Interactive Discourse Analysis and Empirical Study on the Teaching Effectiveness of English Speaking Classes in the Context of Human-Machine Dual-Teacher Collaboration
TL;DR
It is found that classrooms with H-M dual teachers outperform those with human teachers alone, and six types of teacher-AI collaboration have been identified within H-M classrooms.
Abstract
The confluence of LLMs, ASR, and the expertise of human teachers has given rise to a new teaching configuration: a human-machine (H-M) dual teacher. The discourse properties of such a teaching configuration have not yet been investigated. In this paper, we review the relevant research on H-M dual teachers in EFL classrooms in Chinese higher education institutions to provide an overview of the discourse properties of such classrooms. Based on metrics of discourse acts, discourse act length, self-repair rate, and lexical complexity, as well as findings from research on ASR, natural language processing, and intelligent tutoring systems, we find that classrooms with H-M dual teachers outperform those with human teachers alone. More specifically, classrooms with H-M dual teachers have up to 85.7% more discourse acts than classrooms with human teachers alone, have 12.6% fewer instances of self-repair than classrooms with human teachers alone, and exhibit effect sizes of d > 2.0 for speaking-related measures. Furthermore, six types of teacher-AI collaboration have been identified within H-M classrooms. Finally, limitations to the current state of H-M classroom implementation are presented and discussed.