Leveraging Frequent Query Substructures to Generate Formal Queries for Complex Question Answering

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 2614–2622, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 2614 Leveraging Frequent Query Substructures to Generate Formal Queries for Complex Question Answering Jiwei Ding Wei Hu∗ Qixin Xu Yuzhong Qu∗ State Key Laboratory for Novel Software Technology, Nanjing University, China {jwding, qxxu}nju@outlook.com, {whu, yzqu}@nju.edu.cn Abstract Formal query generation aims to generate cor- rect executable queries for question answer- ing over knowledge bases (KBs), given en- tity and relation linking results. Current ap- proaches build universal paraphrasing or rank- ing models for the whole questions, which are likely to fail in generating queries for com- plex, long-tail questions. In this paper, we propose SubQG, a new query generation ap- proach based on frequent query substructures, which helps rank the existing (but nonsignif- icant) query structures or build new query structures. Our experiments on two bench- mark datasets show that our approach signif- icantly outperforms the existing ones, espe- cially for complex questions. Also, it achieves promising performance with limited training data and noisy entity/relation linking results. 1 Introduction Knowledge-based question answering (KBQA) aims to answer natural language questions over knowledge bases (KBs) such as DBpedia and Freebase. Formal query generation is an important component in many KBQA systems (Bao et al., 2016; Cui et al., 2017; Luo et al., 2018), especially for answering complex questions. Given entity and relation linking results, formal query gener- ation aims to generate correct executable queries, e.g., SPARQL queries, for the input natural lan- guage questions. An example question and its for- mal query are shown in Figure 1. Generally speak- ing, formal query generation is expected to include but not be limited to have the capabilities of (i) recognizing and paraphrasing different kinds of constraints, including triple-level constraints (e.g., “movies” corresponds to a typing constraint for the target variable) and higher level constraints (e.g., subgraphs). For instance, “the same ... as” ∗Corresponding authors How many movies have the same director as The Shawshank Redemption? dbo:Film ?Var2 ?Var1 ISA dbo:director dbr:TSR ?Var3 COUNT dbo:director Figure 1: An example for complex question and query represents a complex structure shown in the mid- dle of Figure 1; (ii) recognizing and paraphras- ing aggregations (e.g., “how many” corresponds to COUNT); and (iii) organizing all the above to generate an executable query (Singh et al., 2018; Zafar et al., 2018). There are mainly two kinds of query generation approaches for complex questions. (i) Template- based approaches choose a pre-collected template for query generation (Cui et al., 2017; Abuja- bal et al., 2017). Such approaches highly rely on the coverage of templates, and perform unsta- bly when some complex templates have very few natural language questions as training data. (ii) Approaches based on semantic parsing and neu- ral networks learn entire representations for ques- tions with different query structures, by using a neural network following the encode-and-compare framework (Luo et al., 2018; Zafar et al., …

Content truncated to 500 words