ASSESSING LINGUISTIC NATURALNESS IN CHATGPT-GENERATED ENGLISH: A COMPARATIVE STUDY WITH HUMAN-WRITTEN TEXTS

As large language models (LLMs) like ChatGPT become increasingly integrated into educational, professional, and creative writing domains, questions surrounding the linguistic naturalness of their outputs have gained prominence. While ChatGPT excels in generating grammatically correct and contextually appropriate content, its ability to emulate human-like linguistic naturalness remains underexplored. This study investigates the extent to which ChatGPT-generated English texts align with human expectations of natural language use, focusing on aspects such as fluency, idiomaticity, coherence, and pragmatic appropriateness. Employing a comparative design, the study evaluates AI-generated texts against human-written samples across multiple genres—including essays, emails, and narratives. Data were assessed by both expert human raters and automated linguistic analysis tools. Findings reveal that while ChatGPT performs competitively in terms of surface fluency and grammaticality, it often lacks the nuanced idiomatic usage and context-sensitive expressions that characterize native-level naturalness. The study underscores the limitations of relying solely on automated metrics to assess language quality and emphasizes the need for human-in-the-loop evaluation in AI language applications. These insights contribute to ongoing discussions in applied linguistics and AI ethics, offering practical implications for education, content creation, and machine-mediated communication.