Understanding, Detecting, and Repairing Real-World In-Context-Learning-Based Text-to-SQL Errors

This paper presents a comprehensive study on the errors associated with in-context learning (ICL) in text-to-SQL tasks, identifying 27 error types across seven categories and evaluating existing repair methods. It introduces MapleDoctor, a new framework that improves error detection and repair, achieving a 13.8% increase in successful repairs while reducing repair latency by 67.4%. This advancement is significant for practitioners as it enhances the reliability and efficiency of deploying LLMs for SQL query generation.