Files changed (1) hide show
  1. README.md +14 -4
README.md CHANGED
@@ -14,10 +14,10 @@ size_categories:
14
  source_datasets:
15
  - original
16
  task_categories:
17
- - conversational
18
  - text-generation
19
  - fill-mask
20
  task_ids:
 
21
  - dialogue-generation
22
  - dialogue-modeling
23
  - language-modeling
@@ -162,15 +162,19 @@ dataset_info:
162
 
163
  - **Homepage:** https://worksheets.codalab.org/worksheets/0xa79833f4b3c24f4188cee7131b120a59
164
  - **Repository:** https://github.com/google/airdialogue
165
- - **Paper:** https://www.aclweb.org/anthology/D18-1419/
166
  - **Leaderboard:** https://worksheets.codalab.org/worksheets/0xa79833f4b3c24f4188cee7131b120a59
167
  - **Point of Contact:** [AirDialogue-Google](mailto:airdialogue@gmail.com)
168
- [Aakash Gupta](mailto:aakashg80@gmail.com)
169
 
170
  ### Dataset Summary
171
 
172
  AirDialogue, is a large dataset that contains 402,038 goal-oriented conversations. To collect this dataset, we create a contextgenerator which provides travel and flight restrictions. Then the human annotators are asked to play the role of a customer or an agent and interact with the goal of successfully booking a trip given the restrictions.
173
 
 
 
 
 
174
  ### Supported Tasks and Leaderboards
175
 
176
  We use perplexity and BLEU score to evaluate the quality of the language generated by the model. We also compare the dialogue state generated by the model s and the ground truth state s0. Two categories of the metrics are used: exact match scores and scaled scores
@@ -289,22 +293,28 @@ cc-by-nc-4.0
289
 
290
  ### Citation Information
291
 
 
292
  @inproceedings{wei-etal-2018-airdialogue,
293
  title = "{A}ir{D}ialogue: An Environment for Goal-Oriented Dialogue Research",
294
  author = "Wei, Wei and
295
  Le, Quoc and
296
  Dai, Andrew and
297
  Li, Jia",
 
 
 
 
298
  booktitle = "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
299
  month = oct # "-" # nov,
300
  year = "2018",
301
  address = "Brussels, Belgium",
302
  publisher = "Association for Computational Linguistics",
303
- url = "https://www.aclweb.org/anthology/D18-1419",
304
  doi = "10.18653/v1/D18-1419",
305
  pages = "3844--3854",
306
  abstract = "Recent progress in dialogue generation has inspired a number of studies on dialogue systems that are capable of accomplishing tasks through natural language interactions. A promising direction among these studies is the use of reinforcement learning techniques, such as self-play, for training dialogue agents. However, current datasets are limited in size, and the environment for training agents and evaluating progress is relatively unsophisticated. We present AirDialogue, a large dataset that contains 301,427 goal-oriented conversations. To collect this dataset, we create a context-generator which provides travel and flight restrictions. We then ask human annotators to play the role of a customer or an agent and interact with the goal of successfully booking a trip given the restrictions. Key to our environment is the ease of evaluating the success of the dialogue, which is achieved by using ground-truth states (e.g., the flight being booked) generated by the restrictions. Any dialogue agent that does not generate the correct states is considered to fail. Our experimental results indicate that state-of-the-art dialogue models can only achieve a score of 0.17 while humans can reach a score of 0.91, which suggests significant opportunities for future improvement.",
307
  }
 
308
 
309
  ### Contributions
310
 
 
14
  source_datasets:
15
  - original
16
  task_categories:
 
17
  - text-generation
18
  - fill-mask
19
  task_ids:
20
+ - conversational
21
  - dialogue-generation
22
  - dialogue-modeling
23
  - language-modeling
 
162
 
163
  - **Homepage:** https://worksheets.codalab.org/worksheets/0xa79833f4b3c24f4188cee7131b120a59
164
  - **Repository:** https://github.com/google/airdialogue
165
+ - **Paper:** https://aclanthology.org/D18-1419/
166
  - **Leaderboard:** https://worksheets.codalab.org/worksheets/0xa79833f4b3c24f4188cee7131b120a59
167
  - **Point of Contact:** [AirDialogue-Google](mailto:airdialogue@gmail.com)
168
+ - **Point of Contact:** [Wei Wei](mailto:wewei@google.com)
169
 
170
  ### Dataset Summary
171
 
172
  AirDialogue, is a large dataset that contains 402,038 goal-oriented conversations. To collect this dataset, we create a contextgenerator which provides travel and flight restrictions. Then the human annotators are asked to play the role of a customer or an agent and interact with the goal of successfully booking a trip given the restrictions.
173
 
174
+ News in v1.3:
175
+ - We have included the test split of the AirDialogue dataset.
176
+ - We have included the meta context for OOD2 in the original AirDialogue paper.
177
+
178
  ### Supported Tasks and Leaderboards
179
 
180
  We use perplexity and BLEU score to evaluate the quality of the language generated by the model. We also compare the dialogue state generated by the model s and the ground truth state s0. Two categories of the metrics are used: exact match scores and scaled scores
 
293
 
294
  ### Citation Information
295
 
296
+ ```bibtex
297
  @inproceedings{wei-etal-2018-airdialogue,
298
  title = "{A}ir{D}ialogue: An Environment for Goal-Oriented Dialogue Research",
299
  author = "Wei, Wei and
300
  Le, Quoc and
301
  Dai, Andrew and
302
  Li, Jia",
303
+ editor = "Riloff, Ellen and
304
+ Chiang, David and
305
+ Hockenmaier, Julia and
306
+ Tsujii, Jun{'}ichi",
307
  booktitle = "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
308
  month = oct # "-" # nov,
309
  year = "2018",
310
  address = "Brussels, Belgium",
311
  publisher = "Association for Computational Linguistics",
312
+ url = "https://aclanthology.org/D18-1419",
313
  doi = "10.18653/v1/D18-1419",
314
  pages = "3844--3854",
315
  abstract = "Recent progress in dialogue generation has inspired a number of studies on dialogue systems that are capable of accomplishing tasks through natural language interactions. A promising direction among these studies is the use of reinforcement learning techniques, such as self-play, for training dialogue agents. However, current datasets are limited in size, and the environment for training agents and evaluating progress is relatively unsophisticated. We present AirDialogue, a large dataset that contains 301,427 goal-oriented conversations. To collect this dataset, we create a context-generator which provides travel and flight restrictions. We then ask human annotators to play the role of a customer or an agent and interact with the goal of successfully booking a trip given the restrictions. Key to our environment is the ease of evaluating the success of the dialogue, which is achieved by using ground-truth states (e.g., the flight being booked) generated by the restrictions. Any dialogue agent that does not generate the correct states is considered to fail. Our experimental results indicate that state-of-the-art dialogue models can only achieve a score of 0.17 while humans can reach a score of 0.91, which suggests significant opportunities for future improvement.",
316
  }
317
+ ```
318
 
319
  ### Contributions
320