Learning to Generate Reformulation Actions for Scalable Conversational Query Understanding

The ability of conversational query understanding (CQU) is indispensable to multi-turn QA. However, existing methods are data-driven and expensive to extend to new conversation domains, or under specific frameworks like semantic parsing, and hard to apply to different underlying QA technologies. We propose a novel contextual query reformulation (CQR) module based on reformulation actions for general CQU. The actions are designed to capture syntactic regularities of conversations, so they are domain-independent and scalable. For action generation, we propose a multi-task learning framework based on sequence generation and coreference resolution, and introduce grammar constraints into decoding. Then CQR synthesizes standalone queries based on generated actions, which naturally adapts to original downstream technologies. Experiments on different CQU datasets suggest that action-based methods substantially outperform direct query generation, and the action generation model performs the best among these methods.