USING LLMs FOR CROSS-LINGUAL STYLISTIC ADAPTATION OF CHILDREN’S FICTION

Abstract

Background. In translating children’s literature, not only semantic accuracy but also stylistic fidelity to the original – intonational expressiveness, lexical accessibility, and age appropriateness – play a crucial role. Within cross-lingual adaptation, these factors require a delicate balance between cultural specificity and readability. Contemporary large language models (LLMs), such as GPT, offer new possibilities for automated and semi-automated translation, raising the question: how effective are they in preserving stylistic features when translating children’s fiction?

Purpose. This article explores the potential of large language models for cross-lingual stylistic adaptation of Ukrainian children’s literary texts into French, taking into account intonational, stylistic, and age-related specificities.

Methods. The study employs methods of contrastive analysis, stylistic examination of original texts and their translations, and experimental translation using an LLM (ChatGPT-4). The outcomes are evaluated in terms of register appropriateness, intonation transfer, narrative structure retention, and emotional tone.

Results. LLMs demonstrate a high level of grammatical and lexical precision, stylistic flexibility, and adaptability to child-oriented vocabulary. However, they tend to neutralize local cultural markers, may flatten the author’s unique style, and require well-designed prompts to preserve genre-specific features. At the same time, LLMs can generate multiple stylistically diverse translation options, enhancing the translator’s creative scope.

Discussion. LLMs can serve as effective tools for assisting in the translation of children’s literature, particularly in the initial stages of adaptation or in creating alternative versions. However, achieving stylistically precise and nuanced translation still requires the expertise of a professional human translator. Future research should focus on developing age- and culture-sensitive prompting strategies and improving evaluation frameworks for assessing stylistic adequacy in LLM-generated translations.