metadata
license: cc-by-sa-3.0
Dataset Summary
BoolQ is a question answering dataset created by Google for yes/no questions containing 15942 examples. These questions are naturally occurring ---they are generated in unprompted and unconstrained settings. Each example was a triplet of (question, passage, answer), with the title of the page as optional additional context. I converted it to ShareGPT where the passage and answer are combined into one cohesive response from the AI. I plan to use this as regularization for upcoming datasets that I intend to post to this account.
Licensing Information
BoolQ is released under the Creative Commons Share-Alike 3.0 license.
Citation Information
@inproceedings{clark2019boolq,
title = {BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions},
author = {Clark, Christopher and Lee, Kenton and Chang, Ming-Wei, and Kwiatkowski, Tom and Collins, Michael, and Toutanova, Kristina},
booktitle = {NAACL},
year = {2019},
}
Contributions
Thanks to @lewtun, @lhoestq, @thomwolf, @patrickvonplaten, @albertvillanova for adding this dataset.