NeSy-RAG: Neuro-Symbolic RAG for Explainable Question Answering

2026-08-06

AI Summary

This work addresses the opacity of traditional retrieval-augmented generation (RAG) methods, which hinder verification of intermediate reasoning steps, evidence tracing, and detection of missing user context, often leading to erroneous or incomplete responses. To overcome these limitations, the authors propose a modular neuro-symbolic RAG framework that leverages joint natural language–code embeddings to transform retrieved passages into attributable Prolog logic modules. These modules enable deterministic, interpretable, and traceable inference for answer generation. Additionally, the framework incorporates a symbolic knowledge gap detection mechanism that automatically identifies critical missing facts affecting reasoning and triggers user interaction to resolve them. Evaluated on the ShARC benchmark, the method achieves 61.1% accuracy without domain-specific training, substantially outperforming standard RAG baselines (42.8%) and enabling end-to-end explainable and verifiable question answering.

Abstract

Retrieval-augmented generation (RAG) improves question answering by grounding large language models (LLMs) in external knowledge such as text corpora. However, its reasoning process remains largely opaque: intermediate reasoning steps are difficult to verify and cannot be reliably attributed to specific evidence. Moreover, missing user-specific context is rarely detected systematically, often leading to incomplete or incorrect output. We propose NeSy-RAG, a modular neuro-symbolic RAG framework that synthesizes attributable Prolog modules from retrieved text chunks. For each chunk, the system generates semantically meaningful predicates that encode Boolean claims, which may depend on user facts. Using joint natural language-code embeddings, predicates are retrieved and composed into Prolog queries. To address incomplete user context, we introduce a symbolic knowledge-gap detection mechanism that identifies missing user facts whose truth values affect the query outcome and automatically triggers follow-up interactions. Executing the resulting Prolog queries yields deterministic answers together with transparent execution traces that link each reasoning step to its originating source. On the ShARC benchmark, without domain-specific training, NeSy-RAG achieves 61.1% accuracy, outperforming a same-model RAG baseline that achieves 42.8% accuracy.

Problem

• Retrieval-augmented generation
• Explainable reasoning
• Knowledge gap detection
• Attributable reasoning
• Question answering

Innovation

• Neuro-Symbolic
• Retrieval-Augmented Generation
• Explainable QA
• Prolog-based Reasoning
• Knowledge Gap Detection