Datasets:

Modalities:
Text
ArXiv:
Libraries:
Datasets
License:
Xinyu Crystina ZHANG commited on
Commit
d921ec7
β€’
2 Parent(s): d0e7410 36d9415

Merge branch 'main' of https://huggingface.co/datasets/miracl/miracl-corpus

Browse files
Files changed (2) hide show
  1. README.md +63 -3
  2. miracl-corpus.py +2 -2
README.md CHANGED
@@ -1,4 +1,4 @@
1
-
2
  annotations_creators:
3
  - expert-generated
4
 
@@ -21,7 +21,6 @@ language:
21
  - zh
22
 
23
 
24
-
25
  multilinguality:
26
  - multilingual
27
 
@@ -33,6 +32,67 @@ tags: []
33
  task_categories:
34
  - text-retrieval
35
 
 
 
36
 
37
  task_ids:
38
- - document-retrieval
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
  annotations_creators:
3
  - expert-generated
4
 
 
21
  - zh
22
 
23
 
 
24
  multilinguality:
25
  - multilingual
26
 
 
32
  task_categories:
33
  - text-retrieval
34
 
35
+ license:
36
+ - apache-2.0
37
 
38
  task_ids:
39
+ - document-retrieval
40
+ ---
41
+
42
+ # Dataset Card for MIRACL Corpus
43
+
44
+
45
+ ## Dataset Description
46
+ * **Homepage:** http://miracl.ai
47
+ * **Repository:** https://github.com/project-miracl/miracl
48
+ * **Paper:** https://arxiv.org/abs/2210.09984
49
+
50
+ MIRACL πŸŒπŸ™ŒπŸŒ (Multilingual Information Retrieval Across a Continuum of Languages) is a multilingual retrieval dataset that focuses on search across 18 different languages, which collectively encompass over three billion native speakers around the world.
51
+
52
+ This dataset contains the collection data of the 16 "known languages". The remaining 2 "surprise languages" will not be released until later.
53
+
54
+ The corpus for each language is prepared from a Wikipedia dump, where we keep only the plain text and discard images, tables, etc. Each article is segmented into multiple passages using WikiExtractor based on natural discourse units (e.g., `\n\n` in the wiki markup). Each of these passages comprises a "document" or unit of retrieval. We preserve the Wikipedia article title of each passage.
55
+
56
+ ## Dataset Structure
57
+ Each retrieval unit contains three fields: `docid`, `title`, and `text`. Consider an example from the English corpus:
58
+
59
+ ```
60
+ {
61
+ "docid": "39#0",
62
+ "title": "Albedo",
63
+ "text": "Albedo (meaning 'whiteness') is the measure of the diffuse reflection of solar radiation out of the total solar radiation received by an astronomical body (e.g. a planet like Earth). It is dimensionless and measured on a scale from 0 (corresponding to a black body that absorbs all incident radiation) to 1 (corresponding to a body that reflects all incident radiation)."
64
+ }
65
+ ```
66
+ The `docid` has the schema `X#Y`, where all passages with the same `X` come from the same Wikipedia article, whereas `Y` denotes the passage within that article, numbered sequentially. The text field contains the text of the passage. The title field contains the name of the article the passage comes from.
67
+
68
+
69
+ The collection can be loaded using:
70
+ ```
71
+ lang='ar' # or any of the 16 languages
72
+ miracl_corpus = datasets.load_dataset('miracl/miracl-corpus', lang)['train']
73
+ for doc in miracl_corpus:
74
+ docid = doc['docid']
75
+ title = doc['title']
76
+ text = doc['text']
77
+ ```
78
+
79
+ ## Dataset Statistics and Links
80
+ The following table contains the number of passage and Wikipedia articles in the collection of each language, along with the links to the datasets and raw Wikipedia dumps.
81
+ | Language | # of Passages | # of Articles | Links | Raw Wiki Dump |
82
+ |:----------------|--------------:|--------------:|:------|:------|
83
+ | Arabic (ar) | 2,061,414 | 656,982 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-ar) | [🌏](https://archive.org/download/arwiki-20190201/arwiki-20190201-pages-articles-multistream.xml.bz2)
84
+ | Bengali (bn) | 297,265 | 63,762 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-bn) | [🌏](https://archive.org/download/bnwiki-20190201/bnwiki-20190201-pages-articles-multistream.xml.bz2)
85
+ | English (en) | 32,893,221 | 5,758,285 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-en) | [🌏](https://archive.org/download/enwiki-20190201/enwiki-20190201-pages-articles-multistream.xml.bz2)
86
+ | Spanish (es) | 10,373,953 | 1,669,181 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-es) | [🌏](https://archive.org/download/eswiki-20220301/eswiki-20220301-pages-articles-multistream.xml.bz2)
87
+ | Persian (fa) | 2,207,172 | 857,827 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-fa) | [🌏](https://archive.org/download/fawiki-20220301/fawiki-20220301-pages-articles-multistream.xml.bz2)
88
+ | Finnish (fi) | 1,883,509 | 447,815 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-fi) | [🌏](https://archive.org/download/fiwiki-20190201/fiwiki-20190201-pages-articles-multistream.xml.bz2)
89
+ | French (fr) | 14,636,953 | 2,325,608 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-fr) | [🌏](https://archive.org/download/frwiki-20220301/frwiki-20220301-pages-articles-multistream.xml.bz2)
90
+ | Hindi (hi) | 506,264 | 148,107 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-hi) | [🌏](https://archive.org/download/hiwiki-20220301/hiwiki-20220301-pages-articles-multistream.xml.bz2)
91
+ | Indonesian (id) | 1,446,315 | 446,330 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-id) | [🌏](https://archive.org/download/idwiki-20190201/idwiki-20190201-pages-articles-multistream.xml.bz2)
92
+ | Japanese (ja) | 6,953,614 | 1,133,444 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-ja) | [🌏](https://archive.org/download/jawiki-20190201/jawiki-20190201-pages-articles-multistream.xml.bz2)
93
+ | Korean (ko) | 1,486,752 | 437,373 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-ko) | [🌏](https://archive.org/download/kowiki-20190201/kowiki-20190201-pages-articles-multistream.xml.bz2)
94
+ | Russian (ru) | 9,543,918 | 1,476,045 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-ru) | [🌏](https://archive.org/download/ruwiki-20190201/ruwiki-20190201-pages-articles-multistream.xml.bz2)
95
+ | Swahili (sw) | 131,924 | 47,793 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-sw) | [🌏](https://archive.org/download/swwiki-20190201/swwiki-20190201-pages-articles-multistream.xml.bz2)
96
+ | Telugu (te) | 518,079 | 66,353 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-te) | [🌏](https://archive.org/download/tewiki-20190201/tewiki-20190201-pages-articles-multistream.xml.bz2)
97
+ | Thai (th) | 542,166 | 128,179 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-th) | [🌏](https://archive.org/download/thwiki-20190101/thwiki-20190101-pages-articles-multistream.xml.bz2)
98
+ | Chinese (zh) | 4,934,368 | 1,246,389 | [πŸ€—](https://huggingface.co/datasets/miracl/miracl-corpus/tree/main/miracl-corpus-v1.0-zh) | [🌏](https://archive.org/download/zhwiki-20220301/zhwiki-20220301-pages-articles-multistream.xml.bz2)
miracl-corpus.py CHANGED
@@ -49,7 +49,7 @@ _DESCRIPTION = 'dataset load script for MIRACL'
49
  _DATASET_URLS = {
50
  lang: {
51
  'train': [
52
- f'https://huggingface.co/datasets/MIRACL/miracl-corpus/resolve/main/miracl-corpus-v1.0-{lang}/docs-{i}.jsonl.gz' for i in range(n)
53
  ]
54
  } for lang, n in languages2filesize.items()
55
  }
@@ -58,7 +58,7 @@ _DATASET_URLS = {
58
  class MIRACLCorpus(datasets.GeneratorBasedBuilder):
59
  BUILDER_CONFIGS = [
60
  datasets.BuilderConfig(
61
- version=datasets.Version('1.1.0'),
62
  name=lang,
63
  description=f'MIRACL dataset in language {lang}.'
64
  ) for lang in languages2filesize
 
49
  _DATASET_URLS = {
50
  lang: {
51
  'train': [
52
+ f'https://huggingface.co/datasets/miracl/miracl-corpus/resolve/main/miracl-corpus-v1.0-{lang}/docs-{i}.jsonl.gz' for i in range(n)
53
  ]
54
  } for lang, n in languages2filesize.items()
55
  }
 
58
  class MIRACLCorpus(datasets.GeneratorBasedBuilder):
59
  BUILDER_CONFIGS = [
60
  datasets.BuilderConfig(
61
+ version=datasets.Version('1.0.0'),
62
  name=lang,
63
  description=f'MIRACL dataset in language {lang}.'
64
  ) for lang in languages2filesize