Unlocking Legal Knowledge with Multi-Layered Embedding-Based Retrieval

TL;DR

The paper tackles the difficult problem of retrieving nuanced legal knowledge from hierarchical texts by introducing a multi-layered embedding-based retrieval method within a Retrieval Augmented Generation (RAG) framework. It defines embeddings, aboutness, and semantic chunking, and argues that representing legal texts at multiple structural levels (document, article, paragraph, sections, and groupings) enables context-aware responses. Through a Brazil-focused case study on the Brazilian Constitution, it demonstrates that multi-layered chunking yields denser, more relevant representations, improves chunk selection via cosine similarity filtering, and outperforms flat chunking in identifying essential passages. The approach promises more accurate, efficient legal information retrieval and can extend to other domains with hierarchical textual data, offering potential for improved accessibility and decision support in legislative and regulatory tasks.

Abstract

This work addresses the challenge of capturing the complexities of legal knowledge by proposing a multi-layered embedding-based retrieval method for legal and legislative texts. Creating embeddings not only for individual articles but also for their components (paragraphs, clauses) and structural groupings (books, titles, chapters, etc), we seek to capture the subtleties of legal information through the use of dense vectors of embeddings, representing it at varying levels of granularity. Our method meets various information needs by allowing the Retrieval Augmented Generation system to provide accurate responses, whether for specific segments or entire sections, tailored to the user's query. We explore the concepts of aboutness, semantic chunking, and inherent hierarchy within legal texts, arguing that this method enhances the legal information retrieval. Despite the focus being on Brazil's legislative methods and the Brazilian Constitution, which follow a civil law tradition, our findings should in principle be applicable across different legal systems, including those adhering to common law traditions. Furthermore, the principles of the proposed method extend beyond the legal domain, offering valuable insights for organizing and retrieving information in any field characterized by information encoded in hierarchical text.