AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Addressing Semantic Blind Spots in Text-to-SQL via Component Pre-generation and AST Matching Rewards

Xingyu Ma, Xin Tian, Lingxiang Wu, Xuepeng Wang, Zhilin Zhang, Xueming Tang, Jinqiao Wang

ICML 2026 · Large language models › Language, safety and interpretability · approach: Text generation and evaluation metrics

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add SQL component pre-generation reward (ablation)base reward GRPO · BIRD-DEV (challenging)EX: 55.56 -> 59.03Improved
add AST matching reward (ablation)base reward GRPO · BIRD-DEVEX (total): 63.89 -> 64.02Improved
add IDF node weighting to AST reward (ablation)base + AST reward · BIRD-DEVEX (total): 64.02 -> 64.21Improved
add IDF weighting to pre-gen + AST reward (ablation)pre-gen + AST · BIRD-DEV (challenging)EX: 60.42 -> 63.19Improved
add IDF weighting to pre-gen + AST reward (ablation)pre-gen + AST · BIRD-DEVEX (total): 64.99 -> 65.78Improved
replace n-gram with AST matching reward (ablation)pre-generation · BIRD-DEVEX (total): 63.80 -> 64.99Improved
add pre-generation with n-gram matching vs. base alone (ablation)base reward GRPO · BIRD-DEVEX (total): 63.89 -> 63.80No clear effect
apply pre-generation to all questions (remove complexity analysis) (ablation)full method · BIRD-DEVEX (total): 65.78 -> 64.86Hurt
use rule-based instead of model&rule complexity (ablation)full method · BIRD-DEVEX (total): 65.78 -> 65.25Hurt
use model-based complexity only (ablation)full method · BIRD-DEVEX (total): 65.78 -> 64.73Hurt
add n-gram local matching reward (ablation)w/o matching reward · SPIDER-DEVEM: 11.7 -> 37.2Improved
replace n-gram with AST matching reward (ablation)local matching reward · SPIDER-TESTEM: 36.3 -> 45.1Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy