SPOC-SQL: Stage-wise Preference Optimization for Controllable Text-to-SQL
2026-08-24 • Computation and Language
Computation and Language
AI summaryⓘ
The authors address the task of turning natural language questions into SQL queries by breaking the process into four clear steps instead of doing it all at once. They design a method called SPOC-SQL that helps the model learn better decisions at key points in query building by focusing on smaller, manageable parts. This approach also allows humans to intervene and fix mistakes during the process, making the final SQL queries more accurate and reliable. Their experiments show that this step-by-step method improves the performance compared to treating the task as a single step.
Text-to-SQLSQL querynatural language processingstructured reasoningdatabase schemaquery constraintsstage-wise optimizationpreference optimizationintermediate representationhuman-in-the-loop
Authors
Yingnan Chen, Chun Ding, Tianshi Xu, Xu Yang, Si Wu
Abstract
Text-to-SQL aims to translate natural language questions into executable SQL queries over relational databases, requiring multi-stage structured reasoning over database schemas and query constraints. However, existing methods treat this task as single-step generation, where models optimize entire SQL sequences without targeted feedback at key decision points and lack support for interacting with and controlling the intermediate generation process. To address this issue, we propose SPOC-SQL, which decomposes Text-to-SQL into four sequential subtasks following standard SQL execution logic and designs stage-specific optimization strategies for the model to learn key decisions. Specifically, we propose the implementation of fine-grained preference optimisation at key decision points across SQL stages, with the objective of enhancing structured decision-making during query construction. Furthermore, a structured decomposition strategy is designed, facilitating stage-wise intervention and correction through explicit intermediate representations. This results in more controllable and reliable SQL generation. Experiments demonstrate that incorporating stage-wise human knowledge consistently improves performance, validating the effectiveness of stage perception controllable generation.