COAL-SQL is proposed, a unified framework combining Coverage-Guided Augmentation (CGA) and Failure-Driven Learning (FDL) that achieves 64.9% execution accuracy on the BIRD development set and outperforms baselines trained at comparable scale.
Abstract
Text-to-SQL translates natural-language questions into executable SQL queries, but open-source large language models still require task-specific post-training for complex, real-world SQL generation. Effective post-training requires both training data that cover the capabilities demanded by the target task and a learning strategy that enables the model to acquire them. Existing datasets provide valuable supervision but incompletely cover SQL structures, while augmentation methods typically expand data without identifying structural gaps. Moreover, supervised fine-tuning (SFT) or reinforcement learning (RL) alone cannot dynamically address weaknesses exposed during training. We propose COAL-SQL, a unified framework combining Coverage-Guided Augmentation (CGA) and Failure-Driven Learning (FDL). CGA uses greedy selection to identify SQL structures missing from the original dataset and constructs complementary examples, improving structural coverage. FDL retains GRPO as the main optimization objective while supplying targeted supervision for unsolved examples. At the step level, it applies SFT to verified reasoning traces generated by a strong LLM for accumulated failures. At the epoch level, it retrieves structurally related examples based on accumulated failures to create targeted practice, helping the model acquire the corresponding SQL capabilities. With only 12,600 distinct post-training examples, COAL-SQL achieves 64.9% execution accuracy on the BIRD development set and outperforms baselines trained at comparable scale. The code is available at https://github.com/TechNomad-ds/COAL-SQL.
Open-source LLMs provide a flexible alternative to closed-source models for Text-to-SQL but still struggle with complex queries. Finetuning with chain-of-thought (CoT) data improves reasoning, yet standard datasets lack CoT annotations, requiring synthetic generation. Existing synthesis methods only verify the final an...
Rui Zhang, Yu-Jia Su, Rui-Xuan Li et al.· Proceedings of the 2026 Conf...· 0 citations
Text-to-SQL translates natural language questions into SQL queries, enabling database access without writing SQL. Fine-tuning open-weight large language models (LLMs) is cost-effective, but existing methods treat SQL as ordinary text and optimize token-level likelihood without explicitly modeling its syntactic, semanti...
Text-to-SQL models are commonly trained to map questions directly to static queries, whereas real-world database agents operate through stateful, multi-turn interaction with live databases -- inspecting schemas, executing probe queries, diagnosing errors, and revising hypotheses. This creates a critical train-deploy mi...
Hao-Lin Yang, Ji-Peng Zhang, Zhi-Qiang Qian et al.· 0 citations
Natural-language access to structured manufacturing data could let plant personnel query batch records without structured query language (SQL) expertise, yet realistic multi-table industrial schemas remain hard for text-to-SQL systems, and regulated environments demand data sovereignty, traceability, and validatable be...
L. Rabelo, J. Zamora-Aguas, Fahad Alasim et al.· Applied Sciences· 0 citations
Text-to-SQL data synthesis must expand semantic coverage without losing the relational structure of observed queries. Existing SQL offers evidence of plausible join and nesting patterns, but reusing queries alone limits the range of training examples; unconstrained generation, in contrast, risks implausible structures...
Hao-Yu Zhang, Ying-Qi Zou, Xiang-Yu Wang et al.· Information· 0 citations
Text-to-SQL systems typically assume a one-to-one mapping between natural-language queries and correct SQL translations. In practice, however, many queries are inherently ambiguous, admitting multiple valid interpretations that yield distinct outputs. We introduce Adversarial Intent Resolution (AIR), a multi-agent fram...
You-Bing Yin, Xu-Jun Peng, Muhammad Shahin Uddin et al.· Proceedings of the 2026 Conf...· 0 citations
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.