Exploiting Intrinsic Duality for Multi-Hop Question Generation

The QQ framework utilizes a unified architecture to perform multi-hop question generation (MQG) and question answering (QA) simultaneously. It enforces bidirectional alignment constraints and contrastive learning to ensure strict correspondence between generated questions and target answers. This approach leverages the intrinsic duality between MQG and QA tasks to improve question-answer consistency.