An AI Multi-Agent System for Translation Revision Teaching: Developing and Validating the TAM-AS Model

Abstract

Against the backdrop of AIGC, translation education faces a “high-autonomy-low-control” dilemma where translation revision requires learner autonomy but static CAT tools lack structured guidance. This study develops a multi-agent AI system for translation revision teaching and extends the classic TAM into TAM-AS by integrating “agentic support.” It adopted a three-phase design, constructing TAM-AS and three AI agents (dynamic task planning, context-aware error correction, cognitive attunement), implementing an 8-week intervention with 120 translation majors via stratified random assignment to experimental/control groups, and validating effectiveness through statistical and thematic analysis. Key findings include the AI system significantly enhancing self-regulated revision with 58% more independent decisions and 47% higher self-correction accuracy (p<0.001), perceived translation quality, clear feedback and interface simplicity driving TAM-AS acceptance while tool speed being irrelevant, and TAM-AS improving technology acceptance predictability by 23.6% compared to classic TAM. This study provides a scalable AI tool to resolve the dilemma and enriches theories linking AI support to learner autonomy in education.