Exploring the Naturalness of AI-generated Texts

In recent years, the development of Chat Generative Pre-Trained (ChatGPT) enables computers to communicate in human-like language. To understand how closely they approximate the syntactic characteristics of natural human discourse, this book offers a systematic investigation into the readability of texts produced by ChatGPT-4 by adopting cognitive load as the central analytical lens. It takes a corpus-based approach to measure readability through a quantitative analysis of dependency distance, which captures syntactic structures that modulate processing effort to reveal distinct constructional strategies employed by ChatGPT-4, highlighting statistically significant divergences from human writing practices.