Neural Natural Logic Inference for Interpretable Question Answering

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3673–3684 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 3673 Neural Natural Logic Inference for Interpretable Question Answering Jihao Shi, Xiao Ding˚, Li Du, Ting Liu, and Bing Qin Research Center for Social Computing and Information Retrieval Harbin Institute of Technology, China {jhshi, xding, ldu, tliu, qinb}@ir.hit.edu.cn Abstract Many open-domain question answering prob- lems can be cast as a textual entailment task, where a question and candidate answers are concatenated to form hypotheses. A QA sys- tem then determines if the supporting knowl- edge bases, regarded as potential premises, en- tail the hypotheses. In this paper, we investi- gate a neural-symbolic QA approach that in- tegrates natural logic reasoning within deep learning architectures, towards developing ef- fective and yet explainable question answer- ing models. The proposed model gradually bridges a hypothesis and candidate premises following natural logic inference steps to build proof paths. Entailment scores between the acquired intermediate hypotheses and candi- date premises are measured to determine if a premise entails the hypothesis. As the natural logic reasoning process forms a tree-like, hier- archical structure, we embed hypotheses and premises in a Hyperbolic space rather than Eu- clidean space to acquire more precise represen- tations. Empirically, our method outperforms prior work on answering multiple-choice sci- ence questions, achieving the best results on two publicly available datasets. The natural logic inference process inherently provides ev- idence to help explain the prediction process. 1 Introduction Question answering (QA) is an important real-life NLP application but also a challenging task for as- sessing how well AI systems understand human language and perform reasoning to answer ques- tions. A main challenge of QA is that the answers often do not explicitly exist in a supporting knowl- edge base but instead need to be inferred from it. Prior work (Angeli et al., 2016) has viewed QA as a textual entailment problem performed on a large premise set, where a question and candidate answers are formulated as hypotheses that need to be proved. *Corresponding author. Neural networks have recently become the main- stream models for QA (Lukovnikov et al., 2017; Jia et al., 2018; Yang et al., 2019). Most of the models, however, are unable to give explainable inference results. Developing effective and yet explainable question answering models has attracted more at- tention (Abujabal et al., 2017; Yang et al., 2018; Zhou et al., 2018; Sydorova et al., 2019; Weber et al., 2019). In this paper, we investigate a neural-symbolic QA approach that integrates natural logic reason- ing (Lakoff, 1970; Nairn et al., 2006; MacCartney and Manning, 2009) within deep learning architec- tures for QA, aiming to keep the backbone of infer- ence based on the natural logic formalism, while integrating neural networks to make the systems powerful and robust. Conventional natural logic has been designed for natural language inference and question answering (MacCartney and Manning, 2009; Angeli and Manning, 2014). As opposed to performing deduction on an abstract logical form, e.g., …

Content truncated to 500 words