A Comparative Analysis of Style-Specific Linguistic Features in AI-generated Texts: The Case of editGPT

다국어 초록 (Multilingual Abstract)

This study explores the stylistic capabilities of editGPT, an AI-powered text editing tool, by examining how it transforms identical source texts into four distinct registers: academic, formal, casual, and speech. Drawing on a specialized corpus derived from the ICIC Fundraising Discourse dataset, the research investigates style-specific linguistic features using both lexical and syntactic metrics. Lexical features analyzed include overused words, nominalizations, and measures of lexical complexity, such as density, sophistication, and variation. Syntactic analyses address parts of speech distribution, passive constructions, and complexity indices, such as clause embedding and sentence length. Statistical methods, including ANOVA and post-hoc tests, confirm significant differences across styles, aligning with genre conventions while also yielding unexpected patterns, such as high nominal complexity in casual texts. The study highlights how editGPT operationalizes stylistic norms by modifying lexical choices and structural features without altering semantic content. Findings suggest that AI-generated texts can effectively model stylistic variation, making them valuable resources for teaching stylistic awareness in language education. This research advances understanding of AI’s role in language learning and stylistics, and underscores the importance of style-sensitive writing instruction in EFL contexts.