Comparative Analysis of various RAG-based methods for Multilingual AI Legal Document Processing

Stephany Guzman

 Co-Presenters: Individual Presentation
College: Dorothy and George Hennings College of Science, Mathematics and Technology
Major: Computer Science
Faculty Research Mentor: Yulia Kumar, Jenny Li
 

Abstract:

The research investigates the quality of text summarization, content understanding, and Q/A abilities of various RAG-based methods and apps over AI-related multilingual legal documents. The main such method include RAG-based apps vs GraphRAG vs Lazy GraphRAG approaches. Analysis includes cross-lingual comparison of English, Spanish, German, and French documents. By incorporating tools like Swarm, spaCy, and fast Text, the project processes and translates AI-related legal texts, such as EU AI Act, while extracting entities, keywords, and legal clauses. A visualization system is also developed to graph relationships among extracted entities. By identifying relationships and patterns across multilingual legal texts, this project aims to uncover meaningful insights that transcend linguistic and cultural boundaries. The goal is to enable efficient legal document analysis, improve multilingual accessibility, and identify ethical and legal concerns in AI laws. This work ultimately contributes to the development of fair, transparent, and globally aware AI governance systems for future applications.