Bridging Textual and Tabular Data for Cross-Domain Text-to-SQL Semantic Parsing

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4870–4888 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 4870 Bridging Textual and Tabular Data for Cross-Domain Text-to-SQL Semantic Parsing Xi Victoria Lin Richard Socher Caiming Xiong Salesforce Research {xilin,rsocher,cxiong}@salesforce.com Abstract We present BRIDGE, a powerful sequential architecture for modeling dependencies be- tween natural language questions and rela- tional databases in cross-DB semantic pars- ing. BRIDGE represents the question and DB schema in a tagged sequence where a sub- set of the fields are augmented with cell val- ues mentioned in the question. The hybrid sequence is encoded by BERT with minimal subsequent layers and the text-DB contextu- alization is realized via the fine-tuned deep attention in BERT. Combined with a pointer- generator decoder with schema-consistency driven search space pruning, BRIDGE at- tained state-of-the-art performance on the well-studied Spider benchmark (65.5% dev, 59.2% test), despite being much simpler than most recently proposed models for this task. Our analysis shows that BRIDGE effectively captures the desired cross-modal dependen- cies and has the potential to generalize to more text-DB related tasks. Our implemen- tation is available at https://github.com/ salesforce/TabularSemanticParsing. 1 Introduction Text-to-SQL semantic parsing addresses the prob- lem of mapping natural language utterances to exe- cutable relational DB queries. Early work in this area focus on training and testing the semantic parser on a single DB (Hemphill et al., 1990; Dahl et al., 1994; Zelle and Mooney, 1996; Zettlemoyer and Collins, 2005; Dong and Lapata, 2016). How- ever, DBs are widely used in many domains and developing a semantic parser for each individual DB is unlikely to scale in practice. More recently, large-scale datasets consisting of hundreds of DBs and the corresponding question- SQL pairs have been released (Yu et al., 2018; Zhong et al., 2017; Yu et al., 2019b,a) to encourage the development of semantic parsers that can work User ID Follows Follower ID User_Profiles UID Name Email Partition ID … … List the name and number of followers for each user SQL SELECT name, followers FROM User_Profiles Followers SQL SELECT COUNT(DISTINCT t2.title) FROM Publication AS T2 JOIN Journal AS T1 ON T2.JID = T1.JID WHERE T1.name = “PVLDB” JID Journal Homepage Name Publication PID Abstract Title … JID Year Return me the number of papers on PVLDB … … Domain Academic Domain Twitter Figure 1: Two questions from the Spider dataset with similar intent resulted in completely different SQL log- ical forms on two DBs. In cross-DB text-to-SQL se- mantic parsing, the interpretation of a natural language question is strictly grounded in the underlying rela- tional DB schema. well across different DBs (Guo et al., 2019; Bogin et al., 2019b; Zhang et al., 2019; Wang et al., 2019; Suhr et al., 2020; Choi et al., 2020). The setup is challenging as it requires the model to interpret a question conditioned on a relational DB unseen during training and accurately express the question intent via SQL logic. Consider the two examples shown in Figure 1, both questions have the …

Content truncated to 500 words