The emergence of generative AI in translation classrooms has shifted instructional priorities away from text production alone toward issues of quality and judgement. This study reports a within-class longitudinal investigation of post-editing of AI-assisted translation in Japanese–Chinese translation training. Over a single semester, twenty-eight undergraduate Japanese majors completed six translation assignments, each involving an AI-generated draft, manual revision, and a short-written justification of their revision choices. Two learning outcomes were examined: overall translation quality (scoreₜotal, 0–100) and the quality of students’ rationales (rationalequality, 0–2), the latter capturing the extent to which revision decisions were supported by explicit, text-based evidence. Results indicated gradual and non-linear improvement in translation quality, alongside more noticeable changes in students’ justification practices over time. Genre also played a mediating role: news tasks tended to elicit cue-based rationales related to modality and attribution, whereas literary tasks prompted broader but less readily verifiable stylistic reasoning. These findings suggest that the pedagogical value of post-editing of AI-assisted translation lies not only in improving translation products, but also in fostering evaluative judgement through routine, scaffolded post-editing and justification activities.
Gu et al. (Sun,) studied this question.