izhx commited on
Commit
7e3723f
β€’
1 Parent(s): 085e125
Files changed (1) hide show
  1. README.md +210 -0
README.md CHANGED
@@ -1,3 +1,213 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  license: apache-2.0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - af
4
+ - ar
5
+ - az
6
+ - bg
7
+ - bn
8
+ - de
9
+ - el
10
+ - en
11
+ - es
12
+ - et
13
+ - eu
14
+ - fa
15
+ - fi
16
+ - fr
17
+ - gu
18
+ - he
19
+ - hi
20
+ - ht
21
+ - hu
22
+ - id
23
+ - it
24
+ - ja
25
+ - jv
26
+ - ka
27
+ - kk
28
+ - ko
29
+ - lt
30
+ - ml
31
+ - mr
32
+ - ms
33
+ - my
34
+ - nl
35
+ - pa
36
+ - pl
37
+ - pt
38
+ - qu
39
+ - ro
40
+ - ru
41
+ - sw
42
+ - ta
43
+ - te
44
+ - th
45
+ - tl
46
+ - tr
47
+ - uk
48
+ - ur
49
+ - vi
50
+ - wo
51
+ - yo
52
+ - zh
53
  license: apache-2.0
54
+ pretty_name: Mewsli-X
55
+ task_categories:
56
+ - text-retrieval
57
+ task_ids:
58
+ - entity-linking-retrieval
59
+ configs:
60
+ - config_name: wikipedia_pairs
61
+ data_files:
62
+ - split: train
63
+ path: wikipedia_pairs/train.jsonl.tar.gz
64
+ - split: validation
65
+ path: wikipedia_pairs/dev.jsonl.tar.gz
66
+ - config_name: ar
67
+ data_files:
68
+ - split: validation
69
+ path: wikinews_mentions/ar/dev.jsonl
70
+ - split: test
71
+ path: wikinews_mentions/ar/test.jsonl
72
+ - config_name: de
73
+ data_files:
74
+ - split: validation
75
+ path: wikinews_mentions/de/dev.jsonl
76
+ - split: test
77
+ path: wikinews_mentions/de/test.jsonl
78
+ - config_name: en
79
+ data_files:
80
+ - split: validation
81
+ path: wikinews_mentions/en/dev.jsonl
82
+ - split: test
83
+ path: wikinews_mentions/en/test.jsonl
84
+ - config_name: es
85
+ data_files:
86
+ - split: validation
87
+ path: wikinews_mentions/es/dev.jsonl
88
+ - split: test
89
+ path: wikinews_mentions/es/test.jsonl
90
+ - config_name: fa
91
+ data_files:
92
+ - split: validation
93
+ path: wikinews_mentions/fa/dev.jsonl
94
+ - split: test
95
+ path: wikinews_mentions/fa/test.jsonl
96
+ - config_name: ja
97
+ data_files:
98
+ - split: validation
99
+ path: wikinews_mentions/ja/dev.jsonl
100
+ - split: test
101
+ path: wikinews_mentions/ja/test.jsonl
102
+ - config_name: pl
103
+ data_files:
104
+ - split: validation
105
+ path: wikinews_mentions/pl/dev.jsonl
106
+ - split: test
107
+ path: wikinews_mentions/pl/test.jsonl
108
+ - config_name: ro
109
+ data_files:
110
+ - split: validation
111
+ path: wikinews_mentions/ro/dev.jsonl
112
+ - split: test
113
+ path: wikinews_mentions/ro/test.jsonl
114
+ - config_name: ta
115
+ data_files:
116
+ - split: validation
117
+ path: wikinews_mentions/ta/dev.jsonl
118
+ - split: test
119
+ path: wikinews_mentions/ta/test.jsonl
120
+ - config_name: tr
121
+ data_files:
122
+ - split: validation
123
+ path: wikinews_mentions/tr/dev.jsonl
124
+ - split: test
125
+ path: wikinews_mentions/tr/test.jsonl
126
+ - config_name: uk
127
+ data_files:
128
+ - split: validation
129
+ path: wikinews_mentions/uk/dev.jsonl
130
+ - split: test
131
+ path: wikinews_mentions/uk/test.jsonl
132
+ - config_name: candidate_entities
133
+ data_files:
134
+ - split: test
135
+ path: candidate_entities.jsonl.tar.gz
136
+ size_categories:
137
+ - 100K<n<1M
138
  ---
139
+
140
+ I generated the dataset following [mewsli-x.md#getting-started](https://github.com/google-research/google-research/blob/master/dense_representations_for_entity_retrieval/mel/mewsli-x.md#getting-started)
141
+ and converted into different parts (see [`process.py`](process.py)):
142
+ - ar/de/en/es/fa/ja/pl/ro/ta/tr/uk wikinews_mentions dev and test (from `wikinews_mentions-dev/test.jsonl`)
143
+ - candidate entities of 50 languages (from `candidate_set_entities.jsonl`)
144
+ - English wikipedia_pairs to fine-tune models (from `wikipedia_pairs-dev/train.jsonl`)
145
+
146
+ Raw data files are in [`raw.tar.gz`](raw.tar.gz), which contains:
147
+ ```
148
+ [...] 535M Feb 24 22:06 candidate_set_entities.jsonl
149
+ [...] 9.8M Feb 24 22:06 wikinews_mentions-dev.jsonl
150
+ [...] 35M Feb 24 22:06 wikinews_mentions-test.jsonl
151
+ [...] 24M Feb 24 22:06 wikipedia_pairs-dev.jsonl
152
+ [...] 283M Feb 24 22:06 wikipedia_pairs-train.jsonl
153
+ ```
154
+
155
+ **Below is from the original [readme](https://github.com/google-research/google-research/blob/master/dense_representations_for_entity_retrieval/mel/mewsli-x.md)**
156
+
157
+ # Mewsli-X
158
+
159
+ Mewsli-X is a multilingual dataset of entity mentions appearing in
160
+ [WikiNews](https://www.wikinews.org/) and
161
+ [Wikipedia](https://www.wikipedia.org/) articles, that have been automatically
162
+ linked to [WikiData](https://www.wikidata.org/) entries.
163
+
164
+ The primary use case is to evaluate transfer-learning in the zero-shot
165
+ cross-lingual setting of the
166
+ [XTREME-R benchmark suite](https://sites.research.google/xtremer):
167
+
168
+ 1. Fine-tune a pretrained model on English Wikipedia examples;
169
+ 2. Evaluate on WikiNews in other languages &mdash; **given an *entity mention*
170
+ in a WikiNews article, retrieve the correct *entity* from the predefined
171
+ candidate set by means of its textual description.**
172
+
173
+ Mewsli-X constitutes a *doubly zero-shot* task by construction: at test time, a
174
+ model has to contend with different languages and a different set of entities
175
+ from those observed during fine-tuning.
176
+
177
+ πŸ‘‰ For data examples and other editions of Mewsli, see [README.md](https://github.com/google-research/google-research/blob/master/dense_representations_for_entity_retrieval/mel/README.md).
178
+
179
+ πŸ‘‰ Consider submitting to the
180
+ **[XTREME-R leaderboard](https://sites.research.google/xtremer)**. The XTREME-R
181
+ [repository](https://github.com/google-research/xtreme) includes code for
182
+ getting started with training and evaluating a baseline model in PyTorch.
183
+
184
+ πŸ‘‰ Please cite this paper if you use the data/code in your work: *[XTREME-R:
185
+ Towards More Challenging and Nuanced Multilingual Evaluation (Ruder et al.,
186
+ 2021)](https://aclanthology.org/2021.emnlp-main.802.pdf)*.
187
+
188
+ > _**NOTE:** New evaluation results on Mewsli-X are **not** directly comparable to those reported in the paper because the dataset required further updates, as detailed [below](#updated-dataset). This does not affect the overall findings of the paper._
189
+
190
+ ```
191
+ @inproceedings{ruder-etal-2021-xtreme,
192
+ title = "{XTREME}-{R}: Towards More Challenging and Nuanced Multilingual Evaluation",
193
+ author = "Ruder, Sebastian and
194
+ Constant, Noah and
195
+ Botha, Jan and
196
+ Siddhant, Aditya and
197
+ Firat, Orhan and
198
+ Fu, Jinlan and
199
+ Liu, Pengfei and
200
+ Hu, Junjie and
201
+ Garrette, Dan and
202
+ Neubig, Graham and
203
+ Johnson, Melvin",
204
+ booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
205
+ month = nov,
206
+ year = "2021",
207
+ address = "Online and Punta Cana, Dominican Republic",
208
+ publisher = "Association for Computational Linguistics",
209
+ url = "https://aclanthology.org/2021.emnlp-main.802",
210
+ doi = "10.18653/v1/2021.emnlp-main.802",
211
+ pages = "10215--10245",
212
+ }
213
+ ```