Grammar-based Neural Text-to-SQL Generation

TL;DR

This paper introduces grammar-based decoding techniques for neural text-to-SQL models, improving hierarchical SQL generation by handling schema dependencies and reducing errors on challenging datasets.

Contribution

It presents methods to construct schema-dependent grammars with minimal over-generation, enhancing neural text-to-SQL performance.

Findings

14-18% relative error reduction on ATIS and Spider datasets

Effective handling of SQL complexities with grammar-based decoding

Improved hierarchical SQL generation in neural models

Abstract

The sequence-to-sequence paradigm employed by neural text-to-SQL models typically performs token-level decoding and does not consider generating SQL hierarchically from a grammar. Grammar-based decoding has shown significant improvements for other semantic parsing tasks, but SQL and other general programming languages have complexities not present in logical formalisms that make writing hierarchical grammars difficult. We introduce techniques to handle these complexities, showing how to construct a schema-dependent grammar with minimal over-generation. We analyze these techniques on ATIS and Spider, two challenging text-to-SQL datasets, demonstrating that they yield 14--18\% relative reductions in error.