Professor Hanjung Lee of the Department of English Language and Literature at Sungkyunkwan University (SKKU), led by President Yoo Ji-beom, will publish a new scholarly monograph, 『Polysemy in Semantics with Contextualized Language Models』, this December through De Gruyter Brill, one of the world's leading academic publishers. The book introduces an innovative approach to the study of linguistic meaning by integrating artificial intelligence with linguistic theory.
The monograph will appear in 『Language and Computers』, a prestigious international academic book series established in 1988 and known for publishing rigorously reviewed research at the intersection of language and computational technology. By bringing together traditional approaches to lexical semantics and advanced computational models of artificial intelligence, the book offers new insights into the fundamental limitations of contemporary language models, including those underlying systems such as ChatGPT, and suggests promising directions for future research.
Humans naturally distinguish between different meanings of a word depending on the context in which it is used. For example, the Korean word ‘nun’ can refer either to the human eye or to snow, and speakers readily identify the intended meaning from the surrounding context. For AI language models, however, accurately distinguishing subtle and complex differences in word meaning remains a significant challenge. Professor Lee addresses this challenge by integrating linguistic theories of meaning with neural language models that represent words and their contexts numerically. This approach provides a new account of how word meanings vary continuously and dynamically across different contexts.
The book combines large-scale analyses of English corpus data with psycholinguistic experiments examining human interpretations of word meanings and computational analyses probing the internal semantic representations of AI language models. Through systematic comparisons of human judgments and model predictions, the research reveals that contemporary language models do not always capture the full richness and diversity of human linguistic meaning. Instead, they exhibit systematic tendencies to favor more frequent or prototypical meanings, potentially narrowing and distorting the range of semantic variation found in actual language use. The study offers pioneering empirical evidence of these semantic biases and their implications for AI-based language understanding.
The significance of this research was highlighted in Professor Lee's invited keynote lecture at the 2026 Seoul International Conference on Linguistics (SICOL-2026), held in Seoul this August, where the findings were presented to an international audience of linguists. Building on this work, Professor Lee is scheduled to present further research at the Annual Meeting of the Linguistic Society of America (LSA), one of the world's most prestigious linguistics conferences, to be held in San Francisco this coming January.
Professor Lee commented, "Through this book, I sought to explore both the remarkable potential and the inherent limitations of AI language models, while advancing our understanding of how artificial intelligence interacts with human language and thought. I hope to continue pursuing interdisciplinary research that brings together linguistics, artificial intelligence, and cognitive science, contributing to new directions in English linguistics and the humanities."