AI TrackerFrontier desk · AEST
Back to news
normaltech

FALCON: A Model and Dataset Agnostic Framework for Synthetic Data Generation for NL2SQL Pairs

arXiv cs.CL · 3 Oct 2026, 03:19 AEST

FALCON is a framework designed for generating realistic, ambiguity-aware NL-to-SQL data that matches the complexity of real-world benchmarks. It allows organizations to create high-complexity training data locally without relying on external APIs.

Key facts

Read aloud

FALCON is a framework designed for generating realistic, ambiguity-aware NL-to-SQL data that matches the complexity of real-world benchmarks. It allows organizations to create high-complexity training data locally without relying on external A P I s. Relational databases are among the most widely deployed forms of structured knowledge, and natural language access to them requires grounding language onto schema entities and relations while handling the ambiguity inherent in how people phrase requests. Existing synthetic NL-to-SQL data generation methods largely ignore this ambiguity and produce oversimplified queries that fail to prepare models for the complexity of real-world structured knowledge access. Human evaluation confirms consistent high quality across model sizes, and our generated data exceeds existing benchmarks in both SQL complexity and natural language richness. Difficulty-stratified analysis shows models trained on FALCON data increasingly outperform baseline-trained models as query complexity increases, validating our pipeline's success in generating challenging training data.

Sources in this cluster