한국 대학 EFL 학습자의 구두 발표 원고에 나타난 담화 연결어 분석 : AI 보조 수정을 위한 LLM 이전 기준선 연구

As generative AI is increasingly integrated into education, the need to develop learners' ability to critically evaluate and self-directly revise AI-based texts is growing. Accordingly, this present study aims to establish a learner-driven revision model based on digital critical thinking. To this end, oral presentation scripts written by Korean university EFL learners were analyzed to examine the frequency, functional distribution, and lexical concentration of discourse connectives. A learner corpus (KULC) was constructed based on presentation scripts (totaling 21,203 tokens) written by 84 nursing and health science students on major-related topics in 2021, before the widespread adoption of AI-based writing tools. Discourse connectives were classified into additive, adversative, causal, and sequential connectives according to the classification system of Larsen-Freeman and Celce-Murcia (2015) and extracted using AntConc. The analysis results showed that discourse connectives in the additive and sequential categories were most frequently used, particularly high-frequency expressions such as "and," "also," "first," and "second." Conversely, adversative and causal connectives were used less frequently, and lexical diversity was limited due to a reliance on specific expressions such as "but" and "therefore." Based on these results, this study proposed a rubric to promote balanced use of discourse connectives across functional categories. Ultimately, this rubric will help learners evaluate and refine their AI-assisted presentation scripts independently. This self-directed process will strengthen both discourse cohesion and rhetorical clarity, leading to a much more effective presentation delivery.