SOMA-SQL: Resolving Multi-Source Ambiguity in NL-to-SQL via Synthetic Log and Execution Probing

In three linesSOMA-SQL resolves multi-source ambiguity in NL-to-SQL translation via synthetic query logs and execution probing. The method constructs synthetic logs to ground schema interpretation, generates SQL candidates, then executes targeted probing queries driven by an ambiguity taxonomy. Results: +13.0% execution accuracy on average across 6 benchmarks, up to +16.7% on ambiguous questions.

Summary generated by Claude — human-verified