Beyond Stylometry: Measuring Narrative Identity Preservation in AI-Rewritten Literary Fiction Through Computational and Literary Evaluation

Abstract

Background: The increasing use of generative artificial intelligence in literary rewriting raises concerns about whether semantic similarity adequately reflects the preservation of narrative identity, including stylistic distinctiveness, narrative voice, characterization, and interpretive complexity.
Objective: This conceptual study aims to develop a multidimensional framework and an empirically testable validation protocol for evaluating narrative identity preservation in AI-rewritten literary fiction by integrating computational and literary perspectives.
Method: A conceptual integrative review was conducted on six purposively selected empirical studies published in 2025 that address computational stylometry, literary style imitation, fiction evaluation, reader experience, and interpretive assessment, supported by foundational works in narratology, stylometry, and literary natural language processing. Thematic synthesis was used to derive the framework, which was then operationalized into a paired source–rewrite experimental protocol.
Results: The synthesis yielded four themes: the limits of stylometric similarity, the gap between semantic fidelity and narrative structure, discrepancies between automated and human literary judgment, and the need for source-relative evaluation. These themes were integrated into a three-layer framework comprising semantic, stylistic, and narratological continuity, each linked to specified indicators, annotation criteria, and validation tests.
Conclusion: Narrative identity preservation is better conceptualized as a multidimensional, source-relative construct than as semantic similarity alone. The framework offers methodological guidance for computational literary studies and responsible AI-assisted adaptation, although its reliability and predictive validity require empirical confirmation.