ResearchResearch paperRetrieval, RAG & Search1 source · Oct 19, 2026

StrucChunk: Structure-Aware Retrieval for Legal Document RAG

We propose StrucChunk (Structure Aware Chunking) with hybrid dense-sparse retrieval for hierarchically structured legal documents.

Retrieval-Augmented Generation (RAG) can ground LLM outputs in domain-specific knowledge, but legal documents are hard: standard chunking fragments hierarchical argument structure, and generic embeddings struggle with both precise citation matching and semantic understanding.
On LegalBench-RAG (6,858 query-passage pairs), it achieves Char-Recall@8 of 48.18% vs. 23.59% for the best baseline (+104%), surpassing the published state of the art by +83%.
On NitiBench-CCL (3,730 Thai financial law queries), HR@5 is 0.856 vs. 0.594 for naive chunking (+44.1%), showing that section-boundary alignment is the dominant factor across languages, and cross-reference augmentation adds +2.3% on cross-reference queries.
We release the implementation, 5,256 evaluation queries across five legal corpora, annotation methodology and chunking quality metrics.

[1]StrucChunk: Structure-Aware Retrieval for Legal Document RAG
arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 12:56 PM
We propose StrucChunk (Structure Aware Chunking) with hybrid dense-sparse retrieval for hierarchically structured legal documents.
Retrieval-Augmented Generation (RAG) can ground LLM outputs in domain-specific knowledge, but legal documents are hard: standard chunking fragments hierarchical argument structure, and generic embeddings struggle with both precise citation matching and semantic understanding.

Extractive summary: sentences quoted from the sources.

Before this