Datasets:

Multilinguality:
multilingual
Size Categories:
1M<n<10M
Language Creators:
found
Annotations Creators:
found
Source Datasets:
original
ArXiv:
Tags:
License:
adithya7 commited on
Commit
98aff4c
1 Parent(s): 4376b3e

add dataset script

Browse files
Files changed (2) hide show
  1. dataset_infos.json +0 -0
  2. xlel_wd.py +315 -0
dataset_infos.json ADDED
The diff for this file is too large to render. See raw diff
xlel_wd.py ADDED
@@ -0,0 +1,315 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+ # TODO: Address all TODOs and remove all explanatory comments
15
+ """TODO: Add a description here."""
16
+
17
+
18
+ import json
19
+
20
+ import datasets
21
+
22
+ _CITATION = """\
23
+ @article{pratapa-etal-2022-multilingual,
24
+ title = {Multilingual Event Linking to Wikidata},
25
+ author = {Pratapa, Adithya and Gupta, Rishubh and Mitamura, Teruko},
26
+ publisher = {arXiv},
27
+ year = {2022},
28
+ url = {https://arxiv.org/abs/2204.06535},
29
+ }
30
+ """
31
+
32
+ _DESCRIPTION = """\
33
+ XLEL-WD is a multilingual event linking dataset. \
34
+ This dataset contains mention references from multilingual Wikipedia/Wikinews articles to event items in Wikidata. \
35
+ The text descriptions for Wikidata events are compiled from Wikipedia articles.
36
+ """
37
+
38
+ _HOMEPAGE = "https://github.com/adithya7/xlel-wd"
39
+
40
+ _LICENSE = "CC-BY-4.0"
41
+
42
+ # TODO: Add link to the official dataset URLs here
43
+ # The HuggingFace Datasets library doesn't host the datasets but only points to the original files.
44
+ # This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
45
+ _URLS = {
46
+ "wikipedia-zero-shot": {
47
+ "train": "wikipedia.train.jsonl",
48
+ "dev": "wikipedia-zero-shot.dev.jsonl",
49
+ "test": "wikipedia-zero-shot.test.jsonl",
50
+ },
51
+ "wikinews-zero-shot": {"test": "wikinews-zero-shot.test.jsonl"},
52
+ "wikinews-cross-domain": {"test": "wikinews-cross-domain.test.jsonl"},
53
+ }
54
+
55
+ _WIKIPEDIA_ZERO_SHOT_LANGS = [
56
+ "af",
57
+ "ar",
58
+ "be",
59
+ "bg",
60
+ "bn",
61
+ "ca",
62
+ "cs",
63
+ "da",
64
+ "de",
65
+ "el",
66
+ "en",
67
+ "es",
68
+ "fa",
69
+ "fi",
70
+ "fr",
71
+ "he",
72
+ "hi",
73
+ "hu",
74
+ "id",
75
+ "it",
76
+ "ja",
77
+ "ko",
78
+ "ml",
79
+ "mr",
80
+ "ms",
81
+ "nl",
82
+ "no",
83
+ "pl",
84
+ "pt",
85
+ "ro",
86
+ "ru",
87
+ "si",
88
+ "sk",
89
+ "sl",
90
+ "sr",
91
+ "sv",
92
+ "sw",
93
+ "ta",
94
+ "te",
95
+ "th",
96
+ "tr",
97
+ "uk",
98
+ "vi",
99
+ "zh",
100
+ ]
101
+
102
+ _WIKINEWS_CROSS_DOMAIN_LANGS = [
103
+ "ar",
104
+ "bg",
105
+ "ca",
106
+ "cs",
107
+ "de",
108
+ "el",
109
+ "en",
110
+ "es",
111
+ "fi",
112
+ "fr",
113
+ "he",
114
+ "hu",
115
+ "it",
116
+ "ja",
117
+ "ko",
118
+ "nl",
119
+ "no",
120
+ "pl",
121
+ "pt",
122
+ "ro",
123
+ "ru",
124
+ "sr",
125
+ "sv",
126
+ "ta",
127
+ "tr",
128
+ "uk",
129
+ "zh",
130
+ ]
131
+
132
+ _WIKINEWS_ZERO_SHOT_LANGS = [
133
+ "ar",
134
+ "cs",
135
+ "de",
136
+ "en",
137
+ "es",
138
+ "fi",
139
+ "fr",
140
+ "it",
141
+ "ja",
142
+ "ko",
143
+ "nl",
144
+ "no",
145
+ "pl",
146
+ "pt",
147
+ "ru",
148
+ "sr",
149
+ "sv",
150
+ "ta",
151
+ "tr",
152
+ "uk",
153
+ "zh",
154
+ ]
155
+
156
+ _TASK_NAMES = []
157
+ _TASK_DESCRIPTIONS = []
158
+
159
+ # wikipedia based tasks
160
+ _TASK_NAMES += ["wikipedia-zero-shot"]
161
+ _TASK_DESCRIPTIONS += [
162
+ "This task requires linking mentions from multilingual wiki to Wikidata events (zero-shot evaluation)"
163
+ ]
164
+
165
+ for lang in _WIKIPEDIA_ZERO_SHOT_LANGS:
166
+ _TASK_NAMES += [f"wikipedia-zero-shot.{lang}"]
167
+ _TASK_DESCRIPTIONS += [
168
+ f"This task requires linking mentions from {lang}wiki to Wikidata events (zero-shot evaluation)."
169
+ ]
170
+
171
+ # wikinews based tasks (zero-shot)
172
+ _TASK_NAMES += ["wikinews-zero-shot"]
173
+ _TASK_DESCRIPTIONS += [
174
+ "This task requires linking mentions from multilingual wikinews to Wikidata events (zero-shot evaluation)."
175
+ ]
176
+ for lang in _WIKINEWS_ZERO_SHOT_LANGS:
177
+ _TASK_NAMES += [f"wikinews-zero-shot.{lang}"]
178
+ _TASK_DESCRIPTIONS += [
179
+ f"This task requires linking mentions from {lang}wikinews to Wikidata events (zero-shot evaluation)."
180
+ ]
181
+
182
+ # wikinews based tasks (cross-domain)
183
+ _TASK_NAMES += ["wikinews-cross-domain"]
184
+ _TASK_DESCRIPTIONS += [
185
+ "This task requires linking mentions from multilingual wikinews to Wikidata events (cross-domain evaluation)."
186
+ ]
187
+ for lang in _WIKINEWS_CROSS_DOMAIN_LANGS:
188
+ _TASK_NAMES += [f"wikinews-cross-domain.{lang}"]
189
+ _TASK_DESCRIPTIONS += [
190
+ f"This task requires linking mentions from {lang}wikinews to Wikidata events (cross-domain evaluation)."
191
+ ]
192
+
193
+
194
+ class XlelWdConfig(datasets.BuilderConfig):
195
+ """BuilderConfig for XLEL-WD"""
196
+
197
+ def __init__(self, features, citation, url, **kwargs) -> None:
198
+ super(XlelWdConfig, self).__init__(version=datasets.Version("1.0.0"), **kwargs)
199
+ self.features = features
200
+ self.citation = citation
201
+ self.url = url
202
+
203
+
204
+ class XlelWd(datasets.GeneratorBasedBuilder):
205
+ """A dataset for multilingual linking of event mentions to Wikidata."""
206
+
207
+ VERSION = datasets.Version("1.0.0")
208
+ # the features different slightly for Wikipedia and Wikinews
209
+ # Wikinews dataset also contains the article title and publication date
210
+ BUILDER_CONFIGS = [
211
+ XlelWdConfig(
212
+ name=name,
213
+ description=desc,
214
+ features=["mention", "context_left", "context_right", "context_lang"],
215
+ citation=_CITATION,
216
+ url=_URLS[name.split(".")[0]],
217
+ )
218
+ if name.startswith("wikipedia")
219
+ else XlelWdConfig(
220
+ name=name,
221
+ description=desc,
222
+ features=[
223
+ "mention",
224
+ "context_left",
225
+ "context_right",
226
+ "context_lang",
227
+ "context_title",
228
+ "context_date",
229
+ ],
230
+ citation=_CITATION,
231
+ url=_URLS[name.split(".")[0]],
232
+ )
233
+ for name, desc in zip(_TASK_NAMES, _TASK_DESCRIPTIONS)
234
+ ]
235
+
236
+ def _info(self):
237
+
238
+ features = {
239
+ feature: datasets.Value("string") for feature in self.config.features
240
+ }
241
+ features["label_id"] = datasets.Value("string")
242
+ return datasets.DatasetInfo(
243
+ description=_DESCRIPTION + self.config.description,
244
+ features=datasets.Features(features),
245
+ homepage=_HOMEPAGE,
246
+ license=_LICENSE,
247
+ citation=self.config.citation,
248
+ )
249
+
250
+ def _split_generators(self, dl_manager):
251
+
252
+ urls = _URLS[self.config.name.split(".")[0]]
253
+ downloaded_files = dl_manager.download_and_extract(urls)
254
+ if self.config.name.startswith("wikipedia"):
255
+ return [
256
+ datasets.SplitGenerator(
257
+ name=datasets.Split.TRAIN,
258
+ gen_kwargs={
259
+ "filepath": downloaded_files["train"],
260
+ "split": "train",
261
+ },
262
+ ),
263
+ datasets.SplitGenerator(
264
+ name=datasets.Split.VALIDATION,
265
+ gen_kwargs={
266
+ "filepath": downloaded_files["dev"],
267
+ "split": "dev",
268
+ },
269
+ ),
270
+ datasets.SplitGenerator(
271
+ name=datasets.Split.TEST,
272
+ gen_kwargs={
273
+ "filepath": downloaded_files["test"],
274
+ "split": "test",
275
+ },
276
+ ),
277
+ ]
278
+ elif self.config.name.startswith("wikinews"):
279
+ return [
280
+ datasets.SplitGenerator(
281
+ name=datasets.Split.TEST,
282
+ gen_kwargs={
283
+ "filepath": downloaded_files["test"],
284
+ "split": "test",
285
+ },
286
+ ),
287
+ ]
288
+
289
+ def _generate_examples(self, filepath, split):
290
+
291
+ task_domain, *task_langs = self.config.name.split(".")
292
+ with open(filepath, encoding="utf-8") as f:
293
+ for key, row in enumerate(f):
294
+ data = json.loads(row)
295
+ # generate mention references for the specified language
296
+ # if no language is specific in the config, return all
297
+ if len(task_langs) == 0 or task_langs[0] == data["context_lang"]:
298
+ if task_domain.startswith("wikipedia"):
299
+ yield key, {
300
+ "mention": data["mention"],
301
+ "context_left": data["context_left"],
302
+ "context_right": data["context_right"],
303
+ "context_lang": data["context_lang"],
304
+ "label_id": "" if split == "test" else data["label_id"],
305
+ }
306
+ elif task_domain.startswith("wikinews"):
307
+ yield key, {
308
+ "mention": data["mention"],
309
+ "context_left": data["context_left"],
310
+ "context_right": data["context_right"],
311
+ "context_lang": data["context_lang"],
312
+ "context_title": data["context_title"],
313
+ "context_date": data["context_date"],
314
+ "label_id": "" if split == "test" else data["label_id"],
315
+ }