생성형 AI의 스페인어 접어 대명사 se 처리 방식 연구: 번역과 맥락적 적합도 평가를 중심으로 = How Generative AI Processes Spanish Clitic Pronoun se: A Study on Translation Output and Contextual Fit Judgment

다국어 초록 (Multilingual Abstract)

This study examines how generative AI handles the Spanish clitic pronoun se, which functions as an aspectual operator that adjusts specific facets of events beyond simple reflexivity. Since Korean and English lack direct morphosyntactic correspondents to se, its translation requires inferential processing. The research employs a three-stage task design: first, translation of 18 Spanish sentence pairs (with/without se) into Korean and English; second, scenario generation based on the sentences; and third, fit judgment between two sentence options given the generated scenarios. Results show that in translation, se’s semantic contribution tends to be expressed indirectly through lexical choices and may be neutralized when competing cues are present. However, when explicit contextual cues are provided, LLM models actively connect dynamic event indicators with se’s functions. This indicates that LLMs prioritize semantic cues and contextual strength over morphological forms themselves. The findings illustrate how LLMs process linguistic categories that lack cross-linguistic equivalence, suggesting both capabilities and limitations in handling fine-grained semantic distinctions in machine translation.