Effectiveness in Retrieving Legal Precedents: Exploring Text Summarization and Cutting-Edge Language Models Toward a Cost-Efficient Approach [poster]

Automating the retrieval of legal precedents is crucial for streamlining legal research and upholding the principle of stare decisis. With the exponential growth of legal data, traditional methods fail to cope with the demands for efficiency and accuracy. Recent advancements in Artificial Intelligence (AI), particularly in Natural Language Processing (NLP) and Machine Learning (ML), offer promising solutions. However, challenges remain, such as the need for high computational power, costs, scalability, and the lack of universally applicable methodologies across various jurisdictions.

This study explores the potential of combining legal text summarization techniques with cutting-edge language models, such as OpenAI's ADA, to develop an efficient and scalable system for legal precedent retrieval. The focus is on balancing performance and resource consumption, addressing the ongoing need for cost-effective yet reliable AI-driven solutions in the legal domain.

We assessed different methods for summarizing legal text by extracting parts such as person, organization, place, time, statutes, and jurisprudence. The study also compared summaries based on concepts (nouns) and relations (verbs). Additionally, this research compared the performance of text embeddings created from models trained with general-purpose text and legal documents.