Title: Applying Transformer-Based Models for Lexical and Semantic Investigation of Legal Terms
Authors: Khujakulov Sunnatullo
Volume: 9
Issue: 12
Pages: 12-17
Publication Date: 2025/12/28
Abstract:
This research investigates the application of Transformer-based models, including BERT and Legal-BERT, for the lexical and semantic analysis of legal terminology. Legal texts are characterized by complex, context-dependent language, polysemy, and domain-specific expressions, which pose significant challenges for both human interpretation and computational processing. To address these challenges, a curated corpus of legal documents - comprising statutes, case law, and contracts - was compiled. Preprocessing techniques such as tokenization, lemmatization, and multi-word expression recognition were applied to optimize input for the models. Lexical analysis focused on term frequency, collocations, and contextual embeddings, while semantic analysis examined term similarity, clustering, and sense disambiguation. Experimental results demonstrate that Transformer-based models effectively capture subtle semantic distinctions, uncover latent lexical structures, and identify synonymous and contextually related legal terms. Compared to traditional computational approaches, these models provide superior performance in capturing both explicit and implicit relationships within legal language. The study highlights their potential applications in legal translation, automated terminology management, and AI-assisted legal research. This work contributes to the integration of advanced NLP techniques into legal linguistics and provides a foundation for future exploration in multilingual and low-resource legal domains.