AI feedback literacy in translator education: profiling cognitive-affective-ethical engagement, developmental trajectories, and learner patterns

The widespread adoption of Large Language Models (LLMs) in translator education has made AI-generated feedback an increasingly common instructional resource. However, how translation students cognitively interpret, affectively regulate, and ethically evaluate such feedback remains underexplored. This study conceptualizes AI Feedback Literacy (AIFL) as an emerging sub-competence in translation pedagogy and proposes a Cognitive-Affective-Ethical (CAE) framework to examine its development. A mixed-methods study was conducted through a 12-week, three-phase intervention with 60 translation students from two Chinese universities. The intervention integrated LLM-feedback workshops, genre-diverse translation tasks, feedback logs, and reflective writing activities. Quantitative analyses showed significant improvements in all AIFL dimensions (Z = −3.62 to −4.11, p < 0.001; r = 0.47–0.53), with overall scores increasing from 3.21 to 4.07. Confirmatory factor analysis supported the three-factor structure, while structural equation modeling revealed that cognitive understanding predicted affective trust calibration (β = 0.61) and ethical reflection (β = 0.54), explaining 63% of AIFL variance. Qualitative findings revealed four learner profiles and a recurring trust–skepticism–recalibration cycle, indicating a shift from passive AI reliance to informed and responsible engagement. The study highlights AIFL as a teachable construct and offers a human-centered framework for AI-mediated translator education.