Mechanisms of Alleviating Second-Language Writing Anxiety through ‘Creative Imitation’ in the Context of Large Language Models

Abstract

The rapid growth of generative artificial intelligence has made large language models (LLMs) such as ChatGPT a regular presence in second-language (L2) writing. Even so, how these tools act as dynamic scaffolds — and the cognitive and affective mechanisms behind that process—remain relatively unclear. Grounded in Vygotskian sociocultural theory and Cognitive Load Theory, this study examines the effects of LLM interaction and alignment on L2 academic writing. It also proposes a theoretical framework and a three-step collaborative approach to test whether ‘creative imitation’ can reduce L2 writing anxiety. The findings indicate that LLMs work well as dynamic scaffolding tools: they can substantially lower cognitive load, strengthen learners’ writing self-efficacy, and make L2 academic tasks more enjoyable. At the same time, poorly guided artificial intelligence (AI) interaction or excessive dependence on these systems may bring latent language anxiety to the surface. This study concludes that the introduction of LLMs shifts the traditional alignment and dialogic mechanisms in L2 writing. It is recommended that educators dynamically adjust scaffolding strategies based on task complexity to optimize learners’ emotional experiences and sustainably accelerate L2 academic writing proficiency.