What Distinguishes Musun and Etteon? A Corpus-Based Quantitative Analysis

This study investigates the distributional characteristics of the Korean demonstrative prenoun musun and etteon. Previous research on interrogatives has shown that substitution errors between these two words occur at a significant rate; however, little progress has been made in addressing this issue. The distinction seems to arise from the fact that each word appears in particular contexts and carries subtle semantic differences. To clarify these distributional tendencies, this study adopts a quantitative approach. Five experiments were conducted based on extracted examples of musun and etteon: (1) frequency comparison within spoken corpora, (2) collocation analysis of surrounding expressions, (3) word embedding analysis comparing cosine similarity and mean vector similarity between example sentences, (4) analysis of occurrence environments with surprisal-based predictability comparison, and (5) distributional comparison using generative language models. Results show that while the cosine similarity between the words themselves was low, the average vector similarity of their sentences was considerably high. Furthermore, although language models treated etteon as the unmarked form, quantitative evidence revealed that its occurrence was more restricted than that of musun. By systematically analyzing large-scale data with multiple quantitative methods, this study identifies the distributional properties and usage contexts of the two interrogatives, thereby addressing a long-standing question about their differentiation. These findings open up a new method in comparing and contrasting similar words, and are also significant for linguistic research.