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
8520f9e
1 Parent(s): 1fee472

Add droberta data files

Browse files
README.md CHANGED
@@ -147,16 +147,16 @@ dataset_info:
147
  dtype: string
148
  splits:
149
  - name: train
150
- num_bytes: 9270719
151
  num_examples: 10000
152
  - name: validation
153
- num_bytes: 925065
154
  num_examples: 1000
155
  - name: test
156
- num_bytes: 1005406
157
  num_examples: 1000
158
- download_size: 9018914
159
- dataset_size: 11201190
160
  configs:
161
  - config_name: adversarialQA
162
  data_files:
@@ -182,6 +182,14 @@ configs:
182
  path: dbidaf/validation-*
183
  - split: test
184
  path: dbidaf/test-*
 
 
 
 
 
 
 
 
185
  train-eval-index:
186
  - config: adversarialQA
187
  task: question-answering
 
147
  dtype: string
148
  splits:
149
  - name: train
150
+ num_bytes: 9270683
151
  num_examples: 10000
152
  - name: validation
153
+ num_bytes: 925029
154
  num_examples: 1000
155
  - name: test
156
+ num_bytes: 1005242
157
  num_examples: 1000
158
+ download_size: 2815452
159
+ dataset_size: 11200954
160
  configs:
161
  - config_name: adversarialQA
162
  data_files:
 
182
  path: dbidaf/validation-*
183
  - split: test
184
  path: dbidaf/test-*
185
+ - config_name: droberta
186
+ data_files:
187
+ - split: train
188
+ path: droberta/train-*
189
+ - split: validation
190
+ path: droberta/validation-*
191
+ - split: test
192
+ path: droberta/test-*
193
  train-eval-index:
194
  - config: adversarialQA
195
  task: question-answering
dataset_infos.json CHANGED
@@ -238,67 +238,55 @@
238
  },
239
  "droberta": {
240
  "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",
241
- "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",
242
  "homepage": "https://adversarialqa.github.io/",
243
  "license": "",
244
  "features": {
245
  "id": {
246
  "dtype": "string",
247
- "id": null,
248
  "_type": "Value"
249
  },
250
  "title": {
251
  "dtype": "string",
252
- "id": null,
253
  "_type": "Value"
254
  },
255
  "context": {
256
  "dtype": "string",
257
- "id": null,
258
  "_type": "Value"
259
  },
260
  "question": {
261
  "dtype": "string",
262
- "id": null,
263
  "_type": "Value"
264
  },
265
  "answers": {
266
  "feature": {
267
  "text": {
268
  "dtype": "string",
269
- "id": null,
270
  "_type": "Value"
271
  },
272
  "answer_start": {
273
  "dtype": "int32",
274
- "id": null,
275
  "_type": "Value"
276
  }
277
  },
278
- "length": -1,
279
- "id": null,
280
  "_type": "Sequence"
281
  },
282
  "metadata": {
283
  "split": {
284
  "dtype": "string",
285
- "id": null,
286
  "_type": "Value"
287
  },
288
  "model_in_the_loop": {
289
  "dtype": "string",
290
- "id": null,
291
  "_type": "Value"
292
  }
293
  }
294
  },
295
- "post_processed": null,
296
- "supervised_keys": null,
297
  "builder_name": "adversarial_qa",
 
298
  "config_name": "droberta",
299
  "version": {
300
  "version_str": "1.0.0",
301
- "description": null,
302
  "major": 1,
303
  "minor": 0,
304
  "patch": 0
@@ -306,32 +294,25 @@
306
  "splits": {
307
  "train": {
308
  "name": "train",
309
- "num_bytes": 9270719,
310
  "num_examples": 10000,
311
- "dataset_name": "adversarial_qa"
312
  },
313
  "validation": {
314
  "name": "validation",
315
- "num_bytes": 925065,
316
  "num_examples": 1000,
317
- "dataset_name": "adversarial_qa"
318
  },
319
  "test": {
320
  "name": "test",
321
- "num_bytes": 1005406,
322
  "num_examples": 1000,
323
- "dataset_name": "adversarial_qa"
324
- }
325
- },
326
- "download_checksums": {
327
- "https://adversarialqa.github.io/data/aqa_v1.0.zip": {
328
- "num_bytes": 9018914,
329
- "checksum": "f4f3c23224a5060b28c35e35581bd5cf46256dda3665418fb83d036d0e0c93cf"
330
  }
331
  },
332
- "download_size": 9018914,
333
- "post_processing_size": null,
334
- "dataset_size": 11201190,
335
- "size_in_bytes": 20220104
336
  }
337
  }
 
238
  },
239
  "droberta": {
240
  "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",
241
+ "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",
242
  "homepage": "https://adversarialqa.github.io/",
243
  "license": "",
244
  "features": {
245
  "id": {
246
  "dtype": "string",
 
247
  "_type": "Value"
248
  },
249
  "title": {
250
  "dtype": "string",
 
251
  "_type": "Value"
252
  },
253
  "context": {
254
  "dtype": "string",
 
255
  "_type": "Value"
256
  },
257
  "question": {
258
  "dtype": "string",
 
259
  "_type": "Value"
260
  },
261
  "answers": {
262
  "feature": {
263
  "text": {
264
  "dtype": "string",
 
265
  "_type": "Value"
266
  },
267
  "answer_start": {
268
  "dtype": "int32",
 
269
  "_type": "Value"
270
  }
271
  },
 
 
272
  "_type": "Sequence"
273
  },
274
  "metadata": {
275
  "split": {
276
  "dtype": "string",
 
277
  "_type": "Value"
278
  },
279
  "model_in_the_loop": {
280
  "dtype": "string",
 
281
  "_type": "Value"
282
  }
283
  }
284
  },
 
 
285
  "builder_name": "adversarial_qa",
286
+ "dataset_name": "adversarial_qa",
287
  "config_name": "droberta",
288
  "version": {
289
  "version_str": "1.0.0",
 
290
  "major": 1,
291
  "minor": 0,
292
  "patch": 0
 
294
  "splits": {
295
  "train": {
296
  "name": "train",
297
+ "num_bytes": 9270683,
298
  "num_examples": 10000,
299
+ "dataset_name": null
300
  },
301
  "validation": {
302
  "name": "validation",
303
+ "num_bytes": 925029,
304
  "num_examples": 1000,
305
+ "dataset_name": null
306
  },
307
  "test": {
308
  "name": "test",
309
+ "num_bytes": 1005242,
310
  "num_examples": 1000,
311
+ "dataset_name": null
 
 
 
 
 
 
312
  }
313
  },
314
+ "download_size": 2815452,
315
+ "dataset_size": 11200954,
316
+ "size_in_bytes": 14016406
 
317
  }
318
  }
droberta/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c39f63f0489f143f6f49bafd3762f4e53e5eaa98a4d65bd9c29307b81bdfc4e0
3
+ size 255806
droberta/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f0cd24723fab48c4ad1651be52690246af91962a93c506436ecf7939d47c476b
3
+ size 2298073
droberta/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af61198b7ef9b8881b6da922110fa70a64cc1e8a5644da39e7156de424a69a54
3
+ size 261573