AI와 인간의 다의성 해석 양상: ‘어떻게’ 일반의문문과 수사의문문을 중심으로

초록

This study investigates how the syntactic distribution of the Korean interrogative adverb eotteoke (‘how’) influences the semantic interpretation of interrogative sentences and experimentally compares interpretive patterns between human speakers and ChatGPT. A drama corpus was analyzed to examine usage patterns, and the same interpretation tasks administered to human participants were presented to ChatGPT for comparison. The results show that human interpretations vary significantly depending on the position of eotteoke. When it appears immediately before the predicate, [method], [manner], and [result] readings are preferred, whereas sentence-initial placement favors [reason] and rhetorical [negative assertion] interpretations. This indicates that syntactic distribution systematically contributes to meaning interpretation. In contrast, ChatGPT did not show significant variation in meaning selection according to distributional position. These findings suggest that the interpretation of eotteoke-questions arises from the interaction between syntax and pragmatics, and that there is a systematic difference between human and AI mechanisms of semantic interpretation.

키워드