Conversational machine comprehension requires the understanding of the history, such as previous question/answer pairs, the document, and the current question. To enable traditional, single-turn models to the history comprehensively, we introduce Flow, a mechanism that can intermediate representations generated during the process of previous questions, through an alternating parallel processing. Compared to approaches that concatenate previous questions/answers input, Flow integrates the latent semantics of the conversation history more. Our model, FlowQA, shows superior performance on two recently proposed challenges (+7.2% F1 on CoQA and +4.0% on QuAC). The of Flow also shows in other tasks. By reducing sequential understanding to conversational machine comprehension, FlowQA the best models on all three domains in SCONE, with +1.8% to +4.4% in accuracy.
No takes yet. Share an insight, caveat, or question.
Huang et al. (2018) studied this question.