소유격 관계대명사 whose-구조의 비선호와 기계 번역 오류 = Dispreference for Whose-Relative Clauses and Their Implications for Machine Translation

다국어 초록 (Multilingual Abstract)

This study investigates the cognitive processing cost associated with the English possessive relative pronoun whose, often avoided in favor of simpler alternatives such as with or that has despite grammatical acceptability. Using ChatGPT 4.1, this study simulates sentence processing behaviors such as self-paced reading, eye-tracking regressions, comprehension accuracy, and naturalness ratings for sentences containing whose, with, and that has constructions. The results indicate that whose clauses induce the highest cognitive load, whereas with constructions facilitate the most efficient processing due to their linear structure and high predictability. These findings are consistent with theories proposed by Gibson (1998) and Frazier and Rayner (1982) that structural complexity and backtracking elevate working memory load. The study also suggests that whose clauses require non-linear integration and backward inferencing, explaining their dispreference in natural language. Pedagogically, teaching alternative constructions alongside whose could enhance comprehension, particularly for L2 learners.