Disco-RAG: Discourse-Aware Retrieval-Augmented Generation

TL;DR

Disco-RAG introduces a discourse-aware framework for retrieval-augmented generation that leverages discourse structures to improve performance on knowledge-intensive tasks.

Contribution

It proposes constructing discourse trees and rhetorical graphs to explicitly incorporate discourse cues into RAG models, enhancing their ability to synthesize dispersed evidence.

Findings

Achieves state-of-the-art results on question answering benchmarks.

Improves long-document summarization performance without fine-tuning.

Demonstrates the importance of discourse structure in RAG systems.

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.