BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions

In this paper we study yes/no questions that are naturally occurring --- meaning that they are generated in unprompted and unconstrained settings. We build a reading comprehension dataset, BoolQ, of such questions, and show that they are unexpectedly challenging. They often query for complex, non-factoid information, and require difficult entailment-like inference to solve. We also explore the effectiveness of a range of transfer learning baselines. We find that transferring from entailment data is more effective than transferring from paraphrase or extractive QA data, and that it, surprisingly, continues to be very beneficial even when starting from massive pre-trained language models such as BERT. Our best method trains BERT on MultiNLI and then re-trains it on our train set. It achieves 80.4% accuracy compared to 90% accuracy of human annotators (and 62% majority-baseline), leaving a significant gap for future work.

PDF Abstract NAACL 2019 PDF NAACL 2019 Abstract

Results from the Paper

Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Question Answering BoolQ BERT-MultiNLI 340M (fine-tuned) Accuracy 80.4 # 26
Question Answering BoolQ GPT-1 117M (fine-tuned) Accuracy 72.87 # 35
Question Answering BoolQ BiDAF + ELMo (fine-tuned) Accuracy 71.41 # 36
Question Answering BoolQ BiDAF-MultiNLI (fine-tuned) Accuracy 75.57 # 33
Question Answering BoolQ Majority baseline Accuracy 62.17 # 46