PresentationDance of Agency? An investigation on learners' agency in a neural machine translation training course more

The emergence of AI, while establishing MT's role in translation classrooms, presents an existential challenge for translators and trainers concerning trainee engagement and subjectivity. In response, this paper investigates trainees' agency and their 'dance of agency' (i.e., resistance and accommodation) during an undergraduate AI-assisted translation course.In contrast to typical machine translation post-editing studies that prioritize productivity, this project adopts an empowerment-focused design. It observes and documents students' decision-making processes to reveal their agency when collaborating with AI. Participants undertake MT-assisted tasks involving three levels of cognitive complexity (MT error annotation, MT with post-editing, and controlled authoring with MT and post-editing) across three text types (informative, expressive, and operative). Preliminary findings indicate that when trainees are fully empowered to annotate MT errors and pre-edit source texts to enhance MT output, they demonstrate greater awareness of potential MT errors and limitations. Consequently, they exhibit increased willingness and confidence in editing MT suggestions. Drawing upon Rammert's (2008) model of agency levels, the results suggest that this approach shifts students' engagement with machine translation, gradually moving them from 'passive' or 'semi-active' roles in MT training towards 'pro-active' and 'co-operative' ones.