Disco-RAG: Discourse-Aware Retrieval-Augmented Generation

2026-01-07

arXiv.org

AI Summary

This work addresses a key limitation of existing retrieval-augmented generation (RAG) approaches, which process retrieved passages in a flat manner and thus struggle to model discourse structure, hindering effective cross-document evidence integration. To overcome this, the paper proposes the first RAG framework that explicitly incorporates discourse structure by constructing intra-paragraph discourse trees and inter-paragraph rhetorical graphs, thereby forming a structured generation blueprint. This blueprint guides large language models to synthesize multi-source information without requiring fine-tuning. By jointly modeling local hierarchical structure and cross-paragraph coherence, the method significantly enhances knowledge synthesis capabilities, achieving state-of-the-art performance on both question answering and long-document summarization benchmarks. These results underscore the critical role of discourse structure in advancing RAG systems.

Technology Category

• Natural Language Processing: Discourse, Pragmatics & Argument Mining
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Knowledge Representation and Reasoning: Argumentation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Retrieval-Augmented Generation (RAG) has emerged as an important means of enhancing the performance of large language models (LLMs) in knowledge-intensive tasks. However, most existing RAG strategies treat retrieved passages in a flat and unstructured way, which prevents the model from capturing structural cues and constrains its ability to synthesize knowledge from dispersed evidence across documents. To overcome these limitations, we propose Disco-RAG, a discourse-aware framework that explicitly injects discourse signals into the generation process. Our method constructs intra-chunk discourse trees to capture local hierarchies and builds inter-chunk rhetorical graphs to model cross-passage coherence. These structures are jointly integrated into a planning blueprint that conditions the generation. Experiments on question answering and long-document summarization benchmarks show the efficacy of our approach. Disco-RAG achieves state-of-the-art results on the benchmarks without fine-tuning. These findings underscore the important role of discourse structure in advancing RAG systems.

Problem

• Retrieval-Augmented Generation
• discourse structure
• knowledge synthesis
• document coherence
• structured retrieval

Innovation

• Discourse-aware
• Retrieval-Augmented Generation
• Discourse Tree
• Rhetorical Graph
• Knowledge Synthesis