인공지능 기계 번역과 상호참조해결 말뭉치를 활용한 한국어와 중국어 한정성 비교 연구

Abstract

This study integrates theoretical linguistics and natural language processing (NLP) to analyze the semantic and pragmatic aspects of definiteness in Korean and Chinese. Theoretically, it elucidates the unique definiteness systems of each language, identifying structural differences through a comparative analysis that informs how translation models process definiteness. Computationally, a Korean coreference resolution corpus focusing on definiteness was constructed, and Korean-Chinese translations were evaluated using advanced language models (GPT-4 and Papago+). Both models performed well in human and automated evaluations, with notable distinctions. GPT-4 occasionally fell short in reflecting Koreanspecific features, particularly in translating named entities, pronouns, and demonstratives, while Papago+ excelled in pronoun translation and syntactic fluency in Chinese. Human evaluators favored Papago+ for its nuanced pronoun choices, whereas GPT-4 achieved slightly higher scores in machine-based metrics emphasizing clarity. This research underscores the importance of human evaluation in assessing translations of languages like Korean and highlights the need to refine model design and evaluation metrics, contributing to cross-linguistic definiteness studies.