Student-AI interaction in computer-assisted consecutive interpreting: patterns and performance

While student-AI interaction patterns have been explored in self-paced cross-linguistic tasks such as L2 writing and translation post-editing, little is known about how students interact with AI in bilingual tasks under intense cognitive and temporal constraints. This study uses computer-assisted consecutive interpreting (CACI) as a window to examine student-AI interaction patterns in such high-stakes environments. Twenty-two Chinese-native interpreting trainees, grouped by prior AI training experience, performed bidirectional CACI tasks using AI-enabled systems integrating automatic speech recognition (ASR) and machine translation (MT) features. With data collected from eye-tracking, pen-recording, and voice-recording, the study reveals that: (a) four interpretable interaction profiles emerge: Intensive Engagers (heavy AI reliance), Fast Scanners (scanning-based processing), Traditionalists (minimal AI reliance), and Frequent Switchers (shifts between AI support and handwritten notes); (b) these patterns shift dynamically between the comprehension and production stages of interpreting, while targeted CACI training was associated with more stable, AI-oriented patterns; and (c) students' interaction patterns during the comprehension stage, but not the production stage, are significantly associated with performance quality. These findings highlight the heterogeneity and variability of student-AI interaction patterns across different stages of complex bilingual tasks and underscore training effects on shaping students' behaviors and performance.