MPPQA: Structure-Aware Extractive Multi-span Question Answering for Procedural Documents

Bihan Zhou, Haopeng Ren, Yi Cai, Zetao Lian, Pinli Zhu, Yushi Zeng

Abstract: Procedural documents, which provide procedural knowledge with step-by-step instructions to help achieve specific goals, have shown growing popularity in both industry workflows and daily routines. Particularly, answering prerequisite associated questions facilitates thorough preparation for procedures and supports the execution of the subsequent operations. However, most prevailing researches in procedural knowledge primarily concentrate on the task of entity state tracking or “What if” reasoning, typically formulated as single-span extraction or multiple-choice question answering. Due to the limitation of single-span extraction format or disregard for leveraging the structural procedural knowledge within procedural document, existing question answering methods still struggle to accurately answer multi-span prerequisite associated questions. In this paper, we explore the problem of prerequisite associated question answering task to extract multi-span style answers from the procedural documents, and propose a key procedure concentrated question answering model SAPE integrated with structure-aware module. Additionally, a new dataset MPPQA is built and extensive experiments are conducted to evaluate the effectiveness of our proposed model.