Beyond intention: A cognitive-affective-ethical framework for translation students' in-task acceptance of...

Although large language models (LLMs) are increasingly integrated into translator education, empirical evidence on how student translators engage with AI-generated feedback in situated, in-task practice remains limited. This mixed-methods study investigates translation students' behavioral acceptance of AI-generated feedback delivered through an LLM-embedded platform, examining the roles of baseline attitude toward AI, translation proficiency, and text genre. A total of 124 translation majors completed 3596 sentence-level accept/reject decisions, each accompanied by mandatory written rationales. Mixed-effects logistic regression revealed that all three factors independently and additively predicted acceptance: text genre emerged as the strongest within-person driver, with acceptance ranging from 81.2% for technical texts to 49.9% for literary excerpts; skeptical baseline attitudes significantly suppressed acceptance; and higher proficiency was associated with greater evaluative selectivity. Thematic analysis of written rationales identified nine reasoning mechanisms, spanning cognitive verification, affective trust calibration, and ethical-identity gatekeeping, which were integrated into a Cognitive-Affective-Ethical (CAE) framework extending the Technology Acceptance Model. CAE conceptualizes acceptance as a hierarchical, conditional process: cognitive evaluation filters feasibility, affective trust modulates reliance under uncertainty, and ethical considerations function as context-activated constraints that can override both cognitive and affective drivers. By shifting the analytical focus from pre-adoption intention to in-task behavioral judgment, this study advances a mechanism-based account of when, why, and under what ethical conditions AI-generated feedback is accepted in translator education, with implications for genre-aware pedagogy and the design of adaptive AI feedback systems.