Search papers, labs, and topics across Lattice.
DecoSearch introduces a training-free framework for translating natural language to SQL that effectively handles complex queries requiring multi-step reasoning. By employing a Schema Selector to prune irrelevant database elements and a LLM Judger to determine the appropriate decomposition strategy, the method routes queries based on their complexity, utilizing a Directed Acyclic Graph for intricate questions. Achieving execution accuracies of 70.53% on BIRD and 88.31% on Spider, DecoSearch outperforms existing training-free approaches while significantly reducing token consumption.
Routing SQL queries based on complexity allows DecoSearch to achieve unprecedented execution accuracy while using an order of magnitude fewer tokens than traditional methods.
Large Language Models (LLMs) have demonstrated remarkable capabilities in translating natural language to SQL, yet existing methods still falter on complex queries requiring multi-step, data-aware reasoning. We introduce DecoSearch, a training-free framework that addresses this by routing each query to the appropriate level of reasoning effort. A lightweight Schema Selector first prunes the full database schema to the relevant tables and columns. An LLM Judger then decides whether the question requires decomposition: straightforward questions follow a direct generation path and complex ones are escalated to a Directed Acyclic Graph (DAG) of atomic sub-questions, each solved by a targeted SQL generation step. A RAG component grounds the decomposer with semantically similar training examples, and a Topology Refiner restructures the reasoning plan when execution failures signal a flawed decomposition rather than a fixable SQL error. DecoSearch achieves 70.53% execution accuracy on BIRD and 88.31% on Spider with a DeepSeek backbone, surpassing all training-free baselines while consuming an order of magnitude fewer tokens than competing methods. It also functions as a model-agnostic wrapper, consistently improving fine-tuned SQL generation backbones without any modification to the pipeline.
