New GRIPS method boosts Text-to-SQL accuracy by 11.8% with reduced token usage

Researchers have developed GRIPS, a three-stage method to improve Text-to-SQL systems by explicitly grounding question mentions to tables, columns, values, and relations before SQL generation. This approach addresses a significant source of errors in SQL Static failures, particularly grounding errors. When tested with GPT-5-nano and Llama-70B models on the TempTabQA-C and BIRD benchmarks, GRIPS-Adaptive demonstrated an average accuracy gain of 11.8 percentage points across 18 out of 20 settings, while also reducing model-call token usage by 47.8%. The study indicates that while specific grounding information is beneficial, adding excessive pre-SQL information does not consistently improve performance. AI

IMPACT Enhances the accuracy and efficiency of SQL generation from natural language, potentially improving data analysis tools.

RANK_REASON Academic paper detailing a new method for improving Text-to-SQL systems. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• BIRD
• GPT-5 nano
• Hugging Face
• LLaMA-70B
• SQL Static
• TempTabQA-C

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →