Beyond the Algorithm: Limits of Artificial Intelligence in Preserving Literary Style and Subtext

The article investigates the capabilities and fundamental limitations of neural machine translation and large language models (GPT-4, Gemini) in rendering complex literary prose. Based on a comparative analysis of John Fowles’s novel The Magus and its Ukrainian translation by Oleh Korol, the study evaluates artificial intelligence systems across metrics of syntactic accuracy, lexical fluency, stylistic appropriateness, and cultural adequacy. Grounded in the theoretical framework of dynamic equivalence (E. Nida), foreignization (L. Venuti), and Ukrainian translation studies (V. Koptilov, I. Korunets, R. Zorivchak), the research demonstrates that while modern AI tools excel at surface-level grammatical alignment and standard terminology, they consistently fail to preserve metaphorical density, authorial voice, and contextual subtext. The paper concludes that human expertise remains irreplaceable in literary translation, positioning AI as an assistive post-editing tool rather than an autonomous creator.