Get Started
Home
Topics
Search
Library
Research questionHow can Text-to-SQL systems learn from database execution when annotated question-SQL pairs are unavailable?Text-to-SQL training commonly depends on costly annotations tied to particular databases and domains. Database execution provides an alternative correctness signal, but it is unclear how far that signal can support learning and transfer without paired labels.
Code Generation & Program Synthesis
Evaluation & Benchmarks
LLM Pretraining & Post-training
Machine Learning
Multi-agent Systems
Natural Language Processing
Reinforcement Learning
Latest papersRecent research connected to this question, newest first.SQL-Zero: Self-Evolving Text-to-SQLThe source evaluates a self-play training approach in which SQL execution supplies the ground truth, using BIRD databases for training and testing transfer on unseen Spider databases and lexical perturbations. Evidence is limited to the evaluated language-model scales, databases, transfer settings, and training iterations.research paper · Sep 4, 2026
Related questions
How can natural-language interfaces generate reliable SQL and GraphQL without runtime LLM inference?How can natural-language interfaces generate reliable SQL over nested, graph-like enterprise schemas at acceptable cost?How can instruction-tuned LLMs learn corpus-specific knowledge without exhaustive synthetic QA or instruction fine-tuning?How can governed language-model analytics preserve expressive queries while producing replayable, evidence-backed answers?