Human-in-the-Loop AI Feedback in Interpreter Training: An ASR-Based Platform Analysis of Instructor Annotations, Comment Functions, and System Constraints

The use of AI-assisted feedback systems has steadily increased in higher education, yet there has been limited research as to which components of expert feedback can be supported by AI and which require continued human judgment. This issue is particularly important in professional training domains where feedback must rely upon contextual interpretation, source–output comparison, and domain expertise. Accordingly, this study analyzes instructor-generated feedback in TalkTrack, an automatic speech recognition (ASR)-supported digital platform for interpreter training, as a basis for human-in-the-loop AI feedback design. The dataset consists of 44 Korean–Japanese, Japanese–Korean, and Chinese–Korean interpreting performances produced by 12 graduate students and includes 931 feedback tags and 756 substantive qualitative comments manually assigned by 10 instructors. The categories most frequently assigned were expression and translation-error feedback, while corrective feedback was the dominant instructional function. Further, feedback behavior varied substantially by instructor and feedback category, indicating that different types of feedback require different levels of human intervention and system support. Accordingly, this study proposes a human-in-the-loop feedback architecture employing ASR and large language models to support detection, organization, retrieval, and suggestion generation, while instructors retain responsibility for source-text fidelity, discourse coherence, pedagogical judgment, and final feedback validation, from which this study proposes empirical design requirements for scalable AI-supported feedback systems in professional e-learning environments.