The Effect of AI-Generated Texts on the Internal Grammar Formation of L2 English Learners

Abstract

The extensive use of AI-generated texts, chiefly those manufactured by bulky language models (LLMs) like ChatGPT, is swiftly renovating second language (L2) learning settings. These tools are gradually used for reading, writing aid, grammar modification, and vocabulary acquisition. However, their influence on the internal grammar formation—the involuntary grammatical acquaintance system that inspires fluent language use—remains underexplored. Drawing on second language acquisition (SLA) theories such as the Input Hypothesis, Noticing Hypothesis, and Usage-Based Learning, the current study explores how exposure to AI-generated texts may affect L2 English learners' acquisition of grammatical structures. A proposed mixed-method study is introduced, designed to compare learners showed to AI-generated in competition with human-authored texts. Results from preceding and hypothetical findings suggest AI texts may improve accuracy in frequent grammatical structures, but lack of linguistic variability may hinder acquisition of complex or irregular patterns. This paper concludes with pedagogical recommendations and a discussion of implications for SLA theory in the AI era.

Keywords: L2 learners, AI-generated texts, grammar acquisition, internal     grammar, ChatGPT, SLA