Read + Verify: Machine Reading Comprehension with Unanswerable Questions
share
Machine reading comprehension with unanswerable questions aims to abstain from answering when no answer can be inferred. Previous works using an additional no-answer option attempt to extract answers and detect unanswerable questions simultaneously, but they have struggled to produce high-quality answers and often misclassify questions. In this paper, we propose a novel read-then-verify system that combines a base neural reader with a sentence-level answer verifier trained to further validate if the predicted answer is entailed by input snippets. Moreover, we augment the base reader with two auxiliary losses to better handle answer extraction and no-answer detection respectively, and investigate three different architectures for the answer verifier. Our experiments on the SQuAD 2.0 dataset show that our system can achieve a score of 74.8 F1 on the development set, outperforming the previous best published model by more than 7 points.
READ FULL TEXT
