Datasets:

Languages:
English
Multilinguality:
monolingual
Size Categories:
10K<n<100K
Language Creators:
found
Annotations Creators:
crowdsourced
Source Datasets:
original
ArXiv:
Tags:
License:
albertvillanova HF staff commited on
Commit
908855b
1 Parent(s): 2101a45

Convert dataset to Parquet

Browse files

Convert dataset to Parquet.

README.md CHANGED
@@ -20,22 +20,6 @@ task_ids:
20
  - open-domain-qa
21
  paperswithcode_id: adversarialqa
22
  pretty_name: adversarialQA
23
- train-eval-index:
24
- - config: adversarialQA
25
- task: question-answering
26
- task_id: extractive_question_answering
27
- splits:
28
- train_split: train
29
- eval_split: validation
30
- col_mapping:
31
- question: question
32
- context: context
33
- answers:
34
- text: text
35
- answer_start: answer_start
36
- metrics:
37
- - type: squad
38
- name: SQuAD
39
  dataset_info:
40
  - config_name: adversarialQA
41
  features:
@@ -61,17 +45,17 @@ dataset_info:
61
  dtype: string
62
  splits:
63
  - name: train
64
- num_bytes: 27858794
65
  num_examples: 30000
66
  - name: validation
67
- num_bytes: 2757128
68
  num_examples: 3000
69
  - name: test
70
- num_bytes: 2919643
71
  num_examples: 3000
72
- download_size: 9018914
73
- dataset_size: 33535565
74
- - config_name: dbidaf
75
  features:
76
  - name: id
77
  dtype: string
@@ -95,17 +79,17 @@ dataset_info:
95
  dtype: string
96
  splits:
97
  - name: train
98
- num_bytes: 9282518
99
  num_examples: 10000
100
  - name: validation
101
- num_bytes: 917943
102
  num_examples: 1000
103
  - name: test
104
- num_bytes: 947111
105
  num_examples: 1000
106
  download_size: 9018914
107
- dataset_size: 11147572
108
- - config_name: dbert
109
  features:
110
  - name: id
111
  dtype: string
@@ -129,16 +113,16 @@ dataset_info:
129
  dtype: string
130
  splits:
131
  - name: train
132
- num_bytes: 9345557
133
  num_examples: 10000
134
  - name: validation
135
- num_bytes: 918192
136
  num_examples: 1000
137
  - name: test
138
- num_bytes: 971454
139
  num_examples: 1000
140
  download_size: 9018914
141
- dataset_size: 11235203
142
  - config_name: droberta
143
  features:
144
  - name: id
@@ -173,6 +157,31 @@ dataset_info:
173
  num_examples: 1000
174
  download_size: 9018914
175
  dataset_size: 11201190
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
176
  ---
177
 
178
  # Dataset Card for adversarialQA
 
20
  - open-domain-qa
21
  paperswithcode_id: adversarialqa
22
  pretty_name: adversarialQA
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
  dataset_info:
24
  - config_name: adversarialQA
25
  features:
 
45
  dtype: string
46
  splits:
47
  - name: train
48
+ num_bytes: 27858686
49
  num_examples: 30000
50
  - name: validation
51
+ num_bytes: 2757092
52
  num_examples: 3000
53
  - name: test
54
+ num_bytes: 2919479
55
  num_examples: 3000
56
+ download_size: 5301049
57
+ dataset_size: 33535257
58
+ - config_name: dbert
59
  features:
60
  - name: id
61
  dtype: string
 
79
  dtype: string
80
  splits:
81
  - name: train
82
+ num_bytes: 9345557
83
  num_examples: 10000
84
  - name: validation
85
+ num_bytes: 918192
86
  num_examples: 1000
87
  - name: test
88
+ num_bytes: 971454
89
  num_examples: 1000
90
  download_size: 9018914
91
+ dataset_size: 11235203
92
+ - config_name: dbidaf
93
  features:
94
  - name: id
95
  dtype: string
 
113
  dtype: string
114
  splits:
115
  - name: train
116
+ num_bytes: 9282518
117
  num_examples: 10000
118
  - name: validation
119
+ num_bytes: 917943
120
  num_examples: 1000
121
  - name: test
122
+ num_bytes: 947111
123
  num_examples: 1000
124
  download_size: 9018914
125
+ dataset_size: 11147572
126
  - config_name: droberta
127
  features:
128
  - name: id
 
157
  num_examples: 1000
158
  download_size: 9018914
159
  dataset_size: 11201190
160
+ configs:
161
+ - config_name: adversarialQA
162
+ data_files:
163
+ - split: train
164
+ path: adversarialQA/train-*
165
+ - split: validation
166
+ path: adversarialQA/validation-*
167
+ - split: test
168
+ path: adversarialQA/test-*
169
+ train-eval-index:
170
+ - config: adversarialQA
171
+ task: question-answering
172
+ task_id: extractive_question_answering
173
+ splits:
174
+ train_split: train
175
+ eval_split: validation
176
+ col_mapping:
177
+ question: question
178
+ context: context
179
+ answers:
180
+ text: text
181
+ answer_start: answer_start
182
+ metrics:
183
+ - type: squad
184
+ name: SQuAD
185
  ---
186
 
187
  # Dataset Card for adversarialQA
adversarialQA/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1a84cab6ed591ed6024c9eab388131fb40cfcd01b0a76d84735b85115b4bde76
3
+ size 457062
adversarialQA/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0010a888ffe67a4da4afb7a9cbb7e0e77daaf02aaf055ee1697c88519afb94d8
3
+ size 4348721
adversarialQA/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ed5569eb1dedc73ad7a1dc34d9064ca8693216b1b47e5c2773a9b2a81668f1f
3
+ size 495266
dataset_infos.json CHANGED
@@ -1,394 +1,375 @@
1
  {
2
- "adversarialQA": {
3
- "description": "AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n",
4
- "citation": "@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl\\_a\\_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n",
5
- "homepage": "https://adversarialqa.github.io/",
6
- "license": "",
7
- "features": {
8
- "id": {
9
- "dtype": "string",
10
- "id": null,
11
- "_type": "Value"
12
- },
13
- "title": {
14
- "dtype": "string",
15
- "id": null,
16
- "_type": "Value"
17
- },
18
- "context": {
19
- "dtype": "string",
20
- "id": null,
21
- "_type": "Value"
22
- },
23
- "question": {
24
- "dtype": "string",
25
- "id": null,
26
- "_type": "Value"
27
- },
28
- "answers": {
29
- "feature": {
30
- "text": {
31
- "dtype": "string",
32
- "id": null,
33
- "_type": "Value"
34
- },
35
- "answer_start": {
36
- "dtype": "int32",
37
- "id": null,
38
- "_type": "Value"
39
- }
 
 
 
 
 
 
 
40
  },
41
- "length": -1,
42
- "id": null,
43
- "_type": "Sequence"
44
- },
45
- "metadata": {
46
- "split": {
47
- "dtype": "string",
48
- "id": null,
49
- "_type": "Value"
50
  },
51
- "model_in_the_loop": {
52
- "dtype": "string",
53
- "id": null,
54
- "_type": "Value"
55
- }
56
- }
57
- },
58
- "post_processed": null,
59
- "supervised_keys": null,
60
- "builder_name": "adversarial_qa",
61
- "config_name": "adversarialQA",
62
- "version": {
63
- "version_str": "1.0.0",
64
- "description": null,
65
- "major": 1,
66
- "minor": 0,
67
- "patch": 0
68
- },
69
- "splits": {
70
- "train": {
71
- "name": "train",
72
- "num_bytes": 27858794,
73
- "num_examples": 30000,
74
- "dataset_name": "adversarial_qa"
75
- },
76
- "validation": {
77
- "name": "validation",
78
- "num_bytes": 2757128,
79
- "num_examples": 3000,
80
- "dataset_name": "adversarial_qa"
81
- },
82
- "test": {
83
- "name": "test",
84
- "num_bytes": 2919643,
85
- "num_examples": 3000,
86
- "dataset_name": "adversarial_qa"
87
- }
88
- },
89
- "download_checksums": {
90
- "https://adversarialqa.github.io/data/aqa_v1.0.zip": {
91
- "num_bytes": 9018914,
92
- "checksum": "f4f3c23224a5060b28c35e35581bd5cf46256dda3665418fb83d036d0e0c93cf"
93
- }
94
  },
95
- "download_size": 9018914,
96
- "post_processing_size": null,
97
- "dataset_size": 33535565,
98
- "size_in_bytes": 42554479
99
- },
100
- "dbidaf": {
101
- "description": "AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n",
102
- "citation": "@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl\\_a\\_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n",
103
- "homepage": "https://adversarialqa.github.io/",
104
- "license": "",
105
- "features": {
106
- "id": {
107
- "dtype": "string",
108
- "id": null,
109
- "_type": "Value"
110
- },
111
- "title": {
112
- "dtype": "string",
113
- "id": null,
114
- "_type": "Value"
115
- },
116
- "context": {
117
- "dtype": "string",
118
- "id": null,
119
- "_type": "Value"
120
- },
121
- "question": {
122
- "dtype": "string",
123
- "id": null,
124
- "_type": "Value"
125
- },
126
- "answers": {
127
- "feature": {
128
- "text": {
129
- "dtype": "string",
130
- "id": null,
131
- "_type": "Value"
132
- },
133
- "answer_start": {
134
- "dtype": "int32",
135
- "id": null,
136
- "_type": "Value"
137
- }
 
 
 
 
 
 
 
 
 
 
 
 
138
  },
139
- "length": -1,
140
- "id": null,
141
- "_type": "Sequence"
142
- },
143
- "metadata": {
144
- "split": {
145
- "dtype": "string",
146
- "id": null,
147
- "_type": "Value"
 
148
  },
149
- "model_in_the_loop": {
150
- "dtype": "string",
151
- "id": null,
152
- "_type": "Value"
153
- }
154
- }
155
- },
156
- "post_processed": null,
157
- "supervised_keys": null,
158
- "builder_name": "adversarial_qa",
159
- "config_name": "dbidaf",
160
- "version": {
161
- "version_str": "1.0.0",
162
- "description": null,
163
- "major": 1,
164
- "minor": 0,
165
- "patch": 0
166
- },
167
- "splits": {
168
- "train": {
169
- "name": "train",
170
- "num_bytes": 9282518,
171
- "num_examples": 10000,
172
- "dataset_name": "adversarial_qa"
173
- },
174
- "validation": {
175
- "name": "validation",
176
- "num_bytes": 917943,
177
- "num_examples": 1000,
178
- "dataset_name": "adversarial_qa"
179
- },
180
- "test": {
181
- "name": "test",
182
- "num_bytes": 947111,
183
- "num_examples": 1000,
184
- "dataset_name": "adversarial_qa"
185
- }
186
- },
187
- "download_checksums": {
188
- "https://adversarialqa.github.io/data/aqa_v1.0.zip": {
189
- "num_bytes": 9018914,
190
- "checksum": "f4f3c23224a5060b28c35e35581bd5cf46256dda3665418fb83d036d0e0c93cf"
191
- }
192
- },
193
- "download_size": 9018914,
194
- "post_processing_size": null,
195
- "dataset_size": 11147572,
196
- "size_in_bytes": 20166486
197
- },
198
- "dbert": {
199
- "description": "AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n",
200
- "citation": "@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl\\_a\\_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n",
201
- "homepage": "https://adversarialqa.github.io/",
202
- "license": "",
203
- "features": {
204
- "id": {
205
- "dtype": "string",
206
- "id": null,
207
- "_type": "Value"
208
- },
209
- "title": {
210
- "dtype": "string",
211
- "id": null,
212
- "_type": "Value"
213
- },
214
- "context": {
215
- "dtype": "string",
216
- "id": null,
217
- "_type": "Value"
218
- },
219
- "question": {
220
- "dtype": "string",
221
- "id": null,
222
- "_type": "Value"
223
- },
224
- "answers": {
225
- "feature": {
226
- "text": {
227
- "dtype": "string",
228
- "id": null,
229
- "_type": "Value"
230
- },
231
- "answer_start": {
232
- "dtype": "int32",
233
- "id": null,
234
- "_type": "Value"
235
- }
236
  },
237
- "length": -1,
238
- "id": null,
239
- "_type": "Sequence"
240
- },
241
- "metadata": {
242
- "split": {
243
- "dtype": "string",
244
- "id": null,
245
- "_type": "Value"
246
  },
247
- "model_in_the_loop": {
248
- "dtype": "string",
249
- "id": null,
250
- "_type": "Value"
251
- }
252
- }
253
- },
254
- "post_processed": null,
255
- "supervised_keys": null,
256
- "builder_name": "adversarial_qa",
257
- "config_name": "dbert",
258
- "version": {
259
- "version_str": "1.0.0",
260
- "description": null,
261
- "major": 1,
262
- "minor": 0,
263
- "patch": 0
264
- },
265
- "splits": {
266
- "train": {
267
- "name": "train",
268
- "num_bytes": 9345557,
269
- "num_examples": 10000,
270
- "dataset_name": "adversarial_qa"
271
- },
272
- "validation": {
273
- "name": "validation",
274
- "num_bytes": 918192,
275
- "num_examples": 1000,
276
- "dataset_name": "adversarial_qa"
277
- },
278
- "test": {
279
- "name": "test",
280
- "num_bytes": 971454,
281
- "num_examples": 1000,
282
- "dataset_name": "adversarial_qa"
283
- }
284
  },
285
- "download_checksums": {
286
- "https://adversarialqa.github.io/data/aqa_v1.0.zip": {
287
- "num_bytes": 9018914,
288
- "checksum": "f4f3c23224a5060b28c35e35581bd5cf46256dda3665418fb83d036d0e0c93cf"
289
- }
290
- },
291
- "download_size": 9018914,
292
- "post_processing_size": null,
293
- "dataset_size": 11235203,
294
- "size_in_bytes": 20254117
295
- },
296
- "droberta": {
297
- "description": "AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n",
298
- "citation": "@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl\\_a\\_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n",
299
- "homepage": "https://adversarialqa.github.io/",
300
- "license": "",
301
- "features": {
302
- "id": {
303
- "dtype": "string",
304
- "id": null,
305
- "_type": "Value"
306
- },
307
- "title": {
308
- "dtype": "string",
309
- "id": null,
310
- "_type": "Value"
311
- },
312
- "context": {
313
- "dtype": "string",
314
- "id": null,
315
- "_type": "Value"
316
- },
317
- "question": {
318
- "dtype": "string",
319
- "id": null,
320
- "_type": "Value"
321
- },
322
- "answers": {
323
- "feature": {
324
- "text": {
325
- "dtype": "string",
326
- "id": null,
327
- "_type": "Value"
328
- },
329
- "answer_start": {
330
- "dtype": "int32",
331
- "id": null,
332
- "_type": "Value"
333
- }
 
 
 
 
 
 
334
  },
335
- "length": -1,
336
- "id": null,
337
- "_type": "Sequence"
338
- },
339
- "metadata": {
340
- "split": {
341
- "dtype": "string",
342
- "id": null,
343
- "_type": "Value"
 
344
  },
345
- "model_in_the_loop": {
346
- "dtype": "string",
347
- "id": null,
348
- "_type": "Value"
349
- }
350
- }
351
- },
352
- "post_processed": null,
353
- "supervised_keys": null,
354
- "builder_name": "adversarial_qa",
355
- "config_name": "droberta",
356
- "version": {
357
- "version_str": "1.0.0",
358
- "description": null,
359
- "major": 1,
360
- "minor": 0,
361
- "patch": 0
362
- },
363
- "splits": {
364
- "train": {
365
- "name": "train",
366
- "num_bytes": 9270719,
367
- "num_examples": 10000,
368
- "dataset_name": "adversarial_qa"
369
- },
370
- "validation": {
371
- "name": "validation",
372
- "num_bytes": 925065,
373
- "num_examples": 1000,
374
- "dataset_name": "adversarial_qa"
375
- },
376
- "test": {
377
- "name": "test",
378
- "num_bytes": 1005406,
379
- "num_examples": 1000,
380
- "dataset_name": "adversarial_qa"
381
- }
382
- },
383
- "download_checksums": {
384
- "https://adversarialqa.github.io/data/aqa_v1.0.zip": {
385
- "num_bytes": 9018914,
386
- "checksum": "f4f3c23224a5060b28c35e35581bd5cf46256dda3665418fb83d036d0e0c93cf"
387
- }
388
  },
389
- "download_size": 9018914,
390
- "post_processing_size": null,
391
- "dataset_size": 11201190,
392
- "size_in_bytes": 20220104
393
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
394
  }
 
1
  {
2
+ "adversarialQA": {
3
+ "description": "AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n",
4
+ "citation": "@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl_a_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n",
5
+ "homepage": "https://adversarialqa.github.io/",
6
+ "license": "",
7
+ "features": {
8
+ "id": {
9
+ "dtype": "string",
10
+ "_type": "Value"
11
+ },
12
+ "title": {
13
+ "dtype": "string",
14
+ "_type": "Value"
15
+ },
16
+ "context": {
17
+ "dtype": "string",
18
+ "_type": "Value"
19
+ },
20
+ "question": {
21
+ "dtype": "string",
22
+ "_type": "Value"
23
+ },
24
+ "answers": {
25
+ "feature": {
26
+ "text": {
27
+ "dtype": "string",
28
+ "_type": "Value"
29
+ },
30
+ "answer_start": {
31
+ "dtype": "int32",
32
+ "_type": "Value"
33
+ }
34
+ },
35
+ "_type": "Sequence"
36
+ },
37
+ "metadata": {
38
+ "split": {
39
+ "dtype": "string",
40
+ "_type": "Value"
41
+ },
42
+ "model_in_the_loop": {
43
+ "dtype": "string",
44
+ "_type": "Value"
45
+ }
46
+ }
47
  },
48
+ "builder_name": "adversarial_qa",
49
+ "dataset_name": "adversarial_qa",
50
+ "config_name": "adversarialQA",
51
+ "version": {
52
+ "version_str": "1.0.0",
53
+ "major": 1,
54
+ "minor": 0,
55
+ "patch": 0
 
56
  },
57
+ "splits": {
58
+ "train": {
59
+ "name": "train",
60
+ "num_bytes": 27858686,
61
+ "num_examples": 30000,
62
+ "dataset_name": null
63
+ },
64
+ "validation": {
65
+ "name": "validation",
66
+ "num_bytes": 2757092,
67
+ "num_examples": 3000,
68
+ "dataset_name": null
69
+ },
70
+ "test": {
71
+ "name": "test",
72
+ "num_bytes": 2919479,
73
+ "num_examples": 3000,
74
+ "dataset_name": null
75
+ }
76
+ },
77
+ "download_size": 5301049,
78
+ "dataset_size": 33535257,
79
+ "size_in_bytes": 38836306
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
80
  },
81
+ "dbidaf": {
82
+ "description": "AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n",
83
+ "citation": "@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl\\_a\\_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n",
84
+ "homepage": "https://adversarialqa.github.io/",
85
+ "license": "",
86
+ "features": {
87
+ "id": {
88
+ "dtype": "string",
89
+ "id": null,
90
+ "_type": "Value"
91
+ },
92
+ "title": {
93
+ "dtype": "string",
94
+ "id": null,
95
+ "_type": "Value"
96
+ },
97
+ "context": {
98
+ "dtype": "string",
99
+ "id": null,
100
+ "_type": "Value"
101
+ },
102
+ "question": {
103
+ "dtype": "string",
104
+ "id": null,
105
+ "_type": "Value"
106
+ },
107
+ "answers": {
108
+ "feature": {
109
+ "text": {
110
+ "dtype": "string",
111
+ "id": null,
112
+ "_type": "Value"
113
+ },
114
+ "answer_start": {
115
+ "dtype": "int32",
116
+ "id": null,
117
+ "_type": "Value"
118
+ }
119
+ },
120
+ "length": -1,
121
+ "id": null,
122
+ "_type": "Sequence"
123
+ },
124
+ "metadata": {
125
+ "split": {
126
+ "dtype": "string",
127
+ "id": null,
128
+ "_type": "Value"
129
+ },
130
+ "model_in_the_loop": {
131
+ "dtype": "string",
132
+ "id": null,
133
+ "_type": "Value"
134
+ }
135
+ }
136
  },
137
+ "post_processed": null,
138
+ "supervised_keys": null,
139
+ "builder_name": "adversarial_qa",
140
+ "config_name": "dbidaf",
141
+ "version": {
142
+ "version_str": "1.0.0",
143
+ "description": null,
144
+ "major": 1,
145
+ "minor": 0,
146
+ "patch": 0
147
  },
148
+ "splits": {
149
+ "train": {
150
+ "name": "train",
151
+ "num_bytes": 9282518,
152
+ "num_examples": 10000,
153
+ "dataset_name": "adversarial_qa"
154
+ },
155
+ "validation": {
156
+ "name": "validation",
157
+ "num_bytes": 917943,
158
+ "num_examples": 1000,
159
+ "dataset_name": "adversarial_qa"
160
+ },
161
+ "test": {
162
+ "name": "test",
163
+ "num_bytes": 947111,
164
+ "num_examples": 1000,
165
+ "dataset_name": "adversarial_qa"
166
+ }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
167
  },
168
+ "download_checksums": {
169
+ "https://adversarialqa.github.io/data/aqa_v1.0.zip": {
170
+ "num_bytes": 9018914,
171
+ "checksum": "f4f3c23224a5060b28c35e35581bd5cf46256dda3665418fb83d036d0e0c93cf"
172
+ }
 
 
 
 
173
  },
174
+ "download_size": 9018914,
175
+ "post_processing_size": null,
176
+ "dataset_size": 11147572,
177
+ "size_in_bytes": 20166486
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
178
  },
179
+ "dbert": {
180
+ "description": "AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n",
181
+ "citation": "@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl\\_a\\_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n",
182
+ "homepage": "https://adversarialqa.github.io/",
183
+ "license": "",
184
+ "features": {
185
+ "id": {
186
+ "dtype": "string",
187
+ "id": null,
188
+ "_type": "Value"
189
+ },
190
+ "title": {
191
+ "dtype": "string",
192
+ "id": null,
193
+ "_type": "Value"
194
+ },
195
+ "context": {
196
+ "dtype": "string",
197
+ "id": null,
198
+ "_type": "Value"
199
+ },
200
+ "question": {
201
+ "dtype": "string",
202
+ "id": null,
203
+ "_type": "Value"
204
+ },
205
+ "answers": {
206
+ "feature": {
207
+ "text": {
208
+ "dtype": "string",
209
+ "id": null,
210
+ "_type": "Value"
211
+ },
212
+ "answer_start": {
213
+ "dtype": "int32",
214
+ "id": null,
215
+ "_type": "Value"
216
+ }
217
+ },
218
+ "length": -1,
219
+ "id": null,
220
+ "_type": "Sequence"
221
+ },
222
+ "metadata": {
223
+ "split": {
224
+ "dtype": "string",
225
+ "id": null,
226
+ "_type": "Value"
227
+ },
228
+ "model_in_the_loop": {
229
+ "dtype": "string",
230
+ "id": null,
231
+ "_type": "Value"
232
+ }
233
+ }
234
  },
235
+ "post_processed": null,
236
+ "supervised_keys": null,
237
+ "builder_name": "adversarial_qa",
238
+ "config_name": "dbert",
239
+ "version": {
240
+ "version_str": "1.0.0",
241
+ "description": null,
242
+ "major": 1,
243
+ "minor": 0,
244
+ "patch": 0
245
  },
246
+ "splits": {
247
+ "train": {
248
+ "name": "train",
249
+ "num_bytes": 9345557,
250
+ "num_examples": 10000,
251
+ "dataset_name": "adversarial_qa"
252
+ },
253
+ "validation": {
254
+ "name": "validation",
255
+ "num_bytes": 918192,
256
+ "num_examples": 1000,
257
+ "dataset_name": "adversarial_qa"
258
+ },
259
+ "test": {
260
+ "name": "test",
261
+ "num_bytes": 971454,
262
+ "num_examples": 1000,
263
+ "dataset_name": "adversarial_qa"
264
+ }
265
+ },
266
+ "download_checksums": {
267
+ "https://adversarialqa.github.io/data/aqa_v1.0.zip": {
268
+ "num_bytes": 9018914,
269
+ "checksum": "f4f3c23224a5060b28c35e35581bd5cf46256dda3665418fb83d036d0e0c93cf"
270
+ }
271
+ },
272
+ "download_size": 9018914,
273
+ "post_processing_size": null,
274
+ "dataset_size": 11235203,
275
+ "size_in_bytes": 20254117
 
 
 
 
 
 
 
 
 
 
 
 
 
276
  },
277
+ "droberta": {
278
+ "description": "AdversarialQA is a Reading Comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles using an adversarial model-in-the-loop.\nWe use three different models; BiDAF (Seo et al., 2016), BERT-Large (Devlin et al., 2018), and RoBERTa-Large (Liu et al., 2019) in the annotation loop and construct three datasets; D(BiDAF), D(BERT), and D(RoBERTa), each with 10,000 training examples, 1,000 validation, and 1,000 test examples.\nThe adversarial human annotation paradigm ensures that these datasets consist of questions that current state-of-the-art models (at least the ones used as adversaries in the annotation loop) find challenging.\n",
279
+ "citation": "@article{bartolo2020beat,\n author = {Bartolo, Max and Roberts, Alastair and Welbl, Johannes and Riedel, Sebastian and Stenetorp, Pontus},\n title = {Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension},\n journal = {Transactions of the Association for Computational Linguistics},\n volume = {8},\n number = {},\n pages = {662-678},\n year = {2020},\n doi = {10.1162/tacl\\_a\\_00338},\n URL = { https://doi.org/10.1162/tacl_a_00338 },\n eprint = { https://doi.org/10.1162/tacl_a_00338 },\n abstract = { Innovations in annotation methodology have been a catalyst for Reading Comprehension (RC) datasets and models. One recent trend to challenge current RC models is to involve a model in the annotation process: Humans create questions adversarially, such that the model fails to answer them correctly. In this work we investigate this annotation methodology and apply it in three different settings, collecting a total of 36,000 samples with progressively stronger models in the annotation loop. This allows us to explore questions such as the reproducibility of the adversarial effect, transfer from data collected with varying model-in-the-loop strengths, and generalization to data collected without a model. We find that training on adversarially collected samples leads to strong generalization to non-adversarially collected datasets, yet with progressive performance deterioration with increasingly stronger models-in-the-loop. Furthermore, we find that stronger models can still learn from datasets collected with substantially weaker models-in-the-loop. When trained on data collected with a BiDAF model in the loop, RoBERTa achieves 39.9F1 on questions that it cannot answer when trained on SQuAD\u2014only marginally lower than when trained on data collected using RoBERTa itself (41.0F1). }\n}\n",
280
+ "homepage": "https://adversarialqa.github.io/",
281
+ "license": "",
282
+ "features": {
283
+ "id": {
284
+ "dtype": "string",
285
+ "id": null,
286
+ "_type": "Value"
287
+ },
288
+ "title": {
289
+ "dtype": "string",
290
+ "id": null,
291
+ "_type": "Value"
292
+ },
293
+ "context": {
294
+ "dtype": "string",
295
+ "id": null,
296
+ "_type": "Value"
297
+ },
298
+ "question": {
299
+ "dtype": "string",
300
+ "id": null,
301
+ "_type": "Value"
302
+ },
303
+ "answers": {
304
+ "feature": {
305
+ "text": {
306
+ "dtype": "string",
307
+ "id": null,
308
+ "_type": "Value"
309
+ },
310
+ "answer_start": {
311
+ "dtype": "int32",
312
+ "id": null,
313
+ "_type": "Value"
314
+ }
315
+ },
316
+ "length": -1,
317
+ "id": null,
318
+ "_type": "Sequence"
319
+ },
320
+ "metadata": {
321
+ "split": {
322
+ "dtype": "string",
323
+ "id": null,
324
+ "_type": "Value"
325
+ },
326
+ "model_in_the_loop": {
327
+ "dtype": "string",
328
+ "id": null,
329
+ "_type": "Value"
330
+ }
331
+ }
332
+ },
333
+ "post_processed": null,
334
+ "supervised_keys": null,
335
+ "builder_name": "adversarial_qa",
336
+ "config_name": "droberta",
337
+ "version": {
338
+ "version_str": "1.0.0",
339
+ "description": null,
340
+ "major": 1,
341
+ "minor": 0,
342
+ "patch": 0
343
+ },
344
+ "splits": {
345
+ "train": {
346
+ "name": "train",
347
+ "num_bytes": 9270719,
348
+ "num_examples": 10000,
349
+ "dataset_name": "adversarial_qa"
350
+ },
351
+ "validation": {
352
+ "name": "validation",
353
+ "num_bytes": 925065,
354
+ "num_examples": 1000,
355
+ "dataset_name": "adversarial_qa"
356
+ },
357
+ "test": {
358
+ "name": "test",
359
+ "num_bytes": 1005406,
360
+ "num_examples": 1000,
361
+ "dataset_name": "adversarial_qa"
362
+ }
363
+ },
364
+ "download_checksums": {
365
+ "https://adversarialqa.github.io/data/aqa_v1.0.zip": {
366
+ "num_bytes": 9018914,
367
+ "checksum": "f4f3c23224a5060b28c35e35581bd5cf46256dda3665418fb83d036d0e0c93cf"
368
+ }
369
+ },
370
+ "download_size": 9018914,
371
+ "post_processing_size": null,
372
+ "dataset_size": 11201190,
373
+ "size_in_bytes": 20220104
374
+ }
375
  }