Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neuro-Symbolic Query Compiler

Yuyao Zhang, Zhicheng Dou, Xiaoxi Li, Jiajie Jin, Yongkang Wu, Zhonghua Li, Qi Ye, Ji-Rong Wen

TL;DR

The paper tackles the difficulty of precise search intent recognition for complex, nested queries in Retrieval-Augmented Generation by introducing QCompiler, a neuro-symbolic framework built around a minimal yet sufficient grammar $G[q]$. It translates natural language into BNF-based expressions, parses them into Abstract Syntax Trees via a Lexical-Syntax Parser, and executes them with a Recursive Descent Processor, yielding atomic leaf sub-queries for targeted retrieval. Empirically, QCompiler achieves state-of-the-art results on four multi-hop benchmarks, improving efficiency by reducing redundant retrievals while maintaining strong accuracy, and demonstrates robust behavior across model scales and query types. The approach is designed as a plug-in to existing RAG pipelines, offering interpretable, verifiable query decomposition and reasoning with broad practical impact for real-world information systems.

Abstract

Precise recognition of search intent in Retrieval-Augmented Generation (RAG) systems remains a challenging goal, especially under resource constraints and for complex queries with nested structures and dependencies. This paper presents QCompiler, a neuro-symbolic framework inspired by linguistic grammar rules and compiler design, to bridge this gap. It theoretically designs a minimal yet sufficient Backus-Naur Form (BNF) grammar $G[q]$ to formalize complex queries. Unlike previous methods, this grammar maintains completeness while minimizing redundancy. Based on this, QCompiler includes a Query Expression Translator, a Lexical Syntax Parser, and a Recursive Descent Processor to compile queries into Abstract Syntax Trees (ASTs) for execution. The atomicity of the sub-queries in the leaf nodes ensures more precise document retrieval and response generation, significantly improving the RAG system's ability to address complex queries.

Neuro-Symbolic Query Compiler

TL;DR

The paper tackles the difficulty of precise search intent recognition for complex, nested queries in Retrieval-Augmented Generation by introducing QCompiler, a neuro-symbolic framework built around a minimal yet sufficient grammar . It translates natural language into BNF-based expressions, parses them into Abstract Syntax Trees via a Lexical-Syntax Parser, and executes them with a Recursive Descent Processor, yielding atomic leaf sub-queries for targeted retrieval. Empirically, QCompiler achieves state-of-the-art results on four multi-hop benchmarks, improving efficiency by reducing redundant retrievals while maintaining strong accuracy, and demonstrates robust behavior across model scales and query types. The approach is designed as a plug-in to existing RAG pipelines, offering interpretable, verifiable query decomposition and reasoning with broad practical impact for real-world information systems.

Abstract

Precise recognition of search intent in Retrieval-Augmented Generation (RAG) systems remains a challenging goal, especially under resource constraints and for complex queries with nested structures and dependencies. This paper presents QCompiler, a neuro-symbolic framework inspired by linguistic grammar rules and compiler design, to bridge this gap. It theoretically designs a minimal yet sufficient Backus-Naur Form (BNF) grammar to formalize complex queries. Unlike previous methods, this grammar maintains completeness while minimizing redundancy. Based on this, QCompiler includes a Query Expression Translator, a Lexical Syntax Parser, and a Recursive Descent Processor to compile queries into Abstract Syntax Trees (ASTs) for execution. The atomicity of the sub-queries in the leaf nodes ensures more precise document retrieval and response generation, significantly improving the RAG system's ability to address complex queries.
Paper Structure (39 sections, 5 equations, 4 figures, 9 tables, 1 algorithm)

This paper contains 39 sections, 5 equations, 4 figures, 9 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Query Understanding
  4. Neuro-Symbolic AI & Grammar Rule
  5. Retrieval-Augmented Generation System
  6. Methodology
  7. Mathematical Definition of Query Types
  8. Backus-Naur Form (BNF) Grammar
  9. Key Components of QCompiler
  10. Training of Query Expression Translator
  11. Validation and Inference
  12. Why Can QCompiler Improve Existing RAG Systems?
  13. Experiment
  14. Datasets and Metrics
  15. Baselines
  16. ...and 24 more sections

Figures (4)

  • Figure 1: Illustration of how QCompiler enhances the RAG system.
  • Figure 2: Illustration of QCompiler, including the grammars, components, and an example of processing complex queries with QCompiler.
  • Figure 3: Illustration of the atomicity of sub-queries in leaf nodes, comparing with traditional RAG systems and Iterative RAG systems, QCompiler has few documents retrieval and more accurate response.
  • Figure 4: The number of correctly answered queries for each type compiled by QCompiler across different methods.