Beyond anxiety reduction : Psychological transformation and systemic contradictions in generative-AI-mediated language learning – A cultural-historical activity theory perspective

Choi, W. (2026) Beyond anxiety reduction : Psychological transformation and systemic contradictions in generative-AI-mediated language learning – A cultural-historical activity theory perspective. Acta Psychologica, 267: 107055. ISSN 0001-6918

Abstract

This study explores the psychological transformation and contradictions arising from Generative AI (Gen-AI)-mediated language learning. Using Cultural-Historical Activity Theory (CHAT) as an analytical lens—operationalized through Engeström's (2015) activity system triangle to analyze interactions among system components and identify multi-level contradictions—the research examines how Gen-AI reshapes learners' cognitive, affective, and identity processes within the broader learning activity system. A mixed-methods design was employed with 60 Korean high school EFL learners, combining quantitative measures of pronunciation ability and anxiety with qualitative interview data. Results revealed statistically significant changes in pronunciation scores and reductions in speaking and pronunciation anxiety, accompanied by increased metacognitive awareness and shifts in linguistic identity. CHAT-based analysis demonstrated how Gen-AI functioned as more than a mere tool, reshaping learners' goals from native-like pronunciation to effective communication, fostering metacognitive abilities, and facilitating positive identity transformations. However, several contradictions also emerged, including inconsistent feedback, cognitive overload, over-reliance on AI, personalization biases, privacy concerns, and gaps between AI-mediated practice and authentic communication. These tensions highlight both the developmental potential and psychological risks of human–AI interaction in education. Findings suggest that effective Gen-AI integration requires addressing these psychological tensions through digital literacy education, balancing automated practice with authentic interaction, and reconsidering assessment frameworks to align with communicative competence rather than native-like accuracy.