Retrospective Reader for Machine Reading Comprehension

27 Jan 2020  ·  Zhuosheng Zhang, Junjie Yang, Hai Zhao ·

Machine reading comprehension (MRC) is an AI challenge that requires machine to determine the correct answers to questions based on a given passage. MRC systems must not only answer question when necessary but also distinguish when no answer is available according to the given passage and then tactfully abstain from answering... When unanswerable questions are involved in the MRC task, an essential verification module called verifier is especially required in addition to the encoder, though the latest practice on MRC modeling still most benefits from adopting well pre-trained language models as the encoder block by only focusing on the "reading". This paper devotes itself to exploring better verifier design for the MRC task with unanswerable questions. Inspired by how humans solve reading comprehension questions, we proposed a retrospective reader (Retro-Reader) that integrates two stages of reading and verification strategies: 1) sketchy reading that briefly investigates the overall interactions of passage and question, and yield an initial judgment; 2) intensive reading that verifies the answer and gives the final prediction. The proposed reader is evaluated on two benchmark MRC challenge datasets SQuAD2.0 and NewsQA, achieving new state-of-the-art results. Significance tests show that our model is significantly better than the strong ELECTRA and ALBERT baselines. A series of analysis is also conducted to interpret the effectiveness of the proposed reader. read more

PDF Abstract

Datasets


Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Question Answering SQuAD2.0 Retro-Reader on ALBERT (ensemble) EM 90.115 # 17
F1 92.580 # 16
Question Answering SQuAD2.0 Retro-Reader on ELECTRA (single model) EM 89.562 # 26
F1 92.052 # 32
Question Answering SQuAD2.0 Retro-Reader (ensemble) EM 90.578 # 7
F1 92.978 # 5
Question Answering SQuAD2.0 Retro-Reader on ALBERT (single model) EM 88.107 # 61
F1 91.419 # 46

Methods