AI Teaching Assistant’s Transition from Tool to Collaborator: A Literature Review on College Students’ English-Speaking Acquisition

Abstract

As LLMs advance, English speaking instruction is shifting toward AI-assisted human–machine collaboration, yet research remains fragmented. Many studies still frame AI assistants merely as pronunciation checkers or scoring tools, overlooking their potential as learning collaborators. This review examines the literature published between 2015 and 2026, tracing how AI teaching assistants have evolved from tools to collaborators and exploring the mechanisms through which they influence English speaking acquisition. Drawing on the accuracy–fluency– complexity (AFC) three‑ dimensional model and the theoretical trajectory of computer‑ assisted language learning (CALL), the paper constructs an analytical framework structured around role evolution, mechanisms of action, and ability enhancement. The review identifies three developmental stages of AI teaching assistants—tool‑ type, interaction‑ type, and collaborator‑ type—distinguished by interaction mode, feedback nature, and role positioning. Collaborator‑ type AI supports learner autonomy, provides personalized feedback, and creates low‑ anxiety practice environments, thereby improving speaking accuracy, fluency, complexity, and logical coherence while alleviating speaking anxiety. Current research, however, remains constrained by short‑ term designs, a lack of systematic mediation analysis, and limited empirical attention to ethical risks. Future work should prioritize longitudinal tracking, human–machine collaboration mechanisms, and frameworks for responsible AI use.