고등학교 영어 텍스트의 담화 연결어에 대한 코퍼스 기반 비교: 인간 저작물과 생성형 AI 텍스트를 중심으로
A Corpus-based Comparison of Discourse Connectors in Human and AI-authored High School Texts
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study examines the use of discourse markers in Korean high school English textbook texts and their AI-paraphrased counterparts generated by ChatGPT. Employing a corpus-based approach, the analysis focuses on five categories of discourse markers: additive, adversative, causal, temporal, and exemplification. The corpus comprises 47 textbook passages and 47 corresponding GPT-generated texts. The results reveal no statistically significant differences in the use of additive, causal, and temporal markers, suggesting that AI-generated texts can effectively replicate the logical flow in these areas. In contrast, notable discrepancies emerged in the adversative and exemplification categories, where textbook texts employed more explicit markers such as however and for example. These findings suggest that while GPT-generated texts exhibit overall fluency, they may lack pedagogically intentional discourse structures that are essential for fostering learners’ logical reasoning and conceptual clarity. The study underscores the need for instructional scaffolding when integrating AI-generated content into educational settings and offers insights for the development of more pedagogically aligned AI-assisted learning materials.
Abstract
This study examines the use of discourse markers in Korean high school English textbook texts and their AI-paraphrased counterparts generated by ChatGPT. Employing a corpus-based approach, the analysis focuses on five categories of discourse markers: additive, adversative, causal, temporal, and exemplification. The corpus comprises 47 textbook passages and 47 corresponding GPT-generated texts. The results reveal no statistically significant differences in the use of additive, causal, and temporal markers, suggesting that AI-generated texts can effectively replicate the logical flow in these areas. In contrast, notable discrepancies emerged in the adversative and exemplification categories, where textbook texts employed more explicit markers such as however and for example. These findings suggest that while GPT-generated texts exhibit overall fluency, they may lack pedagogically intentional discourse structures that are essential for fostering learners’ logical reasoning and conceptual clarity. The study underscores the need for instructional scaffolding when integrating AI-generated content into educational settings and offers insights for the development of more pedagogically aligned AI-assisted learning materials.