fits in with the .definition of the event/state. Thig will include the way in which the place object will be related to the participants in The everitlstate also has a structure identified by a special 5-C link, tor structural conditions, which is used to identify how the event/stete is struc- The above eqample a s s e r t s t h a t a t each i n s t a n c e during the walking, the walker was "in" with r@spect t o the p o s i t i o n of h i s shoes. Such examples r e q u i r e t h a t t h e o b j e c t ' s p o s i t i o n a t each i n s t a n t during t h e e v e n t l s t a t e be compared t o t h e l o c a t i o n of t h e event a t t h a t i n s t a n t .Ulowing for the preporititma]. concepts applying to place objects produced by the UNIT function must be done differently: A s*le interpretation of the prepositional concept in the above may be problematic. Since your arm could be in motion, a stationary observer would include sare w t i o n attributable to your arm in the ant's path. Further, even if r wanted to take the position of the arm at erne one instant it is unclear wbich t o take. These problems, however, dieappear with. the realization that the motion refereaced is not with respect t o an arbitrary observer but to one on the am.For him, the SR can be treated as involving not a moving arm but one essentially static in space. This can be allowed for by requiring the conceptual definition of the prepositions to project the referenced objects' positions shown in the place object in. the P case onto the base object shown in the G case. This will be like taking the base as a static ground and the referenced object as a figure seen against it.*Since the change into the final state m y be gradual and not dramatic, fuzzy relatioucl (~a d e h , 1973) might be used. For instance, the degree of "into"' ednees could be quantified with the analysis showing that a certain degree was reached. proieqtion function which takes the place object in the F role and the object in the G role and produces the projection, shown by an arc of that name, and an abstract space to compare it to, shown by the Abstraction arc.To summarize, the section has shown how conceptual level interpretation can be given the semantic structures proposed earlier in the paper. Any system that uses the semantic structures can also use the interpretations. Of course, the interpretations are baaed on one way of structuring plade objects. Since there are other ways, other interpretations are possible.There are definite limits to the claims we wish to make. In this concluding section, we point out several half-solved and unsolved problems, one area where we could conceivably expand our claims, and then end with a summary and final defense.Metaphorical usages are i~lportant but difficult subjects for semantic representation. Things like "climbing the ladder of success" are far enough away from spatial reference to be ignorable. However, some SR phenomena appear to be metaphors: 20.1 John yelled his greetings to John.In the above, an imaginary object, "his greetings", seems to be sent through space. In the.following, a hypothetical journey is referenced: Others seem t o coordinate w i t h SR's:10.5 Go s t r a i g h t i n t o t h e house.Here " s t r a i g h t " c a n be shown a a p r e d i c a t i n g t h e p a r t of t h e journey up t o t h e t i m e t h e house was e n t e r e d . However, I do n o t know how many o t h e r terms remain t o b u t t h e l a t t e r s a y s t h a t a c e r t a i n type of e v e n t must occur when t h e person i s i n -New York. I have s o l u t i o n s f o r n e i t h e r problem. W e can o n l y a p p e a l t o t h e f a c t t h a t t h e s e phenomena do p r e s e n t problems i n many o t h e r a r e a s of s e m a n t i c s . e v e n t s w i t h space-time zones" i n s t e a d of times and s p a c e s . How t h i s would be done remains t o be s e e n . However, i f w e have n o t a l r e a d y met our gdal of p u t t i n g s p a c e on a par with t i e , t h a t would c e r t a i n l y do i t .beIn suampry, this paper has ahown how the .emantic rtructure of 8pati.l reference8 can be shown as locating events and states of affairs. within a semantic net, this has the form of showing a location ae an attribute of event/state nodes. In line with this, the concept of a place object, showingwhere events and states of affairs held at instances of time, was developed.Several functions were developed for use in predicating locations. Inferencing of spatial facts, the use of prepositional-like concepts for showing spatial relatibnships, and the overall semantic structure of utterances was also discussed.Throughout the paper, the main jus tif ication has been that the analysis handles phenomena that other analyses do not. However, there are other j u r t i f ications. Only one source for space simplifies the modeling of spatial phenomena.Using only static forms simplifies the interpretation of spatial terms. Also, the use of static forms fits in with proposals for state-based semantic representations (~ercone and Schubert, 1975) . Finally, we can see that the analysis of )I semantic structures, in general, fits in with deeper" analyses of semantic structure such as Schank (1973) and Rumelhart (1975) . In sum, there appears to be a strong case for the analysis.
analyses using physical.objects:
The paradigmatic phenomenon for tte analyses that claim physical objects as the referents of S R 1 s is the noun phrase modifier: This figure shows "the man" being located (LOCI at a time, indicated by the T -link, and at a location which was in "the car". This style of analysis seems simple and direct. It appeals to the intuition that only physical objects take up space. It promises t~ be easy to apply, FIGURE 2 . I "The man i n t h e c a r l e f t " i n t h e st-yle o f Schubert (1936) . s i n c e a l l t h a t i s required i s t o a s s o c i a t e SR's with t h e s e n t e n t i a l elements which a r e modified which r e f e r e n c e physical o b j e c t s . Unfortunately, t h e r e are problems.Often t h e r e i s more t o an event than. j u s t i t s partiripahts' l o c a t i o n s :2.2 John i s playing s o l i t a i r e i n the basement.If we hear example 2.2 t h e n more than John i s known t o be i n the basement. His c a r d s a r e , for. example. F u r t h e r , t h e l o c a t i o n of t h e a c t i o n i s more than t h e i n s t a n t a n e o u s p o s i t i o n of John and his c a r d s . For example, space where t h e cards may, p o t e n t i a l l y be plaoed m u s t be included. S i m i l a r l y , t h e following does not Here, successive changes seem to be appropriate. However, one class of references to motion seems to defeat this entire approach:2.8 The man walked across the puddle. 2.9 The man walked around the puddle.The man walked through the puddle.Examples like the above involve duration in a key way and can not be shown with reference to one position. For example, at no time was the man "across" the puddle like ~aleigh's cloak was across it. Similarly, two points showing the man's change of position are inadequate since the same initial and final positions are a'cceptable in all three cases. Finally, adding an intermediate point will not be adequate, since the man might reach that point while on a path that otherwise holds a different relation to the puddle. As shall be seen, the lesson to be learned from these examples is that in allowing for motion, it is the entire path that must be considered and not selected positions of objects. 2.12 John walked from his car across the yard to the ilouse.How the event of 2.12 can be "from", "across", and "to" simultaneously and also have these aspects temporally ordered is nowhere explained in these analyses.Finally, even if SR's are associated with events and states of affairs, the fact that something is often learned about participants' location must be explained. The t h i r d s t y l e of SR a n a l y s i s i s nonuniform i n nature. These e i t h e r mix the two uniform analyses or e l a b o r a t e on the simple event o r s t a t e a n a l y s i s .Mixed analyses claim t h a t some SR's l o c a t e concrete o b j e c t s while some l o c a t e events or s t a t e s of a f f a i r s (see for example, Winograd, 1972, and Schank, 1973) By s a c r i f i c i n g the s i m p l i c i t y t h a t comes from uniformity , these analyses avoid the uniform analyses' complementary problems. However, the mutual problems, e s p e c i a l l y motion, are l e f t unsolved.The nonuniform analyses t h a t e l a b o r a t e on the nature of events and s t a t e s of a f f a i r s a r e best represented by Case analyses, see Bruce (1975) . . I n terms of events and s t a t e s of a f f a i r s , t h e f i r s t case can e i t h e r be used f o r overall event o r s t a t e l o c a t i o n o r i t may be used t o l o c a t e an a s p e c t of t h e event. The f i n a l t h r e e c a s e s a l l r e l a t e t o d i f f e r e n t a s p e c t s of a motional event. This allows f o r examples l i k e 2.12, with i n h e r e n t temporal o r d e r i n g among t h e cases allowing f o r t h e o r d e r i n g of t h e SR's.The Case a n a l y s e s s t i l l has problems The underlined phrases r e f e r t o motion ordered i n time, e . g . , he walked t h e h i l l before t h e b r i d g e . However, Case a n a l y s i s g i v e s no way t o order i n s t a n c e s of the same c a s e . Gruber (1965) p o i n t s out t h e same problem with t h e Goal case:2.14 I walked t o New York t o my mother's.F i n a l l y , t h e Case proposal rnuat be given some physical i n t e r p r e t a t i o n . Any r e p r e s e n t a t i o n of meaning must a t some p o i n t be r e l a t e d t o a m~d e l of t h e world.I n t h i s i n s t a n c e t h e i d e a of a source, g o a l , and path must be somehow r e l a t e d t o models of motion.This paper p r e s e n t s a proposal f o r an a n a l y s i s t h a t i s nonuniform i n t h e same way t h e Case a n a l y s i s i s . A uniform source for l o c a t i o n s modified by SR's i sgiven, but t h e p r e d i c a t i o n of t h e s e spaces by SR's i s shown t o be much more complex than previously thought. F u t t h e r , sentences are not seen as being as simple with respect to SR's as previously supposed.Before presenting the analysis, two sections will be devoted to preliminary topics: our semantic net formalism and the eyntactic status o f the phenomena considered. (19751, Shapiro (,1971), Simmons (19731, and Woods (1975) All this information is essential to any artificially intelligent entity, just as the model is essential'to any analysis in the predicate calculur. However, for showing the semntic relations in which we are mainly interested, an abbreviation is sufficient just as only the formulas are sufficient in most studies using symbolic logic. Hence a special abbreviation will be used in all sections except fX where the definitional level wili be discwsed.
semantic nets:
Central to our abbreviation will be nodes that collapse types and tokens.These will identigy the verbal concepts that characterize the events and states of affairs. We will call them "event/state" nodes. They will be circled and capital letters will be used for abstract types, such as CAUSING. Nonabstract forms will be shown with names that suggest the interpretation, e.g., Sleeping will suggest the sleeping state. When a node represent8 a physical object, identifying information will be included in quotes, e.g., "the bus". Names placed on ascs will abbreviate and suggest the functional roles of attributes.For example, -ANTE for antecedent and -CONS for consequence will be used with CAUSING. Case names will be used with many event and state of affairs types.T for "Time" showing the time an event occurred or state held.-A for "Agent" showing the instigator of an event or state.-0 for "Object", the neutral case (as Fillmore (1971) explains it "the wastebasket") .Restrictions on types of entities which will be necessary will be shown by nonoval shapes for nodes. For example, time instances will be shown in parentheses and time intervals in square brackets. Finally, because it is not essential for our purposes, specification of time will often be left out of most semantic structures. Similarly, ire will consider only declarative statements. Some concepts that act as functions will also be used. Each of these will look like a relation associating parameters with a value. The value will be identifiied by a -VALUE arc. Inference rules will be presented in the form ofLoving \ \ (I I MOryl l ) [ 1.11 ] FIGURE 3 . 1 "John loved Mary all last year." "subnetll' 2 "subnet2", where on seeing slibnet 1' subnet* is to 'be added to the semantic net. These rules will include variables within nodes, where the variables are to be bound on matching and referenced on inferencing. These variables will be in the form of capital letters, e.g., X.To summarize, our semantic net formalism uses concept names, descriptions of objects, mnemonic arc names, and mnemonic shapes for nodes to abbreviate the two levels in a semantic net. Also used are functions and inference ruLes. This will be enough to represent the semantic relations involving reference to space that are being considered. Unfortunately, it is one more unique formalism. However, it sdds no new structures, only abbreviating others. We leave as an unproven claim that it will fit in with any formalism which shows identifiable event and state of affairs nodes such as Norman add Rumelhart (1975) and Schank (1973) . In this sectian, the syntax function of prepositional phraees will be considered and arguments for their primacy will be presented.Our main interest in syntax is in structuring our didcussion of semantics.However, the problems of parsing and generation make the syntax of' SR's independently imprtant. These are not our topics here. However, in an earlier issue of this journal we presented a parsing scheme that produces semantic from syntactic structure and applied the scheme to current clags of phenomena (Sondheimer and Perry, 1975). There is some controversy on the distinction between these two types. We can present two syntactic and one semantic classification procedures. First, adjuncts are never required for granrmaticality, while locative objects can be: These are the four primary uses of locative prepositions. We claim that the semantic structure of other SK'B can be represented through these forms. We will now show this. In general, this will be done by observing the SR's structure or by paraphrase arguments.Some spatial terms can have syntactic and semantic functions similar to prepositions in that they directly serve to relate two forms:
:
state by arcs: indicating its staus as the spatial attribute. A few SR's are shown as naming these locational entities, which we call place ,object. These SR' s involve examples with "where", "here", and "there" However, most SRts are represented as relating place objects to the position of objects in the manner of prepositional phrases. This primacy ok prepositions is argued for in the paper. Motional references are allowed for by functions represented in the nets which produce parts of place objects which are then positioned by prepositional f c m s . The necessary ordering that'comes with motional references is allowed for by associating temporal elements with the functions.While the positioned elements are simple, the overall semantic structure of the sentences containing SR's is often complicated by the involvement of more than one event or state of affairs. The paper includes a survey of the sentential semantic structures necessary to deal with SR's.A similar complexity is necessary to deal with the informakion on the location of objects which is gained from sentences with SR's. The paper suggests-the use of inference rules to allow for this.The most surprising of the paper's oonclusions is that a strong tie exists between referehces to space and temporal information. In fact, the locations of all events and states of affairs placed by SR's are argued to be locations in both space and time. The effect of this conclusion is most clearly seen in a formalized definition of the primitives of the semantic seructures , which is also presented in semantic nets. There, as ane possible interpretation of the place object, it is shown as a set of pairs of volumes in space and points in time. . . . . This paper presents an analpis in a semantic net formalism of the semantic structure of English sentence8 containing references to spatial location.Spatial referefnce, hereafter -SR, provides either'static location or motional information :1.1 John is at home 1.2 Fred ran across the street to the store.The task for the semantic analysis of sentences with SR's is to make clear what is being positioned. This has been difficult to do. Previous proposals have left unanalyzed many phenomena including important motional references. This paper's main conclusion is that a much.improved analysis can be obtained by representing the SR's as positioning ab~tract events and states of affairs.The analysis in semantic nets has the location of an event or state ~f affairs represented as a node which is linked to the node showing the event or state by arcs indicating its status as the spatial attribute. A few SR's are shown as naming these locational entities, which we call pLaee object. These SR's involve examples with "where", "here", and "therei'. However, most SR' s are represented as relating place objects to the position of objects in the manner of prepositional phrases. This primacy of prepositions is argued for in the paper.Motional references are allowed for by functions represented in the nets which produce parts of place objects which are then positioned by prepositional forms.The necessary ordering that comes with motional references is allowed for by associating temporal elements with the functions.While the positioned elements are simple, the overall semantic structure of the sentences containing SR's is often complicated by the involvement of more than one event or state of affairs. The paper includes a survey of the sentential There is available a discussion in greater detail of a preliminary analysis to the one given here (~ondheimer, 1975) . There is also available for comparison an analy is by this authbr of the same meaning phenomena, in the competing paradigm of model-theoretic semantics (Sondheimer , 1978) The current.paper is distinguishable by its better developed semantic net formalism nnd i t n emphasis 3n producing computationally j u s t i f i e d structures.Ihe pazt has seen many studies of SR phenomena. There ha8 been interest in connecting. language and scenes, e. g. , Coles ( 19681, Kochen ( 19691, Winograd (1972) are claimed to locate. In some cases, the SR'B apply to only physical objects..In athers, they apply to only abstract forme identifying events and states o affairs. A broad third type of analysis shows different sorts of entities being modified. Each has.its limitations.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | Problem: Difficulty in analyzing the semantic structure of English sentences containing spatial-location references, especially in relation to motional information.
Solution: Proposing an improved analysis by representing spatial references as positioning abstract events and states of affairs within semantic nets, where the location of an event or state of affairs is linked to nodes showing the event or state, indicating its spatial attribute. | 566 | 0 | null | null | null | null | null | null | null | null |
db39f56b9c186cc7325575bdaa918b1bd83694a4 | 219309031 | null | Computation in Departments of Linguistics | That computers and linguists meet, for the host part, only in the skill sorne~hat evotic field of computational linguistics is $ sad statement about the st2te of ordinary linguistic research The titre bhen computers were to be considered only the t o o l of the natural scientist or the statist$cally minded social s c i e q t i s t i s long past, 'word processing technology' is now the specialty of a growing number of computer companies Not only can this techrology be of great value i n reducing the clerical burden of the linguigt and linguistics student, but, iiinguists, as specialists who have been studying and manipulating language far years, are i n a position t o b e contributing to t h i s f i e l d Jn fact, in many areas of linguistic research the analysis of particular languages, the search for li~rgui&tic universal s, the analysis of discourse and text, computar technology can bc of help tc the l i n g u i s t , and, in many subfields of computer science automated lnngua~e processing, the deslgn of human/machme i ~t e r f ~c e s , the structuring of data bases, linguistics has much to offer the ccnnputer scientist, vet up until how, relatively few such cross contributions have been made Computer scientists have been slow to discgvei the v d u e of Ilnguistirs to their w o r ~, the tine has come for linguists t o take the initiitivc and t o train themselves (and their students) to hake use of and contribute to the field of computer science, Speci?lized traltning in the us& of t h e corrputer with-ln a particular discipline is not new Students i n mary soclal sciences nok flnd themselves facing Lncreaslng pressure and rnanaatorf rcquirnQents to take c o ~p t s r training wlthin the* department, f i f l g h i ~t i c s is, In fact, unusual III not having such requirements or even oppur t u n i t i e s A t a time wnen graduating l i n g u i s t i c s students a l . facing a shrinking job market, the oppurtunity t o be trained ia s ~commercjallj useful application of lin~uistics ougnt t o be attractive to many students Today, in most u n f v e r s i t i ~, coaput41,g i s dvililable t o linguistics | {
"name": [
"Fritzson, Richard"
],
"affiliation": [
null
]
} | null | null | null | 1978-06-01 | 0 | 0 | null | departments only through the use o f a l a r g e , c e n t r a l university computer which is expected t o he of aervice t o a l l university departments. But, a s computer casts continue t o f a l l , and, as l a r s e computing centers continue t o be unresponsive to the needs of t h e i r new users, i t w i l l not be qneammon t o find more and mbre departments purchasing their owh computing f a c i l i t i e s and buying o r developing their own software This is already happening today, both by externally funded individual researchers and by e n t i r e departments i n need of specialized computing f a c i l i t i e s What kinds of computing equipment a r e available4or a l i n g u i s t i c s department crying t o equip i t s e l f today?My Bnswer is structured, to some extent, by the organization b f language It is widely understood, even by non-stratificational l i n g u i s t s , t h a t the faculty of language is based on a s t a c k of structured systems, each one building a large number of u n i t s above from a smaller number below, i,e a handful of phonetic features combine t o form less than f i f t y phonemic segments which combine t o form thousands of morphemes,tens or hundreds of thousands of words, an i n f i n i t e number of sentences and texts expressing countless ideas and concepts It w i l l not be surprising t o find t h a t as one climbs t h i s &tack, from phonology upward, the amount of computing powet needed t o perform useful t a s k s and research increases i n proportion t o the increasing number of u n i t s and t h e complexity of t h e i r structuring I w i l l concernmyself, mostly, with the p o s s i b i l i t i e s a v a i l a b l e for the study of the lower l e v e l s This PB because the type of linguistic work being done i n t h e study of the semantic and cognitive levels is s t i l l primarily research and the people involved ere more l i k e l y t o already know t h e i r needs and options as far as computing goes Also, since t h e cost of computing i n these areas is somewhat higher, it is less l i k e l y t h a t department& w i l l be doing their own purchasing f o r these purposesThe 8tudent of phonology, morphology and l i n g u i s t i c f f e l d a n a l y s i s is concerned .,primarily with the manfpulation of l i n g u i s t~c t e x t , expressed as a If all or most o f the termint\le i n a department are CRT type termlnala, i t w i l l be necessary to provide some means of producing 'hard copy' output on paper While most interactions with a computer can take p l a c e on a screen, some record of the r e s u l t s of a session will be needed for study and evaluation Printers which can handle the fhexible type fonts needed by l i n g u i s t s are available They are fast they operate in the dame way that copyillg wachines work and simply transfer the contents of the CRY screen to the paper (including Syntax i s , perhaps, the most widely studied sub J ect i n linguistics today Given that t h i s is so, there i g a real need for linguists, both profess$onal and student, to understand the extreme d i f f i c u l t y of the task of writing a grammar for a language That attempts are made to do this without the aid of a computer is perhaps a l l the evidence one needs t o see that the difficulties are not well understood. A formal granmar, particularly one written i n the notations commonly used today, is very much like a computer progrm It is a l i s t of instructions for generating a list o f strings, a computer program is a l i s t of instructions for performing some process (which might be generating a list of strings) Both need to be precise, both are very complex, both suffer from the fact that a change i n one part of the ordered l i s t may cause an unanticipated change in the effect of another part It would be very surprising to find that l i n g u i s t s were better a t producing untested, yet correct, #formal charac'terizations of complex processes than computer programmers I eXpect that testing a newfy written gtammar w i l l be as enlightening as experience far a lihguisaics student as debugging a new codolex program is for a gompuer science student, Furthennore, just as the computer is sf uae i n studying phonology and | null | null | null | null | Main paper:
:
departments only through the use o f a l a r g e , c e n t r a l university computer which is expected t o he of aervice t o a l l university departments. But, a s computer casts continue t o f a l l , and, as l a r s e computing centers continue t o be unresponsive to the needs of t h e i r new users, i t w i l l not be qneammon t o find more and mbre departments purchasing their owh computing f a c i l i t i e s and buying o r developing their own software This is already happening today, both by externally funded individual researchers and by e n t i r e departments i n need of specialized computing f a c i l i t i e s What kinds of computing equipment a r e available4or a l i n g u i s t i c s department crying t o equip i t s e l f today?My Bnswer is structured, to some extent, by the organization b f language It is widely understood, even by non-stratificational l i n g u i s t s , t h a t the faculty of language is based on a s t a c k of structured systems, each one building a large number of u n i t s above from a smaller number below, i,e a handful of phonetic features combine t o form less than f i f t y phonemic segments which combine t o form thousands of morphemes,tens or hundreds of thousands of words, an i n f i n i t e number of sentences and texts expressing countless ideas and concepts It w i l l not be surprising t o find t h a t as one climbs t h i s &tack, from phonology upward, the amount of computing powet needed t o perform useful t a s k s and research increases i n proportion t o the increasing number of u n i t s and t h e complexity of t h e i r structuring I w i l l concernmyself, mostly, with the p o s s i b i l i t i e s a v a i l a b l e for the study of the lower l e v e l s This PB because the type of linguistic work being done i n t h e study of the semantic and cognitive levels is s t i l l primarily research and the people involved ere more l i k e l y t o already know t h e i r needs and options as far as computing goes Also, since t h e cost of computing i n these areas is somewhat higher, it is less l i k e l y t h a t department& w i l l be doing their own purchasing f o r these purposesThe 8tudent of phonology, morphology and l i n g u i s t i c f f e l d a n a l y s i s is concerned .,primarily with the manfpulation of l i n g u i s t~c t e x t , expressed as a If all or most o f the termint\le i n a department are CRT type termlnala, i t w i l l be necessary to provide some means of producing 'hard copy' output on paper While most interactions with a computer can take p l a c e on a screen, some record of the r e s u l t s of a session will be needed for study and evaluation Printers which can handle the fhexible type fonts needed by l i n g u i s t s are available They are fast they operate in the dame way that copyillg wachines work and simply transfer the contents of the CRY screen to the paper (including Syntax i s , perhaps, the most widely studied sub J ect i n linguistics today Given that t h i s is so, there i g a real need for linguists, both profess$onal and student, to understand the extreme d i f f i c u l t y of the task of writing a grammar for a language That attempts are made to do this without the aid of a computer is perhaps a l l the evidence one needs t o see that the difficulties are not well understood. A formal granmar, particularly one written i n the notations commonly used today, is very much like a computer progrm It is a l i s t of instructions for generating a list o f strings, a computer program is a l i s t of instructions for performing some process (which might be generating a list of strings) Both need to be precise, both are very complex, both suffer from the fact that a change i n one part of the ordered l i s t may cause an unanticipated change in the effect of another part It would be very surprising to find that l i n g u i s t s were better a t producing untested, yet correct, #formal charac'terizations of complex processes than computer programmers I eXpect that testing a newfy written gtammar w i l l be as enlightening as experience far a lihguisaics student as debugging a new codolex program is for a gompuer science student, Furthennore, just as the computer is sf uae i n studying phonology and
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 560 | 0 | null | null | null | null | null | null | null | null |
4f14ae96e95235ee99037385c57f90570a9c2171 | 219301746 | null | The {TARGET} {P}roject's Interactive Computerized Multilingual Dictionary | b o l d ' 'bond' 'bonding energy' 'boom' 'bore' 'hor inq b a r p 'boundary pos i t l on' 'bracb e 1,' Nofa lhnt Ihere is nothmg to slop olrlsrpnl rtcrtes for the same ! e r r sharmt f~sld codma A d~ffersnf fiche im needed whenever and only when, the equtvalente are dtet~nct 'breakdown* 'breaking' 'bridge connection' 'broker' 'brokerage' 'buffer' 'bur den' ' b i ~s i n e s ~' 'business prof it' 'butterfly valve' 'by means of' 'by-pa.,., valve' 'by" x-ray diff rac tion' 'capacity' 'c api t al' 'capital and reserves* 'cap1 t al gain' 'capital goods' 'capi tal-intensive' Targrt is the command (or program) we use to find equivalents for a rerm The TERMIN command Target works icleritically to the TARGET program (described above) w i t h tttc exception that in TARGtT i f an abort character cm is typed in answer to the Tcrnt: protnpt, the progam exits; in TERMIN, an r ; ~ at this point gets us back t o the commanrl prompt > 9 Record Transaction This command is 115cd to keep a record of the intcract~on between the program and the r l w r I t is ~rsccl for t,tt~c!yirre, how ur,ers interact with the system in order that i t may l~ecorne better ta~lorcd to tlirir ticcdr, (All of the examples in this document were d r a w n tlirectly from rccortlt, m a d r in cxactly ttiis manner). Each interaction between ttic syr.lem and the user may a1r.o bc t~med in the record by using the Tinung option. This p r o \ / i t l ~s an extra tool for s t ~~t l y i n e haw the system is used in practice. I t is also usefill for some purposes to he able to annotate a record while it is being produced. TEHMIN will ignore afiy line beginning with a semi-colon: > I Thls shows that commonts gnt Into the record | {
"name": [
"Burge, John"
],
"affiliation": [
null
]
} | null | null | null | 1978-06-01 | 0 | 0 | null | null | null | ----TERMlN ---------+-Program -- TERMIN --------isthe narne of the procram used by tcr ntinologists to augment and maintain the di'r t t o n a r~e s . Clearly the f a c l l i t~e s in TERMJN must be more varied than the simple This provicfes on-line access to written comnients on various aspects of the use of the TFRMIN program by terminologists All of the twenty or so tekts which may be acccsscd in this way were w r i t t c r~ h y the termtnologists (who also ordered the list of cornmat~ds as above and provided the brief description beside each one).( t o ex'ecuto a command f i l o ) ( t o s t a r t u s i n g an U p t i o n ) ( t o s t o p u s i n g s n O p t i o n )He lp a n t altorlng A L f E R I N G i s a u a y t o c o t ' r e c t e r r o r s u i t l~o~t t y p i n g t h e whole 1 i n e a g a i n .You c a n use I t hy u s i Most of thc tents pro\~ictccl in t t~~s way glvc hints and reminders i n an informal manner, rat her than a detailed seqiience of it~r~tructions on how 'to use the command, Tllic, command is i~c d to crcatr tirw rntrics in the dictionary for terms which have rjol. yrfQbeen entered Let us quppor.c llrat we have prepared a fiche specifying the e q u i v a l~n t in French of the Encl~qh t~r m b o t~d appropriate for the field of Chemistry. Space coristra~nts prevent a dct ailccl cic~cription of the interactions lead~ng to the entry of this term into the dictionary, but the follnw~ng is a trace of the process: his is how we leave TERMIN:T OX/^ E X I T~n e r r w r IS now no longer using the TERMIN program, but is using the TOPS-10 monitor.Tlds command is used to print wliole entries at our terminal. Here is bond: S~~p p o r~e that further investigation of the term bond has r~v c a l e c f that the same Frcnch cak~~valcnt 15 also used for the "bond" holding the nirclcirs of atoms together as for the clectronlc bond which keeps different atoms together. Had this been known when the orig~nal fiche was created, i t would have contained two field codes, cli4 and at8, We need to update the original entry.The term bortd is also irsed in financial and commercial circles This gives an alphabetrcal l~s t of terms in a specifled language. We can limit the list by spccifiying the f~r s t one or two letters of the f~r s t and last term. It may be aborted by typing m at any time, and will return us to the command level immediately. This command i s *used to re-c:.tablish the links between the various files which coot a~n t t~e dictiot~flry. They can hccorne incorrect when the computer crashes while c e r t a i n operations are being pcrformcd, or when there are problems w i t h the system.W t~r t n a term i~, stored after tlav~nc been Edtted, a new entry is made for the new version of the term Tlic old entry IS, however, still there, and hence takes up space.The qatne is t r u e in t l i~ CWIP of the D r l r t~ command. When a term is delefed, it actually otdy t~ccotncs ~n a r r c s~~i b l c ! -arid r,a ~t 1s taking ilp spatp. Every so often this "wasted" space (wl~ic h actually provides Ihe potential for some backup). is recovered using this command; i t compacts the dictioriaryAs c a n be secn from the forcegoing, to use a command we type i t s name (e.g. help) . llpppr r a s e anrl lowcr c a w a r~ ~q~~a l l y ac~~ptaI>le. The program will ttien begin to prompt 11s for the further spccificationr; necccssary to carry out the command. We may T y p r /Ihcnd the rcc,ponseq to thc~,e qtrcstions, in which case the prompt Is not given Help is u~u a l l y obta~nahle by typtng ? and any command can be aborted by typing @ it, re.,ponc,e to any prornptOne convenience in TERMIN 15 t h a t ~t is o f t m not necessary to type the whole of the rr5pollse to a prornpt In fact, all we need to type is a n unambiguous abbreviation, t huq:rot r for R.ctrl~ue Entry, for instance. If we type an ambiguous abbrevation, the system c a n help u s out: We can also be assisted when we make typ~ng mistakes:. I s l tAs it happens, users of the syqtern often rlnu rhis hind of help confusing inilially and so therc ir* a n O p t m t , callecl Hclpfiil, to control it. Initially this option is turned off, but i t c a n b e turned on simply by typ~ng:s t art helpfulTERMIN can often anticipate the answer to one of its questions. For instance, if we have jilst Rctrtrucd a term and then we issue an Edit command, the chances ere that the t r r n l we just retrieved is Itre term we want to edil. The Defaulting Option makes similar assirmptions To use thifiption, we type:>start daf aultlngand ttien we can utilize it: (Note that we have accepted the cicfault by simply striking the <RETURN> key.)-When a_a:?lpp, terms, we need not specify accents (unless they distinguish between two terms) * Th15 i q a cot~vcnirnce to be used when accessing-and fheir coritcntr;: when we type the tcrt in a fiche we must use the correct cases and accents. what W P t~a v c c_alled "r,eri.,cc, al~ovc arP rcpreqented by their termsnames only both in the figurer, and in Ihr t r w l ; this r l~v i c c is u~d merely for clarity of exposltian. 3 timrs a% many In t t~c worst carac Thts worst carqc occurs when equivalents are prc>scnt i n thc dict~onary for every language, which may not be so i n practice, cspccially while a dict~onary is being cornpllcd In the most favorable! case, !J=2 and DtR tin., tllc advantage by a factor of 2. Dlclionaries prepared for American use may o f t r n t,c Fnglisli'*X, so that N-? and DER has a space advsdage as well as the time advant ,~g c dcmonstr ated above. | The prima y h o t i v a t i o n for the iriterilctive dictionary i s that a technical translator may sperid up to 60% of his or her time simply looking up terms. This may include unsi~crcssful searches in several tfictionaries, partly because these dictionaries are out of date b y ttie time ttiey are publisli~d. An interactive computerized dictionary would provide effectively '%mmediatcw access to entries and moreover could be kept constantly updated at the central computing' facll~ty.Currctitly the diclionary contalris spccial~zcd terminology in English, French and German in a number of fields S p e c~a l~n c ! terminology was chosen because this is o f t e n most helpful in practice to the professional translator and also because this is w l~e r t t the benefits of st ar~dardization could be most immediately apparent The l a n g~i a~c r ,were chosen because they are the most immediately useful in the Local en\lironnicnt, as were the fielcls (mainly finance, business and iron, steel and mining fechnoloey).The n e x t section shows in some deta~l how a translalor would access the dictionary and ck~termine a correct eqirivalcnt The section after that describes the facilities used t o maititain and augment ttic d~ctionary The interface to the dictionary described in the nc.xt two sections r e p r r~~e n t s the fruits of continual close cooperatioh over a n extenclcd period of trme brtween researchers from the Computer Science Department and f r o m t h e Department of Modrr6n Languages Such coopc"ration, while it presents many problems initially, is a sine qttn nor1 of success in a venture such as Target.Wliile perfornilng initial sttrd~er, for the represcntat~on of equivalence between terms tho most central relation in a rnt~ltiline,ual dictionary -we have departed from the common practice of using a n alingcral set of concepts realized differently in different 1angt~ap.c~ Close examination q t i o w~d tlint 'ltii5 could not accommodate some nuances of meariiriy, in disparate langilagcs ancf warb* not precise enough, for making inferences wlicnl a particular equivaler~ce was not already present in the dictionary. Moreover, it was found to be less efficient than another method which was investigated and ult inlatcly adopted. Some arg~rrncnts proposed for adopting this different method are set f o r t l i In the final section of this document 2. The.---TARGET ------Program ---- T A R M T -------is also the name of tlie program used by translators to access the entries in the dic tionary while doing their tr arirlat ~o n work This section describes how it is used. Thc ilttlstrations are exact traces of the interaction between the program (in a roman font) atld the translator (in an italic font).We are first asked for the term names and the languages we wish to translate From (6.e. the source languap,e) and To (i.e, the target language):To Language! f r Now, i f there is only one equivalent for that term between those languages, we shall get that equivalent directly. In this case M e have a choice to make:Term; hond From Lanqtragei enTo Langiraget fr bond C h e m i s t r y i Theore t lca l C h~m l s t l y j (ch4) The N~r c l e a r I t i r l~c s t r y r Nuclear. Enet.qql (at61 F i n a n c i a l F l f f a i r s -T a r a t l o 1 1 -C u s t o m s~ ( 1S e l e c t CodeiLet us say the article we are trarislalitlg is in Chemistry. Then we just rype rne appropiate code. TIIC~C are in parcnthcscs in the example and are the same codes as used in the EEC's Euradicautom sy5tem. Here we select a code:Tat m t bondFrom L s n q u~q e ! en To Langtraqst Ir bond Chemls t r y ! Thooret l c a l Chamistryi (ch4) The Nuclear I n d u s t r y t Ntrclear Energy! ( a t6) F i n a n c i a l l l l f a l r s -Taxation -Customsj ( f l ) S e l e c t Cods: ch4and we shall get the appropriate fiche:Select Coder ch4 bond l i a i s o n (FR)Chemis tt-y: Theoret l c a l Chemistry; Tlie Nuc l enr Indits t r y ! Nt~c l e a r Energy! R e f e r e n c e Terms! bond l ng enerqy,We lia\le been told that the -,amp cql~ivalcnt i s used for both chemical, and nuclear bonding Had ttierc been f u r t l~b r information, such as a usage sample, a definition or a note, w e would have been askcd w l~e t l~c r we wanted to see it with the question More?Answering yes would show t h~ information to us.After this first use, Target arlsumcs ttfat we are translating from English to French Notice t tiat it does not ask us t lie From arid To qtrestions: We can override the assumption by lyp~np. all on onp lrne fhe term name, the sourr* language and the rarget language. t4cre we check on the equivalent just obtained: Term i bond1Irroc~prctivr of these con5itl~raltnnc, a dlrttoriary must remain functional while it is irrtorI~plvtr. To be reaiistic, i t i~ pt ol~ably ur~rommon for a dictionary to be "finished", anti all ailtomated rlicllotiarir~ rnl~~.t t~c butlt incrementally, equivalent by equivalent. Figure 2 be obtarrlcd A tempting, but incorrect, solut~on to this problem for ICSR is to assunle that artb~-Scharlfel, producing Figure 7 whe(her or not it is actually appropriate. This is a kind of risky and i~ncontrollod i n f r r~t~c e wl*ich ICSR mn naturaiiy force upon the user. T h c r c rnay be subtle d~f f c r c n r r r it] nlrat~ing bct w~e n languages, yet ICSR forces tratir1i!ivity of t rrlation of crj(~ivr71~n~e betwren all langaugec,. There is an altcrn;ltlve approach within ICSR, in which t h~s is not the assumption, but this will l e a d to pre( iqrly the p~o l i f c r n t i o n of r~rnr,rs which ICSR was designed to avoid. Fur! t i r r rnore, t tse simplif iciit lor1 of int crmccliate concepts which are found to be recloricJ,rt\t will be a cornpl~catccl prorcdtrre. ( l a 5 p r e v~o~~~l y been entcrcd explicitly. However, q i t t r a t i o t~~ will ocr\rr w l l r r c an inwnc\rfi;ltc anqwcr is not available In that case, forrri of L I L fcr wzrirtg may hrlp. W~t h ICSI?, that infcrcr~tinp, has already. been done in t i t~p tt10 i r~t~r r n~d~a t c r .~t~r~~ /ly trwan~. of ~IIC a~s i~t n p t~o n above, and Hhi~s the rtlforrt?al Ion that it is an infcrc3t1cr i c, l o~~t at rrtrlcval time. W~t l i DER, the pointers must b r followcd ! l i t ough r x p l~c~t l y c7r1cl tlltlcr thr s~T~~P I )~ car1 rcport to the user tke extent of tlrt\ tc.ntatrvenc*.ci of tllc c l~r i v~c l ncar ~q~l i v a l c n t I lie cl~c,atl\/at~taf:r, fur tllr! lntertnr\c!i;~le C o n r~p t Space r?epresentation, then, is that I t t~c orlc hand fincl~ng a n cqi~ivalcnt always t akcs two pointers, while Direct Fsl li\/alct\t R~p r e .~c n t a t~o n n~c.ck. ordy one, and on the other -more importantly --DFH i' nlare able to r c p r c s r~~t nuatlcrs of mcanlrle, acror3s languages and incomplete qtates of the dtct~onary FIer>ce Target uses the Dlrect Equivalence Representation for term eqi17valcnce.> I Thls shows that commonts gnt Into the record | null | Main paper:
the --:
----TERMlN ---------+-Program -- TERMIN --------isthe narne of the procram used by tcr ntinologists to augment and maintain the di'r t t o n a r~e s . Clearly the f a c l l i t~e s in TERMJN must be more varied than the simple This provicfes on-line access to written comnients on various aspects of the use of the TFRMIN program by terminologists All of the twenty or so tekts which may be acccsscd in this way were w r i t t c r~ h y the termtnologists (who also ordered the list of cornmat~ds as above and provided the brief description beside each one).( t o ex'ecuto a command f i l o ) ( t o s t a r t u s i n g an U p t i o n ) ( t o s t o p u s i n g s n O p t i o n )He lp a n t altorlng A L f E R I N G i s a u a y t o c o t ' r e c t e r r o r s u i t l~o~t t y p i n g t h e whole 1 i n e a g a i n .You c a n use I t hy u s i Most of thc tents pro\~ictccl in t t~~s way glvc hints and reminders i n an informal manner, rat her than a detailed seqiience of it~r~tructions on how 'to use the command, Tllic, command is i~c d to crcatr tirw rntrics in the dictionary for terms which have rjol. yrfQbeen entered Let us quppor.c llrat we have prepared a fiche specifying the e q u i v a l~n t in French of the Encl~qh t~r m b o t~d appropriate for the field of Chemistry. Space coristra~nts prevent a dct ailccl cic~cription of the interactions lead~ng to the entry of this term into the dictionary, but the follnw~ng is a trace of the process: his is how we leave TERMIN:T OX/^ E X I T~n e r r w r IS now no longer using the TERMIN program, but is using the TOPS-10 monitor.Tlds command is used to print wliole entries at our terminal. Here is bond: S~~p p o r~e that further investigation of the term bond has r~v c a l e c f that the same Frcnch cak~~valcnt 15 also used for the "bond" holding the nirclcirs of atoms together as for the clectronlc bond which keeps different atoms together. Had this been known when the orig~nal fiche was created, i t would have contained two field codes, cli4 and at8, We need to update the original entry.The term bortd is also irsed in financial and commercial circles This gives an alphabetrcal l~s t of terms in a specifled language. We can limit the list by spccifiying the f~r s t one or two letters of the f~r s t and last term. It may be aborted by typing m at any time, and will return us to the command level immediately. This command i s *used to re-c:.tablish the links between the various files which coot a~n t t~e dictiot~flry. They can hccorne incorrect when the computer crashes while c e r t a i n operations are being pcrformcd, or when there are problems w i t h the system.W t~r t n a term i~, stored after tlav~nc been Edtted, a new entry is made for the new version of the term Tlic old entry IS, however, still there, and hence takes up space.The qatne is t r u e in t l i~ CWIP of the D r l r t~ command. When a term is delefed, it actually otdy t~ccotncs ~n a r r c s~~i b l c ! -arid r,a ~t 1s taking ilp spatp. Every so often this "wasted" space (wl~ic h actually provides Ihe potential for some backup). is recovered using this command; i t compacts the dictioriaryAs c a n be secn from the forcegoing, to use a command we type i t s name (e.g. help) . llpppr r a s e anrl lowcr c a w a r~ ~q~~a l l y ac~~ptaI>le. The program will ttien begin to prompt 11s for the further spccificationr; necccssary to carry out the command. We may T y p r /Ihcnd the rcc,ponseq to thc~,e qtrcstions, in which case the prompt Is not given Help is u~u a l l y obta~nahle by typtng ? and any command can be aborted by typing @ it, re.,ponc,e to any prornptOne convenience in TERMIN 15 t h a t ~t is o f t m not necessary to type the whole of the rr5pollse to a prornpt In fact, all we need to type is a n unambiguous abbreviation, t huq:rot r for R.ctrl~ue Entry, for instance. If we type an ambiguous abbrevation, the system c a n help u s out: We can also be assisted when we make typ~ng mistakes:. I s l tAs it happens, users of the syqtern often rlnu rhis hind of help confusing inilially and so therc ir* a n O p t m t , callecl Hclpfiil, to control it. Initially this option is turned off, but i t c a n b e turned on simply by typ~ng:s t art helpfulTERMIN can often anticipate the answer to one of its questions. For instance, if we have jilst Rctrtrucd a term and then we issue an Edit command, the chances ere that the t r r n l we just retrieved is Itre term we want to edil. The Defaulting Option makes similar assirmptions To use thifiption, we type:>start daf aultlngand ttien we can utilize it: (Note that we have accepted the cicfault by simply striking the <RETURN> key.)-When a_a:?lpp, terms, we need not specify accents (unless they distinguish between two terms) * Th15 i q a cot~vcnirnce to be used when accessing-and fheir coritcntr;: when we type the tcrt in a fiche we must use the correct cases and accents. what W P t~a v c c_alled "r,eri.,cc, al~ovc arP rcpreqented by their termsnames only both in the figurer, and in Ihr t r w l ; this r l~v i c c is u~d merely for clarity of exposltian. 3 timrs a% many In t t~c worst carac Thts worst carqc occurs when equivalents are prc>scnt i n thc dict~onary for every language, which may not be so i n practice, cspccially while a dict~onary is being cornpllcd In the most favorable! case, !J=2 and DtR tin., tllc advantage by a factor of 2. Dlclionaries prepared for American use may o f t r n t,c Fnglisli'*X, so that N-? and DER has a space advsdage as well as the time advant ,~g c dcmonstr ated above.
modifiability:
Irroc~prctivr of these con5itl~raltnnc, a dlrttoriary must remain functional while it is irrtorI~plvtr. To be reaiistic, i t i~ pt ol~ably ur~rommon for a dictionary to be "finished", anti all ailtomated rlicllotiarir~ rnl~~.t t~c butlt incrementally, equivalent by equivalent. Figure 2 be obtarrlcd A tempting, but incorrect, solut~on to this problem for ICSR is to assunle that artb~-Scharlfel, producing Figure 7 whe(her or not it is actually appropriate. This is a kind of risky and i~ncontrollod i n f r r~t~c e wl*ich ICSR mn naturaiiy force upon the user. T h c r c rnay be subtle d~f f c r c n r r r it] nlrat~ing bct w~e n languages, yet ICSR forces tratir1i!ivity of t rrlation of crj(~ivr71~n~e betwren all langaugec,. There is an altcrn;ltlve approach within ICSR, in which t h~s is not the assumption, but this will l e a d to pre( iqrly the p~o l i f c r n t i o n of r~rnr,rs which ICSR was designed to avoid. Fur! t i r r rnore, t tse simplif iciit lor1 of int crmccliate concepts which are found to be recloricJ,rt\t will be a cornpl~catccl prorcdtrre. ( l a 5 p r e v~o~~~l y been entcrcd explicitly. However, q i t t r a t i o t~~ will ocr\rr w l l r r c an inwnc\rfi;ltc anqwcr is not available In that case, forrri of L I L fcr wzrirtg may hrlp. W~t h ICSI?, that infcrcr~tinp, has already. been done in t i t~p tt10 i r~t~r r n~d~a t c r .~t~r~~ /ly trwan~. of ~IIC a~s i~t n p t~o n above, and Hhi~s the rtlforrt?al Ion that it is an infcrc3t1cr i c, l o~~t at rrtrlcval time. W~t l i DER, the pointers must b r followcd ! l i t ough r x p l~c~t l y c7r1cl tlltlcr thr s~T~~P I )~ car1 rcport to the user tke extent of tlrt\ tc.ntatrvenc*.ci of tllc c l~r i v~c l ncar ~q~l i v a l c n t I lie cl~c,atl\/at~taf:r, fur tllr! lntertnr\c!i;~le C o n r~p t Space r?epresentation, then, is that I t t~c orlc hand fincl~ng a n cqi~ivalcnt always t akcs two pointers, while Direct Fsl li\/alct\t R~p r e .~c n t a t~o n n~c.ck. ordy one, and on the other -more importantly --DFH i' nlare able to r c p r c s r~~t nuatlcrs of mcanlrle, acror3s languages and incomplete qtates of the dtct~onary FIer>ce Target uses the Dlrect Equivalence Representation for term eqi17valcnce.> I Thls shows that commonts gnt Into the record
:
The prima y h o t i v a t i o n for the iriterilctive dictionary i s that a technical translator may sperid up to 60% of his or her time simply looking up terms. This may include unsi~crcssful searches in several tfictionaries, partly because these dictionaries are out of date b y ttie time ttiey are publisli~d. An interactive computerized dictionary would provide effectively '%mmediatcw access to entries and moreover could be kept constantly updated at the central computing' facll~ty.Currctitly the diclionary contalris spccial~zcd terminology in English, French and German in a number of fields S p e c~a l~n c ! terminology was chosen because this is o f t e n most helpful in practice to the professional translator and also because this is w l~e r t t the benefits of st ar~dardization could be most immediately apparent The l a n g~i a~c r ,were chosen because they are the most immediately useful in the Local en\lironnicnt, as were the fielcls (mainly finance, business and iron, steel and mining fechnoloey).The n e x t section shows in some deta~l how a translalor would access the dictionary and ck~termine a correct eqirivalcnt The section after that describes the facilities used t o maititain and augment ttic d~ctionary The interface to the dictionary described in the nc.xt two sections r e p r r~~e n t s the fruits of continual close cooperatioh over a n extenclcd period of trme brtween researchers from the Computer Science Department and f r o m t h e Department of Modrr6n Languages Such coopc"ration, while it presents many problems initially, is a sine qttn nor1 of success in a venture such as Target.Wliile perfornilng initial sttrd~er, for the represcntat~on of equivalence between terms tho most central relation in a rnt~ltiline,ual dictionary -we have departed from the common practice of using a n alingcral set of concepts realized differently in different 1angt~ap.c~ Close examination q t i o w~d tlint 'ltii5 could not accommodate some nuances of meariiriy, in disparate langilagcs ancf warb* not precise enough, for making inferences wlicnl a particular equivaler~ce was not already present in the dictionary. Moreover, it was found to be less efficient than another method which was investigated and ult inlatcly adopted. Some arg~rrncnts proposed for adopting this different method are set f o r t l i In the final section of this document 2. The.---TARGET ------Program ---- T A R M T -------is also the name of tlie program used by translators to access the entries in the dic tionary while doing their tr arirlat ~o n work This section describes how it is used. Thc ilttlstrations are exact traces of the interaction between the program (in a roman font) atld the translator (in an italic font).We are first asked for the term names and the languages we wish to translate From (6.e. the source languap,e) and To (i.e, the target language):To Language! f r Now, i f there is only one equivalent for that term between those languages, we shall get that equivalent directly. In this case M e have a choice to make:Term; hond From Lanqtragei enTo Langiraget fr bond C h e m i s t r y i Theore t lca l C h~m l s t l y j (ch4) The N~r c l e a r I t i r l~c s t r y r Nuclear. Enet.qql (at61 F i n a n c i a l F l f f a i r s -T a r a t l o 1 1 -C u s t o m s~ ( 1S e l e c t CodeiLet us say the article we are trarislalitlg is in Chemistry. Then we just rype rne appropiate code. TIIC~C are in parcnthcscs in the example and are the same codes as used in the EEC's Euradicautom sy5tem. Here we select a code:Tat m t bondFrom L s n q u~q e ! en To Langtraqst Ir bond Chemls t r y ! Thooret l c a l Chamistryi (ch4) The Nuclear I n d u s t r y t Ntrclear Energy! ( a t6) F i n a n c i a l l l l f a l r s -Taxation -Customsj ( f l ) S e l e c t Cods: ch4and we shall get the appropriate fiche:Select Coder ch4 bond l i a i s o n (FR)Chemis tt-y: Theoret l c a l Chemistry; Tlie Nuc l enr Indits t r y ! Nt~c l e a r Energy! R e f e r e n c e Terms! bond l ng enerqy,We lia\le been told that the -,amp cql~ivalcnt i s used for both chemical, and nuclear bonding Had ttierc been f u r t l~b r information, such as a usage sample, a definition or a note, w e would have been askcd w l~e t l~c r we wanted to see it with the question More?Answering yes would show t h~ information to us.After this first use, Target arlsumcs ttfat we are translating from English to French Notice t tiat it does not ask us t lie From arid To qtrestions: We can override the assumption by lyp~np. all on onp lrne fhe term name, the sourr* language and the rarget language. t4cre we check on the equivalent just obtained: Term i bond1
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 560 | 0 | null | null | null | null | null | null | null | null |
93896668783ed2b676b67223ac0e4a48910b24bd | 219308139 | null | The Derivation of Answers from Logical Forms in a Question Answering System | This papex descrsbes how the process 05 g e n e ~t i n g a response g i v e n an underlying representation f o x an i n p u t q u e s t i o n is accomplished in t h e Transformatioaal Question Rnswering [ P A ? s y s t e m under development a t IBM Research, a b r i e f description a f which is g i v e n . The last formal level o f representation in this system is called a l o g i c a l form. The b a s ~c method of evaluation of logical forms is the generate and t e s t " paradigm, used, f o r q ~& n p l s in the LUNAR system (Woods, Kaplan and Nash-Webber, 1 9 7 2 1, a l t h b u g h t h a t implementation must be fairly efficient t h a t , we s t i l L have a considerable way t o go, ~h i r p a p e r d e s c r i b e s how t h o p r o c e s s o f g e n e r a t i n g a r a s p o n s e g i v e n a n t l ~l d e ~l y i n g xrsprg!senT:amn fair a n i 1 1 p u t q u e s t i o n i s ~c c c m ~l i s h e d A n t h e ~r a n s f o r p n t i o n n l E u c s t ~o n A n s w ~r i h g ( T P ) 1 s y s trrp u n d o r' c o ~~t i n u ~n g dr,u@kopmmc?nt a t I A N R e s e a r c h , . TQA h a s beert, o p e r a t i o n a l . 1 a l a b o r n t s z y m o d e f o r s e v e r a l yeers.. T h e s y s t e m is noid i n s t a l l e d i n t h e o f f i c e o f the p l a n n i n g d e p a r t m e n t o l a s m a l l c i t y u h e r e it is u s e d t o a c c e s s t h e f i l e o f l a n d u s e f o x e a c h p a r c e l o f l a n d I n t h e c i t y , ( a b o u t 1 0 , 0 0 0 p a r c e l s ~i t h 4 0 p i e c a s o f d a t a f o r e a c h p a r c e l 1 . The sysytcm is trnlilcrgoin,g r n ~d i f i c a t i o n s nncl i n ~p x o v s n e n t p x i s x t o a f o r m a l e v a 1 u a t i 0 1 1 s t a g e I A g e n e r a l i z e d f l o w d i a g r a m o f t h & TQA s y s t e m is g i v e n in F i g u r e 1 . | {
"name": [
"Damerau, Fred J."
],
"affiliation": [
null
]
} | null | null | null | 1978-06-01 | 1 | 9 | null | null | The last formal level o f representation in this system is called a l o g i c a l form. The b a s~c method of evaluation of logical forms is the generate and t e s t " paradigm, used, f o r q~& n p l s in the LUNAR system (Woods, Kaplan and Nash-Webber, 1 9 7 2 1, a l t h b u g h t h a t implementation must be fairly efficient in order t e be j~a c t i c a l on a moderate size d a t a base. The 1 Semant~c, interpreter I <-----Sernhntic r u l e s ( f a r a l l ' X I 1 5-----------------,,J <--I 7 1 I L o g i c a l form(s) 1 I f ---------1 I IEvaluatosl <------------ Data( s e a t % ' X 3 8 ' ( q u a n t i t y( s e t x 'X34 '(and ( t e s t i c t '9v ( T~~~~~ X 3 4 ' 1 9 7 6 ) ' = ) ( p a r c e l X 3 4 ) 1 1 ( g r e a t e x t h a n XI15 ?25) 1 t h a t , we s t i l L have a considerable way t o go,p t o g r s m as w e l l as in the L I S P d e f i n i t i o n 3 f the r u n c t j o na b j e c t i n the answer s e t , t h i s p r o b l e m naturally does not | null | null | null | Main paper:
:
The last formal level o f representation in this system is called a l o g i c a l form. The b a s~c method of evaluation of logical forms is the generate and t e s t " paradigm, used, f o r q~& n p l s in the LUNAR system (Woods, Kaplan and Nash-Webber, 1 9 7 2 1, a l t h b u g h t h a t implementation must be fairly efficient in order t e be j~a c t i c a l on a moderate size d a t a base. The 1 Semant~c, interpreter I <-----Sernhntic r u l e s ( f a r a l l ' X I 1 5-----------------,,J <--I 7 1 I L o g i c a l form(s) 1 I f ---------1 I IEvaluatosl <------------ Data( s e a t % ' X 3 8 ' ( q u a n t i t y( s e t x 'X34 '(and ( t e s t i c t '9v ( T~~~~~ X 3 4 ' 1 9 7 6 ) ' = ) ( p a r c e l X 3 4 ) 1 1 ( g r e a t e x t h a n XI15 ?25) 1 t h a t , we s t i l L have a considerable way t o go,p t o g r s m as w e l l as in the L I S P d e f i n i t i o n 3 f the r u n c t j o na b j e c t i n the answer s e t , t h i s p r o b l e m naturally does not
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 560 | 0.016071 | null | null | null | null | null | null | null | null |
f31d554e3405f3a0377f99359bf29d89c0473a62 | 219301596 | null | One More Step Toward Computer Lexicometry | We describe the continuation of an earlier b70rk on the prbblm of lexical coverage. The objective is to prove experimentally certain mathematical conjectures concerning t h e relationshi? between the sizes of the covering and covered sets of words, an&--maximun lenqth of dictionary d e f i n i t i o n s . The data base on which the experiments are cerried sut bas been also extended t6 the full contents -of an existinq dictionary of computer terminology. The rwults of the previous and present work lay the foundqtions for quantitative studies on lexical valence and its relation to the frequency of usage and other p r i n c i p l e s ofb ditztionary selection. Besides t h e inherent interest in t-hese investigations , the concepts dealt with and the methods of cgantifying dictionary variables may eventually lead to more' efficient dictionaries with respect to precision, compactness, and computer time andmemory needed for processing. | {
"name": [
"Findler, Nicholas V. and",
"Lee, Shu-Hwa"
],
"affiliation": [
null,
null
]
} | null | null | null | 1978-06-01 | 0 | 0 | null | null | In order to approach the problem in definite terns, Findler( 1 970) considered three basic variables : (a4) The increme& ratio never exceeds the size ratio.t i o n co rules ( a l ) and (a21 would occur in a dictionary system, whi'ch does not treat polysemous words or homonyms as i n d i v i d u a l entries, every t i m e a new word with many meanings or homonyms~ i.s introduced into the covered set, Second, the cited case is an exception to ruie ( a l ) but not to (a4) . When N=1, the covering and the covered s e t s are of the same s i z e , i . e , botn the increment ra'tio and the size batio equal one, However, not every word is defined By i t s e l f only. If a new word is introquced that. al-ready has a synbonm in the coverihg s e t , it will be d e f i n e d by that synonym. In t h i s caser the increment r a t i o is 0 and the size ratio becomes less than 1 . (This will be clear w i t h the descript i o n of the data base construction on page 11.) For the seoond general task, (b) t h e followinq conjectures were also mads: Table I . ) code 2 i n d i c a t e s that t h e entry is deiiined by iteelf, i . e . it belongs to t H e basic vocabulary. | null | null | null | Main paper:
:
In order to approach the problem in definite terns, Findler( 1 970) considered three basic variables : (a4) The increme& ratio never exceeds the size ratio.t i o n co rules ( a l ) and (a21 would occur in a dictionary system, whi'ch does not treat polysemous words or homonyms as i n d i v i d u a l entries, every t i m e a new word with many meanings or homonyms~ i.s introduced into the covered set, Second, the cited case is an exception to ruie ( a l ) but not to (a4) . When N=1, the covering and the covered s e t s are of the same s i z e , i . e , botn the increment ra'tio and the size batio equal one, However, not every word is defined By i t s e l f only. If a new word is introquced that. al-ready has a synbonm in the coverihg s e t , it will be d e f i n e d by that synonym. In t h i s caser the increment r a t i o is 0 and the size ratio becomes less than 1 . (This will be clear w i t h the descript i o n of the data base construction on page 11.) For the seoond general task, (b) t h e followinq conjectures were also mads: Table I . ) code 2 i n d i c a t e s that t h e entry is deiiined by iteelf, i . e . it belongs to t H e basic vocabulary.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 560 | 0 | null | null | null | null | null | null | null | null |
70af1eba98f23d67240de2f86c0e61d1e003dbd1 | 111078716 | null | Error analysis of Systran output - a suggested criterion for the {`}internal{'} evaluation of translation quality and a possible corrective for system design | in Birmingham. ' "The only way!" It is my personal conviction that the machine translation system developed under your guidance and improved under government contract represents the only viable translation option available to the nation for servicing bulk translation requirements in the interests of national defense.' 'We feel that computerised machine translation (SIC) is not only feasible and economical but the only way to provide bilingual technical manuals. It provides us with a text which is completely acceptable for use by technicians.' '...present users include the government of the USA, the government of Canada, the European Economic Community, and a range of industrial companies including some of North America's largest multinational corporations.' The above statements are excerpts -slightly adapted, but not substantially altered -from SYSTRAN sales literature distributed by WTC of Canada Ltd.; the first statement comes from the United States Air Force, the second from General Motors of Canada Ltd., and the third is WTC advertising copy. Against this background -indeed in spite of it -I should like to give some details of an "error analysis" which I have recently conducted on Russian texts translated into English by the SYSTRAN system. As I have said in a recent article, I cannot prove statistically that I was dealing, with average texts or indeed with an average machine translation system. 1 I will repeat myself by saying that I hope you will not be too scandalised if I say that I am intuitively satisfied that the texts I dealt with were not out of the ordinary and that the machine translation system which was responsible for the translation of those texts deserves attention firstly because it is one of the very few commercially operational machine translation systems in existence and secondly because the E.E.C. has made a considerable financial investment in it and is continuing to optimise the SYSTRAN system until such time as the proposed new EUROTRA system comes on line. The particular texts that I analysed were made available to me by Professor F. Krückeberg and Mr D. Hoppe of the "Gesellschaft für Mathematik und Datenverarbeitung" in Bonn, Germany, and I should like to record my considerable thanks to these gentlemen for their generous assistance. This institution carried out during 1976, on the basis of a licenced agreement with the WTC of La Jolla (California),U.S.A., a number of tests in order to assess the performance and general capabilities of the 109 | {
"name": [
"Knowles, F."
],
"affiliation": [
null
]
} | null | null | Translating and the Computer | 1978-11-01 | 16 | 7 | null | WTC-developed SYSTRAN machine translation system; these tests were carried out on the SYSTRAN software dealing with translation from Russian into English. My remarks in this paper are confined to SYSTRAN'S performance with this particular language pair. The corpus of sentences which I used in my researches was made up of two distinct categories of language data. On the other hand I investigated SYSTRAN'S performance at translating from Russian into English just over 2,000 sentences which had been used as examples in a pedagogical grammar of Russian written for German students. 2 Secondly, I had available four Russian technical texts containing subject matter on i) scales and weighing, ii) airports, iii) helicopters and iv) eyesight. The total number of sentences involved in this textual corpus was slightly under 500. These texts were part of a sample of texts chosen from the Great Soviet Encyclopaedia ("Bol'šaja Sovetskaja Ènciklopedija"). It was felt by the GMD experts who initiated the original SYSTRAN tests that this source of textual material offered two advantages in a test situation. Firstly, the texts represent technical material of sufficient but not excessive terminological difficulty and they are, furthermore, adequately representative in so far as their grammatical complexities are concerned. Secondly, the original texts were carefully edited and have also been made available to readers of English by professional translators. I present in Figure 1 what I have called 'raw' corpus characteristics. In Figure 2 on the other hand I show what I have chosen to call 'edited' corpus characteristics. The discrepancies in the number of tokens in the corpus are the result of editing procedures designed to eliminate number strings and to expand abbreviations. I generated from the 'edited' corpus alphabetic, reverse-alphabetic and descending frequency lists of the lexis occurring in the four technical texts. I also produced concordances to them and used all these materials to aid my error analysis. I have also included in Figure 2 three statistical indices common in the world of statistical linguistics, namely the logarithmic type-token ratio, the logarithmic lemma-type ratio and the index of vocabulary richness.I carried out an analysis of variance test on the homogeneity of the mean sentence length across the four technical texts. The resulting F value was 1.48 and it is less than the tabular value of 2.60 for the 0.05% confidence level with the given degrees of freedom. On this basis I 'pooled' these thematically differing texts to form one 'technical text' corpus, statistically homogeneous at least from the point of view of sentence length and, in my subjective opinion, homogeneous in a number of other respects as well. By comparison an F test comparing the 'grammar' corpus with the 'technical text' corpus yielded the highly significant ratio of 22.71. I therefore draw a distinction between the two corpora in what follows.My strategy in this particular piece of research was to scan all the sentences at my disposal, simultaneously noting and aggregating errors under various categories. The next step was to try to correlate the errors which were evident with the sequence of events in the processing of texts by the SYSTRAN MT system. I present in Figure 3 a simplified so-called 'sequence of events' on the basis of what I have been able to glean by reading published literature relating to the SYSTRAN software 3-8 and by discussing these matters with colleagues who have had the opportunity to probe more deeply into these matters or whose actual job it is to run SYSTRAN jobs and service the SYSTRAN system. I append a further figure, Figure 4 , relating to SYSTRAN dictionary structure, given its crucial importance within the framework of the SYSTRAN software. In fact, the view is tenable that the success that SYSTRAN has achieved is due in larger measure to the size and comprehensiveness of its data-base rather than to the inherent power of its processing algorithms. I must emphasise that my 'results' are inferences. I have never seen -not for want of trying -the bank of SYSTRAN'S Russian dictionaries, neither have I ever seen the suite of SYSTRAN'S Russian-English analysis programs. I have seen -cursorily -SYSTRAN's English-French documentation but notwithstanding this I viewed SYSTRAN for the purposes of this investigation as a 'black box' for which I had input and output and about the inner workings of which I had but meagre information. I would claim, however, that this does enable me to make certain reasonable deductions about SYSTRAN's successes and, more importantly, its failures. One sideeffect of this is, incidentally, to highlight the urgent need to overcome barriers standing in the way of collaboration, barriers which effectively prevent, so it seems, new ideas arising without from penetrating inwards, so to speak.In Figure 5 I present my inventory of translation errors evident in the 'grammar' corpus, that is in the corpus of sentences excerpted from the above-mentioned pedagogical grammar. I must emphasise that the numbers and their accompanying percentages represent error tokens and not error types. In other words, they represent the total number of errors in that category. In Figure 6 I suggest some possible improvements to SYSTRAN software which in my view would have a pronounced enhancing effect; I classify this remark as a sort of 'long distance speculation'. I concede that the mechanisms required to implement suggestions of this sort have to be very finely balanced so as to lessen the risk of combinatorial explosion. Figure 7 presents a similar inventory of the errors which occurred in the 'technical text' corpus. I adhere to the same format as previously and likewise suggest in Figure 8 a small number of modifications which might eliminate, or at least drastically reduce the incidence of 'theoretically' avoidable errors. I am the first to admit, however, that there is what might be called a 'rump' involving in this case over 200 cruces which would be extremely difficult or costly, if not impossible, to solve programmatically. Investigation of these particular sentences by means of the hexadecimal print diagnostic facility would have helped enormously here but I did not have access to this information. I suspect, however, that the homograph disambiguation routines may be largely to blame. Figure 9 gives details of what I call sentence success rate and it is apparent from this that the success rate is indeed extremely low, with errors occurring every four or five words. Note, however, that this last statement is somewhat misleading because a number of the errors occur at what might be called the supra-word level. I give in Figure 10 a checklist of the problems SYSTRAN appears to be suffering from in the realm of Russian morphological analysis and phrase-structure handling.Most of SYSTRAN's errors catalogued above derive from a failure to implement functioning routines in a global and consistent fashion. It seems as if in many cases one salient or typical example has indeed been incorporated but that it stands alone like a prototype which never entered mass production. The answer to a given problem is often available yet the data needed by the problem-solving routines or by the dictionaries is either missing or is inadequate. The second record is SYSTRAN's attempted translation and this is followed by the third record which is a correct English translation of the sentence involved. The fourth record -present only in the examples drawn from the 'technical text' corpus -is a gloss on the mistakes highlighted. It will be obvious that the sentences quoted often also contain other errors which are not commented on.I take first errors encountered in the 'grammar' corpus; these remain without comment. I concede readily that the automatic translation of this material from Russian into English is a very tough test indeed because by definition the sentences involve the total range of, in this case Russian, grammar. A number of SYSTRAN's facilities such as the topical glossary system cannot be put to use in this case. Use might also be made,in dictionary refinement, of Zasorina's new and major Russian frequency dictionary. 9Turning next to the 'technical text' corpus and addressing myself in a sense to the 'real world' of the professional translator. I give (in Appendix II) a further selection of SYSTRAN-generated translation errors, accompanied by explanatory notes.Turning I am sure that we would all agree that we cannot rely on the reader's knowledge or on his good will to this extent. At one point the translation of the 'scales' text was stated by the expert reader to be seriously misleading and potentially dangerous if the reader should attempt to carry out one of the procedures in the way it is described by the 'machine' ! If this is so, then there ought to be no question of letting SYSTRAN loose, that is to say, letting it off its leash. Rather it appears necessary to have a human reviser holding the leash tightly. However, this of course knocks out one of the system's main pit-props and vitiates many of the claims made about SYSTRAN's translation performance and its throughput.I must now make a statement that might well appear paradoxical. I state that SYSTRAN does appear to have achieved a performance level which is better than any other MT system has attained, and that I therefore owe it my respect on this account. I cannot in fact conclude this paper without revealing my admiration for SYSTRAN's data-processing sophistication. I hazard the guess that SYSTRAN may be suffering because a lot of the linguistics 'know-how' in the system was put there by 'linguist-programmers' who are now at one remove, having been replaced by 'systems programmers'. I believe, as I said above, that a good many -but not all -of the pieces in the jig-saw puzzle of the overall strategy for computerised language analysis are already in their correct places.In summary, what I have been trying to say in this paper is that SYSTRAN's best efforts are being in part frustrated, firstly, by deficient language data in its data-base and secondly, by the fact that some areas of potentially crucial, or -at the very least -promising 'know-how' in semantics have not found their way into SYSTRAN's 'architecture', or have not made their presence felt. I refer to 'tools' such as statistically weighted sub-language glossaries 12 , thesaurus methods for disambiguation 13 or lexeme coding techniques 14 , for instance .I should like to thank Margaret Masterman for discussing with me many of the issues touched upon in this paper. Her comments were always both willingly given and illuminating and I am indebted to her; I accept responsibility, of course, for all the shortcomings of this paper which I hope, nonetheless, may be of some interest and use. An airport has mechanization and transportation bases; technical and other warehouses; and a variety of service buildings, engineering networks, and facilities for water supply, sewerage, heat, gas, and electric power.* False information resident in dictionary. U presmykah5ixs4, isklhca4 gatterih i cerepax, ptiq, isklhca4 kivi, v steklovidnoe telo. vda2ts4 ot mesta vxoda zritel6nogo nerva xarakterny vyrost, obil6no snabj2nny1 sosudami, -greben6.In reptiles, except for the tuatara and tortoises, and in birds, except for the kiwi, a characteristic process abundantly supplied with blood vessels -the pecten -protrudes into the vitreous body from the point of entry of the optic nerve. * The technical term for 'reptiles' not in the dictionary. The computer has parsed the participial form of the appropriate Russian verb for which the dictionary contains incorrect information. The verb involved now means not 'to creep/crawl' (the etymology of the word 'reptile'), but 'to grovel'. S rostom prot4j2nnosti vozduwnyx trass i osvoeniem novyx tipov samol2tov povywahts4 trebovani4 k oborudovanih i osna5enih a3roportov.As the length of air routes increases and as new types of aircraft are put into operation, the requirements for the equipment of airports increase. * The computer is 'unaware' that it has used the same English word twice as equivalents of two different Russian words.Sovremennyl a3roport predstavl4et sobo1 slojny1 kompleks injenernyx soorujeni1, texniceskix sredstv, dl4 razme5eni4 kotorogo trebuets4 territori4, izmer4ema4 v otdel6nyx sluca4x tys4cami gektarov (naprimer moskovski1 a3roport Domodedovo, H6h-1orkski1 a3roport Kennedi). Razrabotka arxitekturno-planirovocnyx sxem a3roportov predusmatrivaet naibolee raqional6noe socetanie zon-l2tno1, slujebno1 i jilo1; pri 3tom kompoziqionnym qentrom 4vl4ets4 a3rovokzal vmeste s drugimi ucastkami slujebno1 zony, neprosredstvenno sv4zannymi s obslujivaniem passajirov.The architectural and planning schemes of airports provide for the most rational combination of flight, service, and residential zones. The central element of its composition is the air terminal and other service zones directly connected with the servicing of passengers.* Grammatical class of word not recognised. False parse route leads to wrong dictionary entry. | null | null | null | null | Main paper:
:
WTC-developed SYSTRAN machine translation system; these tests were carried out on the SYSTRAN software dealing with translation from Russian into English. My remarks in this paper are confined to SYSTRAN'S performance with this particular language pair. The corpus of sentences which I used in my researches was made up of two distinct categories of language data. On the other hand I investigated SYSTRAN'S performance at translating from Russian into English just over 2,000 sentences which had been used as examples in a pedagogical grammar of Russian written for German students. 2 Secondly, I had available four Russian technical texts containing subject matter on i) scales and weighing, ii) airports, iii) helicopters and iv) eyesight. The total number of sentences involved in this textual corpus was slightly under 500. These texts were part of a sample of texts chosen from the Great Soviet Encyclopaedia ("Bol'šaja Sovetskaja Ènciklopedija"). It was felt by the GMD experts who initiated the original SYSTRAN tests that this source of textual material offered two advantages in a test situation. Firstly, the texts represent technical material of sufficient but not excessive terminological difficulty and they are, furthermore, adequately representative in so far as their grammatical complexities are concerned. Secondly, the original texts were carefully edited and have also been made available to readers of English by professional translators. I present in Figure 1 what I have called 'raw' corpus characteristics. In Figure 2 on the other hand I show what I have chosen to call 'edited' corpus characteristics. The discrepancies in the number of tokens in the corpus are the result of editing procedures designed to eliminate number strings and to expand abbreviations. I generated from the 'edited' corpus alphabetic, reverse-alphabetic and descending frequency lists of the lexis occurring in the four technical texts. I also produced concordances to them and used all these materials to aid my error analysis. I have also included in Figure 2 three statistical indices common in the world of statistical linguistics, namely the logarithmic type-token ratio, the logarithmic lemma-type ratio and the index of vocabulary richness.I carried out an analysis of variance test on the homogeneity of the mean sentence length across the four technical texts. The resulting F value was 1.48 and it is less than the tabular value of 2.60 for the 0.05% confidence level with the given degrees of freedom. On this basis I 'pooled' these thematically differing texts to form one 'technical text' corpus, statistically homogeneous at least from the point of view of sentence length and, in my subjective opinion, homogeneous in a number of other respects as well. By comparison an F test comparing the 'grammar' corpus with the 'technical text' corpus yielded the highly significant ratio of 22.71. I therefore draw a distinction between the two corpora in what follows.My strategy in this particular piece of research was to scan all the sentences at my disposal, simultaneously noting and aggregating errors under various categories. The next step was to try to correlate the errors which were evident with the sequence of events in the processing of texts by the SYSTRAN MT system. I present in Figure 3 a simplified so-called 'sequence of events' on the basis of what I have been able to glean by reading published literature relating to the SYSTRAN software 3-8 and by discussing these matters with colleagues who have had the opportunity to probe more deeply into these matters or whose actual job it is to run SYSTRAN jobs and service the SYSTRAN system. I append a further figure, Figure 4 , relating to SYSTRAN dictionary structure, given its crucial importance within the framework of the SYSTRAN software. In fact, the view is tenable that the success that SYSTRAN has achieved is due in larger measure to the size and comprehensiveness of its data-base rather than to the inherent power of its processing algorithms. I must emphasise that my 'results' are inferences. I have never seen -not for want of trying -the bank of SYSTRAN'S Russian dictionaries, neither have I ever seen the suite of SYSTRAN'S Russian-English analysis programs. I have seen -cursorily -SYSTRAN's English-French documentation but notwithstanding this I viewed SYSTRAN for the purposes of this investigation as a 'black box' for which I had input and output and about the inner workings of which I had but meagre information. I would claim, however, that this does enable me to make certain reasonable deductions about SYSTRAN's successes and, more importantly, its failures. One sideeffect of this is, incidentally, to highlight the urgent need to overcome barriers standing in the way of collaboration, barriers which effectively prevent, so it seems, new ideas arising without from penetrating inwards, so to speak.In Figure 5 I present my inventory of translation errors evident in the 'grammar' corpus, that is in the corpus of sentences excerpted from the above-mentioned pedagogical grammar. I must emphasise that the numbers and their accompanying percentages represent error tokens and not error types. In other words, they represent the total number of errors in that category. In Figure 6 I suggest some possible improvements to SYSTRAN software which in my view would have a pronounced enhancing effect; I classify this remark as a sort of 'long distance speculation'. I concede that the mechanisms required to implement suggestions of this sort have to be very finely balanced so as to lessen the risk of combinatorial explosion. Figure 7 presents a similar inventory of the errors which occurred in the 'technical text' corpus. I adhere to the same format as previously and likewise suggest in Figure 8 a small number of modifications which might eliminate, or at least drastically reduce the incidence of 'theoretically' avoidable errors. I am the first to admit, however, that there is what might be called a 'rump' involving in this case over 200 cruces which would be extremely difficult or costly, if not impossible, to solve programmatically. Investigation of these particular sentences by means of the hexadecimal print diagnostic facility would have helped enormously here but I did not have access to this information. I suspect, however, that the homograph disambiguation routines may be largely to blame. Figure 9 gives details of what I call sentence success rate and it is apparent from this that the success rate is indeed extremely low, with errors occurring every four or five words. Note, however, that this last statement is somewhat misleading because a number of the errors occur at what might be called the supra-word level. I give in Figure 10 a checklist of the problems SYSTRAN appears to be suffering from in the realm of Russian morphological analysis and phrase-structure handling.Most of SYSTRAN's errors catalogued above derive from a failure to implement functioning routines in a global and consistent fashion. It seems as if in many cases one salient or typical example has indeed been incorporated but that it stands alone like a prototype which never entered mass production. The answer to a given problem is often available yet the data needed by the problem-solving routines or by the dictionaries is either missing or is inadequate. The second record is SYSTRAN's attempted translation and this is followed by the third record which is a correct English translation of the sentence involved. The fourth record -present only in the examples drawn from the 'technical text' corpus -is a gloss on the mistakes highlighted. It will be obvious that the sentences quoted often also contain other errors which are not commented on.I take first errors encountered in the 'grammar' corpus; these remain without comment. I concede readily that the automatic translation of this material from Russian into English is a very tough test indeed because by definition the sentences involve the total range of, in this case Russian, grammar. A number of SYSTRAN's facilities such as the topical glossary system cannot be put to use in this case. Use might also be made,in dictionary refinement, of Zasorina's new and major Russian frequency dictionary. 9Turning next to the 'technical text' corpus and addressing myself in a sense to the 'real world' of the professional translator. I give (in Appendix II) a further selection of SYSTRAN-generated translation errors, accompanied by explanatory notes.Turning I am sure that we would all agree that we cannot rely on the reader's knowledge or on his good will to this extent. At one point the translation of the 'scales' text was stated by the expert reader to be seriously misleading and potentially dangerous if the reader should attempt to carry out one of the procedures in the way it is described by the 'machine' ! If this is so, then there ought to be no question of letting SYSTRAN loose, that is to say, letting it off its leash. Rather it appears necessary to have a human reviser holding the leash tightly. However, this of course knocks out one of the system's main pit-props and vitiates many of the claims made about SYSTRAN's translation performance and its throughput.I must now make a statement that might well appear paradoxical. I state that SYSTRAN does appear to have achieved a performance level which is better than any other MT system has attained, and that I therefore owe it my respect on this account. I cannot in fact conclude this paper without revealing my admiration for SYSTRAN's data-processing sophistication. I hazard the guess that SYSTRAN may be suffering because a lot of the linguistics 'know-how' in the system was put there by 'linguist-programmers' who are now at one remove, having been replaced by 'systems programmers'. I believe, as I said above, that a good many -but not all -of the pieces in the jig-saw puzzle of the overall strategy for computerised language analysis are already in their correct places.In summary, what I have been trying to say in this paper is that SYSTRAN's best efforts are being in part frustrated, firstly, by deficient language data in its data-base and secondly, by the fact that some areas of potentially crucial, or -at the very least -promising 'know-how' in semantics have not found their way into SYSTRAN's 'architecture', or have not made their presence felt. I refer to 'tools' such as statistically weighted sub-language glossaries 12 , thesaurus methods for disambiguation 13 or lexeme coding techniques 14 , for instance .I should like to thank Margaret Masterman for discussing with me many of the issues touched upon in this paper. Her comments were always both willingly given and illuminating and I am indebted to her; I accept responsibility, of course, for all the shortcomings of this paper which I hope, nonetheless, may be of some interest and use. An airport has mechanization and transportation bases; technical and other warehouses; and a variety of service buildings, engineering networks, and facilities for water supply, sewerage, heat, gas, and electric power.* False information resident in dictionary. U presmykah5ixs4, isklhca4 gatterih i cerepax, ptiq, isklhca4 kivi, v steklovidnoe telo. vda2ts4 ot mesta vxoda zritel6nogo nerva xarakterny vyrost, obil6no snabj2nny1 sosudami, -greben6.In reptiles, except for the tuatara and tortoises, and in birds, except for the kiwi, a characteristic process abundantly supplied with blood vessels -the pecten -protrudes into the vitreous body from the point of entry of the optic nerve. * The technical term for 'reptiles' not in the dictionary. The computer has parsed the participial form of the appropriate Russian verb for which the dictionary contains incorrect information. The verb involved now means not 'to creep/crawl' (the etymology of the word 'reptile'), but 'to grovel'. S rostom prot4j2nnosti vozduwnyx trass i osvoeniem novyx tipov samol2tov povywahts4 trebovani4 k oborudovanih i osna5enih a3roportov.As the length of air routes increases and as new types of aircraft are put into operation, the requirements for the equipment of airports increase. * The computer is 'unaware' that it has used the same English word twice as equivalents of two different Russian words.Sovremennyl a3roport predstavl4et sobo1 slojny1 kompleks injenernyx soorujeni1, texniceskix sredstv, dl4 razme5eni4 kotorogo trebuets4 territori4, izmer4ema4 v otdel6nyx sluca4x tys4cami gektarov (naprimer moskovski1 a3roport Domodedovo, H6h-1orkski1 a3roport Kennedi). Razrabotka arxitekturno-planirovocnyx sxem a3roportov predusmatrivaet naibolee raqional6noe socetanie zon-l2tno1, slujebno1 i jilo1; pri 3tom kompoziqionnym qentrom 4vl4ets4 a3rovokzal vmeste s drugimi ucastkami slujebno1 zony, neprosredstvenno sv4zannymi s obslujivaniem passajirov.The architectural and planning schemes of airports provide for the most rational combination of flight, service, and residential zones. The central element of its composition is the air terminal and other service zones directly connected with the servicing of passengers.* Grammatical class of word not recognised. False parse route leads to wrong dictionary entry.
Appendix:
| null | null | null | null | {
"paperhash": [
"toma|some_semantic_considerations_in_russian-english_machine_translation."
],
"title": [
"Some Semantic Considerations in Russian-English Machine Translation."
],
"abstract": [
"Abstract : The report describes the final RADC supported optimization phase of the SYSTRAN Russian-English translation system. The primary thrust of this effort was directed at implementing the use of semantic analysis in both source language analysis and target language synthesis. This project has shown semo-syntactic analysis to be a highly feasible means of sophisticating machine translation and decreasing the need for post-editing. (Author)"
],
"authors": [
{
"name": [
"Larissa Toma",
"P. Garrett",
"Ludek A. Kozlik",
"Donald G. Perwin",
"Chuck Starr"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null
],
"s2_corpus_id": [
"60048453"
],
"intents": [
[]
],
"isInfluential": [
false
]
} | null | 555 | 0.012613 | null | null | null | null | null | null | null | null |
35e931679cbd4e452654c3b0bba201f5f6958b28 | 236999931 | null | Implementing machine aids to translation | of a specialized text, it is obvious that special-subject languages differ from general language not only in their vocabularies, but also in their morphology, syntagmatic structures, phraseology, and, to some extent, their idiom. These features must therefore be considered in designing, developing, and implementing machine aids to translation. They are as essential to programming as to the acquisition of special-language data. Machine aids to translation -in the broader sense defined at the outset -must take into account two extremes of language and subject-matter competence on the part of users, as well as all gradations in between. One extreme is personified by the translator who exhibits a high level of competence in the source and target languages, but possesses little or no knowledge of the special subject area. At the opposite pole is the scientist, engineer, or programmer -or even the economist or jurist -who is highly knowledgeable and competent in his own field but possesses only a passive knowledge of a given foreign language. | {
"name": [
"Tanke, Eberhard H."
],
"affiliation": [
null
]
} | null | null | Translating and the Computer | 1978-11-01 | 20 | 1 | null | of a specialized text, it is obvious that special-subject languages differ from general language not only in their vocabularies, but also in their morphology, syntagmatic structures, phraseology, and, to some extent, their idiom. These features must therefore be considered in designing, developing, and implementing machine aids to translation. They are as essential to programming as to the acquisition of special-language data.Machine aids to translation -in the broader sense defined at the outset -must take into account two extremes of language and subject-matter competence on the part of users, as well as all gradations in between. One extreme is personified by the translator who exhibits a high level of competence in the source and target languages, but possesses little or no knowledge of the special subject area. At the opposite pole is the scientist, engineer, or programmer -or even the economist or jurist -who is highly knowledgeable and competent in his own field but possesses only a passive knowledge of a given foreign language.In the first case, the computer that is to function as a translation tool will have to provide not only the necessary linguistic elements -such as semantic units, terms, syntagms, or phrases -of the special-subject language, but also as much information as possible on the subject itself. In the second case, the linguistic information is of primary interest to the user. This information embraces not only orthography, including diacritics, but also parts of speech, gender, prepositional usage, and the like. Only when these heterogeneous requirements have been met can a system of machine-aided translation be employed universally with success.What happens in the extreme situation in which the user of the terminology data bank has no extensive knowledge of the field being dealt with? In analyzing the source-language text, he will have difficulty recognizing syntagmatic units. He will likewise be ill-equipped to distinguish between endocentric and exocentric syntagms or idioms. A semantic entity is endocentric if its meaning is identical with the sum total of the meanings of its constituents, and exocentric if it is not. In the first case, the translator can often transfer the constituents of an entity individually from source to target language and link them together without impairing the entity's meaning. In the case of exocentric entities, which are often idiomatic and occur frequently in special-language contexts, he cannot.If the translator does not recognize a semantic entity as such in the source language, but tries to understand the text by linking the meanings of the entity's constituents, the result will be a misunderstanding of the concept. The target-language version of the text will then contain misinterpretations of concepts, often rendering it incoherent or even incomprehensible.But even if we leave idiomatic expressions aside and limit our observations to syntagmatic structures in general, comparative analysis of special-subject language, whatever the field, reveals considerable variance in "semantic construction principles" and concepts between natural languages. Not surprisingly, the prerequisites and rules for transferring the syntagmatic units of special-subject language from one natural language to another are just as variable. However, it is generally true that specialist terms which have evolved in a given language in connection with scientific or technological development in innovation-intensive sectors are introduced directly into other languages by a process of "transliteration." A case in point is data processing, a sector in which the U.S. acquired a huge headstart over Europe and the U.S.S.R. As these simple examples confirm, it is essential for a system of machine-aided translation to provide the user with complete information of the type which helps him to recognize that a queried word may in fact, in a given context, be only a constituent of a syntagmatic element, only a part of a larger language entity in the source-language text. A word-for-word translation of multiword terms or constituents of compounds is a dangerous practice and seldom results in the correct rendering of special-language terms and syntagms in the target language. This consideration is crucial for the user in the opposite camp as well, i.e., the specialist who is versed in the grammar and rules of a foreign language, but is unfamiliar with its idiom. The provision of complete information on the term or syntagm -its concept, its usage, and its environment -helps him to penetrate the idiom of specialist jargon and discover the correlation between semantic entities in the source language and their equivalents in the target language.One of the most notorious traits of specialist language is its propensity to proliferate synonyms. Synonyms evolve and are used for many reasons: because parallel developments are pursued simultaneously by different researchers who describe their innovation in their own terms; because there is a need to find differentiating labels for competing products of the same kind; because style in writing demands variety of expression; because usage within the same language varies from region to region (e.g., English in America, Great Britain, South Africa, Australia; French in France and Belgium; Spanish in Spain and the sundry countries of Latin America), etc.Another difficulty of special language is its numerous homographs. Since the text producers (in the case of multilingual communication, the author-translator team) quite naturally have a different mental set and orientation from those to whom the text is addressed, homographs are ambiguous in the source language. They must therefore be resolved in the translation process. For it is seldom that the ambiguities of a homograph in the source language are congruent with the ambiguities of the equivalent in the target language.In designing the data base and programming for a system of machine-aided translation that is intended to serve not only translators, but anyone able to act as a link in the process of multilingual communication, it is useful to know the statistical distribution of language elements (terms, syntagms, phrases) which are revealed by the linguistic analysis of special-language texts.These were a few of the key factors considered in laying the groundwork for the development of the TEAM program system. The voluminous quantity of special-language terminology and its rate of change give rise to other aspects of no less importance.Since it was obvious from the start that not every future need, application, and technical development could be foreseen, the system was kept open-ended, and was developed pragmatically, step-by-step, with every stage of development being put in practical application immediately.The same is true of the linguistic data stored. As already mentioned, all possible sources of ambiguity were eliminated. To this end, the linguistic data of the various languagesthe terms, syntagms, phrases, etc. -are linked by their concepts or meanings, i.e., by "semantic connectors." Ambiguities are further minimized by supplying supplementary information, such as subject fields, definitions, contexts, explanatory notes, and the like. TEAM entry containing multi-word terms. The * is a control character to provide for automatic permutation of these multi-word terms or of abbreviations and their long forms. As shown in this example these two possibilities can also be combined. In this case the system generates entries automatically and places them in their proper alphabetical order. For example, in a German list the term "ADW ..." would be shown preceding the abbreviation "Analog ... " and "Wandler ..." will be listed under "W ...".The TEAM Terminology Data Bank System A characteristic feature of the system is the variety of its input and output facilities. rogation facilities. The output facilities of the system are many and varied. Lists can be produced by line printers.A CRT* phototypesetter makes it possible to typeset entire dictionaries. Microfiches can be produced with the aid of a COM** device.All output devices, line printers, the COM system and the CRT phototypesetter are directly controlled by the data bank system. This means that the continually updated contents of the data bank can be printed immediately on paper or film in any form required. To illustrate the speed with which this is done, a two-language dictionary with 200,000 entries, whose contents were revised up to the last minute before editorial deadline, was typeset in just a few hours in a format that features running titles, pagination and clusters, etc. as Fig. 4 and 5 show.In this process all typesetting commands determining layout, type of font, type size and face, and so on, are generated by the computer according to parameters which are input prior to the typesetting run.Pig. 6 shows and explains part of the classification labels used in the data systems dictionary, a page of which is shown in Fig. 5 . It should be noted that these labels are generated from the alphanumeric codes in category 06 of the TEAM entry classifying the subject field or fields to which this entry, i.e. the terms and their concept, belong. Fig. 4 A page from a Russian-German-English dictionary, showing the automatically made-up page with pagination, running title, and clusters.A page from the Data Systems Dictionary, an example for data bank-controlled CRT phototypesetting. Glossary covering a special subject field and serving as a basis for terminology standardization in the translation, documentation, and sales departments etc .appears. Compound expressions can be interrogated under any significant element of the term or phrase. He will then be shown the sum total of all relevant entries and can page through the entire inventory.Special-subject terminology is voluminous in every sector of knowledge. By the beginning of this century, the Association of German Engineers (VDI) had already compiled some 3-5 million technical terms. Vocabularies of special-subject languages grow and change rapidly in direct proportion to the rate of innovation-in science and technology, and the rate of change in other fields of learning.In this way, every member of the venture adds his contribution of special-language data to a common intellectual fund which is at the disposal of all members. By transferring the latter to the computer the overall efficiency of the translation process can be increased and the translator's task will become more rewarding. With this in mind, the TEAM terminology data bank system was combined with a text editor and processor as shown in Fig. 11 . On the right is a simplified representation of the TEAM data bank system. It is linked with a text editing and text processing system. Texts can be input using magnetic tape, OCR, or video terminals. The latter can likewise be used for text editing. This, then, is the branch used for translation and revision. Wordprocessors can also be connected to the text system to serve as "intelligent" input or output devices and of course for the editing of text. Standard output is via line printers, CRT phototypesetters, COM equipment, or on magnetic tape. This combined system makes it possible to select terras and phrases from the TEAM data bank and transfer them to the text editing system. Here, specific pools are built up -for example, on a particular system or subject area -and stored on magnetic disks for direct access. In addition, the terminology data bank of the TEAM system can be directly interrogated via the terminal of the text system. The formation of subject or system-specific pools serves to eliminate terms and syntagms not wanted, such as synonyms and wrong terminology. This is done before actual translation begins. Every text is examined in the source language for uniformity of terminology. It is compared automatically with the contents of the system-specific pools and all deviations from standard terminology are shown on lists together with their line numbers in the text. The general language portions of texts are excluded from this comparison. Text analysis and concordance programs of the TEAM program system are employed for determining what is required in the system-oriented pools. The source language text is transferred from the text editing system to the TEAM program system for analysis. If terms not contained in the system-or subjectoriented pools are discovered in the text, they are entered in both the TEAM data bank and the terminology pools of the text system. They are then available for the translating process. Wrong terms in the source language text are replaced by the proper ones. As a result of text analysis, the work entailed in proofreading is also reduced. This applies for every level of the translating process, from source language to target language text.Interconnection of TEAM program system and text editing system In any stage of the translation process, including revision and final typing, the text can be changed in various ways. The translated text can also be changed if modifications of the technical equipment, the system, or the like require it. All these functions are supported by the text editor. The text processor, together with the editor, is used when terminology contained in a text is to be changed, or when translations are to be prepared for automatic typesetting.In this text editor and processor -used in connection with the TEAM terminology data base -all editing functions are controlled on screen. These include typing error correctionwhere the original character(s) are simply overwritten -and insertion or deletion of text of almost any length -words, lines, or paragraphs.For the computer to carry out these functions, the translator or revisor enters commands, which can affect entire files, or delimited vertical and horizontal ranges. The vertical range is delimited by screen columns and the horizontal by means of line numbers, the so-called index.This index can be used to address particular portions of the text, such as lines or paragraphs, for the purpose of restructuring text files, e.g. to change the sequence of text modules. Lines and paragraphs may be copied as often as desired in order to re-use them either in the same text file or in any other file, for example, if the same object or the same facts are described in different places.A similar method is used when text modules are copied into a text that is being translated or revised. In all stages of the translation process the translator or the revisor may either work directly on screen or on paper. In the latter case a hard copy of the text is output either by way of line printer or a hard copy device connected to the terminal. Any changes entered into this hard copy are transferred into the computer-stored file with the index serving as line address.If specific strings are to be checked, substituted, or located within a text, a concordance or string search is carried out. This function is initiated by keying in the string, following a short command. The computer then searches the entire file for this string. If so desired, the strings can be automatically replaced by different strings of any length, either one-by-one throughout the entire file, or within delimited ranges. This string search and substitute facility is also used to prepare a source language text when, for example, the text is to be parcelled out to a number of translators. Before the actual translation process begins, special language terms, phrases, etc. -i.e. strings -are inserted into the source language text.In this case the term or syntagm in the source language text, i.e. the string sought, is not overwritten but the wanted target language equivalent is inserted following the source lan-guage concordance. The target strings are obtained from the special subject terminology pool which is frequently updated. This search/substitute process may again be carried out one by one. If so desired, the string sought can also be overwritten. However, this method is disadvantageous, since the translator is then no longer able to resolve ambiguities such as homographs .Even if a source language text comprises thousands of lines with frequent occurrences of the sought string, this operation, if carried out in one go, only takes a few seconds. This process is illustrated by the following text which is first shown in its original form and then a second time after pre-editing, that is, after the target language strings have been inserted.The text editor and processor greatly facilitates text editing functions. For example, long character strings, such as modules, which are stored in the computer can be identified and retrieved by a specific character or character combination. Each time the translator or revisor enters this character or character combination in a text, the computer will automatically insert this long string. All these facilities do, of course, necessitate file protection, so that text cannot be destroyed accidentally or altered intentionally without authorization of the responsible translator or revisor.As already mentioned, output can be directed to various output media as desired, such as hard copy printers -printing one screen after the other -line printers, magnetic tapes, discs, and phototypesetters. If, following translation and revision, typeset output is requested, the text editor and processor offers additional features. A copy of the target language file is analyzed by the computer and according to input format criteria, typesetting commands are automatically inserted. A specific output format including font is then produced. This output format may, however, be changed by simply altering a few parameters that interpret the text format analysis programs . After these typesetting commands have been inserted in the text, a magnetic tape is produced which controls the CRT phototypesetter DIGISET. As an end product, properly laid out text, including running titles, automatic pagination, syllabification, etc., is received, either on photo paper or on film.In the next stage of program development, this procedure will be automated. All special-language terms and all syntagms contained in the text will be found by the computer. The terminology data pools of the text system will be interrogated, and will supply the target language equivalents, automatically. These in turn will be automatically inserted in the source language text in the manner described. This text will be the one used for translation.Concluding The TEAM program system is likewise used to prepare crash courses aimed at developing the foreign-language reading ability of employees. The assumption underlying these courses is that the specialist will understand foreign-language texts dealing with his own field, if he possesses a minimal vocabulary of combined special and general language terms and is acquainted with the key morphological and syntactical charac- | null | null | null | null | Main paper:
cooperation in terminology acquisition and evaluation.:
Special-subject terminology is voluminous in every sector of knowledge. By the beginning of this century, the Association of German Engineers (VDI) had already compiled some 3-5 million technical terms. Vocabularies of special-subject languages grow and change rapidly in direct proportion to the rate of innovation-in science and technology, and the rate of change in other fields of learning.In this way, every member of the venture adds his contribution of special-language data to a common intellectual fund which is at the disposal of all members. By transferring the latter to the computer the overall efficiency of the translation process can be increased and the translator's task will become more rewarding. With this in mind, the TEAM terminology data bank system was combined with a text editor and processor as shown in Fig. 11 . On the right is a simplified representation of the TEAM data bank system. It is linked with a text editing and text processing system. Texts can be input using magnetic tape, OCR, or video terminals. The latter can likewise be used for text editing. This, then, is the branch used for translation and revision. Wordprocessors can also be connected to the text system to serve as "intelligent" input or output devices and of course for the editing of text. Standard output is via line printers, CRT phototypesetters, COM equipment, or on magnetic tape. This combined system makes it possible to select terras and phrases from the TEAM data bank and transfer them to the text editing system. Here, specific pools are built up -for example, on a particular system or subject area -and stored on magnetic disks for direct access. In addition, the terminology data bank of the TEAM system can be directly interrogated via the terminal of the text system. The formation of subject or system-specific pools serves to eliminate terms and syntagms not wanted, such as synonyms and wrong terminology. This is done before actual translation begins. Every text is examined in the source language for uniformity of terminology. It is compared automatically with the contents of the system-specific pools and all deviations from standard terminology are shown on lists together with their line numbers in the text. The general language portions of texts are excluded from this comparison. Text analysis and concordance programs of the TEAM program system are employed for determining what is required in the system-oriented pools. The source language text is transferred from the text editing system to the TEAM program system for analysis. If terms not contained in the system-or subjectoriented pools are discovered in the text, they are entered in both the TEAM data bank and the terminology pools of the text system. They are then available for the translating process. Wrong terms in the source language text are replaced by the proper ones. As a result of text analysis, the work entailed in proofreading is also reduced. This applies for every level of the translating process, from source language to target language text.Interconnection of TEAM program system and text editing system In any stage of the translation process, including revision and final typing, the text can be changed in various ways. The translated text can also be changed if modifications of the technical equipment, the system, or the like require it. All these functions are supported by the text editor. The text processor, together with the editor, is used when terminology contained in a text is to be changed, or when translations are to be prepared for automatic typesetting.In this text editor and processor -used in connection with the TEAM terminology data base -all editing functions are controlled on screen. These include typing error correctionwhere the original character(s) are simply overwritten -and insertion or deletion of text of almost any length -words, lines, or paragraphs.For the computer to carry out these functions, the translator or revisor enters commands, which can affect entire files, or delimited vertical and horizontal ranges. The vertical range is delimited by screen columns and the horizontal by means of line numbers, the so-called index.This index can be used to address particular portions of the text, such as lines or paragraphs, for the purpose of restructuring text files, e.g. to change the sequence of text modules. Lines and paragraphs may be copied as often as desired in order to re-use them either in the same text file or in any other file, for example, if the same object or the same facts are described in different places.A similar method is used when text modules are copied into a text that is being translated or revised. In all stages of the translation process the translator or the revisor may either work directly on screen or on paper. In the latter case a hard copy of the text is output either by way of line printer or a hard copy device connected to the terminal. Any changes entered into this hard copy are transferred into the computer-stored file with the index serving as line address.If specific strings are to be checked, substituted, or located within a text, a concordance or string search is carried out. This function is initiated by keying in the string, following a short command. The computer then searches the entire file for this string. If so desired, the strings can be automatically replaced by different strings of any length, either one-by-one throughout the entire file, or within delimited ranges. This string search and substitute facility is also used to prepare a source language text when, for example, the text is to be parcelled out to a number of translators. Before the actual translation process begins, special language terms, phrases, etc. -i.e. strings -are inserted into the source language text.In this case the term or syntagm in the source language text, i.e. the string sought, is not overwritten but the wanted target language equivalent is inserted following the source lan-guage concordance. The target strings are obtained from the special subject terminology pool which is frequently updated. This search/substitute process may again be carried out one by one. If so desired, the string sought can also be overwritten. However, this method is disadvantageous, since the translator is then no longer able to resolve ambiguities such as homographs .Even if a source language text comprises thousands of lines with frequent occurrences of the sought string, this operation, if carried out in one go, only takes a few seconds. This process is illustrated by the following text which is first shown in its original form and then a second time after pre-editing, that is, after the target language strings have been inserted.The text editor and processor greatly facilitates text editing functions. For example, long character strings, such as modules, which are stored in the computer can be identified and retrieved by a specific character or character combination. Each time the translator or revisor enters this character or character combination in a text, the computer will automatically insert this long string. All these facilities do, of course, necessitate file protection, so that text cannot be destroyed accidentally or altered intentionally without authorization of the responsible translator or revisor.As already mentioned, output can be directed to various output media as desired, such as hard copy printers -printing one screen after the other -line printers, magnetic tapes, discs, and phototypesetters. If, following translation and revision, typeset output is requested, the text editor and processor offers additional features. A copy of the target language file is analyzed by the computer and according to input format criteria, typesetting commands are automatically inserted. A specific output format including font is then produced. This output format may, however, be changed by simply altering a few parameters that interpret the text format analysis programs . After these typesetting commands have been inserted in the text, a magnetic tape is produced which controls the CRT phototypesetter DIGISET. As an end product, properly laid out text, including running titles, automatic pagination, syllabification, etc., is received, either on photo paper or on film.In the next stage of program development, this procedure will be automated. All special-language terms and all syntagms contained in the text will be found by the computer. The terminology data pools of the text system will be interrogated, and will supply the target language equivalents, automatically. These in turn will be automatically inserted in the source language text in the manner described. This text will be the one used for translation.Concluding The TEAM program system is likewise used to prepare crash courses aimed at developing the foreign-language reading ability of employees. The assumption underlying these courses is that the specialist will understand foreign-language texts dealing with his own field, if he possesses a minimal vocabulary of combined special and general language terms and is acquainted with the key morphological and syntactical charac-
:
of a specialized text, it is obvious that special-subject languages differ from general language not only in their vocabularies, but also in their morphology, syntagmatic structures, phraseology, and, to some extent, their idiom. These features must therefore be considered in designing, developing, and implementing machine aids to translation. They are as essential to programming as to the acquisition of special-language data.Machine aids to translation -in the broader sense defined at the outset -must take into account two extremes of language and subject-matter competence on the part of users, as well as all gradations in between. One extreme is personified by the translator who exhibits a high level of competence in the source and target languages, but possesses little or no knowledge of the special subject area. At the opposite pole is the scientist, engineer, or programmer -or even the economist or jurist -who is highly knowledgeable and competent in his own field but possesses only a passive knowledge of a given foreign language.In the first case, the computer that is to function as a translation tool will have to provide not only the necessary linguistic elements -such as semantic units, terms, syntagms, or phrases -of the special-subject language, but also as much information as possible on the subject itself. In the second case, the linguistic information is of primary interest to the user. This information embraces not only orthography, including diacritics, but also parts of speech, gender, prepositional usage, and the like. Only when these heterogeneous requirements have been met can a system of machine-aided translation be employed universally with success.What happens in the extreme situation in which the user of the terminology data bank has no extensive knowledge of the field being dealt with? In analyzing the source-language text, he will have difficulty recognizing syntagmatic units. He will likewise be ill-equipped to distinguish between endocentric and exocentric syntagms or idioms. A semantic entity is endocentric if its meaning is identical with the sum total of the meanings of its constituents, and exocentric if it is not. In the first case, the translator can often transfer the constituents of an entity individually from source to target language and link them together without impairing the entity's meaning. In the case of exocentric entities, which are often idiomatic and occur frequently in special-language contexts, he cannot.If the translator does not recognize a semantic entity as such in the source language, but tries to understand the text by linking the meanings of the entity's constituents, the result will be a misunderstanding of the concept. The target-language version of the text will then contain misinterpretations of concepts, often rendering it incoherent or even incomprehensible.But even if we leave idiomatic expressions aside and limit our observations to syntagmatic structures in general, comparative analysis of special-subject language, whatever the field, reveals considerable variance in "semantic construction principles" and concepts between natural languages. Not surprisingly, the prerequisites and rules for transferring the syntagmatic units of special-subject language from one natural language to another are just as variable. However, it is generally true that specialist terms which have evolved in a given language in connection with scientific or technological development in innovation-intensive sectors are introduced directly into other languages by a process of "transliteration." A case in point is data processing, a sector in which the U.S. acquired a huge headstart over Europe and the U.S.S.R. As these simple examples confirm, it is essential for a system of machine-aided translation to provide the user with complete information of the type which helps him to recognize that a queried word may in fact, in a given context, be only a constituent of a syntagmatic element, only a part of a larger language entity in the source-language text. A word-for-word translation of multiword terms or constituents of compounds is a dangerous practice and seldom results in the correct rendering of special-language terms and syntagms in the target language. This consideration is crucial for the user in the opposite camp as well, i.e., the specialist who is versed in the grammar and rules of a foreign language, but is unfamiliar with its idiom. The provision of complete information on the term or syntagm -its concept, its usage, and its environment -helps him to penetrate the idiom of specialist jargon and discover the correlation between semantic entities in the source language and their equivalents in the target language.One of the most notorious traits of specialist language is its propensity to proliferate synonyms. Synonyms evolve and are used for many reasons: because parallel developments are pursued simultaneously by different researchers who describe their innovation in their own terms; because there is a need to find differentiating labels for competing products of the same kind; because style in writing demands variety of expression; because usage within the same language varies from region to region (e.g., English in America, Great Britain, South Africa, Australia; French in France and Belgium; Spanish in Spain and the sundry countries of Latin America), etc.Another difficulty of special language is its numerous homographs. Since the text producers (in the case of multilingual communication, the author-translator team) quite naturally have a different mental set and orientation from those to whom the text is addressed, homographs are ambiguous in the source language. They must therefore be resolved in the translation process. For it is seldom that the ambiguities of a homograph in the source language are congruent with the ambiguities of the equivalent in the target language.In designing the data base and programming for a system of machine-aided translation that is intended to serve not only translators, but anyone able to act as a link in the process of multilingual communication, it is useful to know the statistical distribution of language elements (terms, syntagms, phrases) which are revealed by the linguistic analysis of special-language texts.These were a few of the key factors considered in laying the groundwork for the development of the TEAM program system. The voluminous quantity of special-language terminology and its rate of change give rise to other aspects of no less importance.Since it was obvious from the start that not every future need, application, and technical development could be foreseen, the system was kept open-ended, and was developed pragmatically, step-by-step, with every stage of development being put in practical application immediately.The same is true of the linguistic data stored. As already mentioned, all possible sources of ambiguity were eliminated. To this end, the linguistic data of the various languagesthe terms, syntagms, phrases, etc. -are linked by their concepts or meanings, i.e., by "semantic connectors." Ambiguities are further minimized by supplying supplementary information, such as subject fields, definitions, contexts, explanatory notes, and the like. TEAM entry containing multi-word terms. The * is a control character to provide for automatic permutation of these multi-word terms or of abbreviations and their long forms. As shown in this example these two possibilities can also be combined. In this case the system generates entries automatically and places them in their proper alphabetical order. For example, in a German list the term "ADW ..." would be shown preceding the abbreviation "Analog ... " and "Wandler ..." will be listed under "W ...".The TEAM Terminology Data Bank System A characteristic feature of the system is the variety of its input and output facilities. rogation facilities. The output facilities of the system are many and varied. Lists can be produced by line printers.A CRT* phototypesetter makes it possible to typeset entire dictionaries. Microfiches can be produced with the aid of a COM** device.All output devices, line printers, the COM system and the CRT phototypesetter are directly controlled by the data bank system. This means that the continually updated contents of the data bank can be printed immediately on paper or film in any form required. To illustrate the speed with which this is done, a two-language dictionary with 200,000 entries, whose contents were revised up to the last minute before editorial deadline, was typeset in just a few hours in a format that features running titles, pagination and clusters, etc. as Fig. 4 and 5 show.In this process all typesetting commands determining layout, type of font, type size and face, and so on, are generated by the computer according to parameters which are input prior to the typesetting run.Pig. 6 shows and explains part of the classification labels used in the data systems dictionary, a page of which is shown in Fig. 5 . It should be noted that these labels are generated from the alphanumeric codes in category 06 of the TEAM entry classifying the subject field or fields to which this entry, i.e. the terms and their concept, belong. Fig. 4 A page from a Russian-German-English dictionary, showing the automatically made-up page with pagination, running title, and clusters.A page from the Data Systems Dictionary, an example for data bank-controlled CRT phototypesetting. Glossary covering a special subject field and serving as a basis for terminology standardization in the translation, documentation, and sales departments etc .appears. Compound expressions can be interrogated under any significant element of the term or phrase. He will then be shown the sum total of all relevant entries and can page through the entire inventory.
Appendix:
| null | null | null | null | {
"paperhash": [
"schmidt|der_einsatz_des_lexikographischen_zweigs_eines_datenbanksystems_zur_herstellung_eines_phraseologischen_fachglossars"
],
"title": [
"Der Einsatz des lexikographischen Zweigs eines Datenbanksystems zur Herstellung eines phraseologischen Fachglossars"
],
"abstract": [
"Der folgende Aufsatz schildert den Versuch, mit einem bereits im Einsatz befindlichen Datenbanksystem, und zwar mit dem vom Sprachendienst der SIEMENS AG entwickelten System TEAM'), die Information-Retrieval-Probleme des Sprachendienstes des Auswärtigen Amts zu lösen. Man hatte dort das System TEAM anhand mehrerer Publikationen bereits kennengelernt. Es galt zu untersuchen, ob der Übergang von einer rein terminologischen Datenbank mit Einund Mehrwortbenennungen zu einem System, das auch die Verarbeitung fachsprachlicher Phraseologie erlaubt, ohne weiteres möglich ist, da für den Bereich der Phraseologie ja völlig andere strukturelle Bedingungen gelten; man denke hier nur an Flexionsendungen, flektierte unregelmäßige Verben usw. Die Kriterien für das Wiederauffinden der Information (Deskriptoren) müssen ebenfalls anders geartet sein. Verlauf und Ergebnis der Untersuchung werden im folgenden gezeigt."
],
"authors": [
{
"name": [
"R. Schmidt",
"O. Vollnhals"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null
],
"s2_corpus_id": [
"61732519"
],
"intents": [
[]
],
"isInfluential": [
false
]
} | null | 555 | 0.001802 | null | null | null | null | null | null | null | null |
f17dedc5301875441a1ab928103d2a0ab4df478b | 26855299 | null | Computer aided translation - a business viewpoint | Before one starts to look for a particular solution, it is necessary to define the precise needs of the problem. Such is the case with our Company; the solution we are pursuing is tailored to the specific communication needs we have identified and it may well not be the most effective direction for another Company. In order to understand why we have chosen our particular path, it is helpful to explain briefly the Company environment. THE ENVIRONMENT Xerox operates in more than 36 Countries spread throughout the world. The task of our Technical Service function is to install and maintain our products, both rented and sold, in each of these Countries. Although the size of operation differs considerably between Countries, the individual functional support needs, in terms of technical data for our Service representatives, is virtually identical. The data is provided in the form of Service Documentation. | {
"name": [
"Elliston, John S. G."
],
"affiliation": [
null
]
} | null | null | Translating and the Computer | 1978-11-01 | 6 | 13 | null | null | null | null | This documentation is vital for the field service organisations to be able to do their job. The documentation provides the Service Representative with all the technical data that he needs. It comprises maintenance procedures, technical data, diagnostic procedures and spare parts lists. This type of information must be provided for all products and all configurations. To provide all of this data, we go through the following processes: documentation development, validation, translation, production, distribution and maintenance. An operation of this type is complex enough, the pressure related to accuracy and timeliness for individual locations just adds to the difficulty. Materials are developed prior to launch and "in-field" validation tests in English may well be running concurrently with several translation programmes, in order to enable the staggered National launch needs to be met.Of the 36 Countries mentioned, only 7 have English as their first language. Even within this group there are sufficient differences within the languages to cause some misunderstanding. A further 14 Countries are obliged to use English text documentation largely because of economics related to the scale of operation. Within this group the ability to speak English varies from very high to very low. The remaining 15 Countries require that all documentation is translated before it can be used in their field environment.The whole operation is critical and costly and any inefficiency can quickly escalate costs. Consequently, the process needs to be subjected to tight controls.The problem can be broadly expressed under three headings: . Costs .Timeliness or lapse time. . Clarity of communication.The demand on our translation resource grows every year. This demand is related to our increasing product range, refinements to existing products and the normal on-going need to maintain existing documentation. An additional factor is the legal demands placed upon a multinational operation to translate to meet legal requirements. One obvious answer is to increase our resource to handle the growing load. Unfortunately, increasing the translation resource increases our cost base and makes us less competitive. The solution we need must be found in productivity, i.e. using the resources we already have, more efficiently.Service documentation is developed by our headquarters function either in the US or UK. In either case, it will be originated in English. On average, it will be between 3 to 4 months after the first English version has been validated before a translated manual will be available. (The precise figure will depend upon the complexity of the product and translation workload and prioritisation). This lapse time reduces the possibility to field-test products in non-English speaking environments and consequently, puts a heavy burden on the English speaking Companies, who must now do the majority of the field tests. The question of validation in non English speaking markets is further compromised by localised translation. Manual translation will inevitably be tempered with experience and interpretation. This means one is no longer testing the original, therefore results obtained are invalid. This is even more of a problem if the 'subjective' or the 'interpretative' translation actually improves on the original English version. It is one thing to identify a documentation fault and relate it to either an origination or translation error; it is quite another thing when a problem is resolved by the translation. In the latter case, the translation passes the test and the original English version gets printed complete with fault.Timeliness is also a key feature of our documentation corrections and update system. At present, extensive delays can result before translated data is available to the field. Again, the reasons are the same, the complexity of the task and prioritisation.The two major factors that contribute to ambiguity within our multinational environment are:. ambiguity -text must be written in a clear manner.. vocabulary -text should only contain those words that are known to be in the end users vocabulary.A commonly expressed opinion is that if a group of 50 translators were given the same sentence to translate, they would produce 50 different versions. A computer given the same sentence will only give one translation. How can we assume that the one output from the Computer is the right translation. I believe that the question directs our attention to the wrong place. The real problem is in the fact that the original sentence was capable of 50 different interpretations. To the producer of Service Documentation, this is frightening. If one sentence is open to so much interpretation, what chance does a Service Representative have when one realises the permutations of a complete book? Obviously, the first problem to tackle is the generation of source material.Our experience to date has shown that it is extremely difficult to define clarity sufficiently objectively to ensure an author writes clearly. Each writer has his own personalised style. Simply using good grammatical English does not in itself eliminate ambiguity. If a writer has written a piece of text that conforms to grammatical rules, the question as to whether it is ambiguous usually results in, at best, subjective discussion and, at worst, emotive argument.Secondly, with present techniques, it is not possible to ensure that authors only use those words within the vocabulary of the target population. Our target population spreads across 36 Countries and ranges from 18 to 50 years of age.Added to this situation, we find that all too often words or phrases have a different meaning when placed in a different context or worse still, in another cultural environment. Recently, whilst visiting the United States, I purchased a coffee from a secretary who looked after the departmental percolator. The price for the coffee was a very modest 10 cents. When I learned of the low cost, I mentioned that it was very cheap. This comment was followed by a rather obvious silence. My colleague later pointed out that it would have been better to suggest that the coffee was 'inexpensive'. The word 'cheap' in the US is usually used in a derogatory sense. Thus, my rather innocent comment was taken as a criticism of the coffee and could have resulted in my having to find an alternative source of coffee.It is perhaps this type of apparently insignificant interpretation that can so easily result in misunderstanding, or even offence. This is especially risky where we tread the often delicate path of operating across National practices and Customs.The need is relatively straightforward. Our Company needs a means of communicating technical data, instructions and information to our worldwide Field Service. The method chosen must be acceptable in the business sense, that is the costs incurred must be less than the benefits gained. It must be capable of providing the output communication when and where it is required. Thirdly, it must ensure that the end user can retrieve data accurately and quickly.Finally, throughout the complete cycle from generation of source language text to translation into target language text, the needs of three categories of end user must be met.Personnel whose first language is English.. Personnel who are obliged to use English but whose first language is not -English. ELLISTON (This particular requirement adds a third dimension to the discussion of manual vs machine translation, as it automatically forces one to look more closely at the source language).Personnel (or machines) who are required to translate the English text into the target language. (These people differ from category two in so far as they will not have the same depth of technical knowledge and understanding).There was and still is, no instant or obvious solution. The path we have followed has taken us through several potential solutions, each in turn being discarded until we have arrived at our present status.Perhaps the first approach that we looked at, was one based on the "Caterpilla English" concept. At its simplest it is a limited vocabulary with each word being carefully defined. The target population is then taught to recognise the words rather as one would recognise a symbol, then associate it with the defined meaning. The end user is not taught to pronounce the words, just to recognise them, thus he does not actually learn the source language. This method has been successful in many areas, but did not fit our particular situation. It would be true to say that we rejected the system more on social grounds than on the basis of any real scientific testing. A new Company setting up its operation may well find the system workable, although legislation in some Countries might make even that difficult. In our situation, we were dealing with well established Operating Companies who already translated material to a high standard and a Field Service force, used to having their support documentation in their own language. To switch to a limited English language was seen as a retrograde step and totally unacceptable.A second solution considered, was the use of a "Command English". This looked a far more likely solution as one could fairly accurately prepare translation for standard command sentences. This would achieve two things. Firstly, a guarantee that the translation is accepted in advance and secondly, a machine can be used to speed up the process. The difficulty that was encountered in this attempt was the constraints placed upon the source language writer. Much of the Service information can be expressed in the directive manner of command English. The problem starts to show when one writes "descriptive statements" or "test objectives" or even statements relating to judgements. In addition, the potential for developing the Command language for use in the areas of training and Customer documentation seems almost zero.During the period these approaches were being investigated, we also examined some of the claims at that time for existing computer translation systems.These systems by and large claimed to offer unrestricted input translation and seemed promising. Regrettably, these claims seldom lived up to the test and the systems tended to be extremely expensive in development and post edit costs per language.The system that we are currently using to develop our total translation process is SYSTRAN. Initially, we did some research with uncontrolled input text which resulted in unacceptable output in terms of the post edit effort required.The dilemma at this stage was that if one used a totally free form of input, the computer translation output required a massive post edit. Conversely, if the source language was written to permit computer aided translation it became unacceptably restrictive to the author or originator. An additional problem with this tightly controlled input is the acceptance of the user of source language material. The large post edit task was unacceptable in terms of both cost and job satisfaction. The amount of post edit was such that it took almost as long as it would have taken to translate the whole exercise manually. The morale of a translator in this mode of operation is low. After all, the job is reduced to trying to understand and correct a rather badly written document.By now it was clear that computer aided translation was achievable but its acceptability was related to the balance between the control of the source language input and the degree of post edit required of the target language output. On the one side, if the constraints placed on the originator are too severe the increased load would cancel the productivity benefit of the system. In addition, one runs into the -real danger of author motivation. On the other side, if one relaxes the input control on the source text translation too much, the post edit function grows to the point that machine productivity is wasted and a similar motivation problem exists, this time for the translator.Vocabulary -It became necessary to ensure that misunderstanding or ambiguity did not arise out of the use of a particular word, or because of the context in which the word was placed.Writing Rules -Once again, to reduce ambiguity in the source text it was necessary to determine rules to define the required size and construction of sentences, etc.This vocabulary became known as RX Customised Vocabulary. The vocabulary was compared to one that was developed in our US location and from the two sets, we developed our present lists, now known as MCE (Multinational Customised English). This vocabulary is made up of several sub-groups.Firstly, the basic core group vocabulary consisting of approximately 1000 words. This group forms the basic communication word list. The other groups are to permit the specialist communication within our specific Company environment. They fall under the categories of copy quality terminology, publications terminology, abbreviations, weights, measures, etc. In all, this provides a total vocabulary of under 3000 words. The next step in the process was to get each target language user to identify their own language equivalents for each word in the MCE vocabulary. As anyone who has tried will know, selecting one foreign language word for one English word is a tough proposition. The important factor is not to simply look for a word for word equivalent, but define one and only one meaning for each English word and then find the target language word or phrase to relate to that precise meaning. For example, the word "replace" is often used to request two quite different actions, e.g. .Remove part A and replace it with part B.In the first case, we are using the word to mean "exchange" and in the second case to mean "put back". This usually gives little problem to experienced English speaking staff, but does cause problems for those who use English text, but whose first language is not English. It also gives problems to the computer.Again, for the sake of clarity, each word was defined as a specific part of speech and, if possible, never in more than one category, i.e. "Switch" as a noun and not as a verb. Unfortunately, this was not always possible. This statement demonstrates at least two problems. Firstly, the sentence is too complex. It needs to be written in several short sentences. Secondly, in an attempt to reduce the amount of text the technician must read, we tend to leave out the definite article. Again, anybody familiar or trained on the subject and who speaks English fluently will probably have no problem. The computer unfortunately has neither of these two advantages.Imagine reading a telegram, (usually written in abbreviated form to save costs), "SHIP SINKS". Does it mean "THE SHIP SINKS" or "SHIP THE SINKS". The difference in meaning by simply moving the position of the definite article is enormous. To overcome these difficulties, the original statement should be written as follows: "Loosen the Main motor.Loosen the Drive shaft. Slide both parts until they touch the Back plate". Summing up, the input to the computer is controlled to the extent that it must be written within the vocabulary of the computer and written in simple short sentences.One concern that was originally felt by the English speaking service representatives, is that the text that they would be issued with would be written in a form of pidgin English. The example in figure 1 shows this is clearly not the case. Our field test indicated that of the order of 90% of our UK sample found no difference between the ordinary text and the customised text, in terms of usability. Our tests in Sweden, where English is the second language, indicated a 70% response that found the MCE version far easier to understand and use.It is important to stress that the judgement in terms of final acceptability is not that of the originator or translator, but that of the end user, in our case the Service Representative.The judgement is based upon the end user's ability to follow the instructions easily and quickly with no negative impact on job performance.With the computer programmed and the necessary vocabularies or dictionaries and target language loaded, the system is ready to go.The source language text is fed into the computer and the target output can be delivered in either hard copy or displayed on a video display unit. The next stage is to post edit the output. This involves identifying errors, analysing them and determining the cause and, if possible, determine solutions to eliminate similar errors in future. These solutions might fall in one of several areas. It may be necessary to add to, or modify the existing dictionaries. It might be necessary to alter the writing rules for future use or it could be that the computer software needs adjustment. Each of these actions has a cumulative effect, gradually taking the total process nearer to the minimum acceptable productivity targets set against the system. It can be seen that in the early stages of implementation of such a system, it is very much a question of "running in" the system.At this point in time, we are extremely optimistic that computer aided translation, using our input controls and based on SYSTRAN, can be used to significantly improve our translation function. We have already shown translation productivity gains of better than 4:1. This level of productivity includes the post edit function related to computer translation. Evidence to date suggests we will improve this level of productivity as we continue to use the system and reap the cumulative effect of software and file improvement. So far our tests have been limited and "off line". The programme that we are working on now is designed to test the total process. This process will involve on line authors' originating the source text using the writing rules and the MCE vocabulary. The text will then be run through the computer and post edited by a qualified translator for the target language. Translation is only part of the total process of developing and implementing a Service Documentation system. As in other systems, there is little to be gained in speeding up part of the process if you leave a bottleneck in another part of the system. For example, there is little gain in spending millions of pounds to build a motorway if all it does is speed up the traffic to the motorway exist and create a traffic jam at the intersection. So with our approach to translation, it is an integrated part of a total system.Once the post edit stage has been completed, the system will permit further productivity benefits. As all the text, both source and target language, is held in the computer we can electronically file it, update and modify it and print it out. By hooking our translation systems directly into a computerised text editing system, we can automatically select type face, size, etc. and compose the final page on a video screen. This greatly speeds up the total process and eliminates the relatively slow and expensive text creation and composition stages each language has traditionally required.Other language pairs will follow, but in each case it is essential to ensure the end user of the target language is involved in the development process. One obvious example where it is essential to gain acceptance from the end user is where you are selecting one base language that is to be used in more than one Country. e.g.French -France, Canada, Switzerland, Belgium. Spanish -Spain, Latin America Dutch -Holland, Belgium Portuguese -Portugal, Latin AmericaIn each of the above situations, the variations of both languages in each Country are significant. However, it is possible to gain acceptance for a common vocabulary between Countries by careful selection and discussion.Exactly how we will finally install the system in terms of function and location is still under development considerations. The diagram (fig. 2) shows the principal activities that we are 'hooking' in to the system. At present, the post edit function is done from hardcopy, as it is an integral part of the input and computer software preparation. Once the system is up and running and post edit becomes purely a translation/editing function, the work could be done remotely or on site, direct on a video display unit.At this stage, we cannot say for what ranges of application computer aided translation will be suitable. As was said earlier, the system can be used for pure technical communication, where facts are listed for future retrieval. Whether it is possible to extend this into the area of training has yet to be established. Within the definition of training in our Company, we range from pure technical skill-based training to the highly interpretive interpersonal skill training. It seems reasonable to assume that the more straightforward technical training offers the best opportunity to use Computer Aided Translation. However, it must be appreciated that training materials are not written to record data but to enable initial learning. To this end, training material tends to be written in a more personalised style, making use of colloquial expressions and localised examples. Obviously, this type of translation requires a combination of the translator's skills and also those of the Trainer. In short, we are now in the field of interpretation, rather than translation and at this moment in time the computer falls short of that particular target. Already computer aided translation has come a long way, there is every reason to believe it will go still further. Our judgements on its acceptability must be based on realistic performance criteria and not on subjective argument. We are not trying to perfect an automated system that "appreciates" the finer points of a particular language, but a 'tool' to assist us in the functional translation of a specific area of business communications. | null | Main paper:
:
This documentation is vital for the field service organisations to be able to do their job. The documentation provides the Service Representative with all the technical data that he needs. It comprises maintenance procedures, technical data, diagnostic procedures and spare parts lists. This type of information must be provided for all products and all configurations. To provide all of this data, we go through the following processes: documentation development, validation, translation, production, distribution and maintenance. An operation of this type is complex enough, the pressure related to accuracy and timeliness for individual locations just adds to the difficulty. Materials are developed prior to launch and "in-field" validation tests in English may well be running concurrently with several translation programmes, in order to enable the staggered National launch needs to be met.Of the 36 Countries mentioned, only 7 have English as their first language. Even within this group there are sufficient differences within the languages to cause some misunderstanding. A further 14 Countries are obliged to use English text documentation largely because of economics related to the scale of operation. Within this group the ability to speak English varies from very high to very low. The remaining 15 Countries require that all documentation is translated before it can be used in their field environment.The whole operation is critical and costly and any inefficiency can quickly escalate costs. Consequently, the process needs to be subjected to tight controls.The problem can be broadly expressed under three headings: . Costs .Timeliness or lapse time. . Clarity of communication.The demand on our translation resource grows every year. This demand is related to our increasing product range, refinements to existing products and the normal on-going need to maintain existing documentation. An additional factor is the legal demands placed upon a multinational operation to translate to meet legal requirements. One obvious answer is to increase our resource to handle the growing load. Unfortunately, increasing the translation resource increases our cost base and makes us less competitive. The solution we need must be found in productivity, i.e. using the resources we already have, more efficiently.Service documentation is developed by our headquarters function either in the US or UK. In either case, it will be originated in English. On average, it will be between 3 to 4 months after the first English version has been validated before a translated manual will be available. (The precise figure will depend upon the complexity of the product and translation workload and prioritisation). This lapse time reduces the possibility to field-test products in non-English speaking environments and consequently, puts a heavy burden on the English speaking Companies, who must now do the majority of the field tests. The question of validation in non English speaking markets is further compromised by localised translation. Manual translation will inevitably be tempered with experience and interpretation. This means one is no longer testing the original, therefore results obtained are invalid. This is even more of a problem if the 'subjective' or the 'interpretative' translation actually improves on the original English version. It is one thing to identify a documentation fault and relate it to either an origination or translation error; it is quite another thing when a problem is resolved by the translation. In the latter case, the translation passes the test and the original English version gets printed complete with fault.Timeliness is also a key feature of our documentation corrections and update system. At present, extensive delays can result before translated data is available to the field. Again, the reasons are the same, the complexity of the task and prioritisation.The two major factors that contribute to ambiguity within our multinational environment are:. ambiguity -text must be written in a clear manner.. vocabulary -text should only contain those words that are known to be in the end users vocabulary.A commonly expressed opinion is that if a group of 50 translators were given the same sentence to translate, they would produce 50 different versions. A computer given the same sentence will only give one translation. How can we assume that the one output from the Computer is the right translation. I believe that the question directs our attention to the wrong place. The real problem is in the fact that the original sentence was capable of 50 different interpretations. To the producer of Service Documentation, this is frightening. If one sentence is open to so much interpretation, what chance does a Service Representative have when one realises the permutations of a complete book? Obviously, the first problem to tackle is the generation of source material.Our experience to date has shown that it is extremely difficult to define clarity sufficiently objectively to ensure an author writes clearly. Each writer has his own personalised style. Simply using good grammatical English does not in itself eliminate ambiguity. If a writer has written a piece of text that conforms to grammatical rules, the question as to whether it is ambiguous usually results in, at best, subjective discussion and, at worst, emotive argument.Secondly, with present techniques, it is not possible to ensure that authors only use those words within the vocabulary of the target population. Our target population spreads across 36 Countries and ranges from 18 to 50 years of age.Added to this situation, we find that all too often words or phrases have a different meaning when placed in a different context or worse still, in another cultural environment. Recently, whilst visiting the United States, I purchased a coffee from a secretary who looked after the departmental percolator. The price for the coffee was a very modest 10 cents. When I learned of the low cost, I mentioned that it was very cheap. This comment was followed by a rather obvious silence. My colleague later pointed out that it would have been better to suggest that the coffee was 'inexpensive'. The word 'cheap' in the US is usually used in a derogatory sense. Thus, my rather innocent comment was taken as a criticism of the coffee and could have resulted in my having to find an alternative source of coffee.It is perhaps this type of apparently insignificant interpretation that can so easily result in misunderstanding, or even offence. This is especially risky where we tread the often delicate path of operating across National practices and Customs.The need is relatively straightforward. Our Company needs a means of communicating technical data, instructions and information to our worldwide Field Service. The method chosen must be acceptable in the business sense, that is the costs incurred must be less than the benefits gained. It must be capable of providing the output communication when and where it is required. Thirdly, it must ensure that the end user can retrieve data accurately and quickly.Finally, throughout the complete cycle from generation of source language text to translation into target language text, the needs of three categories of end user must be met.Personnel whose first language is English.. Personnel who are obliged to use English but whose first language is not -English. ELLISTON (This particular requirement adds a third dimension to the discussion of manual vs machine translation, as it automatically forces one to look more closely at the source language).Personnel (or machines) who are required to translate the English text into the target language. (These people differ from category two in so far as they will not have the same depth of technical knowledge and understanding).There was and still is, no instant or obvious solution. The path we have followed has taken us through several potential solutions, each in turn being discarded until we have arrived at our present status.Perhaps the first approach that we looked at, was one based on the "Caterpilla English" concept. At its simplest it is a limited vocabulary with each word being carefully defined. The target population is then taught to recognise the words rather as one would recognise a symbol, then associate it with the defined meaning. The end user is not taught to pronounce the words, just to recognise them, thus he does not actually learn the source language. This method has been successful in many areas, but did not fit our particular situation. It would be true to say that we rejected the system more on social grounds than on the basis of any real scientific testing. A new Company setting up its operation may well find the system workable, although legislation in some Countries might make even that difficult. In our situation, we were dealing with well established Operating Companies who already translated material to a high standard and a Field Service force, used to having their support documentation in their own language. To switch to a limited English language was seen as a retrograde step and totally unacceptable.A second solution considered, was the use of a "Command English". This looked a far more likely solution as one could fairly accurately prepare translation for standard command sentences. This would achieve two things. Firstly, a guarantee that the translation is accepted in advance and secondly, a machine can be used to speed up the process. The difficulty that was encountered in this attempt was the constraints placed upon the source language writer. Much of the Service information can be expressed in the directive manner of command English. The problem starts to show when one writes "descriptive statements" or "test objectives" or even statements relating to judgements. In addition, the potential for developing the Command language for use in the areas of training and Customer documentation seems almost zero.During the period these approaches were being investigated, we also examined some of the claims at that time for existing computer translation systems.These systems by and large claimed to offer unrestricted input translation and seemed promising. Regrettably, these claims seldom lived up to the test and the systems tended to be extremely expensive in development and post edit costs per language.The system that we are currently using to develop our total translation process is SYSTRAN. Initially, we did some research with uncontrolled input text which resulted in unacceptable output in terms of the post edit effort required.The dilemma at this stage was that if one used a totally free form of input, the computer translation output required a massive post edit. Conversely, if the source language was written to permit computer aided translation it became unacceptably restrictive to the author or originator. An additional problem with this tightly controlled input is the acceptance of the user of source language material. The large post edit task was unacceptable in terms of both cost and job satisfaction. The amount of post edit was such that it took almost as long as it would have taken to translate the whole exercise manually. The morale of a translator in this mode of operation is low. After all, the job is reduced to trying to understand and correct a rather badly written document.By now it was clear that computer aided translation was achievable but its acceptability was related to the balance between the control of the source language input and the degree of post edit required of the target language output. On the one side, if the constraints placed on the originator are too severe the increased load would cancel the productivity benefit of the system. In addition, one runs into the -real danger of author motivation. On the other side, if one relaxes the input control on the source text translation too much, the post edit function grows to the point that machine productivity is wasted and a similar motivation problem exists, this time for the translator.Vocabulary -It became necessary to ensure that misunderstanding or ambiguity did not arise out of the use of a particular word, or because of the context in which the word was placed.Writing Rules -Once again, to reduce ambiguity in the source text it was necessary to determine rules to define the required size and construction of sentences, etc.This vocabulary became known as RX Customised Vocabulary. The vocabulary was compared to one that was developed in our US location and from the two sets, we developed our present lists, now known as MCE (Multinational Customised English). This vocabulary is made up of several sub-groups.Firstly, the basic core group vocabulary consisting of approximately 1000 words. This group forms the basic communication word list. The other groups are to permit the specialist communication within our specific Company environment. They fall under the categories of copy quality terminology, publications terminology, abbreviations, weights, measures, etc. In all, this provides a total vocabulary of under 3000 words. The next step in the process was to get each target language user to identify their own language equivalents for each word in the MCE vocabulary. As anyone who has tried will know, selecting one foreign language word for one English word is a tough proposition. The important factor is not to simply look for a word for word equivalent, but define one and only one meaning for each English word and then find the target language word or phrase to relate to that precise meaning. For example, the word "replace" is often used to request two quite different actions, e.g. .Remove part A and replace it with part B.In the first case, we are using the word to mean "exchange" and in the second case to mean "put back". This usually gives little problem to experienced English speaking staff, but does cause problems for those who use English text, but whose first language is not English. It also gives problems to the computer.Again, for the sake of clarity, each word was defined as a specific part of speech and, if possible, never in more than one category, i.e. "Switch" as a noun and not as a verb. Unfortunately, this was not always possible. This statement demonstrates at least two problems. Firstly, the sentence is too complex. It needs to be written in several short sentences. Secondly, in an attempt to reduce the amount of text the technician must read, we tend to leave out the definite article. Again, anybody familiar or trained on the subject and who speaks English fluently will probably have no problem. The computer unfortunately has neither of these two advantages.Imagine reading a telegram, (usually written in abbreviated form to save costs), "SHIP SINKS". Does it mean "THE SHIP SINKS" or "SHIP THE SINKS". The difference in meaning by simply moving the position of the definite article is enormous. To overcome these difficulties, the original statement should be written as follows: "Loosen the Main motor.Loosen the Drive shaft. Slide both parts until they touch the Back plate". Summing up, the input to the computer is controlled to the extent that it must be written within the vocabulary of the computer and written in simple short sentences.One concern that was originally felt by the English speaking service representatives, is that the text that they would be issued with would be written in a form of pidgin English. The example in figure 1 shows this is clearly not the case. Our field test indicated that of the order of 90% of our UK sample found no difference between the ordinary text and the customised text, in terms of usability. Our tests in Sweden, where English is the second language, indicated a 70% response that found the MCE version far easier to understand and use.It is important to stress that the judgement in terms of final acceptability is not that of the originator or translator, but that of the end user, in our case the Service Representative.The judgement is based upon the end user's ability to follow the instructions easily and quickly with no negative impact on job performance.With the computer programmed and the necessary vocabularies or dictionaries and target language loaded, the system is ready to go.The source language text is fed into the computer and the target output can be delivered in either hard copy or displayed on a video display unit. The next stage is to post edit the output. This involves identifying errors, analysing them and determining the cause and, if possible, determine solutions to eliminate similar errors in future. These solutions might fall in one of several areas. It may be necessary to add to, or modify the existing dictionaries. It might be necessary to alter the writing rules for future use or it could be that the computer software needs adjustment. Each of these actions has a cumulative effect, gradually taking the total process nearer to the minimum acceptable productivity targets set against the system. It can be seen that in the early stages of implementation of such a system, it is very much a question of "running in" the system.At this point in time, we are extremely optimistic that computer aided translation, using our input controls and based on SYSTRAN, can be used to significantly improve our translation function. We have already shown translation productivity gains of better than 4:1. This level of productivity includes the post edit function related to computer translation. Evidence to date suggests we will improve this level of productivity as we continue to use the system and reap the cumulative effect of software and file improvement. So far our tests have been limited and "off line". The programme that we are working on now is designed to test the total process. This process will involve on line authors' originating the source text using the writing rules and the MCE vocabulary. The text will then be run through the computer and post edited by a qualified translator for the target language. Translation is only part of the total process of developing and implementing a Service Documentation system. As in other systems, there is little to be gained in speeding up part of the process if you leave a bottleneck in another part of the system. For example, there is little gain in spending millions of pounds to build a motorway if all it does is speed up the traffic to the motorway exist and create a traffic jam at the intersection. So with our approach to translation, it is an integrated part of a total system.Once the post edit stage has been completed, the system will permit further productivity benefits. As all the text, both source and target language, is held in the computer we can electronically file it, update and modify it and print it out. By hooking our translation systems directly into a computerised text editing system, we can automatically select type face, size, etc. and compose the final page on a video screen. This greatly speeds up the total process and eliminates the relatively slow and expensive text creation and composition stages each language has traditionally required.Other language pairs will follow, but in each case it is essential to ensure the end user of the target language is involved in the development process. One obvious example where it is essential to gain acceptance from the end user is where you are selecting one base language that is to be used in more than one Country. e.g.French -France, Canada, Switzerland, Belgium. Spanish -Spain, Latin America Dutch -Holland, Belgium Portuguese -Portugal, Latin AmericaIn each of the above situations, the variations of both languages in each Country are significant. However, it is possible to gain acceptance for a common vocabulary between Countries by careful selection and discussion.Exactly how we will finally install the system in terms of function and location is still under development considerations. The diagram (fig. 2) shows the principal activities that we are 'hooking' in to the system. At present, the post edit function is done from hardcopy, as it is an integral part of the input and computer software preparation. Once the system is up and running and post edit becomes purely a translation/editing function, the work could be done remotely or on site, direct on a video display unit.At this stage, we cannot say for what ranges of application computer aided translation will be suitable. As was said earlier, the system can be used for pure technical communication, where facts are listed for future retrieval. Whether it is possible to extend this into the area of training has yet to be established. Within the definition of training in our Company, we range from pure technical skill-based training to the highly interpretive interpersonal skill training. It seems reasonable to assume that the more straightforward technical training offers the best opportunity to use Computer Aided Translation. However, it must be appreciated that training materials are not written to record data but to enable initial learning. To this end, training material tends to be written in a more personalised style, making use of colloquial expressions and localised examples. Obviously, this type of translation requires a combination of the translator's skills and also those of the Trainer. In short, we are now in the field of interpretation, rather than translation and at this moment in time the computer falls short of that particular target. Already computer aided translation has come a long way, there is every reason to believe it will go still further. Our judgements on its acceptability must be based on realistic performance criteria and not on subjective argument. We are not trying to perfect an automated system that "appreciates" the finer points of a particular language, but a 'tool' to assist us in the functional translation of a specific area of business communications.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 555 | 0.023423 | null | null | null | null | null | null | null | null |
98da6ca848580ffbcce605fde15ea8cf7242126b | 236999919 | null | {EURODICAUTOM} | EURODICAUTOM J. GOETSCHALCKX Head of the Terminology Bureau Medium-and Long-term Translation Service Commission of the European Communities Luxembourg The original name for what we now so proudly call EURODICAUTOM was in fact DICAUTOM, standing simply for dictionnaire automatique. This reveals that from the beginning, 15 years ago, when my predecessor Mr. J.A. BACHRACH started this project, we did not intend to set up a clever, sophisticated gadget. We wanted to offer the translator a new, more efficient tool, so that he could do his work as he did before, consulting a dictionary to resolve terminological problems if necessary. EURODICAUTOM is not an automatic device, at least not a penny-in-the-slotmachine. It is just a computer-aided dictionary. But it is of course a dictionary of a special type able to supply information more rapidly and more comprehensively than would a normal lexicographical work. | {
"name": [
"Goetschalckx, J."
],
"affiliation": [
null
]
} | null | null | Translating and the Computer | 1978-11-01 | 9 | 1 | null | The phraseological approach is very important in the scientific and technical field where we have to deal with so-called special languages. Specialisation by our translators is not always possible. For this reason we have to offer full information on the use of the terms in this particular field and very often also a lot of purely technical information. If these phrases or sentences are well chosen they can cater for both the linguistic and technical information needed.For subtle distinctions between scientific or technical terms a definition is very often the best solution, but it is not always easy to find definitions for all terms appearing in a translation.Although mere term-to term equivalents are often dangerous in the hands of a "multi-purpose" translator, we did not want to be cut off from this important source of scientific and technical terminology. We believe that being able to accept terminological information in any form of presentation is an ideal situation for exchanges with other centres.What is the origin of the information we put into a terminology bank? Our own terminological information is the result of an analysis of original documents in each language. The comparative study of these original documents gives real equivalents of professional language usage. This means that the equivalency is very often at the level of the phrase and not necessarily between corresponding words. The semantic content of the corresponding phrases is the same but it can be distributed in a different way from one language to another among the morphemes But we also use terminology compiled by specialized institutions such as AFNOR, the French standards institute, the International Welding Institute, and the European Brewery Convention which is concerned not only about the quality of beer but also about its terminology.In what order is this information presented? In our system each information unit is called an entity. I understand that in recent congresses on applied linguistics the French word "fiche" has been generally and internationally accepted for the same concept.A "fiche" in automated terminology contains purely terminological information, documentation data, and information needed for the electronic data processing and general organisation of the data base and the whole system. In the EURODICAUTOM system each "fiche" is determined by three elements: NI: identification number (as we started from French, the letters are the wrong way round, of course); BE: "bureau émetteur", which is the terminology office which created this terminological information or assumed the responsibility for its input; TY: "type", which is a three-letter code describing an homogeneous collection of terminology or a collection of individual cards designated by the code TFI (fiche individuelle, fiche isolée). This is what we call the "BETYNI" of a fiche.These three elements determine exactly the terminology unit, the "fiche". This apparently unimportant detail has important consequences, It means that a network of terminology centres can feed EURODICAUTOM and, independently of one another, can use their own organisation, as far as choice of subjects for their terminology collections and their own numbering systems are concerned. There is no risk that it might interfere with the collections of other centres belonging to the network.and distribution. This is important for, as you may know, the European Communities have several institutions each with their own translation department and usually a terminology office.When we further examine this EURODICAUTOM fiche we see that it also mentions the author. We respect intellectual property. Nevertheless this possibility has not been used so far. Terminologists are modest people. Or could it be that they are not quite sure of the overall value of their inventions? In most cases we stick to professional usage. Our creative function in terminology is only called upon when there is no other way out.Then we have a reliability code which in fact, has nothing to do with reliability because it is not a real measure of the reliability of the information. Terminologists can also have terminology problems of their own.The so-called reliability code also indicates whether the information has the value of a standard. This is the case of international or national standardized terminology.Extracts from the European Treaties or regulations are of the same type. In quotations they have of course, to be reproduced literally as they stand in the official texts. This type of fiche bears the code 5.If all the language versions are supported by solid sources, the fiche will be given the code 4. As soon as perfection is no longer guaranteed the figure is reduced.This does not mean that the information is bad. We simply are not sure, for lack of bibliographical information. Consequently we know that we still have to work on this fiche to bring it up to the standards of good, reliable terminological information. So it still has something to do with reliability. All these documentary and data processing data are there to highlight and support the real heart of the matter, the purely terminological information in "vedette"or in context, with or without definition.If there is something more to say about the "vedette" that cannot be part of a definition, such as nationally or geographically limited usage, peculiar plural form etc., this can be done in a scope note NT.Until now I have only spoken about the organisation of the fiche, the presentation of the terminological unit with its explanatory documentation. Let us now come to the retrieval stage.As I said in the beginning, EURODICAUTOM should offer the translator a working tool which allows him to achieve higher efficiency and quality without implying fundamental changes in his working methods .After an extremely short and simple sign-on procedure, the terminal invites the user: "Type your question".Although we have only 130,000 "fiches" or entities at the moment, we have taken into account from the beginning the necessity of having many more, 1 million perhaps, and the difficulties arising then because of polysemy etc.That is why we have incorporated a weighting system. The idea was to give first of ail the best answer according to the principle of the longest match.if a multiterm ABC is the subject of a question, the system first gives ABC if it is in the corpus, and then AB, BC or AC. This "partial" information can be useful. If not the translator just stops the interrogation. This reveals again our basic concept: a working tool for a specialist not a "penny-in-the-slot machine" for anybody.Let me give an example: A translator is looking tor the translation of the technical expression "relative cinematic viscosity". The first answer gives the translation of this multiterm. But if the user continues with the interrogation he will obtain consecutively "relative viscosity" and "cinematic viscosity".If the full expression had not been available, it would have been very easy to reconstruct it from these two "partial answers".It is perfectly clear that the higher the number of terms in the expression looked up the more a partial answer is likely to give useful information. With two terms the risk of irrelevant information is much greater, but as as the system is made for translators they must be capable of judging immediately if the partial information for is useful or not.To improve the system we shall reduce the partial answers to those containing not less than n-2 of the terms contained in the question.the partial answers to the question AB (two terms) will give alternatively answers with A only and answers with B only. Let us assume that you are looking for the translation of roll-on-roll-off-ship . If the system gives you a series of partial answers with the term "ship" it is highly unlikely that this will prove useful information. On the other hand any partial answer with the term "roll-onroll-off" will give a useful hint for the right translation of the original expression. So to avoid a long series of poor information containing the term ship, the system will give alternatively both elements of the expression . "Roll-on-roll-off craft" e.g. would be helpful for translating the original "RO/RO ship".Congress attenders of the clever type will have realised that there is a retrieval problem with the phraseological entries, because the words in the phrases are not always in the standard form. This is especially true of languages like German with its numerous inflections and Danish because of the suffixation of the article. To solve this problem we use the truncation device. If, for example, a phrase contains a plural form, truncation will still allow the information to be obtained. Even in Italian it provides the possibility of asking for a form ending in CA or CO and obtaining as an answer the plural form ending in CHE or CHI.We can do even better: the expression "in and outgoing ships" is a form which is not very frequent in English but more so in German and in Dutch.A fiche containing the expression "Stuetz-und Bewegungsapparat" can be the answer to a question requesting the translation of Stuetzapparat.For interrogation regarding a polysemic term or a document concerning a very specific subject field, the interrogation can be made after introducing one or more subject codes.This should not eliminate other information corresponding to the question asked but without the subject code asked for.Coding is often a very subjective matter but wouldn't it be a pity to lose information because of a mere coding error? On the other hand some terms can be common to different fields and have the same equivalent in other languages. Forming techniques in plastics are partly the same as in metals. A terminologist introducing this information on the basis of a document on metal forming could forget to also assign it the general code for mechanical treatment. A user asking about a document dealing with plastics forming might make the same mistake while composing his interrogation parameters. | null | null | null | null | Main paper:
:
The phraseological approach is very important in the scientific and technical field where we have to deal with so-called special languages. Specialisation by our translators is not always possible. For this reason we have to offer full information on the use of the terms in this particular field and very often also a lot of purely technical information. If these phrases or sentences are well chosen they can cater for both the linguistic and technical information needed.For subtle distinctions between scientific or technical terms a definition is very often the best solution, but it is not always easy to find definitions for all terms appearing in a translation.Although mere term-to term equivalents are often dangerous in the hands of a "multi-purpose" translator, we did not want to be cut off from this important source of scientific and technical terminology. We believe that being able to accept terminological information in any form of presentation is an ideal situation for exchanges with other centres.What is the origin of the information we put into a terminology bank? Our own terminological information is the result of an analysis of original documents in each language. The comparative study of these original documents gives real equivalents of professional language usage. This means that the equivalency is very often at the level of the phrase and not necessarily between corresponding words. The semantic content of the corresponding phrases is the same but it can be distributed in a different way from one language to another among the morphemes But we also use terminology compiled by specialized institutions such as AFNOR, the French standards institute, the International Welding Institute, and the European Brewery Convention which is concerned not only about the quality of beer but also about its terminology.In what order is this information presented? In our system each information unit is called an entity. I understand that in recent congresses on applied linguistics the French word "fiche" has been generally and internationally accepted for the same concept.A "fiche" in automated terminology contains purely terminological information, documentation data, and information needed for the electronic data processing and general organisation of the data base and the whole system. In the EURODICAUTOM system each "fiche" is determined by three elements: NI: identification number (as we started from French, the letters are the wrong way round, of course); BE: "bureau émetteur", which is the terminology office which created this terminological information or assumed the responsibility for its input; TY: "type", which is a three-letter code describing an homogeneous collection of terminology or a collection of individual cards designated by the code TFI (fiche individuelle, fiche isolée). This is what we call the "BETYNI" of a fiche.These three elements determine exactly the terminology unit, the "fiche". This apparently unimportant detail has important consequences, It means that a network of terminology centres can feed EURODICAUTOM and, independently of one another, can use their own organisation, as far as choice of subjects for their terminology collections and their own numbering systems are concerned. There is no risk that it might interfere with the collections of other centres belonging to the network.and distribution. This is important for, as you may know, the European Communities have several institutions each with their own translation department and usually a terminology office.When we further examine this EURODICAUTOM fiche we see that it also mentions the author. We respect intellectual property. Nevertheless this possibility has not been used so far. Terminologists are modest people. Or could it be that they are not quite sure of the overall value of their inventions? In most cases we stick to professional usage. Our creative function in terminology is only called upon when there is no other way out.Then we have a reliability code which in fact, has nothing to do with reliability because it is not a real measure of the reliability of the information. Terminologists can also have terminology problems of their own.The so-called reliability code also indicates whether the information has the value of a standard. This is the case of international or national standardized terminology.Extracts from the European Treaties or regulations are of the same type. In quotations they have of course, to be reproduced literally as they stand in the official texts. This type of fiche bears the code 5.If all the language versions are supported by solid sources, the fiche will be given the code 4. As soon as perfection is no longer guaranteed the figure is reduced.This does not mean that the information is bad. We simply are not sure, for lack of bibliographical information. Consequently we know that we still have to work on this fiche to bring it up to the standards of good, reliable terminological information. So it still has something to do with reliability. All these documentary and data processing data are there to highlight and support the real heart of the matter, the purely terminological information in "vedette"or in context, with or without definition.If there is something more to say about the "vedette" that cannot be part of a definition, such as nationally or geographically limited usage, peculiar plural form etc., this can be done in a scope note NT.Until now I have only spoken about the organisation of the fiche, the presentation of the terminological unit with its explanatory documentation. Let us now come to the retrieval stage.As I said in the beginning, EURODICAUTOM should offer the translator a working tool which allows him to achieve higher efficiency and quality without implying fundamental changes in his working methods .After an extremely short and simple sign-on procedure, the terminal invites the user: "Type your question".Although we have only 130,000 "fiches" or entities at the moment, we have taken into account from the beginning the necessity of having many more, 1 million perhaps, and the difficulties arising then because of polysemy etc.That is why we have incorporated a weighting system. The idea was to give first of ail the best answer according to the principle of the longest match.if a multiterm ABC is the subject of a question, the system first gives ABC if it is in the corpus, and then AB, BC or AC. This "partial" information can be useful. If not the translator just stops the interrogation. This reveals again our basic concept: a working tool for a specialist not a "penny-in-the-slot machine" for anybody.Let me give an example: A translator is looking tor the translation of the technical expression "relative cinematic viscosity". The first answer gives the translation of this multiterm. But if the user continues with the interrogation he will obtain consecutively "relative viscosity" and "cinematic viscosity".If the full expression had not been available, it would have been very easy to reconstruct it from these two "partial answers".It is perfectly clear that the higher the number of terms in the expression looked up the more a partial answer is likely to give useful information. With two terms the risk of irrelevant information is much greater, but as as the system is made for translators they must be capable of judging immediately if the partial information for is useful or not.To improve the system we shall reduce the partial answers to those containing not less than n-2 of the terms contained in the question.the partial answers to the question AB (two terms) will give alternatively answers with A only and answers with B only. Let us assume that you are looking for the translation of roll-on-roll-off-ship . If the system gives you a series of partial answers with the term "ship" it is highly unlikely that this will prove useful information. On the other hand any partial answer with the term "roll-onroll-off" will give a useful hint for the right translation of the original expression. So to avoid a long series of poor information containing the term ship, the system will give alternatively both elements of the expression . "Roll-on-roll-off craft" e.g. would be helpful for translating the original "RO/RO ship".Congress attenders of the clever type will have realised that there is a retrieval problem with the phraseological entries, because the words in the phrases are not always in the standard form. This is especially true of languages like German with its numerous inflections and Danish because of the suffixation of the article. To solve this problem we use the truncation device. If, for example, a phrase contains a plural form, truncation will still allow the information to be obtained. Even in Italian it provides the possibility of asking for a form ending in CA or CO and obtaining as an answer the plural form ending in CHE or CHI.We can do even better: the expression "in and outgoing ships" is a form which is not very frequent in English but more so in German and in Dutch.A fiche containing the expression "Stuetz-und Bewegungsapparat" can be the answer to a question requesting the translation of Stuetzapparat.For interrogation regarding a polysemic term or a document concerning a very specific subject field, the interrogation can be made after introducing one or more subject codes.This should not eliminate other information corresponding to the question asked but without the subject code asked for.Coding is often a very subjective matter but wouldn't it be a pity to lose information because of a mere coding error? On the other hand some terms can be common to different fields and have the same equivalent in other languages. Forming techniques in plastics are partly the same as in metals. A terminologist introducing this information on the basis of a document on metal forming could forget to also assign it the general code for mechanical treatment. A user asking about a document dealing with plastics forming might make the same mistake while composing his interrogation parameters.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 555 | 0.001802 | null | null | null | null | null | null | null | null |
40b48fdac5e13c2c0cae96153d674bd72ef83155 | 61821757 | null | Machine translation and computerised terminology systems - a translator{'}s viewpoint | Georges Van Slype (1978): "Deuxième évaluation du système de traduction automatique SYSTRAN anglais-français de la Commission des Communautés européennes". Bureau Marcel van Dijk, Brussels. | {
"name": [
"Arthern, Peter J."
],
"affiliation": [
null
]
} | null | null | Translating and the Computer | 1978-11-01 | 4 | 44 | null | Although I have only a short time available, I want to look at translating and the computer from two points of view. The first is that of a fairly large translating organization which is beginning to use a computerized terminology data base -Eurodicautom -and may become a user of machine translation. The second point of view is that of a translatorand being a staff translator myself I have had to try to put myself into a freelance translator's shoes as well, in order to get a complete picture.I am sure the first question which a translator asks about machine translation is "How will it affect my job?" The question was first asked in the 1950s as machine translation projects proliferated in the United States following a demonstration in 1954 by IBM and a research team at Georgetown University. By 1965 American government agencies are estimated to have spent some 20 million dollars in supporting machine translation research at 17 different institutions. And then the Automatic Language Processing Advisory Committee reported in 1966 that machine translation was slower, less accurate and twice as expensive as human translation and that there was no immediate or predictable prospect of useful machine translation.As Professor Sager has told us, the Commission has bought an American machine-translation system, "Systran", which it is developing in co-operation with the originator, Dr. Peter Toma, to translate texts from French into English and from English into French, and now from English into Italian. A very small number of translators have been assisting with computer programming in this connection, and with the preparation of dictionaries. Other linguists have been revising machine translations into French from English, in development trials. For them, machine translation has become a colleague. If it is true that "to understand is to forgive", this may explain why linguists who have been closely involved in developing the system have been more ready to revise its output than some others. At any rate, if there is a future for machine translation in the European Community institutions, it is obviously going to involve many linguists, either as programmers, lexicologists or revisersor perhaps "pre-editors".It is no part of my present responsibility to make an official assessment of the quality of the Systran translations already done from English into French at the Commission, but as a matter of interest I asked a number of experienced colleagues in our own French Division to evaluate one particular passage from three points of view. I asked one to read the French translation without having sight of the English original, and to tell me how much she understood. This corresponds roughly to the "intelligibility" criterion used in the Commission's own evaluation of Systran.(1) I asked a second colleague, the Head of our French Division, to "mark" the Systran output as if it had been submitted by a candidate in a competition for recruiting French translators.I asked a third colleague -a reviser -to revise the raw Systran output on the basis of the English original in exactly the same way as he would revise a translation made by one of our French translators in the normal course of his work.( 1)Georges Van Slype (1978): "Deuxième évaluation du système de traduction automatique SYSTRAN anglais-français de la Commission des Communautés européennes". Bureau Marcel van Dijk, Brussels.The last two checks correspond to the "fidelity" criterion also used in the Commission's evaluation of Systran, which most translators will instinctively regard as being the principal criterion applicable in judging machine translation, if not the only one.The original English text, the raw Systran French translation, and the French text as revised by our reviser are given in Annex I, together with my French colleagues' comments, in the form in which they were kind enough to note them down for me.Summarizing these comments, the raw translation considered on its own was felt to be quite inadequate for informing a French reader who did not know English about the purposes of the experiment which is described, or the procedure employed. All that he would grasp would be that an experiment with chickens had taken place.As for the translation considered as an entry in a competition to recruit French translators to the Council's staff, the Head of the French Division wrote that if he had had a paper like this to mark he would probably have stopped halfway down the first page, giving the candidate no marks at all.English original felt that the machine which had produced the translation has a memory which is far too rudimentary. He considered that evaluation of the system was premature and could not be conclusive because there was no way of assessing the results which might be achieved by a machine equipped with an adequate memory. This was all the more regrettable in that such results provided ammunition for the detractors of machine translation.Leaving this final conclusion with you, I want now to see how machine translation might affect the operations of the Secretariat of the Council of the European Communities. I must emphasize again that these remarks in no way represent any official position -they simply suggest themselves in the present situation and many if not all will be relevant in any firm or organization which is contemplating the possible use of machine translation.Provided that the quality of the final output was acceptable for the purpose in mind, the Council could have three reasons for using a machine translation system. These are, one, that it was cheaper than translation by the traditional translator-reviser system; two, that it was faster, or three, that there were not enough competent linguistsavailable to produce the necessary translations in any other way. I do not think there is any possibility of the Council Secretariat using raw machine translation for any of its texts, because the only quality we can accept is 100% fidelity to the meaning of the original, even though the style of our translations -as of many of the original textsoften suffers because of the very short deadlines against which we have to work.Leaving aside the possible use of machine translation because of a shortage of translators, our principle criterion would be whether we could produce accurate translations faster by using machine translation plus pre-editing and/or post-editing of texts, than we can at present with translators and revisers. In our particular circumstances -where texts have to be sent to the capitals of all the Member States ahead of meetings of the Council, COREPER (Permanent Representatives Committee) and countless working partiesspeed is of the essence ana cost is a secondary consideration. Which is not to say that a reduction in costs would not be welcome.With any given machine translation system, either Systran or the projected European system, we should need to analyse the types of text drafted in the Secretariat solely from the point of view of the total time taken to produce a 100% accurate translation from the original text. Since we could assume that the central processor time in the computer would be identical for every type of text, and could be neglected in comparison with human translation, this would amount to noting the time taken to pre-edit and/or post-edit different types of machine-translated text and comparing this with the total time taken to translate and revise texts of the same type in the normal way. Unless there were clear savings in time, from the moment the original text reached the Translation Department to the moment the completed translation was finally typed, we should not be interested in machine translation.One would in fact expect that texts which are structured at a superficial level, such as minutes of meetings, would be more amenable to machine translation under our conditions than speeches made by the President of the Council before the European Parliament. Other texts, such as the agendas for meetings, and even the implementing provisions of a Council Regulation or Decision as amended by a working party, can probably be dealt with more efficiently by an extended textprocessing system, than by machine translation as such.In fact, in the Council Secretariat we already employ "translation by photo-copy" to a considerable extent for such things as standard telexes, press releases, appointments to Committees, etc., where it is mainly a question of inserting names, dates, and the titles of documents in a standard format.It is at this point that I go beyond the brief I sketched for myself in the abstract of my paper which you will find in your programme. This is because it has become completely clear to me, since I started preparing for to-day's Seminar, that it is the advent of text-processing systems, not machine translation or even terminology data banks, which is the application of computers which is going to affect professional translators most directly -all of us, freelances and staff translators alike.Professor Sager and Mr. Tanke have both referred to text-processing systems already, so all I need do is to stress their immense flexibility, in that a single translator working on his own can derive many of the advantages which make a large integrated system so attractive for an organization like the Council or the Commission of the European Communities, with our hundreds of translators.the possibility of amending a text repeatedly on the display screen until it is ready for typing in its final format Extra advantages which the Council, or any large organization with an integrated text-processing system, can expect to derive, lie in the possibility of sending texts from one terminal to another for processing. For example, a text which was being amended during a meeting of the Council could be transmitted page by page to terminals in the various Translation Divisions. The translation of the original textcould be called up on the screen and amended, and the amended translation sent straight back to a terminal with its associated printer in a room next to the Council chamber, so that a complete new text in all required languages could be available by the end of the meeting.When we were asked at the Council some months ago to co-operate in an enquiry into post-editing systems for machine translation, I commented that our post-editing system consisted of a red ball-point pen in a reviser's hands. This may in fact continue to be true, since if working with a keyboard to revise translated texts at a visual display unit proves to be uncomfortable, it will be a simple matter to have the text printed out and given to a reviser for revision on paper in the traditional way. The corrections will then be made on the text-processing system by a secretary.This very brief sketch of the possible uses of textprocessing systems shows why all translators must consider their use, and also why machine translation systems and computerized terminology data banks must from now on be integrated into text-processing systems.Before I go on to deal with computerized terminology data banks, however, I would like to look at one or two more aspects of machine translation in general.In a multi-lingual situation such as the one we have in the European Communities, where it is often necessary to produce translations in parallel into several languages from a given original, it will obviously be an attractive proposition -until we get raw machine translation from free-text input which is of almost the same quality as that produced now by our translators -to concentrate on pre-editing texts for machine translation, rather than on post-editing the translations. A good job done on pre-editing a text will save post-editing several translations, and this is a point which those working on the new European machine translation system will presumably have in mind.If one adopts this approach, however, there will be a tendency to go still further back and to attempt to get the authors of texts to draft them in a standardized form which reduces the need for pre-editing. This is where we come up against resistance -we have already met it in the Council Secretariat when attempts have been made to encourage administrators to use standard formats so as to facilitate translation by photo-copy. And of course, this approach is just not on in the European Parliament or the Economic and Social Committee, where elected representatives of the people must obviously be free to express themselves just as they wish.If it were possible to dictate to people how they should write or speak, simply for the sake of making machine translation cheaper or easier, we could end up by making it more difficult for them to express themselves in their own language than it would be for them to learn a second language and use it.Finally, there is a real danger that the widespread use of machine translations which would not be stylistically acceptable if produced by a translator, even if they convey the message of the original, would debase and corrupt the natural languages now in use.On the other hand, it may be possible in some Community operations to replace natural language altogether by computerized information. Indeed, only the other day I was engaged in revising a proposal for a Council Regulation which contained the following clause: "The documents referred to in the preceding paragraph or elsewhere in this Regulation may be replaced by computerized information produced in any form for the same purpose".While translators working outside large firms or organizations are unlikely to come into direct contact with machine translation, and all translators ought to start looking at the use of text-processing machines or systems immediately, computerized terminology data banks fall between these two extremes. Their development and use have so far been restricted to large firms and organizations, but the impending introduction of publicly accessible data-transmission networks such as "Euronet" and systems such as "Teletext" and "Viewdata" which will use the domestic television set as a visual display unit, may mean that any staff or freelance translator will be able to dial for information from a term bank in the not-too-distant future.We translators may regard machine translation systems as competitors, and therefore fear them, but we instinctively feel more at home with something which is obviously not threatening, since all it can do is to help us in our work. Term banks must be "user-oriented", as became abundantly clear at a workshop on "Eurodicautom" which was held in Luxembourg last week, when we found it necessary to spend a con-siderable time discussing who was intended to use the system, and how, before we could look profitably at its content and structure.This question of intended use is paramount, since if it is not settled before a system is developed the resulting confusion may be disastrous. In addition to constituting an aid to translation, term banks can also be used for documentary purposes and for standardization -for example, for maintaining single-language normative dictionaries or as mono-lingual or multi-lingual thesauri for information retrieval systems. However, we are only concerned now with bilingual or multi-lingual term banks -we can also call them electronic dictionaries -specifically intended to assist translators in the same way as traditional dictionaries.he cannot. Firstly, he has a right to expect that the information given to him is clearly and logically presented, and can be read easily and quickly. This also applies to normal dictionaries, and is one of the principal criteria normally applied to such dictionaries. Secondly, he has a right to expect that the information given to him is reliable and accurate. However, he must himself decide on the value of this information and make his choice between alternative translations of a given expression, as he does with a normal dictionary.The basic difference between a printed dictionary and a term bank is that in the term bank all the information is stored electronically and can be added to, updated and amended at will at any time, and that any or all of the information which it contains at a given moment can be made available by a variety of means. It combines the advantages of centralization of information with de-centralization in making it available.The information in a term bank can be made available to the translator in three ways; on paper, in the form of a special subject glossary or a text-related glossary; via a television-type screen in a visual display unit used on line; or on micro-fiche used with a micro-fiche reader. The last two ways of looking up information are normally used to answer single queries arising in the course of translating a text, so that the translator will not need to make more than a mental note, or perhaps a hand-written note, of the answer. In both cases, however, it is possible to make a complete record of what appears on the screen, via a printer connected to the visual display unit or a photo-copier attached to the micro-fiche reader.At this point, it will be worth looking at the advantages and disadvantages of all three systems, both for an individual translator and for an organization using a term bank.At the Bundessprachenamt near Cologne, where some 250 linguists are engaged in translating largely technical texts for the West German Ministry of Defence, a computerized term bank has been in daily use for the last ten years. The philosophy there has always been to keep the translator away from the computer and to give him his information on paper, or on micro-fiche.The Bundessprachenamt's computer produces two basic types of glossary. The first is a special-purpose glossary, printed in a normal type-face, for use by several or many translators who are all working on a large long-term project, perhaps in several places at the same time. The second is a text-related glossary produced in the form of computer print-out for a specific text.In this second case, the translator underlines in his original text the terms he does not know, or on which he wants to check, and returns the text to the administrative office. Here a secretary types these terms into the computer which prints them out, with their equivalents in the target language, either in the order in which they appeared in the original, or in alphabetical order. This list is given to the translator, who in the meantime has been doing another job, a few hours later, or the next day.It is now the translators' responsibility, with the help of subject codes and other information printed out alongside the natural language equivalents, to choose whether the translation offered fits in the context of his text, and which of a number of equivalents does so, if he is offered a choice. If the computer offers no translation, or he is not satisfied with what it provides, the translator has to find the term he wants by other means open to all translators, such as looking up normal dictionaries and reference works, or asking colleagues.He notes on the computer print-out the new terms which he finds and uses, and these are then checked by a terminologist before being entered in the term bank for further use, within a fortnight at the latest.The great advantage of this system for the organization using it is that it gives constant direct feed-back from the translators to the system, so that the latest terms are being recorded all the time and made available to all translators. In practice, nothing like the same level of feedback is produced by the use of visual display units or microfiches.The advantage of the visual display unit used on line, both for the translator and for the organization employing him, is that he can immediately obtain the latest possible information in reply to a question which crops up while he is actually doing a translation. This is particularly important for a Translation Department in an organization like the Secretariat of the Council of the European Communities, where many documents have to be translated against very short deadlines. One can also envisage interpreters consulting such a visual display unit during a meeting, at least when they are working in pairs and one interpreter could interrogate the term bank while his colleague kept talking.Micro-fiche has the advantage that a very large number of terms can be stored in a very small space, that it is cheap to produce, and that it is practicable to distribute the up-dated contents of a term bank to a large number of users, both "in-house" and outside the organization, every six months or so. It would seem at first sight that this might after all be the cheapest and most practical way of distributing the contents of tern banks to freelance translators and to staff translators outside the organizations managing them.One important psychological factor in using visual display units and micro-fiche readers for presenting terminology to translators is that it is not as easy to absorb information from an illuminated screen as from the printed page. If a term bank is designed for use by either of these methods, it is vital that the information which the translator wants should be presented to him clearly in a minimum of words, and without any unnecessary visual clutter. This point in fact is so important that it really means that the presentation of information in a term bank which is going to be used on line at all must be designed for this purpose. If the presentation is acceptable on the screen, it should be completely acceptable on paper, but the reverse is not true.In order to give some idea of the practical considerations involved in consulting a term bank on line from a visual display unit I should like to describe my experience operating a terminal installed in the Council Secretariat in Brussels, and connected via a dedicated telephone line to "Eurodicautom", the Commission's terminology data base at the Computer Centre in Luxembourg.First of all, it is obvious that the technology at present being used for long-distance connections is not yet satisfactory, as there are fairly frequent disturbances and interruptions to the service for technical reasons. For example, during a recent two-hour session at the terminal, it was only possible to interrogate the term bank for about two thirds of the time during which the terminal was connected.The actual operation of the terminal is very simple and it only requires half an hour or so to grasp the mechanical operations involved, many of which are simplified by the provision of special keys for commanding various functions, such as asking a new question, or a decision to operate the truncation of the words requested -of which more lateror to have the associated printer print out the text appearing on the screen.What does require a little practice and -until an operating handbook is available -experimentation, is to discover the optimum way of putting questions in order to get the most helpful answer as quickly as possible. This is because the system is designed to give partial information in reply to a question when it does not contain an equivalent for the whole expression which has been requested, and the user can get bogged down in a mass of irrelevant answers.A question is put by typing on the keyboard the term or expression for which the correct equivalent in the target language is wanted. As the words are typed, they appear on the screen. When the operator has checked that the expression appearing on the screen is correct, he presses a special "enter" key to the right of the space bar and waits for the answer to come up on the screen. If the first answer is not completely satisfactory, further answers, each reproducing the content of a distinct entry in the "dictionary", can be called up by pressing the "entry" key again after each successive answer. When there are no more answers relating in any way to the question which has been put, a message to this effect appears on the screen.Articles or prepositions which appear in the "question" should not be typed, since the system neglects them unless, as is the case with the French preposition "de", confusion is possible (accents not being taken into account) with nouns. In such a case, typing a preposition can call up false answers, and so slow down the operation.On the principle of the longest match, the system willnormally give the correct answer to an expression containing three or four significant words as the first answer, if it contains the expression as such at all. If it does not, one should press the "truncation" key at once, because this will produce the answer if any word or words in the question were in the singular while they are in the plural in the expression recorded in the term bank or vice versa. Even with an expression containing only two significant words, dual or multiple meanings are rare, so that if the term bank contains the answer one is looking for, it will usually come up as the first one.The difficulty starts when one has entered an expression containing more than one significant word, for which the system has no exact match. In this case, in an effort to be helpful, it looks through its memory for any occurrence of any of the single words in the expression, and at present brings them out in an apparently random order, depending on the chronological order of their entry into the term bank.The same random plethora of information is liable to appear when one enters a question consisting of a single word, particularly if it is a common one. But perhaps one should not be asking Eurodicautom simple words?Be that as it may, I have found in practice that if the answer one wants does not appear as the first or second (after truncation) answer to the question, it is rarely worth continuing to press the "enter" key to obtain more than five answers. For this reason, and because it takes the printer one minute and five seconds to print a screen full of information, and it cannot be stopped at the end of the actual text on the screen, so that it may be "printing" empty space for half its time, I have designed a reply form which I use to note relevant information long-hand. This form is shown in Annex II.If one's answer comes up first time, I have found that one obtains it in between 15 and 45 seconds after starting to type the question. As this time includes typing, it obviously depends on the length of the question, and I am only a twofinger typist, so experienced operators will obviously be able to do better. To write out the relevant parts of five answers long-hand in completing one of the special reply forms takes an average of three minutes.Having spent some time in looking at how information can be obtained from a term bank -as this obviously affects translators who are using it -it will be as well to examine how information should be put into it, and by whom.It would be technically possible to allow any user who had access to a visual display unit with keyboard to add new material, or to amend what was already recorded. This is obviously undesirable, but it is equally undesirable to exclude users from contributing to the term bank at all, since the most fruitful way of running a term bank is to have a constant symbiosis or "osmosis" between users and the terminologists who are responsible for what goes in.The principle here must be that users are positively encouraged to submit proposals at all times, either for the translation of expressions which they have not found in the system, or because their experience tells them that their suggestions may be useful. Of course, these proposals must be vetted by the terminologists before they are entered, but this should be done within a fortnight of the proposal being submitted, as experience in systems operating in this way shows that translators want to be able to check that their proposals are in the system within this time, otherwise they become discouraged.Whichever method is used, speed in getting the results into the term bank is of the essence, particularly where one has a large number of translators working on important texts against urgent deadlines. The only acceptable method is now the use of keyboards keying directly into the memory, as in the commercially available text-processing systems. And if it is true, as I saw yesterday in someone else's newspaper, that it is now possible, in principle, to store half a million pages of text on a single memory disc, all of it immediately accessible, we shall have a very simple method of instantly amending and updating very large term banks.Organizations which have already set up term banks, or which are contemplating doing so, will have made their decisions for a variety of reasons, not all of which will be relevant to a freelance translator or a staff translator in a small firm. However, the advent of increasingly flexibletext-processing systems will mean that many small firms may find it worth using their typing equipment in order to set up a private tern bank on the side.What, though, is the market going to be for selling terminology from a term bank to independent "outside" translators, either freelance or staff? If anyone is contemplating doing this, he should do some hard market research first, because people are not going to keep on paying in order to find out, after dialling a term bank, that it doesn't contain the answer they want.I have emphasized dialling for information, i.e. interrogating a term bank on line via "Euronet" or "Viewdata" etc., because this is the only really new development in making information available, with the one prime advantage over the printed word that the information can be constantly up-dated without it being necessary to send subscribers looseleaf addenda or printed supplements to the main body of a glossary. Translators who buy the output from a term bank in the form of printed glossaries or micro-fiches will obviously judge it as they judge a dictionary. They will have paid for their information in advance, probably on the recommendation of colleagues or of professional publications. Their decision as to whether they have got their money's worth cannot cancel their original purchase; at best (or worst) it can only determine whether they place a repeat order or continue their subscription. I imagine that an outside subscriber dialling for instant information from a term bank would be charged for every call he put through, whether or not he found the answer to his question. And even if the service was free, he would not continue dialling if he did not obtain a high proportion of satisfactory answers.In addition to clear presentation of the information they contain, the second essential requirement for term banks designed to be used on line by translators is therefore that they give their users a sufficiently high ratio of satisfactory answers. This criterion applies both to in-house staff in a large organization and to outside subscribers. Possibly one group would accept a lower ratio of satisfactory answers than the other.This need to provide a high ratio of answers has led the managers of existing term banks to look at ways of exchanging information between term banks. "Eurodicautom" has been active in this area, and an ISO working party has been studying possible standards for the exchange of data on magnetic tape. Experience so far seems to indicate that the difficulties in the way of exchanging information are in the main not technical (incompatibility between computer programs and equipment), as was at first thought, but managerial, in the sense that differing term banks have different philosophies and different ways of presenting information, so that information from outside has first to be checked against what is already in the system, in order to prevent duplication, and then tailored to fit.There is a second drawback to the simple exchange of information between term banks in that it will, if carried to its logical conclusion, lead to the existence of several identical term banks all containing the same information. This would at least make it easier for the independent translator -he would simply dial his local term bank, instead of having to find out by trial and error which one gave him the best service.The logical solution is surely that term banks should continue to be set up wherever they meet a particular local need, and that all of them should pass on the terminology which they record to a central term bank for a particular geographical and/or linguistic area. These central term Consideration should also be given to presenting a series of "translators' packages" on the screen simultaneously, one below the other, so that the screen would read like a page in a well-designed glossary. Since experienced translators can very quickly scan a whole page of a glossary or word list, this form of presentation, avoiding the need to key in for successive entries which appear on the screen one at a time, would speed up the process of interrogation very considerably.If everyone operating a term bank, however small, were to use this standard format for presenting their information, allied with strict respect for technical standards for transferring information between term banks on magnetic tape, floppy discs, or other forms of memory yet to be developed, this would be a giant step towards the centralizing of terminology records for which I have already put in a plea. It would also mean that everyone would quickly learn to use information from any term bank, since the technique of interrogation would be the same for all of them.In this crystal-gazing exercise, I have concentrated on access via visual display units, but it seems to me that standardization of presentation would also have advantages for micro-fiches and printed glossaries. The layout of the latter could in any case be varied at will to meet particular requirements by the use of standard text-processing techniques as now applied to typed and printed documents.Having looked at machine translation and terminology data banks separately, with brief references to text-processing systems, I now want to sketch further possible developments based on such systems.In the first place, it has become evident during theSystran trials already carried out by the Commission of the European Communities that machine translation makes no sense unless it can be fitted into the normal production line for translations. As the obvious way of entering, pre-editing and post-editing machine translation texts is now to use a text-processing system, this has led to the realization that the whole production process for translations in the European Community institutions should be re-designed so as to make the maximum use of all the potentialities of large text-processing systems, whether or not machine translation as such is ever used on a routine basis or not."Controlled" situations From this realization it is a short step to the proposal which I now put forward for a new form of machine-aided translation which could give immense benefits in a large "controlled-translation" situation such as that existing in the European Community institutions. In the Community institutions a large number of linguists are employed to translate enormous amounts of written text, in a variety of original languages, into several languages simultaneously. In addition, and this is equally important, all these texts refer to a "controlled" situation, in that the field to which they relate, although very wide, is not infinite, and could in theory be precisely defined at any given moment. Finally, many of the texts involved are highly repetitive, frequently quoting whole passages from existing Community documents.If, as frequently happens, authors do not indicate the source for their quotations, it is easy to imagine how much time is quite unnecessarily wasted by translators in searching for references, or in re-translating texts which have already been translated.Many of these characteristics, if not all, will also be present in other international bodies, government departments and industrial and commercial undertakings. If such bodies are looking at the use of text-processing systems for handling their normal documentation and correspondence, they might also consider their potentialities for dealing, as follows, with their translation problems.The pre-requisite for implementing my proposal is that the text-processing system should have a large enough central memory store. If this is available, the proposal is simply that the organization in question should store all the texts it produces in the system's memory, together with their translations into however many languages are required. This information would have to be stored in such a way that any given portion of text in any of the languages involved can be located immediately, simply from the configuration of the words, without any intermediate coding, together with its translation into any or all of the other languages which the organization employs. This would mean that, simply by entering the final version of a text for printing, as prepared on the screen at the keyboard terminal, and indicating in which languages translations were required, the system would be instructed to compare the new text, probably sentence by sentence, with all the previously recorded texts prepared in the organization in that language, and to print out the nearest available equivalent for each sentence in all the target languages at the same time, on different printers.The result would be a complete text in the original language, plus at least partial translations in as many languages as were required, all grammatically correct as far as they went and all available simultaneously. Depending on how much of the new original was already in store, the subsequent work on the target language texts would range from the insertion of names and dates in standard letters, through light welding at the seams between discrete passages, to the translation of large passages of new text with the aid of a term bank based on the organization's past usage.When the completed translations were typed in the processing system, they would at the same time be entered in the text memory in association with the original, so that the store of translated texts would be automatically updated.The texts stored in this way could also be used as a source of "raw" terminology by calling up individual words or expressions on the screen, with their equivalents in other languages. Terminologists would check and process this information in order to enter it in a separate term bank memory in the internationally agreed format, but if a translator wanted a particular term before it was in the term bank, he could look it up in the text store.Since this form of machine-assisted translation would operate in the context of a complete text-processing system, it could very conveniently be supplemented by "genuine" machine translation, perhaps to translate the missing areas in texts retrieved from the text memory. Whether these mis-sing areas were translated by translators, or by a machine, the terminology used would have to be identical, and must be consistent with the normal terminology employed by the organization. This latter aspect of machine-aided translation has already cropped up in the European Communities, where I and others have been urging for some years now that the machine dictionaries used for the Systran trials should be consistent with the information contained in "Eurodicautom". Those working on these two projects in the Commission are well aware of this requirement, but the same type of considerations apply here as in the exchange of terms between term banks, with the added complication that a machine translation dictionary has to contain vastly more coded information than a term bank for translators or terminologists.Pulling all the scattered aspects of my paper together, what will it be like to work as a translator/reviser/posteditor in the computerized translation bureau or department of tomorrow? Do not forget either that, given reliable telecommunications, a freelance translator will be able to have all the facilities at home which his staff colleague will have at the office.My hunch is that our translator -in many cases, we ourselves -will continue to work at the same type of desk in the same type of office which he (or she) has to-day, with his standard dictionaries and reference works around him. Instead of a traditional type-writer, however, he will have a text-processing terminal with keyboard and screen so that he, or a secretary to whom he dictates, types his translations into the system memory so that they can be corrected on the screen before final "typing" on a separate printer which he will share with a number of colleagues, unless he is working as a lone freelance.If he has access to a local term bank, he will be able to interrogate it simply by typing his question on the keyboard of his text-processing terminal, when the answer will appear on the screen and can also be printed out by the printer. It will also be possible for him or his secretary to get a text-related glossary from the term bank, via the printer, by using the terminal to type questions into a buffer memory for batch processing.In a large organization using my proposed new system of machine-aided "translation by text-retrieval" (let's call it "TERRIER" -an appropriate name, since the Shorter Oxford Dictionary defines this word as "an inventory of property or goods" as well as "a small, active, intelligent variety of dog which pursues its quarry into its burrow or earth") our translator will be given, when he reports for duty, not only the original of the text he is required to "process", but TERRIER'S version of it in the target language, which we hope will be his mother tongue, both presented on paper in normal type-script.Secure in the knowledge that he does not have to do any research for possible hidden references, since TERRIER has done this for him, except for references to documents not already in the system, he will complete the target-language version of the text on paper, using his text-processing terminal to type any completely new passages. He will also use his terminal to get terminological information from the organization's term bank if necessary, either on line or in the form of a text-related glossary if he has enough time.He will then check the complete translation and pass it on, either for revision, if a separate revision stage is required, or straight for typing by a secretary into the textprocessing system for storage in the text-memory and printing out in whatever form is required.It would of course be technically possible to do all translating, editing and revision operations on the screen at the terminal, without printing the texts on paper at all, but I rather suspect that, except for extremely urgent or fairly simple texts, people will prefer to continue getting at least the final versions of their work onto paper so that they can carry out a final check, or so that a reviser can revise the text, with the good old-fashioned pen, pencil or ball-point, unharassed by modern technology.After all, translation is in the end a creative activity, not a mechanical chore. at the time of slaughter (l7th and 18th October, 1977) 39-40 days and 55-56 days respectively. They were White Plymouth Rocks. They have been given a feed with the following composition: 40% maize 20% barley 24% toasted soy bean cakes 3% meat and bone-meal 4% fish meal 3% oats 3% animal fat 3% minerals and vitamins All had been starved for at least 12 hours before slaughter. On each of the two days 3 times 25 chickens (for three weight groups) were caught and transported to the near-by slaughterhouse (distance app. 500 meter). Each weightgroup should contain 24 chickens -the 25th being an extra, that was slaughtered but not used in the study.Slaughtered chickens. All chickens were electrically stunned in a water stunner, killed and bleeded, but removed from the slaughter-line before the scalding tank. They were then handplucked and eviscerated. All organs and the neck were removed. They were dressed according to the definition: "plucked and drawn, without heads and feet, and without hearts, livers and gizzards, called 65% chickens", and with the removal of the neck.The carcasses were then divided in two groups (12 from each weight class) and numbered by application of rubber rings around the thighs, the same number being applied to the two thighs of the same chickens. Those destined for immediate deep-freezing were given a number preceded by a Roman I, those destined for wet chilling with a number preceded by a Roman II. Plusieurs choses sont complètement incompréhensibles. Ainsi, (3ème alinéa de la page 1) "sabot", "camion anglais", "meter" traduit par "compteur" (3ème alinéa de la page 2) etc... ou encore "Roman I" laissé tel quel.Si ,je devais apprécier cette traduction comme une épreuve de concours, ,1e me serais probablement arrêté au milieu de la première page en mettant un zéro au candidat.En tout cas, je me suis beaucoup amusé en lisant que "les poulets soumis à l'épreuve étaient capables de griller des gâteaux de haricots de soja" et que "le 25ème poulet n'était qu'un figurant"("an extra"), tandis que les 24 autres devaient être probablement des artistes, alors que c'est précisément la machine à traduire qui a manqué son numéro de trapèze et s'est écrasée au sol.Première conclusion qui s'impose d'emblée: non seulement cette traduction se situe bien en-dessous du seuil de rentabilité, mais on peut même la considérer comme franchement inutilisable.Cela dit, la machine en question n'était visiblement pas prête à ce genre d'expérience.(1) Son vocabulaire comporte des erreurs matérielles grossières. Je pense aux fautes d'orthographe; ex. : assomer ("assommer), abatoir (abbatoir).Elle manque d'informations : a) vocabulaire : elle ne connaît pas des termes élémentaires comme "bleeded", "hand-plucked", "deep-frozen". b) morphologie : ex. : "23rd novembre","étude pour les poulets", "déterminer de l'eau étrangère". c) syntaxe : ex. : "aider à la planification et en effectuant l'étude étaient également dr. P. Stevens"(3) Il aurait fallu lui apprendre à respecter les ensembles de mots en langue étrangère sans chercher à les interpréter comme s'il s'agissait de mots anglais. ex. : "van" qui est traduit par "camion" alors qu'ils'agit tout simplement de la préposition "de"; "Hoof" qui est traduit par "sabot" alors qu'il s'agit d'un nom propre; "für", qui devient "fourrure" alors que, là aussi, il s'agit d'une simple préposition.On aurait ainsi évité des assemblages de mots hétéroclites tels que "la fourrure Fleischforschung De Bundesanstalt". Ce dernier cas illustre du reste le manque de rigueur de la machine car, optiquement, "für" est différent de "fur". 4Ce qui m'amène à varier des bizarreries de la machine. Dans certains cas, la machine laisse un blanc. Ainsi, "deep-frozen" est traduit par "profond -". Dans d'autres, elle reproduit le mot tel quel; ex. : "bleeded", "hand-plucked". Pourquoi ? 5Si on passe maintenant à un stade de difficulté supérieur, on peut dire : a) que le stock de synonymes dont dispose la machine est insuffisant. Il semblerait que, pour chaque terme, elle ne connaisse qu'un seul équivalent. Ainsi, "implementation" = "exécution", "planning" = "planification", "removal" = "élimination". Il est évident que, dans ces conditions, la machine est vouée à commettre des faux sens . b) Remarque accessoire : s'il fallait à tout prix ne retenir qu'un seul équivalent, encore fallait-il choisir le plus courant et donner pour "live" l'équivalent "vivant" (au lieu de "vif", réservé à certaines tournures). Pour "neck", il fallait, c'est évident, donner "cou" (et non "col", d'un usage plus limité). c) Pour pouvoir fournir une traduction valable, la machine devrait, non seulement disposer d'un stock de synonymes suffisant, mais aussi apprendre à les choisir en fonction du contexte, ce qui, pour un ordinateur incapable de prendre des raccourcis, suppose sans doute toute une séquence d'opérations complexes. Autrement dit, il lui faudrait une mémoire beaucoup plus développée.(6) Passons sur certains "gags" très réussis comme ce "25th" poulet qui doit "être un figurant" ou alors cette autre formule "tout avait été affamé pendant 12 heures avant l'abattage", si drôle qu'elle confine à la poésie.Il reste certains résultats inexplicables tels le "camion anglais" dont, avec la meilleure volonté du monde, on ne parvient pas à retrouver l'origine dans "Laboratorium voor Hygiene en Technologie van Eetwaren van dierlijke Oorsprong". De même, on comprend mal comment "Station de Recherches Avicoles", en français dans l'original, a pu devenir "poste De Recherches Avicoles", après avoir transité par la machine à traduire. | null | null | null | null | Main paper:
the french text revised:
Première conclusion qui s'impose d'emblée: non seulement cette traduction se situe bien en-dessous du seuil de rentabilité, mais on peut même la considérer comme franchement inutilisable.Cela dit, la machine en question n'était visiblement pas prête à ce genre d'expérience.(1) Son vocabulaire comporte des erreurs matérielles grossières. Je pense aux fautes d'orthographe; ex. : assomer ("assommer), abatoir (abbatoir).Elle manque d'informations : a) vocabulaire : elle ne connaît pas des termes élémentaires comme "bleeded", "hand-plucked", "deep-frozen". b) morphologie : ex. : "23rd novembre","étude pour les poulets", "déterminer de l'eau étrangère". c) syntaxe : ex. : "aider à la planification et en effectuant l'étude étaient également dr. P. Stevens"(3) Il aurait fallu lui apprendre à respecter les ensembles de mots en langue étrangère sans chercher à les interpréter comme s'il s'agissait de mots anglais. ex. : "van" qui est traduit par "camion" alors qu'ils'agit tout simplement de la préposition "de"; "Hoof" qui est traduit par "sabot" alors qu'il s'agit d'un nom propre; "für", qui devient "fourrure" alors que, là aussi, il s'agit d'une simple préposition.On aurait ainsi évité des assemblages de mots hétéroclites tels que "la fourrure Fleischforschung De Bundesanstalt". Ce dernier cas illustre du reste le manque de rigueur de la machine car, optiquement, "für" est différent de "fur". 4Ce qui m'amène à varier des bizarreries de la machine. Dans certains cas, la machine laisse un blanc. Ainsi, "deep-frozen" est traduit par "profond -". Dans d'autres, elle reproduit le mot tel quel; ex. : "bleeded", "hand-plucked". Pourquoi ? 5Si on passe maintenant à un stade de difficulté supérieur, on peut dire : a) que le stock de synonymes dont dispose la machine est insuffisant. Il semblerait que, pour chaque terme, elle ne connaisse qu'un seul équivalent. Ainsi, "implementation" = "exécution", "planning" = "planification", "removal" = "élimination". Il est évident que, dans ces conditions, la machine est vouée à commettre des faux sens . b) Remarque accessoire : s'il fallait à tout prix ne retenir qu'un seul équivalent, encore fallait-il choisir le plus courant et donner pour "live" l'équivalent "vivant" (au lieu de "vif", réservé à certaines tournures). Pour "neck", il fallait, c'est évident, donner "cou" (et non "col", d'un usage plus limité). c) Pour pouvoir fournir une traduction valable, la machine devrait, non seulement disposer d'un stock de synonymes suffisant, mais aussi apprendre à les choisir en fonction du contexte, ce qui, pour un ordinateur incapable de prendre des raccourcis, suppose sans doute toute une séquence d'opérations complexes. Autrement dit, il lui faudrait une mémoire beaucoup plus développée.(6) Passons sur certains "gags" très réussis comme ce "25th" poulet qui doit "être un figurant" ou alors cette autre formule "tout avait été affamé pendant 12 heures avant l'abattage", si drôle qu'elle confine à la poésie.Il reste certains résultats inexplicables tels le "camion anglais" dont, avec la meilleure volonté du monde, on ne parvient pas à retrouver l'origine dans "Laboratorium voor Hygiene en Technologie van Eetwaren van dierlijke Oorsprong". De même, on comprend mal comment "Station de Recherches Avicoles", en français dans l'original, a pu devenir "poste De Recherches Avicoles", après avoir transité par la machine à traduire.
i have been asked to give a translator's viewpoint on translating and the computer, and i would like to emphasize straightaway that what i am going to say is exactly thatsimply a personal impression of the present situation and future developments. while i am fortunate in being able to follow what is going on as a representative of the council secretariat on the commission's "cetil" committee (comité d'experts pour le transfer d'information entre langues européennes) i am not speaking on behalf of the council secretariat to-day.:
Although I have only a short time available, I want to look at translating and the computer from two points of view. The first is that of a fairly large translating organization which is beginning to use a computerized terminology data base -Eurodicautom -and may become a user of machine translation. The second point of view is that of a translatorand being a staff translator myself I have had to try to put myself into a freelance translator's shoes as well, in order to get a complete picture.I am sure the first question which a translator asks about machine translation is "How will it affect my job?" The question was first asked in the 1950s as machine translation projects proliferated in the United States following a demonstration in 1954 by IBM and a research team at Georgetown University. By 1965 American government agencies are estimated to have spent some 20 million dollars in supporting machine translation research at 17 different institutions. And then the Automatic Language Processing Advisory Committee reported in 1966 that machine translation was slower, less accurate and twice as expensive as human translation and that there was no immediate or predictable prospect of useful machine translation.As Professor Sager has told us, the Commission has bought an American machine-translation system, "Systran", which it is developing in co-operation with the originator, Dr. Peter Toma, to translate texts from French into English and from English into French, and now from English into Italian. A very small number of translators have been assisting with computer programming in this connection, and with the preparation of dictionaries. Other linguists have been revising machine translations into French from English, in development trials. For them, machine translation has become a colleague. If it is true that "to understand is to forgive", this may explain why linguists who have been closely involved in developing the system have been more ready to revise its output than some others. At any rate, if there is a future for machine translation in the European Community institutions, it is obviously going to involve many linguists, either as programmers, lexicologists or revisersor perhaps "pre-editors".It is no part of my present responsibility to make an official assessment of the quality of the Systran translations already done from English into French at the Commission, but as a matter of interest I asked a number of experienced colleagues in our own French Division to evaluate one particular passage from three points of view. I asked one to read the French translation without having sight of the English original, and to tell me how much she understood. This corresponds roughly to the "intelligibility" criterion used in the Commission's own evaluation of Systran.(1) I asked a second colleague, the Head of our French Division, to "mark" the Systran output as if it had been submitted by a candidate in a competition for recruiting French translators.I asked a third colleague -a reviser -to revise the raw Systran output on the basis of the English original in exactly the same way as he would revise a translation made by one of our French translators in the normal course of his work.( 1)Georges Van Slype (1978): "Deuxième évaluation du système de traduction automatique SYSTRAN anglais-français de la Commission des Communautés européennes". Bureau Marcel van Dijk, Brussels.The last two checks correspond to the "fidelity" criterion also used in the Commission's evaluation of Systran, which most translators will instinctively regard as being the principal criterion applicable in judging machine translation, if not the only one.The original English text, the raw Systran French translation, and the French text as revised by our reviser are given in Annex I, together with my French colleagues' comments, in the form in which they were kind enough to note them down for me.Summarizing these comments, the raw translation considered on its own was felt to be quite inadequate for informing a French reader who did not know English about the purposes of the experiment which is described, or the procedure employed. All that he would grasp would be that an experiment with chickens had taken place.As for the translation considered as an entry in a competition to recruit French translators to the Council's staff, the Head of the French Division wrote that if he had had a paper like this to mark he would probably have stopped halfway down the first page, giving the candidate no marks at all.English original felt that the machine which had produced the translation has a memory which is far too rudimentary. He considered that evaluation of the system was premature and could not be conclusive because there was no way of assessing the results which might be achieved by a machine equipped with an adequate memory. This was all the more regrettable in that such results provided ammunition for the detractors of machine translation.Leaving this final conclusion with you, I want now to see how machine translation might affect the operations of the Secretariat of the Council of the European Communities. I must emphasize again that these remarks in no way represent any official position -they simply suggest themselves in the present situation and many if not all will be relevant in any firm or organization which is contemplating the possible use of machine translation.Provided that the quality of the final output was acceptable for the purpose in mind, the Council could have three reasons for using a machine translation system. These are, one, that it was cheaper than translation by the traditional translator-reviser system; two, that it was faster, or three, that there were not enough competent linguistsavailable to produce the necessary translations in any other way. I do not think there is any possibility of the Council Secretariat using raw machine translation for any of its texts, because the only quality we can accept is 100% fidelity to the meaning of the original, even though the style of our translations -as of many of the original textsoften suffers because of the very short deadlines against which we have to work.Leaving aside the possible use of machine translation because of a shortage of translators, our principle criterion would be whether we could produce accurate translations faster by using machine translation plus pre-editing and/or post-editing of texts, than we can at present with translators and revisers. In our particular circumstances -where texts have to be sent to the capitals of all the Member States ahead of meetings of the Council, COREPER (Permanent Representatives Committee) and countless working partiesspeed is of the essence ana cost is a secondary consideration. Which is not to say that a reduction in costs would not be welcome.With any given machine translation system, either Systran or the projected European system, we should need to analyse the types of text drafted in the Secretariat solely from the point of view of the total time taken to produce a 100% accurate translation from the original text. Since we could assume that the central processor time in the computer would be identical for every type of text, and could be neglected in comparison with human translation, this would amount to noting the time taken to pre-edit and/or post-edit different types of machine-translated text and comparing this with the total time taken to translate and revise texts of the same type in the normal way. Unless there were clear savings in time, from the moment the original text reached the Translation Department to the moment the completed translation was finally typed, we should not be interested in machine translation.One would in fact expect that texts which are structured at a superficial level, such as minutes of meetings, would be more amenable to machine translation under our conditions than speeches made by the President of the Council before the European Parliament. Other texts, such as the agendas for meetings, and even the implementing provisions of a Council Regulation or Decision as amended by a working party, can probably be dealt with more efficiently by an extended textprocessing system, than by machine translation as such.In fact, in the Council Secretariat we already employ "translation by photo-copy" to a considerable extent for such things as standard telexes, press releases, appointments to Committees, etc., where it is mainly a question of inserting names, dates, and the titles of documents in a standard format.It is at this point that I go beyond the brief I sketched for myself in the abstract of my paper which you will find in your programme. This is because it has become completely clear to me, since I started preparing for to-day's Seminar, that it is the advent of text-processing systems, not machine translation or even terminology data banks, which is the application of computers which is going to affect professional translators most directly -all of us, freelances and staff translators alike.Professor Sager and Mr. Tanke have both referred to text-processing systems already, so all I need do is to stress their immense flexibility, in that a single translator working on his own can derive many of the advantages which make a large integrated system so attractive for an organization like the Council or the Commission of the European Communities, with our hundreds of translators.the possibility of amending a text repeatedly on the display screen until it is ready for typing in its final format Extra advantages which the Council, or any large organization with an integrated text-processing system, can expect to derive, lie in the possibility of sending texts from one terminal to another for processing. For example, a text which was being amended during a meeting of the Council could be transmitted page by page to terminals in the various Translation Divisions. The translation of the original textcould be called up on the screen and amended, and the amended translation sent straight back to a terminal with its associated printer in a room next to the Council chamber, so that a complete new text in all required languages could be available by the end of the meeting.When we were asked at the Council some months ago to co-operate in an enquiry into post-editing systems for machine translation, I commented that our post-editing system consisted of a red ball-point pen in a reviser's hands. This may in fact continue to be true, since if working with a keyboard to revise translated texts at a visual display unit proves to be uncomfortable, it will be a simple matter to have the text printed out and given to a reviser for revision on paper in the traditional way. The corrections will then be made on the text-processing system by a secretary.This very brief sketch of the possible uses of textprocessing systems shows why all translators must consider their use, and also why machine translation systems and computerized terminology data banks must from now on be integrated into text-processing systems.Before I go on to deal with computerized terminology data banks, however, I would like to look at one or two more aspects of machine translation in general.In a multi-lingual situation such as the one we have in the European Communities, where it is often necessary to produce translations in parallel into several languages from a given original, it will obviously be an attractive proposition -until we get raw machine translation from free-text input which is of almost the same quality as that produced now by our translators -to concentrate on pre-editing texts for machine translation, rather than on post-editing the translations. A good job done on pre-editing a text will save post-editing several translations, and this is a point which those working on the new European machine translation system will presumably have in mind.If one adopts this approach, however, there will be a tendency to go still further back and to attempt to get the authors of texts to draft them in a standardized form which reduces the need for pre-editing. This is where we come up against resistance -we have already met it in the Council Secretariat when attempts have been made to encourage administrators to use standard formats so as to facilitate translation by photo-copy. And of course, this approach is just not on in the European Parliament or the Economic and Social Committee, where elected representatives of the people must obviously be free to express themselves just as they wish.If it were possible to dictate to people how they should write or speak, simply for the sake of making machine translation cheaper or easier, we could end up by making it more difficult for them to express themselves in their own language than it would be for them to learn a second language and use it.Finally, there is a real danger that the widespread use of machine translations which would not be stylistically acceptable if produced by a translator, even if they convey the message of the original, would debase and corrupt the natural languages now in use.On the other hand, it may be possible in some Community operations to replace natural language altogether by computerized information. Indeed, only the other day I was engaged in revising a proposal for a Council Regulation which contained the following clause: "The documents referred to in the preceding paragraph or elsewhere in this Regulation may be replaced by computerized information produced in any form for the same purpose".While translators working outside large firms or organizations are unlikely to come into direct contact with machine translation, and all translators ought to start looking at the use of text-processing machines or systems immediately, computerized terminology data banks fall between these two extremes. Their development and use have so far been restricted to large firms and organizations, but the impending introduction of publicly accessible data-transmission networks such as "Euronet" and systems such as "Teletext" and "Viewdata" which will use the domestic television set as a visual display unit, may mean that any staff or freelance translator will be able to dial for information from a term bank in the not-too-distant future.We translators may regard machine translation systems as competitors, and therefore fear them, but we instinctively feel more at home with something which is obviously not threatening, since all it can do is to help us in our work. Term banks must be "user-oriented", as became abundantly clear at a workshop on "Eurodicautom" which was held in Luxembourg last week, when we found it necessary to spend a con-siderable time discussing who was intended to use the system, and how, before we could look profitably at its content and structure.This question of intended use is paramount, since if it is not settled before a system is developed the resulting confusion may be disastrous. In addition to constituting an aid to translation, term banks can also be used for documentary purposes and for standardization -for example, for maintaining single-language normative dictionaries or as mono-lingual or multi-lingual thesauri for information retrieval systems. However, we are only concerned now with bilingual or multi-lingual term banks -we can also call them electronic dictionaries -specifically intended to assist translators in the same way as traditional dictionaries.he cannot. Firstly, he has a right to expect that the information given to him is clearly and logically presented, and can be read easily and quickly. This also applies to normal dictionaries, and is one of the principal criteria normally applied to such dictionaries. Secondly, he has a right to expect that the information given to him is reliable and accurate. However, he must himself decide on the value of this information and make his choice between alternative translations of a given expression, as he does with a normal dictionary.The basic difference between a printed dictionary and a term bank is that in the term bank all the information is stored electronically and can be added to, updated and amended at will at any time, and that any or all of the information which it contains at a given moment can be made available by a variety of means. It combines the advantages of centralization of information with de-centralization in making it available.The information in a term bank can be made available to the translator in three ways; on paper, in the form of a special subject glossary or a text-related glossary; via a television-type screen in a visual display unit used on line; or on micro-fiche used with a micro-fiche reader. The last two ways of looking up information are normally used to answer single queries arising in the course of translating a text, so that the translator will not need to make more than a mental note, or perhaps a hand-written note, of the answer. In both cases, however, it is possible to make a complete record of what appears on the screen, via a printer connected to the visual display unit or a photo-copier attached to the micro-fiche reader.At this point, it will be worth looking at the advantages and disadvantages of all three systems, both for an individual translator and for an organization using a term bank.At the Bundessprachenamt near Cologne, where some 250 linguists are engaged in translating largely technical texts for the West German Ministry of Defence, a computerized term bank has been in daily use for the last ten years. The philosophy there has always been to keep the translator away from the computer and to give him his information on paper, or on micro-fiche.The Bundessprachenamt's computer produces two basic types of glossary. The first is a special-purpose glossary, printed in a normal type-face, for use by several or many translators who are all working on a large long-term project, perhaps in several places at the same time. The second is a text-related glossary produced in the form of computer print-out for a specific text.In this second case, the translator underlines in his original text the terms he does not know, or on which he wants to check, and returns the text to the administrative office. Here a secretary types these terms into the computer which prints them out, with their equivalents in the target language, either in the order in which they appeared in the original, or in alphabetical order. This list is given to the translator, who in the meantime has been doing another job, a few hours later, or the next day.It is now the translators' responsibility, with the help of subject codes and other information printed out alongside the natural language equivalents, to choose whether the translation offered fits in the context of his text, and which of a number of equivalents does so, if he is offered a choice. If the computer offers no translation, or he is not satisfied with what it provides, the translator has to find the term he wants by other means open to all translators, such as looking up normal dictionaries and reference works, or asking colleagues.He notes on the computer print-out the new terms which he finds and uses, and these are then checked by a terminologist before being entered in the term bank for further use, within a fortnight at the latest.The great advantage of this system for the organization using it is that it gives constant direct feed-back from the translators to the system, so that the latest terms are being recorded all the time and made available to all translators. In practice, nothing like the same level of feedback is produced by the use of visual display units or microfiches.The advantage of the visual display unit used on line, both for the translator and for the organization employing him, is that he can immediately obtain the latest possible information in reply to a question which crops up while he is actually doing a translation. This is particularly important for a Translation Department in an organization like the Secretariat of the Council of the European Communities, where many documents have to be translated against very short deadlines. One can also envisage interpreters consulting such a visual display unit during a meeting, at least when they are working in pairs and one interpreter could interrogate the term bank while his colleague kept talking.Micro-fiche has the advantage that a very large number of terms can be stored in a very small space, that it is cheap to produce, and that it is practicable to distribute the up-dated contents of a term bank to a large number of users, both "in-house" and outside the organization, every six months or so. It would seem at first sight that this might after all be the cheapest and most practical way of distributing the contents of tern banks to freelance translators and to staff translators outside the organizations managing them.One important psychological factor in using visual display units and micro-fiche readers for presenting terminology to translators is that it is not as easy to absorb information from an illuminated screen as from the printed page. If a term bank is designed for use by either of these methods, it is vital that the information which the translator wants should be presented to him clearly in a minimum of words, and without any unnecessary visual clutter. This point in fact is so important that it really means that the presentation of information in a term bank which is going to be used on line at all must be designed for this purpose. If the presentation is acceptable on the screen, it should be completely acceptable on paper, but the reverse is not true.In order to give some idea of the practical considerations involved in consulting a term bank on line from a visual display unit I should like to describe my experience operating a terminal installed in the Council Secretariat in Brussels, and connected via a dedicated telephone line to "Eurodicautom", the Commission's terminology data base at the Computer Centre in Luxembourg.First of all, it is obvious that the technology at present being used for long-distance connections is not yet satisfactory, as there are fairly frequent disturbances and interruptions to the service for technical reasons. For example, during a recent two-hour session at the terminal, it was only possible to interrogate the term bank for about two thirds of the time during which the terminal was connected.The actual operation of the terminal is very simple and it only requires half an hour or so to grasp the mechanical operations involved, many of which are simplified by the provision of special keys for commanding various functions, such as asking a new question, or a decision to operate the truncation of the words requested -of which more lateror to have the associated printer print out the text appearing on the screen.What does require a little practice and -until an operating handbook is available -experimentation, is to discover the optimum way of putting questions in order to get the most helpful answer as quickly as possible. This is because the system is designed to give partial information in reply to a question when it does not contain an equivalent for the whole expression which has been requested, and the user can get bogged down in a mass of irrelevant answers.A question is put by typing on the keyboard the term or expression for which the correct equivalent in the target language is wanted. As the words are typed, they appear on the screen. When the operator has checked that the expression appearing on the screen is correct, he presses a special "enter" key to the right of the space bar and waits for the answer to come up on the screen. If the first answer is not completely satisfactory, further answers, each reproducing the content of a distinct entry in the "dictionary", can be called up by pressing the "entry" key again after each successive answer. When there are no more answers relating in any way to the question which has been put, a message to this effect appears on the screen.Articles or prepositions which appear in the "question" should not be typed, since the system neglects them unless, as is the case with the French preposition "de", confusion is possible (accents not being taken into account) with nouns. In such a case, typing a preposition can call up false answers, and so slow down the operation.On the principle of the longest match, the system willnormally give the correct answer to an expression containing three or four significant words as the first answer, if it contains the expression as such at all. If it does not, one should press the "truncation" key at once, because this will produce the answer if any word or words in the question were in the singular while they are in the plural in the expression recorded in the term bank or vice versa. Even with an expression containing only two significant words, dual or multiple meanings are rare, so that if the term bank contains the answer one is looking for, it will usually come up as the first one.The difficulty starts when one has entered an expression containing more than one significant word, for which the system has no exact match. In this case, in an effort to be helpful, it looks through its memory for any occurrence of any of the single words in the expression, and at present brings them out in an apparently random order, depending on the chronological order of their entry into the term bank.The same random plethora of information is liable to appear when one enters a question consisting of a single word, particularly if it is a common one. But perhaps one should not be asking Eurodicautom simple words?Be that as it may, I have found in practice that if the answer one wants does not appear as the first or second (after truncation) answer to the question, it is rarely worth continuing to press the "enter" key to obtain more than five answers. For this reason, and because it takes the printer one minute and five seconds to print a screen full of information, and it cannot be stopped at the end of the actual text on the screen, so that it may be "printing" empty space for half its time, I have designed a reply form which I use to note relevant information long-hand. This form is shown in Annex II.If one's answer comes up first time, I have found that one obtains it in between 15 and 45 seconds after starting to type the question. As this time includes typing, it obviously depends on the length of the question, and I am only a twofinger typist, so experienced operators will obviously be able to do better. To write out the relevant parts of five answers long-hand in completing one of the special reply forms takes an average of three minutes.Having spent some time in looking at how information can be obtained from a term bank -as this obviously affects translators who are using it -it will be as well to examine how information should be put into it, and by whom.It would be technically possible to allow any user who had access to a visual display unit with keyboard to add new material, or to amend what was already recorded. This is obviously undesirable, but it is equally undesirable to exclude users from contributing to the term bank at all, since the most fruitful way of running a term bank is to have a constant symbiosis or "osmosis" between users and the terminologists who are responsible for what goes in.The principle here must be that users are positively encouraged to submit proposals at all times, either for the translation of expressions which they have not found in the system, or because their experience tells them that their suggestions may be useful. Of course, these proposals must be vetted by the terminologists before they are entered, but this should be done within a fortnight of the proposal being submitted, as experience in systems operating in this way shows that translators want to be able to check that their proposals are in the system within this time, otherwise they become discouraged.Whichever method is used, speed in getting the results into the term bank is of the essence, particularly where one has a large number of translators working on important texts against urgent deadlines. The only acceptable method is now the use of keyboards keying directly into the memory, as in the commercially available text-processing systems. And if it is true, as I saw yesterday in someone else's newspaper, that it is now possible, in principle, to store half a million pages of text on a single memory disc, all of it immediately accessible, we shall have a very simple method of instantly amending and updating very large term banks.Organizations which have already set up term banks, or which are contemplating doing so, will have made their decisions for a variety of reasons, not all of which will be relevant to a freelance translator or a staff translator in a small firm. However, the advent of increasingly flexibletext-processing systems will mean that many small firms may find it worth using their typing equipment in order to set up a private tern bank on the side.What, though, is the market going to be for selling terminology from a term bank to independent "outside" translators, either freelance or staff? If anyone is contemplating doing this, he should do some hard market research first, because people are not going to keep on paying in order to find out, after dialling a term bank, that it doesn't contain the answer they want.I have emphasized dialling for information, i.e. interrogating a term bank on line via "Euronet" or "Viewdata" etc., because this is the only really new development in making information available, with the one prime advantage over the printed word that the information can be constantly up-dated without it being necessary to send subscribers looseleaf addenda or printed supplements to the main body of a glossary. Translators who buy the output from a term bank in the form of printed glossaries or micro-fiches will obviously judge it as they judge a dictionary. They will have paid for their information in advance, probably on the recommendation of colleagues or of professional publications. Their decision as to whether they have got their money's worth cannot cancel their original purchase; at best (or worst) it can only determine whether they place a repeat order or continue their subscription. I imagine that an outside subscriber dialling for instant information from a term bank would be charged for every call he put through, whether or not he found the answer to his question. And even if the service was free, he would not continue dialling if he did not obtain a high proportion of satisfactory answers.In addition to clear presentation of the information they contain, the second essential requirement for term banks designed to be used on line by translators is therefore that they give their users a sufficiently high ratio of satisfactory answers. This criterion applies both to in-house staff in a large organization and to outside subscribers. Possibly one group would accept a lower ratio of satisfactory answers than the other.This need to provide a high ratio of answers has led the managers of existing term banks to look at ways of exchanging information between term banks. "Eurodicautom" has been active in this area, and an ISO working party has been studying possible standards for the exchange of data on magnetic tape. Experience so far seems to indicate that the difficulties in the way of exchanging information are in the main not technical (incompatibility between computer programs and equipment), as was at first thought, but managerial, in the sense that differing term banks have different philosophies and different ways of presenting information, so that information from outside has first to be checked against what is already in the system, in order to prevent duplication, and then tailored to fit.There is a second drawback to the simple exchange of information between term banks in that it will, if carried to its logical conclusion, lead to the existence of several identical term banks all containing the same information. This would at least make it easier for the independent translator -he would simply dial his local term bank, instead of having to find out by trial and error which one gave him the best service.The logical solution is surely that term banks should continue to be set up wherever they meet a particular local need, and that all of them should pass on the terminology which they record to a central term bank for a particular geographical and/or linguistic area. These central term Consideration should also be given to presenting a series of "translators' packages" on the screen simultaneously, one below the other, so that the screen would read like a page in a well-designed glossary. Since experienced translators can very quickly scan a whole page of a glossary or word list, this form of presentation, avoiding the need to key in for successive entries which appear on the screen one at a time, would speed up the process of interrogation very considerably.If everyone operating a term bank, however small, were to use this standard format for presenting their information, allied with strict respect for technical standards for transferring information between term banks on magnetic tape, floppy discs, or other forms of memory yet to be developed, this would be a giant step towards the centralizing of terminology records for which I have already put in a plea. It would also mean that everyone would quickly learn to use information from any term bank, since the technique of interrogation would be the same for all of them.In this crystal-gazing exercise, I have concentrated on access via visual display units, but it seems to me that standardization of presentation would also have advantages for micro-fiches and printed glossaries. The layout of the latter could in any case be varied at will to meet particular requirements by the use of standard text-processing techniques as now applied to typed and printed documents.Having looked at machine translation and terminology data banks separately, with brief references to text-processing systems, I now want to sketch further possible developments based on such systems.In the first place, it has become evident during theSystran trials already carried out by the Commission of the European Communities that machine translation makes no sense unless it can be fitted into the normal production line for translations. As the obvious way of entering, pre-editing and post-editing machine translation texts is now to use a text-processing system, this has led to the realization that the whole production process for translations in the European Community institutions should be re-designed so as to make the maximum use of all the potentialities of large text-processing systems, whether or not machine translation as such is ever used on a routine basis or not."Controlled" situations From this realization it is a short step to the proposal which I now put forward for a new form of machine-aided translation which could give immense benefits in a large "controlled-translation" situation such as that existing in the European Community institutions. In the Community institutions a large number of linguists are employed to translate enormous amounts of written text, in a variety of original languages, into several languages simultaneously. In addition, and this is equally important, all these texts refer to a "controlled" situation, in that the field to which they relate, although very wide, is not infinite, and could in theory be precisely defined at any given moment. Finally, many of the texts involved are highly repetitive, frequently quoting whole passages from existing Community documents.If, as frequently happens, authors do not indicate the source for their quotations, it is easy to imagine how much time is quite unnecessarily wasted by translators in searching for references, or in re-translating texts which have already been translated.Many of these characteristics, if not all, will also be present in other international bodies, government departments and industrial and commercial undertakings. If such bodies are looking at the use of text-processing systems for handling their normal documentation and correspondence, they might also consider their potentialities for dealing, as follows, with their translation problems.The pre-requisite for implementing my proposal is that the text-processing system should have a large enough central memory store. If this is available, the proposal is simply that the organization in question should store all the texts it produces in the system's memory, together with their translations into however many languages are required. This information would have to be stored in such a way that any given portion of text in any of the languages involved can be located immediately, simply from the configuration of the words, without any intermediate coding, together with its translation into any or all of the other languages which the organization employs. This would mean that, simply by entering the final version of a text for printing, as prepared on the screen at the keyboard terminal, and indicating in which languages translations were required, the system would be instructed to compare the new text, probably sentence by sentence, with all the previously recorded texts prepared in the organization in that language, and to print out the nearest available equivalent for each sentence in all the target languages at the same time, on different printers.The result would be a complete text in the original language, plus at least partial translations in as many languages as were required, all grammatically correct as far as they went and all available simultaneously. Depending on how much of the new original was already in store, the subsequent work on the target language texts would range from the insertion of names and dates in standard letters, through light welding at the seams between discrete passages, to the translation of large passages of new text with the aid of a term bank based on the organization's past usage.When the completed translations were typed in the processing system, they would at the same time be entered in the text memory in association with the original, so that the store of translated texts would be automatically updated.The texts stored in this way could also be used as a source of "raw" terminology by calling up individual words or expressions on the screen, with their equivalents in other languages. Terminologists would check and process this information in order to enter it in a separate term bank memory in the internationally agreed format, but if a translator wanted a particular term before it was in the term bank, he could look it up in the text store.Since this form of machine-assisted translation would operate in the context of a complete text-processing system, it could very conveniently be supplemented by "genuine" machine translation, perhaps to translate the missing areas in texts retrieved from the text memory. Whether these mis-sing areas were translated by translators, or by a machine, the terminology used would have to be identical, and must be consistent with the normal terminology employed by the organization. This latter aspect of machine-aided translation has already cropped up in the European Communities, where I and others have been urging for some years now that the machine dictionaries used for the Systran trials should be consistent with the information contained in "Eurodicautom". Those working on these two projects in the Commission are well aware of this requirement, but the same type of considerations apply here as in the exchange of terms between term banks, with the added complication that a machine translation dictionary has to contain vastly more coded information than a term bank for translators or terminologists.Pulling all the scattered aspects of my paper together, what will it be like to work as a translator/reviser/posteditor in the computerized translation bureau or department of tomorrow? Do not forget either that, given reliable telecommunications, a freelance translator will be able to have all the facilities at home which his staff colleague will have at the office.My hunch is that our translator -in many cases, we ourselves -will continue to work at the same type of desk in the same type of office which he (or she) has to-day, with his standard dictionaries and reference works around him. Instead of a traditional type-writer, however, he will have a text-processing terminal with keyboard and screen so that he, or a secretary to whom he dictates, types his translations into the system memory so that they can be corrected on the screen before final "typing" on a separate printer which he will share with a number of colleagues, unless he is working as a lone freelance.If he has access to a local term bank, he will be able to interrogate it simply by typing his question on the keyboard of his text-processing terminal, when the answer will appear on the screen and can also be printed out by the printer. It will also be possible for him or his secretary to get a text-related glossary from the term bank, via the printer, by using the terminal to type questions into a buffer memory for batch processing.In a large organization using my proposed new system of machine-aided "translation by text-retrieval" (let's call it "TERRIER" -an appropriate name, since the Shorter Oxford Dictionary defines this word as "an inventory of property or goods" as well as "a small, active, intelligent variety of dog which pursues its quarry into its burrow or earth") our translator will be given, when he reports for duty, not only the original of the text he is required to "process", but TERRIER'S version of it in the target language, which we hope will be his mother tongue, both presented on paper in normal type-script.Secure in the knowledge that he does not have to do any research for possible hidden references, since TERRIER has done this for him, except for references to documents not already in the system, he will complete the target-language version of the text on paper, using his text-processing terminal to type any completely new passages. He will also use his terminal to get terminological information from the organization's term bank if necessary, either on line or in the form of a text-related glossary if he has enough time.He will then check the complete translation and pass it on, either for revision, if a separate revision stage is required, or straight for typing by a secretary into the textprocessing system for storage in the text-memory and printing out in whatever form is required.It would of course be technically possible to do all translating, editing and revision operations on the screen at the terminal, without printing the texts on paper at all, but I rather suspect that, except for extremely urgent or fairly simple texts, people will prefer to continue getting at least the final versions of their work onto paper so that they can carry out a final check, or so that a reviser can revise the text, with the good old-fashioned pen, pencil or ball-point, unharassed by modern technology.After all, translation is in the end a creative activity, not a mechanical chore. at the time of slaughter (l7th and 18th October, 1977) 39-40 days and 55-56 days respectively. They were White Plymouth Rocks. They have been given a feed with the following composition: 40% maize 20% barley 24% toasted soy bean cakes 3% meat and bone-meal 4% fish meal 3% oats 3% animal fat 3% minerals and vitamins All had been starved for at least 12 hours before slaughter. On each of the two days 3 times 25 chickens (for three weight groups) were caught and transported to the near-by slaughterhouse (distance app. 500 meter). Each weightgroup should contain 24 chickens -the 25th being an extra, that was slaughtered but not used in the study.Slaughtered chickens. All chickens were electrically stunned in a water stunner, killed and bleeded, but removed from the slaughter-line before the scalding tank. They were then handplucked and eviscerated. All organs and the neck were removed. They were dressed according to the definition: "plucked and drawn, without heads and feet, and without hearts, livers and gizzards, called 65% chickens", and with the removal of the neck.The carcasses were then divided in two groups (12 from each weight class) and numbered by application of rubber rings around the thighs, the same number being applied to the two thighs of the same chickens. Those destined for immediate deep-freezing were given a number preceded by a Roman I, those destined for wet chilling with a number preceded by a Roman II. Plusieurs choses sont complètement incompréhensibles. Ainsi, (3ème alinéa de la page 1) "sabot", "camion anglais", "meter" traduit par "compteur" (3ème alinéa de la page 2) etc... ou encore "Roman I" laissé tel quel.Si ,je devais apprécier cette traduction comme une épreuve de concours, ,1e me serais probablement arrêté au milieu de la première page en mettant un zéro au candidat.En tout cas, je me suis beaucoup amusé en lisant que "les poulets soumis à l'épreuve étaient capables de griller des gâteaux de haricots de soja" et que "le 25ème poulet n'était qu'un figurant"("an extra"), tandis que les 24 autres devaient être probablement des artistes, alors que c'est précisément la machine à traduire qui a manqué son numéro de trapèze et s'est écrasée au sol.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 555 | 0.079279 | null | null | null | null | null | null | null | null |
5b15504bfa99c23baed082c2440ec98219ff46f2 | 236999924 | null | The essential skills to be acquired for machine translation | It is exceedingly timely that there should be, at this seminar, a high-level scrutiny of the relations between "translator" and "machine": because, as we all know, there is also a world-wide expansion of the need to translate. Owing to improvements in telecommunications, the earth is becoming a global village; but in every house of the village, the inhabitants still speak each their different language, and this fact affects both individuals and corporations. In particular, the European Commission is confronted with an escalating translation problem; and that for an exceedingly honourable reason. For, whereas the historical solution for the "problem of Babel" was for there to be, on a naked imperialist basis, one "language of dominance" belonging to the nation which had conquered the other nations in the course of founding its empire -with all other languages, belonging to the nations which had been conquered, "languages of servitude" -the European Commission, by its articles, has created the new conception of "Linguistic equality". Every language of a member nation, within the Commission, is to be regarded as the equal of every other; and every important Commission document is to be issued in every Commission language with no one statement of it being regarded as a translation of any other. The establishment by law of this linguistic equality, on this scale, is something new; like the first step of the first astronaut on the moon, it is a "step forward" for the whole human race. However, such global steps forward tend to be both expensive, and also to require the development of new technology; and the step taken by the Commission is no exception to the general rule. For both it is the case that it makes the translation problem more urgent (there are 72 language pairs between which to translate): and it is also the case that, within a space of, say, two years, this step has caused a need for new technical skills; namely, the skills needed to enable the translator to be genuinely assisted by the machine, in order that a genuine man-machine combination shall enable the speed and range of reliable translation to increase. The current difficulty, however, is that the nature of these skills is not yet understood. The reason for this lack of comprehension is two-fold. The academic world, on the one hand, has not yet conceded that machine translation is emerging as a specialist discipline in its own right: for academics, "M.T." is still a, probably disreputable, off-shoot of linguistics: (the more so as this downgrading covers up the fact that linguistics, predominantly, examines only a limited corpus of unilingual material, and | {
"name": [
"Masterman, Margaret"
],
"affiliation": [
null
]
} | null | null | Translating and the Computer | 1978-11-01 | 10 | 3 | null | therefore lacks any adequate technique for examining the interrelations between two or more whole languages). The data-processing world, on the other hand, sees machine translation as just a special sort of data-processing: the fact that two whole natural languages, (at least) are involved in it, and in what, for data-processors, is a new and unique way -all this passes the systems-analysts by.The result of this double failure, within the existing specialities, to grasp the unique nature of the problem, has caused a bifurcation of the two interests involved, with machine translation itself falling down the gap between the two. It is not going to be until the human translator makes himself felt, not only as user but also as designer and as manual-writer, that further progress in obtaining a genuine man-machine translation interaction is going to be made. And therefore the translator must insist, forthwith, on coming into this new discipline, on an equality with the programmer and in a dual capacity, actively as well as passively: and, to this end, on being provided with the detailed knowhow which he requires.After what I have said, it will come as no surprise to learn that, since the nature of the problem is not recognised, the knowhow for solving it is also not there. Over the long run I think it is the academics who will bear the blame for this; because (it will be said) they both reacted too slowly to the pace of technological change, and also failed to observe the emergence of a new "hard science" of transforming meaning. But, in fact, the current situation is as much the responsibility of those who are too close to the technology as of those who, through other academic preoccupations, are too far away. 1 For those who are too close, the programmers, are predominantly thinking about the nature of language, and the nature of translation, only by writing actual machine translation programs, to which they then append notations intended only for fellow programmers: and so they see the material which they are handling, namely language, only through "the veil of the machine". 2 These programmers, in my view, are indeed bringing to light new facts about language about which academic linguists are going to have to take note, though without realising that they are doing so; if I did not think this, I would not think that M.T. was a discipline in its own right. But it is exceedingly difficult for the academic linguists to find out what these facts about language are, because, even when the programmers do express themselves in discursive prose they use phrases like "part of speech", "syntactic transformation", "multiple meaning", "dictionary structure" to refer back to characteristics of the M.T. programs which they themselves have written, and to nothing else. 3 Whereas the academics specify their use of all these same phrases by referring back to many and various literatures; those of general linguistics language-teaching, philosophy, mathematics, content analysis, psychology of language, Artificial Intelligence and computational linguistics, 4 but never to an acknowledged literature of Machine translation per se. So, within this highly multi-disciplinary academic world, quite apart from already existing difficulties of comprehension caused by the multi-disciplinary nature of linguistics itself, we now have a new way in which, when the systematic study of multilingualism or bilingualism is in question, two sets of people can unknowingly "talk through" one another: namely, by use of a whole range of terms relevant to translation, used by programmers to refer to M.T. programs, and by linguists to refer to academic specialist literatures. , where the machine assists the man to translate mathematical texts from Chinese into English, to a standard which enables these same translated texts to be acceptable without further change to specialist libraries.. The two strategies which have produced these two vary widely; you can opt for either, but you cannot pursue both at the same time, since whichever one you opt for will determine the whole subsequent trend of the design of your system. Contrary to expectation, also, it is the second type of system, rather than the first, which already shows signs of becoming interlingualised; as can be seen overall by looking at the flowchart reproduced in Appendix C (2).In the space left, I will now try to say a little more about the two types of clarification specified above, and about the human translating skills which their development needs.10 and a considerably sophisticated one without any explicit use of semantics 11 : but you cannot have a mechanical translation program which does not mechanically translate; and experience has shown that the essential mechanism for producing such translation is a 1-1 bi-lingual dictionary match operative between some source and target units of some kind.History has tended to obscure this fact. In the '60s for instance, wouldbe machine translation program designers used to consider it sufficient to output many possible variants in the output language of each word in the input language. These multi-outputting M.T. programs were interesting comments on the translation-relation, but, as experience showed, the outputs were not translations: those who were meant to benefit by them could not use them. It was not until Peter Toma, in SYSTRAN, (following in this matter on Gilbert King of I.B.M.) succeeded in causing a machine to output one, and only one, coherent output text from each input text, and moreover (unlike the output from King's system, which used no syntax) an output which was very much more often right than wrong, that Machine Translation came of age as a technology to be reckoned with in its own right (i.e. as a skill to be distinguished from other skills).However, if the essential and central "brute-force" mechanism of M.T. is conceded to be a 1-1 dictionary match, two fundamental questions immediately arise. The first is: if crude M.T. is a 1-1 dictionary match, how do we sophisticate it? The second is: if M.A moment's reflection will show any human translator that indeed a 1-1 dictionary match can quite easily be sophisticated. To make the match, the machine has first to be given the boundaries of the unit of text to be matched; which is why the easiest form of match is word for word, since in this match the machine can "bound the match" by the spaces on each side of the word. But the match can be extended to match longer stretches of text than a word; it can be truncated to match shorter stretches of text than a word. It can be made conditional i.e. made to change consequent on the result of one or more tests, and this is where syntactic testing comes in . And it can be transformed from a 1-stage match to a 2-stage match or an N-stage match by a match, say, into a neutral particularised, context-sensitive programs to produce high-level translations of particular stretches of the text, and then re-running the program so that the input text can benefit from its specially oriented dictionary. It is sometimes thought, by sponsors such as Pankowicz 12 , that this process of orienting dictionary to text is illegitimate. I cannot see why. The object of having a machine to produce translation, after all, is not (as with chess) to take part in international M.T. competitions, but to produce usable translations. If this is achieved by putting money and effort into teaching already trained translators to program particularised dictionary entries in a MACRO-language (and even more important, to use their trained judgment to choose which such entries to program) not only are they taking steps, at one remove, to supplement the machine's low-grade skill by their high-grade skill, they are doing something more, which is very interesting: namely, producing a machine-readable bi-lingual data-base which is contextsensitive (something new in linguistics). And it is subsequent examination of this which may quite possibly enable us to make explicit facts which are at present only subliminally known about the translation relation itself.In Appendix B, this process can be seen going on. For first (in B 1 ) we see the authorised E.E.C. translations of a set of phrases, produced by trained human translators; then (in B 2 ) the raw SYSTRAN output for the same phrases; and lastly (in B 3 ) a sample of the many additional dictionary entries required to make B 2 approximate more nearly to B 1.The second method b), where M.T. is used online and with pre-and posteditors, has already shown that it can produce output of much higher quality than that of batch-programmed M.T.; as can be seen by looking at Appendix A 2. But this second method can be cost-effective, that is it can pay for itself, only if one of two background situations obtain. The first of these is that, by using the machine online, knowledge is made internationally accessible which would not be accessible otherwise; for instance, by translating specialist mathematical papers from Russian, Greek, Arabic or Hindi, where the nature of the script, let alone of the language, constitutes a "knowledge-barrier" which scientists just cannot pass. The second background situation which justifies the expense of online M.T. is the interlingual one. The Translation Institute at Brigham Young University, Utah, has the aim of translating the Mormon texts online into 500 of the world's languages, 13 and this, I think, is a particular foretaste of more general things to come. Moreover, this type of program also requires a new highlevel translation skill: namely that of pre-editing an input text by inserting into it cardinal structural features of the output language in machine-readable form. This is no mean feat as any translator, even a highly-trained one, will find if he or she will make the effort to try it. And here is the potential of it. It is easier to learn, once for all, to mark in, on the input, "neutral" structural features which can then be used to synthesise ANY output language than it is to keep adjusting and varying your conception of what has to be pre-edited in as you keep on altering your target language . And again, as in the first case, that of programming particularised dictionary-entries, there is a research potential implicit in developing this interlingual skill, which is that of bringing to bear the trained intuitive skill of translators, to help us discover more about what a cardinal structural notation, neutral as between N output languages, really is. ................ In conclusion, in order to gather together the rather scattered argumentation of this paper, I will list the basic skills which the human translator needs, if he is to participate, on an equality with the programmer, in the development of man-machine interaction for translation. The need for these skills, in the form in which I will now give them, emerges from consideration of the two basic clarifications which I suggested earlier; but I do not think that I have as yet made sufficiently clear what I think they are.Firstly, the translator has to acquire the ability to see translation as a mechanical process sophisticating itself from a basic 1-1 bi-lingual match: namely, of the simplest case which it is possible to imagine of the translation-relation. This skill requires the further capacity, both to assign boundaries and shapes to translation units in any language, and also the classifying ability to assign to these units, once found, markers which will specify the nature and degree of any translation unit's "idiomaticity"that is, the way and the extent to which it differs from the basic 1-1 match. This classifying effort is cardinal to M.T. since a corresponding type of classification has to be made, in each case, of a type of dictionary, each type of dictionary has to handle a specific type of "idiomaticness"as can be seen by looking, yet once again, at the flowchart in Appendix C 1.This first skill is no mean skill in itself; but the translators could acquire it.Secondly, the translator must learn to recognise classes of awkward translation-situations -"knotty problems" -which will require special dictionary entries to solve them. He must then become able to write flowcharts of these dictionary-entries; it will only take him about an hour to do this last, since the comments on the flowcharts can be made in his own words; the programmers will then be pleased to turn them into patterns of coding. But recognising the awkward translation-situations: there lies the skill.When a special type of awkward translation situation keeps on recurring (as occurs, for instance, when English past participles have to be distinguished from English past tenses of finite verbs) then the flowcharts dealing with this phenomenon cease to be only those of individual dictionary-entries and become general syntactic disambiguation routines (called in SYSTRAN "homograph routines").Once all the awkward translation-situations have been identified and solved syntactic analysis for machine translation -which also, incidentally, becomes very abstract -reduces to almost nothing; whereas if the knotty translation-problems have not first been identified and solved, syntactic analysis by machine cannot be done at all. . | null | null | null | null | Main paper:
:
therefore lacks any adequate technique for examining the interrelations between two or more whole languages). The data-processing world, on the other hand, sees machine translation as just a special sort of data-processing: the fact that two whole natural languages, (at least) are involved in it, and in what, for data-processors, is a new and unique way -all this passes the systems-analysts by.The result of this double failure, within the existing specialities, to grasp the unique nature of the problem, has caused a bifurcation of the two interests involved, with machine translation itself falling down the gap between the two. It is not going to be until the human translator makes himself felt, not only as user but also as designer and as manual-writer, that further progress in obtaining a genuine man-machine translation interaction is going to be made. And therefore the translator must insist, forthwith, on coming into this new discipline, on an equality with the programmer and in a dual capacity, actively as well as passively: and, to this end, on being provided with the detailed knowhow which he requires.After what I have said, it will come as no surprise to learn that, since the nature of the problem is not recognised, the knowhow for solving it is also not there. Over the long run I think it is the academics who will bear the blame for this; because (it will be said) they both reacted too slowly to the pace of technological change, and also failed to observe the emergence of a new "hard science" of transforming meaning. But, in fact, the current situation is as much the responsibility of those who are too close to the technology as of those who, through other academic preoccupations, are too far away. 1 For those who are too close, the programmers, are predominantly thinking about the nature of language, and the nature of translation, only by writing actual machine translation programs, to which they then append notations intended only for fellow programmers: and so they see the material which they are handling, namely language, only through "the veil of the machine". 2 These programmers, in my view, are indeed bringing to light new facts about language about which academic linguists are going to have to take note, though without realising that they are doing so; if I did not think this, I would not think that M.T. was a discipline in its own right. But it is exceedingly difficult for the academic linguists to find out what these facts about language are, because, even when the programmers do express themselves in discursive prose they use phrases like "part of speech", "syntactic transformation", "multiple meaning", "dictionary structure" to refer back to characteristics of the M.T. programs which they themselves have written, and to nothing else. 3 Whereas the academics specify their use of all these same phrases by referring back to many and various literatures; those of general linguistics language-teaching, philosophy, mathematics, content analysis, psychology of language, Artificial Intelligence and computational linguistics, 4 but never to an acknowledged literature of Machine translation per se. So, within this highly multi-disciplinary academic world, quite apart from already existing difficulties of comprehension caused by the multi-disciplinary nature of linguistics itself, we now have a new way in which, when the systematic study of multilingualism or bilingualism is in question, two sets of people can unknowingly "talk through" one another: namely, by use of a whole range of terms relevant to translation, used by programmers to refer to M.T. programs, and by linguists to refer to academic specialist literatures. , where the machine assists the man to translate mathematical texts from Chinese into English, to a standard which enables these same translated texts to be acceptable without further change to specialist libraries.. The two strategies which have produced these two vary widely; you can opt for either, but you cannot pursue both at the same time, since whichever one you opt for will determine the whole subsequent trend of the design of your system. Contrary to expectation, also, it is the second type of system, rather than the first, which already shows signs of becoming interlingualised; as can be seen overall by looking at the flowchart reproduced in Appendix C (2).In the space left, I will now try to say a little more about the two types of clarification specified above, and about the human translating skills which their development needs.10 and a considerably sophisticated one without any explicit use of semantics 11 : but you cannot have a mechanical translation program which does not mechanically translate; and experience has shown that the essential mechanism for producing such translation is a 1-1 bi-lingual dictionary match operative between some source and target units of some kind.History has tended to obscure this fact. In the '60s for instance, wouldbe machine translation program designers used to consider it sufficient to output many possible variants in the output language of each word in the input language. These multi-outputting M.T. programs were interesting comments on the translation-relation, but, as experience showed, the outputs were not translations: those who were meant to benefit by them could not use them. It was not until Peter Toma, in SYSTRAN, (following in this matter on Gilbert King of I.B.M.) succeeded in causing a machine to output one, and only one, coherent output text from each input text, and moreover (unlike the output from King's system, which used no syntax) an output which was very much more often right than wrong, that Machine Translation came of age as a technology to be reckoned with in its own right (i.e. as a skill to be distinguished from other skills).However, if the essential and central "brute-force" mechanism of M.T. is conceded to be a 1-1 dictionary match, two fundamental questions immediately arise. The first is: if crude M.T. is a 1-1 dictionary match, how do we sophisticate it? The second is: if M.A moment's reflection will show any human translator that indeed a 1-1 dictionary match can quite easily be sophisticated. To make the match, the machine has first to be given the boundaries of the unit of text to be matched; which is why the easiest form of match is word for word, since in this match the machine can "bound the match" by the spaces on each side of the word. But the match can be extended to match longer stretches of text than a word; it can be truncated to match shorter stretches of text than a word. It can be made conditional i.e. made to change consequent on the result of one or more tests, and this is where syntactic testing comes in . And it can be transformed from a 1-stage match to a 2-stage match or an N-stage match by a match, say, into a neutral particularised, context-sensitive programs to produce high-level translations of particular stretches of the text, and then re-running the program so that the input text can benefit from its specially oriented dictionary. It is sometimes thought, by sponsors such as Pankowicz 12 , that this process of orienting dictionary to text is illegitimate. I cannot see why. The object of having a machine to produce translation, after all, is not (as with chess) to take part in international M.T. competitions, but to produce usable translations. If this is achieved by putting money and effort into teaching already trained translators to program particularised dictionary entries in a MACRO-language (and even more important, to use their trained judgment to choose which such entries to program) not only are they taking steps, at one remove, to supplement the machine's low-grade skill by their high-grade skill, they are doing something more, which is very interesting: namely, producing a machine-readable bi-lingual data-base which is contextsensitive (something new in linguistics). And it is subsequent examination of this which may quite possibly enable us to make explicit facts which are at present only subliminally known about the translation relation itself.In Appendix B, this process can be seen going on. For first (in B 1 ) we see the authorised E.E.C. translations of a set of phrases, produced by trained human translators; then (in B 2 ) the raw SYSTRAN output for the same phrases; and lastly (in B 3 ) a sample of the many additional dictionary entries required to make B 2 approximate more nearly to B 1.The second method b), where M.T. is used online and with pre-and posteditors, has already shown that it can produce output of much higher quality than that of batch-programmed M.T.; as can be seen by looking at Appendix A 2. But this second method can be cost-effective, that is it can pay for itself, only if one of two background situations obtain. The first of these is that, by using the machine online, knowledge is made internationally accessible which would not be accessible otherwise; for instance, by translating specialist mathematical papers from Russian, Greek, Arabic or Hindi, where the nature of the script, let alone of the language, constitutes a "knowledge-barrier" which scientists just cannot pass. The second background situation which justifies the expense of online M.T. is the interlingual one. The Translation Institute at Brigham Young University, Utah, has the aim of translating the Mormon texts online into 500 of the world's languages, 13 and this, I think, is a particular foretaste of more general things to come. Moreover, this type of program also requires a new highlevel translation skill: namely that of pre-editing an input text by inserting into it cardinal structural features of the output language in machine-readable form. This is no mean feat as any translator, even a highly-trained one, will find if he or she will make the effort to try it. And here is the potential of it. It is easier to learn, once for all, to mark in, on the input, "neutral" structural features which can then be used to synthesise ANY output language than it is to keep adjusting and varying your conception of what has to be pre-edited in as you keep on altering your target language . And again, as in the first case, that of programming particularised dictionary-entries, there is a research potential implicit in developing this interlingual skill, which is that of bringing to bear the trained intuitive skill of translators, to help us discover more about what a cardinal structural notation, neutral as between N output languages, really is. ................ In conclusion, in order to gather together the rather scattered argumentation of this paper, I will list the basic skills which the human translator needs, if he is to participate, on an equality with the programmer, in the development of man-machine interaction for translation. The need for these skills, in the form in which I will now give them, emerges from consideration of the two basic clarifications which I suggested earlier; but I do not think that I have as yet made sufficiently clear what I think they are.Firstly, the translator has to acquire the ability to see translation as a mechanical process sophisticating itself from a basic 1-1 bi-lingual match: namely, of the simplest case which it is possible to imagine of the translation-relation. This skill requires the further capacity, both to assign boundaries and shapes to translation units in any language, and also the classifying ability to assign to these units, once found, markers which will specify the nature and degree of any translation unit's "idiomaticity"that is, the way and the extent to which it differs from the basic 1-1 match. This classifying effort is cardinal to M.T. since a corresponding type of classification has to be made, in each case, of a type of dictionary, each type of dictionary has to handle a specific type of "idiomaticness"as can be seen by looking, yet once again, at the flowchart in Appendix C 1.This first skill is no mean skill in itself; but the translators could acquire it.Secondly, the translator must learn to recognise classes of awkward translation-situations -"knotty problems" -which will require special dictionary entries to solve them. He must then become able to write flowcharts of these dictionary-entries; it will only take him about an hour to do this last, since the comments on the flowcharts can be made in his own words; the programmers will then be pleased to turn them into patterns of coding. But recognising the awkward translation-situations: there lies the skill.When a special type of awkward translation situation keeps on recurring (as occurs, for instance, when English past participles have to be distinguished from English past tenses of finite verbs) then the flowcharts dealing with this phenomenon cease to be only those of individual dictionary-entries and become general syntactic disambiguation routines (called in SYSTRAN "homograph routines").Once all the awkward translation-situations have been identified and solved syntactic analysis for machine translation -which also, incidentally, becomes very abstract -reduces to almost nothing; whereas if the knotty translation-problems have not first been identified and solved, syntactic analysis by machine cannot be done at all. .
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 555 | 0.005405 | null | null | null | null | null | null | null | null |
7df9e89e7e3be14d3a883170f741448f0b24e281 | 61449780 | null | An interactive on-line machine translation system ({C}hinese into {E}nglish) | The present on-line system is a direct conversion of the CULT (batch) machine translation system which has been used since January 1975 on a regular basis in translating Chinese mathematics journals. The pre-editing procedures used in CULT are being implemented in the present on-line system by means of editing programs. The enormous problem of inputting Chinese texts is being solved by keying the text directly with a newly designed Chinese keyboard. | {
"name": [
"Loh, Shiu-Chang and",
"Kong, Luan"
],
"affiliation": [
null,
null
]
} | null | null | Translating and the Computer | 1978-11-01 | 3 | 7 | null | This paper describes briefly an interactive on-line machine translation system capable of translating Chinese into readable English, which has recently been implemented at the Chinese University of Hong Kong.The design of this on-line system is mainly based on the algorithm developed for the much used CULT (Chinese University Language Translator) system which was first developed with the IBM 1130 system in 1969 and then implemented on the ICL 1904A system in 1972 (Loh 1972 . CULT has been used, on a regular basis, to translate the Chinese mathematics journal, Acta Mathematica Sinica, since 1975 . The capability of CULT has also been further developed to assist translation in both directions, that is CULT is now capable of supporting translation from Chinese into readable English as well as from English into readable Chinese (Loh, Hung and Kong 1977) .All the versions of CULT mentioned above are in batch mode. One of the major problems of this mode of translation is the problem of inputting Chinese characters. Firstly, the Chinese text was manually encoded into four-digit standard telegraph codes and punched onto cards before being inputted into the computer for translation. This process is enormous, tedious and liable to errors. In order to solve this problem, a research project was carried out at the Chinese University of Hong Kong to try to find a better solution to this problem. As a result, a Chinese key-board was designed.The newly constructed Chinese key-board makes the design of the present online system possible. The use of this key-board for direct input has eliminated many of the errors arising in coding Chinese characters either by means of numerical digits or alphanumerics, thus facilitating the translation process.The present on-line system is primarily based on the CULT (batch mode) system just mentioned, except that the pre-editing procedures can be replaced by the interactive operations of the Chinese key-board, and also the editing facility is provided.The flowchart of the general translation procedures is given in Figure 1 , similar to CULT (batch mode) system. Figures 2, 3, 4, 5, 6, 7, 8 & 9 give the detailed flowcharts of (i) Source text preparation, (ii) Input, (iii) Lexical Analysis, (iv) Syntactic & Semantic Analysis, (v) Relative order analysis, (vi) Target equivalence analysis, (vii) Output and (viii) Output refinement respectively.The editing facilities for on-line up-dating of dictionary records, i.e. correction, insertion, deletion etc. are provided at appropriate points. This enables the operation to interact with the translation system at all times, while translating. Programs for the system are written in Standard FORTRAN and run on the PDP 11/34 computer system.Computer Configuration A PDP 11/34 computer system has recently been installed at the Hung On-To Research Centre for Machine Translation, to be used exclusively for the Machine Translation Project. The configuration of the system is given in Figure 10 . The Chinese key-board is attached in addition to the standard alphanumeric key-board which is to be used in the implementation of On-line Dual Language Translator in due course. When implemented, the system should be capable of translating Chinese into readable English as well as from English into readable Chinese simultaneously.The Chinese Key-board Many Chinese input systems based either on Corner Shapes of the characters, or on Pin Yin (pronunciation), or on a Chinese typewriter key-board with a few thousand keys are commercially available. The widely publicized cylindrical Chinese typewriter key-board which has been developed by the research workers of the Chinese Language Project, Cambridge University, is the latest one to be introduced to the market.None of these input systems mentioned possesses the most desirable characteristic of being easy to learn and operate. Most, if not all, require extensive training of the operator. For simplicity, the input of Chinese characters to the computer system should follow the natural order of how they are written by hand -the way children are taught to write Chinese character at school. This kind of stroke-approach, though theoretically sound, is impossible to implement due to the richness of the language and the infinite combination of strokes that go to make up Chinese characters. However, a feasible way is to employ the "radical" approach, in other words, to represent on the keyboard the most frequently used radicals of the Chinese characters.For the past few years, the structures of the Chinese characters have been studied and a new set of modified radicals which total barely over 200 in number, has been selected. A key-board based on this design approach has been constructed and initial test results are satisfactory. The arrangement of the keys on the key-board is usually according to their normal position in the character, that is, a radical, if normally at the top of the character is placed on the upper portion of the key-board and so on.The main advantage of this approach is that it is simple and natural for any person with a minimal knowledge of written Chinese to operate the key-board with ease. | null | null | null | null | Main paper:
introduction:
This paper describes briefly an interactive on-line machine translation system capable of translating Chinese into readable English, which has recently been implemented at the Chinese University of Hong Kong.The design of this on-line system is mainly based on the algorithm developed for the much used CULT (Chinese University Language Translator) system which was first developed with the IBM 1130 system in 1969 and then implemented on the ICL 1904A system in 1972 (Loh 1972 . CULT has been used, on a regular basis, to translate the Chinese mathematics journal, Acta Mathematica Sinica, since 1975 . The capability of CULT has also been further developed to assist translation in both directions, that is CULT is now capable of supporting translation from Chinese into readable English as well as from English into readable Chinese (Loh, Hung and Kong 1977) .All the versions of CULT mentioned above are in batch mode. One of the major problems of this mode of translation is the problem of inputting Chinese characters. Firstly, the Chinese text was manually encoded into four-digit standard telegraph codes and punched onto cards before being inputted into the computer for translation. This process is enormous, tedious and liable to errors. In order to solve this problem, a research project was carried out at the Chinese University of Hong Kong to try to find a better solution to this problem. As a result, a Chinese key-board was designed.The newly constructed Chinese key-board makes the design of the present online system possible. The use of this key-board for direct input has eliminated many of the errors arising in coding Chinese characters either by means of numerical digits or alphanumerics, thus facilitating the translation process.The present on-line system is primarily based on the CULT (batch mode) system just mentioned, except that the pre-editing procedures can be replaced by the interactive operations of the Chinese key-board, and also the editing facility is provided.The flowchart of the general translation procedures is given in Figure 1 , similar to CULT (batch mode) system. Figures 2, 3, 4, 5, 6, 7, 8 & 9 give the detailed flowcharts of (i) Source text preparation, (ii) Input, (iii) Lexical Analysis, (iv) Syntactic & Semantic Analysis, (v) Relative order analysis, (vi) Target equivalence analysis, (vii) Output and (viii) Output refinement respectively.The editing facilities for on-line up-dating of dictionary records, i.e. correction, insertion, deletion etc. are provided at appropriate points. This enables the operation to interact with the translation system at all times, while translating. Programs for the system are written in Standard FORTRAN and run on the PDP 11/34 computer system.Computer Configuration A PDP 11/34 computer system has recently been installed at the Hung On-To Research Centre for Machine Translation, to be used exclusively for the Machine Translation Project. The configuration of the system is given in Figure 10 . The Chinese key-board is attached in addition to the standard alphanumeric key-board which is to be used in the implementation of On-line Dual Language Translator in due course. When implemented, the system should be capable of translating Chinese into readable English as well as from English into readable Chinese simultaneously.The Chinese Key-board Many Chinese input systems based either on Corner Shapes of the characters, or on Pin Yin (pronunciation), or on a Chinese typewriter key-board with a few thousand keys are commercially available. The widely publicized cylindrical Chinese typewriter key-board which has been developed by the research workers of the Chinese Language Project, Cambridge University, is the latest one to be introduced to the market.None of these input systems mentioned possesses the most desirable characteristic of being easy to learn and operate. Most, if not all, require extensive training of the operator. For simplicity, the input of Chinese characters to the computer system should follow the natural order of how they are written by hand -the way children are taught to write Chinese character at school. This kind of stroke-approach, though theoretically sound, is impossible to implement due to the richness of the language and the infinite combination of strokes that go to make up Chinese characters. However, a feasible way is to employ the "radical" approach, in other words, to represent on the keyboard the most frequently used radicals of the Chinese characters.For the past few years, the structures of the Chinese characters have been studied and a new set of modified radicals which total barely over 200 in number, has been selected. A key-board based on this design approach has been constructed and initial test results are satisfactory. The arrangement of the keys on the key-board is usually according to their normal position in the character, that is, a radical, if normally at the top of the character is placed on the upper portion of the key-board and so on.The main advantage of this approach is that it is simple and natural for any person with a minimal knowledge of written Chinese to operate the key-board with ease.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 555 | 0.012613 | null | null | null | null | null | null | null | null |
45be00df90369c30a6ef653b89fe2b03324cecb6 | 219309378 | null | Properties of Lexical Relations [Appendix {II} of {``}A Lexicon for a Computer Question-Answering System,{''} {AJCL} Microfiche 83] | Certain properties of l e x i c a l -s e w t i c relations can b e very useful i n deductive inference. For instance, 15 we know that a cheetah is a ki,na or mammal anu a mamm i.s a kind of vertebrate then we can deduce that a cheetah i s a kind of vertebrate. Writing T f o r the taxonomy r e l a t i o n , w e can abbreviate this sentence: if cheetah T mammal and mammal T vertebrate then cheetah T vertebrate. Whenever bTc and cTd, it follcaws that bTd. This fact ran be described much more effi ciently by the stuement that the taxonomy r e l a t i o n i s transitive. Two other commonly menttand properties of relations are refilexivity and syrmnetry. These properties may ppply t o predicates formed from lexical entries as w e l l as to lexical-semantic r e l a t i o n s . To be precise, a relation R defined on a s e t S i s s a i d t o be a t r a n a ~t < v e relation if whenever b and c are R-related and also c and d are I? related then b and d staAd in a r e l a t i o n R a l s o , Synonyniy is a transitive r e l a t i o n just as t r a n s i t i v i t y i s . The preposition in behaves in the same way. If Sam is in the kitchen and the kitchen i s i n the h o t e l , then w e know that Sam i s i n the h o t e l . The t i m e interrelation before behaves like t h i s , too. If Zorro arrived before the posse did and t h e posse arrived before thz explosion, then w e know thgt Zorro arrived before the explosion. A r e l a t i o n R defined on a set S is said to have the refZez<ue property if all the elements of S are R-related to thenl~elves, that is, if mRm is true for all members m of the set S, The synonymy relation has this property a word means the same thin% as itself. The antonymy relation ANTI does not have this property. It is not rrue tha&, hot ANTI be, for example. A relation R d e f i n e d on a set S is said t o be e ~s t r i c if whenever,b and c are R-related then so are c and b; that is, R is symme.tric if and only i f bRc always implies cRb. Synonymy also has t h i s property. If b is synonymous with c , then c is synonymous with b. So has antonfly. Given that hot ANTI sold, we immediately know that= c d d ANTI hot. Taxonomy ie not eymmetric, however. A lion is a kind of mammal, but a mammal is not a kind of lion. In question answering we may b e just as Interested in drawing negative conclusions as positive-ones. Thus i ~r m a y be important to know tliat tf bRc is true then cRb must be falae. The term asynmrstrio is used to describe a r e l a t i o n R f o r w h i c h b R c and cRb are never both true, at $east when b and c are different elements of the stt S. Taxonomy is asymmetric and so is the thug interrelation before. If the question asks, "Did c happen before b?" and we know that b happened before r, we can answer with a confident no. For want of a better term we w i l l say that the r e l r Sion R is mn-synonetrio if it is neither symmetric or &symmetric. In t h i s case bRc and cRb are sometimes both true and sometfmes n o t . S h i l a r l y , he term imefz.exive i s used f o r the case i n which mRm is never true, while the term nonreflexit)e is used for the case in which mRm i s s o m e t f m e s true and sometimes not. In the same way i n t r a n s i t i ~e is taken to mean that if bRc and cRd, we can conclude that b and d are not R-related, while nantrcrnsitive will mean that bRd is sometimes true if bRc and cRd, butnot always. Each lexical relation itself; has a lexical entry. The reflexivity, symmetry, and transitivity properties of the relation are listed in t h i s | {
"name": [
"Evens, Martha W. and",
"Smith, Raoul N."
],
"affiliation": [
null,
null
]
} | null | null | null | 1978-12-01 | 0 | 0 | null | null | null | null | We might also allow substitution of conversives, nominalizations, etc.Nancy was Sally's student.Sally was ~ancy's teacher. Since a relation R is called symmetric if bRc alwaye implies cRb, it follows that a symmetric relation ie its own inverse. The syaonymy relation S and antonymy relation ANTI are both self-inverse i n t h i e sense.For t h i s reason we never need the spnbol ANTI, etc. ANTI is MITI The entry for hot includes ANTI cold, the entry for cold includes ANTI hot.(hrique Linkage.Raphael (1968) has proposed a property which seems extremely u s e f u l . .-( ) ' Y i X) (32 X) {ZRY)Condition Every noneslpty subset has a minLmum.Maximum W X c M ) --(qr X) (3 Z X) (YRZ)Condition Every nonempty subset has a maximum. immediately that i f bQc and cQd then bQ1d. Q ' , the 'successor' relation,4 i e now transitive, f o r if # l t c and cQ1d, then one can find s chain of Q-related objects l i n k i n g b and d j u s t bv cbncatenating the c h a i nl i n k i n g c and d , R a p b l ' s p a i r of relations j r i g h t and right behave t h i s way. The relations "is a c h i l d of" and "is a descendant of" are a l g a pafred in t h i s way, | null | Main paper:
:
We might also allow substitution of conversives, nominalizations, etc.Nancy was Sally's student.Sally was ~ancy's teacher. Since a relation R is called symmetric if bRc alwaye implies cRb, it follows that a symmetric relation ie its own inverse. The syaonymy relation S and antonymy relation ANTI are both self-inverse i n t h i e sense.For t h i s reason we never need the spnbol ANTI, etc. ANTI is MITI The entry for hot includes ANTI cold, the entry for cold includes ANTI hot.(hrique Linkage.Raphael (1968) has proposed a property which seems extremely u s e f u l . .-( ) ' Y i X) (32 X) {ZRY)Condition Every noneslpty subset has a minLmum.Maximum W X c M ) --(qr X) (3 Z X) (YRZ)Condition Every nonempty subset has a maximum. immediately that i f bQc and cQd then bQ1d. Q ' , the 'successor' relation,4 i e now transitive, f o r if # l t c and cQ1d, then one can find s chain of Q-related objects l i n k i n g b and d j u s t bv cbncatenating the c h a i nl i n k i n g c and d , R a p b l ' s p a i r of relations j r i g h t and right behave t h i s way. The relations "is a c h i l d of" and "is a descendant of" are a l g a pafred in t h i s way,
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0 | null | null | null | null | null | null | null | null |
a02ec16edc56c8b72684c228a9cfb065554942b2 | 14638648 | null | Glancing, Referring and Explaining in the Dialogue System {HAM-RPM} | T h i s paper tocusses on t h r e e components oft h e d i a l o g u e system HAM-KYM, which converses i n n a t u r a l language about v i s i b l e scenes. F ~r s t , ~t i s demonstrated how the system's communicative competence i s enhanced by i t s i m i t a t i o n of human v i s u a l -s e a r c h processes. The approach taken t o nounphrase r e s o l u t i o n i s then d e s c r i b e d , and an a l g o r i t h m f o r t h e generation o f noun phrases i s illustrated w i t h a s e r i e s o f examples: Finally, the s y s t e m ' s a b i l i t y to e x p l a i n i t s own reasoning i s d i s c u s s e d , w i t h emphasis on the novel a s p e c t s o f i t s i m p l e m e n t a t ~o n . | {
"name": [
"Wahlster, W. and",
"Jameson, A. and",
"Hoeppner, W."
],
"affiliation": [
null,
null,
null
]
} | null | null | null | 1978-12-01 | 15 | 22 | null | The n a t u r a l language d i a l o g u e system HAM-RPMu converses w i t h a human p a r t n e r about scenes w h i c h e i t h e r one o r b o t h a r e l o o k i n g a t d i r e c t l y ( o r have a photograph o f ) . A t p r e s e n t t h e system, which i s implemented i n FUZZY (~e F a i v r e 1977), 1 s b e i n g t e s t e d on two domains: t h e i n t e r i o r o f a l i v i n g room and a t r a f f i c scene. t h e s y s t e m . When i n t e r p r e t i n g t h e d e f i n~t e d e s c r i p t i o n t h e man I n front of t h e tree, assuming % h a t TREE4 i s t h e one meant, t h e system e n t e r s s e v e r a l CONTEXTS i n f r o n t o f TREE4, w i t h i n each r e t r i e v i n g the i n t e r n a l names o f t h e men r e c o g n i z a b l e f r o m t h a t p o i n t . I t doesn't f l n d MAN^ u n t i l i t has e n t e r e d t h e CONTEXT c o r r e s p o n d i n g t o p o i n t 0 . I t then e n t e r s a c o u p l e more, and, fin- about t h e scene a r e less i m p o r t a n t than t h e way t h e p a r t n e r h i m s e l f would be l i k e l y r o p e r c e i v e them. I f o n l y the f a c t s thenlselves a r e known, i n t o rrnation may be l a c k i n g which i s e s s e n t i a l f o r t h e p r o d u c t i o n o f a connnunicat i v e l y adequate response. For example, t h e d e f i n i t e d e s c r i p t i o n whose i n t e rp r e t a t i o n w a s j u s t sketched was, s t r i c t l y speaking, ambiguous, as t h e r e i s a second man i n f r o n t o f t h e t r e e whom t h e system would have considered t o be t h e r e f e r e n t o f the d e s c r i p t i o n i f MAN2 h a d n ' t been there. I n a d d~t i o n t o t h e i n t e r p r e t a t i o n o f ambiguous u t t e r a n c e s , t h e r e a r e o t h e r s i t u a t i o n s i n which t h i s approach can be a p p l i e d e l e g a n t l y ( F i g . 2 ) . The noun-phrase i n t e r p r e t e r t r i e s t o f i n d a u n i q u e r e f e r e n t for each Another case where a d e f i n i t e noun p h r a s e c a n ' t s i m p l y be r e p i | null | null | null | The Now l e t ' s t u r n t o some more complex problems o f noun-phrase g e n e r a t i o n . So S p e e c h A c t s . N. Y. : A c a d e r n~c , 4 1 -5 , These networks a r e implemented as CONTEXTS i n t h e sense i n t r o d u c e d by tile l anguage CONN I V E R . | Main paper:
, the treatment of v i s u a l data:
The n a t u r a l language d i a l o g u e system HAM-RPMu converses w i t h a human p a r t n e r about scenes w h i c h e i t h e r one o r b o t h a r e l o o k i n g a t d i r e c t l y ( o r have a photograph o f ) . A t p r e s e n t t h e system, which i s implemented i n FUZZY (~e F a i v r e 1977), 1 s b e i n g t e s t e d on two domains: t h e i n t e r i o r o f a l i v i n g room and a t r a f f i c scene. t h e s y s t e m . When i n t e r p r e t i n g t h e d e f i n~t e d e s c r i p t i o n t h e man I n front of t h e tree, assuming % h a t TREE4 i s t h e one meant, t h e system e n t e r s s e v e r a l CONTEXTS i n f r o n t o f TREE4, w i t h i n each r e t r i e v i n g the i n t e r n a l names o f t h e men r e c o g n i z a b l e f r o m t h a t p o i n t . I t doesn't f l n d MAN^ u n t i l i t has e n t e r e d t h e CONTEXT c o r r e s p o n d i n g t o p o i n t 0 . I t then e n t e r s a c o u p l e more, and, fin- about t h e scene a r e less i m p o r t a n t than t h e way t h e p a r t n e r h i m s e l f would be l i k e l y r o p e r c e i v e them. I f o n l y the f a c t s thenlselves a r e known, i n t o rrnation may be l a c k i n g which i s e s s e n t i a l f o r t h e p r o d u c t i o n o f a connnunicat i v e l y adequate response. For example, t h e d e f i n i t e d e s c r i p t i o n whose i n t e rp r e t a t i o n w a s j u s t sketched was, s t r i c t l y speaking, ambiguous, as t h e r e i s a second man i n f r o n t o f t h e t r e e whom t h e system would have considered t o be t h e r e f e r e n t o f the d e s c r i p t i o n i f MAN2 h a d n ' t been there. I n a d d~t i o n t o t h e i n t e r p r e t a t i o n o f ambiguous u t t e r a n c e s , t h e r e a r e o t h e r s i t u a t i o n s i n which t h i s approach can be a p p l i e d e l e g a n t l y ( F i g . 2 ) . The noun-phrase i n t e r p r e t e r t r i e s t o f i n d a u n i q u e r e f e r e n t for each Another case where a d e f i n i t e noun p h r a s e c a n ' t s i m p l y be r e p i
. noun-phrase generation:
The Now l e t ' s t u r n t o some more complex problems o f noun-phrase g e n e r a t i o n . So S p e e c h A c t s . N. Y. : A c a d e r n~c , 4 1 -5 , These networks a r e implemented as CONTEXTS i n t h e sense i n t r o d u c e d by tile l anguage CONN I V E R .
Appendix:
| null | null | null | null | {
"paperhash": [
"hahn|ham-rpm:_natural_dialogues_with_an_artificial_partner",
"kleer|amord_explicit_control_of_reasoning",
"davis|generalized_procedure_calling_and_content-directed_invocation"
],
"title": [
"HAM-RPM: Natural Dialogues with an Artificial Partner",
"AMORD explicit control of reasoning",
"Generalized procedure calling and content-directed invocation"
],
"abstract": [
"This paper introduces the understanding system HAM-RPM, which simulates a dialogue partner conversing about a real-life domain. After outlining the system's overall structure, we discuss three of its distinguishing features: The first is its organization of spatial data in a redundant multiple data base, inspired by certain aspects characteristic of visual search in humans. A new algorithm for noun-phrase generation is then sketched which is sensitive to the conversational state and uses a 'worst-case-first' strategy. Finally, we describe in some detail a specifie operationalisation of the notion of the communicative relevance of objects. The paper concludes with a summary of the objectives of this research.",
"The construction of expert problem-solving systems requires the development of techniques for using modular representations of knowledge without encountering combinatorial explosions in the solution effort. This report describes an approach to dealing with this problem based on making some knowledge which is usually implicitly part of an expert problem solver explicit, thus allowing this knowledge about control to be manipulated and reasoned about. The basic components of this approach involve using explicit representations of the control structure of the problem solver, and linking this and other knowledge manipulated by the expert by means of explicit data dependencies.",
"We suggest that the concept of a strategy can profitably be viewed as knowledge about how to select from among a set of plausibly useful knowledge sources, and explore the framework for knowledge organization which this implies. We describe meta rules, a means of encoding strategies that has been implemented in a program called TEIRESIAS, and explore their utility and contribution to problem solving performance.\n Meta rules are also considered in the broader context of a tool for programming. We show that they can be considered a medium for expressing the criteria for retrieval of knowledge sources in a program, and hence can be used to define control regimes. The utility of this as a programming mechanism is considered.\n Finally, we describe the technique of content-directed invocation used by meta rules, and consider its use as a way of implementing strategies. It is also considered in historical perspective as a knowledge source invocation technique, and its advantage over some existing mechanisms like goal-directed invocation is considered.\n This work was supported in part by the Bureau of Health Sciences Research and Evaluation of HEW under Grant HS-01544 and by the Advanced Research Projects Agency under ARPA Order 2494. It was carried out on the SUMEX Computer System, supported by the NIH under Grant RR-00785. The views expressed are solely those of the author."
],
"authors": [
{
"name": [
"W. V. Hahn",
"W. Hoeppner",
"A. Jameson",
"W. Wahlster"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"J. Kleer",
"J. Doyle",
"G. Steele",
"G. Sussman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Randall Davis"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null
],
"s2_corpus_id": [
"5389282",
"18255129",
"13016456"
],
"intents": [
[],
[
"result"
],
[
"result"
]
],
"isInfluential": [
false,
false,
false
]
} | null | 554 | 0.039711 | null | null | null | null | null | null | null | null |
e0492552c7236574190ddee8e44e85d9e97360d7 | 219303310 | null | Path-Based and Node-Based Inference in Semantic Networks | and S are re T=--r atxons R / S is R composed with S, So, V x , y (xR/Sy <-> aa(xRx 5 a s p ) ) . Domain Restriction --If R and S are r e l a t i o n s , (S a 1 R &a t h e rel a t i o n R w i € h V i t s domain restricted to those objects that b e a r the-relation S t o s. SO, Vz,y,.(z(S s)Ry <-> (zS. C l z ~~) ) . | {
"name": [
"Shapiro, Stuart C."
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | null | null | null | The adfaFtZgees or node-base.# i n f e rence s t e m from t h e g e n e r a l i t y of t h e synt a x of node-based inference r u l e s , Pathbased r u l e s are limited to binary r e l at i o n s , hav6 a r e s t r i c t e d q u a n t i f i c a t i o n s t r u c t u r e and r e q u i r q t h a t an arc between t w o nodes be implied by a path between t h e same t w o nodes. Rule R 2 of I!'iguree2 could n o t be w r i t t e n as a path-based r u l e , and, although t h e t r a n s i t i v i t y of SUPPORTS could be expreaaed b a path-based r u l e I (SUPPORTS + SUPPORTS ) , t h e "second order" r u l e R4 of Figure 3 could n o t .L e t us b r i e f l y consider how r u l e R4 i s constructed, whether it r e a l l y i s o r i s n o t a second oeder r u l e , and why it could n o t be expressed as a path-based rulg.Rule R4 supplies a r u l e for use with t r a n s i t i v e r e l a t i o n s . I n order to assert t h a t a r e l a t i o n is t r a n s i t i v e (e.g. assert i o n node M 1 of Figure 4 ) , the r e l a t i o n must be represented as a node, rather than as an arc. This a l s o allows quantificat i o n over such r e l a t i o n s , s i n c e i n a l l node-based inference r u l e formalisms v a r iables may onLy be substituted f o r nodes, n o t for arcs. Since t h e r e l a t i o n is a node, another node must be used to show t h e r e l a t i o n s h i p of t h e r e l a t i o n t o its arguments (e,g. nodes M2 and M3 i n I n t h i s view, t h e p r e d i c a t e s of semantic networks a r e not the nodes representing conceptual r e l a t i o n s , b u t t h e d i f f e r e n t case frames. Rule R4 cannot be repreaented as a path-based rule because it is a r u l e qbout t h e r e l a t i o n AVO, and AVO is a t r i n a r y , r a t h e r than a binary r e l a t i o n .Although some node-based inference r u l e s cannot be expressed by path-based inference r u l e s , it is easy to see t h a t any path-based inference r u l e can be expressed by a node-based inference rule, as long as w e a r e w i l l i n g to replace some arc-relations by nbdes and higher order predicates. | s i n c e the mid s i x t i e s [ t o ; 11 I as a formali s m for t h e r e p r e s e n t a t i o n of knowlEage. Methods have a l s o been developing f o r performing deductive i n f e r e n c e on t h e knowledge r e p r e s e n t e d i n t h e network. I n t h i s paper, we w i l l compare t w o s t y l e s o f i nference t h a t are used i n semantic networks, path-based infexence and node-based i n f e rence. Xn sections 2 and 3, t h e s e terms w i l l be e x p l a i n e d and r e f e r e n c e s to systems tibt use them w i l l be provided. Xn s e c t i o n s 4 and 5 , t h e advantages and disadvantages o f each w i l l be discu8sed. Section% 6 , 7 and 8 w i l l show how they can be used t o complement each o t h e r i n a s i ngle semantic network system, how pathb a s e d c k f e r e n c e can f a l p represent t h e ext e n s f a n a l equivalence of i n t e n s i o n a l concepts, and how a f o r m a l i s m for writing path-based i n f e r e n c e r u l e s can be used to e x p l i c a t e t h e n o t i o n of " i n h e r i t a n c e a i n a semantic network.Path-Based I n f e r e n c e Let: u s r e f e x to a r e l a t i o n (pegf~xce binary) thati s r e p r e s e n t e d by a n arc in a network as a na r c -r e l a t i o n . I f R i s an a r c -r e l a t i o n , an arc l a b e l l e d R from node a to node b r e p r e s e n t s t h a t the r e l a t i o nship aRb holds. I t may be t h a t this arc is n o t p r e s e n t i n t h e network, b u t aRb may be i n f e r r e d from other i n f o r m a t i o n p r e s e n t i n t h e network and one or more i n f e r e n c e r u l e s . I f the other information i n t h e network is a specified path of arcs ftom a t o b , w e w i l l r e f e r t o the i n f e r e n c e as path-based. The ways i n which such p a t h s may be s p e c i f i e d w i l l be developed as this paper proceeds.The two clearest examples of the gene r a l use of path-based i n f e r e n c e are i n SAMENLAQ X I [I81 and Protosynthex I11 [13]. Both t h e s e systems use what might be c a l led " r e l a t i o n a l " networks rather than i n Protoaynthex I11 t h e r e is a n arc labelled COMMANDED from the node r e p r e s e n t i n g Napoleon t o t h e node r e p r e s e n t i n g t h e French army, and i n SAMENLAQ I1 a n arc l abelled EA$T.OF goes f r o m the node for Albany to t h e node' for Buffalo. Both sya t e m s u s e r e l a t i o n a l c a l c u l u s expresdions to form path-based -i n f e r e n c e r u l e s . The following r e l a t i o n a l operators are employed (we h e r e u s e a varia'nt of t h e earlier n o t a t i o n s ) : The network formalism employed i s t h a t o f Shapiro {15;17]. These deduction r u l e s employ p a t t e r n nodes (PI , P2, P 3 , P 4 , P 5P4, P 7 ) , each one of which r e p r e s e n t s a p a t t e r n of nodes t h a t might occur i n the network. W e w i l l t h e r e f o r e c a l l t h i s kind of i n f e r e n c e r u l e a n o d e -h s e d i n f e r e n c e ru5g. P a t t e r n nods are r e l a t e d t o eachother by rut@ nodee, each of which repres e n t a p r o p o s i t i o n a l o p e r a t o r , or, equival e n t l y , an inference mechanism. For examL p l e , R2 r e p r e s e n t s t h e r u l e t h a t if an i ns t a n c e of P 1 occurs i n the network, an i ns t a n c e of R1 with t h e same s u b s t i t u t i o n for 8 m y be deduced., Q u a n t i f i c a t i o n i s represented i n t h i s notation by an arc-rel a t i o n between a r u l e node and t h e variable nodes bound i n t h e rule. For ex-le, 3: i s bound by a univexsall q u a n t i f i e r i n R2and p ie bound by an e x i s t e n t i a l quantif i e r i n R 1 . To see how a node-based inference proceeds, consider t h e n e t w k of Fig-4 i n conjunction with t h e r u l e of Ftgure 3 , and say that we w i s h t o decide i f 4 S U P P~R T S C. The network t h a t would repr e s e n t t h a t A SUPPORTS C matches It should be noted t h a t set inclusion was represented by an arc ( I S A ) i n Section 2, but set membership i s being represented by a node (with a MEMBER, CLASS "case frame") i n t h i s s e c t i o n , The nodal repres e n t a t i o n fs required by node-based i n f e rence r u l e s and i s c o n s i s t e n t with t h e not i o n t h a t everything t h a t t h e hetwork "knows", and every concept t o which t h e Y o network can refer is xepresented by a -node .The major advantage of path-based inference i s e f f i c i e n c y , Carrying out a pith-based inference involves merely checking t h a t a s p e c i f i e d path e x i s t s i n t h e netwoxk between two given nodes (plus, perhaps, 8ome s i d e paths t o specified nodes required by domain and range rrsstrictiona). This f s a well understood and rel a t i v e l y e f f i c i e n t operation, especially compared to t h e backtracking, intersection, o r u n i f i c a t i o n operations required t o W c , k the consistency of v a r i a b l e substitutione i n n notSEX2seT r?iference rules, Moreover, path following seems t o many people t o be what semantic networks were okfginally designed for. The major search algorithm of Q u i l l i a n ' s SemanticMemory is a bi-directional search for a path connecting two nodes (1 0, p. 2491. Also, t h e a b i l i t y t o do path t r a c i n g is a motivation underlying I SA hierarchies, and is why t h e Collins and Q u i l l f a n r e s u l t s 121 gained such attention. Theae ef f iciencies a r e lost by replacing path-based inference r u l e s by node-based inference rules.6 % Combining Path-Based and Node-Based Inference Let us, therefbre, extend our syntax of path-based inference rules t o allow a 41 path of a r c compositions on t h e l e f t of t h e "+" symbol. The r u l e I S A + 1 SA* states t h a t whenever t h e r e is a path o f " 1 a r c s from m d e n t o mode m, w e can infer a " v i r t u a l n I S A arc d i r e c t l y f r o m n t o m which we may, if wkwish, a c t u a l l y add to t h e network. S;tm&lply, let the r u l e SUB-/SUP 4-(SUB-/SUP)* s t a t e tnat whenever a path of a l t e r n a t i n g SUB-and SUP a r c s goes from node n t o node m, w e can i n f e r a " v i r t u a l n node with SUB to n and SUP to m which w e may, if we Wish, a c t u a l l y add to t h e network.We now have a formalism f o r specifying path-based inference r u l e s i n a nskt work formalism t h a t represents binary conceptual relations by two c a s e case frames. This would allow, f o r example, for a more unified representation i n t h e SNIFFER system [31, i n which node-based inference rules a r e implemented and buult-in path . One i s t h e match routine t h a t i s given a p a t t e r n node a n d f i n d s instances of it i n t h e network, and t h e other i s t h e routine t h a t intrtrpsets t h e q u a n t i f i e r s ana connectives to carry o u t t h e a c t u a l deduction. The match r o u t i n e can be enhanced t o make use of path-based inference rules. Consider a t y p i c a l match routine used i n t h e dedubtion i n Section 3 of A SUPPORTS C from the network of Figure 4 and t h e r u l e of Figure 3 , q d l e t u s fntroduce t h e notation t h a t i f P i s a path of arcs and n i s a node, P [n] represents t h e set of nodes found by following t h e path P from t h e node n. I n t h e example, the match r o u t i n e was given t h e p a t t e r n P 4 t o match i n t h e binding [rjSUPPORTS 1. I t waS 93;rle t o f i n d M 1 by i n t e r s e c t i n g CLASS^ ITRANSITIVEI w i t h MEMBER~~SUPPORTSI .Now, l e t u s suppose t h a t t h e path-basea inference rule CLASS f CLASS/ (SUB-&UP) * has been declared i n such a way t h a t h e match r o u t i n e could use it. T t e match routine would i n t e r s e c t MEMBER [SUPPORTS] with (CLASS/(SUB-/SUP)*)~[TRANSITIVEI and be a b l e t o f i n d a v i r t u a l node a s s e r t i n g t h a t SUPPORTS is TRANSITIVE e m n if a long chain of set inclusions separated them. The proposal, therefore, i s t h i s : any arc-relation i n a semantic network may be defined i n terms of a path-baaed inference r u l e which t h e match r o u t i n e i s capable of using when finding instances of p a t t e r n | null | null | A b s t r a c t Two styles of performing inferehcs i n semantic networks are presenteu and compared. P a t h -b a s d i n w x e n c e allows an arc or a path of arcs between two given-nodes t o be i n f e r r e d from t h e e x i s t e n c e of ano t h e r s p e c i f i e d p a t h between t h e same two nodes. Path-based infexence r u l e s may be w r i t t e n u s h g a binary relational c a l c u l u s notation. Node-based i n f e r e n c e allows a s t r u c t u r e o f nodes t o be i n f e r r e d from t h e e x i s t e n c e of a n i n s t a c c e of a p a t t e r n of node s t r u c t u r e s . Node-based i n f e r e n c e riles can be c o n s t r u c t e d i n a semantic network u s i n g a v a r i a n t o f ' a p r e d i c a t e c a l c u l u s n o t a t i o n . Path-based i n f e r e n c e i s more e f f i c i e n t , while node-based i n f e rence i s more general. A method i s des c r i b e d o f combining t h e t w o s t y l e s i n a s i n g l e system i n order t o take advantage o f t h e s t r e n g t h s of each. App&ications of path-based i n f e r e n c e r u l e s to t h e repres e n t a t i o n of $he ex-sional equivalence o f i n t e n s i o n a l concepts, and t o t h e e x p l ic a t i o n of i n h e r i t a n c e i n h i e r a r c h i e s are sketched. | Main paper:
i n t r o d u c t i o n semantic networks have developed:
s i n c e the mid s i x t i e s [ t o ; 11 I as a formali s m for t h e r e p r e s e n t a t i o n of knowlEage. Methods have a l s o been developing f o r performing deductive i n f e r e n c e on t h e knowledge r e p r e s e n t e d i n t h e network. I n t h i s paper, we w i l l compare t w o s t y l e s o f i nference t h a t are used i n semantic networks, path-based infexence and node-based i n f e rence. Xn sections 2 and 3, t h e s e terms w i l l be e x p l a i n e d and r e f e r e n c e s to systems tibt use them w i l l be provided. Xn s e c t i o n s 4 and 5 , t h e advantages and disadvantages o f each w i l l be discu8sed. Section% 6 , 7 and 8 w i l l show how they can be used t o complement each o t h e r i n a s i ngle semantic network system, how pathb a s e d c k f e r e n c e can f a l p represent t h e ext e n s f a n a l equivalence of i n t e n s i o n a l concepts, and how a f o r m a l i s m for writing path-based i n f e r e n c e r u l e s can be used to e x p l i c a t e t h e n o t i o n of " i n h e r i t a n c e a i n a semantic network.Path-Based I n f e r e n c e Let: u s r e f e x to a r e l a t i o n (pegf~xce binary) thati s r e p r e s e n t e d by a n arc in a network as a na r c -r e l a t i o n . I f R i s an a r c -r e l a t i o n , an arc l a b e l l e d R from node a to node b r e p r e s e n t s t h a t the r e l a t i o nship aRb holds. I t may be t h a t this arc is n o t p r e s e n t i n t h e network, b u t aRb may be i n f e r r e d from other i n f o r m a t i o n p r e s e n t i n t h e network and one or more i n f e r e n c e r u l e s . I f the other information i n t h e network is a specified path of arcs ftom a t o b , w e w i l l r e f e r t o the i n f e r e n c e as path-based. The ways i n which such p a t h s may be s p e c i f i e d w i l l be developed as this paper proceeds.The two clearest examples of the gene r a l use of path-based i n f e r e n c e are i n SAMENLAQ X I [I81 and Protosynthex I11 [13]. Both t h e s e systems use what might be c a l led " r e l a t i o n a l " networks rather than i n Protoaynthex I11 t h e r e is a n arc labelled COMMANDED from the node r e p r e s e n t i n g Napoleon t o t h e node r e p r e s e n t i n g t h e French army, and i n SAMENLAQ I1 a n arc l abelled EA$T.OF goes f r o m the node for Albany to t h e node' for Buffalo. Both sya t e m s u s e r e l a t i o n a l c a l c u l u s expresdions to form path-based -i n f e r e n c e r u l e s . The following r e l a t i o n a l operators are employed (we h e r e u s e a varia'nt of t h e earlier n o t a t i o n s ) : The network formalism employed i s t h a t o f Shapiro {15;17]. These deduction r u l e s employ p a t t e r n nodes (PI , P2, P 3 , P 4 , P 5P4, P 7 ) , each one of which r e p r e s e n t s a p a t t e r n of nodes t h a t might occur i n the network. W e w i l l t h e r e f o r e c a l l t h i s kind of i n f e r e n c e r u l e a n o d e -h s e d i n f e r e n c e ru5g. P a t t e r n nods are r e l a t e d t o eachother by rut@ nodee, each of which repres e n t a p r o p o s i t i o n a l o p e r a t o r , or, equival e n t l y , an inference mechanism. For examL p l e , R2 r e p r e s e n t s t h e r u l e t h a t if an i ns t a n c e of P 1 occurs i n the network, an i ns t a n c e of R1 with t h e same s u b s t i t u t i o n for 8 m y be deduced., Q u a n t i f i c a t i o n i s represented i n t h i s notation by an arc-rel a t i o n between a r u l e node and t h e variable nodes bound i n t h e rule. For ex-le, 3: i s bound by a univexsall q u a n t i f i e r i n R2and p ie bound by an e x i s t e n t i a l quantif i e r i n R 1 . To see how a node-based inference proceeds, consider t h e n e t w k of Fig-4 i n conjunction with t h e r u l e of Ftgure 3 , and say that we w i s h t o decide i f 4 S U P P~R T S C. The network t h a t would repr e s e n t t h a t A SUPPORTS C matches It should be noted t h a t set inclusion was represented by an arc ( I S A ) i n Section 2, but set membership i s being represented by a node (with a MEMBER, CLASS "case frame") i n t h i s s e c t i o n , The nodal repres e n t a t i o n fs required by node-based i n f e rence r u l e s and i s c o n s i s t e n t with t h e not i o n t h a t everything t h a t t h e hetwork "knows", and every concept t o which t h e Y o network can refer is xepresented by a -node .
advantaqes of node-based inference:
The adfaFtZgees or node-base.# i n f e rence s t e m from t h e g e n e r a l i t y of t h e synt a x of node-based inference r u l e s , Pathbased r u l e s are limited to binary r e l at i o n s , hav6 a r e s t r i c t e d q u a n t i f i c a t i o n s t r u c t u r e and r e q u i r q t h a t an arc between t w o nodes be implied by a path between t h e same t w o nodes. Rule R 2 of I!'iguree2 could n o t be w r i t t e n as a path-based r u l e , and, although t h e t r a n s i t i v i t y of SUPPORTS could be expreaaed b a path-based r u l e I (SUPPORTS + SUPPORTS ) , t h e "second order" r u l e R4 of Figure 3 could n o t .L e t us b r i e f l y consider how r u l e R4 i s constructed, whether it r e a l l y i s o r i s n o t a second oeder r u l e , and why it could n o t be expressed as a path-based rulg.Rule R4 supplies a r u l e for use with t r a n s i t i v e r e l a t i o n s . I n order to assert t h a t a r e l a t i o n is t r a n s i t i v e (e.g. assert i o n node M 1 of Figure 4 ) , the r e l a t i o n must be represented as a node, rather than as an arc. This a l s o allows quantificat i o n over such r e l a t i o n s , s i n c e i n a l l node-based inference r u l e formalisms v a r iables may onLy be substituted f o r nodes, n o t for arcs. Since t h e r e l a t i o n is a node, another node must be used to show t h e r e l a t i o n s h i p of t h e r e l a t i o n t o its arguments (e,g. nodes M2 and M3 i n I n t h i s view, t h e p r e d i c a t e s of semantic networks a r e not the nodes representing conceptual r e l a t i o n s , b u t t h e d i f f e r e n t case frames. Rule R4 cannot be repreaented as a path-based rule because it is a r u l e qbout t h e r e l a t i o n AVO, and AVO is a t r i n a r y , r a t h e r than a binary r e l a t i o n .Although some node-based inference r u l e s cannot be expressed by path-based inference r u l e s , it is easy to see t h a t any path-based inference r u l e can be expressed by a node-based inference rule, as long as w e a r e w i l l i n g to replace some arc-relations by nbdes and higher order predicates.
advantages of path-based inference:
The major advantage of path-based inference i s e f f i c i e n c y , Carrying out a pith-based inference involves merely checking t h a t a s p e c i f i e d path e x i s t s i n t h e netwoxk between two given nodes (plus, perhaps, 8ome s i d e paths t o specified nodes required by domain and range rrsstrictiona). This f s a well understood and rel a t i v e l y e f f i c i e n t operation, especially compared to t h e backtracking, intersection, o r u n i f i c a t i o n operations required t o W c , k the consistency of v a r i a b l e substitutione i n n notSEX2seT r?iference rules, Moreover, path following seems t o many people t o be what semantic networks were okfginally designed for. The major search algorithm of Q u i l l i a n ' s SemanticMemory is a bi-directional search for a path connecting two nodes (1 0, p. 2491. Also, t h e a b i l i t y t o do path t r a c i n g is a motivation underlying I SA hierarchies, and is why t h e Collins and Q u i l l f a n r e s u l t s 121 gained such attention. Theae ef f iciencies a r e lost by replacing path-based inference r u l e s by node-based inference rules.6 % Combining Path-Based and Node-Based Inference Let us, therefbre, extend our syntax of path-based inference rules t o allow a 41 path of a r c compositions on t h e l e f t of t h e "+" symbol. The r u l e I S A + 1 SA* states t h a t whenever t h e r e is a path o f " 1 a r c s from m d e n t o mode m, w e can infer a " v i r t u a l n I S A arc d i r e c t l y f r o m n t o m which we may, if wkwish, a c t u a l l y add to t h e network. S;tm&lply, let the r u l e SUB-/SUP 4-(SUB-/SUP)* s t a t e tnat whenever a path of a l t e r n a t i n g SUB-and SUP a r c s goes from node n t o node m, w e can i n f e r a " v i r t u a l n node with SUB to n and SUP to m which w e may, if we Wish, a c t u a l l y add to t h e network.We now have a formalism f o r specifying path-based inference r u l e s i n a nskt work formalism t h a t represents binary conceptual relations by two c a s e case frames. This would allow, f o r example, for a more unified representation i n t h e SNIFFER system [31, i n which node-based inference rules a r e implemented and buult-in path . One i s t h e match routine t h a t i s given a p a t t e r n node a n d f i n d s instances of it i n t h e network, and t h e other i s t h e routine t h a t intrtrpsets t h e q u a n t i f i e r s ana connectives to carry o u t t h e a c t u a l deduction. The match r o u t i n e can be enhanced t o make use of path-based inference rules. Consider a t y p i c a l match routine used i n t h e dedubtion i n Section 3 of A SUPPORTS C from the network of Figure 4 and t h e r u l e of Figure 3 , q d l e t u s fntroduce t h e notation t h a t i f P i s a path of arcs and n i s a node, P [n] represents t h e set of nodes found by following t h e path P from t h e node n. I n t h e example, the match r o u t i n e was given t h e p a t t e r n P 4 t o match i n t h e binding [rjSUPPORTS 1. I t waS 93;rle t o f i n d M 1 by i n t e r s e c t i n g CLASS^ ITRANSITIVEI w i t h MEMBER~~SUPPORTSI .Now, l e t u s suppose t h a t t h e path-basea inference rule CLASS f CLASS/ (SUB-&UP) * has been declared i n such a way t h a t h e match r o u t i n e could use it. T t e match routine would i n t e r s e c t MEMBER [SUPPORTS] with (CLASS/(SUB-/SUP)*)~[TRANSITIVEI and be a b l e t o f i n d a v i r t u a l node a s s e r t i n g t h a t SUPPORTS is TRANSITIVE e m n if a long chain of set inclusions separated them. The proposal, therefore, i s t h i s : any arc-relation i n a semantic network may be defined i n terms of a path-baaed inference r u l e which t h e match r o u t i n e i s capable of using when finding instances of p a t t e r n
:
A b s t r a c t Two styles of performing inferehcs i n semantic networks are presenteu and compared. P a t h -b a s d i n w x e n c e allows an arc or a path of arcs between two given-nodes t o be i n f e r r e d from t h e e x i s t e n c e of ano t h e r s p e c i f i e d p a t h between t h e same two nodes. Path-based infexence r u l e s may be w r i t t e n u s h g a binary relational c a l c u l u s notation. Node-based i n f e r e n c e allows a s t r u c t u r e o f nodes t o be i n f e r r e d from t h e e x i s t e n c e of a n i n s t a c c e of a p a t t e r n of node s t r u c t u r e s . Node-based i n f e r e n c e riles can be c o n s t r u c t e d i n a semantic network u s i n g a v a r i a n t o f ' a p r e d i c a t e c a l c u l u s n o t a t i o n . Path-based i n f e r e n c e i s more e f f i c i e n t , while node-based i n f e rence i s more general. A method i s des c r i b e d o f combining t h e t w o s t y l e s i n a s i n g l e system i n order t o take advantage o f t h e s t r e n g t h s of each. App&ications of path-based i n f e r e n c e r u l e s to t h e repres e n t a t i o n of $he ex-sional equivalence o f i n t e n s i o n a l concepts, and t o t h e e x p l ic a t i o n of i n h e r i t a n c e i n h i e r a r c h i e s are sketched.
Appendix:
| null | null | null | null | {
"paperhash": [
"fikes|a_network-based_knowledge_representation_and_its_natural_deduction_system",
"shapiro|representing_and_locating_deduction_rules_in_a_semantic_network",
"mcdermott|artificial_intelligence_meets_natural_stupidity",
"hendrix|expanding_the_utility_of_semantic_networks_through_partitioning",
"schubert|extending_the_expressive_power_of_semantic_networks",
"shapiro|a_net_structure_for_semantic_information_storage,_deduction_and_retrieval",
"schwarcz|a_deductive_question-answerer_for_natural_language_inference",
"shapiro|a_net_structure_based_relational_question_answerer:_description_and_examples",
"kay|the_mind_system"
],
"title": [
"A Network-Based Knowledge Representation and Its Natural Deduction System",
"Representing and locating deduction rules in a semantic network",
"Artificial intelligence meets natural stupidity",
"Expanding the Utility of Semantic Networks Through Partitioning",
"Extending The Expressive Power Of Semantic Networks",
"A Net Structure for Semantic Information Storage, Deduction and Retrieval",
"A deductive question-answerer for natural language inference",
"A Net Structure Based Relational Question Answerer: Description and Examples",
"The MIND System"
],
"abstract": [
"We describe a knowledge representation scheme called KNET and a problem solving system called SNIFFER designed to answer queries using a K-NET knowledge base. K-NET uses a partitioned semantic net to combine the expressive capabilities of the first-order predicate calculus with linkage to procedural knowledge and with full indexing of objects to the relationships in which they participate. Facilities are also included for representing taxonomies of sets and for maintaining hierarchies of contexts. SNIFFER is a manager and coordinator of deductive and problem-solving processes. The basic system includes a logically complete set of natural deduction facilities that do not require statements to be converted into clause or prenex normal form. Using SNIFFER's coroutine-based control structure, alternative proofs may be constructed in pseudo-parallel and results shared among them. In addition, SNIFFER can also manage the application of specialist procedures that have specific knowledge about a particular domain or about the topology of the K-NET structures, for example, specialist procedures are used to manipulate taxonomic information and to link the system to information in external data bases.",
"A semantic network is defined with its arcs and nodes separated into various sets. Arcs are partitioned into descending, ascending, and auxiliary arcs. Nodes are partitioned into base, variable, assertion, pattern and auxiliary nodes. Nodes can be temporary or permanent.Some pattern and assertion nodes, called rule nodes, represent propositional functions of the nodes they dominate. Rule nodes may bind the variables they dominate with any one of a set of binding relations representing quantifiers. A rule node which dominates variables all of which are bound is a constant deduction rule.Deduction rules may be viewed as pattern-invoked procedures. The type of propositional function determines the procedure, the variables bound by the rule are the local variables, and the quantifier determines the type of binding.A binding is defined as a list of variables associated with the nodes they are bound to. A binding can be used like a substitution, except it is seldom actually applied. Instead, a pattern node and a binding for it are used as a pair.A match routine is defined which is given a source node and a binding and finds target nodes, target bindings and more fully specified source bindings. Target nodes that are patterns provide entrees into relevant rules.",
"As a field, artificial intelligence has always been on the border of respectability, and therefore on the border of crackpottery. Many critics <Dreyfus, 1972>, <Lighthill, 1973> have urged that we are over the border. We have been very defensive toward this charge, drawing ourselves up with dignity when it is made and folding the cloak of Science about us. On the other hand, in private, we have been justifiably proud of our willingness to explore weird ideas, because pursuing them is the only way to make progress.",
"An augmentation of semantic networks is presented in which the various nodes and arcs are partitioned into \"net spaces.\" These net spaces delimit the scopes of quantified variables, distinguish hypothetical and imaginary situations from reality, encode alternative worlds considered in planning, and focus attention at particular levels of detail.",
"\"Factual knowledge\" used by natural language processing systems can be constructively represented in the form of semantic networks. Compared to a \"linear\" representation such as that of the Predicate Calculus however, semantic networks presort racial problems with respect to the use of logical connectives, quantifiers, descriptions, and certain other constructions. Systematic solutions to these problems will be proposed, in the form of extensions to a more or less conventorial network rotation. Predicate Calculus translations of network propositions will frequently be given for comparison, to illustrate the close kinship of the two forms of representation.",
"This paper describes a data structure, MENS (MEmory Net Structure), that is useful for storing semantic information stemming from a natural language, and a system, MENTAL (MEmory Net That Answers and Learns) that interacts with a user (human or program), stores information into and retrieves information from MENS and interprets some information in MENS as rules telling it how to deduce new information from what is already stored. MENTAL can be used as a guestion-answering system with formatted input /output, as a vehicle for experimenting with various theories of semantic structures or as the memory management portion of a natural language question-answering system.",
"The question-answering aspects of the Protosynthex III prototype language processing system are described and exemplified in detail. The system is written in LISP 1.5 and operates on the Q-32 time-sharing system. The system's data structures and their semantic organization, the deductive question-answering formalism of relational properties and complex-relation-forming operators, and the question-answering procedures which employ these features in their operation are all described and illustrated. Examples of the system's performance and of the limitations of its question-answering capability are presented and discussed. It is shown that the use of semantic information in deductive question answering greatly facilitates the process, and that a top-down procedure which works from question to answer enables effective use to be made of this information. It is concluded that the development of Protosynthex III into a practically useful system to work with large data bases is possible but will require changes in both the data structures and the algorithms used for question answering.",
"A question answering system is described which uses a net structure for storage of infor mation. The net structure consists of nodes and labelled edges, which represent relations be tween the nodes. The labels are also nodes, and therefore definitions of relations may be stored in the net. It is demonstrated that the generality and complexity of this memory struc ture allows a surprisingly powerful question an swering system to be constructed using comparit ively simple executive routines. Output from the question answerer, which is currently run ning on an interactive, time sharing system, is included, showing its range of applicabi l i ty i n cluding question answering, inductive and de ductive inference, simple theorem proving and problem solving.",
"The MIND system is a single computer program incorporating an extensible set of fundamental linguistic processors that can be combined on command to carry out a great variety of tasks from grammar testing to question-answering and language translation. The program is controlled from a graphic display console from which the user can specify the sequence of operations, modify rules, edit texts and monitor the details of each operation to any desired extent. Presently available processors include morphological and syntactic analyzers, a semantic file processor, a transformational component, a morphological synthesizer, and an interactive disambiguator."
],
"authors": [
{
"name": [
"R. Fikes",
"G. Hendrix"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"S. Shapiro"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. McDermott"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"G. Hendrix"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Lenhart K. Schubert"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"S. Shapiro"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. Schwarcz",
"John F. Burger",
"R. F. Simmons"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"S. Shapiro",
"G. H. Woodmansee"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"M. Kay",
"G. Martins"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null,
null,
null,
null,
null,
null
],
"s2_corpus_id": [
"9957931",
"2420273",
"28619965",
"9767925",
"60549410",
"33714788",
"14586260",
"46540731",
"43606749"
],
"intents": [
[],
[],
[],
[
"background"
],
[],
[],
[
"methodology"
],
[],
[]
],
"isInfluential": [
false,
false,
false,
false,
false,
false,
false,
false,
false
]
} | null | 554 | null | null | null | null | null | null | null | null | null |
af52675314bee13ba29971c694b5be73b15e9346 | 219307333 | null | A Model for Knowledge and its Application to Discourse Analysis | I s required t o know t h a t w ~e r c n t y p ~t a l t i g e r s a r e s t r i p e d " ( F u t n a n r , 1975, p 2 4 9 ) 1 )!any m o d e l s o f kr,owX.edfie 'have bccn d < w e l o p e b f o r use I n comp u t a t i o r a a l env-dr:,nmcnts . . Samp arc. f c ~ nrcstr i c tt.d dl-~mrai.rrz; ( B l a c k , 1908: | {
"name": [
"Phillips, Brian"
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 0 | 1 | null | null | null | null | l e v e l , wl-rich hna l e d r6 n g r e a t e r awareness uE t h e r o t e o f meaning" in Iangtxage For "A t e x t is best: regarded a s n SEIAhTXG u n i t ; a u n i t n o t a £form but o f meaningt' ( I l a l l i d a y ct Ilasan, 1976, p. 2 ) . This b e i n g so, discsursc a n a l y s i s will d e e p e n u r t l n d c r s t a n d i n g sf meaning andIn t h i s paper 1: p r e s e n t a rnodel, of nzcaning s t r a n~l y inf'l~lencccl h y Ilays (].969a, 3.969b, 1970, 1973) (1972) , where no particular context is prescrfbed* It will be apparent that a t many p o i n t s t h e present model draws upon these earlier syatems.Some of the differences between systems are probably differences i n notation-However no system I s a t a s t a g e of c o~s t a n e y or completeness that makes it worthwhile to devote much effort to e s t a b l i s h i n g the T h i s membership r e l a t i o n is termed i n s t a n c e ( I S T ) . For example, "William Proxmire" i s an i n s t a n c e of "person", F i g u r e 2. Most i n s t a n c e s are n o t named, t a k h g t h e i r name from t h e i r varietal p a r e n t , but a major exception i s people, ebg., "~e t e r " , "Aunt S a l l y " , Figure 1.Any p a t h through the paradigmatic o r g a n i z a t i o n of knowledge which To r e p r e s e p t t e arguments of occas2onal predications, t h e D i c a l (TYP) arc is used. Thus the "blrd" in "bfrds eat worms" is as in Figure 2 . It i s also possible t o use the typical r e l a t i o n t o a t t a c h occasional proper t i e s t o members of c a t e g o r i e s , i .e., t o instances. In The f i n a l paradigmatic r e l a t i o n is manif es-t-ationThis corresponds tot h e phenomenon of o b j e c t constancy: An o b j e c t may undergo change i n space and t i m e , but i t i s still perceived a s t h e same o b j e e t. For -ampla, W i l l i a m Proxmire b e f o r e apd a f t e r h i s h a i r t r a n s p l a n t i s s t i l l W i l l i a m Proxmire. Also an o b j e c t may part i c i p a t e i n many d i f f e r e n t a c t i o n s but s t i l l p r e s e r v e i t s i d e n t i t y , e.g., M a a l f e s t a t i a n s of v a r i e t a l and typical concepts a r e also p o s s i b l e .The latter are used for p r o~e r t i e s t h a t are t r u e o f t h e concept b u t o n l y a t spme p o i n t or period of time, f o r example, "vertebrates are horn", F i g u r e 2. For t y p i c a l a r c s t h i s n o t i o n is redundant a s t y p i c a l embodies s p a t i a l and temporal inde tgrminancy . However m a n i f e s t a t i o n does h a v e a use with the t y p i c a l a r c in representirrg ~o r e f e r e n c e . Suppose i t can happen that a person can t r i p causing him t o b e h u r t .The Uperaon" i n t h e encoded event is a t y p i c a l i s e d "person", but i t must b e t h e same person t h a t t r i p s t h a t i s h u r t . Figure 2 shows the use of m a n i f e s t a t i o n relations t o i n d i c a t e t h i s i d e n t i t y . More w i l l b e s a i d l a t e r about t h e formal r e p r e s e n t a t i o n of t h e c a u s a l r e l a t i o n indicated in h e r i t a n c e " , and so he may clloose t o leave i t a n d r e t u r n l a t e r -Other systems, Q u i l l i a n (1969) , Rumelhart e t a l . (19721, and Schank (1975a) Of t h e f o u r a r c s , v a r i e t y , t y p i c a l , and m a n i f e s t a t i o n can b e i t e r a t e d ; instance c a n n o t . F i g u r e 2and 3 i l l u s t r a t e i t e r a t i v e arrangements of v a r i e t y and m a n i f e s t a t i o n -T h a t t y p i c a l a l The process of matching will be discussed later.The definiendum can also be any concept, the choice is fdiosyncrattc; there i s no reason why t h i s device cannot be used with apparently non-abstract concepts, for example, a dog could be "man's best iriend" for sowk, in contrast with a non-abstrabt d e f i n i t i o n of "canine animal"Wn-abstract d e f i n i t i o n s have the form "genus-specificata".In the encyclapedia, t h e r e p r e s e n t a t i o n i s made up from a node related by v a r i e t y to t h e genus (animal), Subconscious b e l i e f s of self are unmarked i n t h e encyclopedia.Subconscious beliefs of a n o t h e r are i n d i c a t e d by a believe arc between a node r e p r e s e n t i n g t h e believer and a modality node aovering t h e network r e p r e s e n t a i o n of the c o n t e n t o f the b e l i e f s . The subconscious belief of ( 2 ) by "people" is given i n F i g u r e 8. is prevented from being i n h e r i t e d by "penguin" by having e x p l i c i t l y "penguin not fly".The g e n e r a l i t y of i n h e r i t a n c e depends on t h e form of representat i o n of t h e p r o p e r t y a t t h e a n c e s t o r node. P r o p e r t i e s t h a t are univer- ( 3 ) There i s a book that :js read by every scholar, the network form of ( 9 ) , "Marv" i s the agent and caviar" is t h e object i v e of "gobble1'. Flgute 12 encodes (10). (10) Henry t r a v e l s t o o much. H e is g e t t i n g a f o r e i g n a c c e n t .Antecedents may bm nominal, v e r b a l , o r clausal . The second kind of anaphore has a dependent t h a t i s a n a b s t r a c t term f o r the a n t e c e d e n t . involving s p a t i a l ( l e f t , r i g h t , e tc * ) and comnonential (part-whole) o r g a n i e a t i o n .Longacre (1968) n o t e s t h a t i n a g i v e n l a n g u a g e t h e r e i s a f i n i t e happen i n t h e given c i r c u n~s t a n c es .The i n d i c a t i o n s from t h e t e s t i n g of Thorndyke ( 1 9 7 6 ) a r e that.~n i e r e n c e s a r e a p s y c h o l o g i c a l r e a l i t y i n u n d e r s t a n d i n g n a t u r a l l a n g u a g e t e x t s . paradigmatic p a t h p l u s a t y p i c a l arc p l u s m a n i f e s t a t i o n i a followed t o , f o r example, node 2 i n the e n c y c l o p e d i a . R i p p l i p g from "gobble" i n t h e The syntagmatic arcs traversed are noted. From "Marv", node A, a c o n v e r s e paradigmatic p a t h p l u s t y p i c a l p l u s m a n i f e s t a t i o n plus c o n v e r s e a g e n t i s followed, w i t h a Stories 12, 32, and 42 show t h a t t h e success of rescue attempts is a n t i c i p a t e d .-66-( S t o r y 40) On October l l t h , 1974, an u n i d e n t i f i e d man drowned i n his b a t h t u b a t t h e Hotel Sheraton* The drowning was due t o t h e f a c t t h a t h6 f e l l into t h e tub i n t r y i n g t o make himself sobet.( S t o r y 42) An 11 year-old boy drowned today after f a l l i n g i n t o t h e c a n a l where he and his f r i e n d s were playing.The two o t h e r boys, both eleven, I n t i a l s t a t e . (8odes 1, 2, 3, 4 ) . S t e p 1. Fall causes injury. (Node 5 ) .Step 2.Injury causes i n a b i l i t y t o act . (Node 6 ) .S t e p 3. In water and not able to acr: causes fescue. (Node 7 and a l i n k t o node 3 ) . S t e p 4. To rescue someone who is i n the water, get i n t o t h e water. (Nodes 8, 9) S t e p 5. Acting causes weariness. (Node 10).Step 6 . Weariness causes i n a b i l i t y t o act. (Node 11).Step 7. In water and not a b l e t o act causes drowning.(Node 12 and a link to node 4 ) .Note t h a t t h e antcedent condition i n S t e p 3 is t h e same as i n StepBoth r e s u l t a n t s i t u a t i o n s are p o s s i b l e and are notedThe system can select e i t h e r . However, the wrong choice does not lead to a connected s t r u c t u r e and backup t o the a l t e r n a t i v e has t o b e made.After S t e p 7 t h e discourse has an i n f e r r e d caugal s t r u c t u r e connec t i n g all the o r i g i n a l propositions.The theme "tragedy" f i t s , the rescue i s a ( p a r t i a l ) cause of the demisem Rescue is a variety of act and good can apply t o i t and Brown i s a v a r i e t y of d i e . The drowning theme is a l s o present.Although t h e drowning theme i s n o t defined i n terms of the tragedy, i t can be seen t h a t one is properly embedded i n the o t h e r . The Process t h a t performed the a n a l y s i s i s a t present incomplete because the notion of embedding i s nQt well understood f o r the highly s t r u ctured network. The process used the t r a n s i t i v i t y of cause and the conj oining of propositions . Thus Che tragedy encompasses propositions 3, 10, 11, and 4 and the drowning 8, 9, 10, 11, and 4.The tran- ( 1 9 7 6 ) points o u t , one problem w i t h Scripte i s that they are invoked $n t h e i r e n t i a e t y by word association. Thus i t i s suggested t h a t , for example, "I bought some beer from the supermarket, drove home, and drank | null | Main paper:
:
l e v e l , wl-rich hna l e d r6 n g r e a t e r awareness uE t h e r o t e o f meaning" in Iangtxage For "A t e x t is best: regarded a s n SEIAhTXG u n i t ; a u n i t n o t a £form but o f meaningt' ( I l a l l i d a y ct Ilasan, 1976, p. 2 ) . This b e i n g so, discsursc a n a l y s i s will d e e p e n u r t l n d c r s t a n d i n g sf meaning andIn t h i s paper 1: p r e s e n t a rnodel, of nzcaning s t r a n~l y inf'l~lencccl h y Ilays (].969a, 3.969b, 1970, 1973) (1972) , where no particular context is prescrfbed* It will be apparent that a t many p o i n t s t h e present model draws upon these earlier syatems.Some of the differences between systems are probably differences i n notation-However no system I s a t a s t a g e of c o~s t a n e y or completeness that makes it worthwhile to devote much effort to e s t a b l i s h i n g the T h i s membership r e l a t i o n is termed i n s t a n c e ( I S T ) . For example, "William Proxmire" i s an i n s t a n c e of "person", F i g u r e 2. Most i n s t a n c e s are n o t named, t a k h g t h e i r name from t h e i r varietal p a r e n t , but a major exception i s people, ebg., "~e t e r " , "Aunt S a l l y " , Figure 1.Any p a t h through the paradigmatic o r g a n i z a t i o n of knowledge which To r e p r e s e p t t e arguments of occas2onal predications, t h e D i c a l (TYP) arc is used. Thus the "blrd" in "bfrds eat worms" is as in Figure 2 . It i s also possible t o use the typical r e l a t i o n t o a t t a c h occasional proper t i e s t o members of c a t e g o r i e s , i .e., t o instances. In The f i n a l paradigmatic r e l a t i o n is manif es-t-ationThis corresponds tot h e phenomenon of o b j e c t constancy: An o b j e c t may undergo change i n space and t i m e , but i t i s still perceived a s t h e same o b j e e t. For -ampla, W i l l i a m Proxmire b e f o r e apd a f t e r h i s h a i r t r a n s p l a n t i s s t i l l W i l l i a m Proxmire. Also an o b j e c t may part i c i p a t e i n many d i f f e r e n t a c t i o n s but s t i l l p r e s e r v e i t s i d e n t i t y , e.g., M a a l f e s t a t i a n s of v a r i e t a l and typical concepts a r e also p o s s i b l e .The latter are used for p r o~e r t i e s t h a t are t r u e o f t h e concept b u t o n l y a t spme p o i n t or period of time, f o r example, "vertebrates are horn", F i g u r e 2. For t y p i c a l a r c s t h i s n o t i o n is redundant a s t y p i c a l embodies s p a t i a l and temporal inde tgrminancy . However m a n i f e s t a t i o n does h a v e a use with the t y p i c a l a r c in representirrg ~o r e f e r e n c e . Suppose i t can happen that a person can t r i p causing him t o b e h u r t .The Uperaon" i n t h e encoded event is a t y p i c a l i s e d "person", but i t must b e t h e same person t h a t t r i p s t h a t i s h u r t . Figure 2 shows the use of m a n i f e s t a t i o n relations t o i n d i c a t e t h i s i d e n t i t y . More w i l l b e s a i d l a t e r about t h e formal r e p r e s e n t a t i o n of t h e c a u s a l r e l a t i o n indicated in h e r i t a n c e " , and so he may clloose t o leave i t a n d r e t u r n l a t e r -Other systems, Q u i l l i a n (1969) , Rumelhart e t a l . (19721, and Schank (1975a) Of t h e f o u r a r c s , v a r i e t y , t y p i c a l , and m a n i f e s t a t i o n can b e i t e r a t e d ; instance c a n n o t . F i g u r e 2and 3 i l l u s t r a t e i t e r a t i v e arrangements of v a r i e t y and m a n i f e s t a t i o n -T h a t t y p i c a l a l The process of matching will be discussed later.The definiendum can also be any concept, the choice is fdiosyncrattc; there i s no reason why t h i s device cannot be used with apparently non-abstract concepts, for example, a dog could be "man's best iriend" for sowk, in contrast with a non-abstrabt d e f i n i t i o n of "canine animal"Wn-abstract d e f i n i t i o n s have the form "genus-specificata".In the encyclapedia, t h e r e p r e s e n t a t i o n i s made up from a node related by v a r i e t y to t h e genus (animal), Subconscious b e l i e f s of self are unmarked i n t h e encyclopedia.Subconscious beliefs of a n o t h e r are i n d i c a t e d by a believe arc between a node r e p r e s e n t i n g t h e believer and a modality node aovering t h e network r e p r e s e n t a i o n of the c o n t e n t o f the b e l i e f s . The subconscious belief of ( 2 ) by "people" is given i n F i g u r e 8. is prevented from being i n h e r i t e d by "penguin" by having e x p l i c i t l y "penguin not fly".The g e n e r a l i t y of i n h e r i t a n c e depends on t h e form of representat i o n of t h e p r o p e r t y a t t h e a n c e s t o r node. P r o p e r t i e s t h a t are univer- ( 3 ) There i s a book that :js read by every scholar, the network form of ( 9 ) , "Marv" i s the agent and caviar" is t h e object i v e of "gobble1'. Flgute 12 encodes (10). (10) Henry t r a v e l s t o o much. H e is g e t t i n g a f o r e i g n a c c e n t .Antecedents may bm nominal, v e r b a l , o r clausal . The second kind of anaphore has a dependent t h a t i s a n a b s t r a c t term f o r the a n t e c e d e n t . involving s p a t i a l ( l e f t , r i g h t , e tc * ) and comnonential (part-whole) o r g a n i e a t i o n .Longacre (1968) n o t e s t h a t i n a g i v e n l a n g u a g e t h e r e i s a f i n i t e happen i n t h e given c i r c u n~s t a n c es .The i n d i c a t i o n s from t h e t e s t i n g of Thorndyke ( 1 9 7 6 ) a r e that.~n i e r e n c e s a r e a p s y c h o l o g i c a l r e a l i t y i n u n d e r s t a n d i n g n a t u r a l l a n g u a g e t e x t s . paradigmatic p a t h p l u s a t y p i c a l arc p l u s m a n i f e s t a t i o n i a followed t o , f o r example, node 2 i n the e n c y c l o p e d i a . R i p p l i p g from "gobble" i n t h e The syntagmatic arcs traversed are noted. From "Marv", node A, a c o n v e r s e paradigmatic p a t h p l u s t y p i c a l p l u s m a n i f e s t a t i o n plus c o n v e r s e a g e n t i s followed, w i t h a Stories 12, 32, and 42 show t h a t t h e success of rescue attempts is a n t i c i p a t e d .-66-( S t o r y 40) On October l l t h , 1974, an u n i d e n t i f i e d man drowned i n his b a t h t u b a t t h e Hotel Sheraton* The drowning was due t o t h e f a c t t h a t h6 f e l l into t h e tub i n t r y i n g t o make himself sobet.( S t o r y 42) An 11 year-old boy drowned today after f a l l i n g i n t o t h e c a n a l where he and his f r i e n d s were playing.The two o t h e r boys, both eleven, I n t i a l s t a t e . (8odes 1, 2, 3, 4 ) . S t e p 1. Fall causes injury. (Node 5 ) .Step 2.Injury causes i n a b i l i t y t o act . (Node 6 ) .S t e p 3. In water and not able to acr: causes fescue. (Node 7 and a l i n k t o node 3 ) . S t e p 4. To rescue someone who is i n the water, get i n t o t h e water. (Nodes 8, 9) S t e p 5. Acting causes weariness. (Node 10).Step 6 . Weariness causes i n a b i l i t y t o act. (Node 11).Step 7. In water and not a b l e t o act causes drowning.(Node 12 and a link to node 4 ) .Note t h a t t h e antcedent condition i n S t e p 3 is t h e same as i n StepBoth r e s u l t a n t s i t u a t i o n s are p o s s i b l e and are notedThe system can select e i t h e r . However, the wrong choice does not lead to a connected s t r u c t u r e and backup t o the a l t e r n a t i v e has t o b e made.After S t e p 7 t h e discourse has an i n f e r r e d caugal s t r u c t u r e connec t i n g all the o r i g i n a l propositions.The theme "tragedy" f i t s , the rescue i s a ( p a r t i a l ) cause of the demisem Rescue is a variety of act and good can apply t o i t and Brown i s a v a r i e t y of d i e . The drowning theme is a l s o present.Although t h e drowning theme i s n o t defined i n terms of the tragedy, i t can be seen t h a t one is properly embedded i n the o t h e r . The Process t h a t performed the a n a l y s i s i s a t present incomplete because the notion of embedding i s nQt well understood f o r the highly s t r u ctured network. The process used the t r a n s i t i v i t y of cause and the conj oining of propositions . Thus Che tragedy encompasses propositions 3, 10, 11, and 4 and the drowning 8, 9, 10, 11, and 4.The tran- ( 1 9 7 6 ) points o u t , one problem w i t h Scripte i s that they are invoked $n t h e i r e n t i a e t y by word association. Thus i t i s suggested t h a t , for example, "I bought some beer from the supermarket, drove home, and drank
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0.001805 | null | null | null | null | null | null | null | null |
5c31a437cddd3296e76be38091dcb7e43f2310db | 219306983 | null | A Lexicon for a Computer Question-Answering System | An integral part of any natural language understanding sys tern, but one which has received very l i t t l e a t t e n t i o n i n application, i s the lexicon. It i s needed during the parsing of the input text for making inferences, and for generating language output or performing some action. This paper d i s c u s s e s Lhe principal questions concerning the lexicon as i t relates i n particular t o a question-answering system and proposes a specific type of lexicon to f u l f i l l the needs o f t h i s system. Rather than make a d i s t i n c t i o n between dictionary and encyclopedia, w e have a s i n g l e global data base which we c a l l the lexicon. Homographs are differentiated and phrases with fixed meanings are treated as separate entries. A l l the information in this lexicon i s encoded i n the form OZ relations and words or word senses. These form a large network w i t h the words as nodes and the relations as edges. In addition the relations define semsntic f i e l d s and these are used t o treat problems of ambiguity. Relations are use6 to encode both syntactic and s m n t i c information. Axiom schemes are associated with each relation and these are used f o r inferencing. The l e x i c a l r e l a t i o n s then are a t the heart (or brain) o f the system for representation, r e t r i e v a l , and inferencfng. For each relation we describe its semantics and the axioms appropriate t o i t . In the positing of l e x i c a l r e l a t i o n s our a p p r e h has been in£ luenced by the work of 4presfan, Me1 cuk, and Zolkovs~y . The lexical relations we have p o s i t e d are the t r a d i t i o n a l svnonvmy and antonymy, taxonomy, part whole, gradfig and approximately f o r t y o t h e r s . The whole set, deliberately l e f t open ended, is subdiudded i n t o n i n ~ subsets which include attribute relatibns, coLl~cational relations and paradigma ti* onEs. Each relation has its own lexical entry givhng its properties and t e l l i n g how to interpret lexical relationships i n a f i r s t order predicate calculus form. lor example, the tnformatfon for the lextcal entry dog gncludes t h e statement dog T animal, that is, t h a t a dog is a kind of mim~z. The lexical enfry f o r T, the taxonomic relation, i n its turn includes infprmatic , which allows the statement to be interpreted as HoZda(Ncom(dog,X)) -RoZds(lYcorn(anima1,X)). The inventory of relations i s expandable simply by adding l e x i c a l entries for new relations. In a d d i t i o n having both the lexical entries and the relations in the entrdes expressed in the same notat i o n a l form as t h a t of input sentences, namely in a Eirst order predicate calculus notation, allows for a consistent, coherent, and easily modifiable system f o r analysis, inference, | {
"name": [
"Evens, Martha W and",
"Smith, Raoul N"
],
"affiliation": [
null,
null
]
} | null | null | null | 1978-12-01 | 49 | 0 | null | The r e l a t i o n SON was borrowed from the Constituent: X is defined as being a constituent o r part of Y.The example given is cheek-face.Apresyan, M e 1 h k an'd 3olkovsky do not have an e x p l i c i t part-whole r e l a t i o n but they do include tGo r e l a t i o n s in this same area. We have Our current lexicon contains only two axioms f o r the part-whole rel a t i o n . One i s transitivity:i f X PART Y and Y PART Z, then X PART 2.The other, borrowed from Raphael, connects PART and Taxonomy. Essentially i t says t h a t i f a l l X ' s a r e Y ' s and a l l Y 1 s have 2 ' s as parts, then a l l X ' s also have 2 ' 6 as p a r t s . There i s an extensive philosophical l i t e - ". . .with which w e see things" money : I f . . .we buy things with it" (1967 : 175)ratureThe "operational1' c l a s s includes examples in which X is defined as "the characteristic goal or recipient" of action Y bridle: " . . .which they put on horses" (1967 :178) What they c a l l the "spatial" r e l a t i o n a l s o seem t o b e of t h i s same type, grindstone: "...on which a k n i f e is sharpened" (1967:177)Folk d e f i n i t i o n s c o l l e c t e d from speakers of English o f t e n are of t h i s v a r i e t y , sometimes combined with taxonomy, e . g . "a house i s a building i n whfch people reside" (Evens 1975:340) . Children i n p a r t i c u l a r seem Their semantic s t r u c t u r e s are based on grammatical r e l a t i o n s . For verbs these are a subject r e l a t i o n , a d i r e c t object r e l a t i o n , and two kinds of i n d i r e c t object relatiotls. The functions S1, S2, S j , and S4 correspond t o these grammatical r e l a t i o n s . S1 r e l a t e s the verb t o i t s generic subject. The r e s u l t of digging is u s u a l l y a hole; hole TRES'ULT to dig. Wen t h eCubs b e a t t h e Cardinals the Cardinals a r e the l o s e r s ; zoser TCAGENT t o beat2. The thing you sew with is c a l l e d a needle; needte TINST t o , s m . (This i s the Casagrande and Hale operational r e l a t i o n . ) Most p l a n t s sprout from earth.: earth TSOURCE to sprout. One who loves is called a lover; b v e r TEXPR to Zove. People usually bake cakes i n a kitchen;kitclzm m O C t o bake2. It should be noticed that t h e r e l a t i o n TLOC The r e l a t i o n s i n t h i s group, l i k e t h e t y p i c a l case r e l a t i o n s exm amined i n t h e preceding s e c t i o n , are basically ooocurrence r e l a t i o n s .They connect words which cooccur conqtantly and point t o words which have s p e c i a l meanings i n p a r t i c u l a r contexts. This fs a n important pa^ L of t h e l e x i c a l knowledge of the n a t i v e speaker o f t e n neglected i n dict i o n a r i e s . Most of our r e l a t i o n s i n t h i s group are borrowed from the Soviet lexicographers: COPUL, LIQU, PREPAR, DEGRAD. h. Paradigmatic Rehtions .The r e l a t i o n s which we have grouped together as paradigmatic r e l a t i o n s are highly disparate i n kind and importance. CAUSE, BECOME, and Nomare, we believe, essential t o the structlve of the English lexicon; ABLE and @JN seem p o t e n t i a l l y q u i t e useful. There seem t o be very few e m p l e s of BE. /ill except BECOME were influenced by t h e inventory of Apresyan; open1a d j e c t i v e -"the doot is open" open2t o become open1verb i n t r a n s i t t v e -. '%he door opens" open3-to cause t o open2verb transitive -"John opens t h e door" @en is only one of hundreds of verb-adjective homographs i n English. Coo'l behaves like opm. W e start with the adjective coozl, "the j e l l o was cool". me intraneitive wan cool2 means "to become cool " 9 "the j e the r e l a t i o n between czem2 and cZeanl t h e cause to become" r e l a t i o n w i l l be compounded from CAUSE and BECOME4 It will probably occur o f t e n enough t o deserve a name of i t s own, perhaps MAKE. The i n f l e c t i o n a l r e l a t i o n s are, of course, paradigmatic r e l a t i o n s , b u t a r e groupea s e p a r a t e l y because of their s t t o n g family resemblance and p a r t i c u l a r l y u n i n t e r e s t i n g nature. The lexicon i s then used to look f o r cohnections between go and send.The l e x i c a l entry fol; send includes t h e information CAUSE do. The e n t r y f o r t h e l e a r n 1 r e l a t i o n CAUSE contains s e v e r a l axiom schemes. we must f i r s t i d e n t i f y t h e c a t i n t h e s~o-ry, t h a t i s , recognize t h a t a k i t t e n i s a cat and then r e a l i z e t h a t i t is a young one. The l e x i c a l r e l a t i o n CHILD i s e s s e n t i a l t o t h i s task. The d e f i n i t i o n of E t t e n c o n s i s t s of CHLLD cat. The l e x i c a l e n t r y f o r cat contains CHILD k i t t e n .The l e x i c a l entry f o r t h e r e l a t t o n CHILD contains axiom schemes which, when k i t t e n and cat a r e f i l l e d in i n t h e praper places, t e l l us t h a t i f X i s a k i t t e n then i t i s a cat and i t is young. That is, i f IVcom(kitten,X)then Ncom(cat ,X) and P(age,X, young) .In addition some questions force us t o look a t t h e i n t e r a c t i o n between two o r more lexical r e l a t i o n s . To answer the question 6. What animal did P e t e r hear?we need t o know t h a t a meow i s a t y p i c a l c a t soubd, which is expressed by the l e x i c a l r e l a t i o n SON, memo SON oat. W e a l s o need t o know t h a t a c a t i s an animal, cat T anirnaz, and t h a t a k i t t e n is a young cat, a5above k i t t e n CHILD cat. | The Organization of the Lexicon and the SemanticThe l e x i c o n presented h e r e i s being developed a s a n i n t e g r a l p a r t of a computer question-answering system which answers multiple-choice q u e s t i a n s about simple c h i also i h c l u d e s~ T make where T 1s t h e well-known taxonomy r e l a t i o n , so t h a t i f t h e s t o r y says t h a t "Mother baked a cake " w e can1 i n t e r t h a t she inade one add CAUSE bakel s o t h a t w e car deduce that t h e cake has baked he s e l e c t Lon r g s t r i c t i o n s t h a t help us t e l l i n s t a n c e s sf bakel and bdke2apn'rt can a l s o be expressed compactly using t h e T r e l a t i o n . W e a l s o The lexical entry f o r death consJ.sts of NOMV die. There are, as w e l l , l e x i c a l e n t r i e s f o r some m u l t i p l e word expressions such as birthday par ~IJ, baZZ g m , piggy bank, and thank you. The core o f , a l e x i c a l reading copprises a l l and only those semantic s p e c i f i c a t i o n s t h a t determine, roughly speaking, its place w i t h i n t h b system of d i c t i o n a r y e n t r i e s , i. e. d e l j m i t i t from o t h e r (non-synonymous') e n t r i e s . The periphez-8 c o n e l s t s of those m a n t i c s p e c i f i c a t i o n s whikh could be removed from i t s r e a a i n g withour changing i t s r e l a t i o n t c o t h e r l e x i c a l readings w i t h i n t h e same grammar --and thus from t h e encyclopedia t o t h e lexicop. For i n s t a n c e , s b~p o s e a new entry, "2eopard--a l a r g e , wild cat" i s t o be added. The e n t i r e lexicon must be ~ear'ched f o r e n t r i e s which mention large w i l d cass. I f one i s found, s a y "lion--a l a r g e wlld cat", then enough information must be added t h e main types of a c t i J i t i e s connected w i t h skis (a s k i -t r i p , a ski-race . . . I and so on. Even t h e s e scan*yt examples make i t c l e a r t h a t t h e information about the l e x ic a l universe is, a t l e a s t p a r t i a l l y , of an eneyclopaedic nature. W e say "partZallyl' because genuine encyclopaedic information about skis ( t h e i r h i s t o r y , t h e way they are manufactured, etc.) is not supplied hgre: t h e s e c t i o n s contain only such words and phrases as are necessary f o r t a l k i n g on t h e t o p i c , and nothing e l s e . (1970:19,) The problem here is that "what is needed f o r t a l k i n g about t h e toqic" hepends very much on who i s going t o do t h e talking. The d e f i n i t i o nof slEi i n Webster 's Nm tntemationaZ (2nd Edition) begins :One of a p a i r of narrow s t r i p s of wood, metal, o r p l a s t i c , usually i n conibinatian, bound one on each f o o t and used fox gliding over a snow-covered surface.Apresyan, iolkovsky, and Me1 cuk do not provide f o r three of the items,mentioned here: what s k i s a r e made of (wood, p l a s t i c , or metal), what shape they come i n (long and narrow) and where they belong s p a t i a l l y (on t h e f e e t ) . Yet these items could be e s s e n t i a l i n understanding inferences i n a story.It was snowing, Jim took out h i s skis. H e waxed the wooden s t r i p s . . . . This lack of definition cahses tremepdous problem in a dynamic system.A "real" d i c t i o n a r y ar encyclopedia, the one i n a person's b r a i n , is c o n s t a n t l y changing. I n f o m a t i o n i s added, c o r r e c t e d , apd perhaps l o s t .A truly i n t e r e s t i n g memorymodel must be dynamic. The w r k of Kiparsky and Kiparsky (1970) , Lakof f (1971) and KcCdwley, (1968) has shown t h a t syntax and semantics cannot be separated i n t o such n e a t compartments. But i f syntax and semantics a r e interwoven then does i t make sense t o put s y n t a c t i c information i n one box and semadtic informat i o r i n a n o t h e r ? The enswer t o this questgon given a t l e a s t by g e n e r a t i v e semantics c a l l s i n t o question t h e t r a d i t i o n a l distinction between t h e dice".& tionary and t h e encyclopedia.We accept the generative semantikist arguments that syntax a d semant i c s cannot be separated and thus do n u t s e p a r a t e s y n t a c t i c and semantic i n b r m a t i o n . use and how they should b e combined i n l e x i c a l s e n a n t i c s t r u c t u r e s . Two importarit competing models a're provided by componential f e a t u r e a n a l y s i s and by r e l a t i o n a l networks. I n a componential a n a l y s i s model the p r i m e s are semantic features and words are deffried by bundles of features. This is a n a t u r a l extension of t h e d j ~i n c t i v e feature approach t o phoneme d e s c r i p t i o n which h $~; been used to e x p l a i n many phonological phenomena.Certain p r a c t i c a l problems a r i s e . The number of words i n any 1anguage'Fs f a r l a r g e r than t h e number of phonemes. The number of d i s t i n c t i v e feat u r e s which serve t o d i s c r j h i n a t e them must b e l a r g e r too. The wordsemantic feature matrix f o r a given language would be v a s t l y l a r g e r than the phoneme-phonetic f e a t u r e natrix. I n a d d i t i o n , t h i s matrlx woyld b e extremely s p a r s e , Also, i t $8 n o t clear Qhether a l l t h e e n t r i e s i h t h i s malcr* cd'uld be +/as i n a phoneme marix. Axe semantic f e a t u r e s e i t h e r d e f i n i t e l y absent o r b f i n i t e l g p r e s e n t d r a r e some f e a t u r e s p r e s e n t by degrees? The s i z e of the c o m p~n e n t i a l a n a l y s i s matrix would imfnediatelg Ip a r e l a t z a n a l network model, however, t h e primes are r e l a t i o n s and words o r word senses. Relations connect w'osds together i n a network i n wnlch the woPds a r e nodes and t h e r e l a t i o n s a r e edges. In f a c t , wo.rds,are The componential a n a l y s i s model r e q u i r e s t h e discovery of possibly thousaads 0% semantic f e a t u r e s . For a r e l a t i o n a l network m o b 1 an inventc ry of l e x i c a l relatAons and theil; p r o p e r t i e s must be developed. This i s apparent-ly a s i g n i f i c a n t l y simpler task than the discovery of semantic f e a t u r e s , f o r the mooel, t h e i n t r a n s i t i v e pave, must t e l l us about s e l e c t i o n as w e l l as how The hope is tHat t h e system will be a b l e t o "understand" simple metaphors t h i s way. I t would be i n t e r e s t i n g t o t r y t o c r e a t e metaphors by picking noun phrase arpuments c l a w t o but: n o t u n d t r t h e nodcs I n d i c i t t d by the sel e c t i o n information.Words with the same phyeical shape b u t dgffprent meanings constantly cause trouble in n a t u r a l language processing. I n designing a lexicon we must decide whether o r not t o c r e a t e a separate entry f o r each v a r i a t i o n t meaning aid type of use. Quillian is p a r t i c u l a r l y i n t e r e s t e d i n words w'ith multiple meafiings and he experimented with tseveral i n h i s memory model. In Quillian (1968) the word ptmt i s treated as a three-way homonym with three separate ttrpe nodes, each with a s e p a r a t e definition-plane:PLANT Living s t r u c t u r e which is not ah animal, frequently with leaves, g e t t i n g its food from air, water, earth.PUNT2 Apparatus used f o r any process i n industry.W T 3 P u t (seed, p l a n t , etc.) i n e a x h f o r growth.The type node f o r the first forms a d i s j u n c t i v e set with token nodes pointing t o the other twoThe word food has a s i n g l e d e f i n i t i o n with a l t e r n a t i v e formulations:That which J i v i n g being has t o take i n t o keep i t l i v i n g and for growth. Things f omllng meals, e s p e c i a l l y o t h e r than dklnk.A polysemous word l i k e t h i s has a s i n g l e type node and a s i n g l e d e f i n i t i o nplane, but t h e two a l t e r n a t i v e d e f i n i t i o n s a r e combined with an OR l i n k .Apresyan, Me1 'guk and ZolkovsQ a t tack t h e homonymy-polysemy problem with vigor. Graphically coincident worda a r e considered homonyms, given and "come h e l l o r high water" are treated be u n i t s , but "make headway" i s no't. Instead headmy i s defxned as "progress" and marked as appearing Levi t h e underlying structure f o r "birthday boy" i s "boy-have-b5rthday" and the underlying s truc t p r e f o r "Sir thday cake" is "cake-for-bir thday . tld i s t i n c t i v-record(n) , f ieh(v) -f i s h ( n ) ,Then under c e r t a i n conditions h e , for, etc. can be deleted t o give us t h e noun adjunct expressions. Given these r u l e s , she argues, i t is not necessary t o treat these exptessions acCl separate lexical items. While her rules seem sufficient t o allow us t o syntheeiize these compounds c o r r e c t l y , difficlClJties a r i s e when we t r y t o use them for a n a l y s i s . The question-answering system needs t o be able t o infer from "birthday boy" that t h e boy i n question is having a birthday, but t o avoid inferring from "birthday cake" t h a t t h e cake is having a birthday. For correct recognition w e need Noun-noun colppounds have separate e n t r i e s . A birthday cake is treatedas "a cake f o r a birthday." A ball gme is represented as "a game that has a ball". A piggy bank 1 s deflned as "a bank t h a t i s a pj g. I tThe system i s Papago d e f i n i t i o n s . These are "cootdinate" p a i r s like "needle-thread"or "bread-butter", "clang" responses like "table-stable" , o r sequential responses 'bish-bone" and " w h i s t l e -s t~~~~. 'Phey remark about the bread-but ter p a i r that t h e relationship involved between "bread" and "butter" is similar t o t h a t discussed f o r contingency, except that i n Werner Is Lexica2 Re'lations.theThere are two ways to go from the study of f o l k d e f i n i t i o n s . One Most of the r e l a t i o n s given i n these two papers are a s appropriate i n English a s i n Russian. Some, although appropriate i n English, embody more sophistication than seems necessary i n t h i s p l o j e c t . Suppose t h e t e x t says "John wants money" and a question asks "Doeswayi s a synonym of w a n t . The s u b s t i t u t i o n of one f o r t h e other results i n a successful pattern match.The taxonomy r e l a t i o n T i s expressed i n many ways i n English; perhaps "is a kind of" i s t h e most t y p i c a l :A dog is a kind of animal.A dog i s an animal.The notation dog T mhaZ i s used to state 'this r e l a t i o n s h i p . I n t h e lexicon i t is represented by an edge from the dog entry t o the a n k~z e n t r y l a b e l l e d TI Werner's work on t h e taxonomy r e l a t i o n i n memory models has shown t h a t this r e l a t i o n plays a c r u c i a l role i n lexical theory as w e l l as i n ~r a c t i c a l question answering. H e has discussed t h e t h e o r e t i c a l a s p e c t s of the taxonomy r e l a t i o n a t length (Werner 1969 , 1972 , Perchonock and Werner 1969 , Werner and Fenton 1970 and has used it in geveral s t u d i e s Begishe 1969, 1970) . The term taxommy i s the n a t u r a l choice s i n c e it i s now well-known m anthropology.2. ) Synorapy. The synonymy r e l a t i o n poses some d i f f iwit philo-sophicqJ problems. Do two words ever have the same meaning, o r are t h e r e always differences? What criteria can be used t o decide whether two words a r e spnonymourUl Apresyan, ~o l k o v s k y and el ' Euk (1970 : 5) have a t tempted t o state a p r e c i s e c r i t e r i o n : t h e two words should be s a n & c a l l~ sub- has property Adj , then it a l s o has t h e property Not (Adj 2) and v i c e versa.In t h e n o t a t i o n used f o r t h e bemantic representations i n t h e questionanswering system t h i s is s t a t e d :i f , on t h e o t h e r hand, i t has t h e property ~o t the l e x i c a l information husband RECK wife, i.e. i f X1 has X2 a s husbond then X2 has X1 a s wife. The axiom scheme f o r A WCR B says t h a t i f X1 has X2 a s A then X2 has XI as B.HoZdaIR ( Antonymy seems t o be a highly diverse l e x i c a l concept. With f u r t h e r study i t may spawn still more l e x i c a l r e l a t i o n s .Grading r e l a t i o n s l i k e antonymy r e l a t i o n s involve a l t e r n a t i v e s of some kind. Graded alternatives appear t o b e organized i n l i s t s o r o t h e r kinds of formal s t r u c t u r e s . Our c o l l e c t i o n of grading r e l a t i o n s is i n a state of fLux, many a s p e c t s of grading are s t i l l n o t properly defined.1. ) meuzng. The notation Q is borrowed from Werner b u t used i n a very r e s t r i c t e d sense t o connect adjacent items on lists, as i n on day Q h c e s w . It could be read "is immediately followed by." 2.) S e t -e h n e n t . SET r e l a t e s t h e name f o r the set t o t h e name of the elements, e.g. flock SET sheep. This i s t h e r e l a t i o n which the Soviets c a l l Mult, This r e l a t i o n seems t o b e p a r t i c u l a r l y well-founded psycholo-. 3 .) Tern6 for juveniZes. The most common a t t r i b u t e r e l a t i o n i n our vocabulary is CHILD, which r e l a t e s the term f o r t h e offspring t n t h e term f o r i t s parent, as i n puppy CHILD dog, kitten CHILD cat, lamb CHILD sheep. 4.) HaJita*. The habitat r e l a t i o n we have called HOME, so thai Awca HOME t h .The of attribute-value pairs. The l e x i c a l entry f o r puppy contairk CHILD -dog. The l e x i c a l entry f o r dog contains CHILDpuppy. The l e x i c a l e n t r y for the l e x i c a l r e l a t i o n CHILD tells u s how t o i n t e r p r e t these.It contains an axio111 scheme which when f i l l e d i n t e l l s us t h a t X i s a puppy i f and only i f X i s a dog and X i s young Ncbm(puppy,X) means t h a t Ncom(dog,X) and also P(age,X,ygung). Information o f t e n classed as d e r i v a t i o n a l morphology w i l l be included h e r e , t h e l e x i c a l e n t r y for For gzve boch the object and the experiencer may be d e l e t e d . * Sam i s being t a l l .The next item t e l l s us whether 1he verb allows a regular passive o r not. Only thobe whlch do n o t allow a passfve a r e marked. Apresyan, el' Eak, and ?!olkovsky t r e a t this a l s o using l e x i c a l r e l a t i o n s . Eventua l l y t h i s w i l l probably be computable from o t h e r information i n t h e entry. p a r t i c u l a r l y i n use with "woulU l i k e to", "would", " w i l l t t , and "let me".The l a s t item tells whether a verb takes i n d i r e c t question (IQ) . It is probably t h e case that when £ a c t i v i t y and p0rformative s t r u c t u r e are understood, this item w i l l be predictable. The 10 verbs are apparently a l l expositives, but n o t all exposirivee a r e IQ'S and t h e IQ c l a s h i f i c a - SmnpZe entry for bakel:Category: noncopula verb (The cake baked i n the oven.)(The rock baked i n t h e sun.) The second i t e m is t h e selection preference. For red i t is thing;for big i t is t h h g , thought. The s e l e c t i o n preference could probably be s t a t e d once i n t h e e n t r y for the primitive concept and not repeated.Since i t is u s e f u l t o have it r e a d i l y available i n pareing, i t i s included separately i n every a d j e c t i v e entry.With adjectjves as with verbs w e o f t e a have causally r e l a t e d homographs. The z d m i n "warm coat" has a d i f f e r e n t meaning from t h e i n "warm pie." A warm p i e has a temperature greater than room temperature, but a warm coat makes you warm. These a r e c a l l e d warm and W~XPEI I 2 land are connected by CAUSE u-.HOW does one recognize which is which? I f the head noun is cZothing o r one of t h e 'furnace-stove-oven' family o r indeed anything else which has function heat, i s assumed.Adjectives, l i k e verbs, are marked 'Actionyes' o r 'Action -No'Lexical entries €or adverbs are very much l i k e those f o r a d j e c t i v e s .The main s t r a t e g y rfollowed i n the design of t h e l e x i c a l entry has been t o make i t as compact as possible. It seems l i k e l y that more information w i l l have t o be added later. (1) Individual constants of each s o r t .The object constants ate ~i t r e n XI, X2 ....Each corresponds t o a unique object i n the s t o r y . | null | null | null | Main paper:
) ckrmoteristic s o d:
The r e l a t i o n SON was borrowed from the Constituent: X is defined as being a constituent o r part of Y.The example given is cheek-face.Apresyan, M e 1 h k an'd 3olkovsky do not have an e x p l i c i t part-whole r e l a t i o n but they do include tGo r e l a t i o n s in this same area. We have Our current lexicon contains only two axioms f o r the part-whole rel a t i o n . One i s transitivity:i f X PART Y and Y PART Z, then X PART 2.The other, borrowed from Raphael, connects PART and Taxonomy. Essentially i t says t h a t i f a l l X ' s a r e Y ' s and a l l Y 1 s have 2 ' s as parts, then a l l X ' s also have 2 ' 6 as p a r t s . There i s an extensive philosophical l i t e - ". . .with which w e see things" money : I f . . .we buy things with it" (1967 : 175)ratureThe "operational1' c l a s s includes examples in which X is defined as "the characteristic goal or recipient" of action Y bridle: " . . .which they put on horses" (1967 :178) What they c a l l the "spatial" r e l a t i o n a l s o seem t o b e of t h i s same type, grindstone: "...on which a k n i f e is sharpened" (1967:177)Folk d e f i n i t i o n s c o l l e c t e d from speakers of English o f t e n are of t h i s v a r i e t y , sometimes combined with taxonomy, e . g . "a house i s a building i n whfch people reside" (Evens 1975:340) . Children i n p a r t i c u l a r seem Their semantic s t r u c t u r e s are based on grammatical r e l a t i o n s . For verbs these are a subject r e l a t i o n , a d i r e c t object r e l a t i o n , and two kinds of i n d i r e c t object relatiotls. The functions S1, S2, S j , and S4 correspond t o these grammatical r e l a t i o n s . S1 r e l a t e s the verb t o i t s generic subject. The r e s u l t of digging is u s u a l l y a hole; hole TRES'ULT to dig. Wen t h eCubs b e a t t h e Cardinals the Cardinals a r e the l o s e r s ; zoser TCAGENT t o beat2. The thing you sew with is c a l l e d a needle; needte TINST t o , s m . (This i s the Casagrande and Hale operational r e l a t i o n . ) Most p l a n t s sprout from earth.: earth TSOURCE to sprout. One who loves is called a lover; b v e r TEXPR to Zove. People usually bake cakes i n a kitchen;kitclzm m O C t o bake2. It should be noticed that t h e r e l a t i o n TLOC The r e l a t i o n s i n t h i s group, l i k e t h e t y p i c a l case r e l a t i o n s exm amined i n t h e preceding s e c t i o n , are basically ooocurrence r e l a t i o n s .They connect words which cooccur conqtantly and point t o words which have s p e c i a l meanings i n p a r t i c u l a r contexts. This fs a n important pa^ L of t h e l e x i c a l knowledge of the n a t i v e speaker o f t e n neglected i n dict i o n a r i e s . Most of our r e l a t i o n s i n t h i s group are borrowed from the Soviet lexicographers: COPUL, LIQU, PREPAR, DEGRAD. h. Paradigmatic Rehtions .The r e l a t i o n s which we have grouped together as paradigmatic r e l a t i o n s are highly disparate i n kind and importance. CAUSE, BECOME, and Nomare, we believe, essential t o the structlve of the English lexicon; ABLE and @JN seem p o t e n t i a l l y q u i t e useful. There seem t o be very few e m p l e s of BE. /ill except BECOME were influenced by t h e inventory of Apresyan; open1a d j e c t i v e -"the doot is open" open2t o become open1verb i n t r a n s i t t v e -. '%he door opens" open3-to cause t o open2verb transitive -"John opens t h e door" @en is only one of hundreds of verb-adjective homographs i n English. Coo'l behaves like opm. W e start with the adjective coozl, "the j e l l o was cool". me intraneitive wan cool2 means "to become cool " 9 "the j e the r e l a t i o n between czem2 and cZeanl t h e cause to become" r e l a t i o n w i l l be compounded from CAUSE and BECOME4 It will probably occur o f t e n enough t o deserve a name of i t s own, perhaps MAKE. The i n f l e c t i o n a l r e l a t i o n s are, of course, paradigmatic r e l a t i o n s , b u t a r e groupea s e p a r a t e l y because of their s t t o n g family resemblance and p a r t i c u l a r l y u n i n t e r e s t i n g nature. The lexicon i s then used to look f o r cohnections between go and send.The l e x i c a l entry fol; send includes t h e information CAUSE do. The e n t r y f o r t h e l e a r n 1 r e l a t i o n CAUSE contains s e v e r a l axiom schemes. we must f i r s t i d e n t i f y t h e c a t i n t h e s~o-ry, t h a t i s , recognize t h a t a k i t t e n i s a cat and then r e a l i z e t h a t i t is a young one. The l e x i c a l r e l a t i o n CHILD i s e s s e n t i a l t o t h i s task. The d e f i n i t i o n of E t t e n c o n s i s t s of CHLLD cat. The l e x i c a l e n t r y f o r cat contains CHILD k i t t e n .The l e x i c a l entry f o r t h e r e l a t t o n CHILD contains axiom schemes which, when k i t t e n and cat a r e f i l l e d in i n t h e praper places, t e l l us t h a t i f X i s a k i t t e n then i t i s a cat and i t is young. That is, i f IVcom(kitten,X)then Ncom(cat ,X) and P(age,X, young) .In addition some questions force us t o look a t t h e i n t e r a c t i o n between two o r more lexical r e l a t i o n s . To answer the question 6. What animal did P e t e r hear?we need t o know t h a t a meow i s a t y p i c a l c a t soubd, which is expressed by the l e x i c a l r e l a t i o n SON, memo SON oat. W e a l s o need t o know t h a t a c a t i s an animal, cat T anirnaz, and t h a t a k i t t e n is a young cat, a5above k i t t e n CHILD cat.
the form of the lexical entry:
The of attribute-value pairs. The l e x i c a l entry f o r puppy contairk CHILD -dog. The l e x i c a l entry f o r dog contains CHILDpuppy. The l e x i c a l e n t r y for the l e x i c a l r e l a t i o n CHILD tells u s how t o i n t e r p r e t these.It contains an axio111 scheme which when f i l l e d i n t e l l s us t h a t X i s a puppy i f and only i f X i s a dog and X i s young Ncbm(puppy,X) means t h a t Ncom(dog,X) and also P(age,X,ygung). Information o f t e n classed as d e r i v a t i o n a l morphology w i l l be included h e r e , t h e l e x i c a l e n t r y for For gzve boch the object and the experiencer may be d e l e t e d . * Sam i s being t a l l .The next item t e l l s us whether 1he verb allows a regular passive o r not. Only thobe whlch do n o t allow a passfve a r e marked. Apresyan, el' Eak, and ?!olkovsky t r e a t this a l s o using l e x i c a l r e l a t i o n s . Eventua l l y t h i s w i l l probably be computable from o t h e r information i n t h e entry. p a r t i c u l a r l y i n use with "woulU l i k e to", "would", " w i l l t t , and "let me".The l a s t item tells whether a verb takes i n d i r e c t question (IQ) . It is probably t h e case that when £ a c t i v i t y and p0rformative s t r u c t u r e are understood, this item w i l l be predictable. The 10 verbs are apparently a l l expositives, but n o t all exposirivee a r e IQ'S and t h e IQ c l a s h i f i c a - SmnpZe entry for bakel:Category: noncopula verb (The cake baked i n the oven.)(The rock baked i n t h e sun.) The second i t e m is t h e selection preference. For red i t is thing;for big i t is t h h g , thought. The s e l e c t i o n preference could probably be s t a t e d once i n t h e e n t r y for the primitive concept and not repeated.Since i t is u s e f u l t o have it r e a d i l y available i n pareing, i t i s included separately i n every a d j e c t i v e entry.With adjectjves as with verbs w e o f t e a have causally r e l a t e d homographs. The z d m i n "warm coat" has a d i f f e r e n t meaning from t h e i n "warm pie." A warm p i e has a temperature greater than room temperature, but a warm coat makes you warm. These a r e c a l l e d warm and W~XPEI I 2 land are connected by CAUSE u-.HOW does one recognize which is which? I f the head noun is cZothing o r one of t h e 'furnace-stove-oven' family o r indeed anything else which has function heat, i s assumed.Adjectives, l i k e verbs, are marked 'Actionyes' o r 'Action -No'Lexical entries €or adverbs are very much l i k e those f o r a d j e c t i v e s .The main s t r a t e g y rfollowed i n the design of t h e l e x i c a l entry has been t o make i t as compact as possible. It seems l i k e l y that more information w i l l have t o be added later. (1) Individual constants of each s o r t .
summary:
The object constants ate ~i t r e n XI, X2 ....Each corresponds t o a unique object i n the s t o r y .
5.:
The Organization of the Lexicon and the SemanticThe l e x i c o n presented h e r e i s being developed a s a n i n t e g r a l p a r t of a computer question-answering system which answers multiple-choice q u e s t i a n s about simple c h i also i h c l u d e s~ T make where T 1s t h e well-known taxonomy r e l a t i o n , so t h a t i f t h e s t o r y says t h a t "Mother baked a cake " w e can1 i n t e r t h a t she inade one add CAUSE bakel s o t h a t w e car deduce that t h e cake has baked he s e l e c t Lon r g s t r i c t i o n s t h a t help us t e l l i n s t a n c e s sf bakel and bdke2apn'rt can a l s o be expressed compactly using t h e T r e l a t i o n . W e a l s o The lexical entry f o r death consJ.sts of NOMV die. There are, as w e l l , l e x i c a l e n t r i e s f o r some m u l t i p l e word expressions such as birthday par ~IJ, baZZ g m , piggy bank, and thank you. The core o f , a l e x i c a l reading copprises a l l and only those semantic s p e c i f i c a t i o n s t h a t determine, roughly speaking, its place w i t h i n t h b system of d i c t i o n a r y e n t r i e s , i. e. d e l j m i t i t from o t h e r (non-synonymous') e n t r i e s . The periphez-8 c o n e l s t s of those m a n t i c s p e c i f i c a t i o n s whikh could be removed from i t s r e a a i n g withour changing i t s r e l a t i o n t c o t h e r l e x i c a l readings w i t h i n t h e same grammar --and thus from t h e encyclopedia t o t h e lexicop. For i n s t a n c e , s b~p o s e a new entry, "2eopard--a l a r g e , wild cat" i s t o be added. The e n t i r e lexicon must be ~ear'ched f o r e n t r i e s which mention large w i l d cass. I f one i s found, s a y "lion--a l a r g e wlld cat", then enough information must be added t h e main types of a c t i J i t i e s connected w i t h skis (a s k i -t r i p , a ski-race . . . I and so on. Even t h e s e scan*yt examples make i t c l e a r t h a t t h e information about the l e x ic a l universe is, a t l e a s t p a r t i a l l y , of an eneyclopaedic nature. W e say "partZallyl' because genuine encyclopaedic information about skis ( t h e i r h i s t o r y , t h e way they are manufactured, etc.) is not supplied hgre: t h e s e c t i o n s contain only such words and phrases as are necessary f o r t a l k i n g on t h e t o p i c , and nothing e l s e . (1970:19,) The problem here is that "what is needed f o r t a l k i n g about t h e toqic" hepends very much on who i s going t o do t h e talking. The d e f i n i t i o nof slEi i n Webster 's Nm tntemationaZ (2nd Edition) begins :One of a p a i r of narrow s t r i p s of wood, metal, o r p l a s t i c , usually i n conibinatian, bound one on each f o o t and used fox gliding over a snow-covered surface.Apresyan, iolkovsky, and Me1 cuk do not provide f o r three of the items,mentioned here: what s k i s a r e made of (wood, p l a s t i c , or metal), what shape they come i n (long and narrow) and where they belong s p a t i a l l y (on t h e f e e t ) . Yet these items could be e s s e n t i a l i n understanding inferences i n a story.It was snowing, Jim took out h i s skis. H e waxed the wooden s t r i p s . . . . This lack of definition cahses tremepdous problem in a dynamic system.A "real" d i c t i o n a r y ar encyclopedia, the one i n a person's b r a i n , is c o n s t a n t l y changing. I n f o m a t i o n i s added, c o r r e c t e d , apd perhaps l o s t .A truly i n t e r e s t i n g memorymodel must be dynamic. The w r k of Kiparsky and Kiparsky (1970) , Lakof f (1971) and KcCdwley, (1968) has shown t h a t syntax and semantics cannot be separated i n t o such n e a t compartments. But i f syntax and semantics a r e interwoven then does i t make sense t o put s y n t a c t i c information i n one box and semadtic informat i o r i n a n o t h e r ? The enswer t o this questgon given a t l e a s t by g e n e r a t i v e semantics c a l l s i n t o question t h e t r a d i t i o n a l distinction between t h e dice".& tionary and t h e encyclopedia.We accept the generative semantikist arguments that syntax a d semant i c s cannot be separated and thus do n u t s e p a r a t e s y n t a c t i c and semantic i n b r m a t i o n . use and how they should b e combined i n l e x i c a l s e n a n t i c s t r u c t u r e s . Two importarit competing models a're provided by componential f e a t u r e a n a l y s i s and by r e l a t i o n a l networks. I n a componential a n a l y s i s model the p r i m e s are semantic features and words are deffried by bundles of features. This is a n a t u r a l extension of t h e d j ~i n c t i v e feature approach t o phoneme d e s c r i p t i o n which h $~; been used to e x p l a i n many phonological phenomena.Certain p r a c t i c a l problems a r i s e . The number of words i n any 1anguage'Fs f a r l a r g e r than t h e number of phonemes. The number of d i s t i n c t i v e feat u r e s which serve t o d i s c r j h i n a t e them must b e l a r g e r too. The wordsemantic feature matrix f o r a given language would be v a s t l y l a r g e r than the phoneme-phonetic f e a t u r e natrix. I n a d d i t i o n , t h i s matrlx woyld b e extremely s p a r s e , Also, i t $8 n o t clear Qhether a l l t h e e n t r i e s i h t h i s malcr* cd'uld be +/as i n a phoneme marix. Axe semantic f e a t u r e s e i t h e r d e f i n i t e l y absent o r b f i n i t e l g p r e s e n t d r a r e some f e a t u r e s p r e s e n t by degrees? The s i z e of the c o m p~n e n t i a l a n a l y s i s matrix would imfnediatelg Ip a r e l a t z a n a l network model, however, t h e primes are r e l a t i o n s and words o r word senses. Relations connect w'osds together i n a network i n wnlch the woPds a r e nodes and t h e r e l a t i o n s a r e edges. In f a c t , wo.rds,are The componential a n a l y s i s model r e q u i r e s t h e discovery of possibly thousaads 0% semantic f e a t u r e s . For a r e l a t i o n a l network m o b 1 an inventc ry of l e x i c a l relatAons and theil; p r o p e r t i e s must be developed. This i s apparent-ly a s i g n i f i c a n t l y simpler task than the discovery of semantic f e a t u r e s , f o r the mooel, t h e i n t r a n s i t i v e pave, must t e l l us about s e l e c t i o n as w e l l as how The hope is tHat t h e system will be a b l e t o "understand" simple metaphors t h i s way. I t would be i n t e r e s t i n g t o t r y t o c r e a t e metaphors by picking noun phrase arpuments c l a w t o but: n o t u n d t r t h e nodcs I n d i c i t t d by the sel e c t i o n information.Words with the same phyeical shape b u t dgffprent meanings constantly cause trouble in n a t u r a l language processing. I n designing a lexicon we must decide whether o r not t o c r e a t e a separate entry f o r each v a r i a t i o n t meaning aid type of use. Quillian is p a r t i c u l a r l y i n t e r e s t e d i n words w'ith multiple meafiings and he experimented with tseveral i n h i s memory model. In Quillian (1968) the word ptmt i s treated as a three-way homonym with three separate ttrpe nodes, each with a s e p a r a t e definition-plane:PLANT Living s t r u c t u r e which is not ah animal, frequently with leaves, g e t t i n g its food from air, water, earth.PUNT2 Apparatus used f o r any process i n industry.W T 3 P u t (seed, p l a n t , etc.) i n e a x h f o r growth.The type node f o r the first forms a d i s j u n c t i v e set with token nodes pointing t o the other twoThe word food has a s i n g l e d e f i n i t i o n with a l t e r n a t i v e formulations:That which J i v i n g being has t o take i n t o keep i t l i v i n g and for growth. Things f omllng meals, e s p e c i a l l y o t h e r than dklnk.A polysemous word l i k e t h i s has a s i n g l e type node and a s i n g l e d e f i n i t i o nplane, but t h e two a l t e r n a t i v e d e f i n i t i o n s a r e combined with an OR l i n k .Apresyan, Me1 'guk and ZolkovsQ a t tack t h e homonymy-polysemy problem with vigor. Graphically coincident worda a r e considered homonyms, given and "come h e l l o r high water" are treated be u n i t s , but "make headway" i s no't. Instead headmy i s defxned as "progress" and marked as appearing Levi t h e underlying structure f o r "birthday boy" i s "boy-have-b5rthday" and the underlying s truc t p r e f o r "Sir thday cake" is "cake-for-bir thday . tld i s t i n c t i v-record(n) , f ieh(v) -f i s h ( n ) ,Then under c e r t a i n conditions h e , for, etc. can be deleted t o give us t h e noun adjunct expressions. Given these r u l e s , she argues, i t is not necessary t o treat these exptessions acCl separate lexical items. While her rules seem sufficient t o allow us t o syntheeiize these compounds c o r r e c t l y , difficlClJties a r i s e when we t r y t o use them for a n a l y s i s . The question-answering system needs t o be able t o infer from "birthday boy" that t h e boy i n question is having a birthday, but t o avoid inferring from "birthday cake" t h a t t h e cake is having a birthday. For correct recognition w e need Noun-noun colppounds have separate e n t r i e s . A birthday cake is treatedas "a cake f o r a birthday." A ball gme is represented as "a game that has a ball". A piggy bank 1 s deflned as "a bank t h a t i s a pj g. I tThe system i s Papago d e f i n i t i o n s . These are "cootdinate" p a i r s like "needle-thread"or "bread-butter", "clang" responses like "table-stable" , o r sequential responses 'bish-bone" and " w h i s t l e -s t~~~~. 'Phey remark about the bread-but ter p a i r that t h e relationship involved between "bread" and "butter" is similar t o t h a t discussed f o r contingency, except that i n Werner Is Lexica2 Re'lations.theThere are two ways to go from the study of f o l k d e f i n i t i o n s . One Most of the r e l a t i o n s given i n these two papers are a s appropriate i n English a s i n Russian. Some, although appropriate i n English, embody more sophistication than seems necessary i n t h i s p l o j e c t . Suppose t h e t e x t says "John wants money" and a question asks "Doeswayi s a synonym of w a n t . The s u b s t i t u t i o n of one f o r t h e other results i n a successful pattern match.The taxonomy r e l a t i o n T i s expressed i n many ways i n English; perhaps "is a kind of" i s t h e most t y p i c a l :A dog is a kind of animal.A dog i s an animal.The notation dog T mhaZ i s used to state 'this r e l a t i o n s h i p . I n t h e lexicon i t is represented by an edge from the dog entry t o the a n k~z e n t r y l a b e l l e d TI Werner's work on t h e taxonomy r e l a t i o n i n memory models has shown t h a t this r e l a t i o n plays a c r u c i a l role i n lexical theory as w e l l as i n ~r a c t i c a l question answering. H e has discussed t h e t h e o r e t i c a l a s p e c t s of the taxonomy r e l a t i o n a t length (Werner 1969 , 1972 , Perchonock and Werner 1969 , Werner and Fenton 1970 and has used it in geveral s t u d i e s Begishe 1969, 1970) . The term taxommy i s the n a t u r a l choice s i n c e it i s now well-known m anthropology.2. ) Synorapy. The synonymy r e l a t i o n poses some d i f f iwit philo-sophicqJ problems. Do two words ever have the same meaning, o r are t h e r e always differences? What criteria can be used t o decide whether two words a r e spnonymourUl Apresyan, ~o l k o v s k y and el ' Euk (1970 : 5) have a t tempted t o state a p r e c i s e c r i t e r i o n : t h e two words should be s a n & c a l l~ sub- has property Adj , then it a l s o has t h e property Not (Adj 2) and v i c e versa.In t h e n o t a t i o n used f o r t h e bemantic representations i n t h e questionanswering system t h i s is s t a t e d :i f , on t h e o t h e r hand, i t has t h e property ~o t the l e x i c a l information husband RECK wife, i.e. i f X1 has X2 a s husbond then X2 has X1 a s wife. The axiom scheme f o r A WCR B says t h a t i f X1 has X2 a s A then X2 has XI as B.HoZdaIR ( Antonymy seems t o be a highly diverse l e x i c a l concept. With f u r t h e r study i t may spawn still more l e x i c a l r e l a t i o n s .Grading r e l a t i o n s l i k e antonymy r e l a t i o n s involve a l t e r n a t i v e s of some kind. Graded alternatives appear t o b e organized i n l i s t s o r o t h e r kinds of formal s t r u c t u r e s . Our c o l l e c t i o n of grading r e l a t i o n s is i n a state of fLux, many a s p e c t s of grading are s t i l l n o t properly defined.1. ) meuzng. The notation Q is borrowed from Werner b u t used i n a very r e s t r i c t e d sense t o connect adjacent items on lists, as i n on day Q h c e s w . It could be read "is immediately followed by." 2.) S e t -e h n e n t . SET r e l a t e s t h e name f o r the set t o t h e name of the elements, e.g. flock SET sheep. This i s t h e r e l a t i o n which the Soviets c a l l Mult, This r e l a t i o n seems t o b e p a r t i c u l a r l y well-founded psycholo-. 3 .) Tern6 for juveniZes. The most common a t t r i b u t e r e l a t i o n i n our vocabulary is CHILD, which r e l a t e s the term f o r t h e offspring t n t h e term f o r i t s parent, as i n puppy CHILD dog, kitten CHILD cat, lamb CHILD sheep. 4.) HaJita*. The habitat r e l a t i o n we have called HOME, so thai Awca HOME t h .
Appendix:
| null | null | null | null | {
"paperhash": [
"litowitz|learning_to_make_definitions",
"mccawley|lexicography_and_the_count-mass_distinction",
"nash-webber|semantics_and_speech_understanding",
"raphael|sir:_a_computer_program_for_semantic_information_retrieval"
],
"title": [
"Learning to make definitions",
"Lexicography and the Count-mass Distinction",
"Semantics and Speech Understanding",
"SIR: A COMPUTER PROGRAM FOR SEMANTIC INFORMATION RETRIEVAL"
],
"abstract": [
"ABSTRACT The ability to define words by means of other words, which forms a part of many standardized tests, must be learned by a child. The nature of the ‘definitional task’ and the development of language responses (from children 4; 5 to 7; 5) are discussed in terms of a linguistic analysis of the definitional form and its semantic relations. Progress in definitional strategies by children moves along two continua: conceptually from the individually experiential to the socially shared; and syntactically from actual predicates through hypothetical predicates to adult definitional sentence frames. Implications include novel elicitation techniques, psycholinguistically informed evaluation of direction and scoring measures of some common standardized verbal tests, and better understanding of the range of normal development on one specific language task.",
"Proceedings of the First Annual Meeting of the Berkeley Linguistics \nSociety (1975), pp. 314-321",
"Abstract : Syntactic constraints and expectations are based on the patterns formed by a given set of linguistic objects, e.g. nouns, verbs, adjectives, etc. Pragmatic ones arise from notions of conversational structure and the types of linguistic behavior appropriate to a given situation. The bases for semantic constraints and expectations are an a priori sense of what can be meaningful and the ways in which meaningful concepts can be realized in actual language. The paper describes how semantics is being used in several recent speech understanding systems. It then expands the generalities of the first section with a detailed discussion of some actual problems that have arisen in the attempt to understand speech.",
"SIR is a computer system, programmed in the LISP language, which accepts information and answers questions expressed in a restricted form of English. This system demonstrates what can reasonably be called an ability to \"understand\" semantic information. SIR''s semantic and deductive ability is based on the construction of an internal model, which uses word associations and property lists, for the relational information normally conveyed in conversational statements. A format-matching procedure extracts semantic content from English sentences. If an input sentence is declarative, the system adds appropriate information to the model. If an input sentence is a question, the system searches the model until it either finds the answer or determines why it cannot find the answer. In all cases SIR reports its conclusions. The system has some capacity to recognize exceptions to general rules, resolve certain semantic ambiguities, and modify its model structure in order to save computer memory space. Judging from its conversational ability, SIR is more \"intelligent\" than any existing question-answering system. The author describes how this ability was developed and how the basic features of SIR compare with those of other systems. The working system, SIR , is a first step toward intelligent machine communication. The author proposes a next step by describing how to construct a more general system which is less complex and yet more powerful than SIR . This proposed system contains a generalized version of the SIR model, a formal logical system called SIR1 , and a computer program for testing the truth of SIR1 statements with respect to the generalized model by using partial proof procedures in the predicate calculus. The thesis also describes the formal properties of SIR1 and how they relate to the logical structure of SIR ."
],
"authors": [
{
"name": [
"B. Litowitz"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"J. McCawley"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"B. Nash-webber"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"B. Raphael"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null
],
"s2_corpus_id": [
"146693021",
"60191456",
"60296761",
"62082148"
],
"intents": [
[],
[],
[],
[]
],
"isInfluential": [
false,
false,
false,
false
]
} | null | 554 | 0 | null | null | null | null | null | null | null | null |
7e2e03b501568967eb989853c412e72316a73fe4 | 198183589 | null | The Relation of Grammar to Cognition--a Synopsis | A sentence (or other portion of discourse) i s taken to evoke in the 1 is tener a meaning compl ex, here called a "cognitive representation". The lexical elements of the sentence, t o simpli$y, by and large specify the content of the cognitive representation, while the gramatical elements specify i t s structure. Thus, looking systematically a t the actual notions specified by grammatical elements can 9 ive us a handle for ascertaining the very makeup of | {
"name": [
"Talmy, Leonard"
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 0 | 1 | null | A sentence (or other portion of discourse) i s taken to evoke i n the listener a particular kind o f experiential compl ex--here t o be termed a "cognitive representation" or " C P U . l There appears to be a significant way i n which different portions of the language i n p u t specify, g r codk for, different portions of the CR. The major finding, is that--for a f i r s t approximation--the 1 mica1 fraction of a sentence codes mainly for the content, or substance, of a CRY while the grammatical fraction of a sentence codes mainly for the structure of a CR. Determining the structure within a realm of phenomena has been a central concern for analytic science, including 1 inguist i c s and psychology. With grammar seen in the above light, i t can be used in determining the structure, of the 1 anguage-re1 ated portion of human cogrTition, w i t h possi bl e connections t o further poreons. In particular, looking systematically a t the actual notions specified by grammatical elements can give us a handle for ascertaining the ery nakeup of (1 f nguist i c -) cogni tive structuring.!The beqinnings of such a n endeavor are the aims of this paper Several ideas here require some immediate elaboration. The distinction between lexical and grammatical i s made entirely formally--i,e., without any reference t o meaning--on the basis of the distinction between open-cl ass and cl osed-cl ass. 3 A1 1 openclass elem nts--i . e . , the stems of nouns, verbs, and B adjectives --are considered lexical. Everything else i s considered grammatical. Included here are a1 1 closed-cl ass morphemes and words--infl ections , p a rt iclles, adposi tons, conjunctions, demonstratives , etc. --as we1 1 as syntactic constructions, grama tical re1 ations , categorial identi t i e s , word order, and intonation. Terminological ly here, "grammatical element" wi 11 be used to refer t o any o f these.The nature of content and of structure, and the distinction between them, are not understood we1 1 enough t o be addressed analytically i n this paper and must be l e f t t o our intuitive sense of the matter. 5 Taking them for granted, however, we can now more finely characterize the linguistic-cognitive crossrelationships noted earlien While most of a CR's content i s specified by the lexical fraction of a sentence, the lexical items do usually specify some structural notions along with the contentful ones. The gramatical elements Of a sentence more unalloyedly specify only structural notions and specify them more determinately in the case of conflict w i t h a lexical item, e tab1 ishing perhaps the majority of a CR' s structure. 8In other work in the present di recti on--notab1 y Fillmore's {e.g. , 1975, 1976 )--concern has also been with ascertaining structre, b u t the sentence elements used as starting-points have generally been lexical i terns with prominently i nmi xed structural specif ications ( l i k e buy and s e l l ) . The present work, i n part a complement t o the othet, takes advantage of grammar's greater directness and completeness in specifying structure.This paper i s divided into three sections. In the f i r s t , a sampling of grammatical elements i s examined for the notions that they specify, both as an introduction to out method and for the aim of notici n g properties common to such notions as we1 1 as properties excluded from them. In the second, we present a number of the categories i n which grammatically specified noSions have been observed t o pattern. In the third, we speculate on broader cognitve connections.1. The Nature o f Gramnatically Specified Notions I n t h i s section we examine a small sampling o f grammatical elements f o r the p a r t i c u l a r component notions t h a t they specjfy. The sample w i l l give a h e u r i s t i c indication o f the kinds of notions t h a t get grammatically specified as well as o f kinds o f notions t h a t possibly never do. The excluded kinds w i l l be seen as r e a d i l y specifiable by l e x i c a l elements. A further comparison between the characteri s t i c s o f gramnatically specified notions and o f l e x i c a l l y specified ones I s then made. To indicate the major finding a t the outset, i t seems that-grammatical specifications f o r structure are prepon'dera n t l y r e l a t i v i s t i c o r topological, and exclude the f i x e d o r m e t r i c a l l y Euclidean.For a f i r s t simple case, many languages have inf l e c t i o n s for the noun (Engl i s h has -b and -5 ) t h a t specify the u n i~l e x or the m u l t i p~e x instaztiat i o n o f t h e object specified by the noun. By con-' t r a s t , no languages appear t o have i n f l e c t i o n s t h a t speci fy the redness o r b l ueness , etc. -i . e. , the part i c u l a r color--of the object specified by a noun.-.i n the preceding, the underlined are instances o f "notions". the speaker-side o r the non-speaker-side o f a concept u a l p a r t i t i o n drawn through space (or time o r other qua1 i t a t i ve dimension). This i n t e g r a l specification can be analyzed as containing the following component notions (ehcl osed by quotes) :(1 a-b. a ' p a r t i t i o n ' t h a t divides a space i n t o 'regionst/'sides' c-e. the ' locatednes's' (a p a r t i c u l a r r e l a t i o n ) o f a 'point' ( o r object idealizable as a point) 'within' a region f-g. (a side t h a t i s the) 'same' as o r ' d i f f e r e n t ' from h-i. a 'currently indicated' object and a 'currently communicating' e n t i t yNotions t h a t might a t f i r s t be ascribed t o such deict i c s , such as of distance o r perhaps size, prove not t o be, on the evidence o f sentence-pairs 1 i k e (2):(2) a. This speck i s smaller than t h a t speck. b. This planet i s smaller than t h a t planet.The CRs evoked by (2a) and (b) d i f f e r greatly, i nvol ving t i n y objects m i 1 1 imeters apart o r huge objects parsecs apart. Yet the sentences d i f f e r only l e x i ca l l y , not g r a m a t i c a l l y * Hence, the CRs' notions as t o the magnitude o f size o r distsence cannot be traced t o the d e i c t i c s (or t o other gramnatical elements) i n the sentences. Thus, the notional specifications o f a t h i s o r a t h a t appear, i n part, t o be genuinely topologicbl : the establishment of a p a r t i t i o n remains a constant, but i t s p o s i t i o n can vary unlimitedly (or, using topology's characterizabil i t y as "rubber-sheet geometry", the p a r t i t i o n ' s distance away can be (3) a. I wal ked through the water. b. I walked through the timeber ( i . e . p , woods).In t h i s usage, through specifies, broadly, 'motion along a l i n e t h a t i s w i t h i n a medium'. The component notions contained here include:4 ) 1 a-e.motion '--i . e., 'one-to-one correspondences'between 'adjacent' points of ' space' and adjacent points o f ' t i m e ' f. motion t h a t describes a "Line' g.t h e l o c a t e d n e s s o f a l i n e w i t h i n a 'medium' h-i. a medium, i.e., a region o f ekree-dimensional space s e t apart by the lomtedness w i t h in i t d f 'material' t h a t i s i n a pattern o f dis- t r i b u t i o ? ' o f a certain range o f character ( s t i 1 1 t o be determined)Again, w i t h (3a) and ( i t can be f u r t h e r determined t h a t ' r a t e of motion' and 'shape/contour o f 1 inear path' are also n o t specified by the gramnatical element. As one s t e p i n a program t o ascertain any propert i e s comnon t o gramnatical l y specified notions, the notions j u s t found are gathered together i n Table 1 . For h e u r i s t i c purposes, the notions are very provisi o n a l l y divided i n t a three groups on the basis o f t h e i r r e l a t i o n t o topology. I n group (a) are the notions t h a t properly be1 ong , o r are r e a d i l y def inabl e, i n the actual mathematical system o f topology. I n group (b), the notions might not be p a r t o f topology proper but intuitively'seem l i k e those t h a t are--and might be includable i n a related mathematical system t h a t could be constructed. Pn group (c) are the not i o n s t h a t f a l l outside o f any usual conception o f a mathemat'ical system. The number of notions i n the f i r s t two groups combined i s 13, while t h e t h i r d has 6--an i n d i c a t i o n o f a preponderant propensity f o r gramnati cal elements t o specify quasi -topological notions. The ratSo i n t h i s d i r e c t i o n i s i n f a c t improved i f we consider t h a t even several n o t i~n s~i n group (c)--the bottom three--resemble topological ones i n t h e sense o f involving r e l a t i v i s t i c relationships between quantities r a t h e r than absol u t e l y f i x e d quantities.(7) Table 1: Some notions found t o be specified d. one-to-one matter ,correspondences space tSme motion med i urn c u r r e n t l y indicated/ comnunicating e n t i t yFor a compl ementary program o f ascertaining any properties excluded from gramnatical specification, the notions found above n o t t o be specified by the .elerhents investigated are 1 i s t e d i n Table 2 . Rather than topological, topologf-like, o r r e l a t i v i s t i c , these notions involve Eucl idean-geometric concepts (e.g., set distance, Size, contour), q u a n t i f i e d measure, and various p a r t i c u l a r i t i e s o f a quantity--in sum, c h a r a c t e r i s t i c s t h a t a r e absolute o r fixed. These grammatical -1 e x i c a l differences can be s e t i n t o f u r t h e r r e l i e f by i n t u r n varying one elementtype while keeping the other constant. Thus, varying o n l y the gramnatical elements o f ( 9 ) , as i s done i n (13) A machine cancelled t h e stamps.The preceding sampLing of g r a m a t i c a l elements has y i e l d e d a set o f natisns helpful toward discovering comnon properties. But t h e s e t has been small and haphazardly a r r i v e d at. With a broader and more systematic investigation, patterns of organization become evident. Gramnatically s p e c i f i e d notions can be seen t o p a t t e r n Jn categories, and the categories, i n turn, i n integrated systems. I n t h i s section we look a t some o f these categories and systems, The grammatical elements here w i l l not be treated i n isolation, but i n associa%ion w i t h l e x i c a l items. That i s , the grammatically specified structural not i o n s w i l l be considered i n i n t e r a c t i o n w i t h t h a t portiop o f l e x i c a l specification t h a t i s a1 so struct u r a l . This interaction e n t a i l s cognitive processing, and d i f f e r e n t cases o f such processing w i l l be considered along the way.The note on methodology should be made t h a t our d i r e c t i o n o f analysis has been from grammatical speci f i c a t i~n t o category, not the reverse. That i s , the categories considered be1 ow were discovered t o be re1 evant t o the specifications o f various grammatical elements. They were not part o f some a priori concept u a l schema which then sought corrobovative exampl es.The category o f "dimension" has two member notions, 'space' and ' time'. The kind o f "quantity" t h a t exists i n space i s --i n respectively continuous o r discrete form--'matter8 o r 'objects'. The kind o f quantity e x i s t i n g i n time i s 'action' o r 'events' ("action" i s meant t o refeF t o any obtaining circumstance not j u s t ( w i l l e d ) motion). I n tabular form, these notions r e l a t e thus: 13space: matter/objects time:action/events A number of grammatical and l e x i c a l referents are specific w i t h regard t o one o r the other pole o f t h i s category. But s i~c e the category cross-cuts the ones t r e d e d next, we w i l l not exempl i f y i t here but w i l l endeavor i n the following, t o present both space and time examples side by side.The category here t o be termed "plexityaL i s a quantity's s t a t e o f a r t i c u l a t i o n i n t o equivalent elements. Where the quantity consists o f only one such element, i t i s "uniplex", and where i t consists o f more than one, i t i s "multiplex". When the quantity involved i s m a t e r , p l e x i t y is, o f course, equivalent t o the t r a d i t i o n a l category o f "number" w i t h i t s component notions "singulsr" and "plural". But the present notions are intended t o capture the generalizat i o n f r o m matter over t o action, which the t r a d i t i o n a l ones do not.9Specifications as t o p l e x i t y are made by both l e x i c a l items and gramnatical elements, and the interplay between the two when they are both i n associat i o n must be noted. Example English l e x i c a l items t h a t basically specify a unipl ex referent are--for matter and action, respectively--birdand ( t o ) sigh. They q n occur w i t h gramnatical elements t h a t themselves specify a uniplexity, 1 i k e those under1 ined i n (14a) (many languages have here a more regular, overt system o f markers than English). But they can a1 so occur w i t h gramnatical elements t h a t specify a m u l t i p l e x i t y , as i n (14b). I n t h i s association, such elements can be thought t o t r i g g e r a p a r t i c u l a r cogn i t i v e operation--in t h i s case, one o f "mu1 tiplexing". The reverse o f the preceding circumstances i s also t o be found i n language. F i r s t , there are lexi c a l items t h a t i n t r i n s i c a l l y s~e c i f y a mu1 t l p l e x i t y .Engl i s h examples are f u r n i t u r g o r timber ( I l , 'stand i n g t r e e s ' ) f o r m a t t m r e a t h e f o r action, as used i n (15a). And, too, there are gramnatical elements able t o appear i n association here, as i n (15b), t h a t s'ignal an operation the reverse of mu1 tiplexing-one t h a t can be c a l l e d "unit-excerpting". By t h i s operation, a single one o f the specified equivalent u n i t s i s taken and set i n the foreground o f attention. The category o f "state o f dividedness" r e f e r s t o a quantity ' s internal qonsistency. A quantity i s "discrete" (or "particulate") if there are breaks i n i t s 00 inuity. Otherwise, the quantity i s "continu o~~" .~~ Both lexical and grmaatical elements are sensitive. i n t h e i r specifications, t o the d i s t i n ctians o f t h i s category. But there appear t o be n$ gramnatical elements t h a t solely specify discreteness o r continuity f o r a quantity, and also none t h a t signal an operation f o r reversing quantity's lexdcal l y specified state o f dividedness. f 2 I n consequence, there i s d i f f i c u l t y i n demonstrating t h i s category e x p l i c i t l y by i t s e l f , and so we defer i t s treatment u n t i l the next section, where i t can be seen iU interaction with the other categories.The preceding four categories o f a t t r i b u t e s a l l pertain t o a quantity s imul taneously and, taken together, can be considered t o constitute a system of attributes that may be termed a quantity's "dispdsition". The particular intersections o f the several a t t r i b u t e s w i l l be the main object o f attention here.These, f i r s t l y , can be schematized as i n (19): (25) Moving along on the t r a i n i n g course, she c l imbed the f ire-1 adder a t exactly niidday.This s h i f t i n the cognized extensionality O f the event can be thought t o involve a cognitiv'e process o f "reduction" or of "taking the long-range view". The s h i f t carralso go i n the other direction. The event referent can be idealized as an unbounded extent from the e f f e c t o f grammatical elements l i k e "keep -w,"-er -and -er",and "as -+ S", as i n 26 TM prece-event referent was continuous, but a dpscrete case can exhibit the same s h i f t s o f extensiu~ral i t y . One such case, perhaps t o be considered as most basically o f bounded extent, i s shown with that degree o f extensionality i n (27a). But the referent can also be idealized as a point, as i n (27b) ( i t i s clear that the cows here d i d not a l l d i e a t the same moment, and y e t the spread of t h e i r death tl'rnes i s conceptually collapsed i n t o such a single moment). Or, the referent can be idealized as an unbounded extent, as i n (27c):(27) a. The cows a l l died i n a month. b. When the cows a l l died, we sold our farm. c. The cows kept l y i n g (and dying) u n t i l the serum f i n a l l y arrived. (35b). l9 Especially w i t h regard t o i n t e r n a l l y d i sc r e t e quantities--as w i t h a c l u s t e r o f trees--the two NPs can here be seen as coding for two d i f f e r e n t " l e v e l s o f synthesis": The l a t e r NP specifies an unsynthesited m u l t i p l e x i t y , w h i l e the e a r l i e r NP spec i f i e s a p a r t i c u l a r g e a t a l t synthesized therefrom.There i s a f u r t h e r c o g n i t i v e d i s t i n c t i o n involved here t h a t language usually makes: e l t h e r l e v e l of s i n t h e s i s can be placed i n the foreground of p t t e n t i o n while the other l e v e l i s placed i n the background. One grammatical form t h a t specifies t h i s involves p l a c i n g t h e foregrounded NP-type f i r s t , as shown i n (36a). With the use o f t h i s granmatical device, moreover, predications can be made t h a t p e r t a i n s o l e l y t o one l e v e l o f synthesis o r the other, a$ seen i n (36b): (32) a. Thereare houses here and there i n the valley. There a r e c e r t a i n surface forms, furthermore, whose b. There -i s a house every now and t E n through r e f e r e n t s a r e keyed t o applying t o only one o r the t h e valley.other l e v e l o f synthesis. Thus, toggther (toward each o t h e r ) tends t o c o r r e l a t e w i t h mu1 t i p 1 e objects, w h i l e m u p o n i t s e l f ) tends t o c o r r e l a t e w i t h aIn a comparable case, the moving-per-h~ective form,shown i n (33b), i s the o n l y mode t h a t can be spec-composne thereof: i f i e d using everyday language. One must r e s o r t t o s c i e n t i f i c language, as i n (33a), i n order t o estabi s h the synoptic perspective: 33a. The telephone poles' heights form a gradient t h a t c o r r e l a t e s w i t h t h e i r l o c a t i o n s on the road. b. The telephone poles g e t t a l l e r the f u r t h e r down t h e road they are.The reverse o f the preceding circumstances i s a1 so encountered. An example i n v o l v i n g a sequential m u l t i p l e x i t y o f eWnts i s shown i n (34a) w i t h t h e more congruent movi ng-perspec t i v e mode speci f l ed. I n (34b), the same r e f e r e n t instead becomes t h e o b j e c t of syno p t i c viewing. The preceding has involved s h i f t i n g a t t e n t i o n from a mu1 t i p l e x i t y t o the g e s t a l t t h a t i t c o n s t itutes. Also encountered i n language a r c means f o r specifying the reverse: s h i f t i n g a t t e n t i o n from a g e s t a l t t o the components t h a t c o n s t i t u t e i t . This procedure can take place when the s t a r t i n g l e x i c a l item specifies an e n t i t y taken t o be already a t the more s y n t h e t i c l e v e l , as i s t h e case w i t h iceberg i n (38a). By grammatical devices 1 i ke those seen i n (38b), such an e n t i t y can be broken down from conception as a coherent whole and presented i n terms o f component p a r t s and t h e i r i n t e r r e l a t i o n s :(38) a. The iceberg broke i n two.b. The two halves o f the iceberg broke a p a r t ( * i n two).Again we encounter a surface form--in -two--that corr e l a t e s w i t h o n l y one l e v e l o f synthesis and n o t t h e other.20 2.8 Level of Synthesis 2.9 Level o f Exemplarity The category t o be considered now p e r t a i n s t o bounded q u a n t i t i e s , 1 i k e those schematized i n t h e A/B row i n ( 1 9 ) . One form o f l o c u t i o n already seen t o s p e c i f y such q u a n t i t i e s i s t h e p a r t i c u l a r type of "NP of NP" construction i l l u s t r a t e d i n (35a). Here the second NP s p e c i f i e s t h e ,identity o f t h e q u a n t i t y involved, i t s e l f conceptual ized as without i n t r i n s i c bo,unds, while t h e f i r s t NP s p e c i f i e s the bounding ( o r "portion-taking") per se o f t h e quantity: i. She held a gun i n both hands.Sbe held a gun i n e i t h e r hand. 23More notional categories and cognitive processes have been worked up than there i s opportunity t o present here. Some o f t h i s other material i s treated i n an e a r l i e r work, Talmy (1977) (which i t s e l f lacks some o f the material presented here). But we w i l l briefl'y indicate some o f the concepts involved.The adjectives i n a p a i r l i k e s i c k j w e l l behave d i f f e r e n t l y i n association w i t h g r a m n a t i x e l ements specifying vectoral degree, as shown i n (40). I n t h t s they p a r a l l e l the behgvior o f certain s p a t i a l expressions l i ke a t the border/past the border: (41) ( A f t e r e a t i n the shrimp, he f e l t Worse and worse and 3 he was almost sick a t one point/ he f i n a l l y got s i c k i n 5 hrs.Lexical expresdions3 1 i ke cottage and hotel room mav be taken t o have "as'sociated characteristics"--h&e, respectively, those of ' permanent residense' and 'temporary Ibdgihg ' . These a t t r i bytes may mesh o r c o n f l i c t with the specifications o f another element i n the same sentence, e.g., with the d i r e c t i o n a l adverbhome, which specifies a permanent residence. both"hostt and 'guest' are t o be found i n the. "I"):(43) a. The host served me some dessert from the kitchen. b. I served myself some dessert from the kitchen. c. I went and g o t some dessert from the kitchen.A major aim i n cognitive l i n g u i s t i c s must be t o investigate the interactions between 1 e~i c a l and grammatical specifications a r i s i n g i n a s i n g l e sentence. Included here are the cognitive accommodations t h a t take place where there are c o n f l i c t i n g specifccations. A number of interactions have been provision a l l y i d e n t i f i e d , and f o u r seem d e f i n i t e l y established: operations, s h i f t s , blends (of two kinds: superimposed and introjected), and juxtapositions. The l a s t three of these arp t-reatqd a t length i n Talmy (1977) . I n (44a), the l e x i c a l verb flash appears. w i t h i t s basic s t r u c t u r a l specification as a point-durational f u l l -cycle unipl ex event. This undergoes the process o f mu1 t i p l e x i n g , t o y i e l d the unbounded mu1 t i p l e x i t y i n (44b). This then undergops bounding i n (44.c). This bounded mu1 t i p l e x i t y i s then-f i r s t put through the process of reduction to become idealized as a point. and this is i n t u r n multiplexed, yielding (44d). This new unbounded mu1 tip1 exity i s finally then bounded i n (44e). The nesting of structural specifications in this last stage can be represented schematically as in (45):Grammatical ly specified structuring appears to be similar, i n certain of i t s characteristics and functions, t o the structuring i n other cognitive domains, notably that of visual perception. In particul ar, the characteristic of being quasi-topological qan be pointed to, and three major functions can be identified: classification, synoptics, and continuity. The thinking here i s not equally f a r along on a l l these matters, but something of i t s directions can be indicated.Grammatical specifications can be seen to cons t i t u t e a classification with regard to the vast variety of 1 earned,. conceived, and perceived material .They gather different portions of the material together intq subdivisions distinct from each other. By this, any particular currently cognized element i s associated with i t s imp1 i c i t "subdivision-mates". An illustrative case here are the twenty-odd motionrelaied ppepositions in English, such as through and into whlch together subdivide the domain of 'paths considered with respect t o reference-objects' . This domain covers a great and varied range, but any particular "path" fa1 1 s within the purvue of one or another preposition, associated there ~4 t h other "paths" The associations are often language-specific and sometimes seem arbitrary or idiosynchratic. Thus, a s s?n earlier, classed together by through are such dissimi l a r cases a s a straightforward liquid-parting course (walking through water) and a ,zig-zag obstacle-avoidi ng course (wal king through timber). The question arises why such distinctions should be effaced by the grammatical system, while they are observed by the lexical and other cognitive systems. Why are grammaticdl elements--say, such prepostions--not a large and open class marking indefinitely many d i stinctions? One may speculate that the cognitive function of such classificatioh l i e s in rendering contentful material manipulable--i.e., amenable t o transmission, storage, and processing--and that its lack would render content an ineffective agglopergtion.The original assumption made i n this paper about grammatical specification involved the synoptic function. That i s , thO grammatical elements of any particular sentence together specify the structure i f the cognitive representati on evoked by that sentence.Their specifications act as a scaffolding or framework across which contentful material can be splayed or draped. I t can be speculited that such structure is necessavly for a disparate quantity of contentful material to cohere in any sensible way or t o be simultaneously cognized as a gestalt.In the course of discourse, a great welter of notions pass i n rapid succession. B u t there are several ways in which a cognitive continuity i s maintained through this flux and a coherent gestalt i s sumnated over time. For one, there are cognitive processes whereoy the successive notions generally can be sensibly connected together o r f i t -into a conceptual matrix. For another, rhetorical specifications --all the y ' s , on. the other hands, and a num-ber of subtler elements not generally recognized for this--direct the i l locutionary flow and make up the "logical" tissue of the discourse. Through this, grammatical elements appear to play a determinative role. Their specifications establ ish a structural level w i t h greater temporal constancy amidst more f 1 eeting aspects of content.These forms of grammatically specified structuring seem t o parallel forms discernable in the operation of visual perception.24 First, the perception of anv particular object i s mediated by i t s association w i t h re1 a ted objects i n a cl assi f i catory schema.Secondly, the we1 t e r of visual sensations cognized a t any given moment for some who1 e scene i s rendered coherent by the perception of structural delineations running through it. One specialized form of thjs i s discernable when one intends t o move through a space, say, from one to the opposi~te corner of a restaurant. The sensations of tables, chairs ,etc. are, in effect, perceived i n simplified spatial arrangements as i f from an aerial view, and the plot of a course one could follow through that i s sensed.T h i r d l y , i n the course of motion through space over time, there is a great flux of visual sensations rushing past, b u t sense of continuity i s maintained by the perception of structure running through the successive scenes. Two 1 eve1 s of "scene-structure constancy" are maintained. In the f i r s t , the perce~ved del ineations afford greater permanence t h a n the sensory flux, b u t do slowly shift. Thjs i s the level where, say, i n walking past a table, i t s perceivedtoutl ine i s maintained b u t s h i f t s gradually from a quadrilateral t o a trapezoid and back to a quadrilateral. A deeper level of greater constancy i s also maintained, from which the table cont?nues to be perceived as a rectangle no matter where one i s in relation to i t . For a final parallel -w;ith grammatical specification, the topology-1 i ke nature of visual perception i s evident here. For certain abstract characterjstics of a scene and i t s contents are maintained constant while other, more metrical and Eucl idean characteris t i c s are free t o vary without re1 evance thereto.4. Notes 1. The word "evoke" i s used because the relationship i s not direct, The CR i s an emergent, compounded by barious cognitive processes out of the sentence elements' referential meanings, understanding of the present situation, general know1 edge, etc.Our term "cognitive representation" i s similar i n purport t o Fillmore's (1975) "scene" but i s chosen over that more specifically visual term, Ine 1 inguistical l y evok~d somplex can have much from other sense modal i t i e s (notably som/ k i nestheti c and auditory) as we1 1 as meta-modal aspects.2. Comprehension, rather than production, i s the d l rection we limit ourselves to in the i n i t i a l endeavor. This direction would seem t o yield more immediately reliable findings, since i t s starting point i s w i t h more overtly manifest, hence handleab'l e, forms 1 i ke grammatical elements rather than w i t h meanings and experiential Complexes, which rely more on introspection and reports of introspection. Nevertheless , eacl direction does involve both the manifest and the experiential sides of language.. This i s a classical linguistic distinction. A class i n which morphemes are formally gathered i s con Sidered open i f i t i s quite large and easily augment- | null | null | null | 9 ive us a handle for ascertaining the very makeup of linguistic-) cognitive structuring. W e accordingly examine a number of grammatical 1 y spec 'fled notions, observe the categories and systems i n which they pattern, and speculate on broader cognitive connections.Some provisional findings have alreagy emerged. Grammatical specifications for s tFucture are preponderantly re1 a t i v i s t i c or topological, and exclude the fixed or metrical Jy Eucl idean. The categories 7n which grammatical notions pattern irlclude: ,bl exi ty perspectival mode s t a t e of boundedness level of synthesis s t a t e of dividedness 1 eve1 of exemplarity degyee of extensional i t y axial characteristics pattern of distribution scene-breakup " Grammatical spec7 f ication ' o f s t r u c t u r~n g appears t o be the same, in certain abstract characteristics, as the structurinq of visual perception, | Main paper:
introduction:
A sentence (or other portion of discourse) i s taken to evoke i n the listener a particular kind o f experiential compl ex--here t o be termed a "cognitive representation" or " C P U . l There appears to be a significant way i n which different portions of the language i n p u t specify, g r codk for, different portions of the CR. The major finding, is that--for a f i r s t approximation--the 1 mica1 fraction of a sentence codes mainly for the content, or substance, of a CRY while the grammatical fraction of a sentence codes mainly for the structure of a CR. Determining the structure within a realm of phenomena has been a central concern for analytic science, including 1 inguist i c s and psychology. With grammar seen in the above light, i t can be used in determining the structure, of the 1 anguage-re1 ated portion of human cogrTition, w i t h possi bl e connections t o further poreons. In particular, looking systematically a t the actual notions specified by grammatical elements can give us a handle for ascertaining the ery nakeup of (1 f nguist i c -) cogni tive structuring.!The beqinnings of such a n endeavor are the aims of this paper Several ideas here require some immediate elaboration. The distinction between lexical and grammatical i s made entirely formally--i,e., without any reference t o meaning--on the basis of the distinction between open-cl ass and cl osed-cl ass. 3 A1 1 openclass elem nts--i . e . , the stems of nouns, verbs, and B adjectives --are considered lexical. Everything else i s considered grammatical. Included here are a1 1 closed-cl ass morphemes and words--infl ections , p a rt iclles, adposi tons, conjunctions, demonstratives , etc. --as we1 1 as syntactic constructions, grama tical re1 ations , categorial identi t i e s , word order, and intonation. Terminological ly here, "grammatical element" wi 11 be used to refer t o any o f these.The nature of content and of structure, and the distinction between them, are not understood we1 1 enough t o be addressed analytically i n this paper and must be l e f t t o our intuitive sense of the matter. 5 Taking them for granted, however, we can now more finely characterize the linguistic-cognitive crossrelationships noted earlien While most of a CR's content i s specified by the lexical fraction of a sentence, the lexical items do usually specify some structural notions along with the contentful ones. The gramatical elements Of a sentence more unalloyedly specify only structural notions and specify them more determinately in the case of conflict w i t h a lexical item, e tab1 ishing perhaps the majority of a CR' s structure. 8In other work in the present di recti on--notab1 y Fillmore's {e.g. , 1975, 1976 )--concern has also been with ascertaining structre, b u t the sentence elements used as starting-points have generally been lexical i terns with prominently i nmi xed structural specif ications ( l i k e buy and s e l l ) . The present work, i n part a complement t o the othet, takes advantage of grammar's greater directness and completeness in specifying structure.This paper i s divided into three sections. In the f i r s t , a sampling of grammatical elements i s examined for the notions that they specify, both as an introduction to out method and for the aim of notici n g properties common to such notions as we1 1 as properties excluded from them. In the second, we present a number of the categories i n which grammatically specified noSions have been observed t o pattern. In the third, we speculate on broader cognitve connections.1. The Nature o f Gramnatically Specified Notions I n t h i s section we examine a small sampling o f grammatical elements f o r the p a r t i c u l a r component notions t h a t they specjfy. The sample w i l l give a h e u r i s t i c indication o f the kinds of notions t h a t get grammatically specified as well as o f kinds o f notions t h a t possibly never do. The excluded kinds w i l l be seen as r e a d i l y specifiable by l e x i c a l elements. A further comparison between the characteri s t i c s o f gramnatically specified notions and o f l e x i c a l l y specified ones I s then made. To indicate the major finding a t the outset, i t seems that-grammatical specifications f o r structure are prepon'dera n t l y r e l a t i v i s t i c o r topological, and exclude the f i x e d o r m e t r i c a l l y Euclidean.For a f i r s t simple case, many languages have inf l e c t i o n s for the noun (Engl i s h has -b and -5 ) t h a t specify the u n i~l e x or the m u l t i p~e x instaztiat i o n o f t h e object specified by the noun. By con-' t r a s t , no languages appear t o have i n f l e c t i o n s t h a t speci fy the redness o r b l ueness , etc. -i . e. , the part i c u l a r color--of the object specified by a noun.-.i n the preceding, the underlined are instances o f "notions". the speaker-side o r the non-speaker-side o f a concept u a l p a r t i t i o n drawn through space (or time o r other qua1 i t a t i ve dimension). This i n t e g r a l specification can be analyzed as containing the following component notions (ehcl osed by quotes) :(1 a-b. a ' p a r t i t i o n ' t h a t divides a space i n t o 'regionst/'sides' c-e. the ' locatednes's' (a p a r t i c u l a r r e l a t i o n ) o f a 'point' ( o r object idealizable as a point) 'within' a region f-g. (a side t h a t i s the) 'same' as o r ' d i f f e r e n t ' from h-i. a 'currently indicated' object and a 'currently communicating' e n t i t yNotions t h a t might a t f i r s t be ascribed t o such deict i c s , such as of distance o r perhaps size, prove not t o be, on the evidence o f sentence-pairs 1 i k e (2):(2) a. This speck i s smaller than t h a t speck. b. This planet i s smaller than t h a t planet.The CRs evoked by (2a) and (b) d i f f e r greatly, i nvol ving t i n y objects m i 1 1 imeters apart o r huge objects parsecs apart. Yet the sentences d i f f e r only l e x i ca l l y , not g r a m a t i c a l l y * Hence, the CRs' notions as t o the magnitude o f size o r distsence cannot be traced t o the d e i c t i c s (or t o other gramnatical elements) i n the sentences. Thus, the notional specifications o f a t h i s o r a t h a t appear, i n part, t o be genuinely topologicbl : the establishment of a p a r t i t i o n remains a constant, but i t s p o s i t i o n can vary unlimitedly (or, using topology's characterizabil i t y as "rubber-sheet geometry", the p a r t i t i o n ' s distance away can be (3) a. I wal ked through the water. b. I walked through the timeber ( i . e . p , woods).In t h i s usage, through specifies, broadly, 'motion along a l i n e t h a t i s w i t h i n a medium'. The component notions contained here include:4 ) 1 a-e.motion '--i . e., 'one-to-one correspondences'between 'adjacent' points of ' space' and adjacent points o f ' t i m e ' f. motion t h a t describes a "Line' g.t h e l o c a t e d n e s s o f a l i n e w i t h i n a 'medium' h-i. a medium, i.e., a region o f ekree-dimensional space s e t apart by the lomtedness w i t h in i t d f 'material' t h a t i s i n a pattern o f dis- t r i b u t i o ? ' o f a certain range o f character ( s t i 1 1 t o be determined)Again, w i t h (3a) and ( i t can be f u r t h e r determined t h a t ' r a t e of motion' and 'shape/contour o f 1 inear path' are also n o t specified by the gramnatical element. As one s t e p i n a program t o ascertain any propert i e s comnon t o gramnatical l y specified notions, the notions j u s t found are gathered together i n Table 1 . For h e u r i s t i c purposes, the notions are very provisi o n a l l y divided i n t a three groups on the basis o f t h e i r r e l a t i o n t o topology. I n group (a) are the notions t h a t properly be1 ong , o r are r e a d i l y def inabl e, i n the actual mathematical system o f topology. I n group (b), the notions might not be p a r t o f topology proper but intuitively'seem l i k e those t h a t are--and might be includable i n a related mathematical system t h a t could be constructed. Pn group (c) are the not i o n s t h a t f a l l outside o f any usual conception o f a mathemat'ical system. The number of notions i n the f i r s t two groups combined i s 13, while t h e t h i r d has 6--an i n d i c a t i o n o f a preponderant propensity f o r gramnati cal elements t o specify quasi -topological notions. The ratSo i n t h i s d i r e c t i o n i s i n f a c t improved i f we consider t h a t even several n o t i~n s~i n group (c)--the bottom three--resemble topological ones i n t h e sense o f involving r e l a t i v i s t i c relationships between quantities r a t h e r than absol u t e l y f i x e d quantities.(7) Table 1: Some notions found t o be specified d. one-to-one matter ,correspondences space tSme motion med i urn c u r r e n t l y indicated/ comnunicating e n t i t yFor a compl ementary program o f ascertaining any properties excluded from gramnatical specification, the notions found above n o t t o be specified by the .elerhents investigated are 1 i s t e d i n Table 2 . Rather than topological, topologf-like, o r r e l a t i v i s t i c , these notions involve Eucl idean-geometric concepts (e.g., set distance, Size, contour), q u a n t i f i e d measure, and various p a r t i c u l a r i t i e s o f a quantity--in sum, c h a r a c t e r i s t i c s t h a t a r e absolute o r fixed. These grammatical -1 e x i c a l differences can be s e t i n t o f u r t h e r r e l i e f by i n t u r n varying one elementtype while keeping the other constant. Thus, varying o n l y the gramnatical elements o f ( 9 ) , as i s done i n (13) A machine cancelled t h e stamps.The preceding sampLing of g r a m a t i c a l elements has y i e l d e d a set o f natisns helpful toward discovering comnon properties. But t h e s e t has been small and haphazardly a r r i v e d at. With a broader and more systematic investigation, patterns of organization become evident. Gramnatically s p e c i f i e d notions can be seen t o p a t t e r n Jn categories, and the categories, i n turn, i n integrated systems. I n t h i s section we look a t some o f these categories and systems, The grammatical elements here w i l l not be treated i n isolation, but i n associa%ion w i t h l e x i c a l items. That i s , the grammatically specified structural not i o n s w i l l be considered i n i n t e r a c t i o n w i t h t h a t portiop o f l e x i c a l specification t h a t i s a1 so struct u r a l . This interaction e n t a i l s cognitive processing, and d i f f e r e n t cases o f such processing w i l l be considered along the way.The note on methodology should be made t h a t our d i r e c t i o n o f analysis has been from grammatical speci f i c a t i~n t o category, not the reverse. That i s , the categories considered be1 ow were discovered t o be re1 evant t o the specifications o f various grammatical elements. They were not part o f some a priori concept u a l schema which then sought corrobovative exampl es.
dimension / kind o f quantity:
The category o f "dimension" has two member notions, 'space' and ' time'. The kind o f "quantity" t h a t exists i n space i s --i n respectively continuous o r discrete form--'matter8 o r 'objects'. The kind o f quantity e x i s t i n g i n time i s 'action' o r 'events' ("action" i s meant t o refeF t o any obtaining circumstance not j u s t ( w i l l e d ) motion). I n tabular form, these notions r e l a t e thus: 13space: matter/objects time:action/events A number of grammatical and l e x i c a l referents are specific w i t h regard t o one o r the other pole o f t h i s category. But s i~c e the category cross-cuts the ones t r e d e d next, we w i l l not exempl i f y i t here but w i l l endeavor i n the following, t o present both space and time examples side by side.The category here t o be termed "plexityaL i s a quantity's s t a t e o f a r t i c u l a t i o n i n t o equivalent elements. Where the quantity consists o f only one such element, i t i s "uniplex", and where i t consists o f more than one, i t i s "multiplex". When the quantity involved i s m a t e r , p l e x i t y is, o f course, equivalent t o the t r a d i t i o n a l category o f "number" w i t h i t s component notions "singulsr" and "plural". But the present notions are intended t o capture the generalizat i o n f r o m matter over t o action, which the t r a d i t i o n a l ones do not.9Specifications as t o p l e x i t y are made by both l e x i c a l items and gramnatical elements, and the interplay between the two when they are both i n associat i o n must be noted. Example English l e x i c a l items t h a t basically specify a unipl ex referent are--for matter and action, respectively--birdand ( t o ) sigh. They q n occur w i t h gramnatical elements t h a t themselves specify a uniplexity, 1 i k e those under1 ined i n (14a) (many languages have here a more regular, overt system o f markers than English). But they can a1 so occur w i t h gramnatical elements t h a t specify a m u l t i p l e x i t y , as i n (14b). I n t h i s association, such elements can be thought t o t r i g g e r a p a r t i c u l a r cogn i t i v e operation--in t h i s case, one o f "mu1 tiplexing". The reverse o f the preceding circumstances i s also t o be found i n language. F i r s t , there are lexi c a l items t h a t i n t r i n s i c a l l y s~e c i f y a mu1 t l p l e x i t y .Engl i s h examples are f u r n i t u r g o r timber ( I l , 'stand i n g t r e e s ' ) f o r m a t t m r e a t h e f o r action, as used i n (15a). And, too, there are gramnatical elements able t o appear i n association here, as i n (15b), t h a t s'ignal an operation the reverse of mu1 tiplexing-one t h a t can be c a l l e d "unit-excerpting". By t h i s operation, a single one o f the specified equivalent u n i t s i s taken and set i n the foreground o f attention. The category o f "state o f dividedness" r e f e r s t o a quantity ' s internal qonsistency. A quantity i s "discrete" (or "particulate") if there are breaks i n i t s 00 inuity. Otherwise, the quantity i s "continu o~~" .~~ Both lexical and grmaatical elements are sensitive. i n t h e i r specifications, t o the d i s t i n ctians o f t h i s category. But there appear t o be n$ gramnatical elements t h a t solely specify discreteness o r continuity f o r a quantity, and also none t h a t signal an operation f o r reversing quantity's lexdcal l y specified state o f dividedness. f 2 I n consequence, there i s d i f f i c u l t y i n demonstrating t h i s category e x p l i c i t l y by i t s e l f , and so we defer i t s treatment u n t i l the next section, where i t can be seen iU interaction with the other categories.The preceding four categories o f a t t r i b u t e s a l l pertain t o a quantity s imul taneously and, taken together, can be considered t o constitute a system of attributes that may be termed a quantity's "dispdsition". The particular intersections o f the several a t t r i b u t e s w i l l be the main object o f attention here.These, f i r s t l y , can be schematized as i n (19): (25) Moving along on the t r a i n i n g course, she c l imbed the f ire-1 adder a t exactly niidday.This s h i f t i n the cognized extensionality O f the event can be thought t o involve a cognitiv'e process o f "reduction" or of "taking the long-range view". The s h i f t carralso go i n the other direction. The event referent can be idealized as an unbounded extent from the e f f e c t o f grammatical elements l i k e "keep -w,"-er -and -er",and "as -+ S", as i n 26 TM prece-event referent was continuous, but a dpscrete case can exhibit the same s h i f t s o f extensiu~ral i t y . One such case, perhaps t o be considered as most basically o f bounded extent, i s shown with that degree o f extensionality i n (27a). But the referent can also be idealized as a point, as i n (27b) ( i t i s clear that the cows here d i d not a l l d i e a t the same moment, and y e t the spread of t h e i r death tl'rnes i s conceptually collapsed i n t o such a single moment). Or, the referent can be idealized as an unbounded extent, as i n (27c):(27) a. The cows a l l died i n a month. b. When the cows a l l died, we sold our farm. c. The cows kept l y i n g (and dying) u n t i l the serum f i n a l l y arrived. (35b). l9 Especially w i t h regard t o i n t e r n a l l y d i sc r e t e quantities--as w i t h a c l u s t e r o f trees--the two NPs can here be seen as coding for two d i f f e r e n t " l e v e l s o f synthesis": The l a t e r NP specifies an unsynthesited m u l t i p l e x i t y , w h i l e the e a r l i e r NP spec i f i e s a p a r t i c u l a r g e a t a l t synthesized therefrom.There i s a f u r t h e r c o g n i t i v e d i s t i n c t i o n involved here t h a t language usually makes: e l t h e r l e v e l of s i n t h e s i s can be placed i n the foreground of p t t e n t i o n while the other l e v e l i s placed i n the background. One grammatical form t h a t specifies t h i s involves p l a c i n g t h e foregrounded NP-type f i r s t , as shown i n (36a). With the use o f t h i s granmatical device, moreover, predications can be made t h a t p e r t a i n s o l e l y t o one l e v e l o f synthesis o r the other, a$ seen i n (36b): (32) a. Thereare houses here and there i n the valley. There a r e c e r t a i n surface forms, furthermore, whose b. There -i s a house every now and t E n through r e f e r e n t s a r e keyed t o applying t o only one o r the t h e valley.other l e v e l o f synthesis. Thus, toggther (toward each o t h e r ) tends t o c o r r e l a t e w i t h mu1 t i p 1 e objects, w h i l e m u p o n i t s e l f ) tends t o c o r r e l a t e w i t h aIn a comparable case, the moving-per-h~ective form,shown i n (33b), i s the o n l y mode t h a t can be spec-composne thereof: i f i e d using everyday language. One must r e s o r t t o s c i e n t i f i c language, as i n (33a), i n order t o estabi s h the synoptic perspective: 33a. The telephone poles' heights form a gradient t h a t c o r r e l a t e s w i t h t h e i r l o c a t i o n s on the road. b. The telephone poles g e t t a l l e r the f u r t h e r down t h e road they are.The reverse o f the preceding circumstances i s a1 so encountered. An example i n v o l v i n g a sequential m u l t i p l e x i t y o f eWnts i s shown i n (34a) w i t h t h e more congruent movi ng-perspec t i v e mode speci f l ed. I n (34b), the same r e f e r e n t instead becomes t h e o b j e c t of syno p t i c viewing. The preceding has involved s h i f t i n g a t t e n t i o n from a mu1 t i p l e x i t y t o the g e s t a l t t h a t i t c o n s t itutes. Also encountered i n language a r c means f o r specifying the reverse: s h i f t i n g a t t e n t i o n from a g e s t a l t t o the components t h a t c o n s t i t u t e i t . This procedure can take place when the s t a r t i n g l e x i c a l item specifies an e n t i t y taken t o be already a t the more s y n t h e t i c l e v e l , as i s t h e case w i t h iceberg i n (38a). By grammatical devices 1 i ke those seen i n (38b), such an e n t i t y can be broken down from conception as a coherent whole and presented i n terms o f component p a r t s and t h e i r i n t e r r e l a t i o n s :(38) a. The iceberg broke i n two.b. The two halves o f the iceberg broke a p a r t ( * i n two).Again we encounter a surface form--in -two--that corr e l a t e s w i t h o n l y one l e v e l o f synthesis and n o t t h e other.20 2.8 Level of Synthesis 2.9 Level o f Exemplarity The category t o be considered now p e r t a i n s t o bounded q u a n t i t i e s , 1 i k e those schematized i n t h e A/B row i n ( 1 9 ) . One form o f l o c u t i o n already seen t o s p e c i f y such q u a n t i t i e s i s t h e p a r t i c u l a r type of "NP of NP" construction i l l u s t r a t e d i n (35a). Here the second NP s p e c i f i e s t h e ,identity o f t h e q u a n t i t y involved, i t s e l f conceptual ized as without i n t r i n s i c bo,unds, while t h e f i r s t NP s p e c i f i e s the bounding ( o r "portion-taking") per se o f t h e quantity: i. She held a gun i n both hands.Sbe held a gun i n e i t h e r hand. 23More notional categories and cognitive processes have been worked up than there i s opportunity t o present here. Some o f t h i s other material i s treated i n an e a r l i e r work, Talmy (1977) (which i t s e l f lacks some o f the material presented here). But we w i l l briefl'y indicate some o f the concepts involved.The adjectives i n a p a i r l i k e s i c k j w e l l behave d i f f e r e n t l y i n association w i t h g r a m n a t i x e l ements specifying vectoral degree, as shown i n (40). I n t h t s they p a r a l l e l the behgvior o f certain s p a t i a l expressions l i ke a t the border/past the border: (41) ( A f t e r e a t i n the shrimp, he f e l t Worse and worse and 3 he was almost sick a t one point/ he f i n a l l y got s i c k i n 5 hrs.Lexical expresdions3 1 i ke cottage and hotel room mav be taken t o have "as'sociated characteristics"--h&e, respectively, those of ' permanent residense' and 'temporary Ibdgihg ' . These a t t r i bytes may mesh o r c o n f l i c t with the specifications o f another element i n the same sentence, e.g., with the d i r e c t i o n a l adverbhome, which specifies a permanent residence. both"hostt and 'guest' are t o be found i n the. "I"):(43) a. The host served me some dessert from the kitchen. b. I served myself some dessert from the kitchen. c. I went and g o t some dessert from the kitchen.A major aim i n cognitive l i n g u i s t i c s must be t o investigate the interactions between 1 e~i c a l and grammatical specifications a r i s i n g i n a s i n g l e sentence. Included here are the cognitive accommodations t h a t take place where there are c o n f l i c t i n g specifccations. A number of interactions have been provision a l l y i d e n t i f i e d , and f o u r seem d e f i n i t e l y established: operations, s h i f t s , blends (of two kinds: superimposed and introjected), and juxtapositions. The l a s t three of these arp t-reatqd a t length i n Talmy (1977) . I n (44a), the l e x i c a l verb flash appears. w i t h i t s basic s t r u c t u r a l specification as a point-durational f u l l -cycle unipl ex event. This undergoes the process o f mu1 t i p l e x i n g , t o y i e l d the unbounded mu1 t i p l e x i t y i n (44b). This then undergops bounding i n (44.c). This bounded mu1 t i p l e x i t y i s then-f i r s t put through the process of reduction to become idealized as a point. and this is i n t u r n multiplexed, yielding (44d). This new unbounded mu1 tip1 exity i s finally then bounded i n (44e). The nesting of structural specifications in this last stage can be represented schematically as in (45):
further cognitive connections:
Grammatical ly specified structuring appears to be similar, i n certain of i t s characteristics and functions, t o the structuring i n other cognitive domains, notably that of visual perception. In particul ar, the characteristic of being quasi-topological qan be pointed to, and three major functions can be identified: classification, synoptics, and continuity. The thinking here i s not equally f a r along on a l l these matters, but something of i t s directions can be indicated.Grammatical specifications can be seen to cons t i t u t e a classification with regard to the vast variety of 1 earned,. conceived, and perceived material .They gather different portions of the material together intq subdivisions distinct from each other. By this, any particular currently cognized element i s associated with i t s imp1 i c i t "subdivision-mates". An illustrative case here are the twenty-odd motionrelaied ppepositions in English, such as through and into whlch together subdivide the domain of 'paths considered with respect t o reference-objects' . This domain covers a great and varied range, but any particular "path" fa1 1 s within the purvue of one or another preposition, associated there ~4 t h other "paths" The associations are often language-specific and sometimes seem arbitrary or idiosynchratic. Thus, a s s?n earlier, classed together by through are such dissimi l a r cases a s a straightforward liquid-parting course (walking through water) and a ,zig-zag obstacle-avoidi ng course (wal king through timber). The question arises why such distinctions should be effaced by the grammatical system, while they are observed by the lexical and other cognitive systems. Why are grammaticdl elements--say, such prepostions--not a large and open class marking indefinitely many d i stinctions? One may speculate that the cognitive function of such classificatioh l i e s in rendering contentful material manipulable--i.e., amenable t o transmission, storage, and processing--and that its lack would render content an ineffective agglopergtion.The original assumption made i n this paper about grammatical specification involved the synoptic function. That i s , thO grammatical elements of any particular sentence together specify the structure i f the cognitive representati on evoked by that sentence.Their specifications act as a scaffolding or framework across which contentful material can be splayed or draped. I t can be speculited that such structure is necessavly for a disparate quantity of contentful material to cohere in any sensible way or t o be simultaneously cognized as a gestalt.In the course of discourse, a great welter of notions pass i n rapid succession. B u t there are several ways in which a cognitive continuity i s maintained through this flux and a coherent gestalt i s sumnated over time. For one, there are cognitive processes whereoy the successive notions generally can be sensibly connected together o r f i t -into a conceptual matrix. For another, rhetorical specifications --all the y ' s , on. the other hands, and a num-ber of subtler elements not generally recognized for this--direct the i l locutionary flow and make up the "logical" tissue of the discourse. Through this, grammatical elements appear to play a determinative role. Their specifications establ ish a structural level w i t h greater temporal constancy amidst more f 1 eeting aspects of content.These forms of grammatically specified structuring seem t o parallel forms discernable in the operation of visual perception.24 First, the perception of anv particular object i s mediated by i t s association w i t h re1 a ted objects i n a cl assi f i catory schema.Secondly, the we1 t e r of visual sensations cognized a t any given moment for some who1 e scene i s rendered coherent by the perception of structural delineations running through it. One specialized form of thjs i s discernable when one intends t o move through a space, say, from one to the opposi~te corner of a restaurant. The sensations of tables, chairs ,etc. are, in effect, perceived i n simplified spatial arrangements as i f from an aerial view, and the plot of a course one could follow through that i s sensed.T h i r d l y , i n the course of motion through space over time, there is a great flux of visual sensations rushing past, b u t sense of continuity i s maintained by the perception of structure running through the successive scenes. Two 1 eve1 s of "scene-structure constancy" are maintained. In the f i r s t , the perce~ved del ineations afford greater permanence t h a n the sensory flux, b u t do slowly shift. Thjs i s the level where, say, i n walking past a table, i t s perceivedtoutl ine i s maintained b u t s h i f t s gradually from a quadrilateral t o a trapezoid and back to a quadrilateral. A deeper level of greater constancy i s also maintained, from which the table cont?nues to be perceived as a rectangle no matter where one i s in relation to i t . For a final parallel -w;ith grammatical specification, the topology-1 i ke nature of visual perception i s evident here. For certain abstract characterjstics of a scene and i t s contents are maintained constant while other, more metrical and Eucl idean characteris t i c s are free t o vary without re1 evance thereto.4. Notes 1. The word "evoke" i s used because the relationship i s not direct, The CR i s an emergent, compounded by barious cognitive processes out of the sentence elements' referential meanings, understanding of the present situation, general know1 edge, etc.Our term "cognitive representation" i s similar i n purport t o Fillmore's (1975) "scene" but i s chosen over that more specifically visual term, Ine 1 inguistical l y evok~d somplex can have much from other sense modal i t i e s (notably som/ k i nestheti c and auditory) as we1 1 as meta-modal aspects.2. Comprehension, rather than production, i s the d l rection we limit ourselves to in the i n i t i a l endeavor. This direction would seem t o yield more immediately reliable findings, since i t s starting point i s w i t h more overtly manifest, hence handleab'l e, forms 1 i ke grammatical elements rather than w i t h meanings and experiential Complexes, which rely more on introspection and reports of introspection. Nevertheless , eacl direction does involve both the manifest and the experiential sides of language.. This i s a classical linguistic distinction. A class i n which morphemes are formally gathered i s con Sidered open i f i t i s quite large and easily augment-
:
9 ive us a handle for ascertaining the very makeup of linguistic-) cognitive structuring. W e accordingly examine a number of grammatical 1 y spec 'fled notions, observe the categories and systems i n which they pattern, and speculate on broader cognitive connections.Some provisional findings have alreagy emerged. Grammatical specifications for s tFucture are preponderantly re1 a t i v i s t i c or topological, and exclude the fixed or metrical Jy Eucl idean. The categories 7n which grammatical notions pattern irlclude: ,bl exi ty perspectival mode s t a t e of boundedness level of synthesis s t a t e of dividedness 1 eve1 of exemplarity degyee of extensional i t y axial characteristics pattern of distribution scene-breakup " Grammatical spec7 f ication ' o f s t r u c t u r~n g appears t o be the same, in certain abstract characteristics, as the structurinq of visual perception,
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0.001805 | null | null | null | null | null | null | null | null |
05f3c944c3f2e2bdcab7588eadbe5818131efce1 | 219307782 | null | With a Spoon in Hand This Must Be the Eating Frame | A language camprehension program using "framestf " s c r i p t s f ' , etc. must b e a b l e t o d e c i d e which framed are a p p r o p r i a t e to the text. | {
"name": [
"Charniak, Eugene"
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | null | null | null | i s n o t a l q a y s so easy.('%+ woman way@ while-the man on t h e s t a g e sawed her i n h a l f " s w g e a t s MAGICIAN b u t how?) T h i s paper w i l l examine how a program might g o about determining t h e a p p r o p r i a t e frame in such cases.A t a s u f f i c i e n t l y sag* l e v e l the model presented h e r e w i l l reaembge t h a t o f Minsky (1975) i n it's a s s m p t i o n chat one u s u a l l y h a s a v a i l a b l e one o r more c o n t e x t frames.Hence one o n l y needs worry i f information comes i n w h i c h d o e s n o t P i t them.A s opposed t o Minsky however t h e s u g g e s t i o n s f o r new c o n t e x t frames w i l X not come from t h e old ones, but r a t h e r from t h e c o n f l i c t i n 8 information.The problem them becomes how p o t e n t i a l frames a r e indexed under t h e i n f o m a t ion which "suggestsw them.making i n f e r e n t e s from a v e r y large base of common s e n s e knowledge. To avoid d e a t h by c m b i n a t o r i a l explosion our computer must be h b l e t o ~c c e s s t h e knowledqe i t n e d s without i r r e l e v a n t knowledse ~e t t i n g i n i t s m y . A p l a u s i b l e c o n s t r a i n t on t h e knowledge we m i~h t use a t a %iven point i n a s t~r y o r c o n v e r s a t i o n ( I s h a l l henceforth simply assme we are d e a l i n g w i t h a s t o r y ) is' t o =strict c o n s i d e r a t i o n to t h a t p o r t i o n o f our knowledge which is "&bout" t h i n g s which have been mentioned i n t h e d i s c o u r s e . So i f w e have a s t o r y which menttons t r a i n s and t r a i n s t a t i o n s , we w i l l not use our knowledge o f , shy, circuees. T h i s r e q u i r e s , of course. t h a t g i v e n a t o p i c , such a s t r a i n s , or e a t i n g . we must be a b l e to acCess i t s knowledge without going t h r o q h a v e r y t hing we know.Hence we a r e lead i n a natural way t o something approaching a notion of "frame" (Minsky a c o l l e c t i o n o f knowledge about a s i n g l e stereotyped s t t u a t i o n . I n t h e above discussion however I have made a r a t h e r important slight o f hand. Given s t o r y we only want to consider those frames "about" t h i n g s t n the s t o r y . l b w L s i t that we d e c i d e which frames q u a l i t y 'I was a b l e t o g l o s s over t h i s because i n most s i t u a t i o n s t h e problca, a t l e a s t a t a s u r f a c e Iwel. does n o t appear a l l thst d i f f i c u l t .If t h e s t o r y i s abotlt: t r a i n s . i t will s u r e l~ meqtion t r a i n s .So we see t h e word "train". and we assume t h a t t r a i n s ate r e l e v a n t . What could be easier.Unfortunately, this ease 1s d e c e p t i v e f o r t h e s t o r y may mentton many topics of which o n l y a few a r e t r w l y important t o the s t o r y . For example.The lawyer gook a cab t o *the r e s t a u r a n t near the uni'(iereit.y.Here w? have "lawyer", "cab" , " r e s t a u r~n t " and " u n i v e r s i t~" a11 o f which are c a l l i n g f 6 r o u r a t t e n t i o n . Somehow on t h e b a s i s of latef l i n e s we must weed o u t t h o s e which our o n l y incjidentak . To s m a t i z e , t h e l a s t few paragraphs, the problem o f frame determinatioin i n language comprehension $ n v~l v e s t h r e e sub-problms .1) stories will t y p i c a l l y e I d d e t o many h i g h e r fraqks, any of which might serve as the contax& for the incoming Iinee. Nnw d o we choose between them? 2 ) The words used i n a s t o r y may not directly i n d i c a t e t h e proper higher frame. l?uw In tbe paper which follows I w i l l be p r i m a r i l y concentrate 3n ( 2 ) w i t h (3) b e i q mentfoned o c c a s i o n a l l y . I n essence my p o s i t i o n on ( 1 i a t h a t it w i l l not be t o o much o f a problem, provided t h a t &he c o s t o f s e t t i n g up a c o n t e x t like " r w t a u r a n t " i~ small. f f i t ; i s never used then a s Fhe stary g o e s on it w i l l teceeded l n t a the background How t h i s "receed ing " takes place I shall, not s a y . a i n c e f o r one thing i t is 4 problem i n many a r e a s , and f o r a n o t h e r , 1 don't know.Concerning 2and ( 3 ) , we w i l l be lead t o a p 0 8 i t i a h similar t o t h a t o f MLnsky (1975) We w i l l s e e however, t h a t t h e r e a t e s t i l l a l o t o f problema with t h i s * polpMion which d o not a t f i r s t glance me& t h e eye.' 1 THE CLUE INTERSECTION METHOD Rather than immediately presenting my scheme, let m e s t a r t by showing the problems with a n a l b r n a t i v e p o s s i b i l i t y , which I w i l l c a l l t h e "clue i n t e r s e c t i o n " metbod. This a l t e r n a t i v e is by no means a strats man a s one researcher has i n f a c t e x p l i c i t l y suggested i t (Fahlman 1977) and I f o r one find i t a v e r y n a t u r a l way of t h i n k i n g about t h e problem.The i d e a behind t h i s mkghod i s t h a t we a r e given oergain c l u e s i n t h e istory about t h e n a t u r e of the c o r r e c t Erame, and t o find t h e frame we simply i n t The c l u e s h e r e a r e t h i n g s l i k e "aisle", " t u n a f i s h ' e t c .the English w r d s which a r e the c l u e s , but r a t h e r the concepts which underlie the words. I w t l l assuue t h a t we go from one t o the o t h e r v i a a n independent? parsing algorithm.(However t h i s assumes t h a t t h e r e Is no v i c i o u s i n t e r a c t i o n b e t w e n f r b e d s e r m i n a t i o n and d lsambigua t i o n . Given t h a t dieadbiguation depends W U p r i o r frame determinatiop (see (Hayes 1977) for numerous examples) t h i s may be i n c o r r e c t . ) So the input t o the frame d e t e r m i n~r w i l l be something l i k e SUPEWA&KET. The point i s t h a t none o f these e l m s w i l l be unambiguous, but when we take the i n t e r s e c t i o n the qnly t h i n g which vill be l e f t i s SUPERMARKET.ST-1 (WALK JACK-1 AISLE-1) ST-2 ( PERSON JAC K-1 ) ST-3 (EQUAL (NAME JACK-1 ) "JACK") S T 4 (EQUAL (SEX JACK-1 ) MALE) ST-5 (AISLE AISLE-1 ST06 (PUT JACK-I TUNA-FISH<@-1 BASKET-1) ST97 (BASKET B&RET-I)There are, however, problems with t h i s v i e w of t h i n g s .For one t h i n g i t i g n o r e s what I w i l l c a l l t h e "clue s e l e c t i o n " problem. This seemed reasonable given t h a t t h e y d o tend t o s t g g e s t "supermarket", a s d e s i r e d . But there is more information i n t h e sentence. It was Jack who d i d a l l o f t h i s . Why not i n t e r s e c t what we know about Jack with a l l of t h e r e s t , o r WALK? Or a g a i nsupgose something ever so s l i g h t l y odd h a m . such a s t h e basket h i t t i n g a qcrewdriver which i s on t h e Eloor.SCREWDRIVER w i l l have vacioua t h i n g s indexed undqr i t , but more l i k e l y t h a n not t i n t e r k e c t i o n with t h e rest of t h e i t e m s mentioned above w i l l g i v e us t h e n u l l set. For t h a t m a t t e r , i s t h e r e any reason w o n l y i n t e r s e c t t h i n g s i n the same sentence? The answer here i s c l e a r l y n6, s i n c e t h e r e a r e many examples which r e q u i r e j u s t t h e opposite.Jack was walking, down an a i s l e . Fur thermore , t h e r e a r e some problems with t h e c l u e i n t x r s e c t i s n method which go beyond t h e mere a r e themselves dependent on havlng t h e c o n t e x t frames a v a i l a b l e . That i s t o s a y , before we can rule o u t SUPhRMARKET, we need some p i e c e o f information from t h e SUPERMARaT frame which w i l l e n a b l e us t o say t h a t Jack should not b e t u r n i n g on a l i g h t . g i v e n t h a t he is cast i n t h e r o l e o f SHOPPER i n t h a t frame-I n t e r e s t i n g l y e n o w h , Fahlman (who I e a r l i e r noted i s a proponent o f t h e c l u e i n t e r s e c t i o n method) had a major role i n t h e e v o l u t i o n o f t h e Minsky proposal whic4 I advocate. It seems reaeonable: t o a s s m e t h a t we guess even b e f o r e t h e second s e n t e n c e t h a t J a c k w l l l make a c a l l . Tb a n t i c i p a t e t h i s wr, must have TELEF?iOWXNG indexed under TELEPHONE.When we see t h e fir a t l i n e we first t r y t o ' i n t e g r a t e i t i n t o what m a l r e a d y know. S i n c e t h e r e w i l l be nothing there t o i n t e g r a t e i t i n q o , we try A a N C muat b e i n t h e prdximity a f t h e phone, and Jack j u s t accompliqhed t h a t . Hence we a r e a b l e t o i n t e g r a t e (ATTELEPHONIhG frame, and e v e r y t h i n g is f i n e . ; I f we i n s t a n t i a t e t h e ROOM frame t h e n t h e ;HOME-PHONE v a r i a b l e i n i t shquld be bound ; t o the . token which i s bound t o THING.; S i m i l a r l y f o r PUBLIC-LOC and PAY-PHONE . (TELEPHONING (PHONE . mLEPHONE-1)) [ROOM (ROOM . ROOM-1 ) ( HOMELPHONE . T E~ PHONE-1 ) )The syntax h e r e i 8 t h e name o f t h e frame followed [ROOM-1 (UNIT)<SELF ( a ROOM w i t h HOME-PHONE TELEPHONE-1 )> ]So,= a r e hypotheslzlng 1) an i n s t a n c e o f telephoning, where t h e o n l y t h i n g w e know about i t i s t h e telephone involved, and 2 ) a room (ROOM-1 ) which a t t h e moment is o n l y furnished with a telephone. Note Bkat t h i s assumes t h a t i n our room frame we have a n explicit s l o t for a telephone. This i s e q u i v a l e n t t o assuming t h a t rooms t y p i c a l l y have phones i n them. W e can now i n t e g r a t e the f a c t t h a t J a c k i s at; t h e phone i n t o t h e telephoning frame, assuniq t h a t t h i s s t a t e i s e x p l i c i t l y mentioned t h e r e ( i . e . we h o w t h a t a s p a r t o f telephoning t h e A a m must be AT t h e TEUPHONE). With t h i s added o w TELEPHONING statement w i l l now be:(TELEPHONING (AGENT . JACK-1 ) (TELEPHONE . TELEPHONE-1))wnen cne secona Arne cumerr AII 1~e must see how t h i s f i t s i n t o t h e TELEPHONING frame, but this i s a problem of i n t e g r a t i o n Zf we a r e t o l d t h a t J a c k i s i n a r e s t a u r a n t we must a c t i v a t e RESTAURANTING. 1 T o u r c u r r q n t a n a l y s i s (RESTAURANT (THINGOBJECT frame and hence w i l l o n l y be looking for LOCATIONS i n which t h e r e s t a u r a n t w i l l f i t . Hence i n t h i s c a s e t h e I N frame must a c t l i k e t h e GO frame i n looking f o r ACTION i n d e c i e s i n which it might f i t .More g e n e r a l l y , any state which i s t y p i c a l l y modsffed by a n a c t i o n should cause us The car was green. Jack had to be home by t hrce .would happen because t h e f a c t that t h e c a r i s g r e e n would not i n t e g r a t e i n t o DRIVING.) However. much t p my s u r p r i s e , when I gave t h i s example t o peOople t h e y d i d not g e t t h e DRIVING frame e i t h e r . Hoyever, w i t h a modified example t h e y do.The s t e e r i n g wheel was g r e e n . Jack had t o be home by t h r e e . This i s most mysterious. One s u g g e s t i o n (Lehnert personal communication) i s t h a t t o "see" t h e s t e e r i n g wheel t h e "viewer" must be i n t h e c a r . which i n t u r n suggests d r i v i n g ( s i n c e I N would demand a c t i o n i n t e g r a t i o n ) . T h i s may indeed be c o r r e c t , but we must t%en ex p l a i n why i n t h e f i r st example t h e k c t t h a t t h e viewer must be NEAR t h e car does n o t cause t h e same t h i n g . In any c a s e however, t h e s e examples a r e s u f f i c i e n t l y odd that i t seems i n a d v i s a b l e t o mold a t h e o r y around them. | null | null | null | There i s one way i n which t h e telephone example makes t h e problem look simplgr than i t i s . The woman waved t h e man on t h e s t a g e sawed h e r i n h a l f . Here i t would seem t h a t t h e n o t i o n o f sawing a person i n h a l f $ 8 t h e c r u t~a l concept which l e a d s u s t o magic, although t h e f a c t t h a t t h e woman does not seem conc2rned. and t h e e n t i r e t h i n g i s happening on a s t a g e c e r t a i n l y h e l p re-enforce t h i s idea.But p r e s u l a b l y t h e output o f our parser w l l l simply s t a t e t h a t we have here a n If i t were a MOVE w i t h t h e EARTH a s t h e t h i n g moved then EARTHQUAKE.Note however that i f there were few enough t h i n g s a t t a c h e d t o SAWING our n e t would not save s i g n i f i c a n t t i m e . | Main paper:
more complex 1ndt.c es:
There i s one way i n which t h e telephone example makes t h e problem look simplgr than i t i s . The woman waved t h e man on t h e s t a g e sawed h e r i n h a l f . Here i t would seem t h a t t h e n o t i o n o f sawing a person i n h a l f $ 8 t h e c r u t~a l concept which l e a d s u s t o magic, although t h e f a c t t h a t t h e woman does not seem conc2rned. and t h e e n t i r e t h i n g i s happening on a s t a g e c e r t a i n l y h e l p re-enforce t h i s idea.But p r e s u l a b l y t h e output o f our parser w l l l simply s t a t e t h a t we have here a n If i t were a MOVE w i t h t h e EARTH a s t h e t h i n g moved then EARTHQUAKE.Note however that i f there were few enough t h i n g s a t t a c h e d t o SAWING our n e t would not save s i g n i f i c a n t t i m e .
:
i s n o t a l q a y s so easy.('%+ woman way@ while-the man on t h e s t a g e sawed her i n h a l f " s w g e a t s MAGICIAN b u t how?) T h i s paper w i l l examine how a program might g o about determining t h e a p p r o p r i a t e frame in such cases.A t a s u f f i c i e n t l y sag* l e v e l the model presented h e r e w i l l reaembge t h a t o f Minsky (1975) i n it's a s s m p t i o n chat one u s u a l l y h a s a v a i l a b l e one o r more c o n t e x t frames.Hence one o n l y needs worry i f information comes i n w h i c h d o e s n o t P i t them.A s opposed t o Minsky however t h e s u g g e s t i o n s f o r new c o n t e x t frames w i l X not come from t h e old ones, but r a t h e r from t h e c o n f l i c t i n 8 information.The problem them becomes how p o t e n t i a l frames a r e indexed under t h e i n f o m a t ion which "suggestsw them.making i n f e r e n t e s from a v e r y large base of common s e n s e knowledge. To avoid d e a t h by c m b i n a t o r i a l explosion our computer must be h b l e t o ~c c e s s t h e knowledqe i t n e d s without i r r e l e v a n t knowledse ~e t t i n g i n i t s m y . A p l a u s i b l e c o n s t r a i n t on t h e knowledge we m i~h t use a t a %iven point i n a s t~r y o r c o n v e r s a t i o n ( I s h a l l henceforth simply assme we are d e a l i n g w i t h a s t o r y ) is' t o =strict c o n s i d e r a t i o n to t h a t p o r t i o n o f our knowledge which is "&bout" t h i n g s which have been mentioned i n t h e d i s c o u r s e . So i f w e have a s t o r y which menttons t r a i n s and t r a i n s t a t i o n s , we w i l l not use our knowledge o f , shy, circuees. T h i s r e q u i r e s , of course. t h a t g i v e n a t o p i c , such a s t r a i n s , or e a t i n g . we must be a b l e to acCess i t s knowledge without going t h r o q h a v e r y t hing we know.Hence we a r e lead i n a natural way t o something approaching a notion of "frame" (Minsky a c o l l e c t i o n o f knowledge about a s i n g l e stereotyped s t t u a t i o n . I n t h e above discussion however I have made a r a t h e r important slight o f hand. Given s t o r y we only want to consider those frames "about" t h i n g s t n the s t o r y . l b w L s i t that we d e c i d e which frames q u a l i t y 'I was a b l e t o g l o s s over t h i s because i n most s i t u a t i o n s t h e problca, a t l e a s t a t a s u r f a c e Iwel. does n o t appear a l l thst d i f f i c u l t .If t h e s t o r y i s abotlt: t r a i n s . i t will s u r e l~ meqtion t r a i n s .So we see t h e word "train". and we assume t h a t t r a i n s ate r e l e v a n t . What could be easier.Unfortunately, this ease 1s d e c e p t i v e f o r t h e s t o r y may mentton many topics of which o n l y a few a r e t r w l y important t o the s t o r y . For example.The lawyer gook a cab t o *the r e s t a u r a n t near the uni'(iereit.y.Here w? have "lawyer", "cab" , " r e s t a u r~n t " and " u n i v e r s i t~" a11 o f which are c a l l i n g f 6 r o u r a t t e n t i o n . Somehow on t h e b a s i s of latef l i n e s we must weed o u t t h o s e which our o n l y incjidentak . To s m a t i z e , t h e l a s t few paragraphs, the problem o f frame determinatioin i n language comprehension $ n v~l v e s t h r e e sub-problms .1) stories will t y p i c a l l y e I d d e t o many h i g h e r fraqks, any of which might serve as the contax& for the incoming Iinee. Nnw d o we choose between them? 2 ) The words used i n a s t o r y may not directly i n d i c a t e t h e proper higher frame. l?uw In tbe paper which follows I w i l l be p r i m a r i l y concentrate 3n ( 2 ) w i t h (3) b e i q mentfoned o c c a s i o n a l l y . I n essence my p o s i t i o n on ( 1 i a t h a t it w i l l not be t o o much o f a problem, provided t h a t &he c o s t o f s e t t i n g up a c o n t e x t like " r w t a u r a n t " i~ small. f f i t ; i s never used then a s Fhe stary g o e s on it w i l l teceeded l n t a the background How t h i s "receed ing " takes place I shall, not s a y . a i n c e f o r one thing i t is 4 problem i n many a r e a s , and f o r a n o t h e r , 1 don't know.Concerning 2and ( 3 ) , we w i l l be lead t o a p 0 8 i t i a h similar t o t h a t o f MLnsky (1975) We w i l l s e e however, t h a t t h e r e a t e s t i l l a l o t o f problema with t h i s * polpMion which d o not a t f i r s t glance me& t h e eye.' 1 THE CLUE INTERSECTION METHOD Rather than immediately presenting my scheme, let m e s t a r t by showing the problems with a n a l b r n a t i v e p o s s i b i l i t y , which I w i l l c a l l t h e "clue i n t e r s e c t i o n " metbod. This a l t e r n a t i v e is by no means a strats man a s one researcher has i n f a c t e x p l i c i t l y suggested i t (Fahlman 1977) and I f o r one find i t a v e r y n a t u r a l way of t h i n k i n g about t h e problem.The i d e a behind t h i s mkghod i s t h a t we a r e given oergain c l u e s i n t h e istory about t h e n a t u r e of the c o r r e c t Erame, and t o find t h e frame we simply i n t The c l u e s h e r e a r e t h i n g s l i k e "aisle", " t u n a f i s h ' e t c .the English w r d s which a r e the c l u e s , but r a t h e r the concepts which underlie the words. I w t l l assuue t h a t we go from one t o the o t h e r v i a a n independent? parsing algorithm.(However t h i s assumes t h a t t h e r e Is no v i c i o u s i n t e r a c t i o n b e t w e n f r b e d s e r m i n a t i o n and d lsambigua t i o n . Given t h a t dieadbiguation depends W U p r i o r frame determinatiop (see (Hayes 1977) for numerous examples) t h i s may be i n c o r r e c t . ) So the input t o the frame d e t e r m i n~r w i l l be something l i k e SUPEWA&KET. The point i s t h a t none o f these e l m s w i l l be unambiguous, but when we take the i n t e r s e c t i o n the qnly t h i n g which vill be l e f t i s SUPERMARKET.ST-1 (WALK JACK-1 AISLE-1) ST-2 ( PERSON JAC K-1 ) ST-3 (EQUAL (NAME JACK-1 ) "JACK") S T 4 (EQUAL (SEX JACK-1 ) MALE) ST-5 (AISLE AISLE-1 ST06 (PUT JACK-I TUNA-FISH<@-1 BASKET-1) ST97 (BASKET B&RET-I)There are, however, problems with t h i s v i e w of t h i n g s .For one t h i n g i t i g n o r e s what I w i l l c a l l t h e "clue s e l e c t i o n " problem. This seemed reasonable given t h a t t h e y d o tend t o s t g g e s t "supermarket", a s d e s i r e d . But there is more information i n t h e sentence. It was Jack who d i d a l l o f t h i s . Why not i n t e r s e c t what we know about Jack with a l l of t h e r e s t , o r WALK? Or a g a i nsupgose something ever so s l i g h t l y odd h a m . such a s t h e basket h i t t i n g a qcrewdriver which i s on t h e Eloor.SCREWDRIVER w i l l have vacioua t h i n g s indexed undqr i t , but more l i k e l y t h a n not t i n t e r k e c t i o n with t h e rest of t h e i t e m s mentioned above w i l l g i v e us t h e n u l l set. For t h a t m a t t e r , i s t h e r e any reason w o n l y i n t e r s e c t t h i n g s i n the same sentence? The answer here i s c l e a r l y n6, s i n c e t h e r e a r e many examples which r e q u i r e j u s t t h e opposite.Jack was walking, down an a i s l e . Fur thermore , t h e r e a r e some problems with t h e c l u e i n t x r s e c t i s n method which go beyond t h e mere a r e themselves dependent on havlng t h e c o n t e x t frames a v a i l a b l e . That i s t o s a y , before we can rule o u t SUPhRMARKET, we need some p i e c e o f information from t h e SUPERMARaT frame which w i l l e n a b l e us t o say t h a t Jack should not b e t u r n i n g on a l i g h t . g i v e n t h a t he is cast i n t h e r o l e o f SHOPPER i n t h a t frame-I n t e r e s t i n g l y e n o w h , Fahlman (who I e a r l i e r noted i s a proponent o f t h e c l u e i n t e r s e c t i o n method) had a major role i n t h e e v o l u t i o n o f t h e Minsky proposal whic4 I advocate. It seems reaeonable: t o a s s m e t h a t we guess even b e f o r e t h e second s e n t e n c e t h a t J a c k w l l l make a c a l l . Tb a n t i c i p a t e t h i s wr, must have TELEF?iOWXNG indexed under TELEPHONE.When we see t h e fir a t l i n e we first t r y t o ' i n t e g r a t e i t i n t o what m a l r e a d y know. S i n c e t h e r e w i l l be nothing there t o i n t e g r a t e i t i n q o , we try A a N C muat b e i n t h e prdximity a f t h e phone, and Jack j u s t accompliqhed t h a t . Hence we a r e a b l e t o i n t e g r a t e (ATTELEPHONIhG frame, and e v e r y t h i n g is f i n e . ; I f we i n s t a n t i a t e t h e ROOM frame t h e n t h e ;HOME-PHONE v a r i a b l e i n i t shquld be bound ; t o the . token which i s bound t o THING.; S i m i l a r l y f o r PUBLIC-LOC and PAY-PHONE . (TELEPHONING (PHONE . mLEPHONE-1)) [ROOM (ROOM . ROOM-1 ) ( HOMELPHONE . T E~ PHONE-1 ) )The syntax h e r e i 8 t h e name o f t h e frame followed [ROOM-1 (UNIT)<SELF ( a ROOM w i t h HOME-PHONE TELEPHONE-1 )> ]So,= a r e hypotheslzlng 1) an i n s t a n c e o f telephoning, where t h e o n l y t h i n g w e know about i t i s t h e telephone involved, and 2 ) a room (ROOM-1 ) which a t t h e moment is o n l y furnished with a telephone. Note Bkat t h i s assumes t h a t i n our room frame we have a n explicit s l o t for a telephone. This i s e q u i v a l e n t t o assuming t h a t rooms t y p i c a l l y have phones i n them. W e can now i n t e g r a t e the f a c t t h a t J a c k i s at; t h e phone i n t o t h e telephoning frame, assuniq t h a t t h i s s t a t e i s e x p l i c i t l y mentioned t h e r e ( i . e . we h o w t h a t a s p a r t o f telephoning t h e A a m must be AT t h e TEUPHONE). With t h i s added o w TELEPHONING statement w i l l now be:(TELEPHONING (AGENT . JACK-1 ) (TELEPHONE . TELEPHONE-1))wnen cne secona Arne cumerr AII 1~e must see how t h i s f i t s i n t o t h e TELEPHONING frame, but this i s a problem of i n t e g r a t i o n Zf we a r e t o l d t h a t J a c k i s i n a r e s t a u r a n t we must a c t i v a t e RESTAURANTING. 1 T o u r c u r r q n t a n a l y s i s (RESTAURANT (THINGOBJECT frame and hence w i l l o n l y be looking for LOCATIONS i n which t h e r e s t a u r a n t w i l l f i t . Hence i n t h i s c a s e t h e I N frame must a c t l i k e t h e GO frame i n looking f o r ACTION i n d e c i e s i n which it might f i t .More g e n e r a l l y , any state which i s t y p i c a l l y modsffed by a n a c t i o n should cause us The car was green. Jack had to be home by t hrce .would happen because t h e f a c t that t h e c a r i s g r e e n would not i n t e g r a t e i n t o DRIVING.) However. much t p my s u r p r i s e , when I gave t h i s example t o peOople t h e y d i d not g e t t h e DRIVING frame e i t h e r . Hoyever, w i t h a modified example t h e y do.The s t e e r i n g wheel was g r e e n . Jack had t o be home by t h r e e . This i s most mysterious. One s u g g e s t i o n (Lehnert personal communication) i s t h a t t o "see" t h e s t e e r i n g wheel t h e "viewer" must be i n t h e c a r . which i n t u r n suggests d r i v i n g ( s i n c e I N would demand a c t i o n i n t e g r a t i o n ) . T h i s may indeed be c o r r e c t , but we must t%en ex p l a i n why i n t h e f i r st example t h e k c t t h a t t h e viewer must be NEAR t h e car does n o t cause t h e same t h i n g . In any c a s e however, t h e s e examples a r e s u f f i c i e n t l y odd that i t seems i n a d v i s a b l e t o mold a t h e o r y around them.
Appendix:
| null | null | null | null | {
"paperhash": [
"fahlman|netl:_a_system_for_representing_and_using_real-world_knowledge",
"charniak|framed_painting:_the_representation_of_a_common_sense_knowledge_fragment",
"hayes|some_association-based_techniques_for_lexical_disambiguation_by_machine"
],
"title": [
"NETL: A System for Representing and Using Real-World Knowledge",
"Framed PAINTING: The Representation of a Common Sense Knowledge Fragment",
"Some association-based techniques for lexical disambiguation by machine"
],
"abstract": [
"Abstract : This report describes a knowledge-base system in which the information is stored in a network of small parallel processing elements--node and link units--which are controlled by an external serial computer. Discussed is NETL, a language for storing real-world information in such a network. A simulator for the parallel network system has been implemented in MACLISP, and an experimental version of NETL is running on this simulator. A number of test-case results and simulated timings will be presented. (Author)",
"This paper presents a “frame” representation for common sense knowledge and uses it to formalize our knowledge of “mundane” painting (walls, not portraits). These frames, while designed to aid a computer program to understand stories about the painting process, should be of use to programs which attempt to actually carry out the activity. The paper stresses a “deep” understanding of the activity so that the representation indicates not only what steps to carry out, but also how to do them, and why they should be done. To accomplish this, while at the same time preserving modularity and nonredundancy, a system of interframe pointers is introduced (the COMES-FROM and LEADS-TO pointers) which explain how or why something is done in terms of knowledge given in other frames. The paper proceeds by steadily deepening an initial English-like description of the activity, and a context free grammar for the representation is included.",
"These Ecole polytechnique federale de Lausanne EPFL, n° 277 (1977) Reference doi:10.5075/epfl-thesis-277Print copy in library catalog Record created on 2005-03-16, modified on 2016-08-08"
],
"authors": [
{
"name": [
"S. Fahlman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Eugene Charniak"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"P. Hayes"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null
],
"s2_corpus_id": [
"15562838",
"26014138",
"60714577"
],
"intents": [
[
"background"
],
[
"methodology"
],
[]
],
"isInfluential": [
false,
false,
false
]
} | null | 554 | null | null | null | null | null | null | null | null | null |
5f05ba70552b4cf4fe8177bb2634b36e244b7439 | 219310088 | null | An Argument about the Composition of Conceptual Structure | This research was eupported in part by a Fellowship fw Independent Study and Reeearoh Prom the National Endowment f o r the Humanities. | {
"name": [
"Jackendoff, Ray"
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 13 | 0 | null | null | null | null | r e h a s been more-=or less t a k e n f o r g r a n t e d ( e s p e c i a l P y by t h e A1 community), t h e need t o c o n s i d e r i t s e r i o u s l y h a s been brought t o t h e a t t e n t i o n o f 1 i n g u L s t s r a t h e r r e c e n t l y , by such works a s Fodor (1975) and Miller and Johnson-Laird (1976) T h i s paper w i l l p r e s e n t a combination of l i n gu i s t i c and v i s u a l e v i d e n c e which b e a r s o n t h e n a t u r e of c o n c e p t u a l~t r u c t u r e . The who* of t h i s c o n f i g u r a t i o n can form a v i s u a l f i g u r e seen a g a i n s t t h e background of t h e page. P a r t s a£ i t can a l s o emerge s p o n t a n e o u s l y , a s f ig u r e s ; probably t h e most prominent are a s q u a r e and an X, each of which can seen a g a i n s t t h e rest of t h e page ( i n c l u d i n g t h e rese of t h e c o n f i g u r a t i o n ) as background. Among Less n a t u r a l f igiires, which emerge o n l y w i t h more . d e l i b e r a t e e f f o r t from F i g u r e 1, are such c o n f i g u r a t 3. R e l a t i v e s a l i e n c e of p e r c e i v e d f i g u r e s is a f u n c t i o n of b o t h f e a t u r e s o s t h e p h y s i c a l s i g n a l and p r o p e r t i e s of t h e v i s u a l system. 4. F e a t u r e s of t h e v i s u a l c o n t e x t can a f f e c t r e l a t i v e s a l i e n c e of f i g u r e s . .For example, t h e c o n f i g u r a t i o n s i n F i g u r e 2 become much more l i k e l y t o emerge from F i g u r e 1 upon present a t i o n of "I bought t h a t yesterday--isn't i t gorgeous?" Speaker B, w a b l e t o make o u t anything i n t h e p i c t u r e , doesn't f u l l y understand the u t t e r a n c e and responds "What are you t a l k i n g about?" Suppose A then s a y s , "That boat!" B peers at t h e p i c t u r e and s u r e enough t h e f i g u r e of a boat emerges. H e has a minw aha-experience: "Oh, that! How could I m i s s i t ? " H e now has received t h e message and discourse can continue.Every r e a d e r has probably had an experience like t h i s ; i t s relevance i n t h e p r e s e n t s e t t f n g is as follows: i n order f o r a pragmatically c o n t r o l l e d pronoun t o be understood, i t s intended r e f e r e n t must emerge a s a fdgure i n t h e mind of t h e hearer, t h a t is, it must have a represent a t i o n as a f i g u r a l expression i n c m c e p t u a l s t r u c t u r e . Thus we have e s t a b l i s h e d an i m p o r t a~t connection between t h e figure-ground phenomenon ahd pragmatic anaphora.So fHr we have d e a l t only with f i g u r e s t h a t correspond t o thiiags (or t h e i r shapes). By and large t h i s has been t h e kgnd of f i g u r e t h a t has been inv-tigated, (3) Do it:( H a n k e r attempts to stuff a 9" b a l l through a 6" hoop) Sag: l t t s n o t clear y o u ' l l be a b l e t o do i t . If any reduction is t o t a k e place, i t w i l l be i n the theory of perception, which now must e x p l a i n t h e r e l a t i o n of r e t i n a l ( a~d a u d i t o r y , etc.) s t i m u l i t o event-and place-perception as w e l l as, thing-perceptdon, If t h i s view i s c o r t e c t , one would expect t h e s e o t h e r a s p e c t s of perception t o have many of t h e same g e s t a l t p r o p e r t i e s a s thing-perception, dependenct on proximity, clo-'sure, "good form," and s o f o r t h . I n f a c t , t h e few p i e c e s of work I know of on perception of e n t i t i e s o t h e r than t h i n g s (Michotte (1954) on causation, Jenkins, Wald, and P i t tenger (1976) on e v e n t s , remarks of U e x (1947, pp. 89- | The main argument of t h i s paper combined perceptual and l i n g u i s t i c evidence t o show t h a t f i g u r a l e x p r e s s i o n s i n conceptual s t r u c t u r e must i n c l u d e e n t i t i e s of a g r e a t number of o n t o l o g i c a l t y p e s , I t a k e t h i s to be a prototype f o r a novel s o r t of l i n g u i s t i c argumentation-one t h a t treats d e s c r i p t i v e semantics a s fundamentally a psychological r a t h e r than l o g i c a l d i s c i p l i n e , a n d w h i c h seeks t o account for t h e n a t u r e of thought and of human experience through grammatical s t r u c t u r e . I@ is n c t c l e a r t h a t t h i s i s l i n g u i s t i c s An t h e u s u a l sense any more. Rather i t is an attempt t o use l i n g u i s t i c theory a s a t o o l ~f c o g n i t i v e psychology. This seems t o m e t o b e a promising way t o go. | Main paper:
conclusion:
The main argument of t h i s paper combined perceptual and l i n g u i s t i c evidence t o show t h a t f i g u r a l e x p r e s s i o n s i n conceptual s t r u c t u r e must i n c l u d e e n t i t i e s of a g r e a t number of o n t o l o g i c a l t y p e s , I t a k e t h i s to be a prototype f o r a novel s o r t of l i n g u i s t i c argumentation-one t h a t treats d e s c r i p t i v e semantics a s fundamentally a psychological r a t h e r than l o g i c a l d i s c i p l i n e , a n d w h i c h seeks t o account for t h e n a t u r e of thought and of human experience through grammatical s t r u c t u r e . I@ is n c t c l e a r t h a t t h i s i s l i n g u i s t i c s An t h e u s u a l sense any more. Rather i t is an attempt t o use l i n g u i s t i c theory a s a t o o l ~f c o g n i t i v e psychology. This seems t o m e t o b e a promising way t o go.
:
r e h a s been more-=or less t a k e n f o r g r a n t e d ( e s p e c i a l P y by t h e A1 community), t h e need t o c o n s i d e r i t s e r i o u s l y h a s been brought t o t h e a t t e n t i o n o f 1 i n g u L s t s r a t h e r r e c e n t l y , by such works a s Fodor (1975) and Miller and Johnson-Laird (1976) T h i s paper w i l l p r e s e n t a combination of l i n gu i s t i c and v i s u a l e v i d e n c e which b e a r s o n t h e n a t u r e of c o n c e p t u a l~t r u c t u r e . The who* of t h i s c o n f i g u r a t i o n can form a v i s u a l f i g u r e seen a g a i n s t t h e background of t h e page. P a r t s a£ i t can a l s o emerge s p o n t a n e o u s l y , a s f ig u r e s ; probably t h e most prominent are a s q u a r e and an X, each of which can seen a g a i n s t t h e rest of t h e page ( i n c l u d i n g t h e rese of t h e c o n f i g u r a t i o n ) as background. Among Less n a t u r a l f igiires, which emerge o n l y w i t h more . d e l i b e r a t e e f f o r t from F i g u r e 1, are such c o n f i g u r a t 3. R e l a t i v e s a l i e n c e of p e r c e i v e d f i g u r e s is a f u n c t i o n of b o t h f e a t u r e s o s t h e p h y s i c a l s i g n a l and p r o p e r t i e s of t h e v i s u a l system. 4. F e a t u r e s of t h e v i s u a l c o n t e x t can a f f e c t r e l a t i v e s a l i e n c e of f i g u r e s . .For example, t h e c o n f i g u r a t i o n s i n F i g u r e 2 become much more l i k e l y t o emerge from F i g u r e 1 upon present a t i o n of "I bought t h a t yesterday--isn't i t gorgeous?" Speaker B, w a b l e t o make o u t anything i n t h e p i c t u r e , doesn't f u l l y understand the u t t e r a n c e and responds "What are you t a l k i n g about?" Suppose A then s a y s , "That boat!" B peers at t h e p i c t u r e and s u r e enough t h e f i g u r e of a boat emerges. H e has a minw aha-experience: "Oh, that! How could I m i s s i t ? " H e now has received t h e message and discourse can continue.Every r e a d e r has probably had an experience like t h i s ; i t s relevance i n t h e p r e s e n t s e t t f n g is as follows: i n order f o r a pragmatically c o n t r o l l e d pronoun t o be understood, i t s intended r e f e r e n t must emerge a s a fdgure i n t h e mind of t h e hearer, t h a t is, it must have a represent a t i o n as a f i g u r a l expression i n c m c e p t u a l s t r u c t u r e . Thus we have e s t a b l i s h e d an i m p o r t a~t connection between t h e figure-ground phenomenon ahd pragmatic anaphora.So fHr we have d e a l t only with f i g u r e s t h a t correspond t o thiiags (or t h e i r shapes). By and large t h i s has been t h e kgnd of f i g u r e t h a t has been inv-tigated, (3) Do it:( H a n k e r attempts to stuff a 9" b a l l through a 6" hoop) Sag: l t t s n o t clear y o u ' l l be a b l e t o do i t . If any reduction is t o t a k e place, i t w i l l be i n the theory of perception, which now must e x p l a i n t h e r e l a t i o n of r e t i n a l ( a~d a u d i t o r y , etc.) s t i m u l i t o event-and place-perception as w e l l as, thing-perceptdon, If t h i s view i s c o r t e c t , one would expect t h e s e o t h e r a s p e c t s of perception t o have many of t h e same g e s t a l t p r o p e r t i e s a s thing-perception, dependenct on proximity, clo-'sure, "good form," and s o f o r t h . I n f a c t , t h e few p i e c e s of work I know of on perception of e n t i t i e s o t h e r than t h i n g s (Michotte (1954) on causation, Jenkins, Wald, and P i t tenger (1976) on e v e n t s , remarks of U e x (1947, pp. 89-
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0 | null | null | null | null | null | null | null | null |
7babc23ab71c25a78d13e21fd1d614168bb933d9 | 219310175 | null | A Note on Partial Match of Descriptions. Can One Simultaneously Question (Retrieve) and Inform (Update)? | In data base query systems there is a n ~m p l x c i t assumption tha* de-iptions i n queries must rilatch exactly, i.e., queries are f o r r e t r i e v a l | {
"name": [
"Joshi, Aravind K."
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 1 | 0 | null | null | null | null | h v m d K. JoshiThe Moore School, University of Pennsylvania Philadelphia, Pa. 19104 S y : In data base query systems there is a n ~m p l x c i t assumption tha* de-iptions i n queries must rilatch exactly, i.e., queries are f o r r e t r i e v a l only, and not f o r r e t r i e v a l and updating simultaneously. A related assumption (or constraint) that in quest-ions descriptions are used refere n t i a l l y only ( i . e . , a quest ion cannot be used siPlultaneously for questioning and informing) seems t o hold in ordinary conversations also, with sorrlc qualifications. Som issues related to t h e yalidity of spch a con3rain-t and its relatior1 t o partial matching of. aescrifit ions are b r i e f l y discuSsed in t h i s note.In a question-answer sygtem each descriptiorl m a query is used referent ial3 y i .e. , f o r each description one expects t o f iml an e n t i t y in the Gats tase which serves as the unique r e f e r e n t for 'that descrription. For simplicity, hereafter we w i l l consider only d e f i n i t e descript lons (in particular, d e f i n i t e noun phrases consisting of a d e f i n i t e a r t i c l e , an adjective, and a noun). Thusin (1)(1) Is t h e red book on the table?t h e description t h e red book wi2l serve t o i d e n t i f y an e n t i t y , say, el i n t h e d a t a base2 and t h e description t h e table, an e n t i t y , say, e2. The question can be answered a f t e r verifying t h e appropriate r e l a t i o n between el and e2. For t h e purpose of making t h e definiteness tP.anspar&t and also f o r simplifying t h e discussion in t h i s note, let hs assume t h a t t h e r e is exactly nno h k and one t a b l e in the data base.The match for t h e red b30k can succeed i f el -has a color a t t r i b u t e with the value red. The match can f a i l e i t h e r due to a mismatch or a ial m t e h . A mismatch w i f l occur i f e l has a =or value other than red, say green. A partialmatch w i l l occur i f el has an &specified value f o r t h e color a t t r h t e or i f t t h e possession of t h e color a t t r i b u t e i t s e l f has not been specified f o r el.In t h e rest of the discussion, we w i l l not be wncermed with failure due t o mismatch, a l t b u g h many of t h e issues mised below are q u i t e relevant to this case also. W e w i l l be concerned with partial m t c h e s only. A partial match r e a l l y is 3ppart i a l l y successful match, when\ a pdrt of the jesqript ion has mt'ched exactly , a d the reminderr: ha$ f a i l 4 to match due $0 t h e lack of same hlfioml ion, and not due 1s a rnimt ch.k t us consider the case of a partidl m tch where t h e tart of t h e description that wtched 1s sufficient to identify t h e *referent uniquely. I n (7 t h i s is t r i v i a l l y accmplishcd because of our 2. On the other hand, wc may ,sume t h a t whenever we have a partial match and the refemnts are uniquely identified s o l n e b , we should answer the question, and treat that part of the description which was not acmunted for as new infoxmation. This new i n f o m t ion can then be used t o update the d a t a base. Thus for the question (21, i f the partial match is due to t h e f a c t t h a t i n the data base t h e value for t h e color attribute for e l is not specified, then wa can now specify it t o bered. I f , on the other hand, t h~ partial match was due to tM fact t h a t the possession of t h e color a t t r i b u t e itself is not specified for e l , then t l c updating, would invalve ( 6 1 W n did the p u o h leave? ( 7 ) When did the prr.;on Iiwvc? (8) Hc is a r n t l r h .With evaluative information; simd tanebusly queet ioning andinforming appears PO be a b i t more convenient. If (6) is used by the speaker, it appears that the h w can ulpdatd his model, wit]-lout any int-pting respanses, with the attribute p u c h y attached t o the a t ity, as speaker's evaluation* (and the hearer's too if he agrees wiM the speaker). 'Even if the hearer asks for clarif i q t i o n , it is likely to ,be of the form Oh! I didn ' t know that vou _ t h o k h t he was a gmuch mther than Oh! i didn't knw that h e W -a y c h (campare this to the resppse i n t h e p v ious example 1 . b3) Finally, there is an apparent violdtion 6f the hypothesis in examples such as (9). for uphting has to be mehow relevant to the'old1 i n f~r m a t i o r~, either by being i,t$embl~ f w m i t or by being able to f i t it &to the tli~ecurse s t r n c t m cwated sc, far, PZC. 10 1. 'Phis mrk is partifllly supprted py NSf Crdnt MC576-19466. ,I wish to thank Jcmy k p l a n , Lnm*lc levin, Stan Rosenschein, Ivcm !hg, ard R n n n i~ Webbr for valuable d fi,russ ioris h e of the issues =is& here w i l l be. tlowevcr, in geneml, unique reference may be established due to the context, and the strrtctum and content of the data bse. 4. In t h e data b s e context, updates mb t~c , l h~ 1 ly cant ~n t u,dn t~s . Stnlctwrr up'htec: ilre no1 | null | Main paper:
:
h v m d K. JoshiThe Moore School, University of Pennsylvania Philadelphia, Pa. 19104 S y : In data base query systems there is a n ~m p l x c i t assumption tha* de-iptions i n queries must rilatch exactly, i.e., queries are f o r r e t r i e v a l only, and not f o r r e t r i e v a l and updating simultaneously. A related assumption (or constraint) that in quest-ions descriptions are used refere n t i a l l y only ( i . e . , a quest ion cannot be used siPlultaneously for questioning and informing) seems t o hold in ordinary conversations also, with sorrlc qualifications. Som issues related to t h e yalidity of spch a con3rain-t and its relatior1 t o partial matching of. aescrifit ions are b r i e f l y discuSsed in t h i s note.In a question-answer sygtem each descriptiorl m a query is used referent ial3 y i .e. , f o r each description one expects t o f iml an e n t i t y in the Gats tase which serves as the unique r e f e r e n t for 'that descrription. For simplicity, hereafter we w i l l consider only d e f i n i t e descript lons (in particular, d e f i n i t e noun phrases consisting of a d e f i n i t e a r t i c l e , an adjective, and a noun). Thusin (1)(1) Is t h e red book on the table?t h e description t h e red book wi2l serve t o i d e n t i f y an e n t i t y , say, el i n t h e d a t a base2 and t h e description t h e table, an e n t i t y , say, e2. The question can be answered a f t e r verifying t h e appropriate r e l a t i o n between el and e2. For t h e purpose of making t h e definiteness tP.anspar&t and also f o r simplifying t h e discussion in t h i s note, let hs assume t h a t t h e r e is exactly nno h k and one t a b l e in the data base.The match for t h e red b30k can succeed i f el -has a color a t t r i b u t e with the value red. The match can f a i l e i t h e r due to a mismatch or a ial m t e h . A mismatch w i f l occur i f e l has a =or value other than red, say green. A partialmatch w i l l occur i f el has an &specified value f o r t h e color a t t r h t e or i f t t h e possession of t h e color a t t r i b u t e i t s e l f has not been specified f o r el.In t h e rest of the discussion, we w i l l not be wncermed with failure due t o mismatch, a l t b u g h many of t h e issues mised below are q u i t e relevant to this case also. W e w i l l be concerned with partial m t c h e s only. A partial match r e a l l y is 3ppart i a l l y successful match, when\ a pdrt of the jesqript ion has mt'ched exactly , a d the reminderr: ha$ f a i l 4 to match due $0 t h e lack of same hlfioml ion, and not due 1s a rnimt ch.k t us consider the case of a partidl m tch where t h e tart of t h e description that wtched 1s sufficient to identify t h e *referent uniquely. I n (7 t h i s is t r i v i a l l y accmplishcd because of our 2. On the other hand, wc may ,sume t h a t whenever we have a partial match and the refemnts are uniquely identified s o l n e b , we should answer the question, and treat that part of the description which was not acmunted for as new infoxmation. This new i n f o m t ion can then be used t o update the d a t a base. Thus for the question (21, i f the partial match is due to t h e f a c t t h a t i n the data base t h e value for t h e color attribute for e l is not specified, then wa can now specify it t o bered. I f , on the other hand, t h~ partial match was due to tM fact t h a t the possession of t h e color a t t r i b u t e itself is not specified for e l , then t l c updating, would invalve ( 6 1 W n did the p u o h leave? ( 7 ) When did the prr.;on Iiwvc? (8) Hc is a r n t l r h .With evaluative information; simd tanebusly queet ioning andinforming appears PO be a b i t more convenient. If (6) is used by the speaker, it appears that the h w can ulpdatd his model, wit]-lout any int-pting respanses, with the attribute p u c h y attached t o the a t ity, as speaker's evaluation* (and the hearer's too if he agrees wiM the speaker). 'Even if the hearer asks for clarif i q t i o n , it is likely to ,be of the form Oh! I didn ' t know that vou _ t h o k h t he was a gmuch mther than Oh! i didn't knw that h e W -a y c h (campare this to the resppse i n t h e p v ious example 1 . b3) Finally, there is an apparent violdtion 6f the hypothesis in examples such as (9). for uphting has to be mehow relevant to the'old1 i n f~r m a t i o r~, either by being i,t$embl~ f w m i t or by being able to f i t it &to the tli~ecurse s t r n c t m cwated sc, far, PZC. 10 1. 'Phis mrk is partifllly supprted py NSf Crdnt MC576-19466. ,I wish to thank Jcmy k p l a n , Lnm*lc levin, Stan Rosenschein, Ivcm !hg, ard R n n n i~ Webbr for valuable d fi,russ ioris h e of the issues =is& here w i l l be. tlowevcr, in geneml, unique reference may be established due to the context, and the strrtctum and content of the data bse. 4. In t h e data b s e context, updates mb t~c , l h~ 1 ly cant ~n t u,dn t~s . Stnlctwrr up'htec: ilre no1
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0 | null | null | null | null | null | null | null | null |
34fbaa53e12ebb785641d1d62ac751e42dbc219e | 219310277 | null | A Critical Look at a Formal Model for Stratificational Linguistics | We present here a formalization of the straiificational model of linguistics proposed by Sampson C131 and investigate its generative power. In addition to uncovering a number of counterintuitive properties, the results presented here bear on meta-theoretic claims found in the linguistic literature. For example, Postal [ l l j claimed that stratificational theory was equivalent to context-free phrase-structure grammar, and hence not worthy of further interest. We show, however, that Sampson's model, and several of its restricted versions, allow a far wider range of generative powers. In the cases where the model appears to be too powerful, we suggest possible alterations which may make it more acceptable. | {
"name": [
"Borgida, Alexander T."
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 5 | 0 | null | Linguistic theories are at least partially interested in presenting the regularities found in natural languages. Given the current dominance of the Transformational Generative (TG) school in the field of linguistics, it seems necessary for theories competing for attention to possess a formal model, In addition to the advantages normally derived from presenting results through a formalism, such as precision, succinctness and verifiability, one can also comment on the veracity of metatheoretic claims. It was using such formal arguments that Chomsky and his collaborators demonstrated the inability of finite automata and of context-free grammars to describe all natural language constructs. Similarly, the formal work of Peters and Ritchie [ 8 , 9 1 was important in uncovering inadequacies of two notions of TG theory namely, the "recoverability of deletions condition" and the "universal base hypothesis".Finally, since many generative linguists want grammatical theories which characterize natural languages, they fault any theory which is .too powerful" in the sense of being able to describe languages which clearly cannot be natural languages, such as nonrecursive sets. Furthermore, computer scientists working on natural languages will have to give in the future more consideration to the work of linguists, especially on "exotic" languages, in order to be able to observe a wider range of phenomena. Such access will be facilitated if the formalismsin which the grammaTs are prese'nted lend themselves to computer implementation for purpose$ such as parsing, testing, etc. This entails, among other things,that linguists should avoid as much as possible features which make their grammars generate non-recursive sets, and hence it is one of the purposes of the present paper to point out such features and discuss possible ways of avoiding them.In this paper we will discuss one model proposed for the stratificational theory of linguistics. This theory, advanced by S. Lamb, H . A . Gleason Jr. and their collaborators (C51,[61,C71), advocates that langdages be described in terms of several subsystems, known as strata. Each stratum has its own set of units and a tactics specifying the tlcorrectfl ("all~wable'~) structures that stratum. specific grammar might for example have strata corresponding roughly to semantics, syntax-morphology and phonology, although this is by no means standard. Furthermore, the strata are linearly ordered as levels, and there is a realization relation which connects adjacent strata by attaching to every well-formed structure on one stratum, zero or more accompanying structures on the adjacent strata. Note therefore that a particular utterance has simultaneous expression on each stratum.In this paper we examine the formal model for stratificational linguistics proposed by Sampson ( L 1 3 1 ) . This model uses rewrite grammars G1,G2, ... to describe the tactics, while the realization relation is essentially a rewrite system R acting as a transducer between the languages of-the tactics. More specifically, realization connects adjacent tactics G and G j + l j by matching sentences u in the language generated by G with those sentences j v in the language of G j + l which can be derived from u by using rules from R. An important property of the linguistic realization relation is the fact that' every structure on some stratum can have only a finite number of llrealizates" on the next stratum.This means that the rewrite system R must be constrained so that it has no recursive symbols. Such a rewrite system will be called acyclic.We investigate here the effect of acyclic rewrite systems acting as transaucers on axiom sets, varying the type of the derivations and rules allowed.We prove in this paper that regular languages are closed under transduction by acyclic rewrite systems, but that the linear context-free languages are mapped onto the recursively enumerable sets. This implies that stratificational grammars with non-selfembedding ta~ctics would be too weak while those with even one context-free tactics would be too strong. If the realization derivation is restricted to be in some sense "leftmost", then we show that the transduction can be performed by a finite,state device known as an a transducer. Furthermore, if productions with null right-hand sides are not allowed in an acyclic rewrite system then all the derivations can be made leftmost. This provides one possible method of restricting the generative power of acyclic rewrite systems.By deriving a recursive characterization of the languages generated with n-strata in terms of (n-1)-stratal languages, we can show that if the realization is restricted to being leftmost, then the languages described are homomorphic images of the intersections of the languages generated by the tactics. In particular, this means that we can find natural families of stratificational grammars which generate far example the sets recognized in real time by nondeterministic multitape Turing machines. This result partially confirms a hitherto unproven claim by Sampson, and discredits Postal's Clll classiciation of stratlficational grammars as just another variant of context-free phrase-structure grammars.Finally, we investigate the use of ordered rules in linguistic grammars and prove that in several models they allow the generation of sets which are not even recursively enumerable a clearly unsatisfactory situation.The remainder of the paper is structured as follows,In Section 2, we present the formal definitions and notation to be used, including the formal model for stratificational grammars.In Section 3, we examine the properties of "acyclic rewrite systemsw, which form the principal novel component in our definition of stratificational grammars. We then return in Section 4 to examine the generative power of st'ratificational grammars and relate the results to linguistics. e -o u t p u t f r e e i f f o r an$t ( r , u , v , s ) i n T , t h e s t r i n g v cannot be n u l l . To b e g i n w i t h , we remark t h a t i n G t h e n A => dD i n Gi+l i n c a s e bc -+ d is i n R ( s t e p 5 ) . where we used r u l e Xw A'Yz i n s t c p 0. and l a n g u a g e s L1,. . . , L s u c h t h a t f o r i = 1,. . . ,n L(G.) t h e r e e x i s t s a -t r a n s d u c e r O n v l s u c h t h a t L-RSTRAT(RST) i s e q u a l t o *+ v . €VG $Rev ( v ) = v i -1 f o r i = 1,. . . , m -1 3 1 m + i -1 and --a' --IUN N V + L 2 = { % w l a . . .w % $ % w~+~ %...p ln>O, w . a VOn+ 1 (L-RSTRAT(TOP(RST)) n L ( G ) ) n V , . n+ 1 ( 4 )1 Meaning t y p e 3 , t y p e 2 , t y p e 1, t y p e 0 , l i n e a r l a n g u a g e ..,. , L t t such t h a t L-RSTRAT(TOP(RST)) = hTt(L!i n . . . n -Lt\3 . n n S u b s t i t u t i n g t h i s i n (4) and appllying n o t e s ( a ) , (b) and (c) we the realization derivation must be k-leftmost and a homomorphism must be applied to the intersection of the languages. C81 P e t e r s , P .S. and &W. ~i w h i e ( 1 9 7 1 ) . "On ~e s t r i c t i n g the base component of T r a n s~o r m a t i d n a l grammar^'^. Information and "Control -18. -C91 P e t e r s , P.S. and R.W. Ritchiea ('1973) . "On the g e n e r a t i v e power of Transformational ~r a r n m a r~" 1.nformati06 Sciepces 6 49-83. E 1 2 1 Salomaa , A . (1973) . Formal Languages, Academic Press,NewYork.[ 1 3 J Sampson, G . ( 1 9 7 0 ) . S t r a t i f i c a t i o n a l grammar: 9 D e f a n i -t i o n and a n Example, Janua Linguarum, S e r i e s Minor: 8 8 , The Hague Mouton. | null | null | null | null | Main paper:
introduction:
Linguistic theories are at least partially interested in presenting the regularities found in natural languages. Given the current dominance of the Transformational Generative (TG) school in the field of linguistics, it seems necessary for theories competing for attention to possess a formal model, In addition to the advantages normally derived from presenting results through a formalism, such as precision, succinctness and verifiability, one can also comment on the veracity of metatheoretic claims. It was using such formal arguments that Chomsky and his collaborators demonstrated the inability of finite automata and of context-free grammars to describe all natural language constructs. Similarly, the formal work of Peters and Ritchie [ 8 , 9 1 was important in uncovering inadequacies of two notions of TG theory namely, the "recoverability of deletions condition" and the "universal base hypothesis".Finally, since many generative linguists want grammatical theories which characterize natural languages, they fault any theory which is .too powerful" in the sense of being able to describe languages which clearly cannot be natural languages, such as nonrecursive sets. Furthermore, computer scientists working on natural languages will have to give in the future more consideration to the work of linguists, especially on "exotic" languages, in order to be able to observe a wider range of phenomena. Such access will be facilitated if the formalismsin which the grammaTs are prese'nted lend themselves to computer implementation for purpose$ such as parsing, testing, etc. This entails, among other things,that linguists should avoid as much as possible features which make their grammars generate non-recursive sets, and hence it is one of the purposes of the present paper to point out such features and discuss possible ways of avoiding them.In this paper we will discuss one model proposed for the stratificational theory of linguistics. This theory, advanced by S. Lamb, H . A . Gleason Jr. and their collaborators (C51,[61,C71), advocates that langdages be described in terms of several subsystems, known as strata. Each stratum has its own set of units and a tactics specifying the tlcorrectfl ("all~wable'~) structures that stratum. specific grammar might for example have strata corresponding roughly to semantics, syntax-morphology and phonology, although this is by no means standard. Furthermore, the strata are linearly ordered as levels, and there is a realization relation which connects adjacent strata by attaching to every well-formed structure on one stratum, zero or more accompanying structures on the adjacent strata. Note therefore that a particular utterance has simultaneous expression on each stratum.In this paper we examine the formal model for stratificational linguistics proposed by Sampson ( L 1 3 1 ) . This model uses rewrite grammars G1,G2, ... to describe the tactics, while the realization relation is essentially a rewrite system R acting as a transducer between the languages of-the tactics. More specifically, realization connects adjacent tactics G and G j + l j by matching sentences u in the language generated by G with those sentences j v in the language of G j + l which can be derived from u by using rules from R. An important property of the linguistic realization relation is the fact that' every structure on some stratum can have only a finite number of llrealizates" on the next stratum.This means that the rewrite system R must be constrained so that it has no recursive symbols. Such a rewrite system will be called acyclic.We investigate here the effect of acyclic rewrite systems acting as transaucers on axiom sets, varying the type of the derivations and rules allowed.We prove in this paper that regular languages are closed under transduction by acyclic rewrite systems, but that the linear context-free languages are mapped onto the recursively enumerable sets. This implies that stratificational grammars with non-selfembedding ta~ctics would be too weak while those with even one context-free tactics would be too strong. If the realization derivation is restricted to be in some sense "leftmost", then we show that the transduction can be performed by a finite,state device known as an a transducer. Furthermore, if productions with null right-hand sides are not allowed in an acyclic rewrite system then all the derivations can be made leftmost. This provides one possible method of restricting the generative power of acyclic rewrite systems.By deriving a recursive characterization of the languages generated with n-strata in terms of (n-1)-stratal languages, we can show that if the realization is restricted to being leftmost, then the languages described are homomorphic images of the intersections of the languages generated by the tactics. In particular, this means that we can find natural families of stratificational grammars which generate far example the sets recognized in real time by nondeterministic multitape Turing machines. This result partially confirms a hitherto unproven claim by Sampson, and discredits Postal's Clll classiciation of stratlficational grammars as just another variant of context-free phrase-structure grammars.Finally, we investigate the use of ordered rules in linguistic grammars and prove that in several models they allow the generation of sets which are not even recursively enumerable a clearly unsatisfactory situation.The remainder of the paper is structured as follows,In Section 2, we present the formal definitions and notation to be used, including the formal model for stratificational grammars.In Section 3, we examine the properties of "acyclic rewrite systemsw, which form the principal novel component in our definition of stratificational grammars. We then return in Section 4 to examine the generative power of st'ratificational grammars and relate the results to linguistics. e -o u t p u t f r e e i f f o r an$t ( r , u , v , s ) i n T , t h e s t r i n g v cannot be n u l l .
. generative power of a c y c l i c r e w r i t e systems:
To b e g i n w i t h , we remark t h a t i n G t h e n A => dD i n Gi+l i n c a s e bc -+ d is i n R ( s t e p 5 ) . where we used r u l e Xw A'Yz i n s t c p 0. and l a n g u a g e s L1,. . . , L s u c h t h a t f o r i = 1,. . . ,n L(G.) t h e r e e x i s t s a -t r a n s d u c e r O n v l s u c h t h a t L-RSTRAT(RST) i s e q u a l t o *+ v . €VG $Rev ( v ) = v i -1 f o r i = 1,. . . , m -1 3 1 m + i -1 and --a' --IUN N V + L 2 = { % w l a . . .w % $ % w~+~ %...p ln>O, w . a VOn+ 1 (L-RSTRAT(TOP(RST)) n L ( G ) ) n V , . n+ 1 ( 4 )1 Meaning t y p e 3 , t y p e 2 , t y p e 1, t y p e 0 , l i n e a r l a n g u a g e ..,. , L t t such t h a t L-RSTRAT(TOP(RST)) = hTt(L!i n . . . n -Lt\3 . n n S u b s t i t u t i n g t h i s i n (4) and appllying n o t e s ( a ) , (b) and (c) we the realization derivation must be k-leftmost and a homomorphism must be applied to the intersection of the languages. C81 P e t e r s , P .S. and &W. ~i w h i e ( 1 9 7 1 ) . "On ~e s t r i c t i n g the base component of T r a n s~o r m a t i d n a l grammar^'^. Information and "Control -18. -C91 P e t e r s , P.S. and R.W. Ritchiea ('1973) . "On the g e n e r a t i v e power of Transformational ~r a r n m a r~" 1.nformati06 Sciepces 6 49-83. E 1 2 1 Salomaa , A . (1973) . Formal Languages, Academic Press,NewYork.[ 1 3 J Sampson, G . ( 1 9 7 0 ) . S t r a t i f i c a t i o n a l grammar: 9 D e f a n i -t i o n and a n Example, Janua Linguarum, S e r i e s Minor: 8 8 , The Hague Mouton.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0 | null | null | null | null | null | null | null | null |
06fd4efc72d6c80334a4c9e84c5cbb9cba45a1c8 | 219301102 | null | Review: \textit{ {A}bhangigkeitsgrammatik}, by {J}urgen {K}unze | JUrgen Kunze establishes his dependency grammar with four components. The syntactic is the most important. The three non-syntactic components are the paradigmatic component, theselectional component, and the assigning component. In the first chapter of his book ABHANGIGKEITSGRAMMATIK (Dependency Grammar) the reader g e t s introduced to some of the basic concepts useful in understanding the notions explicated later sn. Subordination or dependency is introduced by way of a diagram, known as a tree, consisting of several connected points. A point or node that is connected to one closer to the top of the page is subordinate to it. This is called direct dependency. Indirect subordinat'ion is when two nodes are connected with one or more points in between them. These three nodes comprise a part tree. Kunz review 48 | {
"name": [
"Ballin, Kenneth F."
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 0 | 0 | null | null | null | null | Obviously there are several part trees which combine to make a tree.If the bottom-most node of our lfttle part tree is not superordinate to any other point then the part tree is an end complex. Every node is an en& complex with itself as its only member.Once one decides to attach words to these nodes it changes from a conne-the dots game to some sort of meaningful diaexam T&e first step in this change is to bring order to the diagram. Since language is the object of study here and the language the book was written in procedes from left to right, the author has ordered his tree from left to right. This type of tree is known as a W tree, i.e. Ghere each node is attached to a word. The book deals with M trees. These are trees in which the nodes are connected to signal combinations (Merkmalkombinationen). A marked tree is one in which a11 the connections are subordinate relations on one kind or another from a set containing all the kinds of subordinate relations possible.In making his investigatiods, Kunze has limited his field of study to modern day written German. This suffices as for in any pure theoretical investigation it is acceptable to assume the observed language is a set of given sentences. The practicability of his theory depends on finding a standard of correctness. In this case tapping the knowledge of a native speaker is of no help ventory and the other four don' t .There are demands made on a system of subordinate relations.The first of these is that the marked tree should be an adequate reeresentation of the syntactic structure of the sentence. Secondly, the subordination relations must allow all categories, qualities, and relations in the base structure To represent and differentiate the paradigmatic and selectional relations that can't be expressed through assigning.Affectation ways (Wirknngswege) are dashed lines connecting two nodes dominat+ed by a third node (see diagram). They represent other relations that exist between nodes aside from subordination. That these affection ways of both the paradigmatic and selectional relations must be represented through subordination relations is another demand made on the system. The last demand made is that the conditions for the paradigmatic and selective points (Vorgaben) must also be represented.The principle called the differentiation principle proves these last two are met. The system makes this determination by using a a knowledge of dependency trees, a fixed inventory of paradigmatic and selectional relations, and a fixed language base in a way which yields the required relations.The last concept developed by the author is that of bundles. Chapter 8 is a discussion of some questions that were brought out as a result of this theory. | null | Main paper:
:
Obviously there are several part trees which combine to make a tree.If the bottom-most node of our lfttle part tree is not superordinate to any other point then the part tree is an end complex. Every node is an en& complex with itself as its only member.Once one decides to attach words to these nodes it changes from a conne-the dots game to some sort of meaningful diaexam T&e first step in this change is to bring order to the diagram. Since language is the object of study here and the language the book was written in procedes from left to right, the author has ordered his tree from left to right. This type of tree is known as a W tree, i.e. Ghere each node is attached to a word. The book deals with M trees. These are trees in which the nodes are connected to signal combinations (Merkmalkombinationen). A marked tree is one in which a11 the connections are subordinate relations on one kind or another from a set containing all the kinds of subordinate relations possible.In making his investigatiods, Kunze has limited his field of study to modern day written German. This suffices as for in any pure theoretical investigation it is acceptable to assume the observed language is a set of given sentences. The practicability of his theory depends on finding a standard of correctness. In this case tapping the knowledge of a native speaker is of no help ventory and the other four don' t .There are demands made on a system of subordinate relations.The first of these is that the marked tree should be an adequate reeresentation of the syntactic structure of the sentence. Secondly, the subordination relations must allow all categories, qualities, and relations in the base structure To represent and differentiate the paradigmatic and selectional relations that can't be expressed through assigning.Affectation ways (Wirknngswege) are dashed lines connecting two nodes dominat+ed by a third node (see diagram). They represent other relations that exist between nodes aside from subordination. That these affection ways of both the paradigmatic and selectional relations must be represented through subordination relations is another demand made on the system. The last demand made is that the conditions for the paradigmatic and selective points (Vorgaben) must also be represented.The principle called the differentiation principle proves these last two are met. The system makes this determination by using a a knowledge of dependency trees, a fixed inventory of paradigmatic and selectional relations, and a fixed language base in a way which yields the required relations.The last concept developed by the author is that of bundles. Chapter 8 is a discussion of some questions that were brought out as a result of this theory.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0 | null | null | null | null | null | null | null | null |
fc888e2a687d159f030c697fc579d46d20c6aff5 | 204748627 | null | Subsequent Reference: Syntactic and Rhetorical Considerations | 01ce an oblect is infrsduced 'into r discourse, the form of subsequent references to it are slrongly governed by converrt,i?n Tliis paper discusses how Chose conventions can be represerrled for use by a generation facility. A multistage | {
"name": [
"McDonald, David D."
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 0 | 0 | null | null | null | representation is used, allowing decisions to be made when and w l~e r e the inforn~ation is available. I t is suggested that a specification of rhetorical structure of the inteded message skould he ir~cluded with the present syntactic one, and the conventions eventually reformulated in terms of it.Whenever a speaker wants to refer in text or speech to some object, action, state, etc., slie must find phrase which will both p r~v i d e an adequate description and fit the context. What governs her choice? One way to f~n d out might be to look at the selected phrase after the fact and t f y to develop a static characterizatton of lhe relation between it and its context. This is what most non-computationdl linguisls do.However, relations derived from fin~shed texts are al best incomplete. They will not tell us how the choice was made or even guarentee that the relation(s) was apparent when the c l i o~r e had to be made. The ne3t secl~on looks at the course of the whole generation process as my program models it, and fits the sub,-process o f findlng phrases fos references within it. Then the process of deciding whether or not to use a pronoud will be cxam~ned in some detail. This will lead to the problem of i r c c~~~i t i~ audience moclels and, the idea that tlie relevant i~rfortnatinn r.lioulcl be computed outside the linguistic co~v.trriction pi oress per se. That ~dea is expanded to ~nclude "tLL3~tor1cal slruc ti~res'* like tlie relation "all of a set" that leads t o a plirases i i~e "...a square, ... the other square". Finally, a clesigt~ for Illis rhetorical slruc ture is sketched.St~~,por;e we had a logically n~inded program that wanted to 11iake llie stalc~irent: We usually lli~ink of objectsnoun phrasesas being the oldy I l i l~i p ,~ that nliglrt be refered to more than once, but that 15 not tlicr taw?. Cons~der llre formula mortal(Rome0) A moi tal(Jul1iet). 1 fiat could be rendered In any of several ways ~~~c l~t c i t n~. : . , :Vx man(x) mortal(x)"Romro IS mot f a1 and so 1s Juliet". Here the second tis stance o f mortal() was realtzed by a spec~al, highly restr~cted ~r a~litnat~c clcv~ccexac tly t hc characteristics of a "subsequent refcrwice". From llre p o~n t of view of the language generation cor~rpnncnt, the ~mportanl tlr~ng will be the repetition of some nattrr from tlic rnput formula not, at f~r s t glance at least, the kind o f ol>ject I l l a t name deraotes. (The set of descripttve tornrt~lac. supplicd to the linp,uistics component is called the proi:ra~~i s "tn~r.~ap,e" Sul~formulas or terms wlthtn a message are c ;tllcd "elc~~tcnts" o r "msg-elmls".)1 1 1 r linlrcnal, 0hjects t l i a l appear in a speaker's der.criptions will have defln~nz snd incidental properties assoc~atecl with Ilicm-wh~ch aye accesstble through their names. Upgrading the predicate calculus enough to motivate the use 01 fluent English is a facinating prob\em, but one which 1 will gloss over in l h~s pnper. See Mctknald [1978a] for more detrils. For now, I will nssume that the decisions made by the various entries come out sa as to give the literal version the formulr with the e x p l~l t references just so that we can use it for am exatnple.E e l~w is my progrsm'r representation ~f the situalion just ss it i~ rrbout to choose a phrase for the third instance of x in the formula The polnt of showing this constituent structure k to deinonstrale that w h i the progrem has a greet deal of datr. to bring to bear on the choi ce, it also has &great deal ol data which is utterly irrelevant to ~t . The packaging of the d d rthe size of the search spjceis at least as important as hrnrlng the data ava~lnble in the first place.[clause]1-coord "~f" any thng [subj] [pred] I--.-r- . . -. I [clause] 1-coord 'then" cIaJml=g / I [defxhead] [al[pred-nom] lhat lkng kA k * t l l k A l a manIn the diagram, the names of grammatical categoriec: t\auselr pp, etc , derlote the syntactic nodes of an annotated surface structure. Each node has a set of ' Immediate consliluenls, dgntwzed by a list of named constituent slots. A slot can be empty, hold another node, nold a word or idiom, or hold an element of the input formula which hss yet to be processed, e. 8. rs, or mortd0. T )re words at the l a m s of the tree are given in their root farm. A morphology subroutine specializes lhern for number, lense, etc, when they rro spoken (pr inled on the console). The choice of whet syntactic categories, descriptiva features and conslituen\ slots to maintain is tied up with the cholce of actions rssoctftted with them by the hngusticr compomnt. Tlre [intro] constituent, for example, will act to insurc lhat any introductory clause is realized a a participlb Tl~ere are many trade-offs involved in the design of this grammar, and I will again gloss over them for this paper.The choice of refering phrase for a subsequent reference is delermlned largely by the syntactic relationship between Ihcr current instance nnd the previous instance to the same ebject. In s static, after 'the fed maljrsis, we would detcr)mine this relationship by examining their positiofi in a tree 'fib the orre above This is a simple enough operation for a person usiiig her eyes, but it 1s an awkward mark and sweep style search for I computer program.My program uses a much more efficient, and I would say more porspicbous approach based on recording potentially relevant facts st the time they are first noticed by the I~irguistics component The wwd~ng of the heuristics that are used for 4 k detisions are similar to the wordings used h static anrlys~s. (They almost have to be, given that that is how thd biilk of Iinguishc research has been done to date.) But fhe d+to I' w the heurist~cs is Acquired n a more natural mm_ner.Before discussing the program actual pronominal/zatibn heuristics, 1 w~l l first dieress !a describe the workings ol the generation proceso which collecl~ ( a d creates) tddata. All of Ihe generation componeht's actual knowkdgq is spread about many small, local routines: dictionary entries fDr the abject that will appear in input formulrs; "realization slratcg~cs"the corrstruction routines that those' entries execute to implemenl their decisions~ or brarnmar routinesw - (1) call tlae grammar routine f~r that slot name(2) call Examine-slot-contents contents = nil do nothing contelils 2 <word> call the morphology subroutine with the word p r~n t the result co~itetils = <n~r;g-elmt> use t l~e dic t~onary entry for the element t o find .a phrase for the element; replace the element with that phrase as the contents of the slot; loop tlrrougl~ Ihe cases again.So, having e~nerated 'clause2, in effect by starting the controller on I Ire last case of Examine-slot-contefls, the controllcr will loop around. The contents will now be clause2; the t l~i r d case will be taken and the clause "entered". Its first cons tiluent cbnlains anot her node; the controller recursively re-enters Exainine-node and enters the prepositional phrase.Its first conslituent contains the word " f~r " , which is immcdialedly pr~ntcd out with no changes from the morphology subroutine; the second'contains the f~rst instance of x which is processed wilh the d~ctionary entry common to "issolated variables". Tlrc noun phrase i t constructs replaces the x in the constrtuenf tree; the controller then loops thrqugh the cases The deslgn of the controller guarentees !hat tha generation process WIII nave ,these properties: (1) I t is done in one passthe controller never backs up. (2) Therefore ckcistons, choices of phras~ng, must be made correctly the first time. (3) i t is incremental. When the f~rst part of the text ik being printed out, later parts will be in their internal form. Candy arid Carol, then the oracle would return a nV(I list, and t l~e pronorn~nalizat~on option would go through. If they didn't know tliern, Illen it would return "( Carol I", ana a further round of l i e u r~s t~c s would be tried.To compare tlie relative "pronominalizabil~ty" of several ti~eqsao,e elements, Pronoun? runs them separately through the analyr.1~ atid cvaluation procedure. Buf instead of acting on the evaluatton directly, ~t makes a list of the names of the i n d~v~d u a l hcur~stics that each passes and then compares the t w o 11r.t~. In the current program these would be:Calidy same -sentence proceed-and-command Carol same-s~mplex ;via a trace proceed-and-coninland ppst airs-subjcct n o h t erveet~~niz-d~st rac t ion In this case, tlie relative number of heuristics alone would incl~rate tliat Carol wouid make a "better" interpretation for a pronoutl ill that position, and that, therefore, the possibility of a ~rsiny. s pronoun for Candy should be rejected. But actually, the cllffcrent lieuri~tics are given weightings. Stwe-simplex, for cxa~npie, is much belter evidence than same-serrtence. One of tlic nlore common reasons for rejectlng the use of a pronoun I : , that i t mrglit be missinterpreted as refering to some o t l l c~ bbjcct. The form of subsequent reference eventually cl7oose11 In t l~e s e cases must distinguish the object from the one r t is potentially antb~guous with, but does not have to recap~tt~lato any rnore dc'tail.In particular, one frequent pattern for an Initial reference IS a riocrn ptirar.c with lhc narne of a class of objeds -as its head word, w~t l i a series,,of adjectives, classifiers, or qualifying p l i~a~. c s surot~ncl~np, 11 There IS a simple formula for c o t x t~ ~r c trtly, a non-pronom~nal, subsequent reference to fotlow t h~s I..IIIC~ of NP namely to repeat the class name as the head wat ti atrcl irse cithcr "that" or "the1' as a determmer. In eacli'of these cases, fhe' two objects were both of the S~I~C " s~r t "~ eugcs, corners, brackets, or blocus. By the usual c r~t e r~a , t l i~s w o i~l d mean tliat b e y share dictionary entrres, A T I C~, inclbed, the palred phrases have much in common, and coul(l be seen i s only differing In the chorce of strategy for t lieir acljecllves and/or determiners. This means that the coord~nat~np; mark must be someth~ng other than the "kind-of' poinlcr tlial links objects H;ith !Jdr enhies. 'It will also probably have to be a t e r n~o r d r~ strhcture, since "Ihe opposile corner" is a transient phenomena, defined only at particular moments in ~a c h game dl 't'ic-tac-toe.The sltnplest way to mark the pairs is with an addifional formu(a in the Input message, e.8.(all-of -a-fie1 cornor 1 cornerg) or (codrast-by-size B6 83) W I i r~j the message is initially processed, formulas like these are indexed by their arguments so that, e.g., the dictionary entry for blocks will be able t o notice them and choose its strhtcgies accordingly. 'Pro~omina\ization ~f subsequent references to the focused object is almost always abligatory. (There can be exceptions i f tlie last sever.al references to the object were pronominalized, and the intontion is to "refresh" the audience s memory.) In the example wrtli "Candy" and Carol", if the previous part of the d~ssourse lid been saying 111i ngs about Cbndy, then she would have been established as, the focus of that sentence. The I-hetorrcai context could be very domain specific.The black queen can now take a p a~n . "Notice that i t i s nol necessary to say "a white pawnn because irn~nediate inference thet one makes about what pieces i t is legal for a piece of a given color to "take".S~nce +lie ctlterta for constructing a refet~ng expression for any chess plece will overlap, they will likely share a dictio~lary & d r y . Thus we have a sort of subsequent reference phenomena. The entry tor chess pjeces will be lookinp, f o r the menti011 of a piece s color earlier in the text. I f i t finds one, or rather if it finds one of the Complementary color, and i f the ~l t u a l i o n IS right, 11 can om~t any mention of color from tho phrase i t has aswmbled.How to deternine lliat the situation is "right" is a matter for tho rhctoriral contqxt to specify. The problem is the color of conIrasting piece can be amitled conly i f the choice of verb If there is, it looks to see if its piece is p y t of tlie relation and whether it is the second of the two Id be mchtionad. I f so, i t omits the color name.The power of this representational technique is that it ~o m p l l e s its record of the needed facts at the time when they easily determined. i.e. as the message is being compiled, well belore the reletccrn name has bren rendered into English and the simplicity of the relation obscured.This tec-hn~que should be applicable to many more phenomlicm~a than -simply subsequent reference. Cbnsider sentences like these: I f the source messages for those sentences described only {heir literal content, it would be rmpossible to mottvate the use of also, so, or but in those ways, yet they are what gtve the sentences t hcir naturalness. But i f those rhetorical relations are i n c l t~d~d as part of the linguistic context, with their links to specific phrates and cllctionary enlrres, including these "little" words becomes s~mple. | null | null | Main paper:
:
representation is used, allowing decisions to be made when and w l~e r e the inforn~ation is available. I t is suggested that a specification of rhetorical structure of the inteded message skould he ir~cluded with the present syntactic one, and the conventions eventually reformulated in terms of it.Whenever a speaker wants to refer in text or speech to some object, action, state, etc., slie must find phrase which will both p r~v i d e an adequate description and fit the context. What governs her choice? One way to f~n d out might be to look at the selected phrase after the fact and t f y to develop a static characterizatton of lhe relation between it and its context. This is what most non-computationdl linguisls do.However, relations derived from fin~shed texts are al best incomplete. They will not tell us how the choice was made or even guarentee that the relation(s) was apparent when the c l i o~r e had to be made. The ne3t secl~on looks at the course of the whole generation process as my program models it, and fits the sub,-process o f findlng phrases fos references within it. Then the process of deciding whether or not to use a pronoud will be cxam~ned in some detail. This will lead to the problem of i r c c~~~i t i~ audience moclels and, the idea that tlie relevant i~rfortnatinn r.lioulcl be computed outside the linguistic co~v.trriction pi oress per se. That ~dea is expanded to ~nclude "tLL3~tor1cal slruc ti~res'* like tlie relation "all of a set" that leads t o a plirases i i~e "...a square, ... the other square". Finally, a clesigt~ for Illis rhetorical slruc ture is sketched.St~~,por;e we had a logically n~inded program that wanted to 11iake llie stalc~irent: We usually lli~ink of objectsnoun phrasesas being the oldy I l i l~i p ,~ that nliglrt be refered to more than once, but that 15 not tlicr taw?. Cons~der llre formula mortal(Rome0) A moi tal(Jul1iet). 1 fiat could be rendered In any of several ways ~~~c l~t c i t n~. : . , :Vx man(x) mortal(x)"Romro IS mot f a1 and so 1s Juliet". Here the second tis stance o f mortal() was realtzed by a spec~al, highly restr~cted ~r a~litnat~c clcv~ccexac tly t hc characteristics of a "subsequent refcrwice". From llre p o~n t of view of the language generation cor~rpnncnt, the ~mportanl tlr~ng will be the repetition of some nattrr from tlic rnput formula not, at f~r s t glance at least, the kind o f ol>ject I l l a t name deraotes. (The set of descripttve tornrt~lac. supplicd to the linp,uistics component is called the proi:ra~~i s "tn~r.~ap,e" Sul~formulas or terms wlthtn a message are c ;tllcd "elc~~tcnts" o r "msg-elmls".)1 1 1 r linlrcnal, 0hjects t l i a l appear in a speaker's der.criptions will have defln~nz snd incidental properties assoc~atecl with Ilicm-wh~ch aye accesstble through their names. Upgrading the predicate calculus enough to motivate the use 01 fluent English is a facinating prob\em, but one which 1 will gloss over in l h~s pnper. See Mctknald [1978a] for more detrils. For now, I will nssume that the decisions made by the various entries come out sa as to give the literal version the formulr with the e x p l~l t references just so that we can use it for am exatnple.E e l~w is my progrsm'r representation ~f the situalion just ss it i~ rrbout to choose a phrase for the third instance of x in the formula The polnt of showing this constituent structure k to deinonstrale that w h i the progrem has a greet deal of datr. to bring to bear on the choi ce, it also has &great deal ol data which is utterly irrelevant to ~t . The packaging of the d d rthe size of the search spjceis at least as important as hrnrlng the data ava~lnble in the first place.[clause]1-coord "~f" any thng [subj] [pred] I--.-r- . . -. I [clause] 1-coord 'then" cIaJml=g / I [defxhead] [al[pred-nom] lhat lkng kA k * t l l k A l a manIn the diagram, the names of grammatical categoriec: t\auselr pp, etc , derlote the syntactic nodes of an annotated surface structure. Each node has a set of ' Immediate consliluenls, dgntwzed by a list of named constituent slots. A slot can be empty, hold another node, nold a word or idiom, or hold an element of the input formula which hss yet to be processed, e. 8. rs, or mortd0. T )re words at the l a m s of the tree are given in their root farm. A morphology subroutine specializes lhern for number, lense, etc, when they rro spoken (pr inled on the console). The choice of whet syntactic categories, descriptiva features and conslituen\ slots to maintain is tied up with the cholce of actions rssoctftted with them by the hngusticr compomnt. Tlre [intro] constituent, for example, will act to insurc lhat any introductory clause is realized a a participlb Tl~ere are many trade-offs involved in the design of this grammar, and I will again gloss over them for this paper.The choice of refering phrase for a subsequent reference is delermlned largely by the syntactic relationship between Ihcr current instance nnd the previous instance to the same ebject. In s static, after 'the fed maljrsis, we would detcr)mine this relationship by examining their positiofi in a tree 'fib the orre above This is a simple enough operation for a person usiiig her eyes, but it 1s an awkward mark and sweep style search for I computer program.My program uses a much more efficient, and I would say more porspicbous approach based on recording potentially relevant facts st the time they are first noticed by the I~irguistics component The wwd~ng of the heuristics that are used for 4 k detisions are similar to the wordings used h static anrlys~s. (They almost have to be, given that that is how thd biilk of Iinguishc research has been done to date.) But fhe d+to I' w the heurist~cs is Acquired n a more natural mm_ner.Before discussing the program actual pronominal/zatibn heuristics, 1 w~l l first dieress !a describe the workings ol the generation proceso which collecl~ ( a d creates) tddata. All of Ihe generation componeht's actual knowkdgq is spread about many small, local routines: dictionary entries fDr the abject that will appear in input formulrs; "realization slratcg~cs"the corrstruction routines that those' entries execute to implemenl their decisions~ or brarnmar routinesw - (1) call tlae grammar routine f~r that slot name(2) call Examine-slot-contents contents = nil do nothing contelils 2 <word> call the morphology subroutine with the word p r~n t the result co~itetils = <n~r;g-elmt> use t l~e dic t~onary entry for the element t o find .a phrase for the element; replace the element with that phrase as the contents of the slot; loop tlrrougl~ Ihe cases again.So, having e~nerated 'clause2, in effect by starting the controller on I Ire last case of Examine-slot-contefls, the controllcr will loop around. The contents will now be clause2; the t l~i r d case will be taken and the clause "entered". Its first cons tiluent cbnlains anot her node; the controller recursively re-enters Exainine-node and enters the prepositional phrase.Its first conslituent contains the word " f~r " , which is immcdialedly pr~ntcd out with no changes from the morphology subroutine; the second'contains the f~rst instance of x which is processed wilh the d~ctionary entry common to "issolated variables". Tlrc noun phrase i t constructs replaces the x in the constrtuenf tree; the controller then loops thrqugh the cases The deslgn of the controller guarentees !hat tha generation process WIII nave ,these properties: (1) I t is done in one passthe controller never backs up. (2) Therefore ckcistons, choices of phras~ng, must be made correctly the first time. (3) i t is incremental. When the f~rst part of the text ik being printed out, later parts will be in their internal form. Candy arid Carol, then the oracle would return a nV(I list, and t l~e pronorn~nalizat~on option would go through. If they didn't know tliern, Illen it would return "( Carol I", ana a further round of l i e u r~s t~c s would be tried.To compare tlie relative "pronominalizabil~ty" of several ti~eqsao,e elements, Pronoun? runs them separately through the analyr.1~ atid cvaluation procedure. Buf instead of acting on the evaluatton directly, ~t makes a list of the names of the i n d~v~d u a l hcur~stics that each passes and then compares the t w o 11r.t~. In the current program these would be:Calidy same -sentence proceed-and-command Carol same-s~mplex ;via a trace proceed-and-coninland ppst airs-subjcct n o h t erveet~~niz-d~st rac t ion In this case, tlie relative number of heuristics alone would incl~rate tliat Carol wouid make a "better" interpretation for a pronoutl ill that position, and that, therefore, the possibility of a ~rsiny. s pronoun for Candy should be rejected. But actually, the cllffcrent lieuri~tics are given weightings. Stwe-simplex, for cxa~npie, is much belter evidence than same-serrtence. One of tlic nlore common reasons for rejectlng the use of a pronoun I : , that i t mrglit be missinterpreted as refering to some o t l l c~ bbjcct. The form of subsequent reference eventually cl7oose11 In t l~e s e cases must distinguish the object from the one r t is potentially antb~guous with, but does not have to recap~tt~lato any rnore dc'tail.In particular, one frequent pattern for an Initial reference IS a riocrn ptirar.c with lhc narne of a class of objeds -as its head word, w~t l i a series,,of adjectives, classifiers, or qualifying p l i~a~. c s surot~ncl~np, 11 There IS a simple formula for c o t x t~ ~r c trtly, a non-pronom~nal, subsequent reference to fotlow t h~s I..IIIC~ of NP namely to repeat the class name as the head wat ti atrcl irse cithcr "that" or "the1' as a determmer. In eacli'of these cases, fhe' two objects were both of the S~I~C " s~r t "~ eugcs, corners, brackets, or blocus. By the usual c r~t e r~a , t l i~s w o i~l d mean tliat b e y share dictionary entrres, A T I C~, inclbed, the palred phrases have much in common, and coul(l be seen i s only differing In the chorce of strategy for t lieir acljecllves and/or determiners. This means that the coord~nat~np; mark must be someth~ng other than the "kind-of' poinlcr tlial links objects H;ith !Jdr enhies. 'It will also probably have to be a t e r n~o r d r~ strhcture, since "Ihe opposile corner" is a transient phenomena, defined only at particular moments in ~a c h game dl 't'ic-tac-toe.The sltnplest way to mark the pairs is with an addifional formu(a in the Input message, e.8.(all-of -a-fie1 cornor 1 cornerg) or (codrast-by-size B6 83) W I i r~j the message is initially processed, formulas like these are indexed by their arguments so that, e.g., the dictionary entry for blocks will be able t o notice them and choose its strhtcgies accordingly. 'Pro~omina\ization ~f subsequent references to the focused object is almost always abligatory. (There can be exceptions i f tlie last sever.al references to the object were pronominalized, and the intontion is to "refresh" the audience s memory.) In the example wrtli "Candy" and Carol", if the previous part of the d~ssourse lid been saying 111i ngs about Cbndy, then she would have been established as, the focus of that sentence. The I-hetorrcai context could be very domain specific.The black queen can now take a p a~n . "Notice that i t i s nol necessary to say "a white pawnn because irn~nediate inference thet one makes about what pieces i t is legal for a piece of a given color to "take".S~nce +lie ctlterta for constructing a refet~ng expression for any chess plece will overlap, they will likely share a dictio~lary & d r y . Thus we have a sort of subsequent reference phenomena. The entry tor chess pjeces will be lookinp, f o r the menti011 of a piece s color earlier in the text. I f i t finds one, or rather if it finds one of the Complementary color, and i f the ~l t u a l i o n IS right, 11 can om~t any mention of color from tho phrase i t has aswmbled.How to deternine lliat the situation is "right" is a matter for tho rhctoriral contqxt to specify. The problem is the color of conIrasting piece can be amitled conly i f the choice of verb If there is, it looks to see if its piece is p y t of tlie relation and whether it is the second of the two Id be mchtionad. I f so, i t omits the color name.The power of this representational technique is that it ~o m p l l e s its record of the needed facts at the time when they easily determined. i.e. as the message is being compiled, well belore the reletccrn name has bren rendered into English and the simplicity of the relation obscured.This tec-hn~que should be applicable to many more phenomlicm~a than -simply subsequent reference. Cbnsider sentences like these: I f the source messages for those sentences described only {heir literal content, it would be rmpossible to mottvate the use of also, so, or but in those ways, yet they are what gtve the sentences t hcir naturalness. But i f those rhetorical relations are i n c l t~d~d as part of the linguistic context, with their links to specific phrates and cllctionary enlrres, including these "little" words becomes s~mple.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0 | null | null | null | null | null | null | null | null |
09ddd7010b4225f9afa35d785d5bad0b99f0e6a4 | 219310292 | null | A Computational Account of Some Constraints on Language | In a series of papers over the last several years, Noam Chomsky has argued for several specific properties of lbl~grloge wlilci~ ha claims are universal to all human langi~ages [Cliomsky 73, 75, 76). These properties, wlrich | {
"name": [
"Marcus, Mitchell"
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | null | null | null | null | null | null | forni one of the cornerstones of his current linguistic theory, are eml~oclied in a set of constralnts on language, a set of restrictions on the operation of rules of grammar.Tlils paper wlli outline two argtiments presented a t length i r i [Marcus 771 clemonstrating thctt important subcases of two of tliese constraints, tlie Subjacency Principle oncl tlie Specified Subject Constraint, fall out naturally from the structure of a gramnlar interpreter called PARSIFAL, wliose structure is in tiirn based upon the hypothesis tliat a natural language parser needn't simulate a liondeterministic macliiiio. This "Deterniinism Hypothesistt claims that natural laiigiiaqe can be parsed by a computationally simple nieclion~sm tliat uses neither backtre~king nor pseudopsrailclism, alid in wliicii all grammatical structure c m t e d by the parsor is 'ti~iclelible'v in that it must all be output as part of tlie structural alialysis of the parser's input. Once 15uilt, no grammatical structure can be d~scarded or altered in tlio course of tlie parsing process.In l)prticular, this paper will show that the structurq of the grammar interpreter constrains Its operation in silch a way tliat, by and large, grammar rules cannot parse sentences whlch violate either tlie Specified Subject Cgnstraint or tlre ~tihjacency Principle, The component of the grarnniar interpreter upon wh~clt this. result princlpaily clcpcncls is motivated by the Determinism Hypothesis; this result thus provides inclirect evidence for tlie hypothesis.This result also clepends upbn the use Within . a coni~~trtatlonal framework of tlre dlosely related notions of annotated surface structure and trace theory, which also derive from Cliomskygs recent work? (It should be noted that these constraints are far from universally accepted. They are currently the source of much contraversy; for various critiques of Chomsky's positio~l see [Postal 74; Breslnan 76) . However, what is presented below does not argue for these constralnts, per sa, bill ratiier provides a different sort of explanation, based on a processing model, of why the sorts of sentences which these constraints forbid are. bad. While the exact formulation of these constraints is controverstal, the f a c t that some set of constraints is needed to account for this range of data Is i~enerally agreed upon by most generative graniniclrians. Tlie account which I will present below i s crt~cially Iinkecl l o Clionisky's, however, in that trace theory is at tlie haart of this account.)Because of space lirfiitations, this paper deals only with those gr~nrmatical processes characterized by the con~petence rille ''MOV~ NPg'; tlie constraints imposed by the grammar interpreter upon those processes cliaracteri7ecl by tlie rtrlc "MOVE WH-phrasew are discussed at I~n q t l i in [Marcus 771 wliere I show tlmt tile behavior ctiaracterlzecl by ,Ross' s Complex NP Constraint [Ross 671 ~tsetf follows clirectly from tlie str~rcture of the grammar interpreter for rather cliffere~it reasons than the bhavior considereti In this section. Plqo h c a u s e of space Iiiiiil~itio~is, I will not attempt to show that the two colistralnts I w~l l deal with here necessarily follow from the grtlmmirr interpreter, but rather only that they naturally follow from tlie intarpretar, in particular from a slmple, natural formihatlon of a rule for passlvization, which itself deponcls lieav~ly upon the structure 6f the interpreter. Again, iiecessity is argued for in detail in [Marcus 771 .Tliis paper will first ouNine tlie structure of the graninlilr interpreter, then present the PASSIVE rule, an8tlinn filially show liow Cliomskyts constraints ''fall outM of tlie formulation of PASSIVE.I3efore proceedilig with the body of thls paper, two other kaportnnt properties of the parser should be nie~~t~oriecl wli~ch will not be discirssed here. Both are cliscussecl at lengtli in [Marcus 771 ; the first is sketched as well in [Marcus 78) .1) Simple cuies of grammar can by written for this irlterpreter which elegantly capture the significant gaiieralizatior~s behincl not only passivization, but also such Coiistrtictiolis as yes/no questions, imperatives, and selitetices with existential there. These rules are remiillscent of tlie sorts of rules proposed wjtiiin tlie frnniework of t l~e theory of yeneratlve grammar, despite tlie fact that the rilles presented here must recover underlying structure given only the terminal string of the surface form of the sentence.2)Tlie. grammar interpreter provldes a sirhple exl>lanatkn for tlie difficulty caused by "garclen pathtt sentences, such as V i l e cotton clothing is made of grows in M~ssisslppi.~ Rilles can be written for this interpreter to resolve local strirdtural ambiguities which might seem to require notlcleterminlstic parsing; the power of such rules, howaver, rlspe19ds lirpon a parameter of the mechanism. Most structural aniblgirlties can be resoivecl, given an appropriate setting of this parameter, but those "whlch typically cause garden paths cannot.PARSIFAL mai~itaills two major data structures: a p\isliclown stack of l~icomplete constituents called the active node stack, nricl a sniall tl~r'ee-place constituent buffer whlch contains constituonts whlch are complete, but whose higher level yramniatlcal functlon is as yet uncertain. Figure 1 below stlows a snapshot of the parsergs clatn structures taken wlllle parsing the sentence "John siioi~ld liave schecli~lecl tlie meeting.", N d e tliat the active nocle stnck in sliown growing downward, so that thg strirctirre of the stack reflects the structule of the enrerging parse tree. A t tho bottom of the stack is an ai~xiliary ~iorle Cabelled wit11 the! features modal, p~s t , etc., w l i i c l~ lias as a daughter tlie modal 1tsliouic118. Above the bottoni of tlie stack is an S notla wlth an NP as a daughter. rlonri~iatinc~ the word "Johnt1. There are two words In tile birffer, tiie verb "hovew in the first buffer cell and the word "schedu1oc.l" irr the second: The two word8 "the meetingt1 have notA yetacome to tiie attention of the parser, (The structures of form "(PARSE-AUX CPOQL)' and the like will be explained below.)The Active Nocle Stack - Sl (S DECL MAJOR S) / (PARSE-AUX CPOOL) NP : (John) AUXl (MODAL PAST VSPL AUX) / (BUILD-AUx) MODAL i (should) Tile Buffer 1 : WORD3 ("HAVE VERB TNSLESS AUXOERB PRES V43S) : (iiave) 2 : WORD4 ("SCHEDULE COMP-OBJ VERB tNF-OBJ V-3S ED=EN EN PART PAST ED) : (scheduled)Yet unseen words: the meeting .Figi~re 1 -P'.4RSIFALts two major data structures.The constituent buffer is tiie heart of the grammar interpreter; it is tiie central feature that distinguishes tliis parser froni all others. The words that make up the parser's input first conie to its attonti011 when they appear a t the end of tliis buffer after morpno~ogical analysis. Triggered by tlie worcls at the beginning of the buffer, the parser may clociclo t o create a new grammatical constituent, create a new node at tile bottom of the active rrode stack, and then l~egiri to attach tlie cor~stituents in the buffer to it. After tliis new constituent is conipleted, the parser will then pop tile ~i c w constituent from the active node stack; If the grcllnill~tlticai role of this larger structure is as yet i~nclcterminorl, the parser will insert it into tlie first cell of the I~irffer. The parser is free to examine the constituents its tJic buffcr, to act irpon them, and to otiiewise use the I~u t fer as a workspace.While tlie buffer allows tile parser to examlne some 01 t h e context surrounding .a given constituent, it does not allow arbitrary look-ahead. The length of the buffer Is stl.ictly iimiferl; in tlie version of the parser presented here, the b~l f f e r tins only three cells. (The buffer must be cxtcnclccl to five colis to allow the parser to build NPs in a manner wliich is transparent t o the taclause levelta grammar rilles wllkh will be presented In this paper. This extended parsor still Iias a winclow of only three cells, but the eflcctive start of the buffer can be changed through an "at tenlion slilf ting n~echanisrn~~ whenever tlre parser Is ISuilding an NP. in effect, tills extended parser has two " l~(~i c a i~~.btrf fers of "length three, one for NPs and another for clauses, with these two buffers implemented by allowing an overlap in one larger buffer. For details, see [Marcus 771 .1Note that eacli of tiie tilree cells in the buffer can Iiolcl a grammetical constltuent of any type, where a constlt~reril Is any tree that tlie parser has constructed u~~c l o r a singlo root nocle. The size of the structure ~~ndcr~iocrtlr the tiotle is Immaterial; both ltthatl1 and "that tlie 11i g green cookie monstsras toe got stubbedea are perfectly good col.rstituents once the parser has constri~cterl a srrbordinate clause from tile latter phrase.Tlie constituent buffer and 'the active-node stack are acted ilpon by a grammar wliich is made up of ~mttcrn/action rules; tlris grammar can be viewed as an au$jrnentecl form of Newell and Simont$ production systems [Newell & Simori 721. Each rirle is made up of a pattern, wliich is niatciieti against some subset of the constituents of tlie buffer and the accessible nocles in the active node stack (al~out wliich more will be said belowh and an action, a sequence of operations which acts on these constituents. Etlcii rule is assigned a numerical priority, which the graninlar interpreter uses to arbitrate simulteneous matches.The yrnmmar ns a wiiole is structured into rule packets, clumps of grammar rilles which can be, activated ant1 clcactivatecl as a group1 the grammar interpreter only attenipts to match rules in packets that have been activated by the grammar. Any grammar rule can activate a oachnt by associating that packel with tliq constituent a t tho i~o l tom of tile active nocle stack. As long as that node is at tlie 1)ot'Coni of the stack, the packets associated with it arc! nctive; wlxen tliat node is pushed into the stack, the ~~a c k e t s remain associated with it, but become active again tmly wlieri that node reaches the bottom o l the stack. For sxnmple, in figure 1 above, the packet BUILD-AUX is nssociated with the bottom of the stack, end is thus active, while tlie packet PARSE-AUX is associated with the S node aljove tlie auxiliary. 'u*<worrl>u mean "has tlie root <word>", e.g. "Rhave" means principles. attd their embodiment in PARSIFAL, are as follows:"has -the root tthave"lt. The tokens 181st11, l12ndI1, "3rd8I and 8 t~" (or "c") refer tw the constltuents in the Ist, 2nd, and 1 ) A deterministic parser must be at least partially data The parser (i.e. the grammar interpreter interpreting some grammar) operates by attachiflg constitt~ents w l l i~h are in the buffer to the constituent at tlta l~ottom of tlte sfask: functionally, a,constituent IS in the stack when the parser is attempting to find its daughters, aitd in the buffer when the p a w r is attempting, to find its n~otlier. Once a constituent in She buffer has beeh atlacliecl, tile grpmmar interpreter will a&tomatically remove it from the buffer, filling in tlie gap by shiftihg to the left the constitumts formerly to its right. When the parsei has completed the mnstituent at the bottom of. the stack, it po(,s jliat constityent from tlte adive node stack; the corlstitilent either remeins attached to its parent, if It 'was attoclied to some larger constituent when it was created, or else .it falls into the first cell of the constitu8ntd buffer, 2) A determinislic parser must be able to reflect expectations that follow from the partial structures i~ctijt ctp during the parsing process. 'Packets of ritlcs can be a-ctivated and deactivated by grammar rilles to reflect tlie properties. of the constitue~its .in tlie active node stack.3rd b u f3) A deterministic parser must have some sort of constrained look-ahead facility. PARSIFAL1s buffer pro~ides.tltis constrained look-ahead. Because the bi~ffer can l~oicl several constituents, a grammar :rule can exaniitte tlie cont&xt that follows the first constituent in the buffer before deciding what granrmirtical role it fills in a higher level structure.The key idea is that the site of .the buffer can be sltargly constrained if each location In the buffer cat1 hold .a single complete constituent, regardless of Ilia t constituentls size. It must be stressed that this look-ahead ability must be constrained in tome manner, as i t is here by limiting the length of the-. buffer: otherwise [he "deterrnini~m~~ claim i s vacuous. tlte rrsage o f tlie term here ,can be thought of as a synthesis 01 tlte two cmcepts. Following winograd, this tern? will be used to refer to a notion of surface structure annotated by the addition o l a set of-featyres to each node in a parse tree. Following Chomsky, the term will be used t o refer to a tiotion of surface Structure annotated by the aclclition of an element called trace to indicate the "underlyittg position" of "sltiftecl" NPs-.In current linguistic theory, a trace Js essentially a ~~~~ltonoio<clically null" NP ilt t siltface . structure represetitation of a sentehce that 'has no daughters but is "botylclu to Ilte ' NP that filled that position at some level of i r n c l o r l y~ structure. In a sense, a trace can be viewed as a "dr~mmy" W that serves as a placeholder for. the ,NP t h a t 2ariier filled tll'6t~.position; in the same' sense, the trace's' bintlinfl cen ha viowed as simply a pointer to that NP. It sl~oi~icl be strcssscl at tlis outset, liowever, thbt a tpace is inclistingirisha~~ie froar a nornral NP in terms of normal grnnrtrrotlcal processes; a trace is an NP, even though It is an NF' that clominates no lexical material.There are sovcral reasons for choosing a properly annototoci surface strircturo as a primary output rsprclsantntion for syntactic analysis. While a deeper a~~a l y s i s Otle 11se of trace is to indicate the untleriying position of tlie wh-heacl of a question or relative clause. Tl~us, tho structurr! built by the parser for 3,la would iticlucle tlie trace sl~own in 3.lb, witlFtlle trace's binding shown by the line utlcter the sentence. The position of the trace inclicates tliat 3.1 a has an underlying structure alialogous to tlio overt surface structure of 3 . 1~.Aliother use of trace is to indicate tlie underlying j-lositiol~ of tlio surfnce sttbject of a passivized clause. For cxeml>lo, 3.2a will be parsec1 into a structure tliat includes a trnce as shown as 3.2b: this trace indicates tliat the s\t\,ject of the passive hns the underlying position shown in 3 . 2~. Tlre synlbol 8tVt1 signifies the fact that the subject position of (2c) is fillecl by all NP that clominates rto lexical stritcttlre. (Following Chomsky, I assume that a passive sentence in fact has no trnderlying subject, that an ayr!ntive "by NP" prepositional phrase originates as such in trnclerlyiny s tructurc ) The trace in (3b) inclicates that the phrase "to be happy", wllicl1 the brackets show Is really an ~mbecicleci clarrse, has an underlying subject wl~lch. is iclcnticcll with tlie surface subject of the matrix S, the cibltse tlrat clomiriates the embedded complement. Note that what Is conceptually the underlying subject of the enihatltled clarrse tias been passivized into subject position of tlin matrix S, a phenomenon commonly called "raI~Ing@~ The ntiaiysis of this phenomenon assumed here derlves from [Cl~onrsky 733; it is an alternative to the classic analysis which involves "raising" tlie subject of the embedded clntrscr illto object position o f tlie matrix S before ptlssivization (for details of this later analysis see [Postal Tho Passive Rule in this section ond the next, I will briefly sketch a so111 t ion to tlie plionomena of passivization and 'tralslngtt in tlia cotitoxt of a grammar for PARSIFAL. This section-will prcscnt tlio Passive rule; the next section whL sliow how this rille, wltlrout alteration, l~anrlles the ttralsing@g cases.Let 11s uegln with the parser in the state shown In figr~ro 4 below, in the miclst of parsing 3.2a above. The analysis process for tlie sentence prior t o this point Is essentially parallel to tlie analysis of any simple declarative wllh o11c exception: the rule PASSIVE-AUX in packet BUILD-AUX has clococlecl tile passive morphology In the auxiiia.ry atlcl ylven the auxiliary the feature passive (although this fciaturc? is not visible in figure 4) . At the point we begin our example. tllc packet SUBJ-VERB is active. Figure 6 -After PASSIVE has been executed. Now rilles will run which will activate the two packets SS-VP and IRF-COMP, given that tlie verb of VP17 is 18sclisclule11. These two packets contain rules for parsing simple olljects of non-embeclded Ss, and infinitive coniplements, respectively. Two such rules, each of w l~i c h utilize nn NP immediately following a verb, are given in figtrre 7 Irelow. ?lie rille OBJECTS, in packet SS-VP, picks Up an NP nftcr tlre verb and attaches it to tlie VP node as a simple object. Tlie ruls INF-S-STARTI, in packet INF-COMP, ttmigr_lors when an NP is followecl by @'to1' and a tenseless ver0; it il~iliates an infiliifivo complement and attaches the NP as its strbjoct. (An example of such a sentence is "We warrtecl John to give a seminar next weekg1.) The rule INF-S-STARTI must have a higher priority than OBJECTS beceuse tile pattern of OBJECTS is fulfiiied by any situation that fulfills tlie paLtern of INF-S-STARTI; if both rules ate in active ( x~c k o t s arirl match, the hlglrer priprlty of INF-S-START1 will cause it to be run instead of OBJECTS. here in cletoll, llote that the rule OBJECTS will trigger with tlrc parsor-in tlie state shown In figure 6 above, and wlll atttrclr NP63 as tlro ol)jt?ct of tlie vorb wschedule. OBJECTS is tlrrfs totally Indifferent both to tlie fact that NP63 was not o rngular NP, Dilt rather a trace, and the fact tlrpt f i~5 3 t l i c l not orlgitiata i t 1 tlio i n1,ut string, but was placed into the b i~l l o r by grammatical processes. Whetlier or not thls rule is executecl, is al~solutely unaffected by differences I~etween an actlve sentence and its passlve form; the ntrolysls process for e i t l i~r Is Identical as of thls point In the pnrsinq process. Thus, the arrelysis process will be exactly i~rtrnllal in botli cases alter the PASSIVE rille has been exac!~teci. ( I remind tlie reader that the analysis of passive ass~~nrocl EIOOV~, following Cliomsky, does 170t assume a Irroccss of "agent cioleti~n~~, "subject p~s t p o s i n g~~ or the like.)Passjvcs in En~bodded Complements. -"Raising"Tlie rcnclbr may have wondered why PASSIVE clrops the trace it creates into tlie buffer rather than InrtiiaclioZely attacliing tlie new trace to the VP node. As we will sac below, sirch formulation of PASSIVE also correctly analyzes passives like 3.3a above wl?ich involve nraislng", btlt with no arlclitlo~~al complexity added to tlie grammar, correctly capturing an Important generalization about t l i s l . To sliow the range of the generalization, the exfin~ple wliicli we will investigate in this section, sentence (1) in fig~rre 8 bclow, is yet a level more complnx than 3.3a ahovc; its analysis is sliown sclienratically in 8.2. In this cxtrml>le tliore ore two traces: the first, tlie subject of the entl~cclclecl clause, is bound to the subject of the major clause, tlie seconcl, the object of the embedded S, Is bound to the first trace, and is thus ultimately Qoirnd to the s u l~j c c t of the liiglior S as well. Thus the underlying position of tlie NP "the meeting" can be viewed as being t l~c ohject position of the ombedded S, as shown in 8.3.( 1 )'Tlic lneeting was believed to have been scheduled for Weclnestlay.(2)Tlie nieetilig was believed [s 1 to have been scheduled. t for Waclnesdq)r](3) V believecl Ls V to have scheduled the ineetlng for Wedtiesday J. W e bedin our example, once again, right after "believecl" has been ottached to VP20, tlie current active ~)ocle, as shown it1 figure 9 below. Note that the AUX node has been labelled passive, although this feature is not sliown here. Again, rules will-now be executed which will activnte the packet SS-VP (which contains the rule OB.,IECTS) allti, since "believe" takes infinitive complements, tlie packet INF-COMP ' (which contains INF-S-START1 ): among others. (These rilles will also deactivate tlie packet SUII,I-VEFll3.) Now tlie patterns of OBJECTS and INF-S-START1 wlll both match, ancl INF-S-START1 ,*shown above In figure 7, will be execitted by tlie interpreter since It has tlie liigiier priority. (Note once again that a trace is a perfectly nornial NP from the poiht view of the pattern malcliingl process.) Tliis rille now creates a new S node lal~ellecl Infinitive and attaches tile trace NP66 to the new infinitive as its sui~ject. The resulting state is shown in figure 1 1 below. We are now well on our way to tlie desired analysis. An en\lrecldeci infinitive has been initiated, and a trnco bdunci to tlie subject of the dominating S has been attacl~oci as its suhj8ct. although 110 rule has explicitly itlowsr~rlll tlie trace from one clause into the other.The parser will now proceed exactly as in the previous exa~iple, It will the aux~li~ry. attach It, and attach the verb "schetluled" t o a new VP node. Once again PASSIVE will match and be executed, creating a trace, I~i~irling it l o tlie subject of tile clause (in this case Itself a trace). alicl clropplng the new trace Into the buffer. Again the rule OOJECTS will attach tile trace NP67 as the object of VPZ1, and tlie parse wlll then be completed by clranimatical processes which will not be discussed here. An eclilteci lorin of tlle tree structure which results is shown in figure 1 2 below. A trace is indicated in this tree b y grving tlie tcrnlillal string of its ultimate binding in parentheses. This example demonstrates that the simple formillation of the PASSIVE rule. presented above, interacting with other simply formulated grammatical rules for parsing objects and initiating embedded infinitives, allows a trace to be attacliecl either as tlie object of a verb or as tlie subject of an embedded jnfln!tive, whichever is tlie appropriate analysis f.or a given grammatical situation, Recot~se tlie PASSIVE rule is formutated in such a way that it clrops the. trace it creates into the birffer, later, rules, already fornlyiated to trigger on an NP In' the'buffer;. will analyze sentences with NP-preposing exactly the same as tlioso witliout a preposecl subject. Thus, we see that .the aviriiabiiity of tlie buffer maolteinism is.crucial to chpturlng tliis generalization; such a generalize'tion can only b e s\atr?cl I,y a parser with a mechanism much like the buffer usacl hare.Before ti~rning now to a sketch of q computatlonal account of Clloniskyls constraints, there are several illlportnht liniitations of this work which must be enumerated.First of all, wliile two of Chomskyls constraints soon1 to fall out of the grammar interpreter, there seems ttl Ire no apparent accoiltit of a third, tlie Propositional Island Constraint, in terms of this mechanlsni..Seconcl, Cliomskyts formutation of these coristraitits is irltended to allply to all rules of grammar, both syntactic rules (i.e. transfbrmations) and those rules of seninntic interpretation wliicti' Chomsky calls "rules of c d n s t r~a l~~, a set of shallow senientic rules whlch .govern A~ipplioric processes [~llomsky 773. The discussion here wi(l only tolloh on purely syntactic phenome~ia; the question uf,j~ow rules of 'semantic interpretati~n can be meshed with t l i a fr-aniework presented in this clocument has yet to be investigaterl.'Tliircl, tlie arguments presented below deal only wirlli €nglish, ancl in fact depend strongly upon several facts about Etiglisl~ sflitax, most crucially upon the fact that English i s sul~ject-initial. Whether these argu.ments can be successfully extencled, to other language types 'is an open cluestion, and to this extent tliis work must b e considered exploratory.Ancl finally, I will not show tliat these constraints must bo k u e withoyt exception; as we will see, there are vari-ous situations in wliich tlie .constraints imposed by the gratnninr interpreter can be circumventecl. M O S~ of fliese situations, tliougli, will be shown to demand much more coml)lex grammar forniulations than those typlcally needed in the' flarnaer so, far constructed. This is quite in keeping with tlie suggestion made by Chonisky [Chomsky 77J that tile cotlstrai~its are not necessarily without exception, but ratlier that exceptions will be l'liiyhly marked" and therefore will caulit heavily against any grammar that includes them.The Specifiecl Sul~ject Constraint (SSC), stated informally, says tliat no rule may involve two constituents tliat are Domiriat&d by different cyclic' nodes unless tlie .loweraof the two is tlie subject.of .an S or NP. Thus, no rule may involve constltuents X .and Y in the structure shown in ?igae 13 below, i f and p'are cyclic nodes and Z is the si~bjcct of a, Z distinct from X.No rille can involve X wd Y in this structure. Tne SSC; .explains why tlie surface subject position of vcrl>s' like and lais certahfi' which have no irnclor!ying subject can bo flied only by the subject and not tlie object of tlie enibsdded S: Tlie rule '"MOVE NPgl is free to shift 'ouy NP into the empty-subject -position, but is constrainecl by tlie,SSC so tliat the object of the embedded S canno'f'be nlovktl out bf .hat clause. Thls explains why (n) in figure -1 esseiice, -tlien, the Specified Suljject Constraint bd~istrai~ls tlie rule "MOVE NPu.in' s l~c h a way that only the si~l?ject' of a clei~se con be moved out of that clause into a position in n liigl~er S, Tl~us, If a trace in an annotated stlrfaco structure is bouncl to an NP-Dominated by a higher S, l l~n t iracc mtist fill the subject posiiion of the lower clause,In the reniaincler of tljis s e~t i o n I will show tliat the oramnrnr interpreter constrains grammatical processes in strcti a way that a;iriota ted strrface structures constructed Ily Ilie granlmilr interpreter will have tlits same property, givwi tlic formulation of tile PASSIVE, rule presented-above, 111 tcrais of tlie parsing process, tliis means tliat if a trace Is " I o w~r e c l~~ from one clause to another as a result of a "MOVE NP"-tyl)e or>era tion (luring the parsinq process, then it-will he attacliecl as the subject of tlie second clause. To I)e wore precise, if a trace is attacl~ed so that it is Donrinated by sonie S node S1, and tl1.e trace i s bound to an NP Doniinatecl by sonie other S node S2, tlien that trace will aocessahly bo attached so that it fills the suGect position of S1. This is depictacl in firlure 16 below. This statement of PASSIVE does more, however, tliali slmgly cegtilre a generalizhtlon about a specific construclion. As I .wIII argue it1 detail below, tlie behavior spocifled by both the spocihed Subject Constraint and Sirbjaconcy follows almost immediately from this formulatlon. In [~& c u s 7 7 1~1 BrgUe that tlils formulation of PASSIVE Is the otily simple, tion-ad hoc, formulation of tlils rule possi,ble, dnct tliat qll other rules characterized by tlie competenca rule "MOVE NPtt ~r u~s t operate slmllarly; here, however, I will olily show tliat these cotistrainta,-fallow naturally. from thls formulation of PASSIVE, leaving tlie question of necessity asicle. I will also assume one additional consl~aint below, the Left-to-/light Constraint, which will be brlefly motivated lator in tliis pnper as a natural conclition on the formulation of a ciramniar tor tlils nieclianlsm.Tlie Le,f t-to-Right Constraint: tlie constituents in the buffer aro (almost always) attached to hlglier IeQel constittle~its In left-to-rigl(t order,. 1.e. the first constitl~ent in the buffer is (almost always) attncliecl before tlie seconci constituent.I will now show tliat a trece created by PASSIVE which is bounci to an NP in one clause-can only servd as the subject of a cla11se.dominated by thaf first clause.Given tlie formulption of PASSIVE, a trace can be t'lowerecltt illto one clause from another only by'the indirect route of clropping it illto tlie buffer before the subordinate clause node is created, which ig exactly how the PASSIVE r111n operates. This njoans t l b t the orderlnp of tlie operations is crucially: 1) create a trace and drop it into tho buffer, 2) create o li~borciinate S node, 3) attach the trace to the newly creotecl S, node'. Tlie key pmt Is that at tlie ti~iie that tlie sili~orclinate clause node ig created and boconios ihe current activtFnode, tlie trsce must be ~sitfjng in t11e' l 111f fer, filling one of tlie tliree buffer posltlons. .Tliust tlie pacser will be in tlie state -shown 1n.figure 16' below, with tlie trice, 'in fact, most likely In the first buffer position. Figure 16 -Parser state after embedded S created.Now., given the L-to-R Constraint, a trace which is ill tlie hi~f'feroat the time lliat an en~bedded S node is first creatccl mtlst be one of tlie first several constituents attacl~ecl to file S node ,or Its daughter nodes. From tlie structure of Engli.sli, we know that the leftmost three co~istituents of an enlbedded S node, Ignoring, toplcalized constitrrents, milst .either be COMP NP AUX or NP AUX [", VERB ,.. 1.(Tlie COMP node will'clominate flags like lithatll or' llforll that mark' tlis beglnnlng-of a complement clause.) But then, if a trace, itself an NP, is one of the first 'several constltuents attilcliocl t o an leml)odded clause, tlie only position It can fill will b e tlio subject of tlie clau'se, exactly the empirical c o~i~e y a e n c e of Chomsky8s Specified Subject Constralnt in s i t c l~ p s e s as explnl~iecl .gbove,The L-to-R Constraint Let IJS now return to fhe motivation'for the L-to-R Constraitit. Again, I will not attempt t o prove that thls colistrelnt hyst be true, but merely tq,show why it Is plausible.Enr~~irically, tlie Left-to-Right CCfiTstraint seems to liolcl for tlio triost part: for the grammar of Enyllsh discussed in this I)aper, ancl, it woulcl seem, for atiy grammar of English t l w t attempts to capture tile same range of generalizations as tliis (jramm~r, tlie constituents In the buffei are utllized i;i Inft-to-riglit orclor, witli a small range of exceptions. Thls usiige is clearfy not enforced by the .grammar lnter~reter as proJeiitly iniplen~ent?ee(i; it'is qulle possible to write a set of graniniar rirles that specifically ignores a constituent_in t h e buffor until some arldtrary point h the cl&se, t h o u g h~u c h a set of rilles would be lilghly ad'hoc. However, there rarely seenis to .he a need to remove other than the ' first constituent in tlie buffer.The one B.xception to tlie L-to-R Constraint seems t o Ile that a constituent Ci may be attached 'before the constituetit to* left, C ,I,'if Ci does not appear in surface structi~re in its underlying position (or, if one prefers, In i t s ynniar,kecl l>osItion) and if its rkmoval .from the buffer reestablishes the unmarkec! order of the -remaining constituents, as in tlie caserof tlie AUX-INVERSION rule clis.cussecl~ earlier in tliis paper. To capture .this notion, the L-to-R Constrairit can b e rebtated as follows: All mnstitrrents niilst be attached to higher. level econstituents accorrling'to the .left-'to-right order bf constltuents in the unsarkecl case of tliat constituentls structura.Tliis reformirlation is interesting .In that It would be n natural consecluence of the opereflon of tlie granimar intcjrpreter if packets were associatecl with the phrase strtrct41re rules of .an explicit "base componentI1, and'these r~t l e s \ were used as templates to build up the structure assignccl by tlio gramtiiar interpreter. between Y bticl X) then Y Domlnntes X.Tlre principle of Subjadeticy,' informally stated, says that no rulo can involve constituents that are soper'e'tecl by more .than one cycllc*node. Let us say that a nocie X i s sublacent to a node Y i f tliere is at most ope cycilc nocle, i.e. at niost one NP o j S node, between the cycilc node tliat Don\iliates Y and tlie node X. Given this clatmition. the ~u l l j s c e n c~ prlnciplk says tliat no rule can involve coristltuents tliat are not subjacent.Tlie Subjacency principle l~nplies tliat movemerit .tiles qre co~istrai~iecl so tirat* they can move a constituent ~n i y into positipns tliat tlie' constituent was subjacenj to, i.e. o~ily witl~iri .the clei~se (or NP) in which it originates, or No rille call involve X and Y in this structure.Subjacency inlplles tliat if a constituent is to be "lifted" up more tlian one level in constituent structure, this olioratioti nrtrst. be clone by relreated operations. Thus, to use one of Clio~lskyls exaniples, tlie sentence given in flguro 18a, with a cleep structure analogous to lab, must be tleriveci k s follows (assuming that "is certainl1, like has no subject in uncierlyiny structure): The ,deep structure nlrlst first u~iclergo a niovemeiri operation that results in a ~b i t c t u r e ~nalogous to 18c, and then another movement operation illat results in 18ci, each of these movements leavilill a trace as sliown. That 18c is in fact en intnrniecliato structure is si~pported by tl;e existence of sentmices sucli as 18e, which purporteclly result when the V ill tiie matrix S is replaced by tlie. lexical Item "itvg, and the enil~ecicied S' is tensed rather tlian infinitival. Tlie strttctirre given in 18f is ruled out as a possible annotated sirrface structure, because tlie single trace could only be left if tlie NP was moved in one fell swoop'from'lts untlerlyiti~~ pohition to its.position in surface structure, which woi~lcl .violate Sub/o'cency. IJavlhg statetl Subjaconcy'ln terms of tlie dbstrgct con~potinca tliaory of generative grammar, 'I now will show that a, parsing correlate of St~bjacency follows from ill8 strl~ctitre of tlie. grammar interpreter.. Speciflcaily, I ill slinw tliat tliere are. only .,limited dases in wlilch a trace qeheratotl by a "MOVE-NP" process can be ttiowered; inotg tliaii olio cleuse, i,enqtIiat a trace cieated and bound while any given S is crirrent niust almost always be ettached either tb tligt S or to an.S whlcli Is dominlted by that S.Let us bcglli by exeririning what it would mean to lower a trace more tlian one clause. ~i v e n that a trace can otily be 1810wereclfl by dropping it lnto tlie buffer and tlien croati~ig a sul~orilinhte S node, as discussed above, low&incj a trace nrore tlian one clause necessarily implies tlie -follnwinf~ sequence of events, depicted in figure I 9 below: First, a trace N P~ must (a) be created with some S F i y m 10-Lowering a trace more than 1 clause Dtlt tilis secluence of events is highly unlikely. The essence of the nrglrlment Is tli-is;Nothing in llie buffer can change between tlie time that S 2 is createcl and S3 Is cieated if NPI remains In tlie htrfler. NPI, like nny other nocle that is dropped from tlie active liorls stack illto tile buffer, is lliserted into tlie first Ilulfcr position. But tlien, by tlie L-to;R Constraint, notliing to tI.1e riglit of NP1 can be nttacliecl to a higher level constit'irent utiril-NP1 is ~atiaclidd. (One can show illat it Is ~~i o s t i~nlikclyttliat any constitu6nts will. enter t o the left of NP1 after it is clrol)ped illto tlie buffer, but 'I will .suppress tliis ciotail here; tiie 11111 argumerit is included In [Marcus 771.) 011t if tlte contonts .of the buffer do 'not change between the creation of.S2 and S3i then what.can possibly nintivirte the creation of both S2 and $33 The contents of Hie i j~f f or must necessarily provlde clear evlclerice that botli of these clailses are present, slhce, by the ~etertiiinism tlypothesls, the parser must be correct if it ililtint&s n canstititent. Tiius, the same-three c~ns~kltuents In tiin !)trffer ~attst provide convincing evidence ndt only for the creation of S2 but also'for S3. FurtiTBhnore, j'f NP1 is to becomk athe subject of 83, elid if S 2 Dominates 53, then it woulcl Seem tliat tiie constituents that follow, NP1 In the buffor nirist also he constituents of S3. since ~3 must be ,coni~~ieteci bdoro it is dropped from tiie activg node stack anrl 'constititetits can then be attaci~ei'to 52, But then 52 must be crnnted entirely on tile bast3 of Wfdence ~rovided by tile cotislitile~rts of.andther clause (unless S3 has less than Iliree Constituents). Tliils,lt would seem that the cbntents of tile i~irffer cannot' provide evidence for tile presonce of both clauses itnless the presence of'S3, by itsclf, is cnou(~lr to provide confirnring evidencd-\for the i,resencaof S2. ~i i l s woidd be the case only if there were, say, a clP~tsel c~nstruction. t k t cpiid:~nly appear (perhaps i t i n garliciil6~~environ?j~nt j 'as ,tile lliltial 'constituent of a tiiglier claaso. in this. case, If there Pre such constructions, a vioio t loti of Sui~jecency a tiauld be ibi,s~bie.Wilh. the one exception just mentioned, there is no molivcltion for crcati~lg two clauses in such a situation, and thus the iliitibtion of only one such clause can be motivuty?d. But if only one cldiise is Initiated before NPI is attached, tlicn Ni)l muat bc attaclietb to tliis clause, and this clause is ~~ecpssarily subjacent to the clause which Dominates the 'NP Lo wliicli it is i~ou~id. Thi~s, tlie grammar interpreter wlil behave as if i t enforces the Subjacency Constralnt. ' As a concltrdingl point, it is worthy of note that w\iile the gronrmar interpreter appears to beliave exactly as if it wRra colist~ained by the Subjacency principlei it is in fact constrainotl by a \;crsion of tlie ~laiisemate constraint! (The Clauseniate Constraint, long 'tacitly assumed by linc_~irists but first explicitly stated, I believe, by Postal [Pnstal'64] , states that a transformation can only involve constittrents tlint are Dominated by the same cyclic node.Tliis constraint is i t tiie heart of Postal's attack bn the constraints that nre discussed above and his argument for'a ;'mising" nnnlysis.) The yrehnor interpreter, a* was stated above, litiiits gramnier rules from examining any nocle in the. active nodc stack hiyl;er tiionathe current cyclic hade, wtiic.1; is to say that it call only examine clausemates. The trick is'-that a trace is created and bdund wliile it is a "clauscnr~tc~~ of the NP to wlijch it is boirnd in tiiat the citrrent cyclic~nocie at tliat time Is the nocle to which that NP is atteclied. Tlie. trace is t i ; 6n dropped into the buffer a~i r l &iotlier' S ilotle is created, tilereby' destroying. the clot~scniatc relationslii~~, * The trtice Is then attached to this new S . ~mcle. Thus, it1 a sense, the trace Is lowered from one,clpuse to another. Tl?e.,crucial point is that wiiile tlfis 1 lowering goes on n s a .resuit of the operatlon of,the graminar. i~;tbr~,rct~ri it Is olily lmdcitly lowered in tiiat 1) the trace' was never .atlaclied to the' higher S and 2) it isanot dropped irito the bi~ff.er .because of any reoltzatlon that it. m u~t be "1owDroc1~~; -in fact i t may end upattached as a clausemate of the NP to whicli it is boundas the passive examples preqenteil earlier make clqar. Uie trace is simply dropped into tlie buffer because its grammatical function Is not clear, anti the creation. of tlie second , S . follows' from other Indnpondon tly n~ollvated gramma tics1 'From the poilit of view of tliis orocessing ,tliciory, we. can have our cake ant1 eat i t too; i o the exteni that it makes sense to ninp .resuits from tlie realm of processing into the realm of ~ompotcnce, in 'a sense both the clausemate/'~relslngt~ and tire Subjacency positions are correct.111 closi~lg, I would like to sliow tliat tile properties of t h e grnnimar interpre'ter crucial to capturing the behavior of ~liomsky's constraints were orlglnaliy motivated by the Determlnism Hypdhesis, 'and thus, to some. extent, the Deterniinisni t~ypotl~esis explain-s chomsky's constraints.Tlie strongest form of such,an argi~ment, of course, wotrlcl be to sliow that (a) *ittier (i) tlie grammar Interpreter a c c o i~~i t s for all 01 ~honisky~s. cbnstraints in a inanner which is coucli~sively universal or (ii) the constraints that i t will riot ncco~llit for nre wrong and that (b) tlie properties of the grauiiiiar interpreter wlilch were .crucial for this proof were forced by tiie Determinism Hypotliesis. if such an argument could ho made, it would -show that the ~etermlnlsm Hypotliegis provides a natural processing 'account of the ~i n~i~i s i i c ddte characterized by. chdmskyts constraints, giving strotip confirmation to ttie Determinism Hypothesls. I hsvs, shown none of tlie above, and thus' my clsibs niirst Be proportionately more modest. i have argued 0111 y that irnpor'taiit sub-eases of Chomsky1s constraints follow fiorn the giammar interpreter, bnd while I c m show tlia t tile Onter~nlriism Hypotl~esls strongly mo(lvaies the tiiecl,atiisnis froni wliicli these arguments folloui, I .cannot show hecesliity. The, &tent to .which tiilk argument provicles evidence for the ~eterminism.,Hy~othesis must thus be left to thd reader; no objective measureaxists for such matters.The ability to drop a trace into the buffer Is at the heart. ot tile arguments pre.sented here for Subjacency and the SSC as consecluences of the functioning o f t h e grammar inieroreter; tliii' i ; tlie central operation upon wtitbh the above br(~ilnietits are bosecl. But the'buffer Itself, and the fact thkt a constituent .can be dropped into the buffer'if Its pramnlntical' f i c t i o n is uncertain, are directly motivated by the ~eterhinism. HypdtMsJs. ~i v e n this, it .is fair t o . claim tlirr t if '_~homsky's constraints follow from. the operatlob of -. . . < tltc ranihe her interpreter, t hei i they are strongly linked to the Deterniinism Hypotl~esis, if , Ciiomskyls constraints. are In fact true,, then the arguments prgsented In this paper providq solid evidence in support of the '~eterminlsm Hypothesis. | null | Main paper:
:
forni one of the cornerstones of his current linguistic theory, are eml~oclied in a set of constralnts on language, a set of restrictions on the operation of rules of grammar.Tlils paper wlli outline two argtiments presented a t length i r i [Marcus 771 clemonstrating thctt important subcases of two of tliese constraints, tlie Subjacency Principle oncl tlie Specified Subject Constraint, fall out naturally from the structure of a gramnlar interpreter called PARSIFAL, wliose structure is in tiirn based upon the hypothesis tliat a natural language parser needn't simulate a liondeterministic macliiiio. This "Deterniinism Hypothesistt claims that natural laiigiiaqe can be parsed by a computationally simple nieclion~sm tliat uses neither backtre~king nor pseudopsrailclism, alid in wliicii all grammatical structure c m t e d by the parsor is 'ti~iclelible'v in that it must all be output as part of tlie structural alialysis of the parser's input. Once 15uilt, no grammatical structure can be d~scarded or altered in tlio course of tlie parsing process.In l)prticular, this paper will show that the structurq of the grammar interpreter constrains Its operation in silch a way tliat, by and large, grammar rules cannot parse sentences whlch violate either tlie Specified Subject Cgnstraint or tlre ~tihjacency Principle, The component of the grarnniar interpreter upon wh~clt this. result princlpaily clcpcncls is motivated by the Determinism Hypothesis; this result thus provides inclirect evidence for tlie hypothesis.This result also clepends upbn the use Within . a coni~~trtatlonal framework of tlre dlosely related notions of annotated surface structure and trace theory, which also derive from Cliomskygs recent work? (It should be noted that these constraints are far from universally accepted. They are currently the source of much contraversy; for various critiques of Chomsky's positio~l see [Postal 74; Breslnan 76) . However, what is presented below does not argue for these constralnts, per sa, bill ratiier provides a different sort of explanation, based on a processing model, of why the sorts of sentences which these constraints forbid are. bad. While the exact formulation of these constraints is controverstal, the f a c t that some set of constraints is needed to account for this range of data Is i~enerally agreed upon by most generative graniniclrians. Tlie account which I will present below i s crt~cially Iinkecl l o Clionisky's, however, in that trace theory is at tlie haart of this account.)Because of space lirfiitations, this paper deals only with those gr~nrmatical processes characterized by the con~petence rille ''MOV~ NPg'; tlie constraints imposed by the grammar interpreter upon those processes cliaracteri7ecl by tlie rtrlc "MOVE WH-phrasew are discussed at I~n q t l i in [Marcus 771 wliere I show tlmt tile behavior ctiaracterlzecl by ,Ross' s Complex NP Constraint [Ross 671 ~tsetf follows clirectly from tlie str~rcture of the grammar interpreter for rather cliffere~it reasons than the bhavior considereti In this section. Plqo h c a u s e of space Iiiiiil~itio~is, I will not attempt to show that the two colistralnts I w~l l deal with here necessarily follow from the grtlmmirr interpreter, but rather only that they naturally follow from tlie intarpretar, in particular from a slmple, natural formihatlon of a rule for passlvization, which itself deponcls lieav~ly upon the structure 6f the interpreter. Again, iiecessity is argued for in detail in [Marcus 771 .Tliis paper will first ouNine tlie structure of the graninlilr interpreter, then present the PASSIVE rule, an8tlinn filially show liow Cliomskyts constraints ''fall outM of tlie formulation of PASSIVE.I3efore proceedilig with the body of thls paper, two other kaportnnt properties of the parser should be nie~~t~oriecl wli~ch will not be discirssed here. Both are cliscussecl at lengtli in [Marcus 771 ; the first is sketched as well in [Marcus 78) .1) Simple cuies of grammar can by written for this irlterpreter which elegantly capture the significant gaiieralizatior~s behincl not only passivization, but also such Coiistrtictiolis as yes/no questions, imperatives, and selitetices with existential there. These rules are remiillscent of tlie sorts of rules proposed wjtiiin tlie frnniework of t l~e theory of yeneratlve grammar, despite tlie fact that the rilles presented here must recover underlying structure given only the terminal string of the surface form of the sentence.2)Tlie. grammar interpreter provldes a sirhple exl>lanatkn for tlie difficulty caused by "garclen pathtt sentences, such as V i l e cotton clothing is made of grows in M~ssisslppi.~ Rilles can be written for this interpreter to resolve local strirdtural ambiguities which might seem to require notlcleterminlstic parsing; the power of such rules, howaver, rlspe19ds lirpon a parameter of the mechanism. Most structural aniblgirlties can be resoivecl, given an appropriate setting of this parameter, but those "whlch typically cause garden paths cannot.PARSIFAL mai~itaills two major data structures: a p\isliclown stack of l~icomplete constituents called the active node stack, nricl a sniall tl~r'ee-place constituent buffer whlch contains constituonts whlch are complete, but whose higher level yramniatlcal functlon is as yet uncertain. Figure 1 below stlows a snapshot of the parsergs clatn structures taken wlllle parsing the sentence "John siioi~ld liave schecli~lecl tlie meeting.", N d e tliat the active nocle stnck in sliown growing downward, so that thg strirctirre of the stack reflects the structule of the enrerging parse tree. A t tho bottom of the stack is an ai~xiliary ~iorle Cabelled wit11 the! features modal, p~s t , etc., w l i i c l~ lias as a daughter tlie modal 1tsliouic118. Above the bottoni of tlie stack is an S notla wlth an NP as a daughter. rlonri~iatinc~ the word "Johnt1. There are two words In tile birffer, tiie verb "hovew in the first buffer cell and the word "schedu1oc.l" irr the second: The two word8 "the meetingt1 have notA yetacome to tiie attention of the parser, (The structures of form "(PARSE-AUX CPOQL)' and the like will be explained below.)The Active Nocle Stack - Sl (S DECL MAJOR S) / (PARSE-AUX CPOOL) NP : (John) AUXl (MODAL PAST VSPL AUX) / (BUILD-AUx) MODAL i (should) Tile Buffer 1 : WORD3 ("HAVE VERB TNSLESS AUXOERB PRES V43S) : (iiave) 2 : WORD4 ("SCHEDULE COMP-OBJ VERB tNF-OBJ V-3S ED=EN EN PART PAST ED) : (scheduled)Yet unseen words: the meeting .Figi~re 1 -P'.4RSIFALts two major data structures.The constituent buffer is tiie heart of the grammar interpreter; it is tiie central feature that distinguishes tliis parser froni all others. The words that make up the parser's input first conie to its attonti011 when they appear a t the end of tliis buffer after morpno~ogical analysis. Triggered by tlie worcls at the beginning of the buffer, the parser may clociclo t o create a new grammatical constituent, create a new node at tile bottom of the active rrode stack, and then l~egiri to attach tlie cor~stituents in the buffer to it. After tliis new constituent is conipleted, the parser will then pop tile ~i c w constituent from the active node stack; If the grcllnill~tlticai role of this larger structure is as yet i~nclcterminorl, the parser will insert it into tlie first cell of the I~irffer. The parser is free to examine the constituents its tJic buffcr, to act irpon them, and to otiiewise use the I~u t fer as a workspace.While tlie buffer allows tile parser to examlne some 01 t h e context surrounding .a given constituent, it does not allow arbitrary look-ahead. The length of the buffer Is stl.ictly iimiferl; in tlie version of the parser presented here, the b~l f f e r tins only three cells. (The buffer must be cxtcnclccl to five colis to allow the parser to build NPs in a manner wliich is transparent t o the taclause levelta grammar rilles wllkh will be presented In this paper. This extended parsor still Iias a winclow of only three cells, but the eflcctive start of the buffer can be changed through an "at tenlion slilf ting n~echanisrn~~ whenever tlre parser Is ISuilding an NP. in effect, tills extended parser has two " l~(~i c a i~~.btrf fers of "length three, one for NPs and another for clauses, with these two buffers implemented by allowing an overlap in one larger buffer. For details, see [Marcus 771 .1Note that eacli of tiie tilree cells in the buffer can Iiolcl a grammetical constltuent of any type, where a constlt~reril Is any tree that tlie parser has constructed u~~c l o r a singlo root nocle. The size of the structure ~~ndcr~iocrtlr the tiotle is Immaterial; both ltthatl1 and "that tlie 11i g green cookie monstsras toe got stubbedea are perfectly good col.rstituents once the parser has constri~cterl a srrbordinate clause from tile latter phrase.Tlie constituent buffer and 'the active-node stack are acted ilpon by a grammar wliich is made up of ~mttcrn/action rules; tlris grammar can be viewed as an au$jrnentecl form of Newell and Simont$ production systems [Newell & Simori 721. Each rirle is made up of a pattern, wliich is niatciieti against some subset of the constituents of tlie buffer and the accessible nocles in the active node stack (al~out wliich more will be said belowh and an action, a sequence of operations which acts on these constituents. Etlcii rule is assigned a numerical priority, which the graninlar interpreter uses to arbitrate simulteneous matches.The yrnmmar ns a wiiole is structured into rule packets, clumps of grammar rilles which can be, activated ant1 clcactivatecl as a group1 the grammar interpreter only attenipts to match rules in packets that have been activated by the grammar. Any grammar rule can activate a oachnt by associating that packel with tliq constituent a t tho i~o l tom of tile active nocle stack. As long as that node is at tlie 1)ot'Coni of the stack, the packets associated with it arc! nctive; wlxen tliat node is pushed into the stack, the ~~a c k e t s remain associated with it, but become active again tmly wlieri that node reaches the bottom o l the stack. For sxnmple, in figure 1 above, the packet BUILD-AUX is nssociated with the bottom of the stack, end is thus active, while tlie packet PARSE-AUX is associated with the S node aljove tlie auxiliary. 'u*<worrl>u mean "has tlie root <word>", e.g. "Rhave" means principles. attd their embodiment in PARSIFAL, are as follows:"has -the root tthave"lt. The tokens 181st11, l12ndI1, "3rd8I and 8 t~" (or "c") refer tw the constltuents in the Ist, 2nd, and 1 ) A deterministic parser must be at least partially data The parser (i.e. the grammar interpreter interpreting some grammar) operates by attachiflg constitt~ents w l l i~h are in the buffer to the constituent at tlta l~ottom of tlte sfask: functionally, a,constituent IS in the stack when the parser is attempting to find its daughters, aitd in the buffer when the p a w r is attempting, to find its n~otlier. Once a constituent in She buffer has beeh atlacliecl, tile grpmmar interpreter will a&tomatically remove it from the buffer, filling in tlie gap by shiftihg to the left the constitumts formerly to its right. When the parsei has completed the mnstituent at the bottom of. the stack, it po(,s jliat constityent from tlte adive node stack; the corlstitilent either remeins attached to its parent, if It 'was attoclied to some larger constituent when it was created, or else .it falls into the first cell of the constitu8ntd buffer, 2) A determinislic parser must be able to reflect expectations that follow from the partial structures i~ctijt ctp during the parsing process. 'Packets of ritlcs can be a-ctivated and deactivated by grammar rilles to reflect tlie properties. of the constitue~its .in tlie active node stack.3rd b u f3) A deterministic parser must have some sort of constrained look-ahead facility. PARSIFAL1s buffer pro~ides.tltis constrained look-ahead. Because the bi~ffer can l~oicl several constituents, a grammar :rule can exaniitte tlie cont&xt that follows the first constituent in the buffer before deciding what granrmirtical role it fills in a higher level structure.The key idea is that the site of .the buffer can be sltargly constrained if each location In the buffer cat1 hold .a single complete constituent, regardless of Ilia t constituentls size. It must be stressed that this look-ahead ability must be constrained in tome manner, as i t is here by limiting the length of the-. buffer: otherwise [he "deterrnini~m~~ claim i s vacuous. tlte rrsage o f tlie term here ,can be thought of as a synthesis 01 tlte two cmcepts. Following winograd, this tern? will be used to refer to a notion of surface structure annotated by the addition o l a set of-featyres to each node in a parse tree. Following Chomsky, the term will be used t o refer to a tiotion of surface Structure annotated by the aclclition of an element called trace to indicate the "underlyittg position" of "sltiftecl" NPs-.In current linguistic theory, a trace Js essentially a ~~~~ltonoio<clically null" NP ilt t siltface . structure represetitation of a sentehce that 'has no daughters but is "botylclu to Ilte ' NP that filled that position at some level of i r n c l o r l y~ structure. In a sense, a trace can be viewed as a "dr~mmy" W that serves as a placeholder for. the ,NP t h a t 2ariier filled tll'6t~.position; in the same' sense, the trace's' bintlinfl cen ha viowed as simply a pointer to that NP. It sl~oi~icl be strcssscl at tlis outset, liowever, thbt a tpace is inclistingirisha~~ie froar a nornral NP in terms of normal grnnrtrrotlcal processes; a trace is an NP, even though It is an NF' that clominates no lexical material.There are sovcral reasons for choosing a properly annototoci surface strircturo as a primary output rsprclsantntion for syntactic analysis. While a deeper a~~a l y s i s Otle 11se of trace is to indicate the untleriying position of tlie wh-heacl of a question or relative clause. Tl~us, tho structurr! built by the parser for 3,la would iticlucle tlie trace sl~own in 3.lb, witlFtlle trace's binding shown by the line utlcter the sentence. The position of the trace inclicates tliat 3.1 a has an underlying structure alialogous to tlio overt surface structure of 3 . 1~.Aliother use of trace is to indicate tlie underlying j-lositiol~ of tlio surfnce sttbject of a passivized clause. For cxeml>lo, 3.2a will be parsec1 into a structure tliat includes a trnce as shown as 3.2b: this trace indicates tliat the s\t\,ject of the passive hns the underlying position shown in 3 . 2~. Tlre synlbol 8tVt1 signifies the fact that the subject position of (2c) is fillecl by all NP that clominates rto lexical stritcttlre. (Following Chomsky, I assume that a passive sentence in fact has no trnderlying subject, that an ayr!ntive "by NP" prepositional phrase originates as such in trnclerlyiny s tructurc ) The trace in (3b) inclicates that the phrase "to be happy", wllicl1 the brackets show Is really an ~mbecicleci clarrse, has an underlying subject wl~lch. is iclcnticcll with tlie surface subject of the matrix S, the cibltse tlrat clomiriates the embedded complement. Note that what Is conceptually the underlying subject of the enihatltled clarrse tias been passivized into subject position of tlin matrix S, a phenomenon commonly called "raI~Ing@~ The ntiaiysis of this phenomenon assumed here derlves from [Cl~onrsky 733; it is an alternative to the classic analysis which involves "raising" tlie subject of the embedded clntrscr illto object position o f tlie matrix S before ptlssivization (for details of this later analysis see [Postal Tho Passive Rule in this section ond the next, I will briefly sketch a so111 t ion to tlie plionomena of passivization and 'tralslngtt in tlia cotitoxt of a grammar for PARSIFAL. This section-will prcscnt tlio Passive rule; the next section whL sliow how this rille, wltlrout alteration, l~anrlles the ttralsing@g cases.Let 11s uegln with the parser in the state shown In figr~ro 4 below, in the miclst of parsing 3.2a above. The analysis process for tlie sentence prior t o this point Is essentially parallel to tlie analysis of any simple declarative wllh o11c exception: the rule PASSIVE-AUX in packet BUILD-AUX has clococlecl tile passive morphology In the auxiiia.ry atlcl ylven the auxiliary the feature passive (although this fciaturc? is not visible in figure 4) . At the point we begin our example. tllc packet SUBJ-VERB is active. Figure 6 -After PASSIVE has been executed. Now rilles will run which will activate the two packets SS-VP and IRF-COMP, given that tlie verb of VP17 is 18sclisclule11. These two packets contain rules for parsing simple olljects of non-embeclded Ss, and infinitive coniplements, respectively. Two such rules, each of w l~i c h utilize nn NP immediately following a verb, are given in figtrre 7 Irelow. ?lie rille OBJECTS, in packet SS-VP, picks Up an NP nftcr tlre verb and attaches it to tlie VP node as a simple object. Tlie ruls INF-S-STARTI, in packet INF-COMP, ttmigr_lors when an NP is followecl by @'to1' and a tenseless ver0; it il~iliates an infiliifivo complement and attaches the NP as its strbjoct. (An example of such a sentence is "We warrtecl John to give a seminar next weekg1.) The rule INF-S-STARTI must have a higher priority than OBJECTS beceuse tile pattern of OBJECTS is fulfiiied by any situation that fulfills tlie paLtern of INF-S-STARTI; if both rules ate in active ( x~c k o t s arirl match, the hlglrer priprlty of INF-S-START1 will cause it to be run instead of OBJECTS. here in cletoll, llote that the rule OBJECTS will trigger with tlrc parsor-in tlie state shown In figure 6 above, and wlll atttrclr NP63 as tlro ol)jt?ct of tlie vorb wschedule. OBJECTS is tlrrfs totally Indifferent both to tlie fact that NP63 was not o rngular NP, Dilt rather a trace, and the fact tlrpt f i~5 3 t l i c l not orlgitiata i t 1 tlio i n1,ut string, but was placed into the b i~l l o r by grammatical processes. Whetlier or not thls rule is executecl, is al~solutely unaffected by differences I~etween an actlve sentence and its passlve form; the ntrolysls process for e i t l i~r Is Identical as of thls point In the pnrsinq process. Thus, the arrelysis process will be exactly i~rtrnllal in botli cases alter the PASSIVE rille has been exac!~teci. ( I remind tlie reader that the analysis of passive ass~~nrocl EIOOV~, following Cliomsky, does 170t assume a Irroccss of "agent cioleti~n~~, "subject p~s t p o s i n g~~ or the like.)Passjvcs in En~bodded Complements. -"Raising"Tlie rcnclbr may have wondered why PASSIVE clrops the trace it creates into tlie buffer rather than InrtiiaclioZely attacliing tlie new trace to the VP node. As we will sac below, sirch formulation of PASSIVE also correctly analyzes passives like 3.3a above wl?ich involve nraislng", btlt with no arlclitlo~~al complexity added to tlie grammar, correctly capturing an Important generalization about t l i s l . To sliow the range of the generalization, the exfin~ple wliicli we will investigate in this section, sentence (1) in fig~rre 8 bclow, is yet a level more complnx than 3.3a ahovc; its analysis is sliown sclienratically in 8.2. In this cxtrml>le tliore ore two traces: the first, tlie subject of the entl~cclclecl clause, is bound to the subject of the major clause, tlie seconcl, the object of the embedded S, Is bound to the first trace, and is thus ultimately Qoirnd to the s u l~j c c t of the liiglior S as well. Thus the underlying position of tlie NP "the meeting" can be viewed as being t l~c ohject position of the ombedded S, as shown in 8.3.( 1 )'Tlic lneeting was believed to have been scheduled for Weclnestlay.(2)Tlie nieetilig was believed [s 1 to have been scheduled. t for Waclnesdq)r](3) V believecl Ls V to have scheduled the ineetlng for Wedtiesday J. W e bedin our example, once again, right after "believecl" has been ottached to VP20, tlie current active ~)ocle, as shown it1 figure 9 below. Note that the AUX node has been labelled passive, although this feature is not sliown here. Again, rules will-now be executed which will activnte the packet SS-VP (which contains the rule OB.,IECTS) allti, since "believe" takes infinitive complements, tlie packet INF-COMP ' (which contains INF-S-START1 ): among others. (These rilles will also deactivate tlie packet SUII,I-VEFll3.) Now tlie patterns of OBJECTS and INF-S-START1 wlll both match, ancl INF-S-START1 ,*shown above In figure 7, will be execitted by tlie interpreter since It has tlie liigiier priority. (Note once again that a trace is a perfectly nornial NP from the poiht view of the pattern malcliingl process.) Tliis rille now creates a new S node lal~ellecl Infinitive and attaches tile trace NP66 to the new infinitive as its sui~ject. The resulting state is shown in figure 1 1 below. We are now well on our way to tlie desired analysis. An en\lrecldeci infinitive has been initiated, and a trnco bdunci to tlie subject of the dominating S has been attacl~oci as its suhj8ct. although 110 rule has explicitly itlowsr~rlll tlie trace from one clause into the other.The parser will now proceed exactly as in the previous exa~iple, It will the aux~li~ry. attach It, and attach the verb "schetluled" t o a new VP node. Once again PASSIVE will match and be executed, creating a trace, I~i~irling it l o tlie subject of tile clause (in this case Itself a trace). alicl clropplng the new trace Into the buffer. Again the rule OOJECTS will attach tile trace NP67 as the object of VPZ1, and tlie parse wlll then be completed by clranimatical processes which will not be discussed here. An eclilteci lorin of tlle tree structure which results is shown in figure 1 2 below. A trace is indicated in this tree b y grving tlie tcrnlillal string of its ultimate binding in parentheses. This example demonstrates that the simple formillation of the PASSIVE rule. presented above, interacting with other simply formulated grammatical rules for parsing objects and initiating embedded infinitives, allows a trace to be attacliecl either as tlie object of a verb or as tlie subject of an embedded jnfln!tive, whichever is tlie appropriate analysis f.or a given grammatical situation, Recot~se tlie PASSIVE rule is formutated in such a way that it clrops the. trace it creates into the birffer, later, rules, already fornlyiated to trigger on an NP In' the'buffer;. will analyze sentences with NP-preposing exactly the same as tlioso witliout a preposecl subject. Thus, we see that .the aviriiabiiity of tlie buffer maolteinism is.crucial to chpturlng tliis generalization; such a generalize'tion can only b e s\atr?cl I,y a parser with a mechanism much like the buffer usacl hare.Before ti~rning now to a sketch of q computatlonal account of Clloniskyls constraints, there are several illlportnht liniitations of this work which must be enumerated.First of all, wliile two of Chomskyls constraints soon1 to fall out of the grammar interpreter, there seems ttl Ire no apparent accoiltit of a third, tlie Propositional Island Constraint, in terms of this mechanlsni..Seconcl, Cliomskyts formutation of these coristraitits is irltended to allply to all rules of grammar, both syntactic rules (i.e. transfbrmations) and those rules of seninntic interpretation wliicti' Chomsky calls "rules of c d n s t r~a l~~, a set of shallow senientic rules whlch .govern A~ipplioric processes [~llomsky 773. The discussion here wi(l only tolloh on purely syntactic phenome~ia; the question uf,j~ow rules of 'semantic interpretati~n can be meshed with t l i a fr-aniework presented in this clocument has yet to be investigaterl.'Tliircl, tlie arguments presented below deal only wirlli €nglish, ancl in fact depend strongly upon several facts about Etiglisl~ sflitax, most crucially upon the fact that English i s sul~ject-initial. Whether these argu.ments can be successfully extencled, to other language types 'is an open cluestion, and to this extent tliis work must b e considered exploratory.Ancl finally, I will not show tliat these constraints must bo k u e withoyt exception; as we will see, there are vari-ous situations in wliich tlie .constraints imposed by the gratnninr interpreter can be circumventecl. M O S~ of fliese situations, tliougli, will be shown to demand much more coml)lex grammar forniulations than those typlcally needed in the' flarnaer so, far constructed. This is quite in keeping with tlie suggestion made by Chonisky [Chomsky 77J that tile cotlstrai~its are not necessarily without exception, but ratlier that exceptions will be l'liiyhly marked" and therefore will caulit heavily against any grammar that includes them.The Specifiecl Sul~ject Constraint (SSC), stated informally, says tliat no rule may involve two constituents tliat are Domiriat&d by different cyclic' nodes unless tlie .loweraof the two is tlie subject.of .an S or NP. Thus, no rule may involve constltuents X .and Y in the structure shown in ?igae 13 below, i f and p'are cyclic nodes and Z is the si~bjcct of a, Z distinct from X.No rille can involve X wd Y in this structure. Tne SSC; .explains why tlie surface subject position of vcrl>s' like and lais certahfi' which have no irnclor!ying subject can bo flied only by the subject and not tlie object of tlie enibsdded S: Tlie rule '"MOVE NPgl is free to shift 'ouy NP into the empty-subject -position, but is constrainecl by tlie,SSC so tliat the object of the embedded S canno'f'be nlovktl out bf .hat clause. Thls explains why (n) in figure -1 esseiice, -tlien, the Specified Suljject Constraint bd~istrai~ls tlie rule "MOVE NPu.in' s l~c h a way that only the si~l?ject' of a clei~se con be moved out of that clause into a position in n liigl~er S, Tl~us, If a trace in an annotated stlrfaco structure is bouncl to an NP-Dominated by a higher S, l l~n t iracc mtist fill the subject posiiion of the lower clause,In the reniaincler of tljis s e~t i o n I will show tliat the oramnrnr interpreter constrains grammatical processes in strcti a way that a;iriota ted strrface structures constructed Ily Ilie granlmilr interpreter will have tlits same property, givwi tlic formulation of tile PASSIVE, rule presented-above, 111 tcrais of tlie parsing process, tliis means tliat if a trace Is " I o w~r e c l~~ from one clause to another as a result of a "MOVE NP"-tyl)e or>era tion (luring the parsinq process, then it-will he attacliecl as the subject of tlie second clause. To I)e wore precise, if a trace is attacl~ed so that it is Donrinated by sonie S node S1, and tl1.e trace i s bound to an NP Doniinatecl by sonie other S node S2, tlien that trace will aocessahly bo attached so that it fills the suGect position of S1. This is depictacl in firlure 16 below. This statement of PASSIVE does more, however, tliali slmgly cegtilre a generalizhtlon about a specific construclion. As I .wIII argue it1 detail below, tlie behavior spocifled by both the spocihed Subject Constraint and Sirbjaconcy follows almost immediately from this formulatlon. In [~& c u s 7 7 1~1 BrgUe that tlils formulation of PASSIVE Is the otily simple, tion-ad hoc, formulation of tlils rule possi,ble, dnct tliat qll other rules characterized by tlie competenca rule "MOVE NPtt ~r u~s t operate slmllarly; here, however, I will olily show tliat these cotistrainta,-fallow naturally. from thls formulation of PASSIVE, leaving tlie question of necessity asicle. I will also assume one additional consl~aint below, the Left-to-/light Constraint, which will be brlefly motivated lator in tliis pnper as a natural conclition on the formulation of a ciramniar tor tlils nieclianlsm.Tlie Le,f t-to-Right Constraint: tlie constituents in the buffer aro (almost always) attached to hlglier IeQel constittle~its In left-to-rigl(t order,. 1.e. the first constitl~ent in the buffer is (almost always) attncliecl before tlie seconci constituent.I will now show tliat a trece created by PASSIVE which is bounci to an NP in one clause-can only servd as the subject of a cla11se.dominated by thaf first clause.Given tlie formulption of PASSIVE, a trace can be t'lowerecltt illto one clause from another only by'the indirect route of clropping it illto tlie buffer before the subordinate clause node is created, which ig exactly how the PASSIVE r111n operates. This njoans t l b t the orderlnp of tlie operations is crucially: 1) create a trace and drop it into tho buffer, 2) create o li~borciinate S node, 3) attach the trace to the newly creotecl S, node'. Tlie key pmt Is that at tlie ti~iie that tlie sili~orclinate clause node ig created and boconios ihe current activtFnode, tlie trsce must be ~sitfjng in t11e' l 111f fer, filling one of tlie tliree buffer posltlons. .Tliust tlie pacser will be in tlie state -shown 1n.figure 16' below, with tlie trice, 'in fact, most likely In the first buffer position. Figure 16 -Parser state after embedded S created.Now., given the L-to-R Constraint, a trace which is ill tlie hi~f'feroat the time lliat an en~bedded S node is first creatccl mtlst be one of tlie first several constituents attacl~ecl to file S node ,or Its daughter nodes. From tlie structure of Engli.sli, we know that the leftmost three co~istituents of an enlbedded S node, Ignoring, toplcalized constitrrents, milst .either be COMP NP AUX or NP AUX [", VERB ,.. 1.(Tlie COMP node will'clominate flags like lithatll or' llforll that mark' tlis beglnnlng-of a complement clause.) But then, if a trace, itself an NP, is one of the first 'several constltuents attilcliocl t o an leml)odded clause, tlie only position It can fill will b e tlio subject of tlie clau'se, exactly the empirical c o~i~e y a e n c e of Chomsky8s Specified Subject Constralnt in s i t c l~ p s e s as explnl~iecl .gbove,The L-to-R Constraint Let IJS now return to fhe motivation'for the L-to-R Constraitit. Again, I will not attempt t o prove that thls colistrelnt hyst be true, but merely tq,show why it Is plausible.Enr~~irically, tlie Left-to-Right CCfiTstraint seems to liolcl for tlio triost part: for the grammar of Enyllsh discussed in this I)aper, ancl, it woulcl seem, for atiy grammar of English t l w t attempts to capture tile same range of generalizations as tliis (jramm~r, tlie constituents In the buffei are utllized i;i Inft-to-riglit orclor, witli a small range of exceptions. Thls usiige is clearfy not enforced by the .grammar lnter~reter as proJeiitly iniplen~ent?ee(i; it'is qulle possible to write a set of graniniar rirles that specifically ignores a constituent_in t h e buffor until some arldtrary point h the cl&se, t h o u g h~u c h a set of rilles would be lilghly ad'hoc. However, there rarely seenis to .he a need to remove other than the ' first constituent in tlie buffer.The one B.xception to tlie L-to-R Constraint seems t o Ile that a constituent Ci may be attached 'before the constituetit to* left, C ,I,'if Ci does not appear in surface structi~re in its underlying position (or, if one prefers, In i t s ynniar,kecl l>osItion) and if its rkmoval .from the buffer reestablishes the unmarkec! order of the -remaining constituents, as in tlie caserof tlie AUX-INVERSION rule clis.cussecl~ earlier in tliis paper. To capture .this notion, the L-to-R Constrairit can b e rebtated as follows: All mnstitrrents niilst be attached to higher. level econstituents accorrling'to the .left-'to-right order bf constltuents in the unsarkecl case of tliat constituentls structura.Tliis reformirlation is interesting .In that It would be n natural consecluence of the opereflon of tlie granimar intcjrpreter if packets were associatecl with the phrase strtrct41re rules of .an explicit "base componentI1, and'these r~t l e s \ were used as templates to build up the structure assignccl by tlio gramtiiar interpreter. between Y bticl X) then Y Domlnntes X.Tlre principle of Subjadeticy,' informally stated, says that no rulo can involve constituents that are soper'e'tecl by more .than one cycllc*node. Let us say that a nocie X i s sublacent to a node Y i f tliere is at most ope cycilc nocle, i.e. at niost one NP o j S node, between the cycilc node tliat Don\iliates Y and tlie node X. Given this clatmition. the ~u l l j s c e n c~ prlnciplk says tliat no rule can involve coristltuents tliat are not subjacent.Tlie Subjacency principle l~nplies tliat movemerit .tiles qre co~istrai~iecl so tirat* they can move a constituent ~n i y into positipns tliat tlie' constituent was subjacenj to, i.e. o~ily witl~iri .the clei~se (or NP) in which it originates, or No rille call involve X and Y in this structure.Subjacency inlplles tliat if a constituent is to be "lifted" up more tlian one level in constituent structure, this olioratioti nrtrst. be clone by relreated operations. Thus, to use one of Clio~lskyls exaniples, tlie sentence given in flguro 18a, with a cleep structure analogous to lab, must be tleriveci k s follows (assuming that "is certainl1, like has no subject in uncierlyiny structure): The ,deep structure nlrlst first u~iclergo a niovemeiri operation that results in a ~b i t c t u r e ~nalogous to 18c, and then another movement operation illat results in 18ci, each of these movements leavilill a trace as sliown. That 18c is in fact en intnrniecliato structure is si~pported by tl;e existence of sentmices sucli as 18e, which purporteclly result when the V ill tiie matrix S is replaced by tlie. lexical Item "itvg, and the enil~ecicied S' is tensed rather tlian infinitival. Tlie strttctirre given in 18f is ruled out as a possible annotated sirrface structure, because tlie single trace could only be left if tlie NP was moved in one fell swoop'from'lts untlerlyiti~~ pohition to its.position in surface structure, which woi~lcl .violate Sub/o'cency. IJavlhg statetl Subjaconcy'ln terms of tlie dbstrgct con~potinca tliaory of generative grammar, 'I now will show that a, parsing correlate of St~bjacency follows from ill8 strl~ctitre of tlie. grammar interpreter.. Speciflcaily, I ill slinw tliat tliere are. only .,limited dases in wlilch a trace qeheratotl by a "MOVE-NP" process can be ttiowered; inotg tliaii olio cleuse, i,enqtIiat a trace cieated and bound while any given S is crirrent niust almost always be ettached either tb tligt S or to an.S whlcli Is dominlted by that S.Let us bcglli by exeririning what it would mean to lower a trace more tlian one clause. ~i v e n that a trace can otily be 1810wereclfl by dropping it lnto tlie buffer and tlien croati~ig a sul~orilinhte S node, as discussed above, low&incj a trace nrore tlian one clause necessarily implies tlie -follnwinf~ sequence of events, depicted in figure I 9 below: First, a trace N P~ must (a) be created with some S F i y m 10-Lowering a trace more than 1 clause Dtlt tilis secluence of events is highly unlikely. The essence of the nrglrlment Is tli-is;Nothing in llie buffer can change between tlie time that S 2 is createcl and S3 Is cieated if NPI remains In tlie htrfler. NPI, like nny other nocle that is dropped from tlie active liorls stack illto tile buffer, is lliserted into tlie first Ilulfcr position. But tlien, by tlie L-to;R Constraint, notliing to tI.1e riglit of NP1 can be nttacliecl to a higher level constit'irent utiril-NP1 is ~atiaclidd. (One can show illat it Is ~~i o s t i~nlikclyttliat any constitu6nts will. enter t o the left of NP1 after it is clrol)ped illto tlie buffer, but 'I will .suppress tliis ciotail here; tiie 11111 argumerit is included In [Marcus 771.) 011t if tlte contonts .of the buffer do 'not change between the creation of.S2 and S3i then what.can possibly nintivirte the creation of both S2 and $33 The contents of Hie i j~f f or must necessarily provlde clear evlclerice that botli of these clailses are present, slhce, by the ~etertiiinism tlypothesls, the parser must be correct if it ililtint&s n canstititent. Tiius, the same-three c~ns~kltuents In tiin !)trffer ~attst provide convincing evidence ndt only for the creation of S2 but also'for S3. FurtiTBhnore, j'f NP1 is to becomk athe subject of 83, elid if S 2 Dominates 53, then it woulcl Seem tliat tiie constituents that follow, NP1 In the buffor nirist also he constituents of S3. since ~3 must be ,coni~~ieteci bdoro it is dropped from tiie activg node stack anrl 'constititetits can then be attaci~ei'to 52, But then 52 must be crnnted entirely on tile bast3 of Wfdence ~rovided by tile cotislitile~rts of.andther clause (unless S3 has less than Iliree Constituents). Tliils,lt would seem that the cbntents of tile i~irffer cannot' provide evidence for tile presonce of both clauses itnless the presence of'S3, by itsclf, is cnou(~lr to provide confirnring evidencd-\for the i,resencaof S2. ~i i l s woidd be the case only if there were, say, a clP~tsel c~nstruction. t k t cpiid:~nly appear (perhaps i t i n garliciil6~~environ?j~nt j 'as ,tile lliltial 'constituent of a tiiglier claaso. in this. case, If there Pre such constructions, a vioio t loti of Sui~jecency a tiauld be ibi,s~bie.Wilh. the one exception just mentioned, there is no molivcltion for crcati~lg two clauses in such a situation, and thus the iliitibtion of only one such clause can be motivuty?d. But if only one cldiise is Initiated before NPI is attached, tlicn Ni)l muat bc attaclietb to tliis clause, and this clause is ~~ecpssarily subjacent to the clause which Dominates the 'NP Lo wliicli it is i~ou~id. Thi~s, tlie grammar interpreter wlil behave as if i t enforces the Subjacency Constralnt. ' As a concltrdingl point, it is worthy of note that w\iile the gronrmar interpreter appears to beliave exactly as if it wRra colist~ained by the Subjacency principlei it is in fact constrainotl by a \;crsion of tlie ~laiisemate constraint! (The Clauseniate Constraint, long 'tacitly assumed by linc_~irists but first explicitly stated, I believe, by Postal [Pnstal'64] , states that a transformation can only involve constittrents tlint are Dominated by the same cyclic node.Tliis constraint is i t tiie heart of Postal's attack bn the constraints that nre discussed above and his argument for'a ;'mising" nnnlysis.) The yrehnor interpreter, a* was stated above, litiiits gramnier rules from examining any nocle in the. active nodc stack hiyl;er tiionathe current cyclic hade, wtiic.1; is to say that it call only examine clausemates. The trick is'-that a trace is created and bdund wliile it is a "clauscnr~tc~~ of the NP to wlijch it is boirnd in tiiat the citrrent cyclic~nocie at tliat time Is the nocle to which that NP is atteclied. Tlie. trace is t i ; 6n dropped into the buffer a~i r l &iotlier' S ilotle is created, tilereby' destroying. the clot~scniatc relationslii~~, * The trtice Is then attached to this new S . ~mcle. Thus, it1 a sense, the trace Is lowered from one,clpuse to another. Tl?e.,crucial point is that wiiile tlfis 1 lowering goes on n s a .resuit of the operatlon of,the graminar. i~;tbr~,rct~ri it Is olily lmdcitly lowered in tiiat 1) the trace' was never .atlaclied to the' higher S and 2) it isanot dropped irito the bi~ff.er .because of any reoltzatlon that it. m u~t be "1owDroc1~~; -in fact i t may end upattached as a clausemate of the NP to whicli it is boundas the passive examples preqenteil earlier make clqar. Uie trace is simply dropped into tlie buffer because its grammatical function Is not clear, anti the creation. of tlie second , S . follows' from other Indnpondon tly n~ollvated gramma tics1 'From the poilit of view of tliis orocessing ,tliciory, we. can have our cake ant1 eat i t too; i o the exteni that it makes sense to ninp .resuits from tlie realm of processing into the realm of ~ompotcnce, in 'a sense both the clausemate/'~relslngt~ and tire Subjacency positions are correct.111 closi~lg, I would like to sliow tliat tile properties of t h e grnnimar interpre'ter crucial to capturing the behavior of ~liomsky's constraints were orlglnaliy motivated by the Determlnism Hypdhesis, 'and thus, to some. extent, the Deterniinisni t~ypotl~esis explain-s chomsky's constraints.Tlie strongest form of such,an argi~ment, of course, wotrlcl be to sliow that (a) *ittier (i) tlie grammar Interpreter a c c o i~~i t s for all 01 ~honisky~s. cbnstraints in a inanner which is coucli~sively universal or (ii) the constraints that i t will riot ncco~llit for nre wrong and that (b) tlie properties of the grauiiiiar interpreter wlilch were .crucial for this proof were forced by tiie Determinism Hypotliesis. if such an argument could ho made, it would -show that the ~etermlnlsm Hypotliegis provides a natural processing 'account of the ~i n~i~i s i i c ddte characterized by. chdmskyts constraints, giving strotip confirmation to ttie Determinism Hypothesls. I hsvs, shown none of tlie above, and thus' my clsibs niirst Be proportionately more modest. i have argued 0111 y that irnpor'taiit sub-eases of Chomsky1s constraints follow fiorn the giammar interpreter, bnd while I c m show tlia t tile Onter~nlriism Hypotl~esls strongly mo(lvaies the tiiecl,atiisnis froni wliicli these arguments folloui, I .cannot show hecesliity. The, &tent to .which tiilk argument provicles evidence for the ~eterminism.,Hy~othesis must thus be left to thd reader; no objective measureaxists for such matters.The ability to drop a trace into the buffer Is at the heart. ot tile arguments pre.sented here for Subjacency and the SSC as consecluences of the functioning o f t h e grammar inieroreter; tliii' i ; tlie central operation upon wtitbh the above br(~ilnietits are bosecl. But the'buffer Itself, and the fact thkt a constituent .can be dropped into the buffer'if Its pramnlntical' f i c t i o n is uncertain, are directly motivated by the ~eterhinism. HypdtMsJs. ~i v e n this, it .is fair t o . claim tlirr t if '_~homsky's constraints follow from. the operatlob of -. . . < tltc ranihe her interpreter, t hei i they are strongly linked to the Deterniinism Hypotl~esis, if , Ciiomskyls constraints. are In fact true,, then the arguments prgsented In this paper providq solid evidence in support of the '~eterminlsm Hypothesis.
Appendix:
| null | null | null | null | {
"paperhash": [
"marcus|a_theory_of_syntactic_recognition_for_natural_language",
"pratt|top_down_operator_precedence",
"winograd|procedures_as_a_representation_for_data_in_a_computer_program_for_understanding_natural_language",
"jackendoff|semantic_interpretation_in_generative_grammar"
],
"title": [
"A theory of syntactic recognition for natural language",
"Top down operator precedence",
"Procedures As A Representation For Data In A Computer Program For Understanding Natural Language",
"Semantic Interpretation in Generative Grammar"
],
"abstract": [
"Abstract : Assume that the syntax of natural language can be parsed by a left-to-right deterministic mechanism without facilities for parallelism or backup. It will be shown that this 'determinism' hypothesis, explored within the context of the grammar of English, leads to a simple mechanism, a grammar interpreter. (Author)",
"There is little agreement on the extent to which syntax should be a consideration in the design and implementation of programming languages. At one extreme, it is considered vitat, and one may go to any lengths [Van Wijngaarden 1969, McKeeman 1970] to provide adequate syntactic capabilities. The other extreme is the spartan denial of a need for a rich syntax [Minsky 1970]. In between, we find some language implementers willing to incorporate as much syntax as possible provided they do not have to work hard at it [Wirth 1971].",
"Abstract : The paper describes a system for the computer understanding of English. The system answers questions, executes commands, and accepts information in normal English dialog. It uses semantic information and context to understand discourse and to disambiguate sentences. It combines a complete syntactic analysis of each sentence with a 'heuristic understander' which uses different kinds of information about a sentence, other parts of the discourse, and general information about the world in deciding what the sentence means.",
"Like other recent work in the field of generative-transformational grammar, this book developed from a realization that many problems in linguistics involve semantics too deeply to be solved insightfully within the syntactic theory of Noam Chomsky's Aspect of the Theory of Syntax. Dr Jackendoff has attempted to take a broader view of semantics, studying the important contribution it makes to the syntactic patterns of English.The research is carried out in the framework of an interpretive theory, that is, a theory of grammar in which syntactic structures are given interpretations by an autonomous syntactic component. The book investigates a wide variety of semantic rules, stating them in considerable detail and extensively treating their consequences for the syntactic component of the grammar. In particular, it is shown that the hypothesis that transformations do not change meaning must be abandoned; but equally stringent restrictions on transformations are formulated within the interpretive theory.Among the areas of grammar discussed are the well-known problems of case relations, pronominalization, negation, and quantifiers. In addition, the author presents semantic analyses of such neglected areas as adverbs and intonation contours; he also proposes radically new approaches to the so-called Crossover Principle, the control problem for complement subjects, parentheticals, and the interpretation of nonspecific noun phrases."
],
"authors": [
{
"name": [
"Mitchell P. Marcus"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"V. Pratt"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"T. Winograd"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. Jackendoff"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null
],
"s2_corpus_id": [
"6616065",
"12490113",
"54114373",
"61367317"
],
"intents": [
[],
[
"background"
],
[
"methodology"
],
[
"background"
]
],
"isInfluential": [
false,
false,
false,
false
]
} | null | 554 | null | null | null | null | null | null | null | null | null |
92b82359436b66a51b91c3cf051c38848fb535cc | 17773952 | null | Remarks on Processing, Constraints, and the Lexicon | HKR1 A HKS O N 1'UOC:ESSl N(;, CONS'I'WAIN'l'S, AN11 1'111< LI'SIC'ON* TIlollii~s Ikrnsow Stnriford Uliivcrsity l..inguists hi\ve long recopnizch the desirnhilitj'of cmbqdding a tliu~ry 01. ' graliilnai withill tr tlicory ol' t inguistic pcrf ormanc* (scc, c,g., C l ~o m A y .(19G5;10A1 5)). It lias bccn widcly assulncd by transformationalists that an adcqui~tc' niodcl of ',a language | {
"name": [
"Wasow, Thomas"
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 31 | 5 | null | raised, 1 ittle support cotild B e idduced for the hypothesis that thc oparations of transforn~ational grammar play .a part in speakers' or hearers' processing of scntences.(see Fpdor, et a1 (1974: chapter 5) ). lnitead of concerning themselves with qr~cstions of processing,. tmnsf'ormntioni~lists have concentrated their efforts ( i~t least in the l i~~t dt.~iide o r so) on the problem of constraining thc k w c r of their thcory. T , k goal of much recent research has becn t o construct as. restrictive a theory of granlmar as possiblc, within the bounds set by the known divcrsits of human latigungcs (scc. c.g.. Ross (1007), Chomsky (1973) , Bresnan (1.976) , Emonds (1976) , and Culicover and Wexler (1977) for examples .of this .type of research).Computational linguists, on the other hand, have not e$plicitly concerned rhenisc.lves very a,mi~ch with the probl'etn of constraints (but see Woods (1973; 124-5) 'for an exception), Ilatlicr, their goill has bccn to l'iltd effective procedures for the parsing and proccssitig of natural Ianguagc. Whilc this is implicitly a . rcstricti.on to rcc~~rsive languages, the computational literature has dealt more, with questions of processing . than with how to limit the class .of available grammars o r languages.In previoils papers (~s h e r s o n and Wasow (1976) , Wasow (in press a. 19711)) 1 have argi~cd for the legititnacy of the quest for constrilints ,as a rescarch strategy. 1 have argued that a thcory that ,plrrccs litnits on thc class of possible languages n1ah.c~ significant empirical clai~ns a b~u t --l himan .mental capacitics, and can contribute. to a solution to "the fundamental empirical problem .of linguistics" (as ~h o m s k y has cullcd it) of how cliild ren .are able .to '!car11 languages with such' facility. I have tried to show that such psychological cl.iir~ls Ciirl, LIL' ~~l;ldc, ' i t l~o i~t I I~: \~~I I L ; all) ; t b : t~~~l~ j~t i~l i s i~b~~t what rolc gtammqs play in pc.~~orariulcc. In short, I have a r g~~c d that a tlit'ory 01' gra~linii~r Cat1 ~liak'c sig~lificant col~t~.ibiitions to psycholbgy, i~rdcpcndet~t of tllb ilnswcr to the granli~iatical. rc.i:ill izatioa ~~~o b l c t n .Rccyit work by Joi~tl Bres~l;tll (in prcss) takcs a vory diffcrcnt positiun: s,hc 1135 S L I~~C S L C J illat tratibfor~ii;lti~l~ali~t~ O L I~I I t to pay morc attention to thc gram~natic:~l rcaliz:~tion problem, and thitt .considerations of processing suggest. ' radical modifications in 7he theory of' transfortnational grammar. Furthcr, she argues that thcrc. is-st~iplt: pra~nliiatical evidcnce for tlrcsc modificatioos. In this piryer 1 will suggest solnc dxtcnsions at' hcr proposnlq and will explore somc of their empirical cotiscqucnces.Furthe& 1 will argue that her I'rrrnlcwork ~iiiikcs it possiblc tu imposr: riltllcr rcstrictive consfrir inls vn 'gr;rlrin~i~t icsl tt~cory. 'i'hils. 1 will ilrgllc that the grrtmmaticnl rcaliration problcm and the prablcm of constraitiing trun~fr\r~~liltio~lal thcory, wllilc lo@ally independent, are both addressed by Bcesnan's proposals. If I am correct in this, then L)res~ian's "realistic tratlsformational grammar" rcpiescnts a major-conucrgcnce oi .the concerns of transfor~natiorid and compu~atiorri~l lingnists.My presentation , will consist of three parts. First, I . will briefly sketch Brr'snan's framcxvork. Sccond, I. will suggest some extensions of her proposals and point out .$erne conse'q~icnces .d these e.xtcnsions. Third, -1 will prdpose how her framewo'rk can be constrained, and indicate certain desirable consequences .of m? proposals.The primary innovation of Bresnan's framework is &hat. it eliminates . a large 'class'af transformations in favor of an enriched conception bf the lexicon. The grammar that results is one that Bresnan claims is . f a r more. realistic from a proccsging point of view than other versions. of transformational grammar. She points out striking similarities betwczt~ her proposals and .recent ' conip~rtirtiunal and psycholinguistic work by KapIa11 and Wanncr, and shc argues that Augmented Trr~nsition Nctworks can provide a t least a' partial answer to. the .grammatical realization problem within her framework. now sketch very roughly what Btesnan's "realistic" transt-ormational erammar .is like.. . Rules like passive, dative, and raising rules, which are "structqre-preserving" (in the sense that thcir. outputs are st~ucturally' idei~tical-to indcpcnddntly' rcquircd base-generated structures) and "local" (in tlic scnsr: that the :elein$nts affectcd arc' always in the immediate cnvi'ronni~nt of some governink iexica~ item, ~a o s l i y a . vcrb): are climinutcd from the tritnsforrnational cot~lponent atid relcyutcd to tlic lexicorl. .Lexical entries include, among other things, (strict) subcatcgorization frames arlcl o c absll.~c;l rc~u.usc111.11iyns N 1kic.h I31-cs11an ~.i~lls "f~111~ti011ill S~~U C~L I~C S " ur '*prcdicatc arg11IIicnt structurc9*'. S~bccttc'porizatio~~ 1'ranl~'s' give t l~c sy~itactic e h v j r~t~f i~c~~t s ' in which thc luxical i'tclii lilay appuar: thcse arc cxpresscd in terlas o f i1 hiisic s t grarli~i~aticnl relations, includi(g "sut~ject" a11.d "objcct". 'i'hcsc notiotis. w l i~l r l~nivrrsi~l, are inst;~nliatcd dif'fcrclitly i l l diffcrc~\t Ii\ng.ll;~gcs; for cxamplc, nrcsllnti tt1kt.s c~&nti:llly thc s t r u c t~~~a l dcfi~iitiorls of "subject" . . and ''nhject" prmposrd hy ('hornsky (19115; 71) as langr~ngespccific charilctcri;r.a~io~ls' 01' thcsc notio~ls for English. Futlctionitl structure3 give a .more abstract reprpscntation of tlic elements mcngioncd in the sub~;ltcgoriz;ltion frame, indicating what thcir-"1ogic;lll' rc1;ltionships arc, 'I'hus, thc functional structure correswnds very roughly to the deep structure in the standard theory of transform;iti.onal grammar, and the subcategorization frame corresponds even more roughly to the surface structure.What the standerd theory did with local structure-preserving transformations' Bresnan can do in either of two ways. ~elati'otrships like active/passive are handled by positing two separnte lexical~'entries for active and passivg verb forms. The product\uify of this relationship' can be accounted for by means of a lexical redundancy rule, which would say, in effect, that corresponding to the typical...transitive verb there is an intransitive. verb which looks morpbobgically like the perfect form of "the transitive, and whos'e subject plilys the sitme Iqical role ( i .~. in the functional structure) as the object-of the tra~sitive verb. ~resnan's other way of replaeink local structure-preserving rules is illtatrated most clearly with-the raising rules. Raising to object pcwitioh, for cxarnple, is used to capture the fact that ihe NP which' is syntactically the object c4 one clause is logically not an argument of that clause a1 all, put 4 subject of the sutsordinate'~clausc. Bresnan expresses this simply in terms of the rcli~tionship between the saategorization frame and the'functional structure; that is, the object of the niain clause plays no role in the ful\ctional structure of that clause, but is "prrhsed down". to play 3 role in the next clause down. In the interests of brevity I will not. illustrate Bresnan's framework here. Rather, I will refer the interested rei\dzr to her paper, and go on to indicate my reasons for seeking to modify hl:r proposals.My primary motivittion comes from some earlier work of mine (Wiaow (1977) ). which argued agaiqst the climirta&iotl'of local. strrtcture-priscrving trnnsformations. My qrgumcnt was based on theobscrvation that there are two similar but distinct classes of linguistic relationships whose differences can be expressec) rather nilturally as the differences between tra~rsformational rules and lexical redundancy rules: ' The clearest cxatnple of this is the English passive. ItAss often been suggested that some passive pafiiciples arc aljectives and others verbs: I .pointed out that adjectival passivcs ~~t i d verbal passives differed in certain systemiltic ways. M y c e n M claim was that the surface subject of adjectival passives was always the deep direct object of the correspoiding verb. For example, a passive participlz which is demonstrably adjectival . (e.g.,.because it is prefixed with un-or imnediately follows seem) may not have as its. surface subject the "logical" subject of a lower clause, the indirect object, or a chunk of an idiom: *John is unknown lo be a cornmurist *John seemed told the story, *~d v & r u~e seemed taken o/ John,. A verbal passive, in contrast, could have as its subject any . N P .which could immediately follow the corresponding active verb: John is .known to be a communlsr, John was told rhe story, Advantage was taken of John This, I claimed, would foffow from the hypothesis that adjectival pssives are formed by a lexical redundancy rulc, whereas verbal . passive .are tr~nsformatlonally derived, if lexical redundancy rules arb "relational", in the sense that they are formulated in terms of grammatical relations such as subject and object, whereas transformations are '"struttural", i.e., .they areoperations on phrast! structure tree.It is evident that my earlier position is inconsistent with Bi_esnanls recent proposals. My estensions of her ideas, developed in collaboration with Ron -Kaplan, are'in part an attemipt to capture within her framework ihe distinction -my earliti paper sought. to pxplicatc in terms of the lexicon/transfotmation contrast They are also motivated by the very interesting comments of Anderson (1977) . *~ride-son suggests tliat 1 -was mistaken in -claimin8 that the operative factor in formulating rules like the adjectival. passive r u l~ wag the deep grammatical relation .of the surface subject Rather, he argues, it is thematic relations like "theine". "aeent*'. "goal"and "source-" (see Gruher (1965) and Jackendoff (1972)) which are 'erucinll. Asstiming Andersoh to' he correct, an obvious 1n4qification df Nesnan's systerfi svggests itself,' which would permit the distinctions of my earlier paper to be captured. Let us supposc: thath the functiollsrl structure in,lexical 1:ntues is-a specification of which thiniatic relations should be assigned to the eleme~rts mel!tiorixJ in the subcategorization frame. Then we may distinguish.two types of lexical rules: those that make reference to thematic relatiqns and thdse that. do not. The former would correspond to rules that nty earlier paper called lexical, and -the I9tter to those ihat ' l galled transformations, 'This is the extension of Brcsnan's fratilework that I wish To propose. I will illustrate ky formulating the two pa:;sivc.rules and the dative rille and applying them to a fragment of the lexicon of Engli&.kly formalism is based on the ..rtssrrmplion that the grilfnn~alicsl relations are eivcn latrguagc-wid$ definitions in struct~~ral terms (at !cast in English) along rhe lines in(licnted by Dresn~n. and that a verb's st~bcr\tegori;ration frame inrt.cly itidicates which relations it has, and vrltat grammatical categories lhnso relations are assigncd to. . (Thus, 1 differ from Bresnan in this .respect, for she assumed that. grammatical relations would be limited to NP's). I will adopt the fb\\owing abbreviations: "SS" 2 (surface) subject: SO = (surface) object "S02" ='(suiface) second object; "1" = tl~eme: "2" o agent; "3" t goal; "4" = complemenl The rule forming ,erbal passive participles from the corre~po~ding active lexical entries can now be formulated2 quite simply as SScSO. This is to be interpreted i s follows: eliminate "SS" wherever i t apjears in lhe entry for the active verb (eliminating also au)' assignment it may have to a thematic relatibn) and change all occurrences of "SO to "~5"~. The adjectival passive rule Will differ from this in that it ha? an additional condition on it: if SO=1, then SScSB: This condition insures that the SO is "loka~, in the sense that' it bears a thematic relation to the verb. The dative rule4 also has a "localners" condition: if S02.1, then SOcS02. .Let me illustrate these. rules with a simple example, namely the verb sell. The basic lexical entry' 1 posit for this verb includes the following information: SS=N P, -SO=NP, SOZ~NP;' SS=2, SO.3, SOZ-1. This. 1 claitn, is amon& the information that must be included in a representatipn of sell in such uses as They sold John rwo cars. Appll ing the verbal oassive rule to this entry,' we 'get the following: SS=NP, .SOZ=NP; SS=3, S02=1. This verb appears in examples like John ulas.soId two cnrs. Since the original entry for sell did not meet the condition S0~1, thz idjectival pi~ssive rulc' is not applicable; 'correspondirigly, forms 1ilr.e +John was unsold two cars are impossible. The condition for application of dative, 502-1 is met, so we can derive an entry in which .SS=NP, SO=NP; SS=2, SO=1. This corresponds to examples like They sold two cars. Notice that this last entry does 'satisfy the condition on the adjectiva' passive rule, so .we can derive the fqllowing entry for an iidjectival passive p.artici ple for sell: SSdCP; SS=l. This corresporids to exantples like Two cars were unsold.Let us now turn to some mote complex eKamp1es. Specifically, I now want to look at several different verbs which share the same strict subcategorization frame, namely. SS-NP, SQ=NP, SOZ=VP. The verbs in question differ from one another along two Cirnensjons, ni~nely, thc'as~igr~ment of tlict~irrtic relations, and coritrol properties. Whpt I mean by thSs latter $?rase is qidte simple: !he. ulrderslood st~bjcct of ~h c VP' in .the SO2 position will be.lhe'SS in some cases and the SO in others. I will represent this. in the funct iotial structurc by assigning, a thCm;~ti~ relalio,n irot si~nply to S02. but to SOZ(SS) or SOZ(!iO). depending on the8 co~!lrol properties5.assig~ro\cnts of .thernalic rtlatiotis arc inten2:d 'to reflect certiii n intuitions :I bout the scmaniic rdeb df thc various eieme~its. but I cannot, in gcnerni, provide c*tnpil.ical irrgUnients fot 111y asslpments. othcr than the fact that they give me the right rcsulls. 1 do have an optri~lionnl cri~crion for deciding whether lo call the SO 3 1 or 3 3: when the verb ill question could nppeiir in n donl)lc object const r~~c t ion (i.e., irnmedliitelv followed by two NP's except ~h a c the controller is t l~e st~l~jcct, riot the o!>ject, i.e., ltle f~inctionel srracture is SS=2. SO=3. S02/SS)=l. If we try. to aflply either passive rule, we will get the following fi~nclinnal str-ucttire: SS=3. SOZ()=i. l'his is ill-formed for the snnle reasoti 11131 the dative of tell w:is, namely, lack of a controjlcr. ' 1 lie corrr.spond4ng exaniples are also impossible: *Joftn.nv~~s prvntised 1. 0 mow the l a~~r t or *John seen~ed promTsEd lo nrorcl tlte Iu~clrt. Dative, frowcver. can .apply, yielding an eptry whose hnctiqnal structure 'is SS=2, SO(SS)=l. This corresponds to example3 like I promised io niow rhe lawn.1 hope that this fragment of the lexicon suffices to show that my proposzd modification of Brrsnan's system permi6 an elegant and natural accouni of a number of syntactic distinctioas, incl'udingSome which have not been-discussed in the li teratuie, to my. knowlidge. One nike featwe that I @odd like to emphasize is that my' proposals provide' a rather straightforwafd acco Jnt. of Visser's (1973; 21 18) observation: "A passive .transform .is onlj possible when khe complement relates to the immediately preceding (pro)noun." In my terminology, passive will be impossible when the active has a complement controlled by the SS, as in' the case of piomise; fcm passivization will always lead to an uncontrolled complement. Thus.-to taka another standard example of Visser's general izatibn, we can account for the distinction betwcer, strike and regard much BS we accounted )for the difference. between pramise and tell. Both will have the following subcategorization frame: SS=NP. SOSNP, S02tAP. Their fuctional structures will include the assignments S S d and SO=I; they will differ in that rcagard -will have S02(~0)=4; while strike hu S02(SS)+. These assignments are for examples like John regards/s!r{kes U r y as pompbus. If ne$ipply passive to regard we get SSb=1, SOZ(SS)=4, bs in Mary is regarded as pompous, Applying passive, to strike wZ get SS=l, S02()4, which i s ill-formed, 4s is *Mury is struck as pompous. Notice, i n c i d e n t w , that .this, example ilfdstrates that, in the system 1 advocate here, cdnstituents other than VP's can serve as predicates, ant! be subject to control.This concludes my suggestions, for 'rnodifyMg Bresnan's framework. 1 hope I have succeeded in indicating how a grammar which makes extensive use of the lexicon In place of syntactic tr:insformtiofis can handle an r w of syntactic facts inga satisfying manner. Next, ! wis 5: tb .argue. that a systelii of the sort 'outlined here can be effectively constrained in reaso~iable and i n t e r k i~q ways. Intuitively, it seems quite plausible that such a systcm would be easy to conslrain, for by drastically reducing the role of transformatfuns, it opens the way for reductions in the power of transformalions. A nuniber of candidate constrainls on transformations come. to mind. For example, .within Bresnan's f rimework one might plausibly argue that n o . trnnsformation call creatc new gramtnatical reliltions (e.g., there will be no "subject-creating" ttansforhations, like passive or raising to subjcct), or that no transforma$io~, can change the words in< thc -sentence niorpl~ulogically (c,g., there *ill be no norninalization, agreement, or case-nierkin-trit~~sformalions-Lcf. Rrame (1978)). Various ways ir; which lexical rilles niight be constrai~led also come to mind: most immediately, it seems to me.biet lllaiiy of the' "lavts".of reletiydal grnnilnar propos:d by Posttil and ~erlrnuttcr in recent yea-s could be translaled straibhtforwardly into the kind of framework discussed herc. In this aaper, however, I would like to consider the consequences of a constrai~lt on transformations modeled on the Freezing Principle of Culigover and Wexler :1977). My proposal depends on distinguishing two classes of trapsformations: root transformations (Emonds (1976) ). and what I will call untlounded'rules. Root transformations arc rules like English subject-auxiliary inversion in questions, which apply only to main clauses; unbounded rules are transformations (e.g., wh-movement) which involve a crucial variable, it., they move something over a variable or they delete sqmething under identity with something on the other side of a v3riable7 (see the contributions by Chomsky, .Bath. Bresnan, and Partee in Culicover, et a1 (1977) for discussion of whether unbounded rules are truly unbounded). The constraint l wish to propose, which 1 will call. the. inter:iction constraint'is the following: once a rule of one.of these c l a s s~ has applied Lo a given structure, .no further rule of the same type may apply to that structure. More specifically, when a transformation applie~, the smalles~. constituent containing all of the affected elements becomes' frozen, in the seme that further transformations of tlie same type may analyze i t This means, in effect, that there will be no 'ititeractions anlong root transforrnatibns, nor among unbounded tl.ansfermations (though a root transformation may interact with an unbounded rule, as in the case of English wh-questions). . l believe tl'at there are several desirable consequences of prohibiting such interactions.First df all, let me mention a somewhat conjectural reason f o~ advocating ihe interattion constraint. As noted above, a vcrj similar proposal emerged from the learnability studies of Wexler, Culicover, and Hamburger: they were able to prove that a class of grammars in which nodes were frozen under slmilar conditions was leatnabls by a fairly simple learning device. Hence, it secnis plausible to conjecture that the interactioq constraint might I)e useful in devising a learnability proof for some version of Rresnnn's Chcory. 'In any evefit, it seems that tile in teraxtion constrairl t Would make the language-learner's task easier by limiting the extent a to which surface structures could deviitte from base fortns (seeCoker & Urain (in preparation)).Second, thcre is empirical, support for the interaction consrraint Emonds (1972; 38=40) shows that only one root preposing trarlsforniation can apply p' er scntetlcc. Sincc the snialle;t structure conltlinir~g initiill position in a mot sentence is the whale sentence. Emm~ds's observqrion is ati itnrllediale conscquer.cc of the interaction constraint. Similarly, many of the witys in which unbounded tri~nsfortnations are prollibiled from tnteracting are familiar. For cxnn~plc, thc fact that elctncnts it1 relative clauses are initcccssi ble to irr~bounded transformlr~io~ls has been extensively discussed in the literature (c.g., Ross (1967) , Chonrs&y (1973), tomcite only two accuiiw). This fact follows from the interaction constraint, slnce an unbounded transformation is involved in the formation of relalive clauses. Hence, examples like Who do you know u man w h~s a w ? or 'John i s tallertltanJ know a man who is are excluded by th: interaction constraint.The fact that comparative clauses and embedded questions are also "islands" has been less widely discussed in the literature, but is also a consequence of the interaction constraint Thus, such examples as *Who-is John louder ihan Mary persuaded to be? or *Who does John wonder when Bill will see? are excluded because they involve wh-movernent extracting inaterial from clauscs in which wh-movement or comparative deletion has taken place. Likewise, comparative clauses are impervious to further applications of comparative deletion: John was kind to more people rhan he liked Bill more than / liked (where this would mean, if grammatical, that the number c~f people John was kind to exceeded the number of people liked better by Bitl. ihan by me). In short, the interaction constraint seems to make the right predfctions about a substantial array of data.Finally, I would like to suggest that the interaction constraint serves not only to restrict tllc class of grammars made available by lirlguistic theory, but also to limit the class of languqges generable by the available grammars (see Wasow (in prcss 3) for discussion, of this distinctim). I w.ill not attempt any formal demonstration of this concluSion here, but will sketch briefly why 1 believe it to be the case. Peters and Ritchie (1973) prove that the languagc generated by a transformational grammar is recursibe if it is possible, on the basis of a surface string, to effectively compute a maxinium size of a deep structure from which that string could be derived. The interaction constraint, together with sthe standard condition on recoverability of deletions (see Peters and Rilchie (1973) ). liniit the extent to which deietions may shrink a stfucture. 1'0 show wliy this is the case, it will tle aseful to invent some terminology: let us call A a parcnt of B if B can be derived from A-by a single application of one transformation. A parent's parent will he called a grandparent, and so on. .Now consider a string cjf length n. Because 3f the recoverabill ty condition, its parcnt callnot bc longer than 2n (measuring length-in lernls of number of tertiiinal symbds): Likewise. its grandparent cannot he* longer than 4n However, if the grandparcnl were the full 4 1 r long, then tlie parent would be frozen by.lhc. interaction constraint, and thc original string viould be undcrivable. In fact, each (length 1 1 ) half of the parent must have 3 parent of lensth no morc than 2n-1, if we are to avoid blocking the dtrivation by the intcractbn co~~strainl. TINIS, the .maxinium size of a grandparetlt is 4n-2, By similar reasorting it is not. hard to sce .that he niaxitlitrm s i~e of. any ancestor m t l generations rernovetl is 2In(2n-.m) .Sincc this nl~nihcr becomes xero. whcn m;2n. there is an effective upper b~rrrnd on the sizc of any atlccstor. Ilencc, the it\tcr;lctio~l con.str;~int, togcthcr wit41 lllc ~~i t~\ d : !~~l .condi:ion on tecovurlrhility of dcletionq, li m i t h 4 he class of I~I I I &~I~I~C S wlrich can be. generala to a s~bclass of the recursive sets8. This provides yet another point of convergence with comp~?ational concerns, since, as noted above, a language must be fecursivc in' order to 'be effectively processed.r havp qketched a version of transformational gramma[ which 5eems to hold considerable promise. There are a number of problems with this approach which 1 am aware of and undoubtedly many more I am blisshlly ignorant of. What 1 haye presented here. was intended, more than anything elv, as an indication of a program of research, and 1 have hence felt free to ignore, many important issues. The primary point' J wish to make is that the study of language appears 'to have progressed t a poipt where the concerns of the transformationalist "and the concerns of, the compu@tional linguist need not conflict, and indeed qay be addressed by a single theory.* I wish to express my gratitude to Adrian Akmajian, Joan Bresnan, and especially Ron Kaplon for verb stimulating discussioris of some of the material in this paper. They are, of course,. absolved of an9 responsibility for its .shortcomings. I am also very grateful to the Xerox Corporation for making its resources human and elcctronic, available to me in the preparation of this paper. Some of the research reported on here was begun under a Summer Stipend from the ~afiorlal Endowment for the Humanities. Footnotes 1. No rigorous definition of these. notioris has. cver been offered i n the literature. and certain problems with [he wry lhcy have been used have been pointed out (e.~.. I i~s t and Briime.(1976)1. I do hot wish to conimil myself lo all of thc clr~ms wliicli have been made In the lrterature about t ese notions. and my notali~n below is intended to rcftecl this. 1 do, how f ver. believe that,tholr who have discussed thematic rclauont; are onto something impurhnl 2. Obviously. thcre is nrore to forming passives-than this; for cxample. I ignort morphology.Those .familiar ~r t h . Postal and Pcrllnutler's version. of relational grammar will recugnl7e the resenihlance d last senltnce to the Rclattonal Annih~lalion L; I w.Notice by the way. lhat iry passive rules siky tiothing r b v~t the h. v phrase. 1 sir1 iisstlmlng. with B r c m~n (in press), (hill thcre is an inde ntlcnt rule ass~~,ning':~ecnr status to tihe objecls d some by phrases.Thrs r i~r u u l l l operate no! vnlf i n passives. bul also i n exarnpln. ltke The syn~phony wcls by Becthnvrn. 4. Notirc that 1 :Im fortnuktlinp the dative rule "hackwards", Ih.~t is, with the doublb objccl coi]structron 11s the inpul. My rule says nothing aboul the prcpositicrns to iind / o r brlc~nac I ~ISSUII~~ that thc func(~urt:tl rolc of their objects wlll bc covtred by sep.rt;ile rules, ;a i s thc c:ec w~t h b Examples like Julrn's qall 'wur 10 htdry m d Titis presmr i s Jar yua lm$crlcnce to my acsuniptlon.5. . This is .lo be itndcrstood as saying lhat -tht SO? will be trcalpd :I $ ?prcdicatc, with 11s ow11 assi&nniu~hts of tkcniat~c rtl.lIlons. qnd with the elrmcnl in . p:~rcthescs t;cilted irs ~f rl were the SS of lhnt predicate.6. Jane Kobirison has sujgcstcd to nic 11131 i t mi@ bc awre :ippropriate sernnnl~c;rllp to. lrr;lt llrc st1ltjlfc.t. of bclievc as n 3. I'his. would hc perfectly conipi~tiblr with nly an:llgsls. . ,7. My treatment here ignores anaphora rules Iike'VP deletion and sluicing. I ;In1 .assumin that the* rules arc not transforaa~rons, but 3 separate category of ruyes. subject to their awn unHue cond~truns (see W a r n (in press b) for dtscussion).As ivcn, my argument does not lake into account root trans rmatrons or spec.1 f led deletions (see Warow (in press a)). I : is quite triviaphowever. l o extend .the arzument 01 cover these cases | null | null | null | null | Main paper:
:
raised, 1 ittle support cotild B e idduced for the hypothesis that thc oparations of transforn~ational grammar play .a part in speakers' or hearers' processing of scntences.(see Fpdor, et a1 (1974: chapter 5) ). lnitead of concerning themselves with qr~cstions of processing,. tmnsf'ormntioni~lists have concentrated their efforts ( i~t least in the l i~~t dt.~iide o r so) on the problem of constraining thc k w c r of their thcory. T , k goal of much recent research has becn t o construct as. restrictive a theory of granlmar as possiblc, within the bounds set by the known divcrsits of human latigungcs (scc. c.g.. Ross (1007), Chomsky (1973) , Bresnan (1.976) , Emonds (1976) , and Culicover and Wexler (1977) for examples .of this .type of research).Computational linguists, on the other hand, have not e$plicitly concerned rhenisc.lves very a,mi~ch with the probl'etn of constraints (but see Woods (1973; 124-5) 'for an exception), Ilatlicr, their goill has bccn to l'iltd effective procedures for the parsing and proccssitig of natural Ianguagc. Whilc this is implicitly a . rcstricti.on to rcc~~rsive languages, the computational literature has dealt more, with questions of processing . than with how to limit the class .of available grammars o r languages.In previoils papers (~s h e r s o n and Wasow (1976) , Wasow (in press a. 19711)) 1 have argi~cd for the legititnacy of the quest for constrilints ,as a rescarch strategy. 1 have argued that a thcory that ,plrrccs litnits on thc class of possible languages n1ah.c~ significant empirical clai~ns a b~u t --l himan .mental capacitics, and can contribute. to a solution to "the fundamental empirical problem .of linguistics" (as ~h o m s k y has cullcd it) of how cliild ren .are able .to '!car11 languages with such' facility. I have tried to show that such psychological cl.iir~ls Ciirl, LIL' ~~l;ldc, ' i t l~o i~t I I~: \~~I I L ; all) ; t b : t~~~l~ j~t i~l i s i~b~~t what rolc gtammqs play in pc.~~orariulcc. In short, I have a r g~~c d that a tlit'ory 01' gra~linii~r Cat1 ~liak'c sig~lificant col~t~.ibiitions to psycholbgy, i~rdcpcndet~t of tllb ilnswcr to the granli~iatical. rc.i:ill izatioa ~~~o b l c t n .Rccyit work by Joi~tl Bres~l;tll (in prcss) takcs a vory diffcrcnt positiun: s,hc 1135 S L I~~C S L C J illat tratibfor~ii;lti~l~ali~t~ O L I~I I t to pay morc attention to thc gram~natic:~l rcaliz:~tion problem, and thitt .considerations of processing suggest. ' radical modifications in 7he theory of' transfortnational grammar. Furthcr, she argues that thcrc. is-st~iplt: pra~nliiatical evidcnce for tlrcsc modificatioos. In this piryer 1 will suggest solnc dxtcnsions at' hcr proposnlq and will explore somc of their empirical cotiscqucnces.Furthe& 1 will argue that her I'rrrnlcwork ~iiiikcs it possiblc tu imposr: riltllcr rcstrictive consfrir inls vn 'gr;rlrin~i~t icsl tt~cory. 'i'hils. 1 will ilrgllc that the grrtmmaticnl rcaliration problcm and the prablcm of constraitiing trun~fr\r~~liltio~lal thcory, wllilc lo@ally independent, are both addressed by Bcesnan's proposals. If I am correct in this, then L)res~ian's "realistic tratlsformational grammar" rcpiescnts a major-conucrgcnce oi .the concerns of transfor~natiorid and compu~atiorri~l lingnists.My presentation , will consist of three parts. First, I . will briefly sketch Brr'snan's framcxvork. Sccond, I. will suggest some extensions of her proposals and point out .$erne conse'q~icnces .d these e.xtcnsions. Third, -1 will prdpose how her framewo'rk can be constrained, and indicate certain desirable consequences .of m? proposals.The primary innovation of Bresnan's framework is &hat. it eliminates . a large 'class'af transformations in favor of an enriched conception bf the lexicon. The grammar that results is one that Bresnan claims is . f a r more. realistic from a proccsging point of view than other versions. of transformational grammar. She points out striking similarities betwczt~ her proposals and .recent ' conip~rtirtiunal and psycholinguistic work by KapIa11 and Wanncr, and shc argues that Augmented Trr~nsition Nctworks can provide a t least a' partial answer to. the .grammatical realization problem within her framework. now sketch very roughly what Btesnan's "realistic" transt-ormational erammar .is like.. . Rules like passive, dative, and raising rules, which are "structqre-preserving" (in the sense that thcir. outputs are st~ucturally' idei~tical-to indcpcnddntly' rcquircd base-generated structures) and "local" (in tlic scnsr: that the :elein$nts affectcd arc' always in the immediate cnvi'ronni~nt of some governink iexica~ item, ~a o s l i y a . vcrb): are climinutcd from the tritnsforrnational cot~lponent atid relcyutcd to tlic lexicorl. .Lexical entries include, among other things, (strict) subcatcgorization frames arlcl o c absll.~c;l rc~u.usc111.11iyns N 1kic.h I31-cs11an ~.i~lls "f~111~ti011ill S~~U C~L I~C S " ur '*prcdicatc arg11IIicnt structurc9*'. S~bccttc'porizatio~~ 1'ranl~'s' give t l~c sy~itactic e h v j r~t~f i~c~~t s ' in which thc luxical i'tclii lilay appuar: thcse arc cxpresscd in terlas o f i1 hiisic s t grarli~i~aticnl relations, includi(g "sut~ject" a11.d "objcct". 'i'hcsc notiotis. w l i~l r l~nivrrsi~l, are inst;~nliatcd dif'fcrclitly i l l diffcrc~\t Ii\ng.ll;~gcs; for cxamplc, nrcsllnti tt1kt.s c~&nti:llly thc s t r u c t~~~a l dcfi~iitiorls of "subject" . . and ''nhject" prmposrd hy ('hornsky (19115; 71) as langr~ngespccific charilctcri;r.a~io~ls' 01' thcsc notio~ls for English. Futlctionitl structure3 give a .more abstract reprpscntation of tlic elements mcngioncd in the sub~;ltcgoriz;ltion frame, indicating what thcir-"1ogic;lll' rc1;ltionships arc, 'I'hus, thc functional structure correswnds very roughly to the deep structure in the standard theory of transform;iti.onal grammar, and the subcategorization frame corresponds even more roughly to the surface structure.What the standerd theory did with local structure-preserving transformations' Bresnan can do in either of two ways. ~elati'otrships like active/passive are handled by positing two separnte lexical~'entries for active and passivg verb forms. The product\uify of this relationship' can be accounted for by means of a lexical redundancy rule, which would say, in effect, that corresponding to the typical...transitive verb there is an intransitive. verb which looks morpbobgically like the perfect form of "the transitive, and whos'e subject plilys the sitme Iqical role ( i .~. in the functional structure) as the object-of the tra~sitive verb. ~resnan's other way of replaeink local structure-preserving rules is illtatrated most clearly with-the raising rules. Raising to object pcwitioh, for cxarnple, is used to capture the fact that ihe NP which' is syntactically the object c4 one clause is logically not an argument of that clause a1 all, put 4 subject of the sutsordinate'~clausc. Bresnan expresses this simply in terms of the rcli~tionship between the saategorization frame and the'functional structure; that is, the object of the niain clause plays no role in the ful\ctional structure of that clause, but is "prrhsed down". to play 3 role in the next clause down. In the interests of brevity I will not. illustrate Bresnan's framework here. Rather, I will refer the interested rei\dzr to her paper, and go on to indicate my reasons for seeking to modify hl:r proposals.My primary motivittion comes from some earlier work of mine (Wiaow (1977) ). which argued agaiqst the climirta&iotl'of local. strrtcture-priscrving trnnsformations. My qrgumcnt was based on theobscrvation that there are two similar but distinct classes of linguistic relationships whose differences can be expressec) rather nilturally as the differences between tra~rsformational rules and lexical redundancy rules: ' The clearest cxatnple of this is the English passive. ItAss often been suggested that some passive pafiiciples arc aljectives and others verbs: I .pointed out that adjectival passivcs ~~t i d verbal passives differed in certain systemiltic ways. M y c e n M claim was that the surface subject of adjectival passives was always the deep direct object of the correspoiding verb. For example, a passive participlz which is demonstrably adjectival . (e.g.,.because it is prefixed with un-or imnediately follows seem) may not have as its. surface subject the "logical" subject of a lower clause, the indirect object, or a chunk of an idiom: *John is unknown lo be a cornmurist *John seemed told the story, *~d v & r u~e seemed taken o/ John,. A verbal passive, in contrast, could have as its subject any . N P .which could immediately follow the corresponding active verb: John is .known to be a communlsr, John was told rhe story, Advantage was taken of John This, I claimed, would foffow from the hypothesis that adjectival pssives are formed by a lexical redundancy rulc, whereas verbal . passive .are tr~nsformatlonally derived, if lexical redundancy rules arb "relational", in the sense that they are formulated in terms of grammatical relations such as subject and object, whereas transformations are '"struttural", i.e., .they areoperations on phrast! structure tree.It is evident that my earlier position is inconsistent with Bi_esnanls recent proposals. My estensions of her ideas, developed in collaboration with Ron -Kaplan, are'in part an attemipt to capture within her framework ihe distinction -my earliti paper sought. to pxplicatc in terms of the lexicon/transfotmation contrast They are also motivated by the very interesting comments of Anderson (1977) . *~ride-son suggests tliat 1 -was mistaken in -claimin8 that the operative factor in formulating rules like the adjectival. passive r u l~ wag the deep grammatical relation .of the surface subject Rather, he argues, it is thematic relations like "theine". "aeent*'. "goal"and "source-" (see Gruher (1965) and Jackendoff (1972)) which are 'erucinll. Asstiming Andersoh to' he correct, an obvious 1n4qification df Nesnan's systerfi svggests itself,' which would permit the distinctions of my earlier paper to be captured. Let us supposc: thath the functiollsrl structure in,lexical 1:ntues is-a specification of which thiniatic relations should be assigned to the eleme~rts mel!tiorixJ in the subcategorization frame. Then we may distinguish.two types of lexical rules: those that make reference to thematic relatiqns and thdse that. do not. The former would correspond to rules that nty earlier paper called lexical, and -the I9tter to those ihat ' l galled transformations, 'This is the extension of Brcsnan's fratilework that I wish To propose. I will illustrate ky formulating the two pa:;sivc.rules and the dative rille and applying them to a fragment of the lexicon of Engli&.kly formalism is based on the ..rtssrrmplion that the grilfnn~alicsl relations are eivcn latrguagc-wid$ definitions in struct~~ral terms (at !cast in English) along rhe lines in(licnted by Dresn~n. and that a verb's st~bcr\tegori;ration frame inrt.cly itidicates which relations it has, and vrltat grammatical categories lhnso relations are assigncd to. . (Thus, 1 differ from Bresnan in this .respect, for she assumed that. grammatical relations would be limited to NP's). I will adopt the fb\\owing abbreviations: "SS" 2 (surface) subject: SO = (surface) object "S02" ='(suiface) second object; "1" = tl~eme: "2" o agent; "3" t goal; "4" = complemenl The rule forming ,erbal passive participles from the corre~po~ding active lexical entries can now be formulated2 quite simply as SScSO. This is to be interpreted i s follows: eliminate "SS" wherever i t apjears in lhe entry for the active verb (eliminating also au)' assignment it may have to a thematic relatibn) and change all occurrences of "SO to "~5"~. The adjectival passive rule Will differ from this in that it ha? an additional condition on it: if SO=1, then SScSB: This condition insures that the SO is "loka~, in the sense that' it bears a thematic relation to the verb. The dative rule4 also has a "localners" condition: if S02.1, then SOcS02. .Let me illustrate these. rules with a simple example, namely the verb sell. The basic lexical entry' 1 posit for this verb includes the following information: SS=N P, -SO=NP, SOZ~NP;' SS=2, SO.3, SOZ-1. This. 1 claitn, is amon& the information that must be included in a representatipn of sell in such uses as They sold John rwo cars. Appll ing the verbal oassive rule to this entry,' we 'get the following: SS=NP, .SOZ=NP; SS=3, S02=1. This verb appears in examples like John ulas.soId two cnrs. Since the original entry for sell did not meet the condition S0~1, thz idjectival pi~ssive rulc' is not applicable; 'correspondirigly, forms 1ilr.e +John was unsold two cars are impossible. The condition for application of dative, 502-1 is met, so we can derive an entry in which .SS=NP, SO=NP; SS=2, SO=1. This corresponds to examples like They sold two cars. Notice that this last entry does 'satisfy the condition on the adjectiva' passive rule, so .we can derive the fqllowing entry for an iidjectival passive p.artici ple for sell: SSdCP; SS=l. This corresporids to exantples like Two cars were unsold.Let us now turn to some mote complex eKamp1es. Specifically, I now want to look at several different verbs which share the same strict subcategorization frame, namely. SS-NP, SQ=NP, SOZ=VP. The verbs in question differ from one another along two Cirnensjons, ni~nely, thc'as~igr~ment of tlict~irrtic relations, and coritrol properties. Whpt I mean by thSs latter $?rase is qidte simple: !he. ulrderslood st~bjcct of ~h c VP' in .the SO2 position will be.lhe'SS in some cases and the SO in others. I will represent this. in the funct iotial structurc by assigning, a thCm;~ti~ relalio,n irot si~nply to S02. but to SOZ(SS) or SOZ(!iO). depending on the8 co~!lrol properties5.assig~ro\cnts of .thernalic rtlatiotis arc inten2:d 'to reflect certiii n intuitions :I bout the scmaniic rdeb df thc various eieme~its. but I cannot, in gcnerni, provide c*tnpil.ical irrgUnients fot 111y asslpments. othcr than the fact that they give me the right rcsulls. 1 do have an optri~lionnl cri~crion for deciding whether lo call the SO 3 1 or 3 3: when the verb ill question could nppeiir in n donl)lc object const r~~c t ion (i.e., irnmedliitelv followed by two NP's except ~h a c the controller is t l~e st~l~jcct, riot the o!>ject, i.e., ltle f~inctionel srracture is SS=2. SO=3. S02/SS)=l. If we try. to aflply either passive rule, we will get the following fi~nclinnal str-ucttire: SS=3. SOZ()=i. l'his is ill-formed for the snnle reasoti 11131 the dative of tell w:is, namely, lack of a controjlcr. ' 1 lie corrr.spond4ng exaniples are also impossible: *Joftn.nv~~s prvntised 1. 0 mow the l a~~r t or *John seen~ed promTsEd lo nrorcl tlte Iu~clrt. Dative, frowcver. can .apply, yielding an eptry whose hnctiqnal structure 'is SS=2, SO(SS)=l. This corresponds to example3 like I promised io niow rhe lawn.1 hope that this fragment of the lexicon suffices to show that my proposzd modification of Brrsnan's system permi6 an elegant and natural accouni of a number of syntactic distinctioas, incl'udingSome which have not been-discussed in the li teratuie, to my. knowlidge. One nike featwe that I @odd like to emphasize is that my' proposals provide' a rather straightforwafd acco Jnt. of Visser's (1973; 21 18) observation: "A passive .transform .is onlj possible when khe complement relates to the immediately preceding (pro)noun." In my terminology, passive will be impossible when the active has a complement controlled by the SS, as in' the case of piomise; fcm passivization will always lead to an uncontrolled complement. Thus.-to taka another standard example of Visser's general izatibn, we can account for the distinction betwcer, strike and regard much BS we accounted )for the difference. between pramise and tell. Both will have the following subcategorization frame: SS=NP. SOSNP, S02tAP. Their fuctional structures will include the assignments S S d and SO=I; they will differ in that rcagard -will have S02(~0)=4; while strike hu S02(SS)+. These assignments are for examples like John regards/s!r{kes U r y as pompbus. If ne$ipply passive to regard we get SSb=1, SOZ(SS)=4, bs in Mary is regarded as pompous, Applying passive, to strike wZ get SS=l, S02()4, which i s ill-formed, 4s is *Mury is struck as pompous. Notice, i n c i d e n t w , that .this, example ilfdstrates that, in the system 1 advocate here, cdnstituents other than VP's can serve as predicates, ant! be subject to control.This concludes my suggestions, for 'rnodifyMg Bresnan's framework. 1 hope I have succeeded in indicating how a grammar which makes extensive use of the lexicon In place of syntactic tr:insformtiofis can handle an r w of syntactic facts inga satisfying manner. Next, ! wis 5: tb .argue. that a systelii of the sort 'outlined here can be effectively constrained in reaso~iable and i n t e r k i~q ways. Intuitively, it seems quite plausible that such a systcm would be easy to conslrain, for by drastically reducing the role of transformatfuns, it opens the way for reductions in the power of transformalions. A nuniber of candidate constrainls on transformations come. to mind. For example, .within Bresnan's f rimework one might plausibly argue that n o . trnnsformation call creatc new gramtnatical reliltions (e.g., there will be no "subject-creating" ttansforhations, like passive or raising to subjcct), or that no transforma$io~, can change the words in< thc -sentence niorpl~ulogically (c,g., there *ill be no norninalization, agreement, or case-nierkin-trit~~sformalions-Lcf. Rrame (1978)). Various ways ir; which lexical rilles niight be constrai~led also come to mind: most immediately, it seems to me.biet lllaiiy of the' "lavts".of reletiydal grnnilnar propos:d by Posttil and ~erlrnuttcr in recent yea-s could be translaled straibhtforwardly into the kind of framework discussed herc. In this aaper, however, I would like to consider the consequences of a constrai~lt on transformations modeled on the Freezing Principle of Culigover and Wexler :1977). My proposal depends on distinguishing two classes of trapsformations: root transformations (Emonds (1976) ). and what I will call untlounded'rules. Root transformations arc rules like English subject-auxiliary inversion in questions, which apply only to main clauses; unbounded rules are transformations (e.g., wh-movement) which involve a crucial variable, it., they move something over a variable or they delete sqmething under identity with something on the other side of a v3riable7 (see the contributions by Chomsky, .Bath. Bresnan, and Partee in Culicover, et a1 (1977) for discussion of whether unbounded rules are truly unbounded). The constraint l wish to propose, which 1 will call. the. inter:iction constraint'is the following: once a rule of one.of these c l a s s~ has applied Lo a given structure, .no further rule of the same type may apply to that structure. More specifically, when a transformation applie~, the smalles~. constituent containing all of the affected elements becomes' frozen, in the seme that further transformations of tlie same type may analyze i t This means, in effect, that there will be no 'ititeractions anlong root transforrnatibns, nor among unbounded tl.ansfermations (though a root transformation may interact with an unbounded rule, as in the case of English wh-questions). . l believe tl'at there are several desirable consequences of prohibiting such interactions.First df all, let me mention a somewhat conjectural reason f o~ advocating ihe interattion constraint. As noted above, a vcrj similar proposal emerged from the learnability studies of Wexler, Culicover, and Hamburger: they were able to prove that a class of grammars in which nodes were frozen under slmilar conditions was leatnabls by a fairly simple learning device. Hence, it secnis plausible to conjecture that the interactioq constraint might I)e useful in devising a learnability proof for some version of Rresnnn's Chcory. 'In any evefit, it seems that tile in teraxtion constrairl t Would make the language-learner's task easier by limiting the extent a to which surface structures could deviitte from base fortns (seeCoker & Urain (in preparation)).Second, thcre is empirical, support for the interaction consrraint Emonds (1972; 38=40) shows that only one root preposing trarlsforniation can apply p' er scntetlcc. Sincc the snialle;t structure conltlinir~g initiill position in a mot sentence is the whale sentence. Emm~ds's observqrion is ati itnrllediale conscquer.cc of the interaction constraint. Similarly, many of the witys in which unbounded tri~nsfortnations are prollibiled from tnteracting are familiar. For cxnn~plc, thc fact that elctncnts it1 relative clauses are initcccssi ble to irr~bounded transformlr~io~ls has been extensively discussed in the literature (c.g., Ross (1967) , Chonrs&y (1973), tomcite only two accuiiw). This fact follows from the interaction constraint, slnce an unbounded transformation is involved in the formation of relalive clauses. Hence, examples like Who do you know u man w h~s a w ? or 'John i s tallertltanJ know a man who is are excluded by th: interaction constraint.The fact that comparative clauses and embedded questions are also "islands" has been less widely discussed in the literature, but is also a consequence of the interaction constraint Thus, such examples as *Who-is John louder ihan Mary persuaded to be? or *Who does John wonder when Bill will see? are excluded because they involve wh-movernent extracting inaterial from clauscs in which wh-movement or comparative deletion has taken place. Likewise, comparative clauses are impervious to further applications of comparative deletion: John was kind to more people rhan he liked Bill more than / liked (where this would mean, if grammatical, that the number c~f people John was kind to exceeded the number of people liked better by Bitl. ihan by me). In short, the interaction constraint seems to make the right predfctions about a substantial array of data.Finally, I would like to suggest that the interaction constraint serves not only to restrict tllc class of grammars made available by lirlguistic theory, but also to limit the class of languqges generable by the available grammars (see Wasow (in prcss 3) for discussion, of this distinctim). I w.ill not attempt any formal demonstration of this concluSion here, but will sketch briefly why 1 believe it to be the case. Peters and Ritchie (1973) prove that the languagc generated by a transformational grammar is recursibe if it is possible, on the basis of a surface string, to effectively compute a maxinium size of a deep structure from which that string could be derived. The interaction constraint, together with sthe standard condition on recoverability of deletions (see Peters and Rilchie (1973) ). liniit the extent to which deietions may shrink a stfucture. 1'0 show wliy this is the case, it will tle aseful to invent some terminology: let us call A a parcnt of B if B can be derived from A-by a single application of one transformation. A parent's parent will he called a grandparent, and so on. .Now consider a string cjf length n. Because 3f the recoverabill ty condition, its parcnt callnot bc longer than 2n (measuring length-in lernls of number of tertiiinal symbds): Likewise. its grandparent cannot he* longer than 4n However, if the grandparcnl were the full 4 1 r long, then tlie parent would be frozen by.lhc. interaction constraint, and thc original string viould be undcrivable. In fact, each (length 1 1 ) half of the parent must have 3 parent of lensth no morc than 2n-1, if we are to avoid blocking the dtrivation by the intcractbn co~~strainl. TINIS, the .maxinium size of a grandparetlt is 4n-2, By similar reasorting it is not. hard to sce .that he niaxitlitrm s i~e of. any ancestor m t l generations rernovetl is 2In(2n-.m) .Sincc this nl~nihcr becomes xero. whcn m;2n. there is an effective upper b~rrrnd on the sizc of any atlccstor. Ilencc, the it\tcr;lctio~l con.str;~int, togcthcr wit41 lllc ~~i t~\ d : !~~l .condi:ion on tecovurlrhility of dcletionq, li m i t h 4 he class of I~I I I &~I~I~C S wlrich can be. generala to a s~bclass of the recursive sets8. This provides yet another point of convergence with comp~?ational concerns, since, as noted above, a language must be fecursivc in' order to 'be effectively processed.r havp qketched a version of transformational gramma[ which 5eems to hold considerable promise. There are a number of problems with this approach which 1 am aware of and undoubtedly many more I am blisshlly ignorant of. What 1 haye presented here. was intended, more than anything elv, as an indication of a program of research, and 1 have hence felt free to ignore, many important issues. The primary point' J wish to make is that the study of language appears 'to have progressed t a poipt where the concerns of the transformationalist "and the concerns of, the compu@tional linguist need not conflict, and indeed qay be addressed by a single theory.* I wish to express my gratitude to Adrian Akmajian, Joan Bresnan, and especially Ron Kaplon for verb stimulating discussioris of some of the material in this paper. They are, of course,. absolved of an9 responsibility for its .shortcomings. I am also very grateful to the Xerox Corporation for making its resources human and elcctronic, available to me in the preparation of this paper. Some of the research reported on here was begun under a Summer Stipend from the ~afiorlal Endowment for the Humanities. Footnotes 1. No rigorous definition of these. notioris has. cver been offered i n the literature. and certain problems with [he wry lhcy have been used have been pointed out (e.~.. I i~s t and Briime.(1976)1. I do hot wish to conimil myself lo all of thc clr~ms wliicli have been made In the lrterature about t ese notions. and my notali~n below is intended to rcftecl this. 1 do, how f ver. believe that,tholr who have discussed thematic rclauont; are onto something impurhnl 2. Obviously. thcre is nrore to forming passives-than this; for cxample. I ignort morphology.Those .familiar ~r t h . Postal and Pcrllnutler's version. of relational grammar will recugnl7e the resenihlance d last senltnce to the Rclattonal Annih~lalion L; I w.Notice by the way. lhat iry passive rules siky tiothing r b v~t the h. v phrase. 1 sir1 iisstlmlng. with B r c m~n (in press), (hill thcre is an inde ntlcnt rule ass~~,ning':~ecnr status to tihe objecls d some by phrases.Thrs r i~r u u l l l operate no! vnlf i n passives. bul also i n exarnpln. ltke The syn~phony wcls by Becthnvrn. 4. Notirc that 1 :Im fortnuktlinp the dative rule "hackwards", Ih.~t is, with the doublb objccl coi]structron 11s the inpul. My rule says nothing aboul the prcpositicrns to iind / o r brlc~nac I ~ISSUII~~ that thc func(~urt:tl rolc of their objects wlll bc covtred by sep.rt;ile rules, ;a i s thc c:ec w~t h b Examples like Julrn's qall 'wur 10 htdry m d Titis presmr i s Jar yua lm$crlcnce to my acsuniptlon.5. . This is .lo be itndcrstood as saying lhat -tht SO? will be trcalpd :I $ ?prcdicatc, with 11s ow11 assi&nniu~hts of tkcniat~c rtl.lIlons. qnd with the elrmcnl in . p:~rcthescs t;cilted irs ~f rl were the SS of lhnt predicate.6. Jane Kobirison has sujgcstcd to nic 11131 i t mi@ bc awre :ippropriate sernnnl~c;rllp to. lrr;lt llrc st1ltjlfc.t. of bclievc as n 3. I'his. would hc perfectly conipi~tiblr with nly an:llgsls. . ,7. My treatment here ignores anaphora rules Iike'VP deletion and sluicing. I ;In1 .assumin that the* rules arc not transforaa~rons, but 3 separate category of ruyes. subject to their awn unHue cond~truns (see W a r n (in press b) for dtscussion).As ivcn, my argument does not lake into account root trans rmatrons or spec.1 f led deletions (see Warow (in press a)). I : is quite triviaphowever. l o extend .the arzument 01 cover these cases
Appendix:
| null | null | null | null | {
"paperhash": [
"emonds|a_transformational_approach_to_english_syntax:_root,_structure-preserving,_and_local_transformations",
"jackendoff|semantic_interpretation_in_generative_grammar",
"ross|constraints_on_variables_in_syntax",
"rosenbaum|the_grammar_of_english_predicate_complement_constructions"
],
"title": [
"A Transformational Approach to English Syntax: Root, Structure-Preserving, and Local Transformations",
"Semantic Interpretation in Generative Grammar",
"Constraints on variables in syntax",
"The grammar of English predicate complement constructions"
],
"abstract": [
"Imagine that you get such certain awesome experience and knowledge by only reading a book. How can? It seems to be greater when a book can be the best thing to discover. Books now will appear in printed and soft file collection. One of them is this book a transformational approach to english syntax root structure preserving and local transformations. It is so usual with the printed books. However, many people sometimes have no space to bring the book for them; this is why they can't read the book wherever they want.",
"Like other recent work in the field of generative-transformational grammar, this book developed from a realization that many problems in linguistics involve semantics too deeply to be solved insightfully within the syntactic theory of Noam Chomsky's Aspect of the Theory of Syntax. Dr Jackendoff has attempted to take a broader view of semantics, studying the important contribution it makes to the syntactic patterns of English.The research is carried out in the framework of an interpretive theory, that is, a theory of grammar in which syntactic structures are given interpretations by an autonomous syntactic component. The book investigates a wide variety of semantic rules, stating them in considerable detail and extensively treating their consequences for the syntactic component of the grammar. In particular, it is shown that the hypothesis that transformations do not change meaning must be abandoned; but equally stringent restrictions on transformations are formulated within the interpretive theory.Among the areas of grammar discussed are the well-known problems of case relations, pronominalization, negation, and quantifiers. In addition, the author presents semantic analyses of such neglected areas as adverbs and intonation contours; he also proposes radically new approaches to the so-called Crossover Principle, the control problem for complement subjects, parentheticals, and the interpretation of nonspecific noun phrases.",
"Massachusetts Institute of Technology. Dept. of Modern Languages and Linguistics. Thesis. 1967. Ph.D.",
"A set of phrase structure rules and a set of transformational rules are proposed for which the claim is made that these rules enumerate the underlying and derived sentential structures which exemplify two productive classes of sentential embedding in English. These are sentential embedding in noun phrases and sentential embedding in verb phrases. First, following a statement of the grammatical rules, the phrase structure rules are analyzed and defended. Second, the transformational rules which map the underlying structures generated by the phrase structure rules onto appropriate derived structures are justified with respect to noun phrase and verb phrase complementation. Finally, a brief treatment is offered for the extension of the proposed descriptive apparatus to noun phrase and verb phrase complementation in predicate adjectival constructions. Thesis Supervisor: Noam Chomsky Title: Professor of Modern Languages"
],
"authors": [
{
"name": [
"J. Emonds"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. Jackendoff"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"J. Ross"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"P. Rosenbaum"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null
],
"s2_corpus_id": [
"123465810",
"61367317",
"60624374",
"62173023"
],
"intents": [
[
"background"
],
[
"background"
],
[
"background"
],
[
"background"
]
],
"isInfluential": [
false,
false,
true,
false
]
} | null | 554 | 0.009025 | null | null | null | null | null | null | null | null |
9098edf1c18cd07a65a848f9b9da815467c3ebf9 | 59702952 | null | Models of the Semantic Structure of Dictionaries | These n o t i o n s lead t o t h e ultimate model of a dictionary, where p o i n t s represent concepts (which nay be verbalized and symbolized in more than one lay) and lines represent relations ( s y n t a ~t i , c or aemantk-c) between canoepts. Ba ~e d on these models, procedures f o r f i,nding prirniti-ve concepts are described, using the s e t of verbs and t h e i r definitions from W3. Specific rules are described, based on some elementary graph-th6qre t i c principles, structural characterist i c s o f dictionary de'finitiohs, and the parsing of the d e f i n itions. These rules have thus f a r reduced t h e i n i t i a l e e t of 20,000 verbs t o fewer than 4,000, with further reduction t o cone as a l l rules are applied, I t is argued that this approach bears a~ strong r e l a t i o nship to efforts t o represent knowledge in framecr. Although much work is needed on t h e parser and on a computerized version of t h i s approach, there is some hope t h a t the parser, i f expectations are borne out, w i l l be capable o f transforming ordinary discourse i n t o canonical frame representations, in some cases and ~m ~l l c r t t;n others, ids that lexlcal 9 entries in dictionaries a m unsatisfactory DeCAuse they do n o t contain sufficient i n f o m a t i o n . These formali-sms thus require t h a t semantic f e a t u r e s such as 1lanirnateft or " s t a t e w be appended* to p a r t i c u l a r ent*ies. While it is true t h a t ordinary d i o t i onary e n t r i e s do not o v e r t l y identify a l l appropriate features, t h i s may be lees a dlfficulhg inherent in definitions than the fact t h s t no one has developed the necessary mechanisms f o r surfacing features from definitions. Thus, for examp3.e. ltnurse1' may not have the f e a t u r e llanLmatew i n i t s definition, but t?nuraew is defined as a ltwomanw which fs defined a d a tlpersonw ~h i c h is defined as a 1"beingfl" which "Ys defined as a "living thingw; this string seems sufficieht te estabaish "nurseN as "anirnatell. In general, it seems t h a t , if a semantic feature is essential to the meaning o f a particular entry, it is s i m i l a r l y necessary %Hat the feature be discoverable within the semantic structure of a dictionary, Otherwise, there is a defect in one or more d e f i n i t i o n e , or t h e dictionary-contains some i n t e r n a l inconsistency. (Clearly, it is beyond expectation t h a t any pre-~n t dictionary will be free of these problems.) The p o s s i b i l i t y of defective definitions has a l s o gene^-ated crf t i c i a m s , more direct than above, on the p o t e n t i a l usefulness Of a dictionary. On one Hand d e f i n i t i o n s are viewed as "deficient in the presentation of relevant dataw since they provide meaninbv ueing "substitutable words ( i . e . by synonyms), rather than by listing d i s t i n c t i v e femtureafl (Nida 1975 : 172) . On another hand-, the proliferation of meanings 10 attached to an entry is viewed as only a case of "apparent polyeenyN which obscures the more general meaning of a lexeme by the addition of "redundant features already determined by the environmentft (Bennett 1975:4-1.1). Both objections may have much v a l i d i t y and ts that extent would necessitate revisions to iqdividu& or sets of definitions. However, neither viewpoint is sufficient' t o preclude an analysis of what actually appears in any dictionary. It is p o s s i b l e that a cbmprehensive analysis might more r e a d i l y surface such d i f f i c u l t i e s and make their amelioration (and the consequent improvement of definitions) that mu& e a s i e r , Xven though dictionaries are viewed somewhat askance by many who study meaning, it seems t h a t this viewpdint is i n f l uenced more by the d i f f i c u l t y o* systematically tapping their contents than by m y substantive objections which conclusively e s t a b l i s h themas ~s e l e s s repositories of semantic content. However, it is necessary to demonstrate t h a t a spstematic app~oach e x i s t s and can y i e l d useful results. 3 , PREVIOUS RESXARCN ON DICTXONARIES Notwithstanaing the foregoing direct and i n d i r e c t c r i t icisms. some attempts have been made to probc the nature and& structure of dictionary definitions. A review of relevant aspects QI-two such s t u d u s will help the niaterial presented here s t a n d out in sharper r e l i e f . Olney 1 9 6 8 describes the conceptual b a a i s of many pro$eetted routines for processing a machine-readable transcript of | {
"name": [
"Litkowski, Kenneth C."
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 6 | 21 | null | The process whkch has been o u t l i n e d ifi the preceding sec- It seema C.0 me t h a t this i r nothing ,more t h a n the process which has a l r e a d y been described using a graph-theoretic naproach, except t h a t the generalhxed frame f o r each verb will not be, 12. Schank, R.C. Conceptual dependency: A theory of' n a t u r a l b n g u a g e understanding. Cognitiye P a y c h o l o~ 3( 1 9 7 2 )~ 532-631 3 13. Simmons, R. F. and Slocum, 3 , G e n q a t i n g E n g l i s h discourse from se6antic networks. Commu~cations of the ACM* 15 (.1972) t e x t which must be associ.ated w i t h each sense. I t w p e a r s as ib t h i s Darser wlll have more ~e n e r a l u s e f o r ordlnary discourse.In t h e interests of space, I have glossed over B l a r g e number of i n t r i c a c i e s t h a t would have to be d e a l t with in a r r i v i n g at a machine-readable h n s c r i p t s u i t a b l e for analysis. Several pages would be reqyired t o describe them f u l l y .t h e main e n t r y x appears e x a c t l y or in an i n f l e c t e d form In a d e f i n i t i o n o f y, then xRy. ( T h i s does not preclude a d i s t i n c t line f o r yRx or XRX.) Therefore, we can e s t a b l i s h a point for every main entry in a dictionary and draw he appropriate d ir e c t e d lines t o form a digraph c o n s i s t i n g of the e n t i r e d i c t i m nary. (~h l s digraph may be disconnected, but probably is not.) An example., which 1 s a subggaph of the dictibnary digraph, 1 s shown in Figure 1 on the next page. Xxcept for broadcast, only the l a b e l s of each point a r e shown, b u t each represents a l l t h e d e f i n i t i o n s appearing at i t s respective main entry. The directed line fromact to broadcast corresponds t o the fact tha* "actis used to define broa@castn, since i t s token appears in "tfle a c t o f spreading abroad". In t h i s model, the token "spreadingH is not represented by a point, since it is not a main e r t r y .may be a p p l i e d in tandem; based on those placed in E. Thus, | null | null | system i s to achieve broad applicability, its dictionary lnust cover a substantial p a t of the n a t u r a l language lexicon. F o r this to occur, the developers of a system must e i t h e r c r e a t e a dictionary from scratch or be able to incorporate an existing dictxonary. Given the amount of effort that usually goes into development of an ordinary dictionary, t h e former a1 ternative is r a t h e r impractical. Bowever, l i t t l e has been done toward meetinn the l a t t e r alternative; with w n a t f o l l o w s , I will 7 describe the approach which I b e l i e v e must b e followed in transforming the contents of an ordinary dictionary for us6 In a true naturaX language system, In order t o be used in a language understanding system, a dictionary's semantic contents must be systematized in a way t h a t the sense in which a word i a being used can be identified. ( 1 ) to describe how to use the dictionary i t s e l f to move toward idhntification of the primitives, at the same time necessary %Hat the feature be discoverable within the semantic structure of a dictionary, Otherwise, there is a defect in one or more d e f i n i t i o n e , or t h e dictionary-contains some i n t e r n a l inconsistency. (Clearly, it is beyond expectation t h a t any pre-~n t dictionary will be free of these problems.) A t y p i c a l subgraph of the dlcfionary digraph using the baszc model.ancorning l i n e s which ape not shown.The r e s u l t a n t digraph f o r even a small dictionary i . S ex- or more points can be shown to be equivalent. Ihe concept, *'the a c t of spreading absoadft, has men shown t o be equivalent to "the act o f spreading over a wide arealt. If the l a t t e r phraseology appears under some main e n t r y , say distribution, t h e n bath it and t h e d e f i n i t i o n of broadcast would e v e n t u a l l y be an.slyzed in t h e same way. We will say t h a t both expressions may represent t h e same concept and hence a r e e q u i v a l e n t a t l e a s t | null | Main paper:
r8lationshlp to'efforts to represent knowledge in frames:
The process whkch has been o u t l i n e d ifi the preceding sec- It seema C.0 me t h a t this i r nothing ,more t h a n the process which has a l r e a d y been described using a graph-theoretic naproach, except t h a t the generalhxed frame f o r each verb will not be, 12. Schank, R.C. Conceptual dependency: A theory of' n a t u r a l b n g u a g e understanding. Cognitiye P a y c h o l o~ 3( 1 9 7 2 )~ 532-631 3 13. Simmons, R. F. and Slocum, 3 , G e n q a t i n g E n g l i s h discourse from se6antic networks. Commu~cations of the ACM* 15 (.1972) t e x t which must be associ.ated w i t h each sense. I t w p e a r s as ib t h i s Darser wlll have more ~e n e r a l u s e f o r ordlnary discourse.In t h e interests of space, I have glossed over B l a r g e number of i n t r i c a c i e s t h a t would have to be d e a l t with in a r r i v i n g at a machine-readable h n s c r i p t s u i t a b l e for analysis. Several pages would be reqyired t o describe them f u l l y .t h e main e n t r y x appears e x a c t l y or in an i n f l e c t e d form In a d e f i n i t i o n o f y, then xRy. ( T h i s does not preclude a d i s t i n c t line f o r yRx or XRX.) Therefore, we can e s t a b l i s h a point for every main entry in a dictionary and draw he appropriate d ir e c t e d lines t o form a digraph c o n s i s t i n g of the e n t i r e d i c t i m nary. (~h l s digraph may be disconnected, but probably is not.) An example., which 1 s a subggaph of the dictibnary digraph, 1 s shown in Figure 1 on the next page. Xxcept for broadcast, only the l a b e l s of each point a r e shown, b u t each represents a l l t h e d e f i n i t i o n s appearing at i t s respective main entry. The directed line fromact to broadcast corresponds t o the fact tha* "actis used to define broa@castn, since i t s token appears in "tfle a c t o f spreading abroad". In t h i s model, the token "spreadingH is not represented by a point, since it is not a main e r t r y .may be a p p l i e d in tandem; based on those placed in E. Thus,
:
system i s to achieve broad applicability, its dictionary lnust cover a substantial p a t of the n a t u r a l language lexicon. F o r this to occur, the developers of a system must e i t h e r c r e a t e a dictionary from scratch or be able to incorporate an existing dictxonary. Given the amount of effort that usually goes into development of an ordinary dictionary, t h e former a1 ternative is r a t h e r impractical. Bowever, l i t t l e has been done toward meetinn the l a t t e r alternative; with w n a t f o l l o w s , I will 7 describe the approach which I b e l i e v e must b e followed in transforming the contents of an ordinary dictionary for us6 In a true naturaX language system, In order t o be used in a language understanding system, a dictionary's semantic contents must be systematized in a way t h a t the sense in which a word i a being used can be identified. ( 1 ) to describe how to use the dictionary i t s e l f to move toward idhntification of the primitives, at the same time necessary %Hat the feature be discoverable within the semantic structure of a dictionary, Otherwise, there is a defect in one or more d e f i n i t i o n e , or t h e dictionary-contains some i n t e r n a l inconsistency. (Clearly, it is beyond expectation t h a t any pre-~n t dictionary will be free of these problems.) A t y p i c a l subgraph of the dlcfionary digraph using the baszc model.ancorning l i n e s which ape not shown.The r e s u l t a n t digraph f o r even a small dictionary i . S ex- or more points can be shown to be equivalent. Ihe concept, *'the a c t of spreading absoadft, has men shown t o be equivalent to "the act o f spreading over a wide arealt. If the l a t t e r phraseology appears under some main e n t r y , say distribution, t h e n bath it and t h e d e f i n i t i o n of broadcast would e v e n t u a l l y be an.slyzed in t h e same way. We will say t h a t both expressions may represent t h e same concept and hence a r e e q u i v a l e n t a t l e a s t
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0.037906 | null | null | null | null | null | null | null | null |
07a0d46cc6893db8360daefc30af98f4d3cccb2f | 219308599 | null | Intentlonallty and Human Conversations | Go a1 s , Plans and Under stand ing , Lawr ence | {
"name": [
"Carbonell Jr, Jaime G."
],
"affiliation": [
null
]
} | null | null | null | 1978-12-01 | 0 | 0 | null | and some i l l u s t r a t i v e output , a r e presented , f )hBonnula~ing r u l e s about human c o w e r sa t i a n s .-This paper i s a n empirical approach t o understanding the processes t h a t underlie hmaa conversations. Since t h e t a s k of codifying a l l the knowledge required for modeling hunan dimour SQ: is monumental, we confine our approach t o formulating r u l e s about t h e conversational. doesn t answer John's question; brushing one's t e e t h i s not "something new". Therefore, we could propose aseather simple conversational r u l e : RULFd: If a question i s asked i n the course o f a conversation, t h e otfier p a r t i c i p a n t should answer t h i s quest ion.Rule 1, h o e v e r , i s a l i t t l e too naive. Suppose B i l l ' s answer was? "There a r e a 'few more microns of dust on t h e windowsill than the l a s t time you asked m e t h a t questionv" This i s indeed "something new", but we would t h i n k o f B i l l as a wise guy f o r answering the question P i t e r a l l y r a t h e r than addressing *t John " m u s t have meanti'. % a t i s t h e t r u e s i g g l f i c a n c e o f a q m t i o n ?In Conversation FragmeAt (1 1 , B i l l might have a n s w r e d : '"I'he 3 -p a r t i c l e angular momentm o f +3/2 was confirmed today." John, a l i t e r a t u r e major who does n o t h d e r s t a n d Physics, may n o t be inclined t o continue the conversation. Therefore, c ill' a answer i s not what was c a l l e d f o r , u n l e s s B i l l i n t e n t i o n a l l y wanted to end the conversation, This example suggests t h a t B i l l missed something i n e s t a b l i s h i n g the t r u e s i g n i f i c a n c e o f John's question' John d i d , indeed, e x p l i c i t l y a s k t o hear something new; i m p l i c i t l y he meant some thing important and out of t h e ordinary. The J -p a r t i c l e 3nswer conforms t o these requirements, but it i s s t i l l an inappropriate response. Therefore, t h e true s i g n i f i c a n c e o f John's answer must include John's oanversational g o a l . m y did John a s k "What's new"? The answer is, obviously. t o s t a r t a conversation with B i l l . B i l l , being aware o f t h i s conversational goal. needs t o choose an answer t h a t attempts t o i n i t t -a t e conversation. That i s B i l l should choo'se a topic o f c o n v e r m t f a n t h a t John can t a l k about and that John be i n t e r e s t e d i n . Coqversational Rule ( 3 ) sunmarizes t h i s discussion: The process a f understanding t he :onversational b p o r t o f a n u t t e r a n c e may b e :onceptually diytded i n t o two primary oubprocesses: 1) determine the conversational goal of t h e u t t e r a n c e , and 2 ) e s t a b l i s h t h e r e a l , o f t e n i m p l i c i t , meaning of t h e utterance. k h n e r t I19771 analyzes t h e process o f e s m b l i s h i n g the r e a l meaning o f questions. Our a n a l y s l s focuses on the conversational goals € t h e p a r t i c i p a n t s and the e s t a b l i s h m w t o f a shared knowledge base betweerh! the p a r t i c i p a n t s .It is t h i s shared c u l t u r a l , p e r s o n a l , and f a c t u a l knowledge t h a t the conversational p a r t i c i p a n t s l e a v e i m p l i c i t i n each communication.To i l l u s t r a t e t h i s f a c t , consider Caper sat i o n a l Fragment l i ) My season for n o t wanting t o go i s t h a t 'I made a previous commitment, and I =annot be i n tbm pl,a&es a t once t o n i g h t .iii) The previous commitment! i s a v i s i t t o my f o l k s r i v ) I a* t e l l i q you &bout the reason why 1 cannot 80 d r i n k b g with you r a t h e r than j u s t saylrtg "no" because f d o not want you t o get angry a t me. B i l l knows t h a t John will i n t e r p r e t h i s answer so as t o conclurte its real s i g n i f i c a n c e ; otherwise B i l l wuld have chosen t o e x p l i c i t l y state the r e a l s i g n i f i c a n c e . How d o e s B i l l know t h a t John w i l l understand him c o r r e c t l y ?C l e a r l y B i l l and John must share some common sense knowledqe such as: a ) A person cannot be i n two places a t 6 ice. b) Previous commitments should be honored. Which way were you t r y i n g t o convince me to Vote?was hoping you would h e l p me make up m y mind. ( i i i 3 My parefits mll.g,h!&ggive me some Jnaney if I ask'them. ( i v ) I f I v i s i t them and a s k them i n person I have a b e t t e r chance o f g e t t i n g some money (v) I ' l l v i s i t them t o n i g h t And then I ' l l a s k them for money. p a r t r c i p a n t , addres-slng the fnr?int 0 f t h e u t t e r a n c e s of t h e o t h e r p a r t i c i p a n t .Since the reader assumes t h a t Rule ( 5 ) i a true for Conversation Fragment ( 5 ) , he concludes t h a t t h e r e must be a connection between B i l l needing, money and the v h i t t o h i s parents. The reader then i n f e r s t h e a o s t l i k e l y connection: B i l l w i l l a s k h i s paren'es for money. John must a l s o make t h i s i n f e r e n c e based on Rule ( 5 ) . unless he knows t h a t B i l l r e g u l a r l y v i s i t s h i s p a r e n t s t o a s k f o r money. The s i g n i f i c a n t -point i l l u s t r a t e d i nexample 5 i s t h a t t h e conversation focused the i n f e r e n c e mechanism t o find a connection between t he r ~p e c r l v e u t t e r a n c e s . Therefore, conversational p r i n c i p l e s can play an important r o l e i n focusing hunan reasoning processes. The )r i n c i p l e of focusing inference processes on s i g n i f i c a n t o r i n t e r e s t i n g aspect's o f :onver s a t i o n a l u t t e r a n c e s and events Meav ing of B i l l ' s u t t e r a n c e : Conversation fragments ( 6 ) , 7and 8i l l u s t r a t e the degree t o which t h e understanding of c o n v e r s a t i o n a l u t t e r a n c e s i s expectation-driven.( i ) I d o v i s i t my family. i i ) Supparting evidence: I'm going t o v i s i t them t o n i g h t . ( i i i ) Therefore what -you just said i s n o t t r u e .The e x p e c t a t i o n s a r e generated from previous Jt t e r a n c e s accordingt o r u l e 5;t h e t o p i c , i n t e n t , and c o q v e r s a t i o n a l g o a l s introduced e a r l i e r i n t h e conversation w i l l be addressed by l a t e r u t t e r a n c e s . In e a c h case t h e same u t t e r a n c e on ill's part i s understood Social r e l a t i o n s and the perceived goals o f conversational p a r t i c i p a n t s play an important r o l e i n i n t e r p r e t i n g t h e meaning of conversational utterances.Let US f i r s t consider t h e influence of t h e s o c i a l r e l a t i o n s between the two p a r t i c i p a n t s : RULE 7. The s o c i a l r e l a t i o n s h i p b e t e e n the p a r t i c i p a n t s i n a conversation generates expectations about t h e i n t e n t i o n a l meaning o f utterances i n t h e conversation. These expectations a r e used t o d e t e r r e h e the c o r n e r a kisnal goals o f each p a r t k i p a n t . 12.1) The stranger r e p l i e d : "You go two blooks toward t h a t t a l l building and t u r n r i g h t .I112.2 The question was i n t e r p r e t e d t o mean four d i f f e r e n t t h i n g s , depending on whom John spoke t o .t h e l i s t e n e r usually i n t e r p r e t s t h i s t o mean "I want t o go to X, but I do not know how t o g e t t h e r e ."If you know how t o g e t to X please take me there. c e 8 i n i n t e r p r e t a t i o n can be explained i n terms of d i f f e r e n c e s i n t h e perceived goals, o f t h e p a r t i c i p a n t s i n the conversation. | null | null | A process model of hunarr conversation.The phenomenon o f human cornrefsation i s too complex for any s i n g l e study t o do j u s t i c e to more than a narrow a s p e c t o f t h e problem. En order t o f u l l y understand hunan conversations w may have t o understand a l l hunan cognitive reasoning processes.Our r e s e a r c h a p p r o a c h c a n be o-utlined a s follows: 1) S t d y many sample conversations; 2 ) t~y t o e s t a b l i s h some r e l a t i v e l y general r u l e s o f conversation; 3) encode these r u l e s i n t o a progess model; 4 ) see i f t h i s model accounts f o r c e r t a i n a s p e c t s o f human conversation; 5) r e a l i z e t h a t F F~ solved hardly more than a minute part o f t h e problem, and 6 ) r e i t e r a t e the r e s e a r c h process i n a (hopefully p o s i t i v e ) feed-badc loop. WHAT'S NEW TODAY, DAVE?Nothing, What'a your name?MY NAME IS MXC S.What i s your job? I AM A COMPUTER PROCUMHER That's n i c e .I MICS t a k e s t h e i n i t i a t i v e by a s s e r t i n g When t h e human p a r t i c i p a n t says t h a t he l i k e s t o smoke marijuana, MICS f i n d s i n i t s memory t h a t smoking marijuana i s i l l e g a l and can l e a d t o t h e smoker g e t t i n g stoned.I i t s c~n vAccording t o MICS's beliefs g e t t i n g stoned is undesirable, and doing some thing i l l e g a known t h a t Dave knows ttpt smoking marifuana l e a d s t o g e t t i n g stoned, and t h a t , i n f a c t , g e t t i n g stoned i s b v e s primary purpose f~r smoking marijuana.I f H1CS knew t h i s f a s t t h e n Conversational Syntax Rule (10) would prevent MICS from w r n i n g Too b&, Dave i s ngt too smart anyways.I ' v e got to go and study.The above conversation i l l u s t r a t e s t h e useiulness o f shafed knowledge b e t m e n t h e t m p a r t i c i p a n t s .Pege and MICS both know about taking arams, and t h e y have a common acquaintance The first speaker usually has t h e convers a t i o n a l g o a l of wanting feedback ( u s u a l l y ag;r@ement) on h i s o~i n i o n . The l a s t -change i n the above d i a l o g i l l u s t r a t e s a d e f i c i e n c y i n MICS determination o f t h e conversational g o a l s . When Pete s a y s , "1've g o t t o go and study," he means, "]C want t o end the conversation ,I1 but i n f e r r i n g thi's conversational goal r e q u i r e s undersranding what i t means t o study. One cannot i n f e r from t h e phrase, "I've got t o go . ..", t h a t t h e conversation should t errninate . For i n s t a n c e , i f Pete says, "1've $ot The knowledge t h a t studying and t a l k i n g cannot proceed s i m u l t a n e w u s l y i s necessary i n order t o conclude t h a t Pete wanted t o end t h e conversation.The primary usefulness o f t h e MICS system i s t h a t i t sheds l i g h t on problems about t h e r u l e s of conversation t h a t might otherwise remain unnoticed.The feedback from t e s t i n g the implementation o f our conversational model forces us t o b e more p r e c i s e and complete i n our formulation o f our i d e a s about conversation.The followi* i s a conversation fraprnent from a l a t e r v e r s i o n o f MICS which has a more sophisticated i n f e r e n c e c a p a b i l i t y f o r determining the knowl&e t h a t i t sharee with t h e o t h e r speaker.A s h t r~y s .Ashtrays a r e use ful .Yes.YOU MIGHT 0 i T CANCER IF YOU SMOKE TOO MUCH.Who cares, anyway I smoke dope, n o t tobacco.I SEE, YOU LIKE TO GET HIGH.It i s b e t t e r than g e t t i n g drunk.NOT NECESSARILY; YOU MIGHT GET CAUGHT BY THE COPS, The u t t e r a n c e s a r e generated i n English b y a r a t h e r c r u d e p h r a s a l English g e n e r a t o r , The u t t e r a n c e s from t h e o t h e r speaker a r e analyzed fw t h e i r meaning and c o n v e r s a t i o n a l form by a p r i m i t i v e , key-concept o r i e n t e d a n a l y z e r . i s , on Occasion, e x p l o i t e d i n cateating c e r t a i n types of jokes, a s i n example 4: 4) SON: Dad, L robbed the l i q u o r s t o r e yesterday DAD: How could you ever do such a t h i n g , son. SON: Well, I g o t me t h i s gun, and I pointed a t the c a s h i e r . . .To i l l u s t r a t e t h e fmportance o f t h e i m p l i c i t conversational goals-and shared knowledge b e t m e n t h e p a r t i c i p a n t s i n a conversation, we present a few more dialog framents b e t w e n B % l l and John, t h e two c o l l e g e s t u d e n t s sharing an apartment. In each e x a p l e , a s i n epmre<sations ( 1 ) and ( 2 ) , BiU u t t e r s the same response, but i t s meaning i s s i g n i f i c a n t l y d i f f e r e n t , depending on t h e context o f t h e conversation.5) JOHN: Are you broke again? You a r e going t o h'ave t o eome up with your s h a r e o f t h e rent t h i s donth. BILL: I ' m going t o visit: my f o l k s w n t g h t . | null | Main paper:
mics::
A process model of hunarr conversation.The phenomenon o f human cornrefsation i s too complex for any s i n g l e study t o do j u s t i c e to more than a narrow a s p e c t o f t h e problem. En order t o f u l l y understand hunan conversations w may have t o understand a l l hunan cognitive reasoning processes.Our r e s e a r c h a p p r o a c h c a n be o-utlined a s follows: 1) S t d y many sample conversations; 2 ) t~y t o e s t a b l i s h some r e l a t i v e l y general r u l e s o f conversation; 3) encode these r u l e s i n t o a progess model; 4 ) see i f t h i s model accounts f o r c e r t a i n a s p e c t s o f human conversation; 5) r e a l i z e t h a t F F~ solved hardly more than a minute part o f t h e problem, and 6 ) r e i t e r a t e the r e s e a r c h process i n a (hopefully p o s i t i v e ) feed-badc loop. WHAT'S NEW TODAY, DAVE?Nothing, What'a your name?MY NAME IS MXC S.What i s your job? I AM A COMPUTER PROCUMHER That's n i c e .I MICS t a k e s t h e i n i t i a t i v e by a s s e r t i n g When t h e human p a r t i c i p a n t says t h a t he l i k e s t o smoke marijuana, MICS f i n d s i n i t s memory t h a t smoking marijuana i s i l l e g a l and can l e a d t o t h e smoker g e t t i n g stoned.I i t s c~n vAccording t o MICS's beliefs g e t t i n g stoned is undesirable, and doing some thing i l l e g a known t h a t Dave knows ttpt smoking marifuana l e a d s t o g e t t i n g stoned, and t h a t , i n f a c t , g e t t i n g stoned i s b v e s primary purpose f~r smoking marijuana.I f H1CS knew t h i s f a s t t h e n Conversational Syntax Rule (10) would prevent MICS from w r n i n g Too b&, Dave i s ngt too smart anyways.I ' v e got to go and study.The above conversation i l l u s t r a t e s t h e useiulness o f shafed knowledge b e t m e n t h e t m p a r t i c i p a n t s .Pege and MICS both know about taking arams, and t h e y have a common acquaintance The first speaker usually has t h e convers a t i o n a l g o a l of wanting feedback ( u s u a l l y ag;r@ement) on h i s o~i n i o n . The l a s t -change i n the above d i a l o g i l l u s t r a t e s a d e f i c i e n c y i n MICS determination o f t h e conversational g o a l s . When Pete s a y s , "1've g o t t o go and study," he means, "]C want t o end the conversation ,I1 but i n f e r r i n g thi's conversational goal r e q u i r e s undersranding what i t means t o study. One cannot i n f e r from t h e phrase, "I've got t o go . ..", t h a t t h e conversation should t errninate . For i n s t a n c e , i f Pete says, "1've $ot The knowledge t h a t studying and t a l k i n g cannot proceed s i m u l t a n e w u s l y i s necessary i n order t o conclude t h a t Pete wanted t o end t h e conversation.The primary usefulness o f t h e MICS system i s t h a t i t sheds l i g h t on problems about t h e r u l e s of conversation t h a t might otherwise remain unnoticed.The feedback from t e s t i n g the implementation o f our conversational model forces us t o b e more p r e c i s e and complete i n our formulation o f our i d e a s about conversation.The followi* i s a conversation fraprnent from a l a t e r v e r s i o n o f MICS which has a more sophisticated i n f e r e n c e c a p a b i l i t y f o r determining the knowl&e t h a t i t sharee with t h e o t h e r speaker.A s h t r~y s .Ashtrays a r e use ful .Yes.YOU MIGHT 0 i T CANCER IF YOU SMOKE TOO MUCH.Who cares, anyway I smoke dope, n o t tobacco.I SEE, YOU LIKE TO GET HIGH.It i s b e t t e r than g e t t i n g drunk.NOT NECESSARILY; YOU MIGHT GET CAUGHT BY THE COPS, The u t t e r a n c e s a r e generated i n English b y a r a t h e r c r u d e p h r a s a l English g e n e r a t o r , The u t t e r a n c e s from t h e o t h e r speaker a r e analyzed fw t h e i r meaning and c o n v e r s a t i o n a l form by a p r i m i t i v e , key-concept o r i e n t e d a n a l y z e r . i s , on Occasion, e x p l o i t e d i n cateating c e r t a i n types of jokes, a s i n example 4: 4) SON: Dad, L robbed the l i q u o r s t o r e yesterday DAD: How could you ever do such a t h i n g , son. SON: Well, I g o t me t h i s gun, and I pointed a t the c a s h i e r . . .To i l l u s t r a t e t h e fmportance o f t h e i m p l i c i t conversational goals-and shared knowledge b e t m e n t h e p a r t i c i p a n t s i n a conversation, we present a few more dialog framents b e t w e n B % l l and John, t h e two c o l l e g e s t u d e n t s sharing an apartment. In each e x a p l e , a s i n epmre<sations ( 1 ) and ( 2 ) , BiU u t t e r s the same response, but i t s meaning i s s i g n i f i c a n t l y d i f f e r e n t , depending on t h e context o f t h e conversation.5) JOHN: Are you broke again? You a r e going t o h'ave t o eome up with your s h a r e o f t h e rent t h i s donth. BILL: I ' m going t o visit: my f o l k s w n t g h t .
:
and some i l l u s t r a t i v e output , a r e presented , f )hBonnula~ing r u l e s about human c o w e r sa t i a n s .-This paper i s a n empirical approach t o understanding the processes t h a t underlie hmaa conversations. Since t h e t a s k of codifying a l l the knowledge required for modeling hunan dimour SQ: is monumental, we confine our approach t o formulating r u l e s about t h e conversational. doesn t answer John's question; brushing one's t e e t h i s not "something new". Therefore, we could propose aseather simple conversational r u l e : RULFd: If a question i s asked i n the course o f a conversation, t h e otfier p a r t i c i p a n t should answer t h i s quest ion.Rule 1, h o e v e r , i s a l i t t l e too naive. Suppose B i l l ' s answer was? "There a r e a 'few more microns of dust on t h e windowsill than the l a s t time you asked m e t h a t questionv" This i s indeed "something new", but we would t h i n k o f B i l l as a wise guy f o r answering the question P i t e r a l l y r a t h e r than addressing *t John " m u s t have meanti'. % a t i s t h e t r u e s i g g l f i c a n c e o f a q m t i o n ?In Conversation FragmeAt (1 1 , B i l l might have a n s w r e d : '"I'he 3 -p a r t i c l e angular momentm o f +3/2 was confirmed today." John, a l i t e r a t u r e major who does n o t h d e r s t a n d Physics, may n o t be inclined t o continue the conversation. Therefore, c ill' a answer i s not what was c a l l e d f o r , u n l e s s B i l l i n t e n t i o n a l l y wanted to end the conversation, This example suggests t h a t B i l l missed something i n e s t a b l i s h i n g the t r u e s i g n i f i c a n c e o f John's question' John d i d , indeed, e x p l i c i t l y a s k t o hear something new; i m p l i c i t l y he meant some thing important and out of t h e ordinary. The J -p a r t i c l e 3nswer conforms t o these requirements, but it i s s t i l l an inappropriate response. Therefore, t h e true s i g n i f i c a n c e o f John's answer must include John's oanversational g o a l . m y did John a s k "What's new"? The answer is, obviously. t o s t a r t a conversation with B i l l . B i l l , being aware o f t h i s conversational goal. needs t o choose an answer t h a t attempts t o i n i t t -a t e conversation. That i s B i l l should choo'se a topic o f c o n v e r m t f a n t h a t John can t a l k about and that John be i n t e r e s t e d i n . Coqversational Rule ( 3 ) sunmarizes t h i s discussion: The process a f understanding t he :onversational b p o r t o f a n u t t e r a n c e may b e :onceptually diytded i n t o two primary oubprocesses: 1) determine the conversational goal of t h e u t t e r a n c e , and 2 ) e s t a b l i s h t h e r e a l , o f t e n i m p l i c i t , meaning of t h e utterance. k h n e r t I19771 analyzes t h e process o f e s m b l i s h i n g the r e a l meaning o f questions. Our a n a l y s l s focuses on the conversational goals € t h e p a r t i c i p a n t s and the e s t a b l i s h m w t o f a shared knowledge base betweerh! the p a r t i c i p a n t s .It is t h i s shared c u l t u r a l , p e r s o n a l , and f a c t u a l knowledge t h a t the conversational p a r t i c i p a n t s l e a v e i m p l i c i t i n each communication.To i l l u s t r a t e t h i s f a c t , consider Caper sat i o n a l Fragment l i ) My season for n o t wanting t o go i s t h a t 'I made a previous commitment, and I =annot be i n tbm pl,a&es a t once t o n i g h t .iii) The previous commitment! i s a v i s i t t o my f o l k s r i v ) I a* t e l l i q you &bout the reason why 1 cannot 80 d r i n k b g with you r a t h e r than j u s t saylrtg "no" because f d o not want you t o get angry a t me. B i l l knows t h a t John will i n t e r p r e t h i s answer so as t o conclurte its real s i g n i f i c a n c e ; otherwise B i l l wuld have chosen t o e x p l i c i t l y state the r e a l s i g n i f i c a n c e . How d o e s B i l l know t h a t John w i l l understand him c o r r e c t l y ?C l e a r l y B i l l and John must share some common sense knowledqe such as: a ) A person cannot be i n two places a t 6 ice. b) Previous commitments should be honored. Which way were you t r y i n g t o convince me to Vote?was hoping you would h e l p me make up m y mind. ( i i i 3 My parefits mll.g,h!&ggive me some Jnaney if I ask'them. ( i v ) I f I v i s i t them and a s k them i n person I have a b e t t e r chance o f g e t t i n g some money (v) I ' l l v i s i t them t o n i g h t And then I ' l l a s k them for money. p a r t r c i p a n t , addres-slng the fnr?int 0 f t h e u t t e r a n c e s of t h e o t h e r p a r t i c i p a n t .Since the reader assumes t h a t Rule ( 5 ) i a true for Conversation Fragment ( 5 ) , he concludes t h a t t h e r e must be a connection between B i l l needing, money and the v h i t t o h i s parents. The reader then i n f e r s t h e a o s t l i k e l y connection: B i l l w i l l a s k h i s paren'es for money. John must a l s o make t h i s i n f e r e n c e based on Rule ( 5 ) . unless he knows t h a t B i l l r e g u l a r l y v i s i t s h i s p a r e n t s t o a s k f o r money. The s i g n i f i c a n t -point i l l u s t r a t e d i nexample 5 i s t h a t t h e conversation focused the i n f e r e n c e mechanism t o find a connection between t he r ~p e c r l v e u t t e r a n c e s . Therefore, conversational p r i n c i p l e s can play an important r o l e i n focusing hunan reasoning processes. The )r i n c i p l e of focusing inference processes on s i g n i f i c a n t o r i n t e r e s t i n g aspect's o f :onver s a t i o n a l u t t e r a n c e s and events Meav ing of B i l l ' s u t t e r a n c e : Conversation fragments ( 6 ) , 7and 8i l l u s t r a t e the degree t o which t h e understanding of c o n v e r s a t i o n a l u t t e r a n c e s i s expectation-driven.( i ) I d o v i s i t my family. i i ) Supparting evidence: I'm going t o v i s i t them t o n i g h t . ( i i i ) Therefore what -you just said i s n o t t r u e .The e x p e c t a t i o n s a r e generated from previous Jt t e r a n c e s accordingt o r u l e 5;t h e t o p i c , i n t e n t , and c o q v e r s a t i o n a l g o a l s introduced e a r l i e r i n t h e conversation w i l l be addressed by l a t e r u t t e r a n c e s . In e a c h case t h e same u t t e r a n c e on ill's part i s understood Social r e l a t i o n s and the perceived goals o f conversational p a r t i c i p a n t s play an important r o l e i n i n t e r p r e t i n g t h e meaning of conversational utterances.Let US f i r s t consider t h e influence of t h e s o c i a l r e l a t i o n s between the two p a r t i c i p a n t s : RULE 7. The s o c i a l r e l a t i o n s h i p b e t e e n the p a r t i c i p a n t s i n a conversation generates expectations about t h e i n t e n t i o n a l meaning o f utterances i n t h e conversation. These expectations a r e used t o d e t e r r e h e the c o r n e r a kisnal goals o f each p a r t k i p a n t . 12.1) The stranger r e p l i e d : "You go two blooks toward t h a t t a l l building and t u r n r i g h t .I112.2 The question was i n t e r p r e t e d t o mean four d i f f e r e n t t h i n g s , depending on whom John spoke t o .t h e l i s t e n e r usually i n t e r p r e t s t h i s t o mean "I want t o go to X, but I do not know how t o g e t t h e r e ."If you know how t o g e t to X please take me there. c e 8 i n i n t e r p r e t a t i o n can be explained i n terms of d i f f e r e n c e s i n t h e perceived goals, o f t h e p a r t i c i p a n t s i n the conversation.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 554 | 0 | null | null | null | null | null | null | null | null |
d9508771de1f0e28846ee674544ece0cbe3e07a4 | 57350096 | null | Toward a Computational Theory of Speech Perception | In recent years,a great deal of evidence has been collected which gives substantially increased insight into the nature of human speech perception. It is the author's belief that such data can be effectively used to infer much of the structure of a practical speech recognition system. This paper details a new view of the role of structural constraints within the several structural domains (e.g. articulation, phonetics, phonology, syntax, semantics) that must be utilized to infer the desired percept. | {
"name": [
"All, Jonathan"
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 0 | 0 | null | Each of the structural domains mentioned above has a substantial "internal theory" describing the constraints within that domain, but there are also many interactions between structural domains which must be considered. Thus words llke "incline" and "survey" shift stress with syntactic role, and there is a pragmatic bias for the ambiguous sentence "John called the boy who has smashed his car up." to be interpreted under a strategy that reflects a tendency for local completion of syntactic structures. It is clear, then, that while analysis within a structural domain (e.g. syntactic parsing) can be performed up to a point,lnteraction with other domains and integration of constraint strengths across these domains is needed for correct perception. The various constraints have differing and changing strengths at different points in an utterance, so that no fixed metric can be used to determine their contribution to the wellformedness of the utterance.At the segmental level, many diverse cues for segmental features have been found. As many as 16 cues mark the voicing distinction, for example.We may think of each of these cues as also representing a constraint, and the strength of the constraint varies with the context. For example, stop closure duration must be interpreted in the context of the local rate of speech, and a given value of closure duration can signify either a voiced or an unvoiced stop depending on the surrounding vowel durations. Thus several cues must be integrated to obtain the perceived segmental feature, and the weights assigned to each cue vary with the local context.From the preceding examples, it is seen that in order to model human speech perception, it is necessary to dynamically integrate a wide variety of constraints. The evidence argues strongly for an active focussed search, whereby the perceptual mechanism knows, as the utterance unfolds, where the strongest constraint strengths are, and uses this reliable information, while ignoring "cues" that are unreliable or non-determining in the immediate context.For example, shadowing experiments have shown that listeners (performing the shadowing task) can restore disrupted words to their original form by using semantic and syntactic context, thus demonstrating the integration process.Furthermore, techniques are now available for analytically finding that informatlon in an input stimulus which can maximally discrimlnate between two candidate prototypes, so that the perceptual control structure can focus only on such information co make a choice between the candidates. In this paper, we develop a theory for speech recognition which contains the required dynamic integration capability coupled with the ability to focus on a restricted set of cues which has been contextually selected.The model of speech recognition which we have developed requires, of course, an initial low-level analysis of the speech waveform to get started.We argue from the recent psychollnguistic literature that stressed syllables provide the required entry points.Stressed syllable peaks can be readily located, and use of the phonotactics of segmental distribution within syllables, together with the relatively clear articulation of syllable-initial consonants, allows us to formulate a robust procedure for determining initial segmental "islands", around which further analysis can proceed. In fact, there is evidence to indicate that the human lexicon is organized and accessed via these stressed syllables.The restriction of the original analysis to these stressed syllables can be regarded as another form of focussed search, which in turn leads to additional searches dictated by the relative constraint strengths of the various domains contributing to the percept. We argue that these views are not only consonant with the current knowledge of human speech perceptlon, but form the proper basis for the design of hlgh-performance Speech recognition systems. | null | null | null | null | Main paper:
:
Each of the structural domains mentioned above has a substantial "internal theory" describing the constraints within that domain, but there are also many interactions between structural domains which must be considered. Thus words llke "incline" and "survey" shift stress with syntactic role, and there is a pragmatic bias for the ambiguous sentence "John called the boy who has smashed his car up." to be interpreted under a strategy that reflects a tendency for local completion of syntactic structures. It is clear, then, that while analysis within a structural domain (e.g. syntactic parsing) can be performed up to a point,lnteraction with other domains and integration of constraint strengths across these domains is needed for correct perception. The various constraints have differing and changing strengths at different points in an utterance, so that no fixed metric can be used to determine their contribution to the wellformedness of the utterance.At the segmental level, many diverse cues for segmental features have been found. As many as 16 cues mark the voicing distinction, for example.We may think of each of these cues as also representing a constraint, and the strength of the constraint varies with the context. For example, stop closure duration must be interpreted in the context of the local rate of speech, and a given value of closure duration can signify either a voiced or an unvoiced stop depending on the surrounding vowel durations. Thus several cues must be integrated to obtain the perceived segmental feature, and the weights assigned to each cue vary with the local context.From the preceding examples, it is seen that in order to model human speech perception, it is necessary to dynamically integrate a wide variety of constraints. The evidence argues strongly for an active focussed search, whereby the perceptual mechanism knows, as the utterance unfolds, where the strongest constraint strengths are, and uses this reliable information, while ignoring "cues" that are unreliable or non-determining in the immediate context.For example, shadowing experiments have shown that listeners (performing the shadowing task) can restore disrupted words to their original form by using semantic and syntactic context, thus demonstrating the integration process.Furthermore, techniques are now available for analytically finding that informatlon in an input stimulus which can maximally discrimlnate between two candidate prototypes, so that the perceptual control structure can focus only on such information co make a choice between the candidates. In this paper, we develop a theory for speech recognition which contains the required dynamic integration capability coupled with the ability to focus on a restricted set of cues which has been contextually selected.The model of speech recognition which we have developed requires, of course, an initial low-level analysis of the speech waveform to get started.We argue from the recent psychollnguistic literature that stressed syllables provide the required entry points.Stressed syllable peaks can be readily located, and use of the phonotactics of segmental distribution within syllables, together with the relatively clear articulation of syllable-initial consonants, allows us to formulate a robust procedure for determining initial segmental "islands", around which further analysis can proceed. In fact, there is evidence to indicate that the human lexicon is organized and accessed via these stressed syllables.The restriction of the original analysis to these stressed syllables can be regarded as another form of focussed search, which in turn leads to additional searches dictated by the relative constraint strengths of the various domains contributing to the percept. We argue that these views are not only consonant with the current knowledge of human speech perceptlon, but form the proper basis for the design of hlgh-performance Speech recognition systems.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 548 | 0 | null | null | null | null | null | null | null | null |
28505ffec87c658a643fe66b1573f240cb73fbcc | 218681456 | null | Applications | Truth, like beauty, is in the eye of the beholder, Z offer a few remarks for the use of those who seek a point of view from which to see truth in the six papers assigned to this session. | {
"name": [
"Hays, David G."
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 2 | 0 | null | Linguistic computation is the fundamental and primitive branch of the art of cumputatlon~ as I have remarked off and on. The insight of yon Neumann~ that operations and data can be represented in the same storage device, is the linguistic insight that anything can have a name in any language.(Whether anything can have a definition is a different question.)I recall surprising a couple of colleagues with this r~ark early in the 1960s, when I had to point out the obvious fact that compillng and interpreting are linguistic procedures and therefore that only in rare instances does a computer spend more time on mathematics than on linguistics.By now we all take the central position of our subject matter for granted.I express this overly familiar truth only for the pragmatic reason that some familiar truths are more helpful than others in preparing for a given discourse.Syntax needs semantic Justification, but semantics has the inherent Justification that knowledge is power. The semantic Justification of syntax is easy: Who would try to represent knowledge without a good gr~--,-r? I have not yet found a better illustration than the tlmstable~ an example that I have used for some years now. Without rules of arrangement and interpretation, the timetable collapses into a llst of places, the digits 0,..9, a~d a few speclal symbols. Almost all of the information in a timetable is conveyed by the syntax, and one suspects that the same is true of the languages of brains, minds, and computers. Syntax needs more than semantic Justification, and pra 8m-tlcs is ready to serve. Without pragmatic Justification, the difference between cognitive and syntactic structures is ridiculous.We may find more Justifiers later, but the rediscovery of pragmatlce is a boon to those who grow tired of hearing language maligned, It is easy to make fun of Engllsh , the language of Shakes spears, Bertrand Russell, and modern science. But the humor sometimes depends on the ignorance of the Joker. We find first semantic, then prasmatic, and perhaps later other kinds of Justification for the quirkiness of English and other languages, and the Jokes loss their point.Form, not content, admits of calculation.Since Aristotle proceeded in accordance with this rule, I find it surprising that John Locke omitted mention of the simple ideas in reflectlon.(One may recall that Locke knew of simple ideas in perceptlon~-~ellow, warm# amoot~nd considered knowledge to derive from perception and reflection.)Listing the sidle ideas in reflection selml in fact to be a task for our century, anticipated in part in the L9th century. Predication, Ins~an~isClonp membership, component, g, denoCation~ localization, moralization are some candidates that presently show strength.A more precise statement is that specific and not general knowledge fixes our interpretations of what we encounter, certainly in language and probably also in other channels of peroeptlon. Thus the great body of knowledge of our culture I of the individual mind, or of the ~asslve database makes lends an appearance of fixedness and stability to the world that simpler minds, cultures, and co~uters cannot get. The general rules of syntax, semantics, and pragmatlcs define the thinkable, allowing ambiguity wheQ some specific issue comes up. In a hash house or a conversation, understanding and trust come with complete and exact information.Conversation is a social activity. The thinking computer (Raphael's title) may be an artificial mind, but the conversing computer (William D. Orr's cltle) is an artificial person and must accept the obligations of social converse.Those obligations are massive: "to do justice and love mercy*', "to do unto others as you would have them do unto you", to act only as ic would be well for all to act, to express fully and concisely what is relevant, "to tell the truth, the whole truth, and nothing hut the truth".Lest anyone suppose chat I have listed the precepts of our greatest masters in a spirit of fun, I hasten to add this obvious truth from study of our species. Whether the sciences be called social, behavioral, or human, they tell us that one accepts knowledge for one's own store only from sources that can be trusted. Nor could wisdom dictate the opposite, since internalized knowledge is inaccessible to test and correction.Is the computer worthy of trust?I have asked this question of students, grading the context from simple arithmetic trust (they trust their pocket calculators to give accurate sums and products) co complex personal trust (they would not accept the computer as a friend).We have, I chink, no experience with computers that are functionally worthy of crust in any but simple matters. We may be learning to make computers follow the masters' precepts in conversation.Whether their users will ever accept them for what they are worth is hard to predict.If computers grow trustworthy and are assigned important tasks, then when crisis occurs the issue of trust may determine such outcomes as war or peace. Thus the issue is not frivolous.Trust arises from knowledge of origin as well as from knowledge of functional capacity. Genetic and cultural history provide enormous confirmation that a neighbor can be trusted, beyond even broad experience.We can gain only a little knowledge about a friend in the course of a friendship, but we can bring to bear all of our own inherent mechanisms of trust for those that look and smell llke us when crisis occurs.The six papers in this session, written by human beings and selected by persons of authority~ deserve sufficient true~ that the reader may learn from them. The systems that they describe may grow into knowledgeable, semantically and pragmatically effective, syntactically wellformed conversents.Their contributions are to that end, and have the advantage that, by seeking to apply knowledge they can detect its limits.Science needs application, since contact with reallt 7 tends to realnd us scientists that there are more things out there than are dreamed of in our theories. | null | null | null | null | Main paper:
:
Linguistic computation is the fundamental and primitive branch of the art of cumputatlon~ as I have remarked off and on. The insight of yon Neumann~ that operations and data can be represented in the same storage device, is the linguistic insight that anything can have a name in any language.(Whether anything can have a definition is a different question.)I recall surprising a couple of colleagues with this r~ark early in the 1960s, when I had to point out the obvious fact that compillng and interpreting are linguistic procedures and therefore that only in rare instances does a computer spend more time on mathematics than on linguistics.By now we all take the central position of our subject matter for granted.I express this overly familiar truth only for the pragmatic reason that some familiar truths are more helpful than others in preparing for a given discourse.Syntax needs semantic Justification, but semantics has the inherent Justification that knowledge is power. The semantic Justification of syntax is easy: Who would try to represent knowledge without a good gr~--,-r? I have not yet found a better illustration than the tlmstable~ an example that I have used for some years now. Without rules of arrangement and interpretation, the timetable collapses into a llst of places, the digits 0,..9, a~d a few speclal symbols. Almost all of the information in a timetable is conveyed by the syntax, and one suspects that the same is true of the languages of brains, minds, and computers. Syntax needs more than semantic Justification, and pra 8m-tlcs is ready to serve. Without pragmatic Justification, the difference between cognitive and syntactic structures is ridiculous.We may find more Justifiers later, but the rediscovery of pragmatlce is a boon to those who grow tired of hearing language maligned, It is easy to make fun of Engllsh , the language of Shakes spears, Bertrand Russell, and modern science. But the humor sometimes depends on the ignorance of the Joker. We find first semantic, then prasmatic, and perhaps later other kinds of Justification for the quirkiness of English and other languages, and the Jokes loss their point.Form, not content, admits of calculation.Since Aristotle proceeded in accordance with this rule, I find it surprising that John Locke omitted mention of the simple ideas in reflectlon.(One may recall that Locke knew of simple ideas in perceptlon~-~ellow, warm# amoot~nd considered knowledge to derive from perception and reflection.)Listing the sidle ideas in reflection selml in fact to be a task for our century, anticipated in part in the L9th century. Predication, Ins~an~isClonp membership, component, g, denoCation~ localization, moralization are some candidates that presently show strength.A more precise statement is that specific and not general knowledge fixes our interpretations of what we encounter, certainly in language and probably also in other channels of peroeptlon. Thus the great body of knowledge of our culture I of the individual mind, or of the ~asslve database makes lends an appearance of fixedness and stability to the world that simpler minds, cultures, and co~uters cannot get. The general rules of syntax, semantics, and pragmatlcs define the thinkable, allowing ambiguity wheQ some specific issue comes up. In a hash house or a conversation, understanding and trust come with complete and exact information.Conversation is a social activity. The thinking computer (Raphael's title) may be an artificial mind, but the conversing computer (William D. Orr's cltle) is an artificial person and must accept the obligations of social converse.Those obligations are massive: "to do justice and love mercy*', "to do unto others as you would have them do unto you", to act only as ic would be well for all to act, to express fully and concisely what is relevant, "to tell the truth, the whole truth, and nothing hut the truth".Lest anyone suppose chat I have listed the precepts of our greatest masters in a spirit of fun, I hasten to add this obvious truth from study of our species. Whether the sciences be called social, behavioral, or human, they tell us that one accepts knowledge for one's own store only from sources that can be trusted. Nor could wisdom dictate the opposite, since internalized knowledge is inaccessible to test and correction.Is the computer worthy of trust?I have asked this question of students, grading the context from simple arithmetic trust (they trust their pocket calculators to give accurate sums and products) co complex personal trust (they would not accept the computer as a friend).We have, I chink, no experience with computers that are functionally worthy of crust in any but simple matters. We may be learning to make computers follow the masters' precepts in conversation.Whether their users will ever accept them for what they are worth is hard to predict.If computers grow trustworthy and are assigned important tasks, then when crisis occurs the issue of trust may determine such outcomes as war or peace. Thus the issue is not frivolous.Trust arises from knowledge of origin as well as from knowledge of functional capacity. Genetic and cultural history provide enormous confirmation that a neighbor can be trusted, beyond even broad experience.We can gain only a little knowledge about a friend in the course of a friendship, but we can bring to bear all of our own inherent mechanisms of trust for those that look and smell llke us when crisis occurs.The six papers in this session, written by human beings and selected by persons of authority~ deserve sufficient true~ that the reader may learn from them. The systems that they describe may grow into knowledgeable, semantically and pragmatically effective, syntactically wellformed conversents.Their contributions are to that end, and have the advantage that, by seeking to apply knowledge they can detect its limits.Science needs application, since contact with reallt 7 tends to realnd us scientists that there are more things out there than are dreamed of in our theories.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 548 | 0 | null | null | null | null | null | null | null | null |
bc4d556b7227c355c7fdbb046aef4af72cacb8d8 | 216847731 | null | Where Questions | Consider question (i), and the answers to it, (2)-(h)~ (i) Where is the Empire State Building? (2) In New York. | {
"name": [
"Shanon, Benny"
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 3 | 16 | null | When (i) is posed in California 2is the appropriate answer to it. This is the case even though (3) and (h) are also true characterizations of the location of the Empire State Building. The pattern of appropriateness alters, however, when the locale where the question presented changes. Thus, when (i) is asked in Israel, 3is the appropriate answer, whereas when it is asked in Manhattan, (I~) is the answer that should be given.The foregoing observations, originally made by Rumelhart (197h) and by Norman (1973) , suggest the following. First, it is not enough for answers to questions to be (semantically) true, they have to be (pragmatically) appropriate as well. Second, appropriateness is not solely determined by the content of the particular propositions in question, but also by the identity of the participants in the particular conversational situation and their locale.In other words, for a person--or for a machine, for that matter--to answer questions, it is not enough to survey one's memory and retrieve information pertaining to the query posed, rather--a selection algorithm has to be used so that an appropriate response would be given. The specification of such a selection algorithm is the topic of the present investigation.The following discussion is based on what is known as the Room Theory: the original, albeit preliminary, model proposed by Rumelhart(197~) in order to account for his insightful observations. I try to examine the psychological validity of this model, and to propose amendments and extensions to it on the basis of empirical data.Theory is on two counts. First, there is the seemingly trivial observation that answers tO different questions are given on diff,:rent levels. Specifically, there is a correlation between the level of the object which is queried and the room on which the respective answer is given. Second, and less trivial, is the observation that answers vary not only with the questions, but also with the spatial relationship which holds between the object of the question and the participants in the conversation. Several loci in the data are indicative of this last pattern. First none of the Americans indicated that the Empire State Building was "in the U.S.", bu~ a third of the Israelis did so; further, some of the Americans, but none of the Israelis, indicated that the building was "in Manhattan". Second, asked about New York City, almost all Israelis, but none of the Americans, answered on the country level. Further, the distribution of the answer patterns furnished by the members of each group changed according to whether the queried city was their own, or close/distant from it. Finally, children's answers to questions about objects also vary with how distant the object is.Above, however, I have qualified the correspondence between the data and the theory; this qualification should now be clarified. I don't think it is meaningful to Judge the validity of a model llke the Room Theory by examining the percentage of cases in which its predictions hold. Such a percentage may reflect the structure of the domain (questions) under investigation, and it need not be indicative of the adequacy of the model as such. The term "by and large" is, however, of qualitative significance. It indicates that unless other factors or reasons are operative, answers to where questions do, indeed, follow the Room A/gorithm. The detection of these "other factors and reasons", their classification and the characterization of the answer types that correspond to them is the main theme of this discussion. Following, then, are the answer patterns which do not conform with the Room Theory.The Room Theory "posits the existence of a psychological room relative to which distances are reckoned. The room corresponds to the smallest geographical region that encompasses both the reference location of the conversants and the location of the places in question". When answering where-questions "the rule is to find the smallest room which Just includes the reference location and the answer location. The appropriate answer is the next smallest geographical unit which contains the location in question, but axcludes the reference location". (Rumelhart, 1975) . The answers generated by this algorithm, note, constitute the placing of the item questioned in a room which is larger than it; henceforth answers of this type will be called vertical.In order to examine the Room Theory, questions regarding places in the world as well as objects in a (concrete) room were presented to several subject populations: college students in Israel and the U.S., American children of three age groups, and aphasic patients. The present report concentrates on the adult data, and only cursory remarks will be made on the answers furnished by the other populations.First, I will discuss answers solicited by an open questionnaire, in which subjects were asked to give one answer to the questions posel to them; later, answers solicited by closed questionnaires will be discussed.First, it should be noted that by and large the answers given by subjects were the ones predicted by the Room First, consider questions about landm-~ks in the towns in which the conversation took place. Most of the answers which involved vertical placement were given on the level of the town itself, i.e. on a level which is higher than the one predicted by the Room Theory. The other answers were not vertical, but rather horizontal:the object questioned was related to another object similar to it. In other words, either the level specified by the Room Theory was changed, or the type of answer (i.e. the generation algorithm itself) was altered. These deviant answers are viewed as two 81ternative solutions to the problem of the floor effect. Specifically, as one goes down the place hierarchy, the specification of rooms between the target and the least common room is cumbersome; indeed, there might not be simple names by which reference to these rooms may be made. Subjects solve this problem either by staying on the level of the least common room or by shifting to the horizontal strate~,.The same problem is noted with the ceilin~ effect, namely, with questions regarding objects which are very high on the place hierarchy: continents for adults, countries for children and aphasic patients. The answers in these cases were varied, a feature which attests the algOr-ithm/c difficulty associated with them. Only a minority of the answers conformed with the Room Theory and most answers were horizontal. Other answer types were: vacuous, in which a vertical answer was given on too high a level (e.g. "in the world"), featural, in which a description, rather than a specification of the locale, was given (e.g. "it is a continent"), or tautological (e.g. "Japan is in Japan").The di~'ferent answer types, we shall say, are the products o~ different alternative answer generation algorithms.The numerical distribution of these answers suggest that the order of preference for the application of the algorithms as the one noted above.There were also cases in which subjects gave answers on a level lower than the one predicted by the Room Theory. Thus, half the Israelis placed the Empire State Building Tin New York", and not "in the U.S." Similarly, all the Americans asked about the Eiffel Tower answered "in Paris", and not "in France". These patterns are attributed to p romlnence. Prominent objects are ones which gain a higher ra-~ in the place h/erarchy than would be attributed to them on semantic classificatory grounds alone. As a consequence, these objects are placed in a room which is more specific then the one predicted by the Room Theory. For instance, New York City is not conceived of by non-Americans as Just another American city; it gains an autonomy of its own and is conceived of as independent of the country in which it is located. The prominence effect suggests that rather than interpreting the room-hierarchy in a concrete fashion (i.e. as isomorphic to the spatial relations which hold in the physical world), one should view it as an abstract conceptual representation.In this representation, ob-Jects ere associated with ta~s: usually, objects which are actually contained in objects of order n are assigned a tag of order n÷l, but prominent objects are assigned tags of the same order as the objects which actually contain them. Thus, if the Empire State Building is tagged n÷l both New York and the U.S. are tngged n, for the Israelis the least com-~n room (order n-l) is the northern hemisphere, and the answer is given on the level of the t~o rooms of order n.Thus, the seemingly unexpected answers associated with prominent objects are due to the modified abstract representations, not to a change in the (vertical) algorithm proper.The salience effect is similar, but distinct. Objects which are close to ones which stand ia a particular relation to the respondent (i.e. physically close, emotionally dear, or belonging to the subject) are not placed in a room but receive horizontal answers instead. For example, all the Israelis answered that Lebanon was "north of Israel", and not that it was "in the Mid-East". Similarly, all the Americans (and half of the Israells) placed Canada in relation to the U.S. Unlike the prominence effect, the salience effect does affect the answer generation algorithm itself, and it bears on individual or cultural differences, not on general semantic com-lderatlons. 4 Specifically, items which ere special to the speaker are tagged in the representation as marked, and this triggers a shift from the vertical to the horizontal algorlthm.All questions considered so far involved one configuration: the two conversants and the target were physically distinct, and together they could he contained in one COmmOn room. This, however, is not the only possible con£iguratlon. Other confi&~aratious, are possible as well: (a) The conversants and the target may coincide in place, as in the question '~here are we now?". (b) The conversants ~ay be contained in the target, as in the question "Where is Israel?" when Posed in Jerusalem.(c) The conversants may be in different places, as in phone conversations.Strictly speaklng, the Room Algorithm does not apply to these configuratlons. Thus, in (b) the least common room is one level above that of the target, but on what level would the answer be? The Room Algorithm would either return the respondent to the place queried or else require detailed and perhaps cumbersomDclassificatlons, neither option is taken. All the answers to the questions noted were given on the room immediately above the target. In (a) a least common room may not be circumscribed in the manner outlined by the Room Algorithm, whereas in (c) a distinction between the speaker and the hearer has to be introduced. All these cases suggest that the different confi~Jrations do invoke different generation algorithms.Hence, an appraisal of the con-Flguration is necessary prior to the application of the answer-generation algorithm proper.So far, the discussion was topological, considering only the spatial configuration holding between the conversants and the object questioned. The respondent 's knowledge of the world was not taken into account.In order to prove the psychological validity of an answer generation algorithm it is crucial to demonstrate that the answer given is chosen from a class of several feasible answers, and is not the only one possible due to a limited data base. This was the purpose of the closed questionnaires. Two such questionnaires were administered: first, sub-Jests were asked to choose the best of several answers given to them; then they were asked to mark all the answers they deemed true. Three points were of interest. First, the answers given in the first two conditions were not necessarily the most specified ones marked in the third. Second, there were answers in the multiple option condition which were evidently true and commonly known but which were nonetheless not marked by subjects. These answers included reversed prominence (i.e. the relation of a prominent object to a less prominent one), featural answers and ones which were too high on the place hierarchy. Third, an "~ don't know" answer on the open questionns/re did not necessarily imply a noanswer in the other conditions. In other words, this answer does not signify complete ignorance, but rather an appreciation on the part of the subject that he cannot f~u'nish the answer he deemm appropriate.Together, the three points indicate that there is indeed a psychological process of answer-generation which does not amount to the specification Of the most detailed information one has regarding the object in question.Still another aspect which has to be considered is the speaker's intention when he poses a question. A study ~ this aspect is just on its way now and at this point, I have to limit myself only to a methodological discussion. Evidently, the process of question-answering requires an appraisal of intention (of. Lehnert, !978), one which involves the evaluation of various contertual, personal and sociological factors. In order to make research feasible, as well as constructive, a factorizatlon of the domain of question-answering, I believe, is needed. In this regard the topological, knowledge and intention aspects were noted. The original Room Theory is an attempt to define the topological aspect. The present study shows that even for this 8spect this Theory is not sufficient. The present discussion suggests that an extended topological theory should consist of the following components : I. Semantic and episodic representations, which are not isomorphic to the physically (logically) defined room-hierar chy.2. Determinants of confi~Irations and problematic cases (floor, ceiling).A set of ordered answer-generation algorithms: vertical place~nent (the algorithm proposed by the Room Theory), horizontal relation, featural descrlption and non-informatlve (vacuous, tautolog-Ical) • Definitely, the topological consideration is not sufficient for the characterization of how people answer where questions. Future investigations should ex~cend the research and also include considerations of k~owledge and intention. At this Juncture, however, we can no~e that it is not possible to reduce question answering to knowledge alone, and that some formal selection algorithms have to be postulated. The formal study of such al~orithms is of relevance to the study of both natural and artificial intelligence. | null | null | null | null | Main paper:
:
When (i) is posed in California 2is the appropriate answer to it. This is the case even though (3) and (h) are also true characterizations of the location of the Empire State Building. The pattern of appropriateness alters, however, when the locale where the question presented changes. Thus, when (i) is asked in Israel, 3is the appropriate answer, whereas when it is asked in Manhattan, (I~) is the answer that should be given.The foregoing observations, originally made by Rumelhart (197h) and by Norman (1973) , suggest the following. First, it is not enough for answers to questions to be (semantically) true, they have to be (pragmatically) appropriate as well. Second, appropriateness is not solely determined by the content of the particular propositions in question, but also by the identity of the participants in the particular conversational situation and their locale.In other words, for a person--or for a machine, for that matter--to answer questions, it is not enough to survey one's memory and retrieve information pertaining to the query posed, rather--a selection algorithm has to be used so that an appropriate response would be given. The specification of such a selection algorithm is the topic of the present investigation.The following discussion is based on what is known as the Room Theory: the original, albeit preliminary, model proposed by Rumelhart(197~) in order to account for his insightful observations. I try to examine the psychological validity of this model, and to propose amendments and extensions to it on the basis of empirical data.Theory is on two counts. First, there is the seemingly trivial observation that answers tO different questions are given on diff,:rent levels. Specifically, there is a correlation between the level of the object which is queried and the room on which the respective answer is given. Second, and less trivial, is the observation that answers vary not only with the questions, but also with the spatial relationship which holds between the object of the question and the participants in the conversation. Several loci in the data are indicative of this last pattern. First none of the Americans indicated that the Empire State Building was "in the U.S.", bu~ a third of the Israelis did so; further, some of the Americans, but none of the Israelis, indicated that the building was "in Manhattan". Second, asked about New York City, almost all Israelis, but none of the Americans, answered on the country level. Further, the distribution of the answer patterns furnished by the members of each group changed according to whether the queried city was their own, or close/distant from it. Finally, children's answers to questions about objects also vary with how distant the object is.Above, however, I have qualified the correspondence between the data and the theory; this qualification should now be clarified. I don't think it is meaningful to Judge the validity of a model llke the Room Theory by examining the percentage of cases in which its predictions hold. Such a percentage may reflect the structure of the domain (questions) under investigation, and it need not be indicative of the adequacy of the model as such. The term "by and large" is, however, of qualitative significance. It indicates that unless other factors or reasons are operative, answers to where questions do, indeed, follow the Room A/gorithm. The detection of these "other factors and reasons", their classification and the characterization of the answer types that correspond to them is the main theme of this discussion. Following, then, are the answer patterns which do not conform with the Room Theory.The Room Theory "posits the existence of a psychological room relative to which distances are reckoned. The room corresponds to the smallest geographical region that encompasses both the reference location of the conversants and the location of the places in question". When answering where-questions "the rule is to find the smallest room which Just includes the reference location and the answer location. The appropriate answer is the next smallest geographical unit which contains the location in question, but axcludes the reference location". (Rumelhart, 1975) . The answers generated by this algorithm, note, constitute the placing of the item questioned in a room which is larger than it; henceforth answers of this type will be called vertical.In order to examine the Room Theory, questions regarding places in the world as well as objects in a (concrete) room were presented to several subject populations: college students in Israel and the U.S., American children of three age groups, and aphasic patients. The present report concentrates on the adult data, and only cursory remarks will be made on the answers furnished by the other populations.First, I will discuss answers solicited by an open questionnaire, in which subjects were asked to give one answer to the questions posel to them; later, answers solicited by closed questionnaires will be discussed.First, it should be noted that by and large the answers given by subjects were the ones predicted by the Room First, consider questions about landm-~ks in the towns in which the conversation took place. Most of the answers which involved vertical placement were given on the level of the town itself, i.e. on a level which is higher than the one predicted by the Room Theory. The other answers were not vertical, but rather horizontal:the object questioned was related to another object similar to it. In other words, either the level specified by the Room Theory was changed, or the type of answer (i.e. the generation algorithm itself) was altered. These deviant answers are viewed as two 81ternative solutions to the problem of the floor effect. Specifically, as one goes down the place hierarchy, the specification of rooms between the target and the least common room is cumbersome; indeed, there might not be simple names by which reference to these rooms may be made. Subjects solve this problem either by staying on the level of the least common room or by shifting to the horizontal strate~,.The same problem is noted with the ceilin~ effect, namely, with questions regarding objects which are very high on the place hierarchy: continents for adults, countries for children and aphasic patients. The answers in these cases were varied, a feature which attests the algOr-ithm/c difficulty associated with them. Only a minority of the answers conformed with the Room Theory and most answers were horizontal. Other answer types were: vacuous, in which a vertical answer was given on too high a level (e.g. "in the world"), featural, in which a description, rather than a specification of the locale, was given (e.g. "it is a continent"), or tautological (e.g. "Japan is in Japan").The di~'ferent answer types, we shall say, are the products o~ different alternative answer generation algorithms.The numerical distribution of these answers suggest that the order of preference for the application of the algorithms as the one noted above.There were also cases in which subjects gave answers on a level lower than the one predicted by the Room Theory. Thus, half the Israelis placed the Empire State Building Tin New York", and not "in the U.S." Similarly, all the Americans asked about the Eiffel Tower answered "in Paris", and not "in France". These patterns are attributed to p romlnence. Prominent objects are ones which gain a higher ra-~ in the place h/erarchy than would be attributed to them on semantic classificatory grounds alone. As a consequence, these objects are placed in a room which is more specific then the one predicted by the Room Theory. For instance, New York City is not conceived of by non-Americans as Just another American city; it gains an autonomy of its own and is conceived of as independent of the country in which it is located. The prominence effect suggests that rather than interpreting the room-hierarchy in a concrete fashion (i.e. as isomorphic to the spatial relations which hold in the physical world), one should view it as an abstract conceptual representation.In this representation, ob-Jects ere associated with ta~s: usually, objects which are actually contained in objects of order n are assigned a tag of order n÷l, but prominent objects are assigned tags of the same order as the objects which actually contain them. Thus, if the Empire State Building is tagged n÷l both New York and the U.S. are tngged n, for the Israelis the least com-~n room (order n-l) is the northern hemisphere, and the answer is given on the level of the t~o rooms of order n.Thus, the seemingly unexpected answers associated with prominent objects are due to the modified abstract representations, not to a change in the (vertical) algorithm proper.The salience effect is similar, but distinct. Objects which are close to ones which stand ia a particular relation to the respondent (i.e. physically close, emotionally dear, or belonging to the subject) are not placed in a room but receive horizontal answers instead. For example, all the Israelis answered that Lebanon was "north of Israel", and not that it was "in the Mid-East". Similarly, all the Americans (and half of the Israells) placed Canada in relation to the U.S. Unlike the prominence effect, the salience effect does affect the answer generation algorithm itself, and it bears on individual or cultural differences, not on general semantic com-lderatlons. 4 Specifically, items which ere special to the speaker are tagged in the representation as marked, and this triggers a shift from the vertical to the horizontal algorlthm.All questions considered so far involved one configuration: the two conversants and the target were physically distinct, and together they could he contained in one COmmOn room. This, however, is not the only possible con£iguratlon. Other confi&~aratious, are possible as well: (a) The conversants and the target may coincide in place, as in the question '~here are we now?". (b) The conversants ~ay be contained in the target, as in the question "Where is Israel?" when Posed in Jerusalem.(c) The conversants may be in different places, as in phone conversations.Strictly speaklng, the Room Algorithm does not apply to these configuratlons. Thus, in (b) the least common room is one level above that of the target, but on what level would the answer be? The Room Algorithm would either return the respondent to the place queried or else require detailed and perhaps cumbersomDclassificatlons, neither option is taken. All the answers to the questions noted were given on the room immediately above the target. In (a) a least common room may not be circumscribed in the manner outlined by the Room Algorithm, whereas in (c) a distinction between the speaker and the hearer has to be introduced. All these cases suggest that the different confi~Jrations do invoke different generation algorithms.Hence, an appraisal of the con-Flguration is necessary prior to the application of the answer-generation algorithm proper.So far, the discussion was topological, considering only the spatial configuration holding between the conversants and the object questioned. The respondent 's knowledge of the world was not taken into account.In order to prove the psychological validity of an answer generation algorithm it is crucial to demonstrate that the answer given is chosen from a class of several feasible answers, and is not the only one possible due to a limited data base. This was the purpose of the closed questionnaires. Two such questionnaires were administered: first, sub-Jests were asked to choose the best of several answers given to them; then they were asked to mark all the answers they deemed true. Three points were of interest. First, the answers given in the first two conditions were not necessarily the most specified ones marked in the third. Second, there were answers in the multiple option condition which were evidently true and commonly known but which were nonetheless not marked by subjects. These answers included reversed prominence (i.e. the relation of a prominent object to a less prominent one), featural answers and ones which were too high on the place hierarchy. Third, an "~ don't know" answer on the open questionns/re did not necessarily imply a noanswer in the other conditions. In other words, this answer does not signify complete ignorance, but rather an appreciation on the part of the subject that he cannot f~u'nish the answer he deemm appropriate.Together, the three points indicate that there is indeed a psychological process of answer-generation which does not amount to the specification Of the most detailed information one has regarding the object in question.Still another aspect which has to be considered is the speaker's intention when he poses a question. A study ~ this aspect is just on its way now and at this point, I have to limit myself only to a methodological discussion. Evidently, the process of question-answering requires an appraisal of intention (of. Lehnert, !978), one which involves the evaluation of various contertual, personal and sociological factors. In order to make research feasible, as well as constructive, a factorizatlon of the domain of question-answering, I believe, is needed. In this regard the topological, knowledge and intention aspects were noted. The original Room Theory is an attempt to define the topological aspect. The present study shows that even for this 8spect this Theory is not sufficient. The present discussion suggests that an extended topological theory should consist of the following components : I. Semantic and episodic representations, which are not isomorphic to the physically (logically) defined room-hierar chy.2. Determinants of confi~Irations and problematic cases (floor, ceiling).A set of ordered answer-generation algorithms: vertical place~nent (the algorithm proposed by the Room Theory), horizontal relation, featural descrlption and non-informatlve (vacuous, tautolog-Ical) • Definitely, the topological consideration is not sufficient for the characterization of how people answer where questions. Future investigations should ex~cend the research and also include considerations of k~owledge and intention. At this Juncture, however, we can no~e that it is not possible to reduce question answering to knowledge alone, and that some formal selection algorithms have to be postulated. The formal study of such al~orithms is of relevance to the study of both natural and artificial intelligence.
Appendix:
| null | null | null | null | {
"paperhash": [
"lehnert|the_process_of_question_answering"
],
"title": [
"The Process of Question Answering"
],
"abstract": [
"Abstract : Problems in computational question answering assume a new perspective when question answering is viewed as a problem in natural language processing. A theory of question answering has been proposed which relies on ideas in conceptual information processing and theories of human memory organization. This theory of question answering has been implemented in a computer program, QUALM, currently being used by two story understanding systems to complete a natural language processing system which reads stories and answers questions about what was read. The processes in QUALM are divided into 4 phases: (1) Conceptual categorization which guides subsequent processing by dictating which specific inference mechanisms and memory retrieval strategies should be invoked in the course of answering a question; (2) Inferential analysis which is responsible for understanding what the questioner really meant when a question should not be taken literally; (3) Content specification which determines how much of an answer should be returned in terms of detail and elaborations, and (4) Retrieval heuristics which do the actual digging to extract an answer from memory."
],
"authors": [
{
"name": [
"W. Lehnert"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null
],
"s2_corpus_id": [
"57370597"
],
"intents": [
[]
],
"isInfluential": [
false
]
} | null | 548 | 0.029197 | null | null | null | null | null | null | null | null |
d813e80464eebc7402f95bf6b2f87dcb91279462 | 33888327 | null | Design for Dialogue Comprehension | This paper describes aspects of the design of a dialogue comprehension system, DCS, currently being Implemented. It concentrates on a few design innovations rather than the description of the whole system. The three areas of innovation discussed are: | {
"name": [
"Mann, William C."
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 6 | 2 | null | and Dialogue Game theory, Z. Design assumptions about how to identify the "best" interpretation among several alternatives, and a method, called Preeminence Scheduling, for implementing those assumptions, 3. A now control structure, tlearsay-3, that extends the control structure of llearsay-l[ and makes Preeminence Scheduling fairly straightforward.I. Dialogue Games, Speech Acts and DCS --Examination of actual human dialogue reveals structure extending over • ~overal turns and corresponding to partlcular issues that the participants raise and resolve. Our past work on dialogue has led to an account of this structure, Dialogue Game theory fLorin & Moore 1978; Moore, l,evlu & Mann 1977] . This theory claims that dialogues (and other language uses as well) are comprehensible only because the participants are making available to each other the knowledge of the goals they are pursuing, at ~he p~omcnt, Patterns of these goals recur, representing language conventions: their theoretical representations are called Dialogue Games.If a speaker employs a particular Dialogue Game, that fact must be recognized by the hearer if the speaker is to achieve the desired effect. In other words, Dialogue Game recognition is an essential part of dialogue comprehension. Invoking a game is an act, and terminating the ongoing use of a game is also an act.Dialogue game theory has recently boon extended [Mann 1079 ] in a way makes these game-related acts explicit Acts of Bidding a game, Accepting a bid, and Bidding termination are formally defined as speech acts, comparable to others In speech act theory. So, for example, in the dialogue fragment below, Ct "Morn, l'm hungry." M." "Did you do a good Job on your Geography homework?" the first turn bids a game called the Permission Seeking game, and the second turn refuses that bid and bids the Information 5caking game.DCS is designed to recognize people's use of dialogue /~.ames in transcripts. For each utterance, it builds a hierarchlal structure representing how the utterance performs certain acts, the goals that the acts serve, end thn goal structure that makes the combination of acts coherent. (The data structure holding this information is described holow in the discussion of llearsay-3.)II. Preeminence Scheduling --It seems inevitable that any system capable of forming the "correct" interpretation of most natural langua~,e usage will usually be able to find several other interpretations, given enough opportunity. It is also inevitable that choices bo made, implicitly or explicitly, among interpretations. The choices will correspond to some Internal notion of quality, also possibly implicit. The notion of quality may vary. but the necessity of makin/', such choices does not rest on the particular notion of quality we use. Clearly, it is also important to avoid choosing a single interpretation when there are several nearly equally attractive ones.What methods do we have for making such choices? Consider three approaches.First-find.. The first Interpretation discovered which satisfies well-formcdness is chosen. The effectiveness of first-find depends on having well-informed, selective processes at every choice point, and is only reasonable if one's expectations about what might be said are very good. Even then, this method will select incorrect interpretations.Z. Bounded search and ranked choice. Interpretations are generated by a bounded-effort search, each is assigned an individual quality .score of some sort, and the best is chosen. While this will not miss good but unexpected interpretations missed by first-find, it is wrong in at least two ways: a) it selects an interpretation (and discards others) when the quality difference between interpretations is insignificant, and b) it expends unnecessary resources making absolute quality Judgments where only relative Judgments are needed. These defects suggest an lmprovemenh 3.Preeminence selection= perform a bounded-effort search for interpretations, and then select as beat the one (if any) having a certain threshold amount of demonstrable preferability over its competitors.The key to corre::t choice is determination that such a threshold difference in quality exists. DCS is designed to identify preeminent interpretations.Consider the information content in the fact that the best two interpretations have a quality difference exceeding a fixed threshold. This fact is sufficient to choose an interpretation, and yet it carries less information than is carried in a set of quality scores for the same set of interpretations. C~omputaUonal efficiencies are available because the work of creating the excess information can be avoided by proper design.Given s tentative quality scoring of one's alternatives, several kinds of computations can be avoided. For the highest-ranked interpretation, it is pointless to perform computations whose only effect is to confirm or support the interpretation, (even thongh we expect that for correct interpretations the ways to show confirmation will be numerous), since these will only drive its score higher.For interpretations with inferior ranks, it is likewise pointless to perform computations that refute them (although we expect that refutations of poor interpretations will be numerous), since these will only drive their scores lower. Neither of these is relevant to demonstrating preeminence.Given effective controls, computation can concentrate on refuting good interpretation• and supporting weak ones. (Of" course, such computations will sometimes move 8 new interpretation into the role of highe•t-renked. They may also destroy an eppsrent preeminence.) If the gap in quality rating between the highest ranked interpretation end the next one rams/no significant, then proem/nonce has been demonstrated.Further efficlencles are possible provided that the maximum quality r•ting improvement front untr/ed support computation• can be predicted, since it is then posstblo to find case• for which the m•ximum support of • low-ranked interpretation would not eliminate an existing preeminence. Similar efficlencies can arise from predicting the max/mum loss 6f quality available from untr/ed refuter/one. This approach ls being implemented in DCS, IIL Control Structure --• new AI programming environment called Hearsey-3 is being implemented at ISI for use in development of several systems. It is an augmentation and major revision of some of the control and data structure ideas found in He•rsey-ll [Lesser & Erman 19773, but it is independent of the speech-understandlng task. Hecruy-3 retains lnterprecess communicetion by means of global "blackboards," end it represents its process knowledge in many specialized "knowledge source" (KS) processes, which nominate themselves at appropriate t/rues bY looking at the blackboard, and then are opportunistically scheduled for execution. Blackbcerds are divided into "levels" that typically contain distinct kinds of state knowledge, the distinctions being ~jed as a gross filter on which future KS computation• ere considered.Hearsay-3 retsi,s the idea of a domain-knowledge blackboard (BB), and it adds a knowledge source scheduling blackboard (SBB) as well. Items on the SBB are opportunities to exercise particular scheduling speclslists celled Schedulln~ Knowledge Sources (SKS).The SBB Is •n ideal data structure For implementin~ Prominence scheduling. In DCS the SBB has four levels, calledRefutation, Support, Evaluation and Ordinary-consequence. These correspond to a factoring of the domain K5 into four groups according to their effects. Knowledge sources in each of these groups nominata themselves onto a different level of the SBB. The scheduling-knowledge sources (SKS) perform preeminence scheduling (when a suitable range of alternatives ls available) by selecting available Refutation level opportunities for the highest-ranked interpretation and Support level opportunities for inferior ones. (The SBB and SKS Features of HearMy-3 •re only two of its many innovation•. )The DCS B8 has 6 levels, named Text. Word-sense•, Syntax, Proposition•, Speech-acts •nd Goals. Goals and goal structures, which •re required in any successful analysis, only arise as explanations of speech acts. The KS used for deriving speech acts from utterances •re seperete from those deriving goals from speech acts. | null | null | null | null | Main paper:
the relation of the dcs design to speech act theory:
and Dialogue Game theory, Z. Design assumptions about how to identify the "best" interpretation among several alternatives, and a method, called Preeminence Scheduling, for implementing those assumptions, 3. A now control structure, tlearsay-3, that extends the control structure of llearsay-l[ and makes Preeminence Scheduling fairly straightforward.I. Dialogue Games, Speech Acts and DCS --Examination of actual human dialogue reveals structure extending over • ~overal turns and corresponding to partlcular issues that the participants raise and resolve. Our past work on dialogue has led to an account of this structure, Dialogue Game theory fLorin & Moore 1978; Moore, l,evlu & Mann 1977] . This theory claims that dialogues (and other language uses as well) are comprehensible only because the participants are making available to each other the knowledge of the goals they are pursuing, at ~he p~omcnt, Patterns of these goals recur, representing language conventions: their theoretical representations are called Dialogue Games.If a speaker employs a particular Dialogue Game, that fact must be recognized by the hearer if the speaker is to achieve the desired effect. In other words, Dialogue Game recognition is an essential part of dialogue comprehension. Invoking a game is an act, and terminating the ongoing use of a game is also an act.Dialogue game theory has recently boon extended [Mann 1079 ] in a way makes these game-related acts explicit Acts of Bidding a game, Accepting a bid, and Bidding termination are formally defined as speech acts, comparable to others In speech act theory. So, for example, in the dialogue fragment below, Ct "Morn, l'm hungry." M." "Did you do a good Job on your Geography homework?" the first turn bids a game called the Permission Seeking game, and the second turn refuses that bid and bids the Information 5caking game.DCS is designed to recognize people's use of dialogue /~.ames in transcripts. For each utterance, it builds a hierarchlal structure representing how the utterance performs certain acts, the goals that the acts serve, end thn goal structure that makes the combination of acts coherent. (The data structure holding this information is described holow in the discussion of llearsay-3.)II. Preeminence Scheduling --It seems inevitable that any system capable of forming the "correct" interpretation of most natural langua~,e usage will usually be able to find several other interpretations, given enough opportunity. It is also inevitable that choices bo made, implicitly or explicitly, among interpretations. The choices will correspond to some Internal notion of quality, also possibly implicit. The notion of quality may vary. but the necessity of makin/', such choices does not rest on the particular notion of quality we use. Clearly, it is also important to avoid choosing a single interpretation when there are several nearly equally attractive ones.What methods do we have for making such choices? Consider three approaches.First-find.. The first Interpretation discovered which satisfies well-formcdness is chosen. The effectiveness of first-find depends on having well-informed, selective processes at every choice point, and is only reasonable if one's expectations about what might be said are very good. Even then, this method will select incorrect interpretations.Z. Bounded search and ranked choice. Interpretations are generated by a bounded-effort search, each is assigned an individual quality .score of some sort, and the best is chosen. While this will not miss good but unexpected interpretations missed by first-find, it is wrong in at least two ways: a) it selects an interpretation (and discards others) when the quality difference between interpretations is insignificant, and b) it expends unnecessary resources making absolute quality Judgments where only relative Judgments are needed. These defects suggest an lmprovemenh 3.Preeminence selection= perform a bounded-effort search for interpretations, and then select as beat the one (if any) having a certain threshold amount of demonstrable preferability over its competitors.The key to corre::t choice is determination that such a threshold difference in quality exists. DCS is designed to identify preeminent interpretations.Consider the information content in the fact that the best two interpretations have a quality difference exceeding a fixed threshold. This fact is sufficient to choose an interpretation, and yet it carries less information than is carried in a set of quality scores for the same set of interpretations. C~omputaUonal efficiencies are available because the work of creating the excess information can be avoided by proper design.Given s tentative quality scoring of one's alternatives, several kinds of computations can be avoided. For the highest-ranked interpretation, it is pointless to perform computations whose only effect is to confirm or support the interpretation, (even thongh we expect that for correct interpretations the ways to show confirmation will be numerous), since these will only drive its score higher.For interpretations with inferior ranks, it is likewise pointless to perform computations that refute them (although we expect that refutations of poor interpretations will be numerous), since these will only drive their scores lower. Neither of these is relevant to demonstrating preeminence.Given effective controls, computation can concentrate on refuting good interpretation• and supporting weak ones. (Of" course, such computations will sometimes move 8 new interpretation into the role of highe•t-renked. They may also destroy an eppsrent preeminence.) If the gap in quality rating between the highest ranked interpretation end the next one rams/no significant, then proem/nonce has been demonstrated.Further efficlencles are possible provided that the maximum quality r•ting improvement front untr/ed support computation• can be predicted, since it is then posstblo to find case• for which the m•ximum support of • low-ranked interpretation would not eliminate an existing preeminence. Similar efficlencies can arise from predicting the max/mum loss 6f quality available from untr/ed refuter/one. This approach ls being implemented in DCS, IIL Control Structure --• new AI programming environment called Hearsey-3 is being implemented at ISI for use in development of several systems. It is an augmentation and major revision of some of the control and data structure ideas found in He•rsey-ll [Lesser & Erman 19773, but it is independent of the speech-understandlng task. Hecruy-3 retains lnterprecess communicetion by means of global "blackboards," end it represents its process knowledge in many specialized "knowledge source" (KS) processes, which nominate themselves at appropriate t/rues bY looking at the blackboard, and then are opportunistically scheduled for execution. Blackbcerds are divided into "levels" that typically contain distinct kinds of state knowledge, the distinctions being ~jed as a gross filter on which future KS computation• ere considered.Hearsay-3 retsi,s the idea of a domain-knowledge blackboard (BB), and it adds a knowledge source scheduling blackboard (SBB) as well. Items on the SBB are opportunities to exercise particular scheduling speclslists celled Schedulln~ Knowledge Sources (SKS).The SBB Is •n ideal data structure For implementin~ Prominence scheduling. In DCS the SBB has four levels, calledRefutation, Support, Evaluation and Ordinary-consequence. These correspond to a factoring of the domain K5 into four groups according to their effects. Knowledge sources in each of these groups nominata themselves onto a different level of the SBB. The scheduling-knowledge sources (SKS) perform preeminence scheduling (when a suitable range of alternatives ls available) by selecting available Refutation level opportunities for the highest-ranked interpretation and Support level opportunities for inferior ones. (The SBB and SKS Features of HearMy-3 •re only two of its many innovation•. )The DCS B8 has 6 levels, named Text. Word-sense•, Syntax, Proposition•, Speech-acts •nd Goals. Goals and goal structures, which •re required in any successful analysis, only arise as explanations of speech acts. The KS used for deriving speech acts from utterances •re seperete from those deriving goals from speech acts.
Appendix:
| null | null | null | null | {
"paperhash": [
"levin|dialogue-games:_metacommunication_structures_for_natural_language_interaction",
"lesser|a_retrospective_view_of_the_hearsay-ii_architecture"
],
"title": [
"Dialogue-Games: Metacommunication Structures for Natural Language Interaction",
"A Retrospective View of the Hearsay-II Architecture"
],
"abstract": [
"Our studies of naturally occurring human dialogue have led to the recognition of a class of regularities which characterize important aspects of communication. People appear to interact according to established patterns which span several turns in a dialogue and which recur frequently. These patterns appear to be organized around the goals which the dialogue serves for each participant. Many things which are said later in a dialogue can only be interpreted as pursuit of these goals, established by earlier dialogue. These patterns have been represented by a set of knowledge structures called Dialogue-Games, capturing shared, conventional knowledge that people have about communication and how it can be used to achieve goals. A Dialogue-Game has Parameters, which represent those elements that vary across instances of a particular pattern—the particular dialogue participants and the content topic. The states of the world which must be in effect for a particular Dialogue-Game to be employed successfully are represented by Specifications of these Parameters. Finally, the expected sequence of intermediate states that occur during instances of a particular conventional pattern are represented by the Components of the corresponding Dialogue-Game. Representations for several Dialogue-Games are presented here, based on our analyses of different kinds of naturally occurring dialogue. A process model is discussed, showing Dialogue-Game identification, pursuit, and termination as part of the comprehension of dialogue utterances. This Dialogue-Game model captures some of the important functional aspects of language, especially indirect uses to achieve implicit communication.",
"The Hearsay model has heen presented as a paradigm for attacking errorful knowledge-intensive problems requiring multiple, cooperating knowledge sources. The Hearsay-II architecture is the latest attempt to explore the model. This paper describes experiences gained while successfully applying this architecture to the problem of speech understanding. The major conclusions are: 1. The paradigm of viewing problem solving in terms of hypothesize-and-test actions distributed among distinct representations of the problem has been shown to be computationally feasible. 2. A global working memory (the \"blackboard\"), in which the distinct representations are integrated in a uniform manner, has made it convenient to construct and integrate the individual sources of knowledge needed for the problem solution. 3. The use of a uniform data-directed structure for controlling knowledge-source activity has made the system easy to understand and modify. 4. A solution has been demonstrated to the problem of focus-of-attention in this type of control environment. This solution does not need to be modified when the sources of knowledge in the system are changed."
],
"authors": [
{
"name": [
"J. Levin",
"James A. Moore"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"V. Lesser",
"L. Erman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null
],
"s2_corpus_id": [
"20069944",
"16333285"
],
"intents": [
[],
[]
],
"isInfluential": [
false,
false
]
} | null | 548 | 0.00365 | null | null | null | null | null | null | null | null |
ee113a7bca16acfd24bbffcbcaf0d6def2e260c7 | 5638681 | null | Semantics of Conceptual Graphs | Conceptual graphs are both a language for representing knowledge and patterns for constructing models. They form models in the AI sense of structures that approximate some actual or possible system in the real world. They also form models in the logical sense of structures for which some set of axioms are true. When combined with recent developments in nonstandard logic and semantics, conceptual graphs can form a bridge between heuristic techniques of AI and formal techniques of model theory. | {
"name": [
"Sowa, John F."
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 33 | 53 | null | Semantic networks are often used in AI for representing meaning. But as Woods (1975) and McDermott (1976) observed, the semantic networks themselves have no well-defined semantics. Standard predicate calculus does have a precisely defined, model theoretic semantics; it is adequate for describing mathematical theories with a closed set of axioms. But the real world is messy, incompletely explored, and full of unexpected surprises.Furthermore, the infinite sets commonly used in logic are intractable both for computers and for the human brain.To develop a more realistic semantics, Hintikka (1973) proposed surface models as incomplete, but extendible, finite constructions:Usually, models are thought of as being given through a specification of a number of properties and relations defined on the domain. If the domain is infinite, this specification (as well as many operations with such entities) may require non-trivial settheoretical assumptions. The process is thus often non-finitistic. It is doubtful whether we can realistically expect such structures to be somehow actually involved in our understanding of a sentence or in our contemplation of its meaning, notwithstanding the fact that this meaning is too often thought of as being determined by the class of possible worlds in which the sentence in question is true. It seems to me much likelier that what is involved in one's actual understanding of a sentence S is a mental anticipation of what can happen in one's step-by-step investigation of a world in which S is true. (p. 129) The first stage of constructing a surface model begins with the entities occurring in a sentence or story. During the construction, new facts may he asserted that block certain extensions or facilitate others. A standard model is the limit of a surface model that has been extended infinitely deep, but such infinite processes are not a normal part of understanding.This paper adapts Hintikka's surface models to the formalism of conceptual graphs (Sowa 1976 (Sowa , 1978 . Conceptual graphs serve two purposes: like other forms of semantic networks, they can be used as a canonical representation of meaning in natural language; but they can also be used as building blocks for constructing abstract structures that serve as models in the model-theoretic sense.• Understanding a sentence begins with a translation of that sentence into a conceptual graph.• During the translation, that graph may be joined to framelike (Minsky 1975) or script-like (Schank & Ahelson 1977) graphs that help resolve ambiguities and incorporate background information.• The resulting graph is a nucleus for constructing models of possible worlds in which the sentence is true.• Laws of the world behave like demons or triggers thai monitor the models and block illegal extensions.• If a surface model could be extended infinitely deep, the result would be a complete standard model. This approach leads to an infinite sequence of algorithms ranging from plausible inference to exact deduction; they are analogous to the varying levels of search in game playing programs. Level 0 would simply translate a sentence into a conceptual graph, but do no inference. Level I would do framelike plausible inferences in joining other background graphs. Level 2 would check constraints by testing the model against the laws. Level 3 would join more background graphs. Level 4 would check further constraints, and so on. If the constraints at level n+l are violated, the system would have to backtrack and undo joins at level n. If at some level, all possible extensions are blocked by violations of the laws, then that means the original sentence (or story) was inconsistent with the laws. If the surface model is infinitely extendible, then the original sentence or story was consistent.Exact inference techniques may let the surface models grow indefinitely; but for many applications, they are as impractical as letting a chess playing program search the entire game tree. Plausible inferences with varying degrees of confidence are possible by stopping the surface models at different levels of extension. For story understanding, the initial surface model would be derived completely from the input story. For consistency checks in updating a data base, the initial model would be derived by joining new information to the preexisting data base. For question-answering, a query graph would be joined to the data base; the depth of search permitted in extending the join would determine the limits of complexity of the questions that are answerable. As a result of this theory, algorithms for plausible and exact inference can be compared within the same framework; it is then possible to make informed trade-offs of speed vs. consistency in data base updates or speed vs. completeness in question answering.Canonical formation rules enforce the selection constraints in linguistics: they do not guarantee that all derived graphs are true, but they rule out semantic anomalies. In terms of graph grammars, the canonical formation rules are contextfree. This section defines logical operations that are contextsensitive, They enforce tighter constraints on graph derivations, but they require more complex pattern matching. Formarion rules and logical operations are complementary mechanisms for building models of possible worlds and checking their consistency, Sowa (1976) discussed two ways of handling logical operators in conceptual graphs: the abstract approach, which treats them as functions of truth values, and the direct approach, which treats implications, conjunctions, disjunctions, and negations as operations for building, splitting, and discarding conceptual graphs. That paper, however, merely mentioned the approach; this paper develops a notation adapted from Oantzen's sequents (1934), but with an interpretation based on Beinap's conditional assertions (1973) and with computational techniques similar to Hendrix's partitioned semantic networks (1975, 1979) . Deliyanni and Kowalski (1979) used a similar notation for logic in semantic networks, but with the arrows reversed.Definition: A seq~nt is a collection of conceptual graphs divided into two sets, called the conditions ut ..... Un and the anergons vt,...,v,,, It is written Ul,...,Un "* vl,...,Vm. Several special cases are distinguished:• A simple assertion has no conditions and only one assertion: -.. v.• A disjunction has no conditions and two or more assertions: ..m. PI,...,Vm.• A simple denial has only one condition and no assertions: u -....• A compound denial has two or more conditions and no assertions: ut,...,un -...• A conditianal assertion has one or more conditions and one or more assertions: ut,...,un .... Vl....,v~ • An empty clause has no conditions or assertions: --.,.• A Horn clo,ue has at most one assertion; i.e. it is elther an empty clause, a denial, a simple assertion, or a conditional assertion of the form ut ..... ,% --4, v.For any concept a in an assertion vi, there may be a concept b in a condition u/ that is declared to be coreferent with a.Informally, a sequent states that if all of the conditions are true, then at least one of the assertions must be true. A se. quent with no conditions is an unconditional assertion; if there are two or more assertions, it states that one must be true, hut it doesn't say which.Multiple asserth)ns are necessary for generality, but in deductions, they may cause a model to split into models of multiple altei'native worlds. A sequent with no assertions denies that the combination of conditions can ever occur. The empty clause is an unconditional denial; it is selfcontradictory. Horn clauses are special cases for which deductions are simplified:they have no disjunctions that cause models of the world to split into multiple alternatives.Definition: Let C be a collection of canonical graphs, and let s be the sequent ut ..... Un -', vl ..... vm. • If every condition graph is covered by some graph in C, then the conditions are said to be salisfied.• If some condition graph is not covered by any graph in C, then the sequent s is said to be inapplicable to C.If n---0 (there are no conditions), then the conditions are trivially satisfied.A sequent is like a conditional assertion in Belnap's sense: When its conditions are not satisfied, it asserts nothing. But when they are satisfied, the assertions must be added to the current context. The next axiom states how they are added.Axiom: Let C be a collection of canonical graphs, and let s be the sequent ul ..... u, -,-v~ ..... v,,,. If the conditions of s are satisfied by C, then s may be applied to C as follows:• If m,=l) (a denial or the empty clause), the collection C is said to be blocked.• If m=l (a Horn clause), a copy of each graph ui is joined to some graph in C by a covering join. Then the assertion v is added to the resulting collection C'.• If m>2, a copy of each graph ui is joined to some graph in C by a covering join. Then all graphs in the resulting collection C' are copied to make m disjoint c~)llections identical to C'. Finally, for each j from I to rn, whe assertion v I is added to the j-th copy of C'.After an assertion v is added to one of the collections C', each concept in v that was declared to be coreferent with some concept b in one of the conditions ui is joined to that concept to which b was joined.When a collection of graphs is inconsistent with a sequent, they are blocked by it. If the sequent represents a fundamental law about the world, then the collection represents an impossible situation. When there is only one assertion in an applicable sequent, the collection is extended. But when there are two or more assertions, the collection splits into as many successors as there are assertions; this splitting is typical of algorithms for dealing with disjunctions. The rules for applying sequents are based on Beth's semantic tableaux f1955), but the computational techniques are similar to typical AI methods of production rules, demons, triggers, and monitors.Deliyanni and Kowalski (1979) relate their algorithms for logic in semantic networks to the resolution principle. This relationship is natural because a sequent whose conditions and assertions are all atoms is equivalent to the standard clause form for resolution.But since the sequents defined in this paper may be arbitrary conceptual graphs, they can package a much larger amount of information in each graph than the low level atoms of ordinary resolution. As a result, many fewer steps may be needed to answer a question or do plausible inferences.Infinite families of p~ssible worlds are computationally intractable, hut Dunn (1973) showed that they are not needed for the semantics of modal logic. He considered each possible world w to be characterized by two sets of propositions: laws L and facts F. Every law is also a fact, but some facts are merely contingently true and are not considered laws. A proposition p is necessarily true in w if it follows from the laws of w, and it is possible in w if it is consistent with the laws of w. Dunn proved that semantics in terms of laws and facts is equivalent to the possible worlds semantics.Dunn's approach to modal logic can be combined with Hintikka's surface models and AI methods for handling defaults.Instead of dealing with an infinite set of possible worlds, the system can construct finite, but extendible surface models. The basis for the surface models is a canon that contains the blueprints for assembling models and a set of laws that must be true for each model. The laws impose obligatory constraints on the models, and the canon contains common background information that serves as a heuristic for extending the models. An initial surface model would start as a canonical graph or collection of graphs that represent a given set of facts in a sentence or story. Consider the story, Mary hit the piggy bank with a hammer. She wanted to go to the movies with Janet. but she wouldn't get her allowance until Thursday. And today was only Tuesday.The first sentence would be translated to a conceptual graph like the one in Section 2. Each of the following sentences would be translated into other conceptual graphs and joined to the original graph. But the story as stated is not understandable without a lot of background information: piggy banks normally contain money; piggy banks are usually made of pottery that is easily broken; going to the movies requires money; an allowance is money; and Tuesday precedes Thursday. Charniak (1972) handled such stories with demons that encapsulate knowledge: demons normally lie dormant, but when their associated patterns occur in a story, they wake up and apply their piece of knowledge to the process of understanding. Similar techniques are embodied in production systems, languages like PLANNER (Hewitt 1972) , and knowledge representation systems like KRL (Bobrow & Winograd 1977) . But the trouble with demons is that they are unconstrained: anything can happen when a demon wakes up, no theorems are possible about what a collection of demons can or cannot do, and there is no way of relating plausible reasoning with demons to any of 'the techniques of standard or nonstandard logic.With conceptual graphs, the computational overhead is about the same as with related AI techniques, but the advantage is that the methods can be analyzed by the vast body of techniques that have been developed in logic. The graph for "Mary hit the piggy-bank with a hammer" is a nucleus around which an infinite number of possible worlds can be built. Two individuals, Mary and rlcc~Y-a^NK:iZzloL are fixed, but the particular act of hitting, the hammer Mary used, and all other circumstances are undetermined. As the story continues, some other individuals may be named, graphs from the canon may be joined to add default information, and laws of the world in the form of sequents may be triggered (like demons) to enforce constraints. The next definition introduces the notion of a world bas~ that provides the building material (a canon) and the laws (sequents) for such a family of possible worlds.Definition: A world basis has three components: a canon C, a finite set of sequents L called laws, and one or more finite collections of canonical graphs {Ct ..... Co} called contexts. No context C~ may be blocked by any law in L.A world basis is a collection of nuclei from which complete possible worlds may evolve. The contexts are like Hintikka's surface models: they are finite, but extendible. The graphs in the canon provide default or plausible information that can be joined to extend the contexts, and the laws are constraints on the kinds of extensions that are possible. When a law is violated, it blocks a context as a candidate for a possible world. A default, however, is optional; if contradicted, a default must be undone, and the context restored to the state before the default was applied. In the sample story, the next sentence might continue: "The piggy bank was made of bronze, and when Mary hit it, a genie appeared and gave her two tickets to Animal House." This continuation violates all the default assumptions; it would be unreasonable to assume it in advance, but once given, it forces the system to back up to a context before the defaults were applied and join the new information to it. Several practical issues arise: how much backtracking is necessary, how is the world basis used to develop possible worlds, and what criteria are used to decide when to stop the (possibly infinite) extensions. The next section suggests an answer.The distinction between optional defaults and obligatory laws is reminiscent of the AND-OR trees that often arise in AI, especially in game playing programs. In fact, Hintikka (1973 Hintikka ( , 1974 proposed a game theoretic semantics for testing the truth of a formula in terms of a model and for elaborating a surface model in which that formula is true. Hintikka's approach can be adapted to elaborating a world basis in much the same way that a chess playing program explores the game tree:• Each context represents a position in the game.• The canon defines [Sossible moves by the current player,• Conditional assertions are moves by the opponent.• Denials are checkmating moves by the opponent.• A given context is consistent with the laws if there exists a strategy for avoiding checkmate.By following this suggestion, one can adapt the techniques developed for game playing programs to other kinds of reasoning in AI.Definition: A game over a world basis W is defined by the following rules:• There are two participants named Player and Oppo-m~nt.• For each context in W, Player has the first move.• Player moves in context C either by joining two graphs in C or by selecting any graph in the canon of W that is joinable to some graph u in C and joining it maxi-really to u. If no joins are possible, Player passes. Then Opponent has the right to move in context C.• Opponent moves by checking whether any denials in W are satisfied by C. If so, context C is blocked and is deleted from W. If no denials are satisfied, Opponent may apply any other sequent that is satisfied in C. If no sequent is satisfied, Opponent passes. Then Player has the right to move in context C.• If no contexts are left in W, Player loses.• If both Player and Opponent pass in succession, Player wins.Player wins this game by building a complete model that is consistent with the laws and with the initial information in the problem. But like playing a perfect game of chess, the cost of elaborating a complete model is prohibitive. Yet a computer can play chess as well as most people do by using heuristics to choose moves and terminating the search after a few levels. To develop systematic heuristics for choosing which graphs to join, Sown (1976) stated rules similar to Wilks' preference semantics ( 1975) .The amount of computation required to play this game might be compared to chess: a typical middle game in chess has about 30 or 40 moves on each side, and chess playing programs can consistently beat beginners by searching only 3 levels deep; they can play good games by searching 5 levels. The number of moves in a world basis depends on the number of graphs in the canon, the number of laws in L, and the number of ~aphs in each context. But for many common applications, 30 or 40 moves is a reasonable estimate at any given level, and useful inferences are possible with just a shallow search. The scripts applied by Schank and Abelson (1977) , for example, correspond to a game with only one level of look-ahead; a game with two levels would provide the plausible information of scripts together with a round of consistency checks to eliminate obvious blunders.By deciding how far to search the game tree, one can derive algorithm for plausible inference with varying levels of confidence. Rigorous deduction similar to model elimination (Loveland 1972 ) can be performed by starting with laws and a context that correspond to the negation of what is to be proved and showing that Opponent has a winning strategy. By similar transformations, methods of plausible and exact inference can be related as variations on a general method of reasoning. | null | This section summarizes axioms, definitions, and theorems about conCeptual graphs that are used in this paper. For a more complete discussion and for other features of the theory that are not used here, see the eartier articles by Sown (1976 Sown ( , 1978 .Definition 1: A comcepm~ gmmp& is a finite, connected, bipartite graph with nodes of the first kind called concepu and nodes of the second kind called conceptual relatWn$.Definition 2: Every conceptual relation has one or more arc~, each of which must be attached to a concept. If the relation has n arcs. it is said to be n-adic, and its arcs are labeled I, 2 ..... n.The most common conceptual relations are dyadic (2-adic), but the definition mechanisms can create ones with any number of arcs. Although the formal defin/tion says that the arcs are numbered, for dyadic relations. arc I is drawn as an arrow pointin8 towards the circle, and arc 2 as an arrow point/aS away from the circle.Axiom I: There is a set T of type labeLv and a function type. which maps concepts and conceptual relations into T.• If rypefa)=type(b), then a and b are said to be of the same tXpe.• Type labels are partially ordered:if (vpe(a)<_typefhL then a is said to be a subtype of b.• Type labels of concepts and conceptual relations arc disjoint, noncomparable subsets nf T: if a is a concept and • is a conceptual relation, then a and r may never he of the same type, nor may one be a subtype of the other.Axiom 2: There is a set I=[il, i2, i3 .... } whose elements are called individual markers. The function referent applies to concepts:If a is a concept, then referentla) is either an individual marker in I or the symbol @, which may be read any.• When referentla) ~" l, then a is said to be an individual concept.• When referent(a)=@, then a is said to be a genertc concept.In diagrams, the referent is written after the type label, ~parated by a colon. A concept of a particular cat could be written as ICAT:=41331. A genetic concept, which would refer to any cat, could be written ICA'r:tiiH or simply [CATI. In data base systems, individual markers correspond to the surrogates (Codd 1979) . which serve as unique internal identifiers for external entities. The symbol @ is Codd's notation for null or unknown values in a data base. Externally printable or speakable names are related to the internal surrogates by the next axiom.Axiom 3: There is a dyadic conceptual relation with type label NAME. If a relation of type NAME occurs in a conceptual graph, then the concept attached to arc I must be a subtype of WORD, and the concept attached to arc 2 must be a subtype of ENTITY. If the second concept is individual, then the first concept is called a name of that individual.The following graph states that the word "Mary" is the name of a particular person: ["Mary"]-.=.tNAME)-=.lPERSON:i30741.if there is only one person named Mary in the context, the graph could be abbreviated to just [PERSON:Mary], Axiom 4: The conformity •elation :: relates type labels in T to individual markers in I. If teT, tel. and t::i. then i is said to conform to t.• If t~gs and t::i. then s::i.• For any type t, t::@.• For any concept c. type(c)::referentfc).The conformity relation says that the individual for which the marker i is a surrogate is of type t. In previous papers, the terms permissible or applicable were used instead of conforms to. but the present term and the symbol :: have been adopted from ALGOL-68. Suppose the individual marker i273 is a surrogate for a beagle named Snoopy. Then BEAGLE::i273 is true. By extension, one may also write the name instead of the marker, as BEAGLE=Snoopy. By axiom 4, Snoopy also conforms to at] supertypes of BEAGLE. such as DOG::Snoopy, ANIMAL=Snoopy. or ENTITY::Snoopy.Definition 3: A star graph is a conceptual graph consisting of a single conceptual relation and the concepts attached to each of its arcs. (Two or more arcs of the conceptual relation may be attached to the same concept. )Definition 4: Two concepts a and b are said to be joinable if both of the following properties are true: ingful. Yet to say that some graphs are meaningful and others are not is begging the question, because the purpose of conceptual graphs is to form the basis of a theory of meaning, To avoid prejudging the issue, the term canonical is used for those graphs derivable from a designated set called the canon. For any given domain of discourse, a canon is dcl'incd that rules out anomalous combinations.Definition 5: A canon has thrcc components:• A partially ordered ~et T of type labels.• A set I of individual marker~, with a conformily relation ::.• A finite set of conceptual graphs with type or c~Jnccl)lS and conceptual relations in T and wilh referents either let *~r markers in I.The number of possible canonical graphs may be infinite, but the canon contains a finite number from which all the others can be derived. With an appropriate canon, many undesirable graphs are ruled out as noncanonical, but the canonical graphs are not necessari!y true. T~) ensure that only truc graphs are derived from true graphs, the laws discussed in Section 4 eliminate incnnsistcnt combinations.Axiom 5: A conceptual graph is called canontrol eithcr if it is in the c:tnq)n or if it is derivable from canonical graphs by ()ne of the following canonic'a/formation •ules. I,et u and v be canonical graphs (u and v may be the same graph).• Copy: An exact copy of u is canonical.• Restrict: Let a be a concept in u, and let t be a type label where t<_typela) and t::referenrfa). Then the graph obtained by changing the type label of a to t and leaving •eferent(a) unchanged is canonical.• Join on aconcept: Let a be aconcept in u, and baconcept in v If a and b are joinable, then the graph derived by the followin~ steps is canonical: First delete b from v; then attach to a all arcs of conceptual relations that had been attached to b. If re/'eremfa) e I, then referent(a) is unchanged; otherwise, referent(a) is replaced by referent(b).• Join on a star: Let r be a conceptual relation in u. and x a conceptual relation in v. If the star graphs of r and s are joinable.then the graph derived by the following steps is canonical: First delete s and its arcs from v; then for each i. join the concept attached to arc i of • to the concept that had been attached to arc i of s.Restriction replaces a type label in a graph by the label of a subtype: this rule lets subtypes inherit the structures that apply to more general types. Join on a concept combines graphs that have concepts of the same type: one graph is overlaid on the other so that two concepts of the same type merge into a single concept; as a result, all the arcs that had been connected to either concept arc connected to the single merged concept.Join on a star merges a conceptual relation and all of its attached concepts in a single operation.Definition 6: Let v be a conceptual graph, let v, be a subgraph of v in which every conceptual relation has exactly the same arcs as in v. and let u be a copy of v, in which zero or more concepts may be restricted to subtypes. Then u is called a projection of v. and ¢, is called a projective ortgin of u in v.The main purpose of projections is to define the rule of join on a common projection, which is a generalization of the rules for joining on a concept or a star.Definition 7: If a conceptual graph u is a projection of both v and w. it is called a common projection of v and w, Theorem l: If u is a common projection of canonical graphs t, and w, then v and w may be joined on the common projection u to form a canonical graph by the following steps:• Let v' be a projective origin of u in v. and let w, be a projective origin of u in w.• Restrict each concept of v, and ~ to the type label of the corresponding concept in u.• Join each concept of v, to the corresponding concept of w,.• Join each star graph of ¢ to the corresponding star of ~The concepts and conceptual relations in the resulting graph consist of those in v-t~, w-~, and a copy of u.Definition 8: If v and w are joined on a common projection u. then all concepts and conceptual relations in the projective origin of u in v and the projective origin of u in ~v are said to be covered by the join. in particular, if the projective origin of u in v includes all of v. then the entire graph v is covered by the join. and the join is called a covering join of v by w,Definition 9: Let v and w be joined on a common projection u. The join is called extendible if there exist some concepts a in v and b in w with the following properties:• The concepts a and b were joined to each other.• a is attached to a conceptual relation • that was not covered by the join.• b is attached to a conceptual relation s that was not covered by the join.• The star graphs of r and s are joinable.If a join is not extendible, it is called mn.ximal.The definition of maximal join given here is simpler than the one given in Sown (1976) , but it has the same result. Maximal joins have the effect of Wilks' preference rules (1975) in forcing a maximum connectivity of the graphs. Covering joins are used in Section 3 in the rules for applying sequeots.Theorem 2: Every covering join is maximal. Sown (1976) continued with further material on quantifiers and procedural attachments, and Sown (1978) continued with mechanisms for defining new types of concepts, conceptual relations, and composite entities that have other entities as parts. Note that the terms sort, aubaort, and well-formed in Sown (1976) have now been replaced by the terms type, subtype, and canonical. | The following conceptual graph shows the concepts and relationships in the sentence "Mary hit the piggy hank with a hammer." The boxes are concepts and the circles are conceptual relations. Inside each box or circle is a type label that designates the type of concept or relation. The conceptual relations labeled AONI". INST. and PTNT represent the linguistic cases agent, instrument, and patient of case grammar.Conceptual graphs are a kind of semantic network. See Findler (1979) for surveys of a variety of such networks that have been used in AI. The diagram above illustrates some features of the conceptual graph notation:• Some concepts are generic. They have only a type label inside the box, e.g. mT or HAMMEa• Other concepts are individuaL They have a colon after the type label, followed by a name (Mary) or a unique identifier called an individual marker (i22103).To keep the diagram from looking overly busy, the hierarchy of types and subtypes is not drawn explicitly, but is determined by a separate partial ordering of type labels. The type labels are used by the formation rules to enforce selection constraints and to support the inheritance of properties from a supertype to a subtype.For convenience, the diagram could be linearized by using square brackets for concepts and parentheses for conceptual relations:[ PERSON:Mary]-.~ AGNT)-~( HIT:c I ]~--4 INST).~-(HAMMEI~.] [HIT:c I ]4--( PTNT).~---[P[ GO Y-B A NK:i22 I03]Linearizing the diagram requires a coreference index, el, on the generic concept HiT. The index shows that the two occurrences designate the same act of hitting. If mT had been an individual concept, its name or individual marker would be sufficient to indicate the same act.Besides the features illustrated in the diagram, the theory of conceptual graphs includes the following:• For any particular domain of discourse, a specially designated set of conceptual graphs called the canon,• Four canonical formation rules for deriving new canonical graphs from any given canon,• A method for defining new concept types: some canonical graph is specified as the differentia and a concept in that graph is designated the genus of the new type,• A method for defining new types of Conceptual relations: some canonical graph is specified as the relator and one or more concepts in that graph are specified as parameters,• A method for defining composite entities as structures having other entities as parts,• Optional quantifiers on generic concepts,• Scope of quantifiers specified either by embedding them inside type definitions or by linking them with functional dependency arcs,• Procedural attachments associated with the functional dependency arcs,• Control marks that determine when attached procedures should be invoked. These features have been described in the earlier papers; for completeness, the appendix recapitulates the axioms and definitions that are explicitly used in this paper. Heidorn's (1972 Heidorn's ( , 1975 Natural Language Processor (NLP) is being used to implement the theory of conceptual graphs. The NLP system processes two kinds of Augmented Phrase Structure rules: decoding rules parse language inputs and create graphs that represent their meaning, and encoding ru/es scan the graphs to generate language output. Since the NLP structures are very similar to conceptual graphs, much of the implementation amounts to identifying some feature or combination of features in NLP for each construct in conceptual graphs. Constructs that would be difficult or inefficient to implement directly in NLP rules can be supported by LISP functions. The inference algorithms in this paper, however, have not yet been implemented. | I would like to thank Charles Bontempo, Jon Handel, and George Heidorn for helpful comments on earlier versions of this paper. | Main paper:
conceptual graphs:
The following conceptual graph shows the concepts and relationships in the sentence "Mary hit the piggy hank with a hammer." The boxes are concepts and the circles are conceptual relations. Inside each box or circle is a type label that designates the type of concept or relation. The conceptual relations labeled AONI". INST. and PTNT represent the linguistic cases agent, instrument, and patient of case grammar.Conceptual graphs are a kind of semantic network. See Findler (1979) for surveys of a variety of such networks that have been used in AI. The diagram above illustrates some features of the conceptual graph notation:• Some concepts are generic. They have only a type label inside the box, e.g. mT or HAMMEa• Other concepts are individuaL They have a colon after the type label, followed by a name (Mary) or a unique identifier called an individual marker (i22103).To keep the diagram from looking overly busy, the hierarchy of types and subtypes is not drawn explicitly, but is determined by a separate partial ordering of type labels. The type labels are used by the formation rules to enforce selection constraints and to support the inheritance of properties from a supertype to a subtype.For convenience, the diagram could be linearized by using square brackets for concepts and parentheses for conceptual relations:[ PERSON:Mary]-.~ AGNT)-~( HIT:c I ]~--4 INST).~-(HAMMEI~.] [HIT:c I ]4--( PTNT).~---[P[ GO Y-B A NK:i22 I03]Linearizing the diagram requires a coreference index, el, on the generic concept HiT. The index shows that the two occurrences designate the same act of hitting. If mT had been an individual concept, its name or individual marker would be sufficient to indicate the same act.Besides the features illustrated in the diagram, the theory of conceptual graphs includes the following:• For any particular domain of discourse, a specially designated set of conceptual graphs called the canon,• Four canonical formation rules for deriving new canonical graphs from any given canon,• A method for defining new concept types: some canonical graph is specified as the differentia and a concept in that graph is designated the genus of the new type,• A method for defining new types of Conceptual relations: some canonical graph is specified as the relator and one or more concepts in that graph are specified as parameters,• A method for defining composite entities as structures having other entities as parts,• Optional quantifiers on generic concepts,• Scope of quantifiers specified either by embedding them inside type definitions or by linking them with functional dependency arcs,• Procedural attachments associated with the functional dependency arcs,• Control marks that determine when attached procedures should be invoked. These features have been described in the earlier papers; for completeness, the appendix recapitulates the axioms and definitions that are explicitly used in this paper. Heidorn's (1972 Heidorn's ( , 1975 Natural Language Processor (NLP) is being used to implement the theory of conceptual graphs. The NLP system processes two kinds of Augmented Phrase Structure rules: decoding rules parse language inputs and create graphs that represent their meaning, and encoding ru/es scan the graphs to generate language output. Since the NLP structures are very similar to conceptual graphs, much of the implementation amounts to identifying some feature or combination of features in NLP for each construct in conceptual graphs. Constructs that would be difficult or inefficient to implement directly in NLP rules can be supported by LISP functions. The inference algorithms in this paper, however, have not yet been implemented.
log/caj connect/yes:
Canonical formation rules enforce the selection constraints in linguistics: they do not guarantee that all derived graphs are true, but they rule out semantic anomalies. In terms of graph grammars, the canonical formation rules are contextfree. This section defines logical operations that are contextsensitive, They enforce tighter constraints on graph derivations, but they require more complex pattern matching. Formarion rules and logical operations are complementary mechanisms for building models of possible worlds and checking their consistency, Sowa (1976) discussed two ways of handling logical operators in conceptual graphs: the abstract approach, which treats them as functions of truth values, and the direct approach, which treats implications, conjunctions, disjunctions, and negations as operations for building, splitting, and discarding conceptual graphs. That paper, however, merely mentioned the approach; this paper develops a notation adapted from Oantzen's sequents (1934), but with an interpretation based on Beinap's conditional assertions (1973) and with computational techniques similar to Hendrix's partitioned semantic networks (1975, 1979) . Deliyanni and Kowalski (1979) used a similar notation for logic in semantic networks, but with the arrows reversed.Definition: A seq~nt is a collection of conceptual graphs divided into two sets, called the conditions ut ..... Un and the anergons vt,...,v,,, It is written Ul,...,Un "* vl,...,Vm. Several special cases are distinguished:• A simple assertion has no conditions and only one assertion: -.. v.• A disjunction has no conditions and two or more assertions: ..m. PI,...,Vm.• A simple denial has only one condition and no assertions: u -....• A compound denial has two or more conditions and no assertions: ut,...,un -...• A conditianal assertion has one or more conditions and one or more assertions: ut,...,un .... Vl....,v~ • An empty clause has no conditions or assertions: --.,.• A Horn clo,ue has at most one assertion; i.e. it is elther an empty clause, a denial, a simple assertion, or a conditional assertion of the form ut ..... ,% --4, v.For any concept a in an assertion vi, there may be a concept b in a condition u/ that is declared to be coreferent with a.Informally, a sequent states that if all of the conditions are true, then at least one of the assertions must be true. A se. quent with no conditions is an unconditional assertion; if there are two or more assertions, it states that one must be true, hut it doesn't say which.Multiple asserth)ns are necessary for generality, but in deductions, they may cause a model to split into models of multiple altei'native worlds. A sequent with no assertions denies that the combination of conditions can ever occur. The empty clause is an unconditional denial; it is selfcontradictory. Horn clauses are special cases for which deductions are simplified:they have no disjunctions that cause models of the world to split into multiple alternatives.Definition: Let C be a collection of canonical graphs, and let s be the sequent ut ..... Un -', vl ..... vm. • If every condition graph is covered by some graph in C, then the conditions are said to be salisfied.• If some condition graph is not covered by any graph in C, then the sequent s is said to be inapplicable to C.If n---0 (there are no conditions), then the conditions are trivially satisfied.A sequent is like a conditional assertion in Belnap's sense: When its conditions are not satisfied, it asserts nothing. But when they are satisfied, the assertions must be added to the current context. The next axiom states how they are added.Axiom: Let C be a collection of canonical graphs, and let s be the sequent ul ..... u, -,-v~ ..... v,,,. If the conditions of s are satisfied by C, then s may be applied to C as follows:• If m,=l) (a denial or the empty clause), the collection C is said to be blocked.• If m=l (a Horn clause), a copy of each graph ui is joined to some graph in C by a covering join. Then the assertion v is added to the resulting collection C'.• If m>2, a copy of each graph ui is joined to some graph in C by a covering join. Then all graphs in the resulting collection C' are copied to make m disjoint c~)llections identical to C'. Finally, for each j from I to rn, whe assertion v I is added to the j-th copy of C'.After an assertion v is added to one of the collections C', each concept in v that was declared to be coreferent with some concept b in one of the conditions ui is joined to that concept to which b was joined.When a collection of graphs is inconsistent with a sequent, they are blocked by it. If the sequent represents a fundamental law about the world, then the collection represents an impossible situation. When there is only one assertion in an applicable sequent, the collection is extended. But when there are two or more assertions, the collection splits into as many successors as there are assertions; this splitting is typical of algorithms for dealing with disjunctions. The rules for applying sequents are based on Beth's semantic tableaux f1955), but the computational techniques are similar to typical AI methods of production rules, demons, triggers, and monitors.Deliyanni and Kowalski (1979) relate their algorithms for logic in semantic networks to the resolution principle. This relationship is natural because a sequent whose conditions and assertions are all atoms is equivalent to the standard clause form for resolution.But since the sequents defined in this paper may be arbitrary conceptual graphs, they can package a much larger amount of information in each graph than the low level atoms of ordinary resolution. As a result, many fewer steps may be needed to answer a question or do plausible inferences.
laws, facts, and possible worlds:
Infinite families of p~ssible worlds are computationally intractable, hut Dunn (1973) showed that they are not needed for the semantics of modal logic. He considered each possible world w to be characterized by two sets of propositions: laws L and facts F. Every law is also a fact, but some facts are merely contingently true and are not considered laws. A proposition p is necessarily true in w if it follows from the laws of w, and it is possible in w if it is consistent with the laws of w. Dunn proved that semantics in terms of laws and facts is equivalent to the possible worlds semantics.Dunn's approach to modal logic can be combined with Hintikka's surface models and AI methods for handling defaults.Instead of dealing with an infinite set of possible worlds, the system can construct finite, but extendible surface models. The basis for the surface models is a canon that contains the blueprints for assembling models and a set of laws that must be true for each model. The laws impose obligatory constraints on the models, and the canon contains common background information that serves as a heuristic for extending the models. An initial surface model would start as a canonical graph or collection of graphs that represent a given set of facts in a sentence or story. Consider the story, Mary hit the piggy bank with a hammer. She wanted to go to the movies with Janet. but she wouldn't get her allowance until Thursday. And today was only Tuesday.The first sentence would be translated to a conceptual graph like the one in Section 2. Each of the following sentences would be translated into other conceptual graphs and joined to the original graph. But the story as stated is not understandable without a lot of background information: piggy banks normally contain money; piggy banks are usually made of pottery that is easily broken; going to the movies requires money; an allowance is money; and Tuesday precedes Thursday. Charniak (1972) handled such stories with demons that encapsulate knowledge: demons normally lie dormant, but when their associated patterns occur in a story, they wake up and apply their piece of knowledge to the process of understanding. Similar techniques are embodied in production systems, languages like PLANNER (Hewitt 1972) , and knowledge representation systems like KRL (Bobrow & Winograd 1977) . But the trouble with demons is that they are unconstrained: anything can happen when a demon wakes up, no theorems are possible about what a collection of demons can or cannot do, and there is no way of relating plausible reasoning with demons to any of 'the techniques of standard or nonstandard logic.With conceptual graphs, the computational overhead is about the same as with related AI techniques, but the advantage is that the methods can be analyzed by the vast body of techniques that have been developed in logic. The graph for "Mary hit the piggy-bank with a hammer" is a nucleus around which an infinite number of possible worlds can be built. Two individuals, Mary and rlcc~Y-a^NK:iZzloL are fixed, but the particular act of hitting, the hammer Mary used, and all other circumstances are undetermined. As the story continues, some other individuals may be named, graphs from the canon may be joined to add default information, and laws of the world in the form of sequents may be triggered (like demons) to enforce constraints. The next definition introduces the notion of a world bas~ that provides the building material (a canon) and the laws (sequents) for such a family of possible worlds.Definition: A world basis has three components: a canon C, a finite set of sequents L called laws, and one or more finite collections of canonical graphs {Ct ..... Co} called contexts. No context C~ may be blocked by any law in L.A world basis is a collection of nuclei from which complete possible worlds may evolve. The contexts are like Hintikka's surface models: they are finite, but extendible. The graphs in the canon provide default or plausible information that can be joined to extend the contexts, and the laws are constraints on the kinds of extensions that are possible. When a law is violated, it blocks a context as a candidate for a possible world. A default, however, is optional; if contradicted, a default must be undone, and the context restored to the state before the default was applied. In the sample story, the next sentence might continue: "The piggy bank was made of bronze, and when Mary hit it, a genie appeared and gave her two tickets to Animal House." This continuation violates all the default assumptions; it would be unreasonable to assume it in advance, but once given, it forces the system to back up to a context before the defaults were applied and join the new information to it. Several practical issues arise: how much backtracking is necessary, how is the world basis used to develop possible worlds, and what criteria are used to decide when to stop the (possibly infinite) extensions. The next section suggests an answer.
game th~ se~md~:
The distinction between optional defaults and obligatory laws is reminiscent of the AND-OR trees that often arise in AI, especially in game playing programs. In fact, Hintikka (1973 Hintikka ( , 1974 proposed a game theoretic semantics for testing the truth of a formula in terms of a model and for elaborating a surface model in which that formula is true. Hintikka's approach can be adapted to elaborating a world basis in much the same way that a chess playing program explores the game tree:• Each context represents a position in the game.• The canon defines [Sossible moves by the current player,• Conditional assertions are moves by the opponent.• Denials are checkmating moves by the opponent.• A given context is consistent with the laws if there exists a strategy for avoiding checkmate.By following this suggestion, one can adapt the techniques developed for game playing programs to other kinds of reasoning in AI.Definition: A game over a world basis W is defined by the following rules:• There are two participants named Player and Oppo-m~nt.• For each context in W, Player has the first move.• Player moves in context C either by joining two graphs in C or by selecting any graph in the canon of W that is joinable to some graph u in C and joining it maxi-really to u. If no joins are possible, Player passes. Then Opponent has the right to move in context C.• Opponent moves by checking whether any denials in W are satisfied by C. If so, context C is blocked and is deleted from W. If no denials are satisfied, Opponent may apply any other sequent that is satisfied in C. If no sequent is satisfied, Opponent passes. Then Player has the right to move in context C.• If no contexts are left in W, Player loses.• If both Player and Opponent pass in succession, Player wins.Player wins this game by building a complete model that is consistent with the laws and with the initial information in the problem. But like playing a perfect game of chess, the cost of elaborating a complete model is prohibitive. Yet a computer can play chess as well as most people do by using heuristics to choose moves and terminating the search after a few levels. To develop systematic heuristics for choosing which graphs to join, Sown (1976) stated rules similar to Wilks' preference semantics ( 1975) .The amount of computation required to play this game might be compared to chess: a typical middle game in chess has about 30 or 40 moves on each side, and chess playing programs can consistently beat beginners by searching only 3 levels deep; they can play good games by searching 5 levels. The number of moves in a world basis depends on the number of graphs in the canon, the number of laws in L, and the number of ~aphs in each context. But for many common applications, 30 or 40 moves is a reasonable estimate at any given level, and useful inferences are possible with just a shallow search. The scripts applied by Schank and Abelson (1977) , for example, correspond to a game with only one level of look-ahead; a game with two levels would provide the plausible information of scripts together with a round of consistency checks to eliminate obvious blunders.By deciding how far to search the game tree, one can derive algorithm for plausible inference with varying levels of confidence. Rigorous deduction similar to model elimination (Loveland 1972 ) can be performed by starting with laws and a context that correspond to the negation of what is to be proved and showing that Opponent has a winning strategy. By similar transformations, methods of plausible and exact inference can be related as variations on a general method of reasoning.
appendix: summary of the formalism:
This section summarizes axioms, definitions, and theorems about conCeptual graphs that are used in this paper. For a more complete discussion and for other features of the theory that are not used here, see the eartier articles by Sown (1976 Sown ( , 1978 .Definition 1: A comcepm~ gmmp& is a finite, connected, bipartite graph with nodes of the first kind called concepu and nodes of the second kind called conceptual relatWn$.Definition 2: Every conceptual relation has one or more arc~, each of which must be attached to a concept. If the relation has n arcs. it is said to be n-adic, and its arcs are labeled I, 2 ..... n.The most common conceptual relations are dyadic (2-adic), but the definition mechanisms can create ones with any number of arcs. Although the formal defin/tion says that the arcs are numbered, for dyadic relations. arc I is drawn as an arrow pointin8 towards the circle, and arc 2 as an arrow point/aS away from the circle.Axiom I: There is a set T of type labeLv and a function type. which maps concepts and conceptual relations into T.• If rypefa)=type(b), then a and b are said to be of the same tXpe.• Type labels are partially ordered:if (vpe(a)<_typefhL then a is said to be a subtype of b.• Type labels of concepts and conceptual relations arc disjoint, noncomparable subsets nf T: if a is a concept and • is a conceptual relation, then a and r may never he of the same type, nor may one be a subtype of the other.Axiom 2: There is a set I=[il, i2, i3 .... } whose elements are called individual markers. The function referent applies to concepts:If a is a concept, then referentla) is either an individual marker in I or the symbol @, which may be read any.• When referentla) ~" l, then a is said to be an individual concept.• When referent(a)=@, then a is said to be a genertc concept.In diagrams, the referent is written after the type label, ~parated by a colon. A concept of a particular cat could be written as ICAT:=41331. A genetic concept, which would refer to any cat, could be written ICA'r:tiiH or simply [CATI. In data base systems, individual markers correspond to the surrogates (Codd 1979) . which serve as unique internal identifiers for external entities. The symbol @ is Codd's notation for null or unknown values in a data base. Externally printable or speakable names are related to the internal surrogates by the next axiom.Axiom 3: There is a dyadic conceptual relation with type label NAME. If a relation of type NAME occurs in a conceptual graph, then the concept attached to arc I must be a subtype of WORD, and the concept attached to arc 2 must be a subtype of ENTITY. If the second concept is individual, then the first concept is called a name of that individual.The following graph states that the word "Mary" is the name of a particular person: ["Mary"]-.=.tNAME)-=.lPERSON:i30741.if there is only one person named Mary in the context, the graph could be abbreviated to just [PERSON:Mary], Axiom 4: The conformity •elation :: relates type labels in T to individual markers in I. If teT, tel. and t::i. then i is said to conform to t.• If t~gs and t::i. then s::i.• For any type t, t::@.• For any concept c. type(c)::referentfc).The conformity relation says that the individual for which the marker i is a surrogate is of type t. In previous papers, the terms permissible or applicable were used instead of conforms to. but the present term and the symbol :: have been adopted from ALGOL-68. Suppose the individual marker i273 is a surrogate for a beagle named Snoopy. Then BEAGLE::i273 is true. By extension, one may also write the name instead of the marker, as BEAGLE=Snoopy. By axiom 4, Snoopy also conforms to at] supertypes of BEAGLE. such as DOG::Snoopy, ANIMAL=Snoopy. or ENTITY::Snoopy.Definition 3: A star graph is a conceptual graph consisting of a single conceptual relation and the concepts attached to each of its arcs. (Two or more arcs of the conceptual relation may be attached to the same concept. )Definition 4: Two concepts a and b are said to be joinable if both of the following properties are true: ingful. Yet to say that some graphs are meaningful and others are not is begging the question, because the purpose of conceptual graphs is to form the basis of a theory of meaning, To avoid prejudging the issue, the term canonical is used for those graphs derivable from a designated set called the canon. For any given domain of discourse, a canon is dcl'incd that rules out anomalous combinations.Definition 5: A canon has thrcc components:• A partially ordered ~et T of type labels.• A set I of individual marker~, with a conformily relation ::.• A finite set of conceptual graphs with type or c~Jnccl)lS and conceptual relations in T and wilh referents either let *~r markers in I.The number of possible canonical graphs may be infinite, but the canon contains a finite number from which all the others can be derived. With an appropriate canon, many undesirable graphs are ruled out as noncanonical, but the canonical graphs are not necessari!y true. T~) ensure that only truc graphs are derived from true graphs, the laws discussed in Section 4 eliminate incnnsistcnt combinations.Axiom 5: A conceptual graph is called canontrol eithcr if it is in the c:tnq)n or if it is derivable from canonical graphs by ()ne of the following canonic'a/formation •ules. I,et u and v be canonical graphs (u and v may be the same graph).• Copy: An exact copy of u is canonical.• Restrict: Let a be a concept in u, and let t be a type label where t<_typela) and t::referenrfa). Then the graph obtained by changing the type label of a to t and leaving •eferent(a) unchanged is canonical.• Join on aconcept: Let a be aconcept in u, and baconcept in v If a and b are joinable, then the graph derived by the followin~ steps is canonical: First delete b from v; then attach to a all arcs of conceptual relations that had been attached to b. If re/'eremfa) e I, then referent(a) is unchanged; otherwise, referent(a) is replaced by referent(b).• Join on a star: Let r be a conceptual relation in u. and x a conceptual relation in v. If the star graphs of r and s are joinable.then the graph derived by the following steps is canonical: First delete s and its arcs from v; then for each i. join the concept attached to arc i of • to the concept that had been attached to arc i of s.Restriction replaces a type label in a graph by the label of a subtype: this rule lets subtypes inherit the structures that apply to more general types. Join on a concept combines graphs that have concepts of the same type: one graph is overlaid on the other so that two concepts of the same type merge into a single concept; as a result, all the arcs that had been connected to either concept arc connected to the single merged concept.Join on a star merges a conceptual relation and all of its attached concepts in a single operation.Definition 6: Let v be a conceptual graph, let v, be a subgraph of v in which every conceptual relation has exactly the same arcs as in v. and let u be a copy of v, in which zero or more concepts may be restricted to subtypes. Then u is called a projection of v. and ¢, is called a projective ortgin of u in v.The main purpose of projections is to define the rule of join on a common projection, which is a generalization of the rules for joining on a concept or a star.Definition 7: If a conceptual graph u is a projection of both v and w. it is called a common projection of v and w, Theorem l: If u is a common projection of canonical graphs t, and w, then v and w may be joined on the common projection u to form a canonical graph by the following steps:• Let v' be a projective origin of u in v. and let w, be a projective origin of u in w.• Restrict each concept of v, and ~ to the type label of the corresponding concept in u.• Join each concept of v, to the corresponding concept of w,.• Join each star graph of ¢ to the corresponding star of ~The concepts and conceptual relations in the resulting graph consist of those in v-t~, w-~, and a copy of u.Definition 8: If v and w are joined on a common projection u. then all concepts and conceptual relations in the projective origin of u in v and the projective origin of u in ~v are said to be covered by the join. in particular, if the projective origin of u in v includes all of v. then the entire graph v is covered by the join. and the join is called a covering join of v by w,Definition 9: Let v and w be joined on a common projection u. The join is called extendible if there exist some concepts a in v and b in w with the following properties:• The concepts a and b were joined to each other.• a is attached to a conceptual relation • that was not covered by the join.• b is attached to a conceptual relation s that was not covered by the join.• The star graphs of r and s are joinable.If a join is not extendible, it is called mn.ximal.The definition of maximal join given here is simpler than the one given in Sown (1976) , but it has the same result. Maximal joins have the effect of Wilks' preference rules (1975) in forcing a maximum connectivity of the graphs. Covering joins are used in Section 3 in the rules for applying sequeots.Theorem 2: Every covering join is maximal. Sown (1976) continued with further material on quantifiers and procedural attachments, and Sown (1978) continued with mechanisms for defining new types of concepts, conceptual relations, and composite entities that have other entities as parts. Note that the terms sort, aubaort, and well-formed in Sown (1976) have now been replaced by the terms type, subtype, and canonical.
acknowledgment:
I would like to thank Charles Bontempo, Jon Handel, and George Heidorn for helpful comments on earlier versions of this paper.
i. surface models:
Semantic networks are often used in AI for representing meaning. But as Woods (1975) and McDermott (1976) observed, the semantic networks themselves have no well-defined semantics. Standard predicate calculus does have a precisely defined, model theoretic semantics; it is adequate for describing mathematical theories with a closed set of axioms. But the real world is messy, incompletely explored, and full of unexpected surprises.Furthermore, the infinite sets commonly used in logic are intractable both for computers and for the human brain.To develop a more realistic semantics, Hintikka (1973) proposed surface models as incomplete, but extendible, finite constructions:Usually, models are thought of as being given through a specification of a number of properties and relations defined on the domain. If the domain is infinite, this specification (as well as many operations with such entities) may require non-trivial settheoretical assumptions. The process is thus often non-finitistic. It is doubtful whether we can realistically expect such structures to be somehow actually involved in our understanding of a sentence or in our contemplation of its meaning, notwithstanding the fact that this meaning is too often thought of as being determined by the class of possible worlds in which the sentence in question is true. It seems to me much likelier that what is involved in one's actual understanding of a sentence S is a mental anticipation of what can happen in one's step-by-step investigation of a world in which S is true. (p. 129) The first stage of constructing a surface model begins with the entities occurring in a sentence or story. During the construction, new facts may he asserted that block certain extensions or facilitate others. A standard model is the limit of a surface model that has been extended infinitely deep, but such infinite processes are not a normal part of understanding.This paper adapts Hintikka's surface models to the formalism of conceptual graphs (Sowa 1976 (Sowa , 1978 . Conceptual graphs serve two purposes: like other forms of semantic networks, they can be used as a canonical representation of meaning in natural language; but they can also be used as building blocks for constructing abstract structures that serve as models in the model-theoretic sense.• Understanding a sentence begins with a translation of that sentence into a conceptual graph.• During the translation, that graph may be joined to framelike (Minsky 1975) or script-like (Schank & Ahelson 1977) graphs that help resolve ambiguities and incorporate background information.• The resulting graph is a nucleus for constructing models of possible worlds in which the sentence is true.• Laws of the world behave like demons or triggers thai monitor the models and block illegal extensions.• If a surface model could be extended infinitely deep, the result would be a complete standard model. This approach leads to an infinite sequence of algorithms ranging from plausible inference to exact deduction; they are analogous to the varying levels of search in game playing programs. Level 0 would simply translate a sentence into a conceptual graph, but do no inference. Level I would do framelike plausible inferences in joining other background graphs. Level 2 would check constraints by testing the model against the laws. Level 3 would join more background graphs. Level 4 would check further constraints, and so on. If the constraints at level n+l are violated, the system would have to backtrack and undo joins at level n. If at some level, all possible extensions are blocked by violations of the laws, then that means the original sentence (or story) was inconsistent with the laws. If the surface model is infinitely extendible, then the original sentence or story was consistent.Exact inference techniques may let the surface models grow indefinitely; but for many applications, they are as impractical as letting a chess playing program search the entire game tree. Plausible inferences with varying degrees of confidence are possible by stopping the surface models at different levels of extension. For story understanding, the initial surface model would be derived completely from the input story. For consistency checks in updating a data base, the initial model would be derived by joining new information to the preexisting data base. For question-answering, a query graph would be joined to the data base; the depth of search permitted in extending the join would determine the limits of complexity of the questions that are answerable. As a result of this theory, algorithms for plausible and exact inference can be compared within the same framework; it is then possible to make informed trade-offs of speed vs. consistency in data base updates or speed vs. completeness in question answering.
Appendix:
| null | null | null | null | {
"paperhash": [
"codd|extending_the_database_relational_model_to_capture_more_meaning",
"codd|extending_the_data_base_relational_model_to_capture_more_meaning",
"deliyanni|logic_and_semantic_networks",
"bobrow|an_overview_of_krl,_a_knowledge_representation_language",
"sowa|conceptual_graphs_for_a_data_base_interface",
"mcdermott|artificial_intelligence_meets_natural_stupidity",
"hendrix|expanding_the_utility_of_semantic_networks_through_partitioning",
"heidorn|augmented_phrase_structure_grammars",
"charniak|toward_a_model_of_children's_story_comprehension",
"loveland|a_unifying_view_of_some_linear_herbrand_procedures",
"heidorn|natural_language_inputs_to_a_simulation_programming_system:_an_introduction",
"hintikka|semantical_games_and_the_bach-peters_paradox"
],
"title": [
"Extending the database relational model to capture more meaning",
"Extending the data base relational model to capture more meaning",
"Logic and semantic networks",
"An overview of KRL, a Knowledge Representation Language",
"Conceptual Graphs for a Data Base Interface",
"Artificial intelligence meets natural stupidity",
"Expanding the Utility of Semantic Networks Through Partitioning",
"Augmented Phrase Structure Grammars",
"Toward a model of children's story comprehension",
"A Unifying View of Some Linear Herbrand Procedures",
"Natural language inputs to a simulation programming system: An introduction",
"SEMANTICAL GAMES AND THE BACH-PETERS PARADOX"
],
"abstract": [
"During the last three or four years several investigators have been exploring “semantic models” for formatted databases. The intent is to capture (in a more or less formal way) more of the meaning of the data so that database design can become more systematic and the database system itself can behave more intelligently. Two major thrusts are clear. (1) the search for meaningful units that are as small as possible—atomic semantics; (2) the search for meaningful units that are larger than the usual n-ary relation—molecular semantics. In this paper we propose extensions to the relational model to support certain atomic and molecular semantics. These extensions represent a synthesis of many ideas from the published work in semantic modeling plus the introduction of new rules for insertion, update, and deletion, as well as new algebraic operators.",
"During the last three or four years several investigators have been exploring 'semantic models' for formatted data bases. The intent is to capture (in a more or less formal way) more of the meaning of the data, so that data base design can become more systematic and the data base system itself can behave more intelligently. Two major thrusts are clear:1) the search for meaningful units that are as small as possible --- atomic semantics2) the search for meaningful units that are larger than the usual n-ary relation --- molecular semantics.In this paper we propose extensions to the relational model to support certain atomic and molecular semantics. These extensions represent a synthesis of many ideas from the published work in semantic modeling.",
"An extended form of semantic network is defined, which can be regarded as a syntactic variant of the clausal form of logic. By virtue of its relationship with logic, the extended semantic network is provided with a precise semantics, inference rules, and a procedural interpretation. On the other hand, by regarding semantic networks as an abstract data structure for the representation of clauses, we provide a theorem-prover with a potentially useful indexing scheme and path-following strategy for guiding the search for a proof.",
"This paper describes KRL, a Knowledge Representation Language designed for use in understander systems. It outlines both the general concepts which underlie our research and the details of KRL-0, an experimental implementation of some of these concepts. KRL is an attempt to integrate procedural knowledge with a broad base of declarative forms. These forms provide a variety of ways to express the logical structure of the knowledge, in order to give flexibility in associating procedures (for memory and reasoning) with specific pieces of knowledge, and to control the relative accessibility of different facts and descriptions. The formalism for declarative knowledge is based on structured conceptual objects with associated descriptions. These objects form a network of memory units with several different sorts of linkages, each having well-specified implications for the retrieval process. Procedures can be associated directly with the internal structure of a conceptual object. This procedural attachment allows the steps for a particular operation to be determined by characteristics of the specific entities involved. The control structure of KRL is based on the belief that the next generation of intelligent programs will integrate data-directed and goal-directed processing by using multi-processing. It provides for a priority-ordered multi-process agenda with explicit (user-provided) strategies for scheduling and resource allocation. It provides procedure directories which operate along with process frameworks to allow procedural parameterization of the fundamental system processes for building, comparing, and retrieving memory structures. Future development of KRL will include integrating procedure definition with the descriptive formalism.",
"A data base system that supports natural language queries is not really natural if it requires the user to know how the data are represented. This paper defines a formalism, called conceptual graphs, that can describe data according to the user's view and access data according to the system's view. In addition, the graphs can represent functional dependencies in the data base and support inferences and computations that are not explicit in the initial query.",
"As a field, artificial intelligence has always been on the border of respectability, and therefore on the border of crackpottery. Many critics <Dreyfus, 1972>, <Lighthill, 1973> have urged that we are over the border. We have been very defensive toward this charge, drawing ourselves up with dignity when it is made and folding the cloak of Science about us. On the other hand, in private, we have been justifiably proud of our willingness to explore weird ideas, because pursuing them is the only way to make progress.",
"An augmentation of semantic networks is presented in which the various nodes and arcs are partitioned into \"net spaces.\" These net spaces delimit the scopes of quantified variables, distinguish hypothetical and imaginary situations from reality, encode alternative worlds considered in planning, and focus attention at particular levels of detail.",
"Augmented phrase structure grammars consist of phrase structure rules with embedded conditions and structure-building actions written in a specially developed language. An attribute-value, record-oriented information structure is an integral part of the theory.",
"Massachusetts Institute of Technology. Dept. of Electrical Engineering. Thesis. 1972. Ph.D.",
"The linked conjunct, resolution, matrix reduction, and model elimination proof procedures constitute a nearly exhaustive list of the basic Herbrand proof procedures introduced in the 1960's. Each was introduced as a hopefully efficient complete procedure for the first order predicate calculus for the purpose of mechanical theorem proving. This paper contains a demonstration that versions of these procedures can be highly related in their design. S-linear resolution, a particular strategy of resolution previously proposed, is seen to possess a natural refinement isomorphic at ground level to a refinement of the model elimination procedure. There is also an isomorphism at the general level between a less natural s-linear resolution refinement and the model elimination refinement. The model elimination procedure is also interpreted within the linked conjunct and matrix reduction procedures. An alternate interpretation of these results is that, very roughly, the procedures other than resolution can be viewed as forms of linear resolution. K E Y W O R D S A N D PHRASES: mechanical theorem proving, Herbrand proof procedure, linked coniunct, resolution, model elimination, matrix reduction procedure, linear resolution procedure CR CATEGORIES: 3.64, 3.66, 5.21",
"A simulation programming system with which models for simple queuing problems can be built through naturallanguage interaction with a computer is described. In this system the English statement of a problem is first translated into a language -independent entity-attribute-value information structure, which can then be translated back into an equivalent English description and into a GPSS simulation program for the problem. This processing is done on an IBM 360/67 by a FORTRAN program which is guided by a set of stratified decoding and encoding rules written in a grammar-rule language developed for this system. A detailed example of the use of the system is included. This task was supported by the Information Systems Program of the Office of Naval Research as Project NR 049314, under Project Order PO 1-0177. The facilities of the W.R. Church Computer Center were utilized for this research.",
"In this paper the game-theoretical semantics of Hintikka is extended so as to cover some typical uses of the English definite article. It is shown that in this approach the semantical problems involved in the so-called Bach-Peters paradox admit a straightforward solution. The game-theoretical approach is compared with certain other approaches to the paradox. A \"logicians' Bach-Peters paradox\" is formulated and a solution to it proposed. Finally it is argued that the game-theoretical semantics might even suggest an approach to syntax that discriminates between the different Bach-Peters sentences in a way the transformational syntax does not."
],
"authors": [
{
"name": [
"E. Codd"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"E. Codd"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Amaryllis Deliyanni",
"R. Kowalski"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. Bobrow",
"T. Winograd"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"J. Sowa"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. McDermott"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"G. Hendrix"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"George E. Heidorn"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Eugene Charniak"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. Loveland"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"George E. Heidorn"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"J. Hintikka",
"E. Saarinen"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null,
null,
null,
null,
null,
null,
null,
null,
null
],
"s2_corpus_id": [
"17517212",
"6524183",
"1205815",
"7965074",
"12092832",
"28619965",
"9767925",
"2658668",
"62620723",
"14244283",
"60214583",
"62219579"
],
"intents": [
[
"background"
],
[
"methodology"
],
[],
[
"methodology"
],
[
"background",
"methodology"
],
[],
[
"methodology"
],
[],
[
"background"
],
[],
[],
[]
],
"isInfluential": [
false,
false,
false,
false,
true,
false,
false,
false,
false,
false,
false,
false
]
} | Problem: The paper discusses the use of conceptual graphs as a bridge between heuristic techniques of AI and formal techniques of model theory.
Solution: By adapting Hintikka's surface models to the formalism of conceptual graphs, the paper proposes a method for constructing abstract structures that serve as models in the model-theoretic sense, aiming to enhance the understanding and representation of meaning in natural language. | 548 | 0.096715 | null | null | null | null | null | null | null | null |
f3f09d77332979d8315c775c1e6654323ff661cd | 16742497 | null | Towards a Self-Extending Parser | This paper discusses an approach to incremental learning in natural language processing. The technique of projecting and integrating semantic constraints to learn word definitions is analyzed as Implemented in the POLITICS system. Extensions and improvements of this technique are developed. The problem of generalizing existing word meanings and understanding metaphorical uses of words Is addressed In terms of semantic constraint Integration. | {
"name": [
"Carbonell, Jaime G."
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 9 | 56 | null | Natural language analysis, like most other subfields of Artificial Intelligence and Computational Linguistics, suffers from the fact that computer systems are unable to automatically better themselves.Automated learning ia considered a very difficult problem, especially when applied to natural language understanding. Consequently, little effort ha8 been focused on this problem. Some pioneering work in Artificial intelligence, such as AM [I] and Winston's learning system 1"2] strove to learn or discover concept descriptions in well-defined domains. Although their efforts produced interesting Ideas and techniques, these techniques do not fully extend to • domain as complex as natural language analysis.Rather than attempting the formidable task of creating a language learning system, I will discuss techniques for Incrementally Increasing the abilities of a flexible language analyzer. There are many tasks that can be considered "Incremental language learning". Initially the learning domain Is restricted to learning the meaning of new words and generalizing existing word definitions. There ere a number of A.I. techniques, and combinations of these techniques capable of exhibiting incremental learning behavior. I first discuss FOULUP and POLITICS, two programs that exhibit a limited capability for Incremental word learning. Secondly, the technique of semantic constraint projection end Integration, as Implemented in POLITICS, Is analyzed in some detail. Finally, I discuss the application of some general learning techniques to the problem of generalizing word definitions end understanding metaphors. | null | Words can have many senses, some more n"neral than others. Let us look at the problem of gen lizlng the semantic definition of a word. Consider the case where "barrier" is defined to be a physical object that dlsenables a transfer of location. (e.g. "The barrier on the road Is blocking my way.") Now, let us interpret the sentence, "Import quotas form a barrier to International trade." Clearly, an Import quota Is not • physical object. Thus, one can minimally generalize "barrier" to mean "anything that disc.shies s physical transfer of location." Let us substitute "tariff" for "quota" In our example. This suggests that our meaning for "barrier" is insufficiently general. A tariff cannot disensble physical transfer; tariffs dime.able willingness to buy or sell goods. Thus, one can further generalize the meaning of barrier to be: "anything that dlaenablee any type of transfer", Yet, Urea trace of the generalization process must be remembered because the original meaning is often preferred, or metaphorically referenced. Consider: "The trade barriers were lifted. • and "The new legislation bulldozed existing trade barriers. • rheas sentences can only be understood metaphorically. rhat is, one needs to refer to the original meaning of ~barrier" as a physical object, In order for •lifting" or 'bulldozing" to make sense. After understanding the literal leaning of a "bulldozed barrier", the next step Is to infer he consequence of such aft action, namely, the barrier no )nger exists. Finally, one can refer to the generalized leaning of "barrier" to interpret the proPoaltion that •The ew legislation caused the trade barriers to be no longer In xietence."propose the *ollowing rules to generalize word definitions ld understand metaphorical references to their ortglnol, mmel definition: 1 ) If the definition of a word violates the semantic constraints projected from an interpretation of the rest of the sentence, create a new word-sense definition that copies the old deflnltiml minimally relaxing (I.e., generalizing) the violated constraint.2) In Interpreting new sentences always prefer the mast specific definition if applicable.3) If the generalized definition Is encountered again in Interpreting text, make It part of the permanent dictionary.• word definition requires further generalization, choose the existing most general definition and minimally relax Its violated semantic constraints until a new, yet more general definition Is formed.5) If the case frame formulated in interpreting a sentence projects more specific semantic constraints onto the word meaning than those consistent with rite entire sentence, Interpret the word usln(! the most specific definition conslste.t with the case frame. If the resultant meaning of the case frame Is inconsistent with the interpretation of the whole sentence, Infer the most likely consequence of the pMtlally-build Conceptual Dependency case frame, and use this consequence In Interpreting the rest of the sentence.The process described by rule 5 enables one to Interpret the metaphorical uses of words like "lifted" and "bulldozed" In our earlier examples. The literal meaning of each word i8 applied to the object case, (i.e., "barrier•), and the Inferred consequence (i.e., destruction of the barrier) i8 used to Interpret the full sentence. | Learning word definitions In semantically-rich contexts Is perhaps one of the simpler tasks of incremental learning. Initially I confine my discussion to situations where the meaning of a word can be learned from the Immediately surrounding context. Later I relax this criterion to see how global context and multiple examples can help to learn the meaning of unknown words.The FOULUP program [3] learned the meaning of some unknown words in the context of applying s script to understand a story. Scripts [4, 5] are frame-like knowledge representations abstracting the important features and causal structure of mundane events. Scripts have general expectations of the actions and objects that will be encountered in processing a story. For Instance, the restaurant script expects to see menus, waitresses, and customers ordering and eating food (at different pre-specifled times In the story).FOULUP took advantage of these script expectations to conclude that Items referenced in the story, which were part of expected actions, were Indeed names of objects that the script expected to see. These expectations were used to form definitions of new words. For instance, FOULUP induced the meaning of "Rabbit" in, "A Rabbit veered off the road and struck a tree," to be a self-propelled vehicle. The system used information about the automobile accident script to match the unknown word with the script-role "VEHICLE", because the script knows that the only objects that veer off roads to smash Into road-side obstructions ere self propelled vehicles.The POLITICS system E6, 7] induces the meanings of unknown words by a one*pass syntactic and semantic constraint projection followed by conceptual enrichment from planning and world-knowledge inferences. Consider how POLITICS proceeds when It encounters the unknown word "MPLA" In analyzing the sentence: "Russia sent massive arms shipments to the MPLA In Angola."Since "MPLA" follows the article '*the N it must be a noun, adjective or adverb. After the word "MPLA", the preposition "in" Is encountered, thus terminating the current prepositional phrase begun with "to". Hence, since all well-formed prepositional phrases require a head noun, and the "to" phrase has no other noun, "MPLA" must be the head noun.Thus, by projecting the syntactic constraints necessary for the sentence to be well formed, one learn8 the syntactic category of an unknown word. it Is not always possible to narrow the categorization of a word to a single syntactic category from one example. In such cases, I propose Intersecting the sets of possible syntactic categories from more then one sample use of the unknown word until the Intersection has a single element.POLITICS learns the meaning of the unknown word by a similar, but substantially more complex, application of the same principle of projecting constraints from other parts of the sentence and subsequently Integrating these constraints to oonetruot a meaning representation.In the example above, POLITICS analyzes the verb "to send" as either in ATRANS or s PTRAflS. (Schank [8] J~ERPONe <ls~ NWISER vii (, llOMI) What has the analyzer learned about "MPLA" as s result of formulating the CD case frame? Clearly the MPLA can only be an actor (I.e., s person, an Institution or s political entity in the POLITICS domain) or s location. Anything else would violate the constraints for the recipient case In both ATRANS end PTRANS. Furthermore, the analyzer knows that the location of the MPLA Is Inside Angola. This Item of Information is integrated with the case constraints to form a partial definition of "MPLA". Unfortunately both Iocatlcms and actors can be located inside countries; thus, the identity of the MPLA is still not uniquely resolved. POLITICS assigns the name RECIP01 to the partial definition of "MPLA" and proceeds to apply Its Inference rules tO understand the political Implications of the event. Here I discuss only the Inferences relevant for further specifying the meaning of -MPLA m .POLITICS Is a goal-driven tnferencer. It must explain ell actions In terms of the goals of the actors and recipients. The emphasis on inducing the goals of actors and relating their actions to means of achieving these goals is Integral to the theory of subjective understanding embodied in POLITICS. (See [7] for a detailed discussion.) Thus, POLITICS tries to determine how the action of sending weapons can be related to the goals of the Soviet Union or any other possible actors involved in the situation. POLITICS k~s that Angola was Jn a state of civto war; that Is, a state where political factions were .'xerclstng their goals of taking military and, therefore, political control of a country. Since po6ssssing weapons Is a precondition to military actions, POLITICS infers that the recipient of the weapons may have been one of the poliUcal factions. (Weapons ere s means to fulfUllng the goal of • political faction, therefore POLITICS Is able to explain why the faction wants to receive weapons.) Thus, MPLA Is Inferred to be a political faction. This Inference is Integrated with the existing partial definition and found to be consistent. Finally, the original action Is refined to be an ATRANS, as transfer of possession of the weapons (not merely their k:mation) helps the political faction to achieve Its military goal.Next, POLITICS tries to determine how sending weapons to s military faction can further the goals of the Soviet Union. Communist countries have the goal of spreading their ' Ideology. POLITICS concludes that this goal can be fulfilled only if the government of Angola becomes communist. Military aid to s political faction has the standard goal of military takeover of the government.Putting these two facts together, POLITICS concludes that the Russian goal can be fulfilled if the MPLA, which may become the new Angeles government, is Communist. The definition formed for MPLA Is ae follows: The reason why memory entries are distinct from dictionary definitions is that there is no one-to-one mapping between the two. For Instance, "Russia" and "Soviet Union" are two separate dictionary entries that refer to the same concept in memory. Similarly, the concept of SCONT (social or political control) abstracts Information useful for the goal-driven inferences, but has no corresponding entry in the lexicon, as I found no example where such concept was explicitly mentioned In newspaper headlines of political conflicts (i.e., POLITICS' domain).QI~'I i~a1"~Some of the Inferences that POLITICS made are much more prone to error than others. More specifically, the syntactic constraint projections and the CD case-frame projections ere quite certain, but the goal-driven Inferences are only reasonable guesses. For Instance, the MPLA coWd have been • plateau where Russia dePosited Its weapons for later delivery.Given such possibilities for error, two possible strategies to deei with the problem of uncertain inference come to mind. First, the system could be restricted to making only the more certain constraint projection and integration inferences. This does not usually produce s complete definition, but the process may be Iterated for other exemplars where the unknown word Is used in different semantic contexts. Each time the new word Is encountered, the semantic constraints are integrated with the previous partial definition until a complete definition is formulated. The problem with this process Is that it may require a substantial number of iterations to converge upon s meaning representation, end when it eventually does, this representation wtll not be as rich as the representation resulting from the less certain goal-driven inferences. For Instance, it would be impossible to conclude that the MPLA was Communist and wanted to take over Angola only by projecting semantic constraints.The second method is based on the system's ability to recover from inaccurate inferences. This is the method i implemented in POLITICS. The first step requires the deteotlon of contradictions between the Inferred Information end new Incoming information. The next step is to assign blame to the appropriate culprit, i.e., the inference rule that asserted the incorrect conclusion. Subsequently, the system must delete the inaccurate assertion and later inferences that depended upon it. (See [9] for a model of truth maintenance.) The final step is to use the new information to correct the memory entry. The optimal system within my paradigm would use a combination of both strategies -It would use Its maximal Inference capability, recover when Inconsistencies arise, and iterate over many exemplars to refine and confirm the meaning of the new word. The first two criteria are present in the POLITICS implementation, but the system sto~s building a new definition after processing a single exemplar unless it detects a contradiction.Let us briefly trace through an example where PC~.ITICS la told that the MPLA is indeed a pisteau after it inferred the meaning to be a political faction. Interpretation when It tries to integrate "the MPLA plateau" with its previous definition of "MPLA". Political factions and plateaus ere different conceptual classes. Furthermore, the new Input states that the Zungsbl received the weapons, not the MPLA. Assuming that the Input Its correct, POLITICS searches for an Inference rule to assign blame for the present contradiction. This Is done simply by temporarily deleting the result of each inference rule that was activated in the original interpretation until the contradiction no longer exists. The rule that concluded that the MPLA was a political faction Is found to resolve both contradictions If deleted.Since recipients of military aid must be political entitles, the MPLA being s geographical location no longer qualifies as a military aid recipient.Finally, POLITICS must check whether the inference rules that depended upon the result of the deleted rule are no longer applicable. Rules, such as the one that concluded that the political faction was communist, depended upon there being a political faction receiving military aid from Russia. The Zungabi now fulfll:s this role; therefore, the inferences about the MPLA are transfered to the Zungabl, and th~ MPLA Is redefined to be a plateau. (Note: the word "Zungabl" was constructed for this example. The MPLA is the present ruling body of Angola.)The POL)TICS Implementation of the project-and-integrate technique ts by no means complete. POLITICS can only Induce the meaning of concrete or proper nouns when there Is sufficient contextual information In a single exemplar. Furthermore, POLITICS assumes that each unknown word will have only one meaning. In general It is useful to realize when a word Is used to mean something other than Its definition, and subsequently formulate an alternative definition.I Illustrate the case where many examples are required to narrow down the meaning of s word with the following example: "Johnny told Mary that If she didn't give him the toy, he would <unknown-word) her." One can induce that the unknown word Is a verb, but its meaning can only be guessed at, In general terms, to be something unfavorable to Mary. For Instance, the unknown word could mean "take the object from", or "cause injury to". One needs more then one example of the unknown word used to mean the same thing In different contexts. Then one has s much richer, combined context from which the meaning can be projected with greater precision.1 diagrams the general project-and-integrate algorithm. This extended version of POLITICS' word-learning technique addresses the problems of iterating over many examples, multiple word definitions, and does not restrict its Input to certain classes of nouns.There are a multitude of ways to incrementally Improve the language understanding capabilities of a system. In this paper I discussed in some detail the process of learning new w~rde. In lesser detail I presented some ideas on how to generalize word meanings and Interpret metaphorical uses of individual words. There are many more aspects to learning language and understanding metaphors that I have not touched upon, For Instance, many metaphors transcend Individual words and phrases. Their Interpretation may require detailed cultural knowledge [10] .In order to place some perspective on project-and-integrate learning method, consider throe general learning mechanisms capable of implementing different aspects of Incremental language learning.Learning hy example. This Is perhaps the most general learning strategy. From several exemplars, one can intersect the common concept by, If necessary, minimally generalizing the meaning of the known part of each example until a common aubpart Is found by Intersection. This common eubpart Is likely to be the meaning of the unknown section of each exemplar.Learning by near-miss analysis. Winston [2] takes full advantage of this technique, it may be usefully applied to a natural language system that can Interactlveiy generate utterances using the words it learned, and later be told whether It used those words correctly, whether It erred seriously, or whether It came close but failed to understand a subtle nuance In meaning.Learning by contextual expectation. EasanUally FOULUP and POLITICS use the method of projecting contextual expectations to the linguistic element whose meaning Is to be Induced. Much more mileage can be gotten from this method, especially If one uses strong syntactic constraints and expectations from other knowledge sources, such as s discourse model, s narrative model, knowledge about who is providing the information, and why the information Is being provided. | null | Main paper:
introduction:
Natural language analysis, like most other subfields of Artificial Intelligence and Computational Linguistics, suffers from the fact that computer systems are unable to automatically better themselves.Automated learning ia considered a very difficult problem, especially when applied to natural language understanding. Consequently, little effort ha8 been focused on this problem. Some pioneering work in Artificial intelligence, such as AM [I] and Winston's learning system 1"2] strove to learn or discover concept descriptions in well-defined domains. Although their efforts produced interesting Ideas and techniques, these techniques do not fully extend to • domain as complex as natural language analysis.Rather than attempting the formidable task of creating a language learning system, I will discuss techniques for Incrementally Increasing the abilities of a flexible language analyzer. There are many tasks that can be considered "Incremental language learning". Initially the learning domain Is restricted to learning the meaning of new words and generalizing existing word definitions. There ere a number of A.I. techniques, and combinations of these techniques capable of exhibiting incremental learning behavior. I first discuss FOULUP and POLITICS, two programs that exhibit a limited capability for Incremental word learning. Secondly, the technique of semantic constraint projection end Integration, as Implemented in POLITICS, Is analyzed in some detail. Finally, I discuss the application of some general learning techniques to the problem of generalizing word definitions end understanding metaphors.
learning from script expectations:
Learning word definitions In semantically-rich contexts Is perhaps one of the simpler tasks of incremental learning. Initially I confine my discussion to situations where the meaning of a word can be learned from the Immediately surrounding context. Later I relax this criterion to see how global context and multiple examples can help to learn the meaning of unknown words.The FOULUP program [3] learned the meaning of some unknown words in the context of applying s script to understand a story. Scripts [4, 5] are frame-like knowledge representations abstracting the important features and causal structure of mundane events. Scripts have general expectations of the actions and objects that will be encountered in processing a story. For Instance, the restaurant script expects to see menus, waitresses, and customers ordering and eating food (at different pre-specifled times In the story).FOULUP took advantage of these script expectations to conclude that Items referenced in the story, which were part of expected actions, were Indeed names of objects that the script expected to see. These expectations were used to form definitions of new words. For instance, FOULUP induced the meaning of "Rabbit" in, "A Rabbit veered off the road and struck a tree," to be a self-propelled vehicle. The system used information about the automobile accident script to match the unknown word with the script-role "VEHICLE", because the script knows that the only objects that veer off roads to smash Into road-side obstructions ere self propelled vehicles.
constraint projection in politics:
The POLITICS system E6, 7] induces the meanings of unknown words by a one*pass syntactic and semantic constraint projection followed by conceptual enrichment from planning and world-knowledge inferences. Consider how POLITICS proceeds when It encounters the unknown word "MPLA" In analyzing the sentence: "Russia sent massive arms shipments to the MPLA In Angola."Since "MPLA" follows the article '*the N it must be a noun, adjective or adverb. After the word "MPLA", the preposition "in" Is encountered, thus terminating the current prepositional phrase begun with "to". Hence, since all well-formed prepositional phrases require a head noun, and the "to" phrase has no other noun, "MPLA" must be the head noun.Thus, by projecting the syntactic constraints necessary for the sentence to be well formed, one learn8 the syntactic category of an unknown word. it Is not always possible to narrow the categorization of a word to a single syntactic category from one example. In such cases, I propose Intersecting the sets of possible syntactic categories from more then one sample use of the unknown word until the Intersection has a single element.POLITICS learns the meaning of the unknown word by a similar, but substantially more complex, application of the same principle of projecting constraints from other parts of the sentence and subsequently Integrating these constraints to oonetruot a meaning representation.In the example above, POLITICS analyzes the verb "to send" as either in ATRANS or s PTRAflS. (Schank [8] J~ERPONe <ls~ NWISER vii (, llOMI) What has the analyzer learned about "MPLA" as s result of formulating the CD case frame? Clearly the MPLA can only be an actor (I.e., s person, an Institution or s political entity in the POLITICS domain) or s location. Anything else would violate the constraints for the recipient case In both ATRANS end PTRANS. Furthermore, the analyzer knows that the location of the MPLA Is Inside Angola. This Item of Information is integrated with the case constraints to form a partial definition of "MPLA". Unfortunately both Iocatlcms and actors can be located inside countries; thus, the identity of the MPLA is still not uniquely resolved. POLITICS assigns the name RECIP01 to the partial definition of "MPLA" and proceeds to apply Its Inference rules tO understand the political Implications of the event. Here I discuss only the Inferences relevant for further specifying the meaning of -MPLA m .
uncertain inference in learning:
POLITICS Is a goal-driven tnferencer. It must explain ell actions In terms of the goals of the actors and recipients. The emphasis on inducing the goals of actors and relating their actions to means of achieving these goals is Integral to the theory of subjective understanding embodied in POLITICS. (See [7] for a detailed discussion.) Thus, POLITICS tries to determine how the action of sending weapons can be related to the goals of the Soviet Union or any other possible actors involved in the situation. POLITICS k~s that Angola was Jn a state of civto war; that Is, a state where political factions were .'xerclstng their goals of taking military and, therefore, political control of a country. Since po6ssssing weapons Is a precondition to military actions, POLITICS infers that the recipient of the weapons may have been one of the poliUcal factions. (Weapons ere s means to fulfUllng the goal of • political faction, therefore POLITICS Is able to explain why the faction wants to receive weapons.) Thus, MPLA Is Inferred to be a political faction. This Inference is Integrated with the existing partial definition and found to be consistent. Finally, the original action Is refined to be an ATRANS, as transfer of possession of the weapons (not merely their k:mation) helps the political faction to achieve Its military goal.Next, POLITICS tries to determine how sending weapons to s military faction can further the goals of the Soviet Union. Communist countries have the goal of spreading their ' Ideology. POLITICS concludes that this goal can be fulfilled only if the government of Angola becomes communist. Military aid to s political faction has the standard goal of military takeover of the government.Putting these two facts together, POLITICS concludes that the Russian goal can be fulfilled if the MPLA, which may become the new Angeles government, is Communist. The definition formed for MPLA Is ae follows: The reason why memory entries are distinct from dictionary definitions is that there is no one-to-one mapping between the two. For Instance, "Russia" and "Soviet Union" are two separate dictionary entries that refer to the same concept in memory. Similarly, the concept of SCONT (social or political control) abstracts Information useful for the goal-driven inferences, but has no corresponding entry in the lexicon, as I found no example where such concept was explicitly mentioned In newspaper headlines of political conflicts (i.e., POLITICS' domain).QI~'I i~a1"~Some of the Inferences that POLITICS made are much more prone to error than others. More specifically, the syntactic constraint projections and the CD case-frame projections ere quite certain, but the goal-driven Inferences are only reasonable guesses. For Instance, the MPLA coWd have been • plateau where Russia dePosited Its weapons for later delivery.
a strategy for dealing with uncertainty:
Given such possibilities for error, two possible strategies to deei with the problem of uncertain inference come to mind. First, the system could be restricted to making only the more certain constraint projection and integration inferences. This does not usually produce s complete definition, but the process may be Iterated for other exemplars where the unknown word Is used in different semantic contexts. Each time the new word Is encountered, the semantic constraints are integrated with the previous partial definition until a complete definition is formulated. The problem with this process Is that it may require a substantial number of iterations to converge upon s meaning representation, end when it eventually does, this representation wtll not be as rich as the representation resulting from the less certain goal-driven inferences. For Instance, it would be impossible to conclude that the MPLA was Communist and wanted to take over Angola only by projecting semantic constraints.The second method is based on the system's ability to recover from inaccurate inferences. This is the method i implemented in POLITICS. The first step requires the deteotlon of contradictions between the Inferred Information end new Incoming information. The next step is to assign blame to the appropriate culprit, i.e., the inference rule that asserted the incorrect conclusion. Subsequently, the system must delete the inaccurate assertion and later inferences that depended upon it. (See [9] for a model of truth maintenance.) The final step is to use the new information to correct the memory entry. The optimal system within my paradigm would use a combination of both strategies -It would use Its maximal Inference capability, recover when Inconsistencies arise, and iterate over many exemplars to refine and confirm the meaning of the new word. The first two criteria are present in the POLITICS implementation, but the system sto~s building a new definition after processing a single exemplar unless it detects a contradiction.Let us briefly trace through an example where PC~.ITICS la told that the MPLA is indeed a pisteau after it inferred the meaning to be a political faction. Interpretation when It tries to integrate "the MPLA plateau" with its previous definition of "MPLA". Political factions and plateaus ere different conceptual classes. Furthermore, the new Input states that the Zungsbl received the weapons, not the MPLA. Assuming that the Input Its correct, POLITICS searches for an Inference rule to assign blame for the present contradiction. This Is done simply by temporarily deleting the result of each inference rule that was activated in the original interpretation until the contradiction no longer exists. The rule that concluded that the MPLA was a political faction Is found to resolve both contradictions If deleted.Since recipients of military aid must be political entitles, the MPLA being s geographical location no longer qualifies as a military aid recipient.Finally, POLITICS must check whether the inference rules that depended upon the result of the deleted rule are no longer applicable. Rules, such as the one that concluded that the political faction was communist, depended upon there being a political faction receiving military aid from Russia. The Zungabi now fulfll:s this role; therefore, the inferences about the MPLA are transfered to the Zungabl, and th~ MPLA Is redefined to be a plateau. (Note: the word "Zungabl" was constructed for this example. The MPLA is the present ruling body of Angola.)
extending the project and integrate method:
The POL)TICS Implementation of the project-and-integrate technique ts by no means complete. POLITICS can only Induce the meaning of concrete or proper nouns when there Is sufficient contextual information In a single exemplar. Furthermore, POLITICS assumes that each unknown word will have only one meaning. In general It is useful to realize when a word Is used to mean something other than Its definition, and subsequently formulate an alternative definition.I Illustrate the case where many examples are required to narrow down the meaning of s word with the following example: "Johnny told Mary that If she didn't give him the toy, he would <unknown-word) her." One can induce that the unknown word Is a verb, but its meaning can only be guessed at, In general terms, to be something unfavorable to Mary. For Instance, the unknown word could mean "take the object from", or "cause injury to". One needs more then one example of the unknown word used to mean the same thing In different contexts. Then one has s much richer, combined context from which the meaning can be projected with greater precision.1 diagrams the general project-and-integrate algorithm. This extended version of POLITICS' word-learning technique addresses the problems of iterating over many examples, multiple word definitions, and does not restrict its Input to certain classes of nouns.
generalizing word definitions.:
Words can have many senses, some more n"neral than others. Let us look at the problem of gen lizlng the semantic definition of a word. Consider the case where "barrier" is defined to be a physical object that dlsenables a transfer of location. (e.g. "The barrier on the road Is blocking my way.") Now, let us interpret the sentence, "Import quotas form a barrier to International trade." Clearly, an Import quota Is not • physical object. Thus, one can minimally generalize "barrier" to mean "anything that disc.shies s physical transfer of location." Let us substitute "tariff" for "quota" In our example. This suggests that our meaning for "barrier" is insufficiently general. A tariff cannot disensble physical transfer; tariffs dime.able willingness to buy or sell goods. Thus, one can further generalize the meaning of barrier to be: "anything that dlaenablee any type of transfer", Yet, Urea trace of the generalization process must be remembered because the original meaning is often preferred, or metaphorically referenced. Consider: "The trade barriers were lifted. • and "The new legislation bulldozed existing trade barriers. • rheas sentences can only be understood metaphorically. rhat is, one needs to refer to the original meaning of ~barrier" as a physical object, In order for •lifting" or 'bulldozing" to make sense. After understanding the literal leaning of a "bulldozed barrier", the next step Is to infer he consequence of such aft action, namely, the barrier no )nger exists. Finally, one can refer to the generalized leaning of "barrier" to interpret the proPoaltion that •The ew legislation caused the trade barriers to be no longer In xietence."propose the *ollowing rules to generalize word definitions ld understand metaphorical references to their ortglnol, mmel definition: 1 ) If the definition of a word violates the semantic constraints projected from an interpretation of the rest of the sentence, create a new word-sense definition that copies the old deflnltiml minimally relaxing (I.e., generalizing) the violated constraint.2) In Interpreting new sentences always prefer the mast specific definition if applicable.3) If the generalized definition Is encountered again in Interpreting text, make It part of the permanent dictionary.• word definition requires further generalization, choose the existing most general definition and minimally relax Its violated semantic constraints until a new, yet more general definition Is formed.5) If the case frame formulated in interpreting a sentence projects more specific semantic constraints onto the word meaning than those consistent with rite entire sentence, Interpret the word usln(! the most specific definition conslste.t with the case frame. If the resultant meaning of the case frame Is inconsistent with the interpretation of the whole sentence, Infer the most likely consequence of the pMtlally-build Conceptual Dependency case frame, and use this consequence In Interpreting the rest of the sentence.The process described by rule 5 enables one to Interpret the metaphorical uses of words like "lifted" and "bulldozed" In our earlier examples. The literal meaning of each word i8 applied to the object case, (i.e., "barrier•), and the Inferred consequence (i.e., destruction of the barrier) i8 used to Interpret the full sentence.
coral.cling remarks:
There are a multitude of ways to incrementally Improve the language understanding capabilities of a system. In this paper I discussed in some detail the process of learning new w~rde. In lesser detail I presented some ideas on how to generalize word meanings and Interpret metaphorical uses of individual words. There are many more aspects to learning language and understanding metaphors that I have not touched upon, For Instance, many metaphors transcend Individual words and phrases. Their Interpretation may require detailed cultural knowledge [10] .In order to place some perspective on project-and-integrate learning method, consider throe general learning mechanisms capable of implementing different aspects of Incremental language learning.Learning hy example. This Is perhaps the most general learning strategy. From several exemplars, one can intersect the common concept by, If necessary, minimally generalizing the meaning of the known part of each example until a common aubpart Is found by Intersection. This common eubpart Is likely to be the meaning of the unknown section of each exemplar.Learning by near-miss analysis. Winston [2] takes full advantage of this technique, it may be usefully applied to a natural language system that can Interactlveiy generate utterances using the words it learned, and later be told whether It used those words correctly, whether It erred seriously, or whether It came close but failed to understand a subtle nuance In meaning.Learning by contextual expectation. EasanUally FOULUP and POLITICS use the method of projecting contextual expectations to the linguistic element whose meaning Is to be Induced. Much more mileage can be gotten from this method, especially If one uses strong syntactic constraints and expectations from other knowledge sources, such as s discourse model, s narrative model, knowledge about who is providing the information, and why the information Is being provided.
Appendix:
| null | null | null | null | {
"paperhash": [
"carbonell|politics:_automated_ideological_reasoning",
"doyle|truth_maintenance_systems_for_problem_solving",
"granger|foul-up:_a_program_that_figures_out_meanings_of_words_from_context",
"winston|learning_structural_descriptions_from_examples",
"carbonell|subjective_understanding,_computer_models_of_belief_systems",
"cullingford|script_application:_computer_understanding_of_newspaper_stories."
],
"title": [
"POLITICS: Automated Ideological Reasoning",
"Truth Maintenance Systems for Problem Solving",
"FOUL-UP: A Program that Figures Out Meanings of Words from Context",
"Learning structural descriptions from examples",
"Subjective understanding, computer models of belief systems",
"Script application: computer understanding of newspaper stories."
],
"abstract": [
"POLITICS is a system of computer programs which simulates humans in comprehending and responding to world events from a given political or ideological perspective. The primary theoretical motivations were: (1) the implementation of a functional system which applies the knowledge structures of Schank and Abelson (1977) to the domain of simulating political belief systems; (2) the development of a tentative theory of intentional goal conflicts and counterplanning. Secondary goals of the POLITICS project include developing a representation for belief systems, investigating cognitive processes such as goal-directed inferencing, and the integration of several types of knowledge representations into a functional system.",
"Abstract : The thesis developed in this paper is that reasoning programs which take care to record the logical justifications for program beliefs can apply several powerful, but simple, domain-independent algorithms to: (1) maintain the consistency of program beliefs; (2) realize substantial search efficiencies; and (3) automatically summarize explanations of program beliefs. This report describes techniques for representing, recording, maintaining, and using justifications for beliefs. Also presented is an annotated implementation of a domain-independent program.",
"The inferencing task of figuring out words from context is implemented in the presence of a large database of world knowledge. The program does not require interaction with the user, but rather uses internal parser expectations and knowledge embodied in scripts to figure out likely definitions for unknown words, and to create context-specific definitions for such words.",
"Massachusetts Institute of Technology. Dept. of Electrical Engineering. Thesis. 1970. Ph.D.",
"Abstract : Modeling human understanding of natural language requires a model of the processes underlying human thought. No two people think exactly alike; different people subscribe to different beliefs and are motivated by different goals in their activities. A theory of subjective understanding has been proposed to account for subjectively-motivated human thinking ranging from ideological belief to human discourse and personality traits. A process-model embodying this theory has been implemented in a computer system, POLITICS. POLITICS models human ideological reasoning in understanding the natural language text of international political events. POLITICS can model either liberal or conservative ideologies. Each ideology produces a different interpretation of the input event. POLITICS demonstrates its understanding by answering questions in natural language question-answer dialogs.",
"Abstract : The report describes a computer story understander which applies knowledge of the world to comprehend what it reads. The system, called SAM, reads newspaper articles from a variety of domains, then demonstrates its understanding by summarizing or paraphrasing the text, or answering questions about it. (Author)"
],
"authors": [
{
"name": [
"J. Carbonell"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"J. Doyle"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. Granger"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"P. Winston"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"J. Carbonell"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"R. E. Cullingford"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null,
null,
null
],
"s2_corpus_id": [
"16520160",
"1340440",
"9255668",
"106617047",
"142895805",
"60708295"
],
"intents": [
[],
[],
[],
[],
[],
[]
],
"isInfluential": [
false,
false,
false,
false,
false,
false
]
} | null | 548 | 0.10219 | null | null | null | null | null | null | null | null |
f651b3ab439547070f75926456bc33f34088b982 | 30393211 | null | The Role Of Focussing in Interpretation of Pronouns | In this p;,per I [ discuss the formal relationship between the process of focussing and interpret;ition of pronominal anaphora. The discussion of focussing extends the work of Grosz [1977]. Foct,ssing is defined algorithmical]y as a process which chooses a focus of attention in a discourse and moves it around as the speaker's focus ch'mges. The paper shows how to use the focussing algorithm by ;m extended example given below. DI-I Alfred a,ld Zohar liked to play baseball. 2 They played it everyday after school before dinner. 3 After their game, the two usually went for ice cream cones. 4 They tasted really good. 5 Alfred always had the vanilla super scooper, 6 while Zohar tried the flavor of the day cone. 7 After the cones had been eaten, 8 the boys went home to study. | {
"name": [
"Sidner, Candace L."
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 13 | 5 | null | null | In this p;,per I [ discuss the formal relationship between the process of focussing and interpret;ition of pronominal anaphora. The discussion of focussing extends the work of Grosz [1977] . Foct,ssing is defined algorithmical]y as a process which chooses a focus of attention in a discourse and moves it around as the speaker's focus ch'mges. The paper shows how to use the focussing algorithm by ;m extended example given below.DI-I Alfred a,ld Zohar liked to play baseball. 2 They played it everyday after school before dinner. 3 After their game, the two usually went for ice cream cones. 4 They tasted really good. 5 Alfred always had the vanilla super scooper, 6 while Zohar tried the flavor of the day cone. 7 After the cones had been eaten, 8 the boys went home to study.In this example, the discourse focusses initially on baseball. The focus moves in DI-3 to the ice cream cone. Using this example, I show how the formal algorithm computes focus and determines how the focus moves according to the signals which the speaker uses in discourse to indicate the movement.Given a process notion of focus, the paper reviews the difficulties with previous approaches (Rieger [1974] , Charniak [1972] , Winograd [1971] , Hobbs [1975] and Lockman [1978] ). Briefly, the first four authors all point out the need for inferencing as part of anaphora disambiguation, but each of their schemes for inferencing suffer from the need for control which will reduce the combinatorial search or which will insure only one search path is taken. In addition, Winograd and Lockman are aware of pronopn phenomena which cannot be treated strictly by inference, as shown below.D2-1 I haven't seen Jeff for several days.2 Carl thinks h e's studying for his exams.3 Oscar says hj is sick, 4 but I think he went to the Cape with Linda.1. This report describes research done at the Artificial Intelligence Laboratory of the Massachusetts Institute of Technology. Support for the laboratory's artificial intelligence research is provided in part by the Advanced Research Projects Agency of the Department of Defense under the Office of Naval Research under Contract Number N00014-73-C4)643.However, their approaches are either simple heuristics which offer no unified treatment (Winograd) or require the computation of a structure which must assume the pronouns have previously been resolved (Lockman) .In order to state formal rules for pronoun interpretation, the concept of antecedence is defined computationally as a relationship among elements represented in a database. Using this framework, the paper supports two claims by means of rules for antecedence.The focus provides a source of antecedence in rules for interpreting pronominal anaphora. 2. Focussing provides a control for the inferencing necessary for some kinds of anaphora.The use of D3 below.for pronominal anaphora rely on three sources of information: syntactic criteria, semantic selectional and consistency checks from inferencing procedures. these rules are presented for examples D2 above and Whitimore isn't such a good thief. The man whose watch he stole called the police. 3 They catzght him.These examples show how to use the three sources of information to support or reject a predicted antecedence. In particular, inferencing is controlled by checking for consistency on a predicted choice rather than by search ~lsing general inference.The paper also indicates what additional requirements are needed for a full treatment of pronominal anphora. These include use of a representation such as that of Webber [197g] ; linguistic rules such as the disjoint reference rules of Lasnik [[976] and Reinhart [[976] as well as rules of anapbora in logical form given by Cbomsky [1976] ; and presence of actor loci such as they in D3. The nature of these requirements is discussed, while the computational inclusion of them is found in $idner [ 1979] ."77 | null | null | null | Main paper:
:
In this p;,per I [ discuss the formal relationship between the process of focussing and interpret;ition of pronominal anaphora. The discussion of focussing extends the work of Grosz [1977] . Foct,ssing is defined algorithmical]y as a process which chooses a focus of attention in a discourse and moves it around as the speaker's focus ch'mges. The paper shows how to use the focussing algorithm by ;m extended example given below.DI-I Alfred a,ld Zohar liked to play baseball. 2 They played it everyday after school before dinner. 3 After their game, the two usually went for ice cream cones. 4 They tasted really good. 5 Alfred always had the vanilla super scooper, 6 while Zohar tried the flavor of the day cone. 7 After the cones had been eaten, 8 the boys went home to study.In this example, the discourse focusses initially on baseball. The focus moves in DI-3 to the ice cream cone. Using this example, I show how the formal algorithm computes focus and determines how the focus moves according to the signals which the speaker uses in discourse to indicate the movement.Given a process notion of focus, the paper reviews the difficulties with previous approaches (Rieger [1974] , Charniak [1972] , Winograd [1971] , Hobbs [1975] and Lockman [1978] ). Briefly, the first four authors all point out the need for inferencing as part of anaphora disambiguation, but each of their schemes for inferencing suffer from the need for control which will reduce the combinatorial search or which will insure only one search path is taken. In addition, Winograd and Lockman are aware of pronopn phenomena which cannot be treated strictly by inference, as shown below.D2-1 I haven't seen Jeff for several days.2 Carl thinks h e's studying for his exams.3 Oscar says hj is sick, 4 but I think he went to the Cape with Linda.1. This report describes research done at the Artificial Intelligence Laboratory of the Massachusetts Institute of Technology. Support for the laboratory's artificial intelligence research is provided in part by the Advanced Research Projects Agency of the Department of Defense under the Office of Naval Research under Contract Number N00014-73-C4)643.However, their approaches are either simple heuristics which offer no unified treatment (Winograd) or require the computation of a structure which must assume the pronouns have previously been resolved (Lockman) .In order to state formal rules for pronoun interpretation, the concept of antecedence is defined computationally as a relationship among elements represented in a database. Using this framework, the paper supports two claims by means of rules for antecedence.The focus provides a source of antecedence in rules for interpreting pronominal anaphora. 2. Focussing provides a control for the inferencing necessary for some kinds of anaphora.The use of D3 below.for pronominal anaphora rely on three sources of information: syntactic criteria, semantic selectional and consistency checks from inferencing procedures. these rules are presented for examples D2 above and Whitimore isn't such a good thief. The man whose watch he stole called the police. 3 They catzght him.These examples show how to use the three sources of information to support or reject a predicted antecedence. In particular, inferencing is controlled by checking for consistency on a predicted choice rather than by search ~lsing general inference.The paper also indicates what additional requirements are needed for a full treatment of pronominal anphora. These include use of a representation such as that of Webber [197g] ; linguistic rules such as the disjoint reference rules of Lasnik [[976] and Reinhart [[976] as well as rules of anapbora in logical form given by Cbomsky [1976] ; and presence of actor loci such as they in D3. The nature of these requirements is discussed, while the computational inclusion of them is found in $idner [ 1979] ."77
Appendix:
| null | null | null | null | {
"paperhash": [
"sidner|towards_a_computational_theory_of_definite_anaphora_comprehension_in_english_discourse",
"hobbs|pronoun_resolution",
"charniak|toward_a_model_of_children's_story_comprehension",
"winograd|procedures_as_a_representation_for_data_in_a_computer_program_for_understanding_natural_language",
"grosz|the_representation_and_use_of_focus_in_dialogue_understanding.",
"reinhart|the_syntactic_domain_of_anaphora",
"klappholz|contextual_reference_resolution",
"rieger|conceptual_memory:_a_theory_and_computer_program_for_processing_the_meaning_content_of_natural_langu"
],
"title": [
"Towards a computational theory of definite anaphora comprehension in English discourse",
"Pronoun resolution",
"Toward a model of children's story comprehension",
"Procedures As A Representation For Data In A Computer Program For Understanding Natural Language",
"The representation and use of focus in dialogue understanding.",
"The syntactic domain of anaphora",
"Contextual Reference Resolution",
"Conceptual memory: a theory and computer program for processing the meaning content of natural langu"
],
"abstract": [
"Abstract : This report investigates the process of focussing as a description and explanation of the comprehension of certain anaphoric expressions in English discourse. The investigation centers on the interpretation of definite anaphora, that is, on the personal pronouns, and noun phrases used with a definite article the, this, or that. Focussing is formalized as a process in which a speaker centers attention on a particular aspect of the discourse. An algorithmic description specifies what the speaker can focus on and how the speaker may change the focus of the discourse as the discourse unfolds. The algorithm allows for a simple focussing mechanism to be constructed: an element in focus, an ordered collection of alternate foci, and a stack of old foci. The data structure for the element in focus is a representation which encodes a limited set of associations between it and other elements from the discourse as well as from general knowledge. This report also establishes other constraints which are needed for the successful comprehension of anaphoric expressions. The focussing mechanism is designed to take advantage of syntactic and semantic information encoded as constraints on the choice of anaphora interpretation. These constraints are due to the work of language researchers; and the focussing mechanism provides a principled means for choosing when to apply the constraints in the comprehension process.",
"Two approaches to the problem of pronoun resolution are presented. The first is a naive algorithm that works by traversing the surface parse trees of the sentences of the text in a particular order looking for noun phrases of the correct gender and number. The algorithm is shown to incorporate many, though not all, of the constraints on co-referentiality between a nonreflective pronoun and a possible antecedent, which have been discovered recently by linguists. The algorithm clearly does not work in all cases, but the results of an examination of several hundred examples from published texts show that it performs remarkably well.In the second approach, it is shown how pronoun resolution is handled in a comprehensive system for semantic analysis of English texts. The system consists of four basic semantic operations which work by accessing a data base of 'World knowledge\" inferences, which are drawn selectively and in a context-dependent way in response to the operations. The first two operations seek to satisfy the demands made by predicates on the nature of their arguments and to discover the relations between sentences. The third operation - knitting - recognizes and merges redundant expressions. These three operations frequently result in a pronoun reference being resolved as a by-product. The fourth operation seeks to resolve those pronouns not resolved by the first three. It involves a bidirectional search of the text and 'World knowledge\" for an appropriate chain of inference and utilizes the efficiency of the naive algorithm.Four examples, including the classic examples of Winograd and Charniak, are presented that demonstrate pronoun resolution within the semantic approach.",
"Massachusetts Institute of Technology. Dept. of Electrical Engineering. Thesis. 1972. Ph.D.",
"Abstract : The paper describes a system for the computer understanding of English. The system answers questions, executes commands, and accepts information in normal English dialog. It uses semantic information and context to understand discourse and to disambiguate sentences. It combines a complete syntactic analysis of each sentence with a 'heuristic understander' which uses different kinds of information about a sentence, other parts of the discourse, and general information about the world in deciding what the sentence means.",
"Abstract : This report develops a representation of focus of attention thatcircumscribes discourse contexts within a general representation ofknowledge. Focus of attention is essential to any comprehension processbecause what and how a person understands is strongly influenced bywhere his attention is directed at a given moment. To formalize thenotion of focus, the need for and the use of focus mechanisms areconsidered from the standpoint of building a computer system that canparticipate in a natural language dialogue with a ser, Two ranges offocus, global and immediate, are investigated, and representations forincorporating them in a computer system are developed.The global focus in which an utterance is interpreted is determinedby the total discourse and situational setting of the utterance. Itinfluences what is talked about, how different concepts are introduced,and how concepts are referenced. To encode global focuscomputationally, a representation is developed that highlights thoseitems that are relevant at a given place in a dialogue. The underlyingknowledge representation is segmented into subunits, called focusspaces, that contain those items that are in the focus of attention of adialogue participant during a particular part of the dialogue.Mechanisms are required for updating the focus representation,because, as a dialogue progresses, the objects and actions that arerelevant to the conversation, and therefore in the participants' focusof attention, change. Procedures are described for deciding when andhow to shift focus in task-oriented dialogues, i.e., in dialogues inwhich the participants are cooperating in a shared task. Theseprocedures are guided by a representation of the task being performed.The ability to represent focus of attention in a languageunderstanding system results in a new approach to an important problemin discourse comprehension -- the identification of the referents ofdefinite noun phrases.",
"Thesis. 1976. Ph.D.--Massachusetts Institute of Technology. Dept. of Foreign Literatures and Linguistics.",
"With the exception of pranomial reference, little, has been written (in the field of computational linguistics) about the phenomenon of reference i n natural language. This paper investigates the power and use of reference i n natural language. and the problems involved in its resolution. An algorithm is sketched for accomplishing reference resolution using a notion of cross-sentential focus, a mechanism f o r hypothesizing a l l possible contextual references, and a judgment mechanism f o r dis - ~ r i r n i n a t i ng among the hypotheses.",
"Abstract : Humans perform vast quantities of spontaneous, subconscious computation in order to understand even the simplest language utterances. The computation is principally meaning-based. With syntax and traditional semantics playing insignificant roles. This thesis supports this conjecture by synthesis of a theory and computer program which account for many aspects of language behavior in humans. It is a theory of language and memory."
],
"authors": [
{
"name": [
"C. Sidner"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Jerry R. Hobbs"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"Eugene Charniak"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"T. Winograd"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"B. Grosz"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"T. Reinhart"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. Klappholz",
"A. Lockman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"C. Rieger"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null,
null,
null,
null,
null
],
"s2_corpus_id": [
"41092026",
"268074020",
"62620723",
"54114373",
"61114426",
"60920028",
"219304841",
"53880811"
],
"intents": [
[],
[],
[],
[],
[],
[],
[],
[]
],
"isInfluential": [
false,
false,
false,
false,
false,
false,
false,
false
]
} | - Problem: The paper discusses the formal relationship between the process of focusing and interpretation of pronominal anaphora, extending previous work by Grosz [1977].
- Solution: The paper proposes an algorithmic definition of focusing as a process that selects a focus of attention in discourse and shifts it as the speaker's focus changes. It demonstrates how this focusing algorithm can be applied through an extended example, showing how formal rules for pronoun interpretation can be established based on the concept of antecedence and the control provided by focusing for inferencing in anaphora interpretation. | 548 | 0.009124 | null | null | null | null | null | null | null | null |
1c066300af5ff1af97054ddc26759e490eccf943 | 220401 | null | Natural Language Input to a Computer-Based Glaucoma Consultation System | A "Front End" for a Computer-Based Glaucoma Consultation System is described. The system views a case as a description of a particular instance of a class of concepts called "structured objects" and builds up a representation of the instance from the sentences in the case. The information required by the consultation system is then extracted and passed on to the consultation system in the appropriately coded form. A core of syntactlc, semantic end contextual rules which are applicable to all structured objects is being developed together with a representation of the structured object GLAUCOMA-PATIENT. There is also a facility for adding domain dependent syntax, abbreviations and defaults. | {
"name": [
"Ciesielski, Victor B."
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 9 | 1 | null | Abstract: A "Front End" for a Computer-Based Glaucoma Consultation System is described. The system views a case as a description of a particular instance of a class of concepts called "structured objects" and builds up a representation of the instance from the sentences in the case.The information required by the consultation system is then extracted and passed on to the consultation system in the appropriately coded form. A core of syntactlc, semantic end contextual rules which are applicable to all structured objects is being developed together with a representation of the structured object GLAUCOMA-PATIENT.There is also a facility for adding domain dependent syntax, abbreviations and defaults. system that has a core of syntax and semantics that is applicable to all structured objects and which can be extended by domain specific syntax, idioms and defaults.work on the interpretation of hospital discharge summaries, which are very similar to case descriptions, has been done by a group at NYU [Sager 1978] . Their work has focused on the creation of formatted data bases for subsequent question answering and is syntax based. The research reported here is concerned with extracting from the case the information understandable by a consultation system and is primarily knowledge based.During the past decade a number of Medical Consultation systems have been developed, for example INTERNIST [Pople. Myers and Miller 1973] ,CASNET/GLAUCOMA [Weiss st. al. 1978] , MYCIN [Shortliffe 1976 ]. Currently still others are being developed. Some of these programs are reaching a stage where they are being used in hospitals and clinics. Such use brings with it the need for fast and natural communication with these programs for the reporting of the "clinical state" of the patient. This includes laboratory findings, symptoms, medications and certain history data. Ideally the reporting would be done by speech but this is currently beyond the state of the art in speech understanding. A more reasonable goal is to try to capture the physicians" written "Natural Language" for describing patients and to write programs to convert these descriptions to the appropriate coded input to the consultation systems.The original motivation for this research came from the desire to have natural language input of cases to CASNET/GLAUCOMA a computer-based glaucoma consultation system developed at Retgers University.A case is several paragraphs of sentences , written by a physician, which describe a patient who has glaucoma or who is suspected of having glaucoma.It was desired to have a "Natural Language Front-End" which could interpret the cases and pass the content to the consultation system. In the beginning stages it was by no means clear that it would even be possible to have a "front end" since it was expected that some sophisticated knowledge of Glaucoma would be necessary and that feedback from the consultation system would be required in understanding the input sentences. However during the course of the investigation it became clear that certain generalizations could be made from the domain of Glaucoma.The key discovery was that under some reasonable assumptions the physic iane notes could be viewed as descriptions of instances of a class of concepts called structured oblects and the knowledge needed to interpret the notes was mostly knowledge of the relationship between language and structured objects rather than knowledge of Glaucoma. The graph has a distinguished node, analogous to the root of a tree, whose label is the name of the concept. All incoming errs to the concept enter only at this distinguished or "head" node. Although the relation between PATIENT and PATIENT-MEDICATION has some surface forms that make it look like an ATTR relation this is not really the case. A "true" structured object would not have ASS links but they must be introduced to deal with GLAUCOMA-PATIENT. the formal semantics of the ASS relation are very similar to those of the ATTR and PART relations. PART SI~C The nunbers after the C prefix in Fisure l donate levels of "sub-conceptln8". Level I £s the lowest level, those concepts do not have any sub-concepts only £natancao. Note that CI-PATIENT-KIGHT-EYE is a sub-concept of C2-PATIENT-gYE, not an Instanceo CI-PATIENT-LEFT-gYE and C2-PATTENT-~IGHT-EYE are two different concepts t that is they have d/~Joint sub-structure; they are as different to the system as C-AiM and C-LEG. There is 8nod reason for this. It is possible that a different Instrument will be needed to measure the value of an attribute in the right eye than in the taft aye. Thls means that the measurement concepts got these attrlbutee will have to he different for the left and right eyes.C I-PAT-LE • C2-PAT-EYE j q S~E ! C I-PAT-LE PRESSURE M. ~c~-PAT-~YE [ C I-PAT-LE , PRESSURE-MSMT nESSURE-"S~'T, I SUBC C l-PAT-RE J ATI"R C I-PAT-P.E PRESSURE C I-PAT-~E- PRESS~E-MSMT ~C~-~AT-I PART ....~S- J MEDICATION j C I-PATIENT ATTR C I-PAT-NED- DL~MOX i c x-~ATIENT-i MET .~ c X-~AT~NT-i ATT~ c,-,ATI,.NT- ,Ic -pAT ' NT: i SEX JH (@1 SEX.-~T l /i -T dAnother example from the d~ain of slancoma show this more vividly. CI-PATIENT-LEYT-~YE-VISUAL-FIELD-~COTCMA denotes a scotoma in the left eye.A particular type of scotoma is the arcuate (bow-shaped) scotoma. This must be a separate concept since it is meaninsful to suty "double arcuste scotoma" but not "doubte scotoma", This means that the concept C .... -FIELD-AACUATE-SCOTflMA has an attribute ~hat cannot be inherited from C..,-~IELD-SCOTOMA. If a measurement concept is the alune for hor~ eyes ( The rules of instantlatlon are embedded in the core.A partial instantiation of CI-PATIENT can be done before the first sentence is processed by tracing links marked NECESSARY.Any component or attribute ins,an,laced at this stage will be introduced by a definite noun phrase while optional components will be introduced by indefinite noun phrases. | null | null | A fundamental assumption that has been made and one that is Justlfled by examination of several sets of cases is that the sentences dascrlbe an instance of a patient with the assumption that the reader already knows the concept. None of the sentences in the notes examined had an interpretation which would requlre updating the concept GLAUCCMA-PATIENT.The interpretation of a case is thus consldared to be the construction of the the corresponding instance of GLAUCOMA-PATIENT.The nature of structured objects as outlined above dlccataa that only two fundamental kinds of assertions are expected in sentences. There wlll either be an assertion about the existence of an optional component as in (5) or about the value of an attribute as in (6) and 7• There Is an arcuete scotoma od.** The pressure is 20 in the left eye. The pressure is normal os.(5) (6)Vary few of the sentences contain Just one assertion, most contain several as in (8) and (9) .There is a nasal step and an arcuete scotoma in the left eye and a central island in the right eye (8) ~he medication is I0 percent pilocarplne daily in both eyes.Even though sentences are viewed as containing assertions their meanings can be represented as sets of instances, Non-nmnerlcal measurements differ from numerical given that there is a procedure which takes these measurements in that RANGE, UNIT and QVALSET are replaced instances and incorporates them into the growing instance by VALSET. One or more members of VALSET are to be of GLAUCOMA-PATIENT. Ibis is due to the tree structure selected in creating an instance of the measurement of instances since Instantlatlon of a concept involves concept, for example: Instantlatlon of all concepts between itself and the root.In fact, many sentences in the cases do not even CI-PATIENT-SEX-MSMT VALSET (ONEOF K-MALE K-FEMALE) contain a relation but merely assert the existence of an instance or of an attribute value as in (I0) and ([1).An instance of a structured object is represented as a tree.Instances are created piece-meal as the Information trickles in from the case.In some cases the Nasal step od. (I0) a I0 year old white male.(II) ** Opthalmologlsts frequently use the abbreviations "ed" for "in the right eye", "os" for "in the left eye" and "ou" for "in hor/1 ayes" 2.2 PROVISIONAL INSTANCES Any particular noun or adjective could refer to a number of different concepts. "Medication" for" example could refer to CI-PATIENT-MEDICATION, CI-PATIENT-&IGHT-EYE-MEDICATION or (I-PATIENT-LEFT-EYE-MEDICATION. Moreover in any particular use it could be referring Co one or more of its possible referents.In t2Medicacion consists of diamox and pllocarpine drops in both eyes."medication" refers co all of its possible referents since diamox is not given to the eye but is taken orally. In addition to this, ic £s generally not possible to know at the clme of encountering a word whether it refers to an existing Instance or to a new instance. This is due to the fact thaC at the time of encountering a reference to a concept all of the values of the instance dimensions mlghc not be known. The mechanism for dealing with these problems is Co assign "provisional Instances" as the referents of words end phrases when they are scanned during the parse and to turn these provisional instances Into "real" instances when the correct parse has been found. This involves finding the values of the instance dimensions from rest of the sentence, from knowledge of defaults or perhaps from values in previous sentences. The most common Instance dimension is TIME and its value is readily obtained from the tense of the verb or from a clme phrase.If the instance dimensions indicate an existing instance then the partial provisional instance from the sentence is incorporated into the existing real instance, otherwise a new instance is created.Several mappings can be made from the representation of structured objects to syntactic classes.For example, all nodes will be referred to by nouns and noun phrases, links will be referred to by prepositions and verbs and members of a VALSET or a 0VALSET will ba referred to by adjectives.The links between concepts and cha ~rds that can be used to refer to them are made at system build time when che structured object is constructed. Some words such as "both" and "very" refer to procedures whose actions are the same no matter what the structured object.The nature of structured objects and of the sentences in cases Indicate thac a "case'* [Bruce 1975 ] approach to semantic analysis is a "natural". A case syecsm ham in fact been implemented with such cases as ATTRIBUTE, OBJECT, VALUE, and UNIT. One case that is particularly useful is FOCUS.It is used to record references Co left eye or right eye for use in embedded or conjoined sentences such as (13).The pressure in the left eye is 27 and there is an arcuate scocoma.For the reasons discussed in section 2.2 ic is necessary co assign sacs of candidate referents to soma of the case values during the course of the parse. These sacs are pruned as higher levels of the parse tree are built.It is noc really possible to vlew cha sentences comprising a case as a subset of English since many of the elementary grammatical rules are broken (e.g. frequent omission of verbs). Rather the sentences are in a medical dialect and parr of the task of wrlClng an interpreter for cases involves an anthropological investlgaclon of the dialect and its definition in some formal way.An analysls of a nt~"ber of cases revealed the following characteristics (see also [Sangscer 1978] ): I) Frequent omission of verbs and punctuation.2) ~ch use of abbreviations local to the domain.3) Two kinds of ellipsis are evident.In one kind the constituents left ouC are co be recovered from knowledge of the structured object; the ocher kind is the standard kind of textual ellipsis where the missing macerisl is recovered from previous sentences.uses of adjectival and prepositional qualifiers can be distinguished. There is a referenclal use as in "in Left eye" in (14) and also an attributive use as in "of elevated pressure"in (14)There is a history of elevated pressure in the left eye.An adjective can only have a referential use if iC has previously been used attrlbucively or if it refers to a focussing attribute.several assertions tend to tak~a one of two forms. In one of these cha focus is on an eye and several measurements are given for that eye as in (15).In the left eye chars is a pressure of 27, .5 cupping and an ercuaCe ecotome.(:5)In the other form the focus is on an attribute and values for both eyes are given as in (16).the pressure is I0 od and 20 os.A good deal of extra syntactic complexity is introduced by the fact chat there are 2 eyes (a particular ex-,.pla of the general phenomenon of multiple idanclcal sub-parts). The problm-is chac (ha qualifying phrases "in the left / rlghc/boch eyes" appear in many different places in the sentences and conslderabla work must be done to find the correct scope. 1.3.4.Pigure 3 Some (edited) output from a run of a case | null | Main paper:
semantics:
A fundamental assumption that has been made and one that is Justlfled by examination of several sets of cases is that the sentences dascrlbe an instance of a patient with the assumption that the reader already knows the concept. None of the sentences in the notes examined had an interpretation which would requlre updating the concept GLAUCCMA-PATIENT.The interpretation of a case is thus consldared to be the construction of the the corresponding instance of GLAUCOMA-PATIENT.The nature of structured objects as outlined above dlccataa that only two fundamental kinds of assertions are expected in sentences. There wlll either be an assertion about the existence of an optional component as in (5) or about the value of an attribute as in (6) and 7• There Is an arcuete scotoma od.** The pressure is 20 in the left eye. The pressure is normal os.(5) (6)Vary few of the sentences contain Just one assertion, most contain several as in (8) and (9) .There is a nasal step and an arcuete scotoma in the left eye and a central island in the right eye (8) ~he medication is I0 percent pilocarplne daily in both eyes.Even though sentences are viewed as containing assertions their meanings can be represented as sets of instances, Non-nmnerlcal measurements differ from numerical given that there is a procedure which takes these measurements in that RANGE, UNIT and QVALSET are replaced instances and incorporates them into the growing instance by VALSET. One or more members of VALSET are to be of GLAUCOMA-PATIENT. Ibis is due to the tree structure selected in creating an instance of the measurement of instances since Instantlatlon of a concept involves concept, for example: Instantlatlon of all concepts between itself and the root.In fact, many sentences in the cases do not even CI-PATIENT-SEX-MSMT VALSET (ONEOF K-MALE K-FEMALE) contain a relation but merely assert the existence of an instance or of an attribute value as in (I0) and ([1).An instance of a structured object is represented as a tree.Instances are created piece-meal as the Information trickles in from the case.In some cases the Nasal step od. (I0) a I0 year old white male.(II) ** Opthalmologlsts frequently use the abbreviations "ed" for "in the right eye", "os" for "in the left eye" and "ou" for "in hor/1 ayes" 2.2 PROVISIONAL INSTANCES Any particular noun or adjective could refer to a number of different concepts. "Medication" for" example could refer to CI-PATIENT-MEDICATION, CI-PATIENT-&IGHT-EYE-MEDICATION or (I-PATIENT-LEFT-EYE-MEDICATION. Moreover in any particular use it could be referring Co one or more of its possible referents.In t2Medicacion consists of diamox and pllocarpine drops in both eyes."medication" refers co all of its possible referents since diamox is not given to the eye but is taken orally. In addition to this, ic £s generally not possible to know at the clme of encountering a word whether it refers to an existing Instance or to a new instance. This is due to the fact thaC at the time of encountering a reference to a concept all of the values of the instance dimensions mlghc not be known. The mechanism for dealing with these problems is Co assign "provisional Instances" as the referents of words end phrases when they are scanned during the parse and to turn these provisional instances Into "real" instances when the correct parse has been found. This involves finding the values of the instance dimensions from rest of the sentence, from knowledge of defaults or perhaps from values in previous sentences. The most common Instance dimension is TIME and its value is readily obtained from the tense of the verb or from a clme phrase.If the instance dimensions indicate an existing instance then the partial provisional instance from the sentence is incorporated into the existing real instance, otherwise a new instance is created.Several mappings can be made from the representation of structured objects to syntactic classes.For example, all nodes will be referred to by nouns and noun phrases, links will be referred to by prepositions and verbs and members of a VALSET or a 0VALSET will ba referred to by adjectives.The links between concepts and cha ~rds that can be used to refer to them are made at system build time when che structured object is constructed. Some words such as "both" and "very" refer to procedures whose actions are the same no matter what the structured object.The nature of structured objects and of the sentences in cases Indicate thac a "case'* [Bruce 1975 ] approach to semantic analysis is a "natural". A case syecsm ham in fact been implemented with such cases as ATTRIBUTE, OBJECT, VALUE, and UNIT. One case that is particularly useful is FOCUS.It is used to record references Co left eye or right eye for use in embedded or conjoined sentences such as (13).The pressure in the left eye is 27 and there is an arcuate scocoma.For the reasons discussed in section 2.2 ic is necessary co assign sacs of candidate referents to soma of the case values during the course of the parse. These sacs are pruned as higher levels of the parse tree are built.
syntax:
It is noc really possible to vlew cha sentences comprising a case as a subset of English since many of the elementary grammatical rules are broken (e.g. frequent omission of verbs). Rather the sentences are in a medical dialect and parr of the task of wrlClng an interpreter for cases involves an anthropological investlgaclon of the dialect and its definition in some formal way.An analysls of a nt~"ber of cases revealed the following characteristics (see also [Sangscer 1978] ): I) Frequent omission of verbs and punctuation.2) ~ch use of abbreviations local to the domain.3) Two kinds of ellipsis are evident.In one kind the constituents left ouC are co be recovered from knowledge of the structured object; the ocher kind is the standard kind of textual ellipsis where the missing macerisl is recovered from previous sentences.uses of adjectival and prepositional qualifiers can be distinguished. There is a referenclal use as in "in Left eye" in (14) and also an attributive use as in "of elevated pressure"in (14)There is a history of elevated pressure in the left eye.An adjective can only have a referential use if iC has previously been used attrlbucively or if it refers to a focussing attribute.several assertions tend to tak~a one of two forms. In one of these cha focus is on an eye and several measurements are given for that eye as in (15).In the left eye chars is a pressure of 27, .5 cupping and an ercuaCe ecotome.(:5)In the other form the focus is on an attribute and values for both eyes are given as in (16).the pressure is I0 od and 20 os.A good deal of extra syntactic complexity is introduced by the fact chat there are 2 eyes (a particular ex-,.pla of the general phenomenon of multiple idanclcal sub-parts). The problm-is chac (ha qualifying phrases "in the left / rlghc/boch eyes" appear in many different places in the sentences and conslderabla work must be done to find the correct scope. 1.3.4.Pigure 3 Some (edited) output from a run of a case
:
Abstract: A "Front End" for a Computer-Based Glaucoma Consultation System is described. The system views a case as a description of a particular instance of a class of concepts called "structured objects" and builds up a representation of the instance from the sentences in the case.The information required by the consultation system is then extracted and passed on to the consultation system in the appropriately coded form. A core of syntactlc, semantic end contextual rules which are applicable to all structured objects is being developed together with a representation of the structured object GLAUCOMA-PATIENT.There is also a facility for adding domain dependent syntax, abbreviations and defaults. system that has a core of syntax and semantics that is applicable to all structured objects and which can be extended by domain specific syntax, idioms and defaults.work on the interpretation of hospital discharge summaries, which are very similar to case descriptions, has been done by a group at NYU [Sager 1978] . Their work has focused on the creation of formatted data bases for subsequent question answering and is syntax based. The research reported here is concerned with extracting from the case the information understandable by a consultation system and is primarily knowledge based.During the past decade a number of Medical Consultation systems have been developed, for example INTERNIST [Pople. Myers and Miller 1973] ,CASNET/GLAUCOMA [Weiss st. al. 1978] , MYCIN [Shortliffe 1976 ]. Currently still others are being developed. Some of these programs are reaching a stage where they are being used in hospitals and clinics. Such use brings with it the need for fast and natural communication with these programs for the reporting of the "clinical state" of the patient. This includes laboratory findings, symptoms, medications and certain history data. Ideally the reporting would be done by speech but this is currently beyond the state of the art in speech understanding. A more reasonable goal is to try to capture the physicians" written "Natural Language" for describing patients and to write programs to convert these descriptions to the appropriate coded input to the consultation systems.The original motivation for this research came from the desire to have natural language input of cases to CASNET/GLAUCOMA a computer-based glaucoma consultation system developed at Retgers University.A case is several paragraphs of sentences , written by a physician, which describe a patient who has glaucoma or who is suspected of having glaucoma.It was desired to have a "Natural Language Front-End" which could interpret the cases and pass the content to the consultation system. In the beginning stages it was by no means clear that it would even be possible to have a "front end" since it was expected that some sophisticated knowledge of Glaucoma would be necessary and that feedback from the consultation system would be required in understanding the input sentences. However during the course of the investigation it became clear that certain generalizations could be made from the domain of Glaucoma.The key discovery was that under some reasonable assumptions the physic iane notes could be viewed as descriptions of instances of a class of concepts called structured oblects and the knowledge needed to interpret the notes was mostly knowledge of the relationship between language and structured objects rather than knowledge of Glaucoma. The graph has a distinguished node, analogous to the root of a tree, whose label is the name of the concept. All incoming errs to the concept enter only at this distinguished or "head" node. Although the relation between PATIENT and PATIENT-MEDICATION has some surface forms that make it look like an ATTR relation this is not really the case. A "true" structured object would not have ASS links but they must be introduced to deal with GLAUCOMA-PATIENT. the formal semantics of the ASS relation are very similar to those of the ATTR and PART relations. PART SI~C The nunbers after the C prefix in Fisure l donate levels of "sub-conceptln8". Level I £s the lowest level, those concepts do not have any sub-concepts only £natancao. Note that CI-PATIENT-KIGHT-EYE is a sub-concept of C2-PATIENT-gYE, not an Instanceo CI-PATIENT-LEFT-gYE and C2-PATTENT-~IGHT-EYE are two different concepts t that is they have d/~Joint sub-structure; they are as different to the system as C-AiM and C-LEG. There is 8nod reason for this. It is possible that a different Instrument will be needed to measure the value of an attribute in the right eye than in the taft aye. Thls means that the measurement concepts got these attrlbutee will have to he different for the left and right eyes.C I-PAT-LE • C2-PAT-EYE j q S~E ! C I-PAT-LE PRESSURE M. ~c~-PAT-~YE [ C I-PAT-LE , PRESSURE-MSMT nESSURE-"S~'T, I SUBC C l-PAT-RE J ATI"R C I-PAT-P.E PRESSURE C I-PAT-~E- PRESS~E-MSMT ~C~-~AT-I PART ....~S- J MEDICATION j C I-PATIENT ATTR C I-PAT-NED- DL~MOX i c x-~ATIENT-i MET .~ c X-~AT~NT-i ATT~ c,-,ATI,.NT- ,Ic -pAT ' NT: i SEX JH (@1 SEX.-~T l /i -T dAnother example from the d~ain of slancoma show this more vividly. CI-PATIENT-LEYT-~YE-VISUAL-FIELD-~COTCMA denotes a scotoma in the left eye.A particular type of scotoma is the arcuate (bow-shaped) scotoma. This must be a separate concept since it is meaninsful to suty "double arcuste scotoma" but not "doubte scotoma", This means that the concept C .... -FIELD-AACUATE-SCOTflMA has an attribute ~hat cannot be inherited from C..,-~IELD-SCOTOMA. If a measurement concept is the alune for hor~ eyes ( The rules of instantlatlon are embedded in the core.A partial instantiation of CI-PATIENT can be done before the first sentence is processed by tracing links marked NECESSARY.Any component or attribute ins,an,laced at this stage will be introduced by a definite noun phrase while optional components will be introduced by indefinite noun phrases.
Appendix:
| null | null | null | null | {
"paperhash": [
"brachman|a_structural_paradigm_for_representing_knowledge.",
"bobrow|an_overview_of_krl,_a_knowledge_representation_language",
"pople|dialog:_a_model_of_diagnostic_logic_for_internal_medicine"
],
"title": [
"A Structural Paradigm for Representing Knowledge.",
"An overview of KRL, a Knowledge Representation Language",
"DIALOG: A Model Of Diagnostic Logic For Internal Medicine"
],
"abstract": [
"Abstract : This report presents on associative network formalism for representing conceptual knowledge. While many similar formalisms have been developed since the introduction of the semantic network in 1966, they have often suffered from inconsistent interpretation of their links, lack of appropriate structure in their nodes, and general expressive inadequacy. In this paper, we take a detailed look at the history of these semantic nets and begin to understand their inadequacies by examining closely what their representational pieces have been intended to model. Based on this analysis, a new type of network is presented - the Structured Inheritance Network (SI-NET) - designed to circumvent common expressive shortcomings.",
"This paper describes KRL, a Knowledge Representation Language designed for use in understander systems. It outlines both the general concepts which underlie our research and the details of KRL-0, an experimental implementation of some of these concepts. KRL is an attempt to integrate procedural knowledge with a broad base of declarative forms. These forms provide a variety of ways to express the logical structure of the knowledge, in order to give flexibility in associating procedures (for memory and reasoning) with specific pieces of knowledge, and to control the relative accessibility of different facts and descriptions. The formalism for declarative knowledge is based on structured conceptual objects with associated descriptions. These objects form a network of memory units with several different sorts of linkages, each having well-specified implications for the retrieval process. Procedures can be associated directly with the internal structure of a conceptual object. This procedural attachment allows the steps for a particular operation to be determined by characteristics of the specific entities involved. The control structure of KRL is based on the belief that the next generation of intelligent programs will integrate data-directed and goal-directed processing by using multi-processing. It provides for a priority-ordered multi-process agenda with explicit (user-provided) strategies for scheduling and resource allocation. It provides procedure directories which operate along with process frameworks to allow procedural parameterization of the fundamental system processes for building, comparing, and retrieving memory structures. Future development of KRL will include integrating procedure definition with the descriptive formalism.",
"A system for computer assisted medical diagnosis has been developed, which incorporates an innovative model of diagnostic logic. A supporting medical data base has also been assembled, now comprising approximately fifty percent of the major diseases of internal medicine. Using weighted associations between disease entitles and their manifestations, and employing a powerful attention focusing heuristic, the system has demonstrated competence In dealing with difficult clinical problems involving multiple diagnoses."
],
"authors": [
{
"name": [
"R. Brachman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. Bobrow",
"T. Winograd"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"H. E. Pople",
"J. Myers",
"R. Miller"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null
],
"s2_corpus_id": [
"58814991",
"7965074",
"17723951"
],
"intents": [
[],
[],
[]
],
"isInfluential": [
false,
false,
false
]
} | null | 548 | 0.001825 | null | null | null | null | null | null | null | null |
aa6f8945d0b781b5466c2be7f08a0929c1effac3 | 41559124 | null | A Snapshot of {KDS}: A Knowledge Delivery System | KDS Is a computer program which creates multl-par~raph, Natural Language text from a computer representation of knowledge to be delivered. We have addressed a number of Issues not previously encountered In the generation of Natural Language st the multi-sentence level, vlz: ordering among sentences and the scope of each, quality comparisons between alternative 8~regations of sub-sententJal units, the coordination of communication with non-linguistic activities by • gcel-pursuin~ planner, end the use of dynamic models of speaker and hearer to shape the text to the task at hand. STATEMENT OF THE PROBLEM The task of KDS is to generate English text under the following constraints: 1. The source of information Is a semantic net, having no a priori structuring to facilitate the outputtlng task. | {
"name": [
"Moore, James A. and",
"Mann, William C."
],
"affiliation": [
null,
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 6 | 7 | null | null | null | Is a computer program which creates multl-par~raph, Natural Language text from a computer representation of knowledge to be delivered. We have addressed a number of Issues not previously encountered In the generation of Natural Language st the multi-sentence level, vlz: ordering among sentences and the scope of each, quality comparisons between alternative 8~regations of sub-sententJal units, the coordination of communication with non-linguistic activities by • gcel-pursuin~ planner, end the use of dynamic models of speaker and hearer to shape the text to the task at hand.The task of KDS is to generate English text under the following constraints:1. The source of information Is a semantic net, having no a priori structuring to facilitate the outputtlng task.This represents the most elaborate performance of KDS to date.The KDS organization reflects our novel paradigm: FRAGMENT-AND-COMPOSE. KDS decomposes the original network into fragments then orders and 8~regatas these according to the dictates of the text-producing task, not according to the needs for which the internal representation was originally conceived. KDS has shown the feasibility of this approach.The KDS organization Is a simple pipeline: FRAGMENT, PLAN, FILTER, HILL-CLIMB, and OUTPUT.FRAGMENT transforms the selected portion of the semantic net into an unordered set of propositions which correspond, roughly, to minimal sentences.2. The text is produced to satisfy an explicit goal held by the text generating system, which describes a desired cognitive state of the reader.3. To achieve the desired state of the reader requires more than a single sentence.This is not the forum for a extensive analysis of our results; for details, see Mann and Moore [ 1979] . However, to communicate the flavor of what ~ve have accomplished--from the motivating goal:and about two pages of formal propositions describing the "Fire-alarm scene', KDS generated the following: "When conveying a scene in which the hearer is to identify himself with one of the actors, express ell propositions involving that actor AFTER those which do not, and separate these two partitions by a paragraph break'.W AeneeorFILTER, deletes from the set, ell propositions currently represented as known by the hearer.coordinates two sub-activities: AGGREGATOR applies rules to combine two or three fragments into a single one. A typical aggregation rule is:"The two fragments 'x does A' and 'x does B' can be combin~! into a single fragment: 'x does A and B'". PREFERENCER evaluates each proposed new fragment, producing a numerical measure of its "goodness". A typical preference rule is:"When instructing the hearer, lncremm the accumulating measure by 10 for each occurrence of the symbol 'YOU'". HILL-CLIMB uses AGGREGATOR to generate new candidate sets of fregments, and PREFERENCER, to determine which new set presents the best one-step improvement over the current set.The objective function of HILL-CLIMB has been enlarged to also take into ecceunt the COST OF FOREGONE OPPORTUNITIES. This has drastically improved the initial performance, since the topology abounds wtth local maxima.KDS has used, at one time or another, on the order of 10 planning rules, 30 aggregation rules and 7 preference rules.The aggregation and preference rules are directly analogoua to the capabilities of linguistic eempotence and performance, respectively.OUTPUT lsa simple (two pages of LISP) text generator driven by a context free grammar. | null | null | Main paper:
kds:
Is a computer program which creates multl-par~raph, Natural Language text from a computer representation of knowledge to be delivered. We have addressed a number of Issues not previously encountered In the generation of Natural Language st the multi-sentence level, vlz: ordering among sentences and the scope of each, quality comparisons between alternative 8~regations of sub-sententJal units, the coordination of communication with non-linguistic activities by • gcel-pursuin~ planner, end the use of dynamic models of speaker and hearer to shape the text to the task at hand.The task of KDS is to generate English text under the following constraints:1. The source of information Is a semantic net, having no a priori structuring to facilitate the outputtlng task.This represents the most elaborate performance of KDS to date.The KDS organization reflects our novel paradigm: FRAGMENT-AND-COMPOSE. KDS decomposes the original network into fragments then orders and 8~regatas these according to the dictates of the text-producing task, not according to the needs for which the internal representation was originally conceived. KDS has shown the feasibility of this approach.The KDS organization Is a simple pipeline: FRAGMENT, PLAN, FILTER, HILL-CLIMB, and OUTPUT.FRAGMENT transforms the selected portion of the semantic net into an unordered set of propositions which correspond, roughly, to minimal sentences.2. The text is produced to satisfy an explicit goal held by the text generating system, which describes a desired cognitive state of the reader.3. To achieve the desired state of the reader requires more than a single sentence.This is not the forum for a extensive analysis of our results; for details, see Mann and Moore [ 1979] . However, to communicate the flavor of what ~ve have accomplished--from the motivating goal:and about two pages of formal propositions describing the "Fire-alarm scene', KDS generated the following: "When conveying a scene in which the hearer is to identify himself with one of the actors, express ell propositions involving that actor AFTER those which do not, and separate these two partitions by a paragraph break'.W AeneeorFILTER, deletes from the set, ell propositions currently represented as known by the hearer.coordinates two sub-activities: AGGREGATOR applies rules to combine two or three fragments into a single one. A typical aggregation rule is:"The two fragments 'x does A' and 'x does B' can be combin~! into a single fragment: 'x does A and B'". PREFERENCER evaluates each proposed new fragment, producing a numerical measure of its "goodness". A typical preference rule is:"When instructing the hearer, lncremm the accumulating measure by 10 for each occurrence of the symbol 'YOU'". HILL-CLIMB uses AGGREGATOR to generate new candidate sets of fregments, and PREFERENCER, to determine which new set presents the best one-step improvement over the current set.The objective function of HILL-CLIMB has been enlarged to also take into ecceunt the COST OF FOREGONE OPPORTUNITIES. This has drastically improved the initial performance, since the topology abounds wtth local maxima.KDS has used, at one time or another, on the order of 10 planning rules, 30 aggregation rules and 7 preference rules.The aggregation and preference rules are directly analogoua to the capabilities of linguistic eempotence and performance, respectively.OUTPUT lsa simple (two pages of LISP) text generator driven by a context free grammar.
Appendix:
| null | null | null | null | {
"paperhash": [
"levin|process_models_of_reference_in_context",
"levin|dialogue-games:_metacommunication_structures_for_natural_language_interaction",
"mann|a_comprehension_model_for_human_dialogue"
],
"title": [
"Process Models of Reference in Context",
"Dialogue-Games: Metacommunication Structures for Natural Language Interaction",
"A Comprehension Model for Human Dialogue"
],
"abstract": [
"Abstract : Reference is a central issue for language comprehension and generation. After reviewing existing process models for comprehending and generating referring expressions, we present a general framework for context and reference processing. The context for reference processing is represented as a Public Workspace. Reference processes access Public Workspace and modify its content, which is the set of concepts currently on the table as far as the current language interaction is concerned. Information from many different sources can be integrated in comprehending or generating referring expressions. Within this general framework, a new system for selectively generating referring phrases is developed. This system decides how much to express about a given concept in a given context.",
"Our studies of naturally occurring human dialogue have led to the recognition of a class of regularities which characterize important aspects of communication. People appear to interact according to established patterns which span several turns in a dialogue and which recur frequently. These patterns appear to be organized around the goals which the dialogue serves for each participant. Many things which are said later in a dialogue can only be interpreted as pursuit of these goals, established by earlier dialogue. These patterns have been represented by a set of knowledge structures called Dialogue-Games, capturing shared, conventional knowledge that people have about communication and how it can be used to achieve goals. A Dialogue-Game has Parameters, which represent those elements that vary across instances of a particular pattern—the particular dialogue participants and the content topic. The states of the world which must be in effect for a particular Dialogue-Game to be employed successfully are represented by Specifications of these Parameters. Finally, the expected sequence of intermediate states that occur during instances of a particular conventional pattern are represented by the Components of the corresponding Dialogue-Game. Representations for several Dialogue-Games are presented here, based on our analyses of different kinds of naturally occurring dialogue. A process model is discussed, showing Dialogue-Game identification, pursuit, and termination as part of the comprehension of dialogue utterances. This Dialogue-Game model captures some of the important functional aspects of language, especially indirect uses to achieve implicit communication.",
"The comprehension of dialogue is an important concern for those interested in natural language processing for several reasons: dialogue gives particularly good access to human communication phenomena, it is less contrived than authored text, and human dialogue provides useful analogies for improving man-machine communication. In naturally occurring dialogues, the goals of the participants play a key role in structuring their language interactions. People know how dialogue is used to achieve goals, and they use this knowledge to comprehend what they hear."
],
"authors": [
{
"name": [
"J. Levin",
"N. Goldman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"J. Levin",
"James A. Moore"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"W. Mann",
"James A. Moore",
"J. Levin"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null
],
"s2_corpus_id": [
"59988461",
"20069944",
"2424018"
],
"intents": [
[],
[],
[]
],
"isInfluential": [
false,
false,
false
]
} | null | 548 | 0.012774 | null | null | null | null | null | null | null | null |
66d933eba55216125bb2d4d5ae3ea83ff4e37d35 | 19653448 | null | The Structure and Process of Talking About Doing | People talk •bout what they do, often •t the same tame a• they are doing. This reporting has •n important function in coordinating aotlon between people working together on real eve~/day problems. Zt is also •n important acts'ca o£ data for social scientists sttu~ylng people's behavior. Xn this paper, we report on some •tudle• we are doing on report dialogues. We describe two kinds of phenomena we have identified, outline a preliminary process model that int•grat•• the report generation with the processes that are generating the actions being reported upon, and specify a systematic methodology For extracting relevant evidence bearing on these phenomena t~om text trenscrlpts of talk about doing to use in evaluating the model. ~OZW~W Reports of problm solving actions are often used a• evident• about the und•rlying cognitive processes involved in generating a problem solution, as "problem solving protocols" (Howell & Simon, 1972). However, these reports ere obviously a kind of language interaction in their own right, in which the subject i• reportlns on hls/hor own actions to the experimenter. We have analyzed problem solving protocols of people solving a puzzle called "Hlsslonaries and Cannibals" and have found that in their report•, people adopt • • point or view" with respect to the problan, through • con•latent use of spatial detxts, For example, when a subject lays: ".., X can't send another cannibal across with another alssioflary or he will he outnumbered when he gets to the other side .., " | {
"name": [
"Levin, James A. and",
"Hutchins, Edwin L."
],
"affiliation": [
null,
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 10 | 1 | null | People talk •bout what they do, often •t the same tame a• they are doing. This reporting has •n important function in coordinating aotlon between people working together on real eve~/day problems. Zt is also •n important acts'ca o£ data for social scientists sttu~ylng people's behavior.Xn this paper, we report on some •tudle• we are doing on report dialogues.We describe two kinds of phenomena we have identified, outline a preliminary process model that int•grat•• the report generation with the processes that are generating the actions being reported upon, and specify a systematic methodology For extracting relevant evidence bearing on these phenomena t~om text trenscrlpts of talk about doing to use in evaluating the model.Reports of problm solving actions are often used a• evident• about the und•rlying cognitive processes involved in generating a problem solution, as "problem solving protocols" (Howell & Simon, 1972) .However, these reports ere obviously a kind of language interaction in their own right, in which the subject i• reportlns on hls/hor own actions to the experimenter. We have analyzed problem solving protocols of people solving a puzzle called "Hlsslonaries and Cannibals" and have found that in their report•, people adopt • • point or view" with respect to the problan, through • con•latent use of spatial detxts, For example, when a subject lays: ".., X can't send another cannibal across with another alssioflary or he will he outnumbered when he gets to the other side .., " the deixis In her report places her as speaker off the • from" side of the considered action, This is indicated both by the choice of the verb "send" and by the description of "the other side". ?he same suhjeot indicated the "to n slde as her point or view in another part or her protocol: ",..'cause you've gotta have one person to hri~ back the boat..." 4 Here, both the verb "bring" and the adverb "back" indicate "point ot view".Although people almost always unmmbiguoualy specify • "point of view" within the problem they are solving in their reports, they also deny awareness of takAn| such • point cF view, However, this point or view is important to the underlying problem solving procesmas. The strongest evidence for this comes ~om the hi|h correlation ~etween Point or view and errors in problem solving actions. Subjects in the ~tlsslonarles and Cannibals task can make errors by Cabin| actions that violate the constraints or the task. Host of these errors occur on the side away frm their current "point st view", even theuah their point of view changes From one physical side to the other during the course or solving the punle, mre interesting is that most of the "undetected" errors emcur on the side •way from their point or view. Some errors arm spontaneously detected by the subject ~mmediately otter askant the action that leads to • violation! others ere "undetected".After the experimenter interrupts topolnt out these undetected errors, the subjects often switch point of view so that the violation condition is now on the same side a• the subjects' point of view.We see the point of view indicated by •patlal delxls In the report of problem solving as reflec~ir~ an underlying allocation of effort (or attention). Pew errors occur with problem elements that •re given processing effort, while constraints that •re given little attention are more often violated. %n this way, these reports are reflecting changes Zn the organization of the problem element• that occur over the course of reaching a solution. We have also identified other ways in which report• embody the use of different conceptual organizations of the problem, including org•nlzatlons that vary from abstract to concrete and from perception oriented to action oriented.There •re multi-utterance structures that occur regularly in problem solving talk that we call • justltAoatlon argument structures." These structures have the form of:(did ) (do ) (since ) (could)+(not do)+(aotion).(bsc•use)->(Justitloatlon (will) argument)(Alternatively, these two segments san he reversed in order, by using connectiv•s like "theretora" or "so".) For example, these kinds of dialogue units occur in many ot the protocols studied by Newall & Simon (1972):"hen letter has one and only one numerical value ,e. Another argument rcrm Is one we call "prapatio ark,sent". (We have borrowed many or our naaes for arlmemt etruoturoe Prom a rhotor%o boo~ (Perelmsn & O%hreohta-Tyteea, I~5g).) Altho~h 5hie book Ln I "noru51vo" aooount oF erlmentatLon, we PLnd £t valuable ae a ~Ado to our atSempt ~o l~VO i deaerlptlve aooount or naturaAly eeeurr~ng %nFomal "or|lentat¢en" eoe~rr%n| An our eub~eoto' reports oF theAr problem eolv~ng,) ?he prqitAe er|~mente %at MAng lOt%on A would Lead to relult R (Imonll ocher ~hlnSe). ROltA~t fl Le undoeLrlble. Therefore don't do aotlon A.enemple Prom our pretooole lot 0,,, Hoth ~LeeAonar£ea are IO£ng 5o have to eema boom beoauee. 'oauea %T 5hey don't eemo booM, veil, one ~e~d pt left and eaten. So beth mAeeloaar~oe oeme book.... "One XntoreatAng ~Ant about 5h~a ~rt~e~ar example %8 5hat ~t %e embedded wlthAn an "el~mlnatAon of alternat£voo" arll~Nnt etruoture. The5 %a, 5hAm "prqitAo arluaent" 18 used 50 el~aAnaSe one oF 5he alternatives, leav£ng only one 50 5eke.A third kAad oF arguaen5 atruoSure we have ideaS%Fled 18 railed "ende-moane"t %P erase S oooure, then there ~e an aotAen A to set 5o seal O. ?herefore eesamLAah orate S am a eubleal. for exemplar "... 3e Lr ever % oould |e5 ~hoee ever 5here, % Obviously, 5ham ar|wJent Fern %o similar 50 5he olaJe%o "means-ends oflalye£o" proposed ae ~rt of many serpent 5hear%re oF problem aolv%ng. The arlmmon5 Peru we hive identified bOOer v~en oIPCIAn k~ndo ot underlying oolnit%ve prooeeoing Is IoLng on, end thAI ~,~nd oF protooo% 5ext h~e been Ulld ll evLdonoe for 5his ~ndorXying prooeaming. Some people have lllUmOd t~H|5 5hal ~nd of languap Anteraotion oorreepondo to a euboea of 5he underly£n8 prooeseee (Nevoll i SAmon, ~973). Other people have questioned whether there %e any oorreapondenoe between vha5 people do Imd what 5hey say (NLsbett & ~llmon, 1977) . Our position le 5hat 5here %e a Fairly rieh ~nSereoSien between motion and report o~ aot~cn, mioh we will doeorihe %n our report OF our prel~m~ary proaese node/, of doing and rlportinl.(This poeitinn %8 oin£1or 5o one outlined reeently hy Rrloeaon ~ SAmoa (1979).)A ~a~csaa ~ O~ nn~aq AH~ ~L~ He have been oonatruotlng a proeeee model oF problem solving ~thin an aot~vatlon preeemo ~unevork (Seven, 1976; 1970) . ~15hAn 5hie FremevorK, nultAple proneness are 8%nultaneoue~y aot/.ve, end 5he 5he %nteraoC~ona between 5he aatlve prooeasea %o epeo~tAed by 5heir re~eeonCotiona %n a netvorM otruotur~ %one term memory. Emoh prooeoe %e so+lYe a oct+sAn aununS, with a oor~aAn smotmt oF nalIAenoln, and ~he more oaIAent a preoeaa As, 5he lar|er %5e %npae5 on oSher presences (and therefore on the overall prooeaoLng).There ere prooeleee tha~ ere oloeely relltld ~O the ~r~romnoe oF 5he problem tooK, lad o~here tt~c are oZoeely related to the report of the task aoClona. ~n the psr~lo,,~ar problem demean of the H~aeionsr£ee and CannLbae8 pusxle, 5he ~aek POliCed ao~Lonl and obJeo~e are defined as oonoopco An the long 5emnemory thaC beoome aoClve durlng the ?rob/.em solving. The oonsCrl£n58 of 5he problem ape represented Ln 5he name way, and leC aoSivoSed 5o varying delrwee during 5he problem so/.v~ng. ~-roro ooour when the oonacre/.n~e are %neutrlolenC%y 8aAAenC to prevent an notion wh£oh landl ~o a v~o/.oS/.on oF 5ha& oonecraAnC.Report related proeeaeoe impost 5he tao~ behavior by mod%Fy£n| 5he distribution o~ lelAenoe 5o 5he 5ask related proneness, "Point oP v~ew" of 5he problem lOlver hll L51 Ampeo~ on the presses%n| by add%n| ealAenea 5o 5hose ooSAvo eonoopte Jesse%sand ~th looatLon where the problem ~lver ham oonooptuaLly looated hAmthereelF, ~uet%P%eat%on arjUmlmt etruoturea l~l~lirly Ampao5 5he d%etr%butAon of emlAenoo by ~noreao%ng 5he sa%%enoe st or 5.see LnFerenee prooealol defined to be 8llJOO£lted wAth the arlumen5 structures.~n 5h~e ~sy, ~aKua|e san lad 5he problem solving, by addle| 50 5he roeouraoe of 5he 5nAked soon5 proooeaeo, %t ann sees h~nder %t %t looks the problem solver into a psrtAoular orlenLutLon oF the problem 5hit %On't f~U~tFul, rap example, to 5he extent 5h~t llmlus|e use Foouaeea eaIAenoo sway From oonetrlAntl t~5 ire beL~J v%olated oaul~ng effete, end elpoOLlALy LF 5hAl ooourl to euoh In extent 5ha5 5hone IPrOrl Ire undeSeoSed, thin the FoOUSlL~ll eFFeot oF languilo elm be l bert%It 5o solving the problem. 80 tar, we have deoor%ba none phenomena ve Mve observed In our solleetion or problem solvlM reports, and also m prel~,,%nary proooea model st problem eo%v~nJ aat~on and report, How san we use sup data 5o evalu|te our model??here are Ray Levels oF evaluative tent:Leg that we could use. At one extreme, 5heor4eo sin be strongly evaluated by doriv:Ln| prodAot~ono Prom 5hem of' epoolF:Lo da5a, vhAoh Le 5hen eo%leo5ed. I~peo~ally when 5he prod:Late4 da5a are unexpeotedt th:Lo prov:Ldee a r~Joroua 5net OF s theory. At another extreme ~0 a "ouFtAo~enoy teacn (Howell & 81mona 19TO) , A model oF an orpn£mD porform.'Lnl name tael¢ pasha 5he euf'Fio%enoy 5on5 %F Lt aloe san perForu ohm name tank. Than %e the evaluatAon 5eat oemmon%y used today For strafe.sial AntelIA|enee models.A more r~l;orous 5eat %e 5n Cry to F~.5 a mode/. 50 • emma OF data. ?hAs ~e the evaluate.on 5eahn~quo moot often UJld today An evaluat~,ng ooln~tlve poyoholo|y 5hoor~ea. A Fourth Ceo~lqua %e to %denSity a set of "or£cioxl" phenomena In 5he data spinet MtAoh to evaluate a mode% OF the5 data, AI ~lluotratod ~.n the liJ~ below, th:Li £s a more powerful evaluation teoiutique 5hit e~nple euFt~o£enoy, but lees povorF~ 5hsn 5he other two 5eohnLquee. Vo Fee). 5hat It 5has point In 5he scats OF the opt, 5hie te 5he appropriate evoluat~on 500hnique to use 5o evalunto our presell modll ~n IAIh5 of our dltl. "point of view" for a problem solver at each point in the problem solving from a record of the problem solving report and a record of moves made. Then, we use this extracted trace to evaluate our model of the role of point of view in problem solving.We have reported here a three pronged approach to the study of problem solving action and report: I) the collected of data on problem solving and talk about problem solving, 2) development of a process model of these behaviors, and 3) use of coding techniques to extract traces of "critical phenomena" from the transcripts for evaluating the model. So far, we have focussed our efforts on two types of problem solving phenomena: the changes in the problem solver's organization of the problem ("point of view"), and systematic multl-utterance structures used to express the forms of inference used to solve the problem ("Justificatlon argument structures").Ericsson, K.A., & Simon, H.A. Thlnking-aloud protocols as data: Effects of verbalization. Pittsburgh, PA: Carnegle-Mellon University, C.I.P. Working Paper | null | null | null | null | Main paper:
:
People talk •bout what they do, often •t the same tame a• they are doing. This reporting has •n important function in coordinating aotlon between people working together on real eve~/day problems. Zt is also •n important acts'ca o£ data for social scientists sttu~ylng people's behavior.Xn this paper, we report on some •tudle• we are doing on report dialogues.We describe two kinds of phenomena we have identified, outline a preliminary process model that int•grat•• the report generation with the processes that are generating the actions being reported upon, and specify a systematic methodology For extracting relevant evidence bearing on these phenomena t~om text trenscrlpts of talk about doing to use in evaluating the model.Reports of problm solving actions are often used a• evident• about the und•rlying cognitive processes involved in generating a problem solution, as "problem solving protocols" (Howell & Simon, 1972) .However, these reports ere obviously a kind of language interaction in their own right, in which the subject i• reportlns on hls/hor own actions to the experimenter. We have analyzed problem solving protocols of people solving a puzzle called "Hlsslonaries and Cannibals" and have found that in their report•, people adopt • • point or view" with respect to the problan, through • con•latent use of spatial detxts, For example, when a subject lays: ".., X can't send another cannibal across with another alssioflary or he will he outnumbered when he gets to the other side .., " the deixis In her report places her as speaker off the • from" side of the considered action, This is indicated both by the choice of the verb "send" and by the description of "the other side". ?he same suhjeot indicated the "to n slde as her point or view in another part or her protocol: ",..'cause you've gotta have one person to hri~ back the boat..." 4 Here, both the verb "bring" and the adverb "back" indicate "point ot view".Although people almost always unmmbiguoualy specify • "point of view" within the problem they are solving in their reports, they also deny awareness of takAn| such • point cF view, However, this point or view is important to the underlying problem solving procesmas. The strongest evidence for this comes ~om the hi|h correlation ~etween Point or view and errors in problem solving actions. Subjects in the ~tlsslonarles and Cannibals task can make errors by Cabin| actions that violate the constraints or the task. Host of these errors occur on the side away frm their current "point st view", even theuah their point of view changes From one physical side to the other during the course or solving the punle, mre interesting is that most of the "undetected" errors emcur on the side •way from their point or view. Some errors arm spontaneously detected by the subject ~mmediately otter askant the action that leads to • violation! others ere "undetected".After the experimenter interrupts topolnt out these undetected errors, the subjects often switch point of view so that the violation condition is now on the same side a• the subjects' point of view.We see the point of view indicated by •patlal delxls In the report of problem solving as reflec~ir~ an underlying allocation of effort (or attention). Pew errors occur with problem elements that •re given processing effort, while constraints that •re given little attention are more often violated. %n this way, these reports are reflecting changes Zn the organization of the problem element• that occur over the course of reaching a solution. We have also identified other ways in which report• embody the use of different conceptual organizations of the problem, including org•nlzatlons that vary from abstract to concrete and from perception oriented to action oriented.There •re multi-utterance structures that occur regularly in problem solving talk that we call • justltAoatlon argument structures." These structures have the form of:(did ) (do ) (since ) (could)+(not do)+(aotion).(bsc•use)->(Justitloatlon (will) argument)(Alternatively, these two segments san he reversed in order, by using connectiv•s like "theretora" or "so".) For example, these kinds of dialogue units occur in many ot the protocols studied by Newall & Simon (1972):"hen letter has one and only one numerical value ,e. Another argument rcrm Is one we call "prapatio ark,sent". (We have borrowed many or our naaes for arlmemt etruoturoe Prom a rhotor%o boo~ (Perelmsn & O%hreohta-Tyteea, I~5g).) Altho~h 5hie book Ln I "noru51vo" aooount oF erlmentatLon, we PLnd £t valuable ae a ~Ado to our atSempt ~o l~VO i deaerlptlve aooount or naturaAly eeeurr~ng %nFomal "or|lentat¢en" eoe~rr%n| An our eub~eoto' reports oF theAr problem eolv~ng,) ?he prqitAe er|~mente %at MAng lOt%on A would Lead to relult R (Imonll ocher ~hlnSe). ROltA~t fl Le undoeLrlble. Therefore don't do aotlon A.enemple Prom our pretooole lot 0,,, Hoth ~LeeAonar£ea are IO£ng 5o have to eema boom beoauee. 'oauea %T 5hey don't eemo booM, veil, one ~e~d pt left and eaten. So beth mAeeloaar~oe oeme book.... "One XntoreatAng ~Ant about 5h~a ~rt~e~ar example %8 5hat ~t %e embedded wlthAn an "el~mlnatAon of alternat£voo" arll~Nnt etruoture. The5 %a, 5hAm "prqitAo arluaent" 18 used 50 el~aAnaSe one oF 5he alternatives, leav£ng only one 50 5eke.A third kAad oF arguaen5 atruoSure we have ideaS%Fled 18 railed "ende-moane"t %P erase S oooure, then there ~e an aotAen A to set 5o seal O. ?herefore eesamLAah orate S am a eubleal. for exemplar "... 3e Lr ever % oould |e5 ~hoee ever 5here, % Obviously, 5ham ar|wJent Fern %o similar 50 5he olaJe%o "means-ends oflalye£o" proposed ae ~rt of many serpent 5hear%re oF problem aolv%ng. The arlmmon5 Peru we hive identified bOOer v~en oIPCIAn k~ndo ot underlying oolnit%ve prooeeoing Is IoLng on, end thAI ~,~nd oF protooo% 5ext h~e been Ulld ll evLdonoe for 5his ~ndorXying prooeaming. Some people have lllUmOd t~H|5 5hal ~nd of languap Anteraotion oorreepondo to a euboea of 5he underly£n8 prooeseee (Nevoll i SAmon, ~973). Other people have questioned whether there %e any oorreapondenoe between vha5 people do Imd what 5hey say (NLsbett & ~llmon, 1977) . Our position le 5hat 5here %e a Fairly rieh ~nSereoSien between motion and report o~ aot~cn, mioh we will doeorihe %n our report OF our prel~m~ary proaese node/, of doing and rlportinl.(This poeitinn %8 oin£1or 5o one outlined reeently hy Rrloeaon ~ SAmoa (1979).)A ~a~csaa ~ O~ nn~aq AH~ ~L~ He have been oonatruotlng a proeeee model oF problem solving ~thin an aot~vatlon preeemo ~unevork (Seven, 1976; 1970) . ~15hAn 5hie FremevorK, nultAple proneness are 8%nultaneoue~y aot/.ve, end 5he 5he %nteraoC~ona between 5he aatlve prooeasea %o epeo~tAed by 5heir re~eeonCotiona %n a netvorM otruotur~ %one term memory. Emoh prooeoe %e so+lYe a oct+sAn aununS, with a oor~aAn smotmt oF nalIAenoln, and ~he more oaIAent a preoeaa As, 5he lar|er %5e %npae5 on oSher presences (and therefore on the overall prooeaoLng).There ere prooeleee tha~ ere oloeely relltld ~O the ~r~romnoe oF 5he problem tooK, lad o~here tt~c are oZoeely related to the report of the task aoClona. ~n the psr~lo,,~ar problem demean of the H~aeionsr£ee and CannLbae8 pusxle, 5he ~aek POliCed ao~Lonl and obJeo~e are defined as oonoopco An the long 5emnemory thaC beoome aoClve durlng the ?rob/.em solving. The oonsCrl£n58 of 5he problem ape represented Ln 5he name way, and leC aoSivoSed 5o varying delrwee during 5he problem so/.v~ng. ~-roro ooour when the oonacre/.n~e are %neutrlolenC%y 8aAAenC to prevent an notion wh£oh landl ~o a v~o/.oS/.on oF 5ha& oonecraAnC.Report related proeeaeoe impost 5he tao~ behavior by mod%Fy£n| 5he distribution o~ lelAenoe 5o 5he 5ask related proneness, "Point oP v~ew" of 5he problem lOlver hll L51 Ampeo~ on the presses%n| by add%n| ealAenea 5o 5hose ooSAvo eonoopte Jesse%sand ~th looatLon where the problem ~lver ham oonooptuaLly looated hAmthereelF, ~uet%P%eat%on arjUmlmt etruoturea l~l~lirly Ampao5 5he d%etr%butAon of emlAenoo by ~noreao%ng 5he sa%%enoe st or 5.see LnFerenee prooealol defined to be 8llJOO£lted wAth the arlumen5 structures.~n 5h~e ~sy, ~aKua|e san lad 5he problem solving, by addle| 50 5he roeouraoe of 5he 5nAked soon5 proooeaeo, %t ann sees h~nder %t %t looks the problem solver into a psrtAoular orlenLutLon oF the problem 5hit %On't f~U~tFul, rap example, to 5he extent 5h~t llmlus|e use Foouaeea eaIAenoo sway From oonetrlAntl t~5 ire beL~J v%olated oaul~ng effete, end elpoOLlALy LF 5hAl ooourl to euoh In extent 5ha5 5hone IPrOrl Ire undeSeoSed, thin the FoOUSlL~ll eFFeot oF languilo elm be l bert%It 5o solving the problem. 80 tar, we have deoor%ba none phenomena ve Mve observed In our solleetion or problem solvlM reports, and also m prel~,,%nary proooea model st problem eo%v~nJ aat~on and report, How san we use sup data 5o evalu|te our model??here are Ray Levels oF evaluative tent:Leg that we could use. At one extreme, 5heor4eo sin be strongly evaluated by doriv:Ln| prodAot~ono Prom 5hem of' epoolF:Lo da5a, vhAoh Le 5hen eo%leo5ed. I~peo~ally when 5he prod:Late4 da5a are unexpeotedt th:Lo prov:Ldee a r~Joroua 5net OF s theory. At another extreme ~0 a "ouFtAo~enoy teacn (Howell & 81mona 19TO) , A model oF an orpn£mD porform.'Lnl name tael¢ pasha 5he euf'Fio%enoy 5on5 %F Lt aloe san perForu ohm name tank. Than %e the evaluatAon 5eat oemmon%y used today For strafe.sial AntelIA|enee models.A more r~l;orous 5eat %e 5n Cry to F~.5 a mode/. 50 • emma OF data. ?hAs ~e the evaluate.on 5eahn~quo moot often UJld today An evaluat~,ng ooln~tlve poyoholo|y 5hoor~ea. A Fourth Ceo~lqua %e to %denSity a set of "or£cioxl" phenomena In 5he data spinet MtAoh to evaluate a mode% OF the5 data, AI ~lluotratod ~.n the liJ~ below, th:Li £s a more powerful evaluation teoiutique 5hit e~nple euFt~o£enoy, but lees povorF~ 5hsn 5he other two 5eohnLquee. Vo Fee). 5hat It 5has point In 5he scats OF the opt, 5hie te 5he appropriate evoluat~on 500hnique to use 5o evalunto our presell modll ~n IAIh5 of our dltl. "point of view" for a problem solver at each point in the problem solving from a record of the problem solving report and a record of moves made. Then, we use this extracted trace to evaluate our model of the role of point of view in problem solving.We have reported here a three pronged approach to the study of problem solving action and report: I) the collected of data on problem solving and talk about problem solving, 2) development of a process model of these behaviors, and 3) use of coding techniques to extract traces of "critical phenomena" from the transcripts for evaluating the model. So far, we have focussed our efforts on two types of problem solving phenomena: the changes in the problem solver's organization of the problem ("point of view"), and systematic multl-utterance structures used to express the forms of inference used to solve the problem ("Justificatlon argument structures").Ericsson, K.A., & Simon, H.A. Thlnking-aloud protocols as data: Effects of verbalization. Pittsburgh, PA: Carnegle-Mellon University, C.I.P. Working Paper
Appendix:
| null | null | null | null | {
"paperhash": [
"nisbett|telling_more_than_we_can_know:_verbal_reports_on_mental_processes."
],
"title": [
"Telling more than we can know: Verbal reports on mental processes."
],
"abstract": [
"Evidence is reviewed which suggests that there may be little or no direct introspective access to higher order cognitive processes. Subjects are sometimes (a) unaware of the existence of a stimulus that importantly influenced a response, (b) unaware of the existence of the response, and (c) unaware that the stimulus has affected the response. It is proposed that when people attempt to report on their cognitive processes, that is, on the processes mediating the effects of a stimulus on a response, they do not do so on the basis of any true introspection. Instead, their reports are based on a priori, implicit causal theories, or judgments about the extent to which a particular stimulus is a plausible cause of a given response. This suggests that though people may not be able to observe directly their cognitive processes, they will sometimes be able to report accurately about them. Accurate reports will occur when influential stimuli are salient and are plausible causes of the responses they produce, and will not occur when stimuli are not salient or are not plausible causes."
],
"authors": [
{
"name": [
"R. Nisbett",
"Timothy D. Wilson"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null
],
"s2_corpus_id": [
"7742203"
],
"intents": [
[]
],
"isInfluential": [
false
]
} | null | 548 | 0.001825 | null | null | null | null | null | null | null | null |
e1302ef7ca9275cd2c7bd12582a7c48f34105878 | 31505541 | null | Schank/Riesbeck vs. Norman/{R}umelhart: What{'}s the Difference? | This paper explores the fundamental differences between two sentence-parsers developed in the early 1970's: Riesbeck's parser for $chank's'conceptual dependency' theory (4, 5), and the 'LNR' parser for Norman and Rumelhart's 'active :~emantic network' theory (3). The Riesbeck parser and the I,NR parser share a common goalthat of trsnsforming an input sentence into a canonical form for later use by memory~inference~paraphrase | {
"name": [
"Eisenstadt, Marc"
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 7 | 0 | null | null | null | Riesbeck's parser i~ implemented as n production system, in which input text can either ssti~{y the condition side of any production rule within ~ packet of currently-active rules, or else interrupt processing by disabling the current packet of rules and enabling ('triggering') a new packet of rules. In operation, the main verb of each segment of text is located, and a pointer to its lexical decomposition (canonical form) is established in memory. The surrounding text, primerily noun phrases, is then systematically mapped onto vacant case frame slots within the memory representation of the decomposed verb. Case information is signposted by a verb-triggered packet of production rules which expects certain cldsses of entity (e.g. animate recipient) to be encountered in the text. Phrase boundaries are handled by keyword-triggered packets of rules which initiate and terminate the parsing of phrases.In contrast to this, the LNR parser is implemented as an augmented transition network, in which input text can either satisfy a current expectation or cause backtracking to a point at which an alternative expectation can be satisfied. In operation, input text is mapped onto a surface case frame, which is an n-ary predicate containing a pointer to the appropriate code responsible for decomposing the predicate into canonical form. Case information is signposted by property-list indicators stored in the lexical entry for verbs. These indicators act as signals or flags which are inspected by augmented tests on PUSH NP and PUSH PP arcs in order to decide whether such transitions are to be allowed. Phrase boundaries are handled by the standard ATN PUSH and POP mechanisms, with provision for backtracking if an initially-fulfilled expectation later turns out to have been incorrect.In order to determine which differences are due to notational conventions, I have implemented versions of both parsers in Kaplan's General Syntactic Processor (GSP) formalism (2)~ a simple but elegant generalization of ATNs. In GSP terms, Riesbeck's active packets of production rules are grammar states, and each rule is represented as a grammar arc. Rule-packet triggering is handled by storing in the lexicon the GSP code which transfers control to a new grammar state when an interrupt is called for. Each packet is in effect a sub-grammar of the type handled normally by an ATN PUSH and POP. The important difference is that the expensive actions normally associated with PUSH and POP (e.g. saving registers, building structures) only occur after it is safe to perform them. That is, bottom-up interrupts and very cheap 'lookahead' ensure that wasteful backtracking is largely avoided.Riesbeck's verb-triggered packet of rules (i.e. the entire sub-grammar which is entered after the verb is encountered) is isomorphic to the LNR-style use of lexical flags, which are in effect 'raised' and 'lowered' ~olely for the benefit of augmented tests on verb-independent ~rcs. Where Riesbeck depicts a 'satisfied expectation' by deleting the relevant production rule from the currently-active packet, LNR achieves the same effect by using augmented tests on PUSH NP and PUSII PP arcz to determine whether a particular case frame Slot has already been filled. Both approaches are handled with equal ease by GSP.In actual practice, Riesbeck's case frame expectations are typically tests for simple selectional restrictions, whereas LNR's case frame expectations are typically tests for the order in which noun phrases are encountered. Prepositions, naturally, are used by both parsers as important case frame clues: Riesbeck has a verbtriggered action alter the interrupt code associated with prepositions so that they 'behave' in precisely the right way; this is isomorphic to LNR's flags which are stored in the lexical entry for a verb and examined by augmented tests on verb-independent prepositional phrase arcs in the grammar.The behaviour of Riesbeck's verb-triggered packets (verb-dependent sub-grammars) is actually independent of when a pointer to the lexical decomposition of the verb is established (i.e. whether a pointer is added as soon as the verb is encountered or whether it is added after the end of the sentence has been reached). Thus, any claims about the possible advantages of 'early' or 'instantaneous' decomposition are moot. Since Riesbeck's cases are filled primarily on the basis of fairly simple selectional restrictions, there is no obvious reason why his parser couldn't have built some other kind of internal representation, based on any one of several linguistic theories of lexical decomposition. Although Riesbeck's decomposition could occur after the entire sentence has been parsed, LNR's decomposition must occur at this point, because it uses a networkmatching algorithm to find already-present structures in memory, and relies upon the arguments of the main n-ary predicate of the sentence being as fully specified as possible.Computationally, the major difference between the two parsers is that Riesbeck's parser uses interrupts to initiate 'safe' PUSHes and POPs to and from sub-gra,s,ars, whereas the L~R parser performs 'risky' PUSHes and POPs like any purely top-down parser. Riesbeck's mechanism is potentially very powerful, and the performance of the LNR parser can be improved by allowing this mechanism to be added automatically by the compiler which transforms an LNR augmented transition network into GSP ~chine code. Each parser can thus be mapped fairly clesJnly onto the other, with the only irreconcilable difference between them being the degree to which they rely on verb-dependent selectional restrictions to guide the process of filling in case frames. This characterization of the differences between them, based on implementing them within a common GSP framework, is somewhat surprising, since (a) the differences have nothing to do with 'conceptual dependency' or 'active septic networks' s~ud (b) the computational difference between them immediately suggests a way to auton~tically incorporate bottom-up processing into the LNR parser to improve not only its efficiency, but also its psychological plausibility. A GSP implementation of a 'hybrid' version of the two parsers is outlined in (I). | null | null | Main paper:
:
Riesbeck's parser i~ implemented as n production system, in which input text can either ssti~{y the condition side of any production rule within ~ packet of currently-active rules, or else interrupt processing by disabling the current packet of rules and enabling ('triggering') a new packet of rules. In operation, the main verb of each segment of text is located, and a pointer to its lexical decomposition (canonical form) is established in memory. The surrounding text, primerily noun phrases, is then systematically mapped onto vacant case frame slots within the memory representation of the decomposed verb. Case information is signposted by a verb-triggered packet of production rules which expects certain cldsses of entity (e.g. animate recipient) to be encountered in the text. Phrase boundaries are handled by keyword-triggered packets of rules which initiate and terminate the parsing of phrases.In contrast to this, the LNR parser is implemented as an augmented transition network, in which input text can either satisfy a current expectation or cause backtracking to a point at which an alternative expectation can be satisfied. In operation, input text is mapped onto a surface case frame, which is an n-ary predicate containing a pointer to the appropriate code responsible for decomposing the predicate into canonical form. Case information is signposted by property-list indicators stored in the lexical entry for verbs. These indicators act as signals or flags which are inspected by augmented tests on PUSH NP and PUSH PP arcs in order to decide whether such transitions are to be allowed. Phrase boundaries are handled by the standard ATN PUSH and POP mechanisms, with provision for backtracking if an initially-fulfilled expectation later turns out to have been incorrect.In order to determine which differences are due to notational conventions, I have implemented versions of both parsers in Kaplan's General Syntactic Processor (GSP) formalism (2)~ a simple but elegant generalization of ATNs. In GSP terms, Riesbeck's active packets of production rules are grammar states, and each rule is represented as a grammar arc. Rule-packet triggering is handled by storing in the lexicon the GSP code which transfers control to a new grammar state when an interrupt is called for. Each packet is in effect a sub-grammar of the type handled normally by an ATN PUSH and POP. The important difference is that the expensive actions normally associated with PUSH and POP (e.g. saving registers, building structures) only occur after it is safe to perform them. That is, bottom-up interrupts and very cheap 'lookahead' ensure that wasteful backtracking is largely avoided.Riesbeck's verb-triggered packet of rules (i.e. the entire sub-grammar which is entered after the verb is encountered) is isomorphic to the LNR-style use of lexical flags, which are in effect 'raised' and 'lowered' ~olely for the benefit of augmented tests on verb-independent ~rcs. Where Riesbeck depicts a 'satisfied expectation' by deleting the relevant production rule from the currently-active packet, LNR achieves the same effect by using augmented tests on PUSH NP and PUSII PP arcz to determine whether a particular case frame Slot has already been filled. Both approaches are handled with equal ease by GSP.In actual practice, Riesbeck's case frame expectations are typically tests for simple selectional restrictions, whereas LNR's case frame expectations are typically tests for the order in which noun phrases are encountered. Prepositions, naturally, are used by both parsers as important case frame clues: Riesbeck has a verbtriggered action alter the interrupt code associated with prepositions so that they 'behave' in precisely the right way; this is isomorphic to LNR's flags which are stored in the lexical entry for a verb and examined by augmented tests on verb-independent prepositional phrase arcs in the grammar.The behaviour of Riesbeck's verb-triggered packets (verb-dependent sub-grammars) is actually independent of when a pointer to the lexical decomposition of the verb is established (i.e. whether a pointer is added as soon as the verb is encountered or whether it is added after the end of the sentence has been reached). Thus, any claims about the possible advantages of 'early' or 'instantaneous' decomposition are moot. Since Riesbeck's cases are filled primarily on the basis of fairly simple selectional restrictions, there is no obvious reason why his parser couldn't have built some other kind of internal representation, based on any one of several linguistic theories of lexical decomposition. Although Riesbeck's decomposition could occur after the entire sentence has been parsed, LNR's decomposition must occur at this point, because it uses a networkmatching algorithm to find already-present structures in memory, and relies upon the arguments of the main n-ary predicate of the sentence being as fully specified as possible.Computationally, the major difference between the two parsers is that Riesbeck's parser uses interrupts to initiate 'safe' PUSHes and POPs to and from sub-gra,s,ars, whereas the L~R parser performs 'risky' PUSHes and POPs like any purely top-down parser. Riesbeck's mechanism is potentially very powerful, and the performance of the LNR parser can be improved by allowing this mechanism to be added automatically by the compiler which transforms an LNR augmented transition network into GSP ~chine code. Each parser can thus be mapped fairly clesJnly onto the other, with the only irreconcilable difference between them being the degree to which they rely on verb-dependent selectional restrictions to guide the process of filling in case frames. This characterization of the differences between them, based on implementing them within a common GSP framework, is somewhat surprising, since (a) the differences have nothing to do with 'conceptual dependency' or 'active septic networks' s~ud (b) the computational difference between them immediately suggests a way to auton~tically incorporate bottom-up processing into the LNR parser to improve not only its efficiency, but also its psychological plausibility. A GSP implementation of a 'hybrid' version of the two parsers is outlined in (I).
Appendix:
| null | null | null | null | {
"paperhash": [
"eisenstadt|alternative_parsers_for_conceptual_dependency:_getting_there_is_half_the_fun",
"riesbeck|computational_understanding_:_analysis_of_sentences_and_context"
],
"title": [
"Alternative Parsers for Conceptual Dependency: Getting There is Half the Fun",
"Computational understanding : analysis of sentences and context"
],
"abstract": [
"This paper takes a look behind the scenes at two conceptually-based parsers in order to shed light on the true differences between them. The first one is Riesbeck's parser for Schank's conceptual dependency; the second is the 'LNR' parser for Norman and Rume hart's active semantic networks. Both are described in terms of Kaplan's General Syntactic Processor formalism. This analysis shows that 'conceptual dependency' and 'active semantic networks' have little or nothing to do with the actual functioning of the parsers. Computationally, the two parsers differ only in terms of (a) effective use of interrupts and (b) reliance on selectional restrictions to guide parsing. A synthesis of the best features of both is suggested.",
"Abstract : The goal of this thesis was to develop a system for the computer analysis of written natural language texts that could also serve as a theroy of human comprehension of natural language. Therefore the construction of this system was guided by four basic assumptions about natural language comprehension. First, the primary goal of comprehension is always to find meanings as soon as possible, Other tasks, such as discovering the syntactic relationships, are performed only when essential to decisions about meaning. Second, an attempt is made to understand each word as soon as it is read, to decide what it means and how it relates to the rest of the text. Third, comprehension means not only understanding what has been seen but also predicting what is likely to be seen next. Fourth, the words of a text provide the cues for finding the information necessary for comprehending that text."
],
"authors": [
{
"name": [
"M. Eisenstadt"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"C. Riesbeck"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null
],
"s2_corpus_id": [
"29629704",
"60975873"
],
"intents": [
[],
[]
],
"isInfluential": [
false,
false
]
} | null | 548 | 0 | null | null | null | null | null | null | null | null |
a2b1ba8f624d81a46c901db6b8d9104d8a608af0 | 6738324 | null | Powerful ideas in computational linquistics - Implications for problem solving, and education | It is our firm belief that solving problems in the domain of computational linguistics (CL) can provide a set of metaphors or powerful ideas which are of great importance to many fields. We have taught several experimental | {
"name": [
"Fischer, Gerhard"
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 7 | 3 | null | It is our firm belief that solving problems in the domain of computational linguistics (CL) can provide a set of metaphors or powerful ideas which are of great importance to many fields. We have taught several experimental classes to students from high schools and universities and s major part of our work was centered around problems dealing with language. We have set up an experimental Language Laboratory in which the students can explore existing computer programs, modify them, design new ones and implement them. The goal was that the student should gain a deeper understanding of language itself and that he/she should learn general and transferable problem solving skills. exercise in pattern matching and symbol manipulation,where certain keywords trigger a few prestored answers. It may also serve as an example for how little machinery is necessary to create the illusion of understanding.[n our interdisciplinary research project (KLING eL el, 1977) we have tried to overcome these problems by providing opportunities for the student to explore powerful ideas in the context of non-trivial problems and by showing that the computer prescence can do much more for education than improve the delivery system for curricula established independently of it.Problem solving with the computer for the non-computer expert is slowly recognized as an important activity in our educational system. It is done best in a project-oriented course in which the student learns to solve problems in different domains.In the past, activities of this sort have been centered around numerical problems, physics problems and the standard computer science problems (eg like writing a sorting procedure).A matching capability can be a key element for many problem solving tasks involving the computer to make otherwise large, complicated efforts reachable. The following powerful ideas can be investigated in the context of this project: I) incremental design: we can start with s pattern marcher which is basicly en EQUAL predicate.The next steps could be: a membership predicate, s pattern with slots of fixed size, s pattern with slots of arbitrary size (which creates the need for back-up), the possibilty for simultanous assignment of matched elements to pattern variables, the restriction of matching by using predicates etc Z) the problem is ill-defineds the specification of the pattern marcher should be derived from the needs of using it to simplify problem solving tasks. A partial implementation can be an important help for a further specification or for a revision of already existing parts, ie the problem formulation is an important part of the problem solving process 3) definition of a new language layer: the pattern matcher can be used as a new language layer between the problem and the programming language and it can help to reduce the distance between the two. 4) glass-box approach: in many situations, we are primarily interested in using the pattern marcher.But by making use of an already existing program the student is not confined to a black box (like it would be in CAI environment); at any time he/she can look inside the program, open it up, change it to his/her own needs etc. A prerequisite for a program to be a glass-box is that it is implemented in a formalism the student is familiar with. 5) recursive control structure, a pattern marcher is a good example for the power of recursive definitions and control structures which can be used in many other situations A pattern marcher can be used in all projects where symbolic structures have to be dissected and identified, eg for the translation from infix to prefix, for parsing and translating processes, for morphological analysis, for simple I/0 routines (eg the identification of keywords), for ELIZA like programs and for symbolic manipulation of algebraic expressions. We have chosen this application specifically to support our claim that many problems considered to be mathematical can be more clearly understood by looking st them from • linguistic viewpoint (and the APL experience shows that changing the precedence rules for the evaluation of arithmetic expressions poses a non-trivial problem).Another application of the pattern marcher would be to parse sentences in a language where the grammar is given. For this purpose we assume that the pattern may contain predicates (which ere marked by "<" and ">"):The following grammer may serve ss an example (it describes the language of st least one "O" followed by at least one "1"):<$8~1~ --> <SO> <$1> <SB> --> 0 I 0<$1> <51> --> 11 1<51>SENT, SO and SI can be implemented with the pattern mstcher aa lollowed: compartments. ~By working on some of the proJects described above our students found that the knowledge which they acquired or discovered was not only useful in the context of a specific task but could be successfully used to understand end solve problems in other domalns as well, which should be illustrated through the following two specific examples: 1) the students became aware that the evaluation of arithmetic expressions (as it is commonly used in mathematics) is not something determined by God but that it is only s convention and that the laws behind it can be easily explained by the use of a grammar.Io2) s student discovered why mathematicians talk about one-to-one mappings (whlch never made any sense to him in mathematics) by trying to design secret codes in some of the language games (eg Pig Latin and other ones) Another important feature of our approach wee that the students extended the range of their "subjectively computable" problems, which helped them to replace their view of the computer being a giant adding machine with the more adequate view of being s general information processing device. We challenged their views thinking about the computer. Despite the fact that computation is still in its infancy there are many strong beliefs whet computers are, whet they can do and what they can not do. Working on the projects described above, the students can do work which is close to the research front (if they would have done their work ten years earlier they could have earned e PhO degree with it). This makes this subject material once again more interesting than much of mathematics where the students have to think about what is not even close to the current research front. | null | null | null | The relevance of problems from linguistics has been ignored. The reasons for this fact are easy to explain: I) the educational community in the language-oriented fields has very little knowledge about using a computer to write interesting programs to gain a deeper understanding of the problems in their domain 2) the computer experts were not familiar with linguisticsthe most commonly used programming lsnguegee end eyetemo ere inadequate to deal with the data structures and dialog requirements which are relevant for language processing 4) new fields like artificial intelligence, cognitive science end computational llnguletlee were not widely known The level of ignorance can best be seen by using ELIZA as an example: many people thought that it was a program which would "understand" the contents of a dialog. It was not evident to them that ELIZA represents nothing more than anIn recent years the view has emerged that the language of computation is the proper dialect The design of learning environments is an important goal for the educational theorist and the teacher. The computer as a new technology has created almost unlimited possibilities to create new and challenging environments. The Turtle world (PAPERT 1979) and the simulation world of Smslltslk (KAY 1977) provide good models of what can be done.In our project we have set up an experimental Language Laboratory in which the students can explore existing programs, modify them, dealgn new ones and implement them. We took great care in our design (by following the tradition of the LOGO projets as opposed to CAI approaches) that the students could work in an active mode end develop ideas in 8 personal way (not limited by the teachers approach). Our teaching style was not to provide answers but the learners were encouraged to use their own language knowledge to find e solution. Their work had to rely on self motivation which seems a more reasonable goal in CL where the products (eg poems, horoscopes, question/answering systems etc) can be more interesting and aesthetically pleasing then a set of numbers appearing as s result in numerical mathematics.Laboratory we wanted to create an environment in which the student's task is not to learn a set of formal rules (eg about the syntax of a programming language), but to give them s world in which they could develop sufficient inside into the way they used language to allow the transposition of this self-knowledge into programs.The students were exposed to different formalisms (primarily to LOGO, but also to LISP, ATNs, semantic networks, MICRO-PLANNER) and could explore the range of possible models which could be implemented in a cognitively efficient way with these formalisms. We tried to engage them in problems of moderate complexity (the students ware no researchern working full-time in a project) and we crested micro-verslons of programs by ommltlng feoturea which were not essential for a conceptual understanding.There is little doubt that we will be unable to solve the problems of coverage in our school and university subjects and of predicting whet specific knowledge our students will need in thirty or forty years. representation: eg to derive implicit knowledge and to study the impact of processing at read-time (antecendent theorems) versus question-time (consequent theorems) in a system which dealt with family relations (a system of this sort can be contrasted with ELIZA or a program to cast horoscopes); the following diagram illustrates how ? implicit relationships (---) can be derived from 3 explicit ones within a family of four persons: general computational ideas (eg like backtracking, which is encounterd in parsing non-deterministc grammars and which could be applied to pattern matching and tree like data structures)', ~/ ~,~ "\ |1 ", \,Most of the hypotheses and assertions of the previous sections ere supported by the empirical work in our project. We have not made an effort to do any kind of formal evaluation, but we have carried out a large number of informal investigations to understand the impact of our approach. Students filled out questionaires, participated in think-aloud protocols for many problem solving situations end we tried to understand their programs and the bugs they produced during the solution of a complex problem. There is no space here to talk about this in detail; the information is documented in KLING et al (1977 ) end FISCHER (1978 end 1979 .We believe that our approach turned out to be very successful.The students enjoyed working in our laboratory and they learned a lot about language as well as general problem solving snd programming skills. Especially students with little interest in mathematical problems were motivated by language-oriented applications. They could work in an active mode and investigate arbitrary formalisms and conjectures.They could see that ideas from linguistics could help them to understand problems in other domains, which supports our hypothesis that problems from CL can serve as an entry point and as a transient object to the world of problem solving, programming end mathematics. | Main paper:
the state of the art:
The relevance of problems from linguistics has been ignored. The reasons for this fact are easy to explain: I) the educational community in the language-oriented fields has very little knowledge about using a computer to write interesting programs to gain a deeper understanding of the problems in their domain 2) the computer experts were not familiar with linguisticsthe most commonly used programming lsnguegee end eyetemo ere inadequate to deal with the data structures and dialog requirements which are relevant for language processing 4) new fields like artificial intelligence, cognitive science end computational llnguletlee were not widely known The level of ignorance can best be seen by using ELIZA as an example: many people thought that it was a program which would "understand" the contents of a dialog. It was not evident to them that ELIZA represents nothing more than anIn recent years the view has emerged that the language of computation is the proper dialect
oesiqn of a lanquaqe laboratory:
The design of learning environments is an important goal for the educational theorist and the teacher. The computer as a new technology has created almost unlimited possibilities to create new and challenging environments. The Turtle world (PAPERT 1979) and the simulation world of Smslltslk (KAY 1977) provide good models of what can be done.In our project we have set up an experimental Language Laboratory in which the students can explore existing programs, modify them, dealgn new ones and implement them. We took great care in our design (by following the tradition of the LOGO projets as opposed to CAI approaches) that the students could work in an active mode end develop ideas in 8 personal way (not limited by the teachers approach). Our teaching style was not to provide answers but the learners were encouraged to use their own language knowledge to find e solution. Their work had to rely on self motivation which seems a more reasonable goal in CL where the products (eg poems, horoscopes, question/answering systems etc) can be more interesting and aesthetically pleasing then a set of numbers appearing as s result in numerical mathematics.Laboratory we wanted to create an environment in which the student's task is not to learn a set of formal rules (eg about the syntax of a programming language), but to give them s world in which they could develop sufficient inside into the way they used language to allow the transposition of this self-knowledge into programs.The students were exposed to different formalisms (primarily to LOGO, but also to LISP, ATNs, semantic networks, MICRO-PLANNER) and could explore the range of possible models which could be implemented in a cognitively efficient way with these formalisms. We tried to engage them in problems of moderate complexity (the students ware no researchern working full-time in a project) and we crested micro-verslons of programs by ommltlng feoturea which were not essential for a conceptual understanding.There is little doubt that we will be unable to solve the problems of coverage in our school and university subjects and of predicting whet specific knowledge our students will need in thirty or forty years. representation: eg to derive implicit knowledge and to study the impact of processing at read-time (antecendent theorems) versus question-time (consequent theorems) in a system which dealt with family relations (a system of this sort can be contrasted with ELIZA or a program to cast horoscopes); the following diagram illustrates how ? implicit relationships (---) can be derived from 3 explicit ones within a family of four persons: general computational ideas (eg like backtracking, which is encounterd in parsing non-deterministc grammars and which could be applied to pattern matching and tree like data structures)', ~/ ~,~ "\ |1 ", \,
pattern matchinqan example for the.deslqn 9nd implementation of s minirlsnquaqe:
A matching capability can be a key element for many problem solving tasks involving the computer to make otherwise large, complicated efforts reachable. The following powerful ideas can be investigated in the context of this project: I) incremental design: we can start with s pattern marcher which is basicly en EQUAL predicate.The next steps could be: a membership predicate, s pattern with slots of fixed size, s pattern with slots of arbitrary size (which creates the need for back-up), the possibilty for simultanous assignment of matched elements to pattern variables, the restriction of matching by using predicates etc Z) the problem is ill-defineds the specification of the pattern marcher should be derived from the needs of using it to simplify problem solving tasks. A partial implementation can be an important help for a further specification or for a revision of already existing parts, ie the problem formulation is an important part of the problem solving process 3) definition of a new language layer: the pattern matcher can be used as a new language layer between the problem and the programming language and it can help to reduce the distance between the two. 4) glass-box approach: in many situations, we are primarily interested in using the pattern marcher.But by making use of an already existing program the student is not confined to a black box (like it would be in CAI environment); at any time he/she can look inside the program, open it up, change it to his/her own needs etc. A prerequisite for a program to be a glass-box is that it is implemented in a formalism the student is familiar with. 5) recursive control structure, a pattern marcher is a good example for the power of recursive definitions and control structures which can be used in many other situations A pattern marcher can be used in all projects where symbolic structures have to be dissected and identified, eg for the translation from infix to prefix, for parsing and translating processes, for morphological analysis, for simple I/0 routines (eg the identification of keywords), for ELIZA like programs and for symbolic manipulation of algebraic expressions. We have chosen this application specifically to support our claim that many problems considered to be mathematical can be more clearly understood by looking st them from • linguistic viewpoint (and the APL experience shows that changing the precedence rules for the evaluation of arithmetic expressions poses a non-trivial problem).Another application of the pattern marcher would be to parse sentences in a language where the grammar is given. For this purpose we assume that the pattern may contain predicates (which ere marked by "<" and ">"):The following grammer may serve ss an example (it describes the language of st least one "O" followed by at least one "1"):<$8~1~ --> <SO> <$1> <SB> --> 0 I 0<$1> <51> --> 11 1<51>SENT, SO and SI can be implemented with the pattern mstcher aa lollowed: compartments. ~By working on some of the proJects described above our students found that the knowledge which they acquired or discovered was not only useful in the context of a specific task but could be successfully used to understand end solve problems in other domalns as well, which should be illustrated through the following two specific examples: 1) the students became aware that the evaluation of arithmetic expressions (as it is commonly used in mathematics) is not something determined by God but that it is only s convention and that the laws behind it can be easily explained by the use of a grammar.Io2) s student discovered why mathematicians talk about one-to-one mappings (whlch never made any sense to him in mathematics) by trying to design secret codes in some of the language games (eg Pig Latin and other ones) Another important feature of our approach wee that the students extended the range of their "subjectively computable" problems, which helped them to replace their view of the computer being a giant adding machine with the more adequate view of being s general information processing device. We challenged their views thinking about the computer. Despite the fact that computation is still in its infancy there are many strong beliefs whet computers are, whet they can do and what they can not do. Working on the projects described above, the students can do work which is close to the research front (if they would have done their work ten years earlier they could have earned e PhO degree with it). This makes this subject material once again more interesting than much of mathematics where the students have to think about what is not even close to the current research front.
[mpiricel findinqe:
Most of the hypotheses and assertions of the previous sections ere supported by the empirical work in our project. We have not made an effort to do any kind of formal evaluation, but we have carried out a large number of informal investigations to understand the impact of our approach. Students filled out questionaires, participated in think-aloud protocols for many problem solving situations end we tried to understand their programs and the bugs they produced during the solution of a complex problem. There is no space here to talk about this in detail; the information is documented in KLING et al (1977 ) end FISCHER (1978 end 1979 .We believe that our approach turned out to be very successful.The students enjoyed working in our laboratory and they learned a lot about language as well as general problem solving snd programming skills. Especially students with little interest in mathematical problems were motivated by language-oriented applications. They could work in an active mode and investigate arbitrary formalisms and conjectures.They could see that ideas from linguistics could help them to understand problems in other domains, which supports our hypothesis that problems from CL can serve as an entry point and as a transient object to the world of problem solving, programming end mathematics.
:
It is our firm belief that solving problems in the domain of computational linguistics (CL) can provide a set of metaphors or powerful ideas which are of great importance to many fields. We have taught several experimental classes to students from high schools and universities and s major part of our work was centered around problems dealing with language. We have set up an experimental Language Laboratory in which the students can explore existing computer programs, modify them, design new ones and implement them. The goal was that the student should gain a deeper understanding of language itself and that he/she should learn general and transferable problem solving skills. exercise in pattern matching and symbol manipulation,where certain keywords trigger a few prestored answers. It may also serve as an example for how little machinery is necessary to create the illusion of understanding.[n our interdisciplinary research project (KLING eL el, 1977) we have tried to overcome these problems by providing opportunities for the student to explore powerful ideas in the context of non-trivial problems and by showing that the computer prescence can do much more for education than improve the delivery system for curricula established independently of it.Problem solving with the computer for the non-computer expert is slowly recognized as an important activity in our educational system. It is done best in a project-oriented course in which the student learns to solve problems in different domains.In the past, activities of this sort have been centered around numerical problems, physics problems and the standard computer science problems (eg like writing a sorting procedure).
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 548 | 0.005474 | null | null | null | null | null | null | null | null |
2d9b284e331ccb32f314bdf2c27a8f56254223cb | 19192295 | null | Knowledge Organization and Application: Brief Comments on Papers in the Session | My brief comments on the papers in this session are based on the abstracts available to me and not on the complete papers. Hence, it is quite possible that some of the comments may turn out to be inappropriate or else they have already been taken care of in the full texts. In a couple of cases~ I had the benefit of reading some earlier longer related reports, which were very helpful. | {
"name": [
"Joshi, Aravind K."
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 0 | 0 | null | null | All the papers (except by Sangster) deal with either knowledge representation, particular types of knowledge to be represented, or how certain types of knowledge are to be used.Brackman describes a lattice-like structured inheritance network (KLONE) as a language for explicit representation of natural language conceptual information. Multiple descriptions can be represented. How does the facility differ from a similar one in KRL? Belief representations appear to be only implicit. Quantification is handled through a set of "structural descriptions." It is not clear how negation is handled. The main application is for the command and control of advanced graphics manioulators through natural language. Is there an implicit claim here that the KLONE representations are suitable for both natural language concepts as well as for those in the visual domain?Sowa also presents a network like representation (conceptual graphs). It is a representation that is apparently based on some ideas of Hintikka on incomplete but extensible models called surface models. Sowa also uses some ideas of graph grammars. It is not clear how multiple descriptions and beliefs can be represented in this framework. Perhaps the detailed paper will clarify some of these issues. This paper does not describe any application.Sangster's paper is not concerned, directly with knowledge representation. It is concerned with complete and partial matching procedures, especially for determining whether a particular instance satisfies the criteria for membership in a particular class. Matching procedures, especially partial matching procedures, are highly relevant to the use of any knowledge representation. Partial matching procedures have received considerable attention in the rule-based systems. This does not appear to be the case for other representations.Moore and Mann do not deal with knowledge representation per se, but rather with the generation of natural language texts from a given knowledge representation. They are more concerned with the problem of generating a text (which includes questions of ordering among sentences, their scopes, etc.) which satisfies a goal held by the system, describing a (cognitive) state of the reader. The need for resorting to multi-sentence structures arises from the fact that for achieving a desired state of the reader, a single sentence may not be adequate. ~cDonald's work on generation appears to be relevant, but it is not mentioned by the authors.Burnstein is primarily concerned with knowledge about (physical) objects and its role in the comprehension process. The interest here is the need for a particular type of knowledge rather than the representation scheme itself, which he takes to be that of Schank. Knowledge about objects, their normal uses, and the kinds of actions they are normally involved in is necessary for interjretation of sentences dealing with objects. In sentence (1) John opened the bottle and poured the wine, Burnstein's analysis indicates that the inference is driven largely by our knowledge about open bottles. In this instance, this need not be the case. We have the same situation in John took the bottle out of the refrioerator and poured the--w-Tne. The inference here is dependent on knowing something about wine bottles and their normal uses; knowledge of the fact that the bottle was open is not necessary.Given the normal reading of (1), (l') John opened the bottle and ~ured the wine out of it will be judged as re~u'n-~an--t~-, be-Te't'~o'n'~f--redundant material in (l') gives (1). Deletion of redundant and recoverable material is a device that language exploits. The recoverability here, however, is dependent on the knowledge about the objects and their normal uses.lf a non-normal reading of (1) is intended (e.g., the wine bein 0 poured into the bottle) then (l") John opened the bottle and poured the wine into it is not felt redundant. This suggests that a prediction that a normal reading is intended can be made (not, of course, with complete certainty) by recognizing that we are dealing with reduced forms. (Of course, context can always override such a prediction.) Some further questions are: Knowledge about objects is essential for comprehension. The paper does not discuss, however, how this knowledge and its particular representation helps in controlling the inferences in a uniform manner. Is there any relationship of this work to the common sense algorithms of Rieger?Lebowitz is also concerned with a particular type of knowledge rather than a representation scheme. Knowledge about the reader's purpose is essential for comprehension. The role played by the "interest" of the reader is also explored. The application is for the comprehension of newspaper stories. There is considerable work beyond the indicated references in the analysis of goal-directed discoursep but this has not been mentioned~ Finally, there are other issues which are important for knowledge representation but which have been either left out or only peripherally mentioned by some of the authors. Some of these are as follows. (i) A representation has to be adequate to support the desired inference. But this is not enough. It is also important to know how inferences are made (e.g., with what ease or difficulty). The interaction of the nature of a representation and the structure of the sentence or discourse will make certain inferences go through more easily than others.(ii) Knowledge has to be updated. Again the nature of the representation would make certain kinds of updates or modifications easy and others difficult.The previous issue also has a bearing on the relationship between knowledge representation and knowledge acquisition. At some level, these two aspects have to be viewed together. | null | null | null | Main paper:
:
All the papers (except by Sangster) deal with either knowledge representation, particular types of knowledge to be represented, or how certain types of knowledge are to be used.Brackman describes a lattice-like structured inheritance network (KLONE) as a language for explicit representation of natural language conceptual information. Multiple descriptions can be represented. How does the facility differ from a similar one in KRL? Belief representations appear to be only implicit. Quantification is handled through a set of "structural descriptions." It is not clear how negation is handled. The main application is for the command and control of advanced graphics manioulators through natural language. Is there an implicit claim here that the KLONE representations are suitable for both natural language concepts as well as for those in the visual domain?Sowa also presents a network like representation (conceptual graphs). It is a representation that is apparently based on some ideas of Hintikka on incomplete but extensible models called surface models. Sowa also uses some ideas of graph grammars. It is not clear how multiple descriptions and beliefs can be represented in this framework. Perhaps the detailed paper will clarify some of these issues. This paper does not describe any application.Sangster's paper is not concerned, directly with knowledge representation. It is concerned with complete and partial matching procedures, especially for determining whether a particular instance satisfies the criteria for membership in a particular class. Matching procedures, especially partial matching procedures, are highly relevant to the use of any knowledge representation. Partial matching procedures have received considerable attention in the rule-based systems. This does not appear to be the case for other representations.Moore and Mann do not deal with knowledge representation per se, but rather with the generation of natural language texts from a given knowledge representation. They are more concerned with the problem of generating a text (which includes questions of ordering among sentences, their scopes, etc.) which satisfies a goal held by the system, describing a (cognitive) state of the reader. The need for resorting to multi-sentence structures arises from the fact that for achieving a desired state of the reader, a single sentence may not be adequate. ~cDonald's work on generation appears to be relevant, but it is not mentioned by the authors.Burnstein is primarily concerned with knowledge about (physical) objects and its role in the comprehension process. The interest here is the need for a particular type of knowledge rather than the representation scheme itself, which he takes to be that of Schank. Knowledge about objects, their normal uses, and the kinds of actions they are normally involved in is necessary for interjretation of sentences dealing with objects. In sentence (1) John opened the bottle and poured the wine, Burnstein's analysis indicates that the inference is driven largely by our knowledge about open bottles. In this instance, this need not be the case. We have the same situation in John took the bottle out of the refrioerator and poured the--w-Tne. The inference here is dependent on knowing something about wine bottles and their normal uses; knowledge of the fact that the bottle was open is not necessary.Given the normal reading of (1), (l') John opened the bottle and ~ured the wine out of it will be judged as re~u'n-~an--t~-, be-Te't'~o'n'~f--redundant material in (l') gives (1). Deletion of redundant and recoverable material is a device that language exploits. The recoverability here, however, is dependent on the knowledge about the objects and their normal uses.lf a non-normal reading of (1) is intended (e.g., the wine bein 0 poured into the bottle) then (l") John opened the bottle and poured the wine into it is not felt redundant. This suggests that a prediction that a normal reading is intended can be made (not, of course, with complete certainty) by recognizing that we are dealing with reduced forms. (Of course, context can always override such a prediction.) Some further questions are: Knowledge about objects is essential for comprehension. The paper does not discuss, however, how this knowledge and its particular representation helps in controlling the inferences in a uniform manner. Is there any relationship of this work to the common sense algorithms of Rieger?Lebowitz is also concerned with a particular type of knowledge rather than a representation scheme. Knowledge about the reader's purpose is essential for comprehension. The role played by the "interest" of the reader is also explored. The application is for the comprehension of newspaper stories. There is considerable work beyond the indicated references in the analysis of goal-directed discoursep but this has not been mentioned~ Finally, there are other issues which are important for knowledge representation but which have been either left out or only peripherally mentioned by some of the authors. Some of these are as follows. (i) A representation has to be adequate to support the desired inference. But this is not enough. It is also important to know how inferences are made (e.g., with what ease or difficulty). The interaction of the nature of a representation and the structure of the sentence or discourse will make certain inferences go through more easily than others.(ii) Knowledge has to be updated. Again the nature of the representation would make certain kinds of updates or modifications easy and others difficult.The previous issue also has a bearing on the relationship between knowledge representation and knowledge acquisition. At some level, these two aspects have to be viewed together.
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 548 | 0 | null | null | null | null | null | null | null | null |
87809fd88e6dc4c8eeac40a7d8cc7e11d808b301 | 5317323 | null | Word Expert Parsing | This paper describes an approach to conceptual analysis and understanding of natural language in which linguistic knowledge centers on individual words, and the analysis mechanisms consist of interactions among distributed procedural experts representing that knowledge. Each word expert models the process of diagnosing the intended usage of a particular word in context. The Word Expert Parser performs conceptual analysis through the Interactlons of tl~e individual experts, which ask questions and exchange information in converging on a single mutually acceptable sentence meaning. The Word Expert theory is advanced as a better cognitive model of natural language understanding than the traditional rule-based approaches. The Word Expert Parser models parts o~ tSe theory, and the important issues of control and representation that arise in developing such a model [orm the basis of the technical discussion. An example from the prototype LISP implementation helps explain the theoretical results presented. | {
"name": [
"Small, Steven L."
],
"affiliation": [
null
]
} | null | null | 17th Annual Meeting of the Association for Computational Linguistics | 1979-06-01 | 19 | 42 | null | Computational understanding of natural language requires complex Interactions among a variety of distinct yet redundant mechanisms.The construction of a computer program to perform such a task begins with the development of an organizational framework which Inherently .incorporates certain assumptions about the nature ot these processes and the environment in which they take place. Such cognitive premises affect nro?oundly the scope and substance of computational ~nalysis for comprehension as found in the program.This paper describes a theory of conceptual parsing which considers knowledge about language to be distributed across a collection of procedural experts centered on individual words. Natural language parsing with word experts entails several new hypotheses about the organization and representation of linguistic and pragmatic knowledge for computational language comprenension.The Word Expert Parser [1] demonstrates hpw the word expert qTt~T~ed w£~h certain ocher choices oaseo on previous work, affect structure and process in a cognitive model of parsing.The Word Expert Parser is a cognitive model of conceptual language analysis in which the unit of ltngu~stic knowledge is the word and the fqcu~ o~ research ts the set or processes unoerlyinR comprehension.The model is aimed directly at problem~ of word sense ambiguity and idiomatic expressions, and in greatly generalizing the notion of wora sense, promotes these issues to a central place in the study of language parsing.Parsing models typically cope unsatisfactorily with the wide heterogeneity of usages of particular words.If a sentence contains a standard form of a word, it can usually be parsed; if it involves a less prevalent form which has a different part of speech, perhaps it too can be parsed. Disti.nguishing amen 8 the ~any senses of a common vero, adjective, or pronoun, tar example, or correctly translating idioms are rarely possible, At the source of this difficulty is the reliance on rule-based formalisms, whethar syntactic or semantic (e.g.. cases), which attempt to capture ~he linguistic contributions inherent in constituent chunks or sentences that consist of more than single words.A crucial assumption underlying work on the Word Expert Parser is that the ~undamental unit of linguistic Knowledge is the word. and that understanding its sense or role in a particular context is the central parsing process. In the parser to be described, the word expert constitutes the kernel of linguistic knowled~nd zts representation the e~emental data structure.IE is procedural in nature and executes directly as a process, cooperating with the other experts for a given sentence to arrive at a mutually acceptable sentence meaning.Certaln principles behind the parser d 9 nqt follow directly from the view or worn primacy, out ~rom other recent theories of parsing. The cognitive processes involved in language comprehension comprise the focus of linguistic study of the word expert approach. Parsin8 is viewea as an inferential process where linguistic knowledge of syntax and semantics and general pragmatic knowledge are applied in a uniform manner during IThe research described in this renor~ .is funded by the National Aeronautics and Space Admzn~stratton under grant , n umbe, r NSC-7255. Their support is gratefully acKnowleageG,This methodological position closely follows that of Rlosbeck (see [2] and [3 ]) and Schank [4] . The central concern with word usage and word sense ambiguity follows similar motivatlons of Wllks [5] . The control structure of the Word Expert Parser results from agreqment .with ~he hypothesis of .Harcus that parsing can he none aetermzntsttcally and ~n a way tn Dhlcn information ,gained through interpretation is permanent [6] . Rieger rne Importance at these mechanisms tar wore usage diagnosis derives from the ubiquity of local ambiguities, and brought about the notion chat ~hey be made the central processes of computational analysls an 9 understanding, Consideration of almost any Engllsn content word leads to a realization of the scope of the problem --with a little time and perhaps help from the dlctlonaFy , man~.dlstinct usages can ee.id~ntifl~d.As.a stmpie lllustrarzon, several usages earn tar the worus "heavy" and "ice" appear in Figure I . Each of. these seemingly" benign words exhibits a rich depth of contextual use, An earlier paper contains.a list at almost sixty verbal usages for the word "take" [llJ.The representation of all contextual word usages in an active way t~at insures their utility for linguistic dlagnasis led to the notion of word experts.Each word expert is a procedural entit~~f all posslblq contextual interpretations of the -word it represents. = Whe~ placed in a context formed by.expqrts for thg.othe ~ wares In a sentence, earn expert ShOUld De capaole or sufficient context-problng and self-examination to determine successfully' its functional or semantic role, and further, to realize the nature of that function or the precise meaning of the word. The representation and control issues involved in basing a parser on word experts are discussed below, following presentation of an example execution of the existing Word Expert Parser.The organization of the parser centers around data repositories on two levels --the sentence level workspace contains a word bin for each word (and sub-lexical morpheme) of the input and the concept level workspace contains a concept bin (described above) for each concept referred to in the input sentence. A third level of processing, the schema level workspaee, while not yet implemented, will contain a schema for each conceptual action of the input sentence.All actions affecting the contents of these data bins are carried out by the word expert processes, one of which is associated with each word bin in the wo rkspace.In addition to this first order information about lexical and conceptual objects, the parser contains a central tableau of control state descriptions available to any expert that can make use of self referential knowledge about its own processing or the states of processing of other model components.The availability of such control state information improves considerably both the performance and the psychological appeal of the model --each word expert attempting to disambiguate its contextual usage knows precisely t~e progress of its neighbors and the state of convergence (or the lack thereof) of the entire parsing process.The principal knowledge structure of the model is the word sense discrimination expert.A word expert represents the the linguistic knowledge required to dlsamblguate the meaning of a single word in any context. Although represented cumputationslly as coroutlnes, these experts differ considerably from ad hoc LISP programs and have approximately the same ~elatlon ~o LISP as an augmented transition network [15] grammar. ° 2use as rh~ graphic represeptatlon of an augmented transltlon networ~ aemonstrates the basic control paradigm of the ATN parsing approach, a graphic representation for word experts exists which embodies its functional framework. Each word expert derives from a branching discrimination structure called a word sense discrimination network or sense net. A sense nec consists of an ordered se~ of • /~tr~Ti~g (the nodes of the network), and for each one, the set of possible answers to that question (the branches emanating from each node).Traversal of a sense network represents the process of converging on a single contextual usage of a word.The terminal nodes of a sense net represent distinct word senses of the word modeled by the network. A sense net for the word "heavy" appears in part (a) of Figure 2 .Examination of this network reveals that four senses are represented --the three adjective usages shown in Figure 1 plus the numinal sense of "thug" as In "Joe's heavy told me to beat it."The network representation of a word expert leaves out certain computational necessities of actually using it for parsing.A word expert has two fundamental activities.(I) An expert asks questions about the lexical and conceptual data being amassed by its neighbors, the control states of various model components, and more general issues requiring common sense or knowledge of the physical world.(2) In addition, at each node an expert performs actions to affect the lexical and conceptual contents of the workspaces, the control states of itself, concept bins, 6An ATN without arbitrarily complex LISP computations on each arc and at each node, that is. If sense discrimination by a word expert results in the knowledge that a word to its right, either not yet executed or suspended, must map to a specific sense or conceptual category, then it should constrain it to do so, thus helping it avoid unnecessary processing or fallacious reasoning.Since word experts are represented as processes, constraining an expert consists of altering the pointer to the address at which it expects to continue execution.Through its descriptive header, an expert conditions this activity and insures that it takes place without disastrous consequences.Each node in the body of the expert has a type deslgnated by a letter following the node name. either Q (question), A (action), S (suspend), or T (terminal). By tracing through the question nodes (treating the others as vacuous except for their gore pointers), a sense network for each word expert process can be derived.The graphical framework of a word expert (and thus the questions it asks) represents its principal linguistic task of word sense disamblguatlon.Each question node has a type, shown following the Q in the.node --MC tmultiple choice), C (conditional), YN (yes/no/, and PI (posslble/Imposslble).In the example expert for "heavy", node nl represents a conditional query into the state of the entire parsing process, and n?de n[2 a multiple choice question involving the conceptual nature of the word to "heavy"s right in the input sentence.Multiple choice questions typically delve into the aslc relations among ob3ects ann actions zn the world. For example, the question asked at node n12 of the "heavy" expert is typical:"Is the object to my right better described as an artistic object a a form of precipitation, or a physical object?Action nodes in the "heavy" expert perform such tasks as determining the concept bin to which it contributes, and pqstin 8 expectations for the word to its right.In terms ot its side effects, the "heavy" expert is fairly simple. A full account of the word expert representation language will be available next year [12] .The basic structure of the Word Expert Parser depends principally on the role of individual word experts in affectlug.(1) each other:s actions and ~2) the neclaratlve result or computatlonal analysis. ~xperts affect each other by posting expectations on the central bulletin board, constraining each other, changing control states of model components (most notably themselves), and augmenting data. structures in. the workspeces. ° .They contribute to the conceptua£ ans ecnematlc result ot toe parse by contrlbuting object names, descrlptions~ schemata, ane other useful data to the concept level workspace. To determine exactly what contributions .to make, i.e.j the accurate ones In the particular run-tlme context at handj the experts as~ questions ot various kinds about the processe sot the model and the world at large.Four types of questions may be asked by an expert, and whereas some queries can be made in more than one way, the several question types solicit different kinds of information.Some questions requlre fairly involved inference to be answered adequately, and others demand no more than simple register lookup. This variety corresponds well, in my opinion, with human processing involved in conceptual analysis.Certain contextual clues to meaning are structural; taking advantage of them requires solel~ knowledge of the state of the parsing process (e.g., 'building a noun prase").Other clues subtly present themselves through more global evidence, usually having to do with linking together high order information about the specific domain at hand.In story comprehension, this involves the plot, characters, focus of attention, and general social psychology as well as common sense knowledge about the world.Understanding texts uealing with specialized subject matter requires knowledge about that particular subject, other subjects related to it, and of course, common sense. The questions asked by a word expert in arriving at the correct contextual interpretation of a word probe sources of both kinds of information, and take different forms. The automobile in "Joanie parked." is an example.could either be one that already exists in the workspace or a new one created by the expert at the time of its decision.After deciding on a concept, the principal role of a (content) word expert is to discriminate among the possibly many remaining senses of the word. Note that a good deal of this disambiguation may take place during the initial phase of concept determination. After asking enough questions to discover some piece of conceptual data, this data augments what already exists in the word's concept 5in, including declarative structures put there both by itself and by the other lexical participants in that concept.The parse completes when each word expert in the .workspace nas terminated.At this point, the concept ievez worKspace contains a complete conceptual interpretation ot the input text.Adequate conceptual parsing of input text regulres a stage missing from this dlscusslon and constituting the current phase of research ---the attachment of each picture and setting concept (bin) to the appropriate conceptual case of an event concept. Such a mechanism can be viewed in an entirely analogous fashion to the mechanisms just described for performln 8 local disamblguation of word senses. Rather ~han word experts, however, the experts on this level are conceptual in nature. The concept level thus becomes the main level of activity and a new level, call it the schema level workspace, turns into the ma~n repository rot inferred Information.When a concept bin has closed, a concept expert is retrieved from a disk file, and initialized. If it is an event concept, its function is to fill its conceptual cases with settings and pictures; if it is a setting or picture, it must aetermlne its schematic role. The activity on this level, therefore, involves higher order processing than sense discrimination, but occurs in Just about the same way.The ambiguities involved in mapping known concepts into conceptual case schemata appear identical to those having to do with ma2ping words into concepts.Discovering that the word "pit maps in a certain context to the notion of a "fruit pit" requires the same abilities and knowledge as realizing that "the red house" maps in some context to the notion of "a ~ocation for smoking pot and listening to records". The implementation of the mechanisms to carry out this next level of inferential disambiguation has already begun. It should be quite clear that this schematic level is by no means the end of the line --active expert-baseo p~ot following and general text understanding flt nicely Int? the word expert framework and constitute its loglca~ extension. | null | null | The Word Expert Parser successfully parses the sentence "The deep ~hilosopher throws the peach pit into the aeep pit," through cooperation among the appropriate word. experts, Initialization of ~he parser consists or retrlevln~ tr~ experts for "the", "deep', "philosopher", "throw", s", ~2An Important aeeumption of the word expert viewpoint is that the set or sucn contextual wars usages is not only finite, but fairly small as well.3The verspectlve of viewing language through lexlcal contribution~ to structure a~d meaning has naEurallv led to the development of wold experts for co~mon m?rphemes that are not war as ~ana even, experimentally, for ~unctuatlos),Especially important is the word expert tar "-ins', which aids significantly i n helpinR co Some word senses of "heavy"1. An overweight person is politely called "heavy":"He has become quite heavy."Emotional music is referred to as "heavy":"Mahler writes heavy music."~.An intensity of precipitation is "heavy":"A heavy snow is expected today."Some word senses of "ice"I.The solid state of water is called "ice":"Ice melts at 0Oc. "2. "Ice" participates In an idiomatic neminal describing a favorite delight:"Homemade ice cream is delicious."3. "Dry Ice" is the solid state of carbon dioxide:"Dry ice will keep that cool ;11 day." ~. "Ice" or "iced" describes things that have been cooled (sometimes with ice):"One iced tea to go please.""Ice" also describes things made of ice:"The ice sculptures are beautiful~" 6,7. "Ice hockey" is the name of a popular sport which has a rule penelizln~ an action called "icing":"Re iced the puck causing a face-off." ~. The term "ice box" refers to both a box containing ice used for cooling foods end a refrigerator:"This ice box isn't plugged in~"Flsure 1: Example contextual word usages ".over", and ~o forth, from a dis~ flle~ and .or~anizin 8 them along with data repositories cal~e~ wor~ oIns in a left to right order in ~he sentence level wo~k~pace. Note that three copies ot t T~-3R~...t ~or "the" anb c.~o cop.ies of each expert for "deep" and "pit" appear in th~ worKspace.Since each expert executes as a process, each process Inetantlatlon in the workspa..ce must be put into an executaole state.At this point, the parse is ready to begin.The word expert for "the" runs first, and is able to terminate immediately, creating a new concept designator (called a concept bin and participating in the concept level worksp~f~"~iclT-'will eventually hold the data the intellectual philosopher described in the input.Next the "deep" expert runs, and since "deep" has a number of word senses,5 is unable to terNinate (i.e~, complete its dlscriminetlgn task)..Instead,it ~uspenas its execution, stating the conditions upon winch it should be resumed.These conditions take the form of associative trigger patterns, and are referred to as disambiguate expressions Involving gerunds or participles such as "the man eat ir~ tiger". A full discussion ot thls will appear in 51t should be clear that the notion of "word sense" as used here encompasses what might more traditionally be ~escr.ibea as "contextua~ ~orn usage", Aspects o~ a word token's linguistic envlromnent constitute Its broadened "sense". restart demons.The "deep" expert creates .a restart demon co wake l'C up when the sense ot the nominal to its right ( l .e., "~hllosopher") becomes knoWn. The exper~ f.or "philosopher now runs, observes the co.ntrol state ot the parser, ant contributes the tact Chat One new concept refers to a person e.ngaged in the study of philosophy. As this expert terminates, the expert tot "=eep" resumes spontaneously, and, constrained by the fact chat "deep" must describe an entity that can be viewed as a person, it finally terminates successfully, contributing the fact that the person is intellectual.The "throw" expert runs next and successfully prunes away several usages of "throw" for contextua, reasons. A major reason for the semantic richness of verbs such as "throw", "cake", and "Jump", is that In context, each interacts strongly with a number of succeedin8 pre~ositions and adverbs to form distinct meaninBs, The woro expert approach easily handles this grouping together or words to torn larger word-like entities.In the particular case of verbs, the expert for a word like ."throw" simply exam.ines.i~.s rSght lex ical n.eighbor, an~ oases its oWn sense alscrtmlnet2on on the co(Rolnetlon or ~ at it .expects co find there, what It actually finds ere, an~ what this neighbor tells it (if It Soas so rat as to ask).No interesting p.article follows throw" in the current exampze, out It snoulo oe easy to conceive or th.e basic expert probes to discriminate the sense of "throw" wnen ;ol-owed by "away", "up", "out" ~ "in the towel", or other woras or wore groups, when no such word rollows "throw". as Is the case nere, its expert slmp-y waits for the existence of an entire concept to Its right, to determine if it meets any of the requirements .~hat would make the correct contextual interpretation of ' throw" different trom the expected "propel by moving ones arm" (e.g.,"throw a party'.').Before any such substantive conceptual activity takes place~ however, .t~ "S" expert ~uns arm ~ontri~uCes Its stannaro morphological information to throw "s data bin. This execution of the "s" expert does not, of course, affect "throw"' s suspended status.The "the" expert for the second "the" in the sentence runs next, and as in the previous case, creates a new con.cep~ bin to represent the da.~a about the no nina~ and des crlptlo.n, to come.Lne "peecn" expert realizes thatIt coulo oe either a noun or an adjective, and thus attempts what ~ call a "pairing" operation with its right neighbor. It essentially asks the expert for "pit" if the two ot them form a noun-noun pair.To determine the answer, ooth "pit" and "peach" have access to the entire model of linguistic and pragmatic knowledBe.Durtn~ this time. ~peach" is in a st.a~e called "attempting pairing" which Is nlzrerent trom the "suspended" state of the "throw" ex.~.ert. "Pit" answers back that it does pair up with "peach' (since "pit" is aware of its run-time context) and enters the "rea.dy" state. "Peach".now ned:ermines its c.orre~t sense and t;erm~netee:An.d ~nc~ only one mean%ngrul sense ~or'plt remains, the pit expert executes quickly, . t.ermlnattng with the contextually a~pro~riace "trulC pit" sense.As ic terminates, the piC. expert closes off the concept b.in In which It part~cipaces, spontaneously resumins the "throw" expert.An examination of the nature of fruit pit.a reveals that they are pergect.ly suited to propelling with ones. arm, ar~ thus, the "th.row" expert terminates successzul~y, contributing its wore| sense to its event concept bin..The "lnto~ expert, runs next, opens a concept bin ~of t~pe 'setting") rot the time, location, or situation about to be described, and suspends itself. On suspension, "lnto"'s expert posts an associative restart condition that will e.nable .its re.sumptlon when a new p~cture concept ~s opened to the right.This initial action CaKes p~ace rot most prepositions.In certain cases, if the end of a sentence is reached before an appropriate expected concept is opened, an expert will take alternative action.For example, one of the "in" experts restart trigger patterns consists of control state data of Just this kind --if the end of a sentence is rear.had .and no. conceptuql object, for the sect.ing creaceo oy "In" has oeen round, the "in" expert wxl~ resume nonetheless, and create a default concept t or perform some kind of intelligent reference aeterminatlon. The sentence "The doctor is In." illustrates this point.In the current example~ the. "the" expert that executes lm.med~ately alter t_.nto"'s suspension creates the exporter.picture concept.The These situations usuaA.~y resolve themes+yes wl~_h a ca §qadlns o~ expert res,-,ptlons and terminations. In our seep ~c example, "deep" ~oets expectations on the central tableau of global control state Knowledge, and waits rot "pit" to terminate • "PIt"' s expert now runs, and since thls bulletin board contains "deep"'s expectations of a ~.oI~, or printed matter, "pit" maps immediately onto a large hole in the ground. This in turn, causes both the resumption and termination of the "deep" expert as well as the closure of the concept bin to whlch the~ oelong.At the closing of the concept bin, the "into expert resumes, marks its concept as a location, and terminates.With all the word experts completed and all concept bins closed, the expert for ".'" runs and completes the parse.The concept level workspace now contains five concepts: a picture concept designating an intellectual philosopher, an event concept representing the throwing action, another picture concept describing a fruit pit which came from a peach, a setting concept representing a location, and the picture concept which describes precisely the nature of this location. Work on the mechanism to determine the schematic roles of the concepts has just begun, and is described briefl~ later. A program trace that shows the actions ot the Nora Expert Parser on the example just presented is available on request. | The Word Expert Parser is a theory of o rganization and cgntro ~ for a conceptual, lansuage an@.~yzer. Th~ contro~ envlrosment ts cnaracter~zeo ny a co£~ectlon ot generator-like coroutines, called word experts, which cooperatively arrive at a conceptual interpretation of an ~nput sentence. Many torms of linguistic ann non-lln~uistlc knowledge are available to these experts In performing their task, including control state Knowledge and knowledge of the world, and by eliminating all but the mpst persistent forms of ambiguity, the parser models numan processing.This new model of parsin£ claims a number of theoretical advantages: (I) Its representations of linguistic knowledge reflect the enormous redundancy in natural languages --without this redundancy in the model, the inter-expert handshaking (seen in many..forms in the example parse) would not be possible. ~z) ~ne model suggests some interesting approaches to language acquisition.Since much of a word expert's knowledge Is encoded in a branching discrimination structure,, addlng new information about a word involves the addition oz a new branch. This branch would be placed in the expert at the point where the contextual clues for dlsambiguatlng the new usage differ from those present for a known usage. (3) Idiosyncratic uses of langua8@ are easily e ncooea, s~nce the wore expert provides a c~esr way to no so.These uses are indistinguishable from other uses in their encodings in the model. (4) The parser represents a cognltively plausible model or se~uentlal coroutine-like processing in human ~anguage understanding.The organization of linguistic knowledge around the word, rather than the rewrite rule, motivates interesting conjectures about the flow of control In a human language understander. | Main paper:
model overview:
The Word Expert Parser successfully parses the sentence "The deep ~hilosopher throws the peach pit into the aeep pit," through cooperation among the appropriate word. experts, Initialization of ~he parser consists or retrlevln~ tr~ experts for "the", "deep', "philosopher", "throw", s", ~2An Important aeeumption of the word expert viewpoint is that the set or sucn contextual wars usages is not only finite, but fairly small as well.3The verspectlve of viewing language through lexlcal contribution~ to structure a~d meaning has naEurallv led to the development of wold experts for co~mon m?rphemes that are not war as ~ana even, experimentally, for ~unctuatlos),Especially important is the word expert tar "-ins', which aids significantly i n helpinR co Some word senses of "heavy"1. An overweight person is politely called "heavy":"He has become quite heavy."Emotional music is referred to as "heavy":"Mahler writes heavy music."~.An intensity of precipitation is "heavy":"A heavy snow is expected today."Some word senses of "ice"I.The solid state of water is called "ice":"Ice melts at 0Oc. "2. "Ice" participates In an idiomatic neminal describing a favorite delight:"Homemade ice cream is delicious."3. "Dry Ice" is the solid state of carbon dioxide:"Dry ice will keep that cool ;11 day." ~. "Ice" or "iced" describes things that have been cooled (sometimes with ice):"One iced tea to go please.""Ice" also describes things made of ice:"The ice sculptures are beautiful~" 6,7. "Ice hockey" is the name of a popular sport which has a rule penelizln~ an action called "icing":"Re iced the puck causing a face-off." ~. The term "ice box" refers to both a box containing ice used for cooling foods end a refrigerator:"This ice box isn't plugged in~"Flsure 1: Example contextual word usages ".over", and ~o forth, from a dis~ flle~ and .or~anizin 8 them along with data repositories cal~e~ wor~ oIns in a left to right order in ~he sentence level wo~k~pace. Note that three copies ot t T~-3R~...t ~or "the" anb c.~o cop.ies of each expert for "deep" and "pit" appear in th~ worKspace.Since each expert executes as a process, each process Inetantlatlon in the workspa..ce must be put into an executaole state.At this point, the parse is ready to begin.The word expert for "the" runs first, and is able to terminate immediately, creating a new concept designator (called a concept bin and participating in the concept level worksp~f~"~iclT-'will eventually hold the data the intellectual philosopher described in the input.Next the "deep" expert runs, and since "deep" has a number of word senses,5 is unable to terNinate (i.e~, complete its dlscriminetlgn task)..Instead,it ~uspenas its execution, stating the conditions upon winch it should be resumed.These conditions take the form of associative trigger patterns, and are referred to as disambiguate expressions Involving gerunds or participles such as "the man eat ir~ tiger". A full discussion ot thls will appear in 51t should be clear that the notion of "word sense" as used here encompasses what might more traditionally be ~escr.ibea as "contextua~ ~orn usage", Aspects o~ a word token's linguistic envlromnent constitute Its broadened "sense". restart demons.The "deep" expert creates .a restart demon co wake l'C up when the sense ot the nominal to its right ( l .e., "~hllosopher") becomes knoWn. The exper~ f.or "philosopher now runs, observes the co.ntrol state ot the parser, ant contributes the tact Chat One new concept refers to a person e.ngaged in the study of philosophy. As this expert terminates, the expert tot "=eep" resumes spontaneously, and, constrained by the fact chat "deep" must describe an entity that can be viewed as a person, it finally terminates successfully, contributing the fact that the person is intellectual.The "throw" expert runs next and successfully prunes away several usages of "throw" for contextua, reasons. A major reason for the semantic richness of verbs such as "throw", "cake", and "Jump", is that In context, each interacts strongly with a number of succeedin8 pre~ositions and adverbs to form distinct meaninBs, The woro expert approach easily handles this grouping together or words to torn larger word-like entities.In the particular case of verbs, the expert for a word like ."throw" simply exam.ines.i~.s rSght lex ical n.eighbor, an~ oases its oWn sense alscrtmlnet2on on the co(Rolnetlon or ~ at it .expects co find there, what It actually finds ere, an~ what this neighbor tells it (if It Soas so rat as to ask).No interesting p.article follows throw" in the current exampze, out It snoulo oe easy to conceive or th.e basic expert probes to discriminate the sense of "throw" wnen ;ol-owed by "away", "up", "out" ~ "in the towel", or other woras or wore groups, when no such word rollows "throw". as Is the case nere, its expert slmp-y waits for the existence of an entire concept to Its right, to determine if it meets any of the requirements .~hat would make the correct contextual interpretation of ' throw" different trom the expected "propel by moving ones arm" (e.g.,"throw a party'.').Before any such substantive conceptual activity takes place~ however, .t~ "S" expert ~uns arm ~ontri~uCes Its stannaro morphological information to throw "s data bin. This execution of the "s" expert does not, of course, affect "throw"' s suspended status.The "the" expert for the second "the" in the sentence runs next, and as in the previous case, creates a new con.cep~ bin to represent the da.~a about the no nina~ and des crlptlo.n, to come.Lne "peecn" expert realizes thatIt coulo oe either a noun or an adjective, and thus attempts what ~ call a "pairing" operation with its right neighbor. It essentially asks the expert for "pit" if the two ot them form a noun-noun pair.To determine the answer, ooth "pit" and "peach" have access to the entire model of linguistic and pragmatic knowledBe.Durtn~ this time. ~peach" is in a st.a~e called "attempting pairing" which Is nlzrerent trom the "suspended" state of the "throw" ex.~.ert. "Pit" answers back that it does pair up with "peach' (since "pit" is aware of its run-time context) and enters the "rea.dy" state. "Peach".now ned:ermines its c.orre~t sense and t;erm~netee:An.d ~nc~ only one mean%ngrul sense ~or'plt remains, the pit expert executes quickly, . t.ermlnattng with the contextually a~pro~riace "trulC pit" sense.As ic terminates, the piC. expert closes off the concept b.in In which It part~cipaces, spontaneously resumins the "throw" expert.An examination of the nature of fruit pit.a reveals that they are pergect.ly suited to propelling with ones. arm, ar~ thus, the "th.row" expert terminates successzul~y, contributing its wore| sense to its event concept bin..The "lnto~ expert, runs next, opens a concept bin ~of t~pe 'setting") rot the time, location, or situation about to be described, and suspends itself. On suspension, "lnto"'s expert posts an associative restart condition that will e.nable .its re.sumptlon when a new p~cture concept ~s opened to the right.This initial action CaKes p~ace rot most prepositions.In certain cases, if the end of a sentence is reached before an appropriate expected concept is opened, an expert will take alternative action.For example, one of the "in" experts restart trigger patterns consists of control state data of Just this kind --if the end of a sentence is rear.had .and no. conceptuql object, for the sect.ing creaceo oy "In" has oeen round, the "in" expert wxl~ resume nonetheless, and create a default concept t or perform some kind of intelligent reference aeterminatlon. The sentence "The doctor is In." illustrates this point.In the current example~ the. "the" expert that executes lm.med~ately alter t_.nto"'s suspension creates the exporter.picture concept.The These situations usuaA.~y resolve themes+yes wl~_h a ca §qadlns o~ expert res,-,ptlons and terminations. In our seep ~c example, "deep" ~oets expectations on the central tableau of global control state Knowledge, and waits rot "pit" to terminate • "PIt"' s expert now runs, and since thls bulletin board contains "deep"'s expectations of a ~.oI~, or printed matter, "pit" maps immediately onto a large hole in the ground. This in turn, causes both the resumption and termination of the "deep" expert as well as the closure of the concept bin to whlch the~ oelong.At the closing of the concept bin, the "into expert resumes, marks its concept as a location, and terminates.With all the word experts completed and all concept bins closed, the expert for ".'" runs and completes the parse.The concept level workspace now contains five concepts: a picture concept designating an intellectual philosopher, an event concept representing the throwing action, another picture concept describing a fruit pit which came from a peach, a setting concept representing a location, and the picture concept which describes precisely the nature of this location. Work on the mechanism to determine the schematic roles of the concepts has just begun, and is described briefl~ later. A program trace that shows the actions ot the Nora Expert Parser on the example just presented is available on request.
structure of the model:
The organization of the parser centers around data repositories on two levels --the sentence level workspace contains a word bin for each word (and sub-lexical morpheme) of the input and the concept level workspace contains a concept bin (described above) for each concept referred to in the input sentence. A third level of processing, the schema level workspaee, while not yet implemented, will contain a schema for each conceptual action of the input sentence.All actions affecting the contents of these data bins are carried out by the word expert processes, one of which is associated with each word bin in the wo rkspace.In addition to this first order information about lexical and conceptual objects, the parser contains a central tableau of control state descriptions available to any expert that can make use of self referential knowledge about its own processing or the states of processing of other model components.The availability of such control state information improves considerably both the performance and the psychological appeal of the model --each word expert attempting to disambiguate its contextual usage knows precisely t~e progress of its neighbors and the state of convergence (or the lack thereof) of the entire parsing process.The principal knowledge structure of the model is the word sense discrimination expert.A word expert represents the the linguistic knowledge required to dlsamblguate the meaning of a single word in any context. Although represented cumputationslly as coroutlnes, these experts differ considerably from ad hoc LISP programs and have approximately the same ~elatlon ~o LISP as an augmented transition network [15] grammar. ° 2use as rh~ graphic represeptatlon of an augmented transltlon networ~ aemonstrates the basic control paradigm of the ATN parsing approach, a graphic representation for word experts exists which embodies its functional framework. Each word expert derives from a branching discrimination structure called a word sense discrimination network or sense net. A sense nec consists of an ordered se~ of • /~tr~Ti~g (the nodes of the network), and for each one, the set of possible answers to that question (the branches emanating from each node).Traversal of a sense network represents the process of converging on a single contextual usage of a word.The terminal nodes of a sense net represent distinct word senses of the word modeled by the network. A sense net for the word "heavy" appears in part (a) of Figure 2 .Examination of this network reveals that four senses are represented --the three adjective usages shown in Figure 1 plus the numinal sense of "thug" as In "Joe's heavy told me to beat it."The network representation of a word expert leaves out certain computational necessities of actually using it for parsing.A word expert has two fundamental activities.(I) An expert asks questions about the lexical and conceptual data being amassed by its neighbors, the control states of various model components, and more general issues requiring common sense or knowledge of the physical world.(2) In addition, at each node an expert performs actions to affect the lexical and conceptual contents of the workspaces, the control states of itself, concept bins, 6An ATN without arbitrarily complex LISP computations on each arc and at each node, that is. If sense discrimination by a word expert results in the knowledge that a word to its right, either not yet executed or suspended, must map to a specific sense or conceptual category, then it should constrain it to do so, thus helping it avoid unnecessary processing or fallacious reasoning.Since word experts are represented as processes, constraining an expert consists of altering the pointer to the address at which it expects to continue execution.Through its descriptive header, an expert conditions this activity and insures that it takes place without disastrous consequences.Each node in the body of the expert has a type deslgnated by a letter following the node name. either Q (question), A (action), S (suspend), or T (terminal). By tracing through the question nodes (treating the others as vacuous except for their gore pointers), a sense network for each word expert process can be derived.The graphical framework of a word expert (and thus the questions it asks) represents its principal linguistic task of word sense disamblguatlon.Each question node has a type, shown following the Q in the.node --MC tmultiple choice), C (conditional), YN (yes/no/, and PI (posslble/Imposslble).In the example expert for "heavy", node nl represents a conditional query into the state of the entire parsing process, and n?de n[2 a multiple choice question involving the conceptual nature of the word to "heavy"s right in the input sentence.Multiple choice questions typically delve into the aslc relations among ob3ects ann actions zn the world. For example, the question asked at node n12 of the "heavy" expert is typical:"Is the object to my right better described as an artistic object a a form of precipitation, or a physical object?Action nodes in the "heavy" expert perform such tasks as determining the concept bin to which it contributes, and pqstin 8 expectations for the word to its right.In terms ot its side effects, the "heavy" expert is fairly simple. A full account of the word expert representation language will be available next year [12] .The basic structure of the Word Expert Parser depends principally on the role of individual word experts in affectlug.(1) each other:s actions and ~2) the neclaratlve result or computatlonal analysis. ~xperts affect each other by posting expectations on the central bulletin board, constraining each other, changing control states of model components (most notably themselves), and augmenting data. structures in. the workspeces. ° .They contribute to the conceptua£ ans ecnematlc result ot toe parse by contrlbuting object names, descrlptions~ schemata, ane other useful data to the concept level workspace. To determine exactly what contributions .to make, i.e.j the accurate ones In the particular run-tlme context at handj the experts as~ questions ot various kinds about the processe sot the model and the world at large.Four types of questions may be asked by an expert, and whereas some queries can be made in more than one way, the several question types solicit different kinds of information.Some questions requlre fairly involved inference to be answered adequately, and others demand no more than simple register lookup. This variety corresponds well, in my opinion, with human processing involved in conceptual analysis.Certain contextual clues to meaning are structural; taking advantage of them requires solel~ knowledge of the state of the parsing process (e.g., 'building a noun prase").Other clues subtly present themselves through more global evidence, usually having to do with linking together high order information about the specific domain at hand.In story comprehension, this involves the plot, characters, focus of attention, and general social psychology as well as common sense knowledge about the world.Understanding texts uealing with specialized subject matter requires knowledge about that particular subject, other subjects related to it, and of course, common sense. The questions asked by a word expert in arriving at the correct contextual interpretation of a word probe sources of both kinds of information, and take different forms. The automobile in "Joanie parked." is an example.could either be one that already exists in the workspace or a new one created by the expert at the time of its decision.After deciding on a concept, the principal role of a (content) word expert is to discriminate among the possibly many remaining senses of the word. Note that a good deal of this disambiguation may take place during the initial phase of concept determination. After asking enough questions to discover some piece of conceptual data, this data augments what already exists in the word's concept 5in, including declarative structures put there both by itself and by the other lexical participants in that concept.The parse completes when each word expert in the .workspace nas terminated.At this point, the concept ievez worKspace contains a complete conceptual interpretation ot the input text.Adequate conceptual parsing of input text regulres a stage missing from this dlscusslon and constituting the current phase of research ---the attachment of each picture and setting concept (bin) to the appropriate conceptual case of an event concept. Such a mechanism can be viewed in an entirely analogous fashion to the mechanisms just described for performln 8 local disamblguation of word senses. Rather ~han word experts, however, the experts on this level are conceptual in nature. The concept level thus becomes the main level of activity and a new level, call it the schema level workspace, turns into the ma~n repository rot inferred Information.When a concept bin has closed, a concept expert is retrieved from a disk file, and initialized. If it is an event concept, its function is to fill its conceptual cases with settings and pictures; if it is a setting or picture, it must aetermlne its schematic role. The activity on this level, therefore, involves higher order processing than sense discrimination, but occurs in Just about the same way.The ambiguities involved in mapping known concepts into conceptual case schemata appear identical to those having to do with ma2ping words into concepts.Discovering that the word "pit maps in a certain context to the notion of a "fruit pit" requires the same abilities and knowledge as realizing that "the red house" maps in some context to the notion of "a ~ocation for smoking pot and listening to records". The implementation of the mechanisms to carry out this next level of inferential disambiguation has already begun. It should be quite clear that this schematic level is by no means the end of the line --active expert-baseo p~ot following and general text understanding flt nicely Int? the word expert framework and constitute its loglca~ extension.
summary and conclusions:
The Word Expert Parser is a theory of o rganization and cgntro ~ for a conceptual, lansuage an@.~yzer. Th~ contro~ envlrosment ts cnaracter~zeo ny a co£~ectlon ot generator-like coroutines, called word experts, which cooperatively arrive at a conceptual interpretation of an ~nput sentence. Many torms of linguistic ann non-lln~uistlc knowledge are available to these experts In performing their task, including control state Knowledge and knowledge of the world, and by eliminating all but the mpst persistent forms of ambiguity, the parser models numan processing.This new model of parsin£ claims a number of theoretical advantages: (I) Its representations of linguistic knowledge reflect the enormous redundancy in natural languages --without this redundancy in the model, the inter-expert handshaking (seen in many..forms in the example parse) would not be possible. ~z) ~ne model suggests some interesting approaches to language acquisition.Since much of a word expert's knowledge Is encoded in a branching discrimination structure,, addlng new information about a word involves the addition oz a new branch. This branch would be placed in the expert at the point where the contextual clues for dlsambiguatlng the new usage differ from those present for a known usage. (3) Idiosyncratic uses of langua8@ are easily e ncooea, s~nce the wore expert provides a c~esr way to no so.These uses are indistinguishable from other uses in their encodings in the model. (4) The parser represents a cognltively plausible model or se~uentlal coroutine-like processing in human ~anguage understanding.The organization of linguistic knowledge around the word, rather than the rewrite rule, motivates interesting conjectures about the flow of control In a human language understander.
[. introduction:
Computational understanding of natural language requires complex Interactions among a variety of distinct yet redundant mechanisms.The construction of a computer program to perform such a task begins with the development of an organizational framework which Inherently .incorporates certain assumptions about the nature ot these processes and the environment in which they take place. Such cognitive premises affect nro?oundly the scope and substance of computational ~nalysis for comprehension as found in the program.This paper describes a theory of conceptual parsing which considers knowledge about language to be distributed across a collection of procedural experts centered on individual words. Natural language parsing with word experts entails several new hypotheses about the organization and representation of linguistic and pragmatic knowledge for computational language comprenension.The Word Expert Parser [1] demonstrates hpw the word expert qTt~T~ed w£~h certain ocher choices oaseo on previous work, affect structure and process in a cognitive model of parsing.The Word Expert Parser is a cognitive model of conceptual language analysis in which the unit of ltngu~stic knowledge is the word and the fqcu~ o~ research ts the set or processes unoerlyinR comprehension.The model is aimed directly at problem~ of word sense ambiguity and idiomatic expressions, and in greatly generalizing the notion of wora sense, promotes these issues to a central place in the study of language parsing.Parsing models typically cope unsatisfactorily with the wide heterogeneity of usages of particular words.If a sentence contains a standard form of a word, it can usually be parsed; if it involves a less prevalent form which has a different part of speech, perhaps it too can be parsed. Disti.nguishing amen 8 the ~any senses of a common vero, adjective, or pronoun, tar example, or correctly translating idioms are rarely possible, At the source of this difficulty is the reliance on rule-based formalisms, whethar syntactic or semantic (e.g.. cases), which attempt to capture ~he linguistic contributions inherent in constituent chunks or sentences that consist of more than single words.A crucial assumption underlying work on the Word Expert Parser is that the ~undamental unit of linguistic Knowledge is the word. and that understanding its sense or role in a particular context is the central parsing process. In the parser to be described, the word expert constitutes the kernel of linguistic knowled~nd zts representation the e~emental data structure.IE is procedural in nature and executes directly as a process, cooperating with the other experts for a given sentence to arrive at a mutually acceptable sentence meaning.Certaln principles behind the parser d 9 nqt follow directly from the view or worn primacy, out ~rom other recent theories of parsing. The cognitive processes involved in language comprehension comprise the focus of linguistic study of the word expert approach. Parsin8 is viewea as an inferential process where linguistic knowledge of syntax and semantics and general pragmatic knowledge are applied in a uniform manner during IThe research described in this renor~ .is funded by the National Aeronautics and Space Admzn~stratton under grant , n umbe, r NSC-7255. Their support is gratefully acKnowleageG,This methodological position closely follows that of Rlosbeck (see [2] and [3 ]) and Schank [4] . The central concern with word usage and word sense ambiguity follows similar motivatlons of Wllks [5] . The control structure of the Word Expert Parser results from agreqment .with ~he hypothesis of .Harcus that parsing can he none aetermzntsttcally and ~n a way tn Dhlcn information ,gained through interpretation is permanent [6] . Rieger rne Importance at these mechanisms tar wore usage diagnosis derives from the ubiquity of local ambiguities, and brought about the notion chat ~hey be made the central processes of computational analysls an 9 understanding, Consideration of almost any Engllsn content word leads to a realization of the scope of the problem --with a little time and perhaps help from the dlctlonaFy , man~.dlstinct usages can ee.id~ntifl~d.As.a stmpie lllustrarzon, several usages earn tar the worus "heavy" and "ice" appear in Figure I . Each of. these seemingly" benign words exhibits a rich depth of contextual use, An earlier paper contains.a list at almost sixty verbal usages for the word "take" [llJ.The representation of all contextual word usages in an active way t~at insures their utility for linguistic dlagnasis led to the notion of word experts.Each word expert is a procedural entit~~f all posslblq contextual interpretations of the -word it represents. = Whe~ placed in a context formed by.expqrts for thg.othe ~ wares In a sentence, earn expert ShOUld De capaole or sufficient context-problng and self-examination to determine successfully' its functional or semantic role, and further, to realize the nature of that function or the precise meaning of the word. The representation and control issues involved in basing a parser on word experts are discussed below, following presentation of an example execution of the existing Word Expert Parser.
Appendix:
| null | null | null | null | {
"paperhash": [
"rieger|word_expert_parsing",
"rieger|the_importance_of_multiple_choice.",
"bobrow|an_overview_of_krl,_a_knowledge_representation_language",
"riesbeck|comprehension_by_computer_:_expectation-based_analysis_of_sentences_in_context",
"erman|a_multi-level_organization_for_problem_solving_using_many,_diverse,_cooperating_sources_of_knowledge",
"mcdermott|the_conniver_reference_manual",
"riesbeck|computational_understanding_:_analysis_of_sentences_and_context",
"bobrow|computational_linguistics_transition_network_grammars_for_natural_language_analysis"
],
"title": [
"Word Expert Parsing",
"The Importance of Multiple Choice.",
"An overview of KRL, a Knowledge Representation Language",
"Comprehension by computer : expectation-based analysis of sentences in context",
"A Multi-Level Organization For Problem Solving Using Many, Diverse, Cooperating Sources Of Knowledge",
"The Conniver Reference Manual",
"Computational understanding : analysis of sentences and context",
"Computational Linguistics Transition Network Grammars for Natural Language Analysis"
],
"abstract": [
"An approach to natural language meaning-based parsing in which the unit linguistic knowledge is the word rather than the rewrite rule is described. In the Word Expert Parser, knowledge about language is distributed across a population of procedural experts, each representing a word of the language, and each an expert at diagnosing that word a intended usage in context. The parser is structured around a coroutine control environment in which the generator-like word experts ask questions and exchange information in coming to collective agreement on sentence meaning. The word Expert theory is advanced as a better cognitive model of human language expertise than the traditional rule-based approach. The technical discussion is organized around examples taken from the prototype LISP system which implements parts of the theory.",
"Abstract : A level of organization of inferences in which competing plausible alternatives can be compared, and all but one actively rejected, is a very important aspect of any comprehension model. Multiple choice inference structures help the model stay tuned to the comprehension context, and help establish a framework in which inference producers are less likely to outstrip inference consumers, a common problem of inference systems. I look at four different modelling areas in which the same issues dominate, and suggest that the structure of knowledge is greatly influenced if one adopts the point of view which places emphasis on multiple choice. (Author)",
"This paper describes KRL, a Knowledge Representation Language designed for use in understander systems. It outlines both the general concepts which underlie our research and the details of KRL-0, an experimental implementation of some of these concepts. KRL is an attempt to integrate procedural knowledge with a broad base of declarative forms. These forms provide a variety of ways to express the logical structure of the knowledge, in order to give flexibility in associating procedures (for memory and reasoning) with specific pieces of knowledge, and to control the relative accessibility of different facts and descriptions. The formalism for declarative knowledge is based on structured conceptual objects with associated descriptions. These objects form a network of memory units with several different sorts of linkages, each having well-specified implications for the retrieval process. Procedures can be associated directly with the internal structure of a conceptual object. This procedural attachment allows the steps for a particular operation to be determined by characteristics of the specific entities involved. The control structure of KRL is based on the belief that the next generation of intelligent programs will integrate data-directed and goal-directed processing by using multi-processing. It provides for a priority-ordered multi-process agenda with explicit (user-provided) strategies for scheduling and resource allocation. It provides procedure directories which operate along with process frameworks to allow procedural parameterization of the fundamental system processes for building, comparing, and retrieving memory structures. Future development of KRL will include integrating procedure definition with the descriptive formalism.",
"Abstract : ELI (English Language Interpreter) is a natural language parsing program currently used by several story understanding systems. ELI differs from most other parsers in that it: produces meaning representations (using Schank's Conceptual Dependency system) rather than syntactic structures; uses syntactic information only when the meaning can not be obtained directly; talks to other programs that make high level inferences that tie individual events into coherent episodes; uses context-based exceptions (conceptual and syntactic) to control its parsing routines. Examples of texts that ELI has understood, and details of how it works are given.",
"An organization is presented for implementing solutions to knowledge-based AI problems. The hypothesize-and-test paradigm is used as the basis for cooperation among many diverse and independent knowledge sources (KS's). The KS's are assumed individually to be errorful and incomplete. \n \nA uniform and integrated multi-level structure, the blackboard, holds the current state of the system. Knowledge sources cooperate by creating, accessing, and modifying elements in the blackboard. The activation of a KS is data-driven, based on the occurrence of patterns in the blackboard which match templates specified by the knowledge source. \n \nEach level in the blackboard specifies a different representation of the problem space; the sequence of levels forms a loose hierarchy in which the elements at each level can approximately be described as abstractions of elements at the next lower level. This decomposition can be thought of as an a prion framework of a plan for solving the problem; each level is a generic stage in the plan. \n \nThe elements at each level in the blackboard are hypotheses about some aspect of that level. The internal structure of an hypothesis consists of a fixed set of attributes; this set is the same for hypotheses at all levels of representation in the blackboard. These attributes are selected to serve as mechanisms for implementing the data-directed hypothesize-and-test paradigm and for efficient goal-directed scheduling of KS's. Knowledge sources may create networks of structural relationships among hypotheses. These relationships, which are explicit in the blackboard, serve to represent inferences and deductions made by the KS's about the hypotheses; they also allow competing and overlapping partial solutions to be handled in an integrated manner. \n \nThe Hearsay II speech-understanding system is an implementation of this organization; it is used here as an example for descriptive purposes.",
"Abstract : The manual is an introduction and reference to the latest version of the Conniver programming language, an artificial intelligence language with general control and data-base structures. (Author)",
"Abstract : The goal of this thesis was to develop a system for the computer analysis of written natural language texts that could also serve as a theroy of human comprehension of natural language. Therefore the construction of this system was guided by four basic assumptions about natural language comprehension. First, the primary goal of comprehension is always to find meanings as soon as possible, Other tasks, such as discovering the syntactic relationships, are performed only when essential to decisions about meaning. Second, an attempt is made to understand each word as soon as it is read, to decide what it means and how it relates to the rest of the text. Third, comprehension means not only understanding what has been seen but also predicting what is likely to be seen next. Fourth, the words of a text provide the cues for finding the information necessary for comprehending that text.",
"The use of augmented transition network grammars for the analysis of natural language sentences is described. Structure building actions associated with the arcs of the grammar network allow for the reordering, restructuring, and copying of constituents necessary to produce deep-structure representations of the type normally obtained from a transforma-tional analysis, and conditions on the arcs allow for a powerful selectivity which can rule out meaningless analyses and take advantage of semantic information to guide the parsing. The advantages of this model for natural language analysis are discussed in detail and illustrated by examples. An implementation of an experimental parsing system for transition network grammars is briefly described. One of the early models for natural language grammars was the finite state transition graph. This model consists of a network of nodes and directed arcs connecting them, where the nodes correspond to states in a finite state machine and the arcs represent transitions from state to state. Each arc is labeled with a symbol whose input can cause a transition from the state at the tail of the arc to the state at its head. This model has the attractive feature that the sequences of words which make up a sentence can be read off directly by following the paths through the grammar from the initial state to some final state. Unfortunately , the model is grossly inadequate for the representation of natural language grammars due to its failure to capture many of their regularities. A most notable inadequacy is the absence of a pushdown mechanism that permits one to suspend the processing of a constituent at a given level while using the same grammar to process an embedded constituent. Suppose, however, that one added the mechanism of re-cursion directly to the transition graph model by fiat. That is, suppose one took a collection of transition graphs each with a name, and permitted as labels on the arcs not only terminal symbols but also nonterminal symbols naming complex constructions which must be present in order for the transition to be followed. The determination of whether such a construction was in fact present in a sentence would be done by a \"subroutine call\" to another transition graph (or the same one). The resulting model of grammar, which we will call a recursive transition network, is equivalent in generative power to that of a context-free grammar or pushdown store automaton, but as we will …"
],
"authors": [
{
"name": [
"Chuck Rieger",
"Steven L. Small"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"C. Rieger"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. Bobrow",
"T. Winograd"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"C. Riesbeck",
"R. Schank"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"L. Erman",
"V. Lesser"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. McDermott",
"G. Sussman"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"C. Riesbeck"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
}
]
},
{
"name": [
"D. G. Bobrow",
"W. A. Woods"
],
"affiliation": [
{
"laboratory": null,
"institution": null,
"location": null
},
{
"laboratory": null,
"institution": null,
"location": null
}
]
}
],
"arxiv_id": [
null,
null,
null,
null,
null,
null,
null,
null
],
"s2_corpus_id": [
"267896024",
"116487884",
"7965074",
"60546035",
"8524471",
"59106434",
"60975873",
"267890696"
],
"intents": [
[],
[
"background"
],
[],
[],
[],
[],
[],
[
"methodology"
]
],
"isInfluential": [
false,
false,
false,
false,
false,
false,
false,
false
]
} | Problem: The paper aims to address the challenge of natural language understanding by proposing a theory of conceptual parsing using distributed procedural experts centered on individual words.
Solution: The hypothesis posited is that the Word Expert Parser, based on the Word Expert theory, provides a more effective cognitive model for natural language understanding compared to traditional rule-based approaches. | 548 | 0.076642 | null | null | null | null | null | null | null | null |
b4d0377b0f05765c7dbb8f7ed5b382838b7ec166 | 9840518 | null | Lexikologi som datalingvistik (Lexicology as computational linguistics) [In {S}wedish] | Alla teoretiska språkmodeller brukar ha en komponent med språkli ga byggstenar, ett lexikon, och en komponent med regler för bygg stenarnas sammanfogning till satser och texter, en grammatik. Vid lemmatisering -angett som ett huvudtema vid datalingvistikdagarna -aktualiseras ett par grundläggande frågor som rör den lexika liska komponenten. Med lemmatisering syftar man i första hand till att samklassificera formella och funktionella varianter till ab straktare lexikaliska enheter och upprätta kanoniska grundformer. Jag vill med det här bidraget lägga några generella synpunkter på lexikaliska enheters egenskaper och inbördes relationer, dvs. på den lexikaliska komponentens allmänna struktur. Det lexikaliska modellbyggandet är i och för sig inte någon spe cifikt datalingvistisk angelägenhet. Det finns emellertid flera skäl att diskutera lexikologi i ett datalingvistiskt sammanhang. För det första har inte allmänlingvisterna ägnat sig i särskilt hög grad åt lexikaliska frågor. Huvudvikten har alltid legat vid andra grammatikkomponenter, såsom den syntaktiska, den semantiska, den fonologiska och i någon mån den morfologiska, olika starkt betonade under olika perioder. Eftersom den teoretiska modellen ändå förutsätter en lexikalisk komponent har denna tenderat att bli något slags "garbage component", som man hänskjutit proble men till då man inte velat ta itu med dem i det sammanhang som just varit aktuellt. Som kontrast har datalingvister ofta haft att lösa lexikaliska problem inom ramen för praktisk verksamhet. Datalingvisten har ofta nog fått bli sin egen lexikolog. Som en följd har rätt mycket av den teoretiskt inriktade lexikologi som över huvud taget bedrivits på senare tid presterats av dataling vister . Det finns emellertid skäl även för andra lexikologer att närma sig datalingvistiken. Själva dot lexikaliska maLorialots art gör doL nödvänduiL all lexikolo<|i lu'drivr. i n.ii.i rMimräd med data-Lexikologi som datalingvistik | {
"name": [
"Ralph, Bo"
],
"affiliation": [
null
]
} | null | null | Proceedings of the 2nd Nordic Conference of Computational Linguistics ({NODALIDA} 1979) | 1979-10-01 | 0 | 0 | null | null | null | null | null | Proceedings of NODALIDA 1979 | Main paper:
:
Proceedings of NODALIDA 1979
Appendix:
| null | null | null | null | {
"paperhash": [],
"title": [],
"abstract": [],
"authors": [],
"arxiv_id": [],
"s2_corpus_id": [],
"intents": [],
"isInfluential": []
} | null | 544 | 0 | null | null | null | null | null | null | null | null |
Subsets and Splits
No saved queries yet
Save your SQL queries to embed, download, and access them later. Queries will appear here once saved.