A Natural Language Interactive Interface for SQL Query GenerationDownload PDF

Anonymous

04 Mar 2022 (modified: 05 May 2023)ICLR 2022 Workshop DL4C Blind SubmissionReaders: Everyone
Keywords: SQL, program synthesis, natural language processing, semantic parsing, human-centered design, human-computer interaction
TL;DR: Using an interactive interface that allows a user to specify their desired table column-by-column, our tool can generate complex multi-sentence SQL queries.
Abstract: We contribute \texttt{nalini}, an natural-language based interactive interface for SQL query generation. Motivated by a lack of usability of existing systems, \texttt{nalini} was built with the intention of using it for complex query generation. The interface allows users to use natural language fragments with minimal structure; users separately describe each desired column in the output table and can optionally describe additional filters. We evaluated \texttt{nalini} with a first-use study with five participants, where participants were asked to generate queries from the TPC-H decision support benchmark. Our study showed that users were able to use \texttt{nalini} to generate complex queries, including TPC-H queries which cannot be generated by any of the leading natural-language-to-SQL tools, and points to promising areas of future work.
1 Reply

Loading