Scalable Multi-Hop Relational Reasoning for Knowledge-Aware Question Answering

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 1295–1309, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 1295 Scalable Multi-Hop Relational Reasoning for Knowledge-Aware Question Answering Yanlin Feng∫ò Xinyue Chenπò Bill Yuchen Lin∏ Peifeng Wang∏ Jun Yan∏ Xiang Ren∏ fengyanlin@pku.edu.cn, xinyuech@andrew.cmu.edu, {yuchen.lin, peifengw, yanjun, xiangren}@usc.edu ∏University of Southern California ∫Peking University πCarnegie Mellon University Abstract Existing work that augment question answer- ing (QA) models with external knowledge (e.g., knowledge graphs) either struggle to model multi-hop relations efficiently, or lack transparency into the model’s prediction ra- tionale. In this paper, we propose a novel knowledge-aware approach that equips pre- trained language models (PTLMs) with a multi-hop relational reasoning module, named multi-hop graph relation network (MHGRN). It performs multi-hop, multi-relational reason- ing over subgraphs extracted from external knowledge graphs. The proposed reasoning module unifies path-based reasoning methods and graph neural networks and results in better interpretability and scalability. We also empir- ically show its effectiveness and scalability on CommonsenseQA and OpenbookQA datasets, and interpret its behaviors with case studies, with the code for experiments released1. 1 Introduction Many recently proposed question answering tasks require not only machine comprehension of the question and context, but also relational reason- ing over entities (concepts) and their relationships by referencing external knowledge (Talmor et al., 2019; Sap et al., 2019; Clark et al., 2018; Mihaylov et al., 2018). For example, the question in Fig. 1 requires a model to perform relational reasoning over mentioned entities, i.e., to infer latent rela- tions among the concepts: {CHILD, SIT, DESK, SCHOOLROOM}. Background knowledge such as “a child is likely to appear in a schoolroom” may not be readily contained in the questions themselves, but are commonsensical to humans. Despite the success of large-scale pre-trained language models (PTLMs) (Devlin et al., 2019; ò The first two authors contributed equally. The major work was done when both authors interned at USC. 1https://github.com/INK-USC/MHGRN Desires Learn Child Schoolroom Classroom Si� Desk AtLocation Where does a child likely sit at a desk? A� Schoolroom � B� Furniture store C� Patio D� Office building E� Library Figure 1: Illustration of knowledge-aware QA. A sample question from CommonsenseQA can be better answered if a relevant subgraph of ConceptNet is pro- vided as evidence. Blue nodes correspond to entities mentioned in the question, and pink nodes correspond to those in the answer. The other nodes are associated entities introduced in subgraph extraction. ì indicates the correct answer. Liu et al., 2019b), these models fall short of pro- viding interpretable predictions, as the knowledge in their pre-training corpus is not explicitly stated, but rather is implicitly learned. It is thus difficult to recover the evidence used in the reasoning process. This has led many to leverage knowledge graphs (KGs) (Mihaylov and Frank, 2018; Lin et al., 2019; Wang et al., 2019; Yang et al., 2019). KGs represent relational knowledge between en- tities with multi-relational edges for models to acquire. Incorporating KGs brings the potential of interpretable and trustworthy predictions, as the knowledge is …

Content truncated to 500 words