DuoRAT: Towards Simpler Text-to-SQL Models

TL;DR

This paper introduces DuoRAT, a simplified text-to-SQL model based on relation-aware transformers, demonstrating that certain complex features are redundant, thus streamlining the model without sacrificing performance.

Contribution

DuoRAT is a simplified, re-implemented state-of-the-art text-to-SQL model using only relation-aware or vanilla transformers, with ablation studies identifying essential and redundant features.

Findings

Simplified model maintains performance with fewer features

Structural SQL features and question-schema links are often redundant

Relation-aware transformers are effective building blocks

Abstract

Recent neural text-to-SQL models can effectively translate natural language questions to corresponding SQL queries on unseen databases. Working mostly on the Spider dataset, researchers have proposed increasingly sophisticated solutions to the problem. Contrary to this trend, in this paper we focus on simplifications. We begin by building DuoRAT, a re-implementation of the state-of-the-art RAT-SQL model that unlike RAT-SQL is using only relation-aware or vanilla transformers as the building blocks. We perform several ablation experiments using DuoRAT as the baseline model. Our experiments confirm the usefulness of some techniques and point out the redundancy of others, including structural SQL features and features that link the question with the schema.