Evaluating Co-reference Chains Based Conversation History in Conversational Question Answering

This paper examines the effect of using co-reference chains based conversational history against the use of entire conversation history for conversational question answering (CoQA) task. The QANet model is modified to include conversational history and NeuralCoref is used to obtain co-reference chains based conversation history. The results of the study indicates that in spite of the availability of a large proportion of co-reference links in CoQA, the abstract nature of questions in CoQA renders it difficult to obtain correct mapping of co-reference related conversation history, and thus results in lower performance compared to systems that use entire conversation history. The effect of co-reference resolution examined on various domains and different conversation length, shows that co-reference resolution across questions is helpful for certain domains and medium-length conversations.