Valid Text-to-SQL Generation with Unification-Based DeepStochLog

2025-03-17

International Workshop on Neural-Symbolic Learning and Reasoning

AI Summary

To address SQL syntax errors, semantic unexecutability, and logical misalignment between natural language and SQL in text-to-SQL generation, this paper introduces DeepStochLog—a deep probabilistic logic programming framework—to the task for the first time. It intrinsically encodes SQL syntactic constraints via first-order logic unification and integrates neural-symbolic reasoning with grammar-guided decoding, thereby ensuring generated SQL queries are syntactically correct, database-executable, and semantically aligned with the input utterance. Evaluated on the Spider benchmark, our approach improves the valid SQL rate by 12.6%, achieves 100% executability of generated queries, and attains state-of-the-art logical consistency. These results significantly enhance the reliability and practical deployability of text-to-SQL systems.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Knowledge Representation and Reasoning: Logic Programming
• Machine Learning: Statistical Relational/Logic Learning

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Querying, indexing, and retrieval in Web-related graphs

Problem

• Ensures valid SQL query generation using unification-based grammars.
• Addresses invalid SQL queries from large language models.
• Improves execution accuracy and alignment with ground truth.

Innovation

• Neurosymbolic framework for SQL generation
• Unification-based definite clause grammars
• Bi-directional interface to language models