Reasoning over Hierarchical Question Decomposition Tree for Explainable Question Answering

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 14556–14570 July 9-14, 2023 ©2023 Association for Computational Linguistics Reasoning over Hierarchical Question Decomposition Tree for Explainable Question Answering Jiajie Zhang1∗, Shulin Cao2∗, Tingjian Zhang2, Xin Lv2, Jiaxin Shi3, Qi Tian3, Juanzi Li2†, Lei Hou2 1Institute for Interdisciplinary Information Sciences, Tsinghua University, Beijing, China 2Department of Computer Science and Technology, Tsinghua University, Beijing, China 3Cloud BU, Huawei Technologies {jiajie-z19, caosl19}@mails.tsinghua.edu.cn {houlei, lijuanzi}@tsinghua.edu.cn Abstract Explainable question answering (XQA) aims to answer a given question and provide an ex- planation why the answer is selected. Existing XQA methods focus on reasoning on a single knowledge source, e.g., structured knowledge bases, unstructured corpora, etc. However, inte- grating information from heterogeneous knowl- edge sources is essential to answer complex questions. In this paper, we propose to lever- age question decomposing for heterogeneous knowledge integration, by breaking down a complex question into simpler ones, and se- lecting the appropriate knowledge source for each sub-question. To facilitate reasoning, we propose a novel two-stage XQA framework, Reasoning over Hierarchical Question Decom- position Tree (RoHT). First, we build the Hier- archical Question Decomposition Tree (HQDT) to understand the semantics of a complex ques- tion; then, we conduct probabilistic reasoning over HQDT from root to leaves recursively, to aggregate heterogeneous knowledge at differ- ent tree levels and search for a best solution considering the decomposing and answering probabilities. The experiments on complex QA datasets KQA Pro and Musique show that our framework outperforms SOTA methods signifi- cantly, demonstrating the effectiveness of lever- aging question decomposing for knowledge in- tegration and our RoHT framework. 1 Introduction Explainable question answering (XQA) is the task of (i) answering a question and (ii) providing an explanation that enables the user to understand why the answer is selected (Neches et al., 1985; Schuff et al., 2020). It provides a qualified way to test the reasoning ability and interpretability of intelligent systems, and plays an important role in artificial intelligence (Lu et al., 2022). ∗Indicates equal contribution. †Corresponding author. 𝒒𝟎: Which is higher, the highest mountain in North America or the highest mountain in Africa? 𝒒𝟏: How high is the highest mountain in North America? 𝒒𝟐: How high is the highest mountain in Africa? 𝒒𝟓: How high is #4? 𝒒𝟔: Which mountain is the highest in Africa? 𝒒𝟕: How high is #6? 𝒒𝟒: Which mountain is the highest in North America? 𝒒𝟑:[SelectBetween] [greater] #1 #2 Figure 1: An example of Hierarchical Question Decom- position Tree (HQDT). qi represents the index of node in its BFS ordering enumeration. Recent work in XQA can be grouped into two di- rections: 1) neuro-symbolic methods (Berant et al., 2013; Liang et al., 2017; Cao et al., 2022b) translate natural language questions into formal representa- tions (e.g., SPARQL (Sun et al., 2020), KoPL (Cao et al., 2022a), lambda-DCS (Liang, 2013), etc.), whose execution on structured knowledge bases (KBs) gives the answer. Here, the formal repre- sentation acts as an explanation of the final …

Content truncated to 500 words