The Era of Specialized Language Models in Justice 4.0

Abstract

The Brazilian judicial system faces a monumental challenge due to the volume of procedures, requiring technological solutions to improve speed and efficiency. This article explores the use of Natural Language Processing (NLP) applied to Law, discussing how the evolution of neural network architectures, from static embeddings to Large Language Models (LLMs), has been transforming legal analysis. Through practical examples, it demonstrates that models specifically trained or adapted with Brazilian legal data outperform generalist models in complex tasks such as legal document classification, summarization, and entity extraction. These models serve as fundamental components for the consolidation of the Justice 4.0 program.