UNJOIN: Enhancing Multi-Table Text-to-SQL Generation via Schema Simplification

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

UNJOIN is a two-stage framework designed to enhance Text-to-SQL generation for multi-table databases, addressing challenges posed by complex schemas and relational operations. The framework decouples schema element retrieval from SQL logic generation. In its first stage, UNJOIN simplifies the database schema by merging all column names into a single-table representation, prefixing each column with its original table name. This allows the model to focus solely on accurate retrieval. The second stage generates the SQL query on this simplified schema, subsequently mapping it back to the original schema by reconstructing necessary JOINs, UNIONs, and relational logic. Evaluated on the SPIDER and BIRD datasets, UNJOIN matches or exceeds state-of-the-art baselines. A key advantage is its reliance solely on schema information, eliminating the need for data access or fine-tuning, which ensures scalability and adaptability across diverse databases.

Key takeaway

For Machine Learning Engineers developing Text-to-SQL solutions for complex multi-table databases, consider adopting UNJOIN's two-stage schema simplification approach. This method can significantly improve query accuracy by allowing your models to focus on schema element retrieval before reconstructing relational logic. You can achieve state-of-the-art performance without requiring extensive data access or fine-tuning, making it a scalable option for diverse database environments.

Key insights

The UNJOIN framework simplifies multi-table Text-to-SQL by decoupling schema retrieval from SQL logic generation via a two-stage schema simplification and reconstruction process.

Principles

• Decouple schema retrieval from SQL logic.
• Simplify complex schemas for LLM processing.
• Reconstruct relational logic post-generation.

Method

UNJOIN merges all database column names into a single-table representation, prefixing each with its table name. SQL is generated on this simplified schema, then mapped back by reconstructing JOINs and UNIONs.

In practice

• Apply schema simplification for multi-table queries.
• Use prefixing for column disambiguation.
• Evaluate Text-to-SQL on SPIDER/BIRD.

Topics

• Text-to-SQL
• Large Language Models
• Schema Simplification
• Multi-Table Databases
• SQL Generation
• SPIDER Dataset

Code references

• coral-lab-asu/unjoin

Best for: AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, NLP Engineer

Related on AIssential

• SIRIUS-SQL: Anchoring Multi-Candidate Text-to-SQL in Execution Feedback — Multi-candidate Text-to-SQL reliability improves by diversifying generation, targeting repairs, and hybrid selection.
• Executable Schema Contracts: From Automatic Ingestion to Multi-Source Retrieval — The system automatically derives an executable schema from multi-source data to unify knowledge graph construction and multi-tool query retrieval.
• Text-to-SQL solution powered by Amazon Bedrock — Text-to-SQL solutions using Amazon Bedrock democratize data access by translating natural language questions into actionable insights.
• LeGo-Code: Can Modular Curriculum Learning Advance Complex Code Generation? Insights from Text-to-SQL — Modular curriculum learning with tier-specific adapters improves LLM performance on complex Text-to-SQL tasks.
• Understanding, Detecting, and Repairing Real-World In-Context-Learning-Based Text-to-SQL Errors — LLM-generated SQL queries have widespread, categorized errors, requiring targeted, efficient repair solutions.
• Progress-SQL: Improving Reinforcement Learning for Text-to-SQL via Progressive Rewards — Multi-turn reinforcement learning with progressive rewards and ODT feedback significantly improves Text-to-SQL generation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.