Journal DOI 10.68956/journal.legal-engineering-technology

Creative Commons source reading · Evaluation, bias and explainability

Enhancing Legal Expertise in Large Language Models through Composite Model Integration: The Development and Evaluation of Law-Neo

Zhihao Liu; Yanzhen Zhu; Mengyuan Lu

Originally published in Proceedings of the Natural Legal Language Processing Workshop 2024 · 2024-11

From the article

Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.

Abstract

Although large language models (LLMs) like ChatGPT have demonstrated considerable capabilities in general domains, they often lack proficiency in specialized fields. Enhancing a model’s performance in a specific domain, such as law, while maintaining low costs, has been a significant challenge. Existing methods, such as fine-tuning or building mixture of experts (MoE) models, often struggle to balance model parameters, training costs, and domain-specific performance. Inspired by composition to augment language models, we have developed Law-Neo, a novel model designed to enhance legal LLMs. This model significantly improves the model’s legal domain expertise at minimal training costs, while retaining the logical capabilities of a large-scale anchor model. Our Law-Neo model outperformed other models in comprehensive experiments on multiple legal task benchmarks, demonstrating the effectiveness of this approach.

Editorial reading guide

  • Which legal corpus, jurisdiction and evaluation task does the article use?
  • What evidence supports the reported performance, and where could the method fail in practice?