Leveraging Adjective-Noun Phrasing Knowledge for Comparison Relation Prediction in Text-to-SQL

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 3515–3520, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 3515 Leveraging Adjective-Noun Phrasing Knowledge for Comparison Relation Prediction in Text-to-SQL Haoyan Liu1∗, Lei Fang2, Qian Liu3, Bei Chen2, Jian-Guang Lou2, Zhoujun Li1 1State Key Lab of Software Development Environment, Beihang University, China 2Microsoft Research, Beijing, China 3State Key Lab of Virtual Reality Technology and Systems, Beihang University, China {haoyan.liu, qian.liu, lizj}@buaa.edu.cn; {leifa, beichen, jlou}@microsoft.com Abstract One key component in text-to-SQL is to pre- dict the comparison relations between column- s and their values. To the best of our knowl- edge, no existing models explicitly introduce external common knowledge to address this problem, thus their capabilities of predicting comparison relations are limited beyond train- ing data. In this paper, we propose to lever- age adjective-noun phrasing knowledge mined from the web to predict the comparison rela- tions in text-to-SQL. Experimental results on both the original and the re-split Spider dataset show that our approach achieves significant improvement over state-of-the-art methods on comparison relation prediction. 1 Introduction Text-to-SQL (Yaghmazadeh et al., 2017; Zhong et al., 2017), which aims at mapping natural lan- guage to SQL queries, is one of the most important tasks in natural language processing. Most state- of-the-art models are end-to-end neural network based models (Zhang et al., 2017; Xu et al., 2017; Yu et al., 2018a; Herzig and Berant, 2018; Dong and Lapata, 2018; Yu et al., 2018b), which mainly extend the Seq2Seq architecture with some com- plicated network structures. As shown by Yu et al. (2018b), the performances of most methods are overstated, because they just match semantic pars- ing results, rather than truly understand the mean- ings of inputs (Finegan-Dollak et al., 2018). In this paper, we study the comparison relation predic- tion problem, as the comparison relations between columns and their values are not well understood by existing methods. In SQL queries, comparison relations are ex- pressed by the comparison operators (=, ̸=, <, ≤ , >, ≥) and value ordering keywords (ASC, DESC ∗This work was done when the first author was an intern at Microsoft Research Asia. in ORDER BY expression). In most text-to-SQL models, the comparison relations are either gen- erated using Seq2Seq architectures (Zhong et al., 2017; Dong and Lapata, 2018), or predicted us- ing classifiers trained with output decoding fea- tures (Xu et al., 2017; Yu et al., 2018b). We give an example to show that external common knowl- edge is indispensable to truly understand the com- parison relations on unseen data. rank name time birth date 1 Martina 11′9 19800930 2 Mirjana 12′5 19820309 3 Justine 12′9 19820601 Table 1: Player Basic Information. Table 1 shows the basic information about the athletes of the 100-meter sprint. Given the query “what is the name of the oldest player ?”, the goal is to generate the SQL “SELECT name FROM players ORDER BY birth date ASC …

Content truncated to 500 words