On Repairing Natural Language to SQL Queries
Authors:
Aidan Z. H. Yang,
Ricardo Brancas,
Pedro Esteves,
Sofia Aparicio,
Joao Pedro Nadkarni,
Miguel Terra-Neves,
Vasco Manquinho,
Ruben Martins
Abstract:
Data analysts use SQL queries to access and manipulate data on their databases. However, these queries are often challenging to write, and small mistakes can lead to unexpected data output. Recent work has explored several ways to automatically synthesize queries based on a user-provided specification. One promising technique called text-to-SQL consists of the user providing a natural language des…
▽ More
Data analysts use SQL queries to access and manipulate data on their databases. However, these queries are often challenging to write, and small mistakes can lead to unexpected data output. Recent work has explored several ways to automatically synthesize queries based on a user-provided specification. One promising technique called text-to-SQL consists of the user providing a natural language description of the intended behavior and the database's schema. Even though text-to-SQL tools are becoming more accurate, there are still many instances where they fail to produce the correct query.
In this paper, we analyze when text-to-SQL tools fail to return the correct query and show that it is often the case that the returned query is close to a correct query. We propose to repair these failing queries using a mutation-based approach that is agnostic to the text-to-SQL tool being used. We evaluate our approach on two recent text-to-SQL tools, RAT-SQL and SmBoP, and show that our approach can repair a significant number of failing queries.
△ Less
Submitted 5 October, 2023;
originally announced October 2023.
