AI Large Language Models and Second Language Writing Development: A SLA Perspective
TL;DR
The study concludes that LLMs can serve as effective SLA mediators within structured, teacher-guided pedagogy.
Abstract
: This study examines how AI large language models (LLMs) affect second language (L2) writing development from a second language acquisition (SLA) perspective. Drawing on the Input, Noticing, Interaction, and Output Hypotheses and Sociocultural Theory, it investigates whether LLM-mediated instruction supports accuracy, lexical complexity, syntactic variety, coherence, and metalinguistic awareness. A mixed-methods quasi-experimental design involved 82 Chinese EFL undergraduates in an LLM-assisted experimental group (n = 42) and a teacher-feedback control group (n = 40) over 16 weeks. Writing tests were analyzed with t-tests and ANCOVA, while interviews and journals supplied qualitative evidence. Results showed significantly stronger gains in the experimental group (p < .001), especially in accuracy, lexical diversity, and coherence. Learners also reported stronger noticing, richer input, more pushed output, and lower anxiety, though over-reliance, generic feedback, and integrity concerns remained. The study concludes that LLMs can serve as effective SLA mediators within structured, teacher-guided pedagogy.