Stylistic Corridors: How Generative AI Narrows Expressive Freedom In Fanfiction

Abstract

This presentation explores the results of a pilot study in which style in fanfiction was assessed using a battery of stylometrics in conjunction with an exploritory factor analysis in order to identify key stylistic dimensions common in fanfiction. Then, these scores were compared to AI continuations of these fanfictions in order to determine the ways in which large language models collapse style in the context of fanfiction. Findings suggest that human and AI writing is reliably separated across all stylistic dimensions identified, this gap is directional and dimension specific, AI writing tends to be more generic, and that these gaps are shaped by the rating of the work.