Span Selection Pre-training for Question Answering

BERT (Bidirectional Encoder Representations from Transformers) and related pre-trained Transformers have provided large gains across many language understanding tasks, achieving a new state-of-the-art (SOTA). BERT is pre-trained on two auxiliary tasks: Masked Language Model and Next Sentence Prediction. In this paper we introduce a new pre-training task inspired by reading comprehension and an effort to avoid encoding general knowledge in the transformer network itself. We find significant and consistent improvements over both BERT-BASE and BERT-LARGE on multiple reading comprehension (MRC) and paraphrasing datasets. Specifically, our proposed model has strong empirical evidence as it obtains SOTA results on Natural Questions, a new benchmark MRC dataset, outperforming BERT-LARGE by 3 F1 points on short answer prediction. We also establish a new SOTA in HotpotQA, improving answer prediction F1 by 4 F1 points and supporting fact prediction by 1 F1 point. Moreover, we show that our pre-training approach is particularly effective when training data is limited, improving the learning curve by a large amount.