안상희
국어교육학연구 60(5): 5-39
This study explores how teacher educators (TEs) and ChatGPT construct and apply evaluation criteria in assessing pre-service teachers’ (PTs) writing, focusing on university newspaper editorials. Nineteen PTs majoring in Korean language education participated in the study. Three TEs collaboratively developed evaluation criteria and a rubric to rate the texts, while ChatGPT (Plus version) was instructed to create its own rubric and later assess the same texts using the TEs’ rubric. Both TEs and ChatGPT emphasized logical reasoning and persuasiveness as core elements of the editorial genre. However, TEs developed detailed, content-oriented criteria highlighting argumentative depth and reader awareness, whereas ChatGPT presented integrated, form-oriented criteria emphasizing clarity, structure, and media adaptability. When applying the same rubric, TEs exhibited a wider scoring range with greater qualitative sensitivity, while ChatGPT showed a convergent mid-range scoring pattern. These results indicate that ChatGPT can apply teacher-developed rubrics consistently but lacks contextual and interpretive nuance. The scoring results of ChatGPT also showed low discriminatory power of the scores. It may serve as a supplementary tool that helps teachers refine criteria & rubric design or cross-check their practice of evaluation, suggesting potential for AI–human collaboration in writing evaluation.