system HF staff commited on
Commit
55d61b8
0 Parent(s):

Update files from the datasets library (from 1.11.0)

Browse files

Release notes: https://github.com/huggingface/datasets/releases/tag/1.11.0

.gitattributes ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bin.* filter=lfs diff=lfs merge=lfs -text
5
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.model filter=lfs diff=lfs merge=lfs -text
12
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
13
+ *.onnx filter=lfs diff=lfs merge=lfs -text
14
+ *.ot filter=lfs diff=lfs merge=lfs -text
15
+ *.parquet filter=lfs diff=lfs merge=lfs -text
16
+ *.pb filter=lfs diff=lfs merge=lfs -text
17
+ *.pt filter=lfs diff=lfs merge=lfs -text
18
+ *.pth filter=lfs diff=lfs merge=lfs -text
19
+ *.rar filter=lfs diff=lfs merge=lfs -text
20
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
21
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
22
+ *.tflite filter=lfs diff=lfs merge=lfs -text
23
+ *.tgz filter=lfs diff=lfs merge=lfs -text
24
+ *.xz filter=lfs diff=lfs merge=lfs -text
25
+ *.zip filter=lfs diff=lfs merge=lfs -text
26
+ *.zstandard filter=lfs diff=lfs merge=lfs -text
27
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,683 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: Russian SuperGLUE
3
+ annotations_creators:
4
+ - crowdsourced
5
+ - expert-generated
6
+ language_creators:
7
+ - crowdsourced
8
+ - expert-generated
9
+ languages:
10
+ - ru-RU
11
+ licenses:
12
+ - mit
13
+ multilinguality:
14
+ - monolingual
15
+ size_categories:
16
+ - 100K<n<1M
17
+ - 1M<n<10M
18
+ - 10M<n<100M
19
+ - 100M<n<1B
20
+ source_datasets:
21
+ - original
22
+ task_categories:
23
+ - text-classification
24
+ task_ids:
25
+ - natural-language-inference
26
+ - multi-class-classification
27
+ ---
28
+
29
+ # Dataset Card for [Russian SuperGLUE]
30
+
31
+ ## Table of Contents
32
+ - [Table of Contents](#table-of-contents)
33
+ - [Dataset Description](#dataset-description)
34
+ - [Dataset Summary](#dataset-summary)
35
+ - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
36
+ - [Languages](#languages)
37
+ - [Dataset Structure](#dataset-structure)
38
+ - [Data Instances](#data-instances)
39
+ - [Data Fields](#data-fields)
40
+ - [Data Splits](#data-splits)
41
+ - [Dataset Creation](#dataset-creation)
42
+ - [Curation Rationale](#curation-rationale)
43
+ - [Source Data](#source-data)
44
+ - [Annotations](#annotations)
45
+ - [Personal and Sensitive Information](#personal-and-sensitive-information)
46
+ - [Considerations for Using the Data](#considerations-for-using-the-data)
47
+ - [Social Impact of Dataset](#social-impact-of-dataset)
48
+ - [Discussion of Biases](#discussion-of-biases)
49
+ - [Other Known Limitations](#other-known-limitations)
50
+ - [Additional Information](#additional-information)
51
+ - [Dataset Curators](#dataset-curators)
52
+ - [Licensing Information](#licensing-information)
53
+ - [Citation Information](#citation-information)
54
+ - [Contributions](#contributions)
55
+
56
+ ## Dataset Description
57
+
58
+ - **Homepage:** https://russiansuperglue.com/
59
+ - **Repository:** https://github.com/RussianNLP/RussianSuperGLUE
60
+ - **Paper:** https://russiansuperglue.com/download/main_article
61
+ - **Leaderboard:** https://russiansuperglue.com/leaderboard/2
62
+ - **Point of Contact:** [More Information Needed]
63
+
64
+ ### Dataset Summary
65
+
66
+ Modern universal language models and transformers such as BERT, ELMo, XLNet, RoBERTa and others need to be properly
67
+ compared and evaluated. In the last year, new models and methods for pretraining and transfer learning have driven
68
+ striking performance improvements across a range of language understanding tasks.
69
+
70
+
71
+ We offer testing methodology based on tasks, typically proposed for “strong AI” — logic, commonsense, reasoning.
72
+ Adhering to the GLUE and SuperGLUE methodology, we present a set of test tasks for general language understanding
73
+ and leaderboard models.
74
+
75
+
76
+ For the first time a complete test for Russian language was developed, which is similar to its English analog.
77
+ Many datasets were composed for the first time, and a leaderboard of models for the Russian language with comparable
78
+ results is also presented.
79
+
80
+ ### Supported Tasks and Leaderboards
81
+
82
+ Supported tasks, barring a few additions, are equivalent to the original SuperGLUE tasks.
83
+
84
+ |Task Name|Equiv. to|
85
+ |----|---:|
86
+ |Linguistic Diagnostic for Russian|Broadcoverage Diagnostics (AX-b)|
87
+ |Russian Commitment Bank (RCB)|CommitmentBank (CB)|
88
+ |Choice of Plausible Alternatives for Russian language (PARus)|Choice of Plausible Alternatives (COPA)|
89
+ |Russian Multi-Sentence Reading Comprehension (MuSeRC)|Multi-Sentence Reading Comprehension (MultiRC)|
90
+ |Textual Entailment Recognition for Russian (TERRa)|Recognizing Textual Entailment (RTE)|
91
+ |Russian Words in Context (based on RUSSE)|Words in Context (WiC)|
92
+ |The Winograd Schema Challenge (Russian)|The Winograd Schema Challenge (WSC)|
93
+ |Yes/no Question Answering Dataset for the Russian (DaNetQA)|BoolQ|
94
+ |Russian Reading Comprehension with Commonsense Reasoning (RuCoS)|Reading Comprehension with Commonsense Reasoning (ReCoRD)|
95
+
96
+ ### Languages
97
+
98
+ All tasks are in Russian.
99
+
100
+ ## Dataset Structure
101
+
102
+ ### Data Instances
103
+
104
+ Note that there are no labels in the `test` splits. This is signified by the `-1` value.
105
+
106
+ #### LiDiRus
107
+
108
+ - **Size of downloaded dataset files:** 0.047 MB
109
+ - **Size of the generated dataset:** 0.47 MB
110
+ - **Total amount of disk used:** 0.517 MB
111
+
112
+ An example of 'test' looks as follows
113
+ ```
114
+ {
115
+ "sentence1": "Новая игровая консоль доступна по цене.",
116
+ "sentence2": "Новая игровая консоль недоступна по цене.",
117
+ "knowledge": "",
118
+ "lexical-semantics": "Morphological negation",
119
+ "logic": "Negation",
120
+ "predicate-argument-structure": "",
121
+ "idx": 10,
122
+ "label": 1
123
+ }
124
+ ```
125
+
126
+ #### RCB
127
+
128
+ - **Size of downloaded dataset files:** 0.134 MB
129
+ - **Size of the generated dataset:** 0.504 MB
130
+ - **Total amount of disk used:** 0.641 MB
131
+
132
+ An example of 'train'/'dev' looks as follows
133
+ ```
134
+ {
135
+ "premise": "— Пойдём пообедаем. Я с утра ничего не ел. Отель, как видишь, весьма посредственный, но мне сказали,
136
+ что в здешнем ресторане отлично готовят.",
137
+ "hypothesis": "В здешнем ресторане отлично готовят.",
138
+ "verb": "сказать",
139
+ "negation": "no_negation",
140
+ "idx": 10,
141
+ "label": 2
142
+ }
143
+ ```
144
+
145
+ An example of 'test' looks as follows
146
+ ```
147
+ {
148
+ "premise": "Я уверен, что вместе мы победим. Да, парламентское большинство думает иначе.",
149
+ "hypothesis": "Вместе мы проиграем.",
150
+ "verb": "думать",
151
+ "negation": "no_negation",
152
+ "idx": 10,
153
+ "label": -1
154
+ }
155
+ ```
156
+
157
+ #### PARus
158
+
159
+ - **Size of downloaded dataset files:** 0.057 MB
160
+ - **Size of the generated dataset:** 0.187 MB
161
+ - **Total amount of disk used:** 0.245 MB
162
+
163
+ An example of 'train'/'dev' looks as follows
164
+ ```
165
+ {
166
+ "premise": "Женщина чинила кран.",
167
+ "choice1": "Кран подтекал.",
168
+ "choice2": "Кран был выключен.",
169
+ "question": "cause",
170
+ "idx": 10,
171
+ "label": 0
172
+ }
173
+ ```
174
+
175
+ An example of 'test' looks as follows
176
+ ```
177
+ {
178
+ "premise": "Ребятам было страшно.",
179
+ "choice1": "Их вожатый рассказал им историю про призрака.",
180
+ "choice2": "Они жарили маршмеллоу на костре.",
181
+ "question": "cause",
182
+ "idx": 10,
183
+ "label": -1
184
+ }
185
+ ```
186
+ #### MuSeRC
187
+
188
+ - **Size of downloaded dataset files:** 1.2 MB
189
+ - **Size of the generated dataset:** 57 MB
190
+ - **Total amount of disk used:** 59 MB
191
+
192
+ An example of 'train'/'dev' looks as follows
193
+ ```
194
+ {
195
+ "paragraph": "(1) Но люди не могут существовать без природы, поэтому в парке стояли железобетонные скамейки —
196
+ деревянные моментально ломали. (2) В парке бегали ребятишки, водилась шпана, которая развлекалась игрой в карты,
197
+ пьянкой, драками, «иногда насмерть». (3) «Имали они тут и девок...» (4) Верховодил шпаной Артемка-мыло, с
198
+ вспененной белой головой. (5) Людочка сколько ни пыталась усмирить лохмотья на буйной голове Артемки, ничего у
199
+ неё не получалось. (6) Его «кудри, издали напоминавшие мыльную пену, изблизя оказались что липкие рожки из
200
+ вокзальной столовой — сварили их, бросили комком в пустую тарелку, так они, слипшиеся, неподъёмно и лежали.
201
+ (7) Да и не ради причёски приходил парень к Людочке. (8) Как только её руки становились занятыми ножницами
202
+ и расчёской, Артемка начинал хватать её за разные места. (9) Людочка сначала увёртывалась от хватких рук Артемки,
203
+ а когда не помогло, стукнула его машинкой по голове и пробила до крови, пришлось лить йод на голову «ухажористого
204
+ человека». (10) Артемка заулюлюкал и со свистом стал ловить воздух. (11) С тех пор «домогания свои хулиганские
205
+ прекратил», более того, шпане повелел Людочку не трогать.",
206
+ "question": "Как развлекались в парке ребята?",
207
+ "answer": "Развлекались игрой в карты, пьянкой, драками, снимали они тут и девок.",
208
+ "idx":
209
+ {
210
+ "paragraph": 0,
211
+ "question": 2,
212
+ "answer": 10
213
+ },
214
+ "label": 1
215
+ }
216
+ ```
217
+
218
+ An example of 'test' looks as follows
219
+ ```
220
+
221
+ {
222
+ "paragraph": "\"(1) Издательство Viking Press совместно с компанией TradeMobile выпустят мобильное приложение,
223
+ посвященное Анне Франк, передает The Daily Telegraph. (2) Программа будет включать в себя фрагменты из дневника
224
+ Анны, озвученные британской актрисой Хеленой Бонэм Картер. (3) Помимо этого, в приложение войдут фотографии
225
+ и видеозаписи, документы из архива Фонда Анны Франк, план здания в Амстердаме, где Анна с семьей скрывались от
226
+ нацистов, и факсимильные копии страниц дневника. (4) Приложение, которое получит название Anne Frank App, выйдет
227
+ 18 октября. (5) Интерфейс программы будет англоязычным. (6) На каких платформах будет доступно Anne Frank App,
228
+ не уточняется. Анна Франк родилась в Германии в 1929 году. (7) Когда в стране начались гонения на евреев, Анна с
229
+ семьей перебрались в Нидерланды. (8) С 1942 года члены семьи Франк и еще несколько человек скрывались от нацистов
230
+ в потайных комнатах дома в Амстердаме, который занимала компания отца Анны. (9) В 1944 году группу по доносу
231
+ обнаружили гестаповцы. (10) Обитатели \"Убежища\" (так Анна называла дом в дневнике) были отправлены в концлагеря;
232
+ выжить удалось только отцу девочки Отто Франку. (11) Находясь в \"Убежище\", Анна вела дневник, в котором описывала
233
+ свою жизнь и жизнь своих близких. (12) После ареста книгу с записями сохранила подруга семьи Франк и впоследствии
234
+ передала ее отцу Анны. (13) Дневник был впервые опубликован в 1947 году. (14) Сейчас он переведен более
235
+ чем на 60 языков.\"",
236
+ "question": "Какая информация войдет в новой мобильное приложение?",
237
+ "answer": "Видеозаписи Анны Франк.",
238
+ "idx":
239
+ {
240
+ "paragraph": 0,
241
+ "question": 2,
242
+ "answer": 10
243
+ },
244
+ "label": -1
245
+ }
246
+ ```
247
+ #### TERRa
248
+
249
+ - **Size of downloaded dataset files:** 0.887 MB
250
+ - **Size of the generated dataset:** 3.28 MB
251
+ - **Total amount of disk used:** 4.19 MB
252
+
253
+ An example of 'train'/'dev' looks as follows
254
+ ```
255
+ {
256
+ "premise": "Музей, расположенный в Королевских воротах, меняет экспозицию. На смену выставке, рассказывающей об
257
+ истории ворот и их реставрации, придет «Аптека трех королей». Как рассказали в музее, посетители попадут в
258
+ традиционный интерьер аптеки.",
259
+ "hypothesis": "Музей закроется навсегда.",
260
+ "idx": 10,
261
+ "label": 1
262
+ }
263
+ ```
264
+
265
+ An example of 'test' looks as follows
266
+ ```
267
+ {
268
+ "premise": "Маршрутка полыхала несколько минут. Свидетели утверждают, что приезду пожарных салон «Газели» выгорел полностью. К счастью, пассажиров внутри не было, а водитель успел выскочить из кабины.",
269
+ "hypothesis": "Маршрутка выгорела.",
270
+ "idx": 10,
271
+ "label": -1
272
+ }
273
+ ```
274
+
275
+ #### RUSSE
276
+
277
+ - **Size of downloaded dataset files:** 3.7 MB
278
+ - **Size of the generated dataset:** 20 MB
279
+ - **Total amount of disk used:** 24 MB
280
+
281
+ An example of 'train'/'dev' looks as follows
282
+ ```
283
+ {
284
+ "word": "дух",
285
+ "sentence1": "Завертелась в доме веселая коловерть: праздничный стол, праздничный дух, шумные разговоры",
286
+ "sentence2": "Вижу: духи собралися / Средь белеющих равнин. // Бесконечны, безобразны, / В мутной месяца игре / Закружились бесы разны, / Будто листья в ноябре",
287
+ "start1": 68,
288
+ "start2": 6,
289
+ "end1": 72,
290
+ "end2": 11,
291
+ "gold_sense1": 3,
292
+ "gold_sense2": 4,
293
+ "idx": 10,
294
+ "label": 0
295
+ }
296
+ ```
297
+
298
+ An example of 'test' looks as follows
299
+ ```
300
+ {
301
+ "word": "доска",
302
+ "sentence1": "На 40-й день после трагедии в переходе была установлена мемориальная доска, надпись на которой гласит: «В память о погибших и пострадавших от террористического акта 8 августа 2000 года».",
303
+ "sentence2": "Фото с 36-летним миллиардером привлекло сеть его необычной фигурой при стойке на доске и кремом на лице.",
304
+ "start1": 69,
305
+ "start2": 81,
306
+ "end1": 73,
307
+ "end2": 85,
308
+ "gold_sense1": -1,
309
+ "gold_sense2": -1,
310
+ "idx": 10,
311
+ "label": -1
312
+ }
313
+ ```
314
+
315
+ #### RWSD
316
+
317
+ - **Size of downloaded dataset files:** 0.04 MB
318
+ - **Size of the generated dataset:** 0.279 MB
319
+ - **Total amount of disk used:** 0.320 MB
320
+
321
+ An example of 'train'/'dev' looks as follows
322
+ ```
323
+ {
324
+ "text": "Женя поблагодарила Сашу за помощь, которую она оказала.",
325
+ "span1_index": 0,
326
+ "span2_index": 6,
327
+ "span1_text": "Ж��ня",
328
+ "span2_text": "она оказала",
329
+ "idx": 10,
330
+ "label": 0
331
+ }
332
+ ```
333
+
334
+ An example of 'test' looks as follows
335
+ ```
336
+ {
337
+ "text": "Мод и Дора видели, как через прерию несутся поезда, из двигателей тянулись клубы черного дыма. Ревущие
338
+ звуки их моторов и дикие, яростные свистки можно было услышать издалека. Лошади убежали, когда они увидели
339
+ приближающийся поезд.",
340
+ "span1_index": 22,
341
+ "span2_index": 30,
342
+ "span1_text": "свистки",
343
+ "span2_text": "они увидели",
344
+ "idx": 10,
345
+ "label": -1
346
+ }
347
+ ```
348
+
349
+ #### DaNetQA
350
+
351
+ - **Size of downloaded dataset files:** 1.3 MB
352
+ - **Size of the generated dataset:** 4.6 MB
353
+ - **Total amount of disk used:** 5.9 MB
354
+
355
+ An example of 'train'/'dev' looks as follows
356
+ ```
357
+ {
358
+ "question": "Вреден ли алкоголь на первых неделях беременности?",
359
+ "passage": "А Бакингем-Хоуз и её коллеги суммировали последствия, найденные в обзорных статьях ранее. Частые случаи
360
+ задержки роста плода, результатом чего является укороченный средний срок беременности и сниженный вес при рождении.
361
+ По сравнению с нормальными детьми, дети 3-4-недельного возраста демонстрируют «менее оптимальную» двигательную
362
+ активность, рефлексы, и ориентацию в пространстве, а дети 4-6 лет показывают низкий уровень работы
363
+ нейроповеденческих функций, внимания, эмоциональной экспрессии, и развития речи и языка. Величина этих влияний
364
+ часто небольшая, частично в связи с независимыми переменными: включая употребление во время беременности
365
+ алкоголя/табака, а также факторы среды . У детей школьного возраста проблемы с устойчивым вниманием и контролем
366
+ своего поведения, а также незначительные с ростом, познавательными и языковыми способностями.",
367
+ "idx": 10,
368
+ "label": 1
369
+ }
370
+ ```
371
+
372
+ An example of 'test' looks as follows
373
+ ```
374
+ {
375
+ "question": "Вредна ли жесткая вода?",
376
+ "passage": "Различают временную жёсткость, обусловленную гидрокарбонатами кальция и магния Са2; Mg2, и постоянную
377
+ жёсткость, вызванную присутствием других солей, не выделяющихся при кипячении воды: в основном, сульфатов и
378
+ хлоридов Са и Mg. Жёсткая вода при умывании сушит кожу, в ней плохо образуется пена при использовании мыла.
379
+ Использование жёсткой воды вызывает появление осадка на стенках котлов, в трубах и т. п. В то же время,
380
+ использование слишком мягкой воды может приводить к коррозии труб, так как, в этом случае отсутствует
381
+ кислотно-щелочная буферность, которую обеспечивает гидрокарбонатная жёсткость. Потребление жёсткой или мягкой
382
+ воды обычно не является опасным для здоровья, однако есть данные о том, что высокая жёсткость способствует
383
+ образованию мочевых камней, а низкая — незначительно увеличивает риск сердечно-сосудистых заболеваний. Вкус
384
+ природной питьевой воды, например, воды родников, обусловлен именно присутствием солей жёсткости.",
385
+ "idx": 100,
386
+ "label": -1
387
+ }
388
+ ```
389
+
390
+ #### RuCoS
391
+
392
+ - **Size of downloaded dataset files:** 54 MB
393
+ - **Size of the generated dataset:** 193 MB
394
+ - **Total amount of disk used:** 249 MB
395
+
396
+ An example of 'train'/'dev' looks as follows
397
+ ```
398
+ {
399
+ "passage": "В Абхазии 24 августа на досрочных выборах выбирают нового президента. Кто бы ни стал победителем,
400
+ возможности его будут ограничены, говорят эксперты, опрошенные DW. В Абхазии 24 августа проходят досрочные выборы
401
+ президента не признанной международным сообществом республики. Толчком к их проведению стали массовые протесты в
402
+ конце мая 2014 года, в результате которых со своего поста был вынужден уйти действующий президент Абхазии Александр
403
+ Анкваб. Эксперты называют среди наиболее перспективных кандидатов находящегося в оппозиции политика Рауля Хаджимбу,
404
+ экс-главу службы безопасности Аслана Бжанию и генерала Мираба Кишмарию, исполняющего обязанности министра обороны.
405
+ У кого больше шансов\n\"Ставки делаются на победу Хаджимбы.\n@highlight\nВ Швеции задержаны двое граждан РФ в связи
406
+ с нападением на чеченского блогера\n@highlight\nТуризм в эпоху коронавируса: куда поехать? И ехать ли
407
+ вообще?\n@highlight\nКомментарий: Россия накануне эпидемии - виноватые назначены заранее",
408
+ "query": "Несмотря на то, что Кремль вложил много денег как в @placeholder, так и в Южную Осетию, об экономическом
409
+ восстановлении данных регионов говорить не приходится, считает Хальбах: \"Многие по-прежнему живут в
410
+ полуразрушенных домах и временных жилищах\".",
411
+ "entities":
412
+ [
413
+ "DW.",
414
+ "Абхазии ",
415
+ "Александр Анкваб.",
416
+ "Аслана Бжанию ",
417
+ "Мираба Кишмарию,",
418
+ "РФ ",
419
+ "Рауля Хаджимбу,",
420
+ "Россия ",
421
+ "Хаджимбы.",
422
+ "Швеции "
423
+ ],
424
+ "answers":
425
+ [
426
+ "Абхазии"
427
+ ],
428
+ "idx":
429
+ {
430
+ "passage": 500,
431
+ "query": 500
432
+ }
433
+ }
434
+ ```
435
+
436
+ An example of 'test' looks as follows
437
+ ```
438
+ {
439
+ "passage": "Почему и как изменится курс белорусского рубля? Какие инструменты следует предпочесть населению, чтобы
440
+ сохранить сбережения, DW рассказали финансовые аналитики Беларуси. На последних валютных торгах БВФБ 2015 года в
441
+ среду, 30 декабря, курс белорусского рубля к доллару - 18569, к евро - 20300, к российскому рублю - 255. В 2016
442
+ году белорусскому рублю пророчат падение как минимум на 12 процентов к корзине валют, к которой привязан его курс.
443
+ А чтобы избежать потерь, белорусам советуют диверсифицировать инвестиционные портфели. Чем обусловлены прогнозные
444
+ изменения котировок белорусского рубля, и какие финансовые инструменты стоит предпочесть, чтобы минимизировать риск
445
+ потерь?\n@highlight\nВ Германии за сутки выявлено более 100 новых заражений коронавирусом\n@highlight\nРыночные цены
446
+ на нефть рухнули из-за провала переговоров ОПЕК+\n@highlight\nВ Италии за сутки произошел резкий скачок смертей от
447
+ COVID-19",
448
+ "query": "Последнее, убежден аналитик, инструмент для узкого круга профессиональных инвесторов, культуры следить за
449
+ финансовым состоянием предприятий - такой, чтобы играть на рынке корпоративных облигаций, - в @placeholder пока нет.",
450
+ "entities":
451
+ [
452
+ "DW ",
453
+ "Беларуси.",
454
+ "Германии ",
455
+ "Италии ",
456
+ "ОПЕК+"
457
+ ],
458
+ "answers": [],
459
+ "idx":
460
+ {
461
+ "passage": 500,
462
+ "query": 500
463
+ }
464
+ }
465
+ ```
466
+
467
+ ### Data Fields
468
+
469
+ #### LiDiRus
470
+
471
+ - `idx`: an `int32` feature
472
+ - `label`: a classification label, with possible values `entailment` (0), `not_entailment` (1)
473
+ - `sentence1`: a `string` feature
474
+ - `sentence2`: a `string` feature
475
+ - `knowledge`: a `string` feature with possible values `''`, `'World knowledge'`, `'Common sense'`
476
+ - `lexical-semantics`: a `string` feature
477
+ - `logic`: a `string` feature
478
+ - `predicate-argument-structure`: a `string` feature
479
+
480
+
481
+ #### RCB
482
+
483
+ - `idx`: an `int32` feature
484
+ - `label`: a classification label, with possible values `entailment` (0), `contradiction` (1), `neutral` (2)
485
+ - `premise`: a `string` feature
486
+ - `hypothesis`: a `string` feature
487
+ - `verb`: a `string` feature
488
+ - `negation`: a `string` feature with possible values `'no_negation'`, `'negation'`, `''`, `'double_negation'`
489
+
490
+ #### PARus
491
+
492
+ - `idx`: an `int32` feature
493
+ - `label`: a classification label, with possible values `choice1` (0), `choice2` (1)
494
+ - `premise`: a `string` feature
495
+ - `choice1`: a `string` feature
496
+ - `choice2`: a `string` feature
497
+ - `question`: a `string` feature with possible values `'cause'`, `'effect'`
498
+
499
+ #### MuSeRC
500
+ - `idx`: an `int32` feature
501
+ - `label` : a classification label, with possible values `false` (0) , `true` (1) (does the provided `answer` contain
502
+ a factual response to the `question`)
503
+ - `paragraph`: a `string` feature
504
+ - `question`: a `string` feature
505
+ - `answer`: a `string` feature
506
+
507
+
508
+ #### TERRa
509
+ - `idx`: an `int32` feature
510
+ - `label`: a classification label, with possible values `entailment` (0), `not_entailment` (1)
511
+ - `premise`: a `string` feature
512
+ - `hypothesis`: a `string` feature
513
+
514
+ #### RUSSE
515
+ - `idx`: an `int32` feature
516
+ - `label` : a classification label, with possible values `false` (0), `true` (1) (whether the given `word` used in the
517
+ same sense in both sentences)
518
+ - `word`: a `string` feature
519
+ - `sentence1`: a `string` feature
520
+ - `sentence2`: a `string` feature
521
+ - `gold_sense1`: an `int32` feature
522
+ - `gold_sense2`: an `int32` feature
523
+ - `start1`: an `int32` feature
524
+ - `start2`: an `int32` feature
525
+ - `end1`: an `int32` feature
526
+ - `end2`: an `int32` feature
527
+
528
+ #### RWSD
529
+
530
+ - `idx`: an `int32` feature
531
+ - `label` : a classification label, with possible values `false` (0), `true` (1) (whether the given spans are
532
+ coreferential)
533
+ - `text`: a `string` feature
534
+ - `span1_index`: an `int32` feature
535
+ - `span2_index`: an `int32` feature
536
+ - `span1_text`: a `string` feature
537
+ - `span2_text`: a `string` feature
538
+
539
+
540
+ #### DaNetQA
541
+ - `idx`: an `int32` feature
542
+ - `label` : a classification label, with possible values `false` (0), `true` (1) (yes/no answer to the `question` found
543
+ in the `passage`)
544
+ - `question`: a `string` feature
545
+ - `passage`: a `string` feature
546
+
547
+ #### RuCoS
548
+
549
+ - `idx`: an `int32` feature
550
+ - `passage`: a `string` feature
551
+ - `query`: a `string` feature
552
+ - `entities`: a `list of strings` feature
553
+ - `answers`: a `list of strings` feature
554
+
555
+
556
+ [More Information Needed]
557
+
558
+ ### Data Splits
559
+
560
+ #### LiDiRus
561
+ | |test|
562
+ |---|---:|
563
+ |LiDiRus|1104|
564
+
565
+ #### RCB
566
+
567
+ | |train|validation|test|
568
+ |----|---:|----:|---:|
569
+ |RCB|438|220|438|
570
+
571
+ #### PARus
572
+
573
+ | |train|validation|test|
574
+ |----|---:|----:|---:|
575
+ |PARus|400|100|500|
576
+
577
+ #### MuSeRC
578
+
579
+ | |train|validation|test|
580
+ |----|---:|----:|---:|
581
+ |MuSeRC|500|100|322|
582
+
583
+
584
+ #### TERRa
585
+
586
+ | |train|validation|test|
587
+ |----|---:|----:|---:|
588
+ |TERRa|2616|307|3198|
589
+
590
+
591
+ #### RUSSE
592
+
593
+ | |train|validation|test|
594
+ |----|---:|----:|---:|
595
+ |RUSSE|19845|8508|18892|
596
+
597
+
598
+ #### RWSD
599
+
600
+ | |train|validation|test|
601
+ |----|---:|----:|---:|
602
+ |RWSD|606|204|154|
603
+
604
+
605
+ #### DaNetQA
606
+
607
+ | |train|validation|test|
608
+ |----|---:|----:|---:|
609
+ |DaNetQA|1749|821|805|
610
+
611
+
612
+ #### RuCoS
613
+
614
+ | |train|validation|test|
615
+ |----|---:|----:|---:|
616
+ |RuCoS|72193|7577|7257|
617
+
618
+ ## Dataset Creation
619
+
620
+ ### Curation Rationale
621
+
622
+ [More Information Needed]
623
+
624
+ ### Source Data
625
+
626
+ #### Initial Data Collection and Normalization
627
+
628
+ [More Information Needed]
629
+
630
+ #### Who are the source language producers?
631
+
632
+ [More Information Needed]
633
+
634
+ ### Annotations
635
+
636
+ #### Annotation process
637
+
638
+ [More Information Needed]
639
+
640
+ #### Who are the annotators?
641
+
642
+ [More Information Needed]
643
+
644
+ ### Personal and Sensitive Information
645
+
646
+ [More Information Needed]
647
+
648
+ ## Considerations for Using the Data
649
+
650
+ ### Social Impact of Dataset
651
+
652
+ [More Information Needed]
653
+
654
+ ### Discussion of Biases
655
+
656
+ [More Information Needed]
657
+
658
+ ### Other Known Limitations
659
+
660
+ [More Information Needed]
661
+
662
+ ## Additional Information
663
+
664
+ ### Dataset Curators
665
+
666
+ [More Information Needed]
667
+
668
+ ### Licensing Information
669
+
670
+ All our datasets are published by MIT License.
671
+
672
+ ### Citation Information
673
+ ```
674
+ @article{shavrina2020russiansuperglue,
675
+ title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},
676
+ author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova, Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and Evlampiev, Andrey},
677
+ journal={arXiv preprint arXiv:2010.15925},
678
+ year={2020}
679
+ }
680
+ ```
681
+ ### Contributions
682
+
683
+ Thanks to [@slowwavesleep](https://github.com/slowwavesleep) for adding this dataset.
dataset_infos.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"lidirus": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\n\"LiDiRus (Linguistic Diagnostic for Russian) is a diagnostic dataset that covers a large volume of linguistic phenomena,\nwhile allowing you to evaluate information systems on a simple test of textual entailment recognition.\nSee more details diagnostics.\n", "citation": "\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/LiDiRus", "license": "", "features": {"sentence1": {"dtype": "string", "id": null, "_type": "Value"}, "sentence2": {"dtype": "string", "id": null, "_type": "Value"}, "knowledge": {"dtype": "string", "id": null, "_type": "Value"}, "lexical-semantics": {"dtype": "string", "id": null, "_type": "Value"}, "logic": {"dtype": "string", "id": null, "_type": "Value"}, "predicate-argument-structure": {"dtype": "string", "id": null, "_type": "Value"}, "idx": {"dtype": "int32", "id": null, "_type": "Value"}, "label": {"num_classes": 2, "names": ["entailment", "not_entailment"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "lidirus", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"test": {"name": "test", "num_bytes": 470306, "num_examples": 1104, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/LiDiRus": {"num_bytes": 47118, "checksum": "e1bc2c55cadbc98b0e77a43e1560af8e2fe336ab4bc372f545f69cb48a479cdc"}}, "download_size": 47118, "post_processing_size": null, "dataset_size": 470306, "size_in_bytes": 517424}, "rcb": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\nThe Russian Commitment Bank is a corpus of naturally occurring discourses whose final sentence contains\na clause-embedding predicate under an entailment canceling operator (question, modal, negation, antecedent\nof conditional).\n", "citation": "\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/RCB", "license": "", "features": {"premise": {"dtype": "string", "id": null, "_type": "Value"}, "hypothesis": {"dtype": "string", "id": null, "_type": "Value"}, "verb": {"dtype": "string", "id": null, "_type": "Value"}, "negation": {"dtype": "string", "id": null, "_type": "Value"}, "idx": {"dtype": "int32", "id": null, "_type": "Value"}, "label": {"num_classes": 3, "names": ["entailment", "contradiction", "neutral"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "rcb", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 199712, "num_examples": 438, "dataset_name": "russian_super_glue"}, "validation": {"name": "validation", "num_bytes": 97993, "num_examples": 220, "dataset_name": "russian_super_glue"}, "test": {"name": "test", "num_bytes": 207031, "num_examples": 438, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/RCB": {"num_bytes": 136700, "checksum": "9ecc0bd0cc04e04922349212452166404b5a75b67be9e6aa996c80c64beb6ddb"}}, "download_size": 136700, "post_processing_size": null, "dataset_size": 504736, "size_in_bytes": 641436}, "parus": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\nChoice of Plausible Alternatives for Russian language\nChoice of Plausible Alternatives for Russian language (PARus) evaluation provides researchers with a tool for assessing\nprogress in open-domain commonsense causal reasoning. Each question in PARus is composed of a premise and two\nalternatives, where the task is to select the alternative that more plausibly has a causal relation with the premise.\nThe correct alternative is randomized so that the expected performance of randomly guessing is 50%.\n", "citation": "\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/PARus", "license": "", "features": {"premise": {"dtype": "string", "id": null, "_type": "Value"}, "choice1": {"dtype": "string", "id": null, "_type": "Value"}, "choice2": {"dtype": "string", "id": null, "_type": "Value"}, "question": {"dtype": "string", "id": null, "_type": "Value"}, "idx": {"dtype": "int32", "id": null, "_type": "Value"}, "label": {"num_classes": 2, "names": ["choice1", "choice2"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "parus", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 74467, "num_examples": 400, "dataset_name": "russian_super_glue"}, "validation": {"name": "validation", "num_bytes": 19397, "num_examples": 100, "dataset_name": "russian_super_glue"}, "test": {"name": "test", "num_bytes": 93192, "num_examples": 500, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/PARus": {"num_bytes": 57585, "checksum": "7093c859a6ab07eab54c86cdd686b7b14afffee46abcd3e2d43876b1fdb8fa59"}}, "download_size": 57585, "post_processing_size": null, "dataset_size": 187056, "size_in_bytes": 244641}, "muserc": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\nWe present a reading comprehension challenge in which questions can only be answered by taking into account information\nfrom multiple sentences. The dataset is the first to study multi-sentence inference at scale, with an open-ended set of\nquestion types that requires reasoning skills.\n", "citation": "@inproceedings{fenogenova-etal-2020-read,\n title = \"Read and Reason with {M}u{S}e{RC} and {R}u{C}o{S}: Datasets for Machine Reading Comprehension for {R}ussian\",\n author = \"Fenogenova, Alena and\n Mikhailov, Vladislav and\n Shevelev, Denis\",\n booktitle = \"Proceedings of the 28th International Conference on Computational Linguistics\",\n month = dec,\n year = \"2020\",\n address = \"Barcelona, Spain (Online)\",\n publisher = \"International Committee on Computational Linguistics\",\n url = \"https://aclanthology.org/2020.coling-main.570\",\n doi = \"10.18653/v1/2020.coling-main.570\",\n pages = \"6481--6497\",\n abstract = \"The paper introduces two Russian machine reading comprehension (MRC) datasets, called MuSeRC and RuCoS,\n which require reasoning over multiple sentences and commonsense knowledge to infer the answer. The former follows\n the design of MultiRC, while the latter is a counterpart of the ReCoRD dataset. The datasets are included\n in RussianSuperGLUE, the Russian general language understanding benchmark. We provide a comparative analysis\n and demonstrate that the proposed tasks are relatively more complex as compared to the original ones for English.\n Besides, performance results of human solvers and BERT-based models show that MuSeRC and RuCoS represent a challenge\n for recent advanced neural models. We thus hope to facilitate research in the field of MRC for Russian and prompt\n the study of multi-hop reasoning in a cross-lingual scenario.\",\n}\n\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/MuSeRC", "license": "", "features": {"paragraph": {"dtype": "string", "id": null, "_type": "Value"}, "question": {"dtype": "string", "id": null, "_type": "Value"}, "answer": {"dtype": "string", "id": null, "_type": "Value"}, "idx": {"paragraph": {"dtype": "int32", "id": null, "_type": "Value"}, "question": {"dtype": "int32", "id": null, "_type": "Value"}, "answer": {"dtype": "int32", "id": null, "_type": "Value"}}, "label": {"num_classes": 2, "names": ["False", "True"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "muserc", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 31651155, "num_examples": 11950, "dataset_name": "russian_super_glue"}, "validation": {"name": "validation", "num_bytes": 5964157, "num_examples": 2235, "dataset_name": "russian_super_glue"}, "test": {"name": "test", "num_bytes": 19850930, "num_examples": 7614, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/MuSeRC": {"num_bytes": 1196720, "checksum": "679d69df99113d8ca6416eda5583a81ab324f5438097216a416d09e745d2f668"}}, "download_size": 1196720, "post_processing_size": null, "dataset_size": 57466242, "size_in_bytes": 58662962}, "terra": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\nTextual Entailment Recognition has been proposed recently as a generic task that captures major semantic inference\nneeds across many NLP applications, such as Question Answering, Information Retrieval, Information Extraction,\nand Text Summarization. This task requires to recognize, given two text fragments, whether the meaning of one text is\nentailed (can be inferred) from the other text.\n", "citation": "\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/TERRa", "license": "", "features": {"premise": {"dtype": "string", "id": null, "_type": "Value"}, "hypothesis": {"dtype": "string", "id": null, "_type": "Value"}, "idx": {"dtype": "int32", "id": null, "_type": "Value"}, "label": {"num_classes": 2, "names": ["entailment", "not_entailment"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "terra", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 1409243, "num_examples": 2616, "dataset_name": "russian_super_glue"}, "validation": {"name": "validation", "num_bytes": 161485, "num_examples": 307, "dataset_name": "russian_super_glue"}, "test": {"name": "test", "num_bytes": 1713499, "num_examples": 3198, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/TERRa": {"num_bytes": 907346, "checksum": "fc7320210b5b6f7087615f13558868de55f46d5e0e365d9d82968c66e6e0dba7"}}, "download_size": 907346, "post_processing_size": null, "dataset_size": 3284227, "size_in_bytes": 4191573}, "russe": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\nWiC: The Word-in-Context Dataset A reliable benchmark for the evaluation of context-sensitive word embeddings.\nDepending on its context, an ambiguous word can refer to multiple, potentially unrelated, meanings. Mainstream static\nword embeddings, such as Word2vec and GloVe, are unable to reflect this dynamic semantic nature. Contextualised word\nembeddings are an attempt at addressing this limitation by computing dynamic representations for words which can adapt\nbased on context.\nRussian SuperGLUE task borrows original data from the Russe project, Word Sense Induction and Disambiguation\nshared task (2018)\n", "citation": "@inproceedings{RUSSE2018,\n author = {Panchenko, Alexander and Lopukhina, Anastasia and Ustalov, Dmitry and Lopukhin, Konstantin and Arefyev,\n Nikolay and Leontyev, Alexey and Loukachevitch, Natalia},\n title = {{RUSSE'2018: A Shared Task on Word Sense Induction for the Russian Language}},\n booktitle = {Computational Linguistics and Intellectual Technologies:\n Papers from the Annual International Conference ``Dialogue''},\n year = {2018},\n pages = {547--564},\n url = {http://www.dialog-21.ru/media/4539/panchenkoaplusetal.pdf},\n address = {Moscow, Russia},\n publisher = {RSUH},\n issn = {2221-7932},\n language = {english},\n}\n\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/RUSSE", "license": "", "features": {"word": {"dtype": "string", "id": null, "_type": "Value"}, "sentence1": {"dtype": "string", "id": null, "_type": "Value"}, "sentence2": {"dtype": "string", "id": null, "_type": "Value"}, "start1": {"dtype": "int32", "id": null, "_type": "Value"}, "start2": {"dtype": "int32", "id": null, "_type": "Value"}, "end1": {"dtype": "int32", "id": null, "_type": "Value"}, "end2": {"dtype": "int32", "id": null, "_type": "Value"}, "gold_sense1": {"dtype": "int32", "id": null, "_type": "Value"}, "gold_sense2": {"dtype": "int32", "id": null, "_type": "Value"}, "idx": {"dtype": "int32", "id": null, "_type": "Value"}, "label": {"num_classes": 2, "names": ["False", "True"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "russe", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 6913280, "num_examples": 19845, "dataset_name": "russian_super_glue"}, "validation": {"name": "validation", "num_bytes": 2957491, "num_examples": 8505, "dataset_name": "russian_super_glue"}, "test": {"name": "test", "num_bytes": 10046000, "num_examples": 18892, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/RUSSE": {"num_bytes": 3806009, "checksum": "60ecf42ea0f3893e857e0a9522ab92a2ae2ec713d1ab361f2f6f594d0f5324a5"}}, "download_size": 3806009, "post_processing_size": null, "dataset_size": 19916771, "size_in_bytes": 23722780}, "rwsd": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\nA Winograd schema is a pair of sentences that differ in only one or two words and that contain an ambiguity that is\nresolved in opposite ways in the two sentences and requires the use of world knowledge and reasoning for its resolution.\nThe schema takes its name from a well-known example by Terry Winograd.\nThe set would then be presented as a challenge for AI programs, along the lines of the Turing test. The strengths of\nthe challenge are that it is clear-cut, in that the answer to each schema is a binary choice; vivid, in that it is\nobvious to non-experts that a program that fails to get the right answers clearly has serious gaps in its understanding;\nand difficult, in that it is far beyond the current state of the art.\n", "citation": "\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/RWSD", "license": "", "features": {"text": {"dtype": "string", "id": null, "_type": "Value"}, "span1_index": {"dtype": "int32", "id": null, "_type": "Value"}, "span2_index": {"dtype": "int32", "id": null, "_type": "Value"}, "span1_text": {"dtype": "string", "id": null, "_type": "Value"}, "span2_text": {"dtype": "string", "id": null, "_type": "Value"}, "idx": {"dtype": "int32", "id": null, "_type": "Value"}, "label": {"num_classes": 2, "names": ["False", "True"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "rwsd", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 132274, "num_examples": 606, "dataset_name": "russian_super_glue"}, "validation": {"name": "validation", "num_bytes": 87959, "num_examples": 204, "dataset_name": "russian_super_glue"}, "test": {"name": "test", "num_bytes": 59051, "num_examples": 154, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/RWSD": {"num_bytes": 40508, "checksum": "65894cf114f022a0469bbd535f045d50880e03aa822cd1f3693a54b3665fa962"}}, "download_size": 40508, "post_processing_size": null, "dataset_size": 279284, "size_in_bytes": 319792}, "danetqa": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\nDaNetQA is a question answering dataset for yes/no questions. These questions are naturally occurring -- they are\ngenerated in unprompted and unconstrained settings.\n\nEach example is a triplet of (question, passage, answer), with the title of the page as optional additional context.\nThe text-pair classification setup is similar to existing natural language inference tasks.\n\nBy sampling questions from a distribution of information-seeking queries (rather than prompting annotators for\ntext pairs), we observe significantly more challenging examples compared to existing NLI datasets.\n", "citation": "@InProceedings{10.1007/978-3-030-72610-2_4,\nauthor=\"Glushkova, Taisia\nand Machnev, Alexey\nand Fenogenova, Alena\nand Shavrina, Tatiana\nand Artemova, Ekaterina\nand Ignatov, Dmitry I.\",\neditor=\"van der Aalst, Wil M. P.\nand Batagelj, Vladimir\nand Ignatov, Dmitry I.\nand Khachay, Michael\nand Koltsova, Olessia\nand Kutuzov, Andrey\nand Kuznetsov, Sergei O.\nand Lomazova, Irina A.\nand Loukachevitch, Natalia\nand Napoli, Amedeo\nand Panchenko, Alexander\nand Pardalos, Panos M.\nand Pelillo, Marcello\nand Savchenko, Andrey V.\nand Tutubalina, Elena\",\ntitle=\"DaNetQA: A Yes/No Question Answering Dataset for the Russian Language\",\nbooktitle=\"Analysis of Images, Social Networks and Texts\",\nyear=\"2021\",\npublisher=\"Springer International Publishing\",\naddress=\"Cham\",\npages=\"57--68\",\nabstract=\"DaNetQA, a new question-answering corpus, follows BoolQ\u00a0[2] design: it comprises natural yes/no questions.\nEach question is paired with a paragraph from Wikipedia and an answer, derived from the paragraph. The task is to take\nboth the question and a paragraph as input and come up with a yes/no answer, i.e. to produce a binary output. In this\npaper, we present a reproducible approach to DaNetQA creation and investigate transfer learning methods for task and\nlanguage transferring. For task transferring we leverage three similar sentence modelling tasks: 1) a corpus of\nparaphrases, Paraphraser, 2) an NLI task, for which we use the Russian part of XNLI, 3) another question answering task,\nSberQUAD. For language transferring we use English to Russian translation together\nwith multilingual language fine-tuning.\",\nisbn=\"978-3-030-72610-2\"\n}\n\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/DaNetQA", "license": "", "features": {"question": {"dtype": "string", "id": null, "_type": "Value"}, "passage": {"dtype": "string", "id": null, "_type": "Value"}, "idx": {"dtype": "int32", "id": null, "_type": "Value"}, "label": {"num_classes": 2, "names": ["False", "True"], "names_file": null, "id": null, "_type": "ClassLabel"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "danetqa", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 2474006, "num_examples": 1749, "dataset_name": "russian_super_glue"}, "validation": {"name": "validation", "num_bytes": 1076455, "num_examples": 821, "dataset_name": "russian_super_glue"}, "test": {"name": "test", "num_bytes": 1023062, "num_examples": 805, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/DaNetQA": {"num_bytes": 1293761, "checksum": "b5b4bcfe17e1eb16aa13a7aab4ca088871e27b0851468e9a07b9b528bb42fb96"}}, "download_size": 1293761, "post_processing_size": null, "dataset_size": 4573523, "size_in_bytes": 5867284}, "rucos": {"description": "Recent advances in the field of universal language models and transformers require the development of a methodology for\ntheir broad diagnostics and testing for general intellectual skills - detection of natural language inference,\ncommonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first\ntime, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from\nscratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating\nmodels and an overall leaderboard of transformer models for the Russian language.\nRussian reading comprehension with Commonsense reasoning (RuCoS) is a large-scale reading comprehension dataset which\nrequires commonsense reasoning. RuCoS consists of queries automatically generated from CNN/Daily Mail news articles;\nthe answer to each query is a text span from a summarizing passage of the corresponding news. The goal of RuCoS is to\nevaluate a machine`s ability of commonsense reasoning in reading comprehension.\n", "citation": "@inproceedings{fenogenova-etal-2020-read,\n title = \"Read and Reason with {M}u{S}e{RC} and {R}u{C}o{S}: Datasets for Machine Reading Comprehension for {R}ussian\",\n author = \"Fenogenova, Alena and\n Mikhailov, Vladislav and\n Shevelev, Denis\",\n booktitle = \"Proceedings of the 28th International Conference on Computational Linguistics\",\n month = dec,\n year = \"2020\",\n address = \"Barcelona, Spain (Online)\",\n publisher = \"International Committee on Computational Linguistics\",\n url = \"https://aclanthology.org/2020.coling-main.570\",\n doi = \"10.18653/v1/2020.coling-main.570\",\n pages = \"6481--6497\",\n abstract = \"The paper introduces two Russian machine reading comprehension (MRC) datasets, called MuSeRC and RuCoS,\n which require reasoning over multiple sentences and commonsense knowledge to infer the answer. The former follows\n the design of MultiRC, while the latter is a counterpart of the ReCoRD dataset. The datasets are included\n in RussianSuperGLUE, the Russian general language understanding benchmark. We provide a comparative analysis\n and demonstrate that the proposed tasks are relatively more complex as compared to the original ones for English.\n Besides, performance results of human solvers and BERT-based models show that MuSeRC and RuCoS represent a challenge\n for recent advanced neural models. We thus hope to facilitate research in the field of MRC for Russian and prompt\n the study of multi-hop reasoning in a cross-lingual scenario.\",\n}\n\n@article{shavrina2020russiansuperglue,\n title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},\n author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,\n Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and\n Evlampiev, Andrey},\n journal={arXiv preprint arXiv:2010.15925},\n year={2020}\n }\n", "homepage": "https://russiansuperglue.com/tasks/task_info/RuCoS", "license": "", "features": {"passage": {"dtype": "string", "id": null, "_type": "Value"}, "query": {"dtype": "string", "id": null, "_type": "Value"}, "entities": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "answers": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "idx": {"passage": {"dtype": "int32", "id": null, "_type": "Value"}, "query": {"dtype": "int32", "id": null, "_type": "Value"}}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "russian_super_glue", "config_name": "rucos", "version": {"version_str": "0.0.1", "description": null, "major": 0, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 160095378, "num_examples": 72193, "dataset_name": "russian_super_glue"}, "validation": {"name": "validation", "num_bytes": 16980563, "num_examples": 7577, "dataset_name": "russian_super_glue"}, "test": {"name": "test", "num_bytes": 15535209, "num_examples": 7257, "dataset_name": "russian_super_glue"}}, "download_checksums": {"https://russiansuperglue.com/tasks/download/RuCoS": {"num_bytes": 56208297, "checksum": "e2f42700122e79cfcce792b54df792630f033eb21e14863cede5852f3aa0d078"}}, "download_size": 56208297, "post_processing_size": null, "dataset_size": 192611150, "size_in_bytes": 248819447}}
dummy/danetqa/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b15154d1567c07c85fbae500bdc18de81ae8f3b5c546316ebe3978392369dd68
3
+ size 8743
dummy/lidirus/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6fc64d495ab2a2414dccd8817ce3d435a0b1a612edfa955a98cb10e0ed6985da
3
+ size 964
dummy/muserc/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:953dc9dcec2f8a8b86f32b31d9ed789189f1a0cbd588780c41197b03610eda24
3
+ size 23763
dummy/parus/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b078382bb5954cfe867a3223c04431d9deb7da03e95e697608120b3ce0454ac0
3
+ size 2283
dummy/rcb/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d5f034be9216b4029830567fdcf1bd32a0182bba60687109a805863b0133122f
3
+ size 3385
dummy/rucos/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a43c0ced7e29dceec7ca06f361812b5067531f6f435bd782745651594c078956
3
+ size 12731
dummy/russe/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9f9dc87fb55bcf9bdf737d9da9cc343fdd7e87639ece885b5a2e5b0bae94169a
3
+ size 3442
dummy/rwsd/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a54f37bf6cad4f4c4df70736828050072b5189cf41216cc7ea0143f496d8828c
3
+ size 2102
dummy/terra/0.0.1/dummy_data.zip ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6448408a6d8170b7a22b4052e1c6199e4749bb31d958d8dbb290f0f2f87bc67a
3
+ size 4062
russian_super_glue.py ADDED
@@ -0,0 +1,531 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ # Lint as: python3
17
+ """The Russian SuperGLUE Benchmark"""
18
+
19
+ import json
20
+ import os
21
+ from typing import List, Union
22
+
23
+ import datasets
24
+
25
+
26
+ _RUSSIAN_SUPER_GLUE_CITATION = """\
27
+ @article{shavrina2020russiansuperglue,
28
+ title={RussianSuperGLUE: A Russian Language Understanding Evaluation Benchmark},
29
+ author={Shavrina, Tatiana and Fenogenova, Alena and Emelyanov, Anton and Shevelev, Denis and Artemova,
30
+ Ekaterina and Malykh, Valentin and Mikhailov, Vladislav and Tikhonova, Maria and Chertok, Andrey and
31
+ Evlampiev, Andrey},
32
+ journal={arXiv preprint arXiv:2010.15925},
33
+ year={2020}
34
+ }
35
+ """
36
+
37
+ _MUSERC_CITATION = """\
38
+ @inproceedings{fenogenova-etal-2020-read,
39
+ title = "Read and Reason with {M}u{S}e{RC} and {R}u{C}o{S}: Datasets for Machine Reading Comprehension for {R}ussian",
40
+ author = "Fenogenova, Alena and
41
+ Mikhailov, Vladislav and
42
+ Shevelev, Denis",
43
+ booktitle = "Proceedings of the 28th International Conference on Computational Linguistics",
44
+ month = dec,
45
+ year = "2020",
46
+ address = "Barcelona, Spain (Online)",
47
+ publisher = "International Committee on Computational Linguistics",
48
+ url = "https://aclanthology.org/2020.coling-main.570",
49
+ doi = "10.18653/v1/2020.coling-main.570",
50
+ pages = "6481--6497",
51
+ abstract = "The paper introduces two Russian machine reading comprehension (MRC) datasets, called MuSeRC and RuCoS,
52
+ which require reasoning over multiple sentences and commonsense knowledge to infer the answer. The former follows
53
+ the design of MultiRC, while the latter is a counterpart of the ReCoRD dataset. The datasets are included
54
+ in RussianSuperGLUE, the Russian general language understanding benchmark. We provide a comparative analysis
55
+ and demonstrate that the proposed tasks are relatively more complex as compared to the original ones for English.
56
+ Besides, performance results of human solvers and BERT-based models show that MuSeRC and RuCoS represent a challenge
57
+ for recent advanced neural models. We thus hope to facilitate research in the field of MRC for Russian and prompt
58
+ the study of multi-hop reasoning in a cross-lingual scenario.",
59
+ }
60
+ """
61
+
62
+ _RUSSE_CITATION = """\
63
+ @inproceedings{RUSSE2018,
64
+ author = {Panchenko, Alexander and Lopukhina, Anastasia and Ustalov, Dmitry and Lopukhin, Konstantin and Arefyev,
65
+ Nikolay and Leontyev, Alexey and Loukachevitch, Natalia},
66
+ title = {{RUSSE'2018: A Shared Task on Word Sense Induction for the Russian Language}},
67
+ booktitle = {Computational Linguistics and Intellectual Technologies:
68
+ Papers from the Annual International Conference ``Dialogue''},
69
+ year = {2018},
70
+ pages = {547--564},
71
+ url = {http://www.dialog-21.ru/media/4539/panchenkoaplusetal.pdf},
72
+ address = {Moscow, Russia},
73
+ publisher = {RSUH},
74
+ issn = {2221-7932},
75
+ language = {english},
76
+ }
77
+ """
78
+
79
+ _DANETQA_CITATION = """\
80
+ @InProceedings{10.1007/978-3-030-72610-2_4,
81
+ author="Glushkova, Taisia
82
+ and Machnev, Alexey
83
+ and Fenogenova, Alena
84
+ and Shavrina, Tatiana
85
+ and Artemova, Ekaterina
86
+ and Ignatov, Dmitry I.",
87
+ editor="van der Aalst, Wil M. P.
88
+ and Batagelj, Vladimir
89
+ and Ignatov, Dmitry I.
90
+ and Khachay, Michael
91
+ and Koltsova, Olessia
92
+ and Kutuzov, Andrey
93
+ and Kuznetsov, Sergei O.
94
+ and Lomazova, Irina A.
95
+ and Loukachevitch, Natalia
96
+ and Napoli, Amedeo
97
+ and Panchenko, Alexander
98
+ and Pardalos, Panos M.
99
+ and Pelillo, Marcello
100
+ and Savchenko, Andrey V.
101
+ and Tutubalina, Elena",
102
+ title="DaNetQA: A Yes/No Question Answering Dataset for the Russian Language",
103
+ booktitle="Analysis of Images, Social Networks and Texts",
104
+ year="2021",
105
+ publisher="Springer International Publishing",
106
+ address="Cham",
107
+ pages="57--68",
108
+ abstract="DaNetQA, a new question-answering corpus, follows BoolQ [2] design: it comprises natural yes/no questions.
109
+ Each question is paired with a paragraph from Wikipedia and an answer, derived from the paragraph. The task is to take
110
+ both the question and a paragraph as input and come up with a yes/no answer, i.e. to produce a binary output. In this
111
+ paper, we present a reproducible approach to DaNetQA creation and investigate transfer learning methods for task and
112
+ language transferring. For task transferring we leverage three similar sentence modelling tasks: 1) a corpus of
113
+ paraphrases, Paraphraser, 2) an NLI task, for which we use the Russian part of XNLI, 3) another question answering task,
114
+ SberQUAD. For language transferring we use English to Russian translation together
115
+ with multilingual language fine-tuning.",
116
+ isbn="978-3-030-72610-2"
117
+ }
118
+ """
119
+
120
+ _RUCOS_CITATION = _MUSERC_CITATION
121
+
122
+ _RUSSIAN_SUPER_GLUE_DESCRIPTION = """\
123
+ Recent advances in the field of universal language models and transformers require the development of a methodology for
124
+ their broad diagnostics and testing for general intellectual skills - detection of natural language inference,
125
+ commonsense reasoning, ability to perform simple logical operations regardless of text subject or lexicon. For the first
126
+ time, a benchmark of nine tasks, collected and organized analogically to the SuperGLUE methodology, was developed from
127
+ scratch for the Russian language. We provide baselines, human level evaluation, an open-source framework for evaluating
128
+ models and an overall leaderboard of transformer models for the Russian language.
129
+ """
130
+
131
+ _HOMEPAGE = "https://russiansuperglue.com/"
132
+
133
+ _LICENSE = "MIT License"
134
+
135
+ _LIDIRUS_DESCRIPTION = """"\
136
+ LiDiRus (Linguistic Diagnostic for Russian) is a diagnostic dataset that covers a large volume of linguistic phenomena,
137
+ while allowing you to evaluate information systems on a simple test of textual entailment recognition.
138
+ See more details diagnostics.
139
+ """
140
+
141
+ _RCB_DESCRIPTION = """\
142
+ The Russian Commitment Bank is a corpus of naturally occurring discourses whose final sentence contains
143
+ a clause-embedding predicate under an entailment canceling operator (question, modal, negation, antecedent
144
+ of conditional).
145
+ """
146
+
147
+ _PARUS_DESCRIPTION = """\
148
+ Choice of Plausible Alternatives for Russian language
149
+ Choice of Plausible Alternatives for Russian language (PARus) evaluation provides researchers with a tool for assessing
150
+ progress in open-domain commonsense causal reasoning. Each question in PARus is composed of a premise and two
151
+ alternatives, where the task is to select the alternative that more plausibly has a causal relation with the premise.
152
+ The correct alternative is randomized so that the expected performance of randomly guessing is 50%.
153
+ """
154
+
155
+ _MUSERC_DESCRIPTION = """\
156
+ We present a reading comprehension challenge in which questions can only be answered by taking into account information
157
+ from multiple sentences. The dataset is the first to study multi-sentence inference at scale, with an open-ended set of
158
+ question types that requires reasoning skills.
159
+ """
160
+
161
+ _TERRA_DESCRIPTION = """\
162
+ Textual Entailment Recognition has been proposed recently as a generic task that captures major semantic inference
163
+ needs across many NLP applications, such as Question Answering, Information Retrieval, Information Extraction,
164
+ and Text Summarization. This task requires to recognize, given two text fragments, whether the meaning of one text is
165
+ entailed (can be inferred) from the other text.
166
+ """
167
+
168
+ _RUSSE_DESCRIPTION = """\
169
+ WiC: The Word-in-Context Dataset A reliable benchmark for the evaluation of context-sensitive word embeddings.
170
+ Depending on its context, an ambiguous word can refer to multiple, potentially unrelated, meanings. Mainstream static
171
+ word embeddings, such as Word2vec and GloVe, are unable to reflect this dynamic semantic nature. Contextualised word
172
+ embeddings are an attempt at addressing this limitation by computing dynamic representations for words which can adapt
173
+ based on context.
174
+ Russian SuperGLUE task borrows original data from the Russe project, Word Sense Induction and Disambiguation
175
+ shared task (2018)
176
+ """
177
+
178
+ _RWSD_DESCRIPTION = """\
179
+ A Winograd schema is a pair of sentences that differ in only one or two words and that contain an ambiguity that is
180
+ resolved in opposite ways in the two sentences and requires the use of world knowledge and reasoning for its resolution.
181
+ The schema takes its name from a well-known example by Terry Winograd.
182
+ The set would then be presented as a challenge for AI programs, along the lines of the Turing test. The strengths of
183
+ the challenge are that it is clear-cut, in that the answer to each schema is a binary choice; vivid, in that it is
184
+ obvious to non-experts that a program that fails to get the right answers clearly has serious gaps in its understanding;
185
+ and difficult, in that it is far beyond the current state of the art.
186
+ """
187
+
188
+ _DANETQA_DESCRIPTION = """\
189
+ DaNetQA is a question answering dataset for yes/no questions. These questions are naturally occurring -- they are
190
+ generated in unprompted and unconstrained settings.
191
+
192
+ Each example is a triplet of (question, passage, answer), with the title of the page as optional additional context.
193
+ The text-pair classification setup is similar to existing natural language inference tasks.
194
+
195
+ By sampling questions from a distribution of information-seeking queries (rather than prompting annotators for
196
+ text pairs), we observe significantly more challenging examples compared to existing NLI datasets.
197
+ """
198
+
199
+ _RUCOS_DESCRIPTION = """\
200
+ Russian reading comprehension with Commonsense reasoning (RuCoS) is a large-scale reading comprehension dataset which
201
+ requires commonsense reasoning. RuCoS consists of queries automatically generated from CNN/Daily Mail news articles;
202
+ the answer to each query is a text span from a summarizing passage of the corresponding news. The goal of RuCoS is to
203
+ evaluate a machine`s ability of commonsense reasoning in reading comprehension.
204
+ """
205
+
206
+
207
+ class RussianSuperGlueConfig(datasets.BuilderConfig):
208
+ """BuilderConfig for the Russian SuperGLUE."""
209
+
210
+ VERSION = datasets.Version("0.0.1")
211
+
212
+ def __init__(
213
+ self,
214
+ features: List[str],
215
+ data_url: str,
216
+ citation: str,
217
+ url: str,
218
+ label_classes: List[str] = ("False", "True"),
219
+ **kwargs,
220
+ ):
221
+ """BuilderConfig for the Russian SuperGLUE.
222
+
223
+ Args:
224
+ features: `list[string]`, list of the features that will appear in the
225
+ feature dict. Should not include "label".
226
+ data_url: `string`, url to download the zip file from.
227
+ citation: `string`, citation for the data set.
228
+ url: `string`, url for information about the data set.
229
+ label_classes: `list[string]`, the list of classes for the label if the
230
+ label is present as a string. Non-string labels will be cast to either
231
+ 'False' or 'True'.
232
+ **kwargs: keyword arguments forwarded to super.
233
+ """
234
+ # 0.0.1: Initial version.
235
+ super(RussianSuperGlueConfig, self).__init__(version=self.VERSION, **kwargs)
236
+ self.features = features
237
+ self.label_classes = label_classes
238
+ self.data_url = data_url
239
+ self.citation = citation
240
+ self.url = url
241
+
242
+
243
+ class RussianSuperGlue(datasets.GeneratorBasedBuilder):
244
+
245
+ BUILDER_CONFIGS = [
246
+ RussianSuperGlueConfig(
247
+ name="lidirus",
248
+ description=_LIDIRUS_DESCRIPTION,
249
+ features=[
250
+ "sentence1",
251
+ "sentence2",
252
+ "knowledge",
253
+ "lexical-semantics",
254
+ "logic",
255
+ "predicate-argument-structure",
256
+ ],
257
+ label_classes=["entailment", "not_entailment"],
258
+ data_url="https://russiansuperglue.com/tasks/download/LiDiRus",
259
+ citation="",
260
+ url="https://russiansuperglue.com/tasks/task_info/LiDiRus",
261
+ ),
262
+ RussianSuperGlueConfig(
263
+ name="rcb",
264
+ description=_RCB_DESCRIPTION,
265
+ features=["premise", "hypothesis", "verb", "negation"],
266
+ label_classes=["entailment", "contradiction", "neutral"],
267
+ data_url="https://russiansuperglue.com/tasks/download/RCB",
268
+ citation="",
269
+ url="https://russiansuperglue.com/tasks/task_info/RCB",
270
+ ),
271
+ RussianSuperGlueConfig(
272
+ name="parus",
273
+ description=_PARUS_DESCRIPTION,
274
+ label_classes=["choice1", "choice2"],
275
+ features=["premise", "choice1", "choice2", "question"],
276
+ data_url="https://russiansuperglue.com/tasks/download/PARus",
277
+ citation="",
278
+ url="https://russiansuperglue.com/tasks/task_info/PARus",
279
+ ),
280
+ RussianSuperGlueConfig(
281
+ name="muserc",
282
+ description=_MUSERC_DESCRIPTION,
283
+ features=["paragraph", "question", "answer"],
284
+ data_url="https://russiansuperglue.com/tasks/download/MuSeRC",
285
+ citation=_MUSERC_CITATION,
286
+ label_classes=["False", "True"],
287
+ url="https://russiansuperglue.com/tasks/task_info/MuSeRC",
288
+ ),
289
+ RussianSuperGlueConfig(
290
+ name="terra",
291
+ description=_TERRA_DESCRIPTION,
292
+ features=["premise", "hypothesis"],
293
+ label_classes=["entailment", "not_entailment"],
294
+ data_url="https://russiansuperglue.com/tasks/download/TERRa",
295
+ citation="",
296
+ url="https://russiansuperglue.com/tasks/task_info/TERRa",
297
+ ),
298
+ RussianSuperGlueConfig(
299
+ name="russe",
300
+ description=_RUSSE_DESCRIPTION,
301
+ features=[
302
+ "word",
303
+ "sentence1",
304
+ "sentence2",
305
+ "start1",
306
+ "start2",
307
+ "end1",
308
+ "end2",
309
+ "gold_sense1",
310
+ "gold_sense2",
311
+ ],
312
+ data_url="https://russiansuperglue.com/tasks/download/RUSSE",
313
+ citation=_RUSSE_CITATION,
314
+ label_classes=["False", "True"],
315
+ url="https://russiansuperglue.com/tasks/task_info/RUSSE",
316
+ ),
317
+ RussianSuperGlueConfig(
318
+ name="rwsd",
319
+ description=_RWSD_DESCRIPTION,
320
+ features=["text", "span1_index", "span2_index", "span1_text", "span2_text"],
321
+ data_url="https://russiansuperglue.com/tasks/download/RWSD",
322
+ citation="",
323
+ label_classes=["False", "True"],
324
+ url="https://russiansuperglue.com/tasks/task_info/RWSD",
325
+ ),
326
+ RussianSuperGlueConfig(
327
+ name="danetqa",
328
+ description=_DANETQA_DESCRIPTION,
329
+ features=["question", "passage"],
330
+ data_url="https://russiansuperglue.com/tasks/download/DaNetQA",
331
+ citation=_DANETQA_CITATION,
332
+ label_classes=["False", "True"],
333
+ url="https://russiansuperglue.com/tasks/task_info/DaNetQA",
334
+ ),
335
+ RussianSuperGlueConfig(
336
+ name="rucos",
337
+ description=_RUCOS_DESCRIPTION,
338
+ features=["passage", "query", "entities", "answers"],
339
+ data_url="https://russiansuperglue.com/tasks/download/RuCoS",
340
+ citation=_RUCOS_CITATION,
341
+ url="https://russiansuperglue.com/tasks/task_info/RuCoS",
342
+ ),
343
+ ]
344
+
345
+ def _info(self):
346
+
347
+ if self.config.name == "russe":
348
+ features = {feature: datasets.Value("string") for feature in ("word", "sentence1", "sentence2")}
349
+ features["start1"] = datasets.Value("int32")
350
+ features["start2"] = datasets.Value("int32")
351
+ features["end1"] = datasets.Value("int32")
352
+ features["end2"] = datasets.Value("int32")
353
+ features["gold_sense1"] = datasets.Value("int32")
354
+ features["gold_sense2"] = datasets.Value("int32")
355
+
356
+ else:
357
+ features = {feature: datasets.Value("string") for feature in self.config.features}
358
+
359
+ if self.config.name == "rwsd":
360
+ features["span1_index"] = datasets.Value("int32")
361
+ features["span2_index"] = datasets.Value("int32")
362
+
363
+ if self.config.name == "muserc":
364
+ features["idx"] = dict(
365
+ {
366
+ "paragraph": datasets.Value("int32"),
367
+ "question": datasets.Value("int32"),
368
+ "answer": datasets.Value("int32"),
369
+ }
370
+ )
371
+ elif self.config.name == "rucos":
372
+ features["idx"] = dict(
373
+ {
374
+ "passage": datasets.Value("int32"),
375
+ "query": datasets.Value("int32"),
376
+ }
377
+ )
378
+ else:
379
+ features["idx"] = datasets.Value("int32")
380
+
381
+ if self.config.name == "rucos":
382
+ # Entities are the set of possible choices for the placeholder.
383
+ features["entities"] = datasets.features.Sequence(datasets.Value("string"))
384
+ # Answers are the subset of entities that are correct.
385
+ features["answers"] = datasets.features.Sequence(datasets.Value("string"))
386
+ else:
387
+ features["label"] = datasets.features.ClassLabel(names=self.config.label_classes)
388
+
389
+ return datasets.DatasetInfo(
390
+ description=_RUSSIAN_SUPER_GLUE_DESCRIPTION + self.config.description,
391
+ features=datasets.Features(features),
392
+ homepage=self.config.url,
393
+ citation=self.config.citation + "\n" + _RUSSIAN_SUPER_GLUE_CITATION,
394
+ )
395
+
396
+ def _split_generators(self, dl_manager: datasets.DownloadManager):
397
+ dl_dir = dl_manager.download_and_extract(self.config.data_url) or ""
398
+ task_name = _get_task_name_from_data_url(self.config.data_url)
399
+ dl_dir = os.path.join(dl_dir, task_name)
400
+ if self.config.name == "lidirus":
401
+ return [
402
+ datasets.SplitGenerator(
403
+ name=datasets.Split.TEST,
404
+ gen_kwargs={
405
+ "data_file": os.path.join(dl_dir, f"{task_name}.jsonl"),
406
+ "split": datasets.Split.TEST,
407
+ },
408
+ ),
409
+ ]
410
+ else:
411
+ return [
412
+ datasets.SplitGenerator(
413
+ name=datasets.Split.TRAIN,
414
+ gen_kwargs={
415
+ "data_file": os.path.join(dl_dir, "train.jsonl"),
416
+ "split": datasets.Split.TRAIN,
417
+ },
418
+ ),
419
+ datasets.SplitGenerator(
420
+ name=datasets.Split.VALIDATION,
421
+ gen_kwargs={
422
+ "data_file": os.path.join(dl_dir, "val.jsonl"),
423
+ "split": datasets.Split.VALIDATION,
424
+ },
425
+ ),
426
+ datasets.SplitGenerator(
427
+ name=datasets.Split.TEST,
428
+ gen_kwargs={
429
+ "data_file": os.path.join(dl_dir, "test.jsonl"),
430
+ "split": datasets.Split.TEST,
431
+ },
432
+ ),
433
+ ]
434
+
435
+ def _generate_examples(self, data_file: str, split: datasets.Split):
436
+ with open(data_file, encoding="utf-8") as file:
437
+ for line in file:
438
+ row = json.loads(line)
439
+
440
+ if self.config.name == "muserc":
441
+
442
+ paragraph = row["passage"]
443
+ for question in paragraph["questions"]:
444
+ for answer in question["answers"]:
445
+ label = answer.get("label")
446
+ key = "%s_%s_%s" % (row["idx"], question["idx"], answer["idx"])
447
+ yield key, {
448
+ "paragraph": paragraph["text"],
449
+ "question": question["question"],
450
+ "answer": answer["text"],
451
+ "label": -1 if label is None else _cast_label(bool(label)),
452
+ "idx": {"paragraph": row["idx"], "question": question["idx"], "answer": answer["idx"]},
453
+ }
454
+
455
+ elif self.config.name == "rucos":
456
+ passage = row["passage"]
457
+ for qa in row["qas"]:
458
+ yield qa["idx"], {
459
+ "passage": passage["text"],
460
+ "query": qa["query"],
461
+ "entities": _get_rucos_entities(passage),
462
+ "answers": _get_rucos_answers(qa),
463
+ "idx": {"passage": row["idx"], "query": qa["idx"]},
464
+ }
465
+ else:
466
+ if self.config.name in ("lidirus", "rcb"):
467
+ # features may be missing
468
+ example = {feature: row.get(feature, "") for feature in self.config.features}
469
+ elif self.config.name == "russe" and split == datasets.Split.TEST:
470
+ # gold senses are not available in `test` split
471
+ example = {
472
+ feature: row[feature]
473
+ for feature in self.config.features
474
+ if feature not in ("gold_sense1", "gold_sense2")
475
+ }
476
+ example["gold_sense1"] = -1
477
+ example["gold_sense2"] = -1
478
+ else:
479
+ if self.config.name == "rwsd":
480
+ row.update(row["target"])
481
+
482
+ example = {feature: row[feature] for feature in self.config.features}
483
+
484
+ example["idx"] = row["idx"]
485
+
486
+ if "label" in row:
487
+ if self.config.name == "parus":
488
+ example["label"] = "choice2" if row["label"] else "choice1"
489
+ else:
490
+ example["label"] = _cast_label(row["label"])
491
+ else:
492
+ assert split == datasets.Split.TEST, row
493
+ example["label"] = -1
494
+
495
+ yield example["idx"], example
496
+
497
+
498
+ def _get_task_name_from_data_url(data_url: str) -> str:
499
+ return data_url.split("/")[-1]
500
+
501
+
502
+ def _cast_label(label: Union[str, bool, int]) -> str:
503
+ """Converts the label into the appropriate string version."""
504
+ if isinstance(label, str):
505
+ return label
506
+ elif isinstance(label, bool):
507
+ return "True" if label else "False"
508
+ elif isinstance(label, int):
509
+ assert label in (0, 1)
510
+ return str(label)
511
+ else:
512
+ raise ValueError("Invalid label format.")
513
+
514
+
515
+ def _get_rucos_entities(passage: dict) -> List[str]:
516
+ """Returns the unique set of entities."""
517
+ text = passage["text"]
518
+ entities = set()
519
+ for entity in passage["entities"]:
520
+ entities.add(text[entity["start"] : entity["end"] + 1])
521
+ return sorted(entities)
522
+
523
+
524
+ def _get_rucos_answers(qa: dict) -> List[str]:
525
+ """Returns the unique set of answers."""
526
+ if "answers" not in qa:
527
+ return []
528
+ answers = set()
529
+ for answer in qa["answers"]:
530
+ answers.add(answer["text"])
531
+ return sorted(answers)