생성형 AI 작성 학술논문 초록의 담화 구조: KCI 등재 번역학 논문 초록을 중심으로

Move Analysis on GAI vs. Human Abstracts: Focusing on KCI-Indexed Translation Studies Abstracts

This study investigates the discourse structures of human- and GPT-4-generated abstracts in Korean translation studies journals using Hyland’s (2004) five-move model. A comparative move analysis was conducted on 60 abstracts from three KCI-indexed journals and GPT-4. The findings reveal three key results. First, human-written abstracts exhibited high variation in move patterns, with only one abstract containing all five moves. Second, all GPT-4-generated abstracts contained four or five moves, consistently including Purpose, Method, and Product, with two highly-used patterns accounting for 63% of all cases. Third, GPT-4 abstracts had significantly higher lexical diversity, average word length, and sentence length. This study contributes to identifying previously unexplored discourse features of translation studies abstract. Another contribution of this study is to highlight the potential of Generative Artificial Intelligence (GAI) for generating structurally sound academic summaries, providing a foundation for future interdisciplinary research and guidelines on the use of GAI in academic research.