A Systematic Approach to Conversational Question Answering

Abstract

Question answering is a well known natural language processing problem that already has extensive research. However, conversational question answering is a fairly novel topic, and has ample room for exploration and analysis. Conversational question answering is a challenging problem compared to other more common forms of question answering. Whereas typical question answering looks to answer a singular question, conversational question answering needs to account for previous question/answer pairs in a conversation to answer. We look into three models: RoBERTa, SDNet, and FlowQA to model the conversational question answering task, and also implement novel techniques onto these models. We aim to provide thorough analysis and experimentation for future conversational question answering literature to build off of.