Speaker
Description
What makes ChatGPT feedback surprising, reassuring, or worrying to an EFL writer? Research on emotional engagement with AI-assisted writing has documented a range of responses, but an emotion label alone may obscure which aspect of a feedback encounter the learner is evaluating. Guided by control-value theory, this study uses a theory-informed qualitative secondary analysis of six Vietnamese-language retrospective interviews with undergraduates who revised an opinion essay using ChatGPT feedback. We identified 63 potentially relevant affective-talk segments, including denials and evaluations without explicit emotion labels, and closely examined 32 segments containing endorsed or qualified emotion expressions. Within-case and cross-case analysis traced the objects of these expressions and whether they were volunteered or elicited by interviewer prompts. The accounts revealed that emotional expressions could concern distinct aspects of the same feedback encounter. One learner expected errors and denied being surprised by them, yet reported surprise at the usefulness of ChatGPT’s feedback. Another described worry about completing the essay and, in a more general account, about AI-generated language exceeding the writer’s proficiency and being recognisable to teachers. Other responses qualified or rejected emotions suggested by interviewers. These findings do not establish emotional effects of ChatGPT; rather, they show why retrospective accounts require attention to both the object of an emotion and how it was elicited. The study offers a practical implication for AI-feedback research and teaching: ask what learners felt about specific aspects of feedback before interpreting their responses as overall anxiety, reassurance, or satisfaction.