Visuals as a Context for NLP Systems When Dealing with Linguistic Polysemy
摘要
Polysemy, the capacity of a single linguistic unit to possess multiple related meanings, is a fundamental phenomenon in natural languages, deeply intertwined with the socio-historical and emotional experiences of language speakers. This study explores the characteristics of polysemous units, the challenges they present to linguistic analysis and computational processing, and recent advancements in addressing these challenges. It examines contemporary research trends from 2020 to 2024 that focus on distinguishing, categorizing, and disambiguating polysemous meanings, highlighting both theoretical insights and technological innovations, particularly in Natural Language Processing (NLP). The article emphasizes the importance of context recognition and word sense disambiguation, outlining the evolution from knowledge-based systems to contextual embeddings and multimodal models such as GPT-4V and CLIP. An illustrative case study of the word apple demonstrates how polysemy is influenced by contextual categories such as characters, space, time, and social values, revealing the dynamic and culturally embedded nature of the construction of meaning. Ultimately, the study argues for an open-ended and flexible categorization framework capable of accommodating new manifestations of polysemy, as language continues to evolve. By integrating linguistic theory, cognitive science and artificial intelligence, this research contributes to a deeper understanding of polysemy and the development of more context-sensitive language technologies.