MODELING SEMANTIC CHANGE THROUGH LARGE LANGUAGE MODELS

This PhD thesis focuses on computational modeling of semantic change through large language models. It investigates the modeling of lexical semantic change, where words change in meaning over time, and extends this study to model text-level semantic change, specifically historical resonance. The thesis addresses three research questions: (1) how to model lexical semantic change using large language models, (2) how to extend the modeling from two time periods to multiple time periods, and (3) how to model historical resonance. The thesis reviews state-of-the-art approaches, proposes novel frameworks, and evaluates different methods across multiple languages. Findings suggest that word embeddings of BERT models can be used to accurately detect semantic change. The thesis also underscores the interpretability and effectiveness of lexical replacements and sense definitions generated by Llama and Flan-T5 models in modeling word meaning. By advancing methodologies and tools in Natural Language Processing and Computational Linguistics, this thesis contributes significantly to modeling the dynamic nature of text semantics over time.