GenAI-mediated immersive dialogue practice in liaison interpreting: a mixed-methods investigation of cognitive-affective outcomes and interactional processes.

Introduction: Liaison interpreting competence encompasses working memory management, executive control, self-efficacy, and affect regulation, yet its development remains constrained by the linguistic inconsistency and variable cognitive-affective demands of conventional peer role-play. Drawing on an integrated framework, this quasi-experimental mixed-methods study investigates whether Generative AI (GenAI)-mediated immersive dialogue practice can create a more cognitively authentic and affectively facilitative practice environment for interpreter-learners. Method: Seventy undergraduate students from two intact classes in a Chinese university were assigned to an experimental group ( Results and discussion: The experimental group scored significantly higher than the control group in overall interpreting performance, particularly in informational accuracy and interactional management. GenAI-supported practice was also associated with significantly lower interpreting-related anxiety with a large effect size. Learning engagement did not differ significantly, indicating that the two modalities may sustain engagement through different psychological mechanisms. Interaction analysis revealed a hub-and-spoke network centered on meaning transfer, with clarification requests and self-repair reflecting active metacognitive monitoring. This pattern represents the aggregate structural profile of interaction across the intervention period rather than a developmental trajectory. Qualitative findings identified psychological safety and repeatable mastery opportunities as key affordances, alongside role instability and limited corrective feedback as constraints, with students exercising psychological agency through prompt negotiation and self-directed scaffolding. These findings offer preliminary empirical support for GenAI as an affectively facilitative practice partner whose effects are theoretically consistent with predictions derived from Cognitive Load Theory, Processing Efficiency Theory, and Self-Efficacy Theory, although the underlying cognitive and motivational mechanisms remain to be directly tested.