LLMs produce structurally diverse SQL queries for identical inputs on the Spider benchmark even when execution is correct, and a compile-style pipeline improves both accuracy and structural consistency.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
SQLStructEval: Structural Evaluation of LLM Text-to-SQL Generation
LLMs produce structurally diverse SQL queries for identical inputs on the Spider benchmark even when execution is correct, and a compile-style pipeline improves both accuracy and structural consistency.