From default output to guided translation: Effect of prompt engineering on AI literary translation quality

Large language models (LLMs) are developing tools for literary translation thanks to breakthroughs in artificial intelligence (AI), but their default outputs frequently fail to capture the stylistic and aesthetic elements of literary texts. This work investigates the use of rapid engineering as mediating method for increasing the quality of AI generated literary translation. Using extracts from selected literary text, the study compares AI outputs to prompt guided translations, focusing on semantic fidelity, stylistic coherence, imagery and tone. The approach, which is based on descriptive translation studies and literary stylistics, examines how focused prompting can affect translation outcomes. The findings show a higher semantic and stylistic correctness in prompt-guided translation with the source text. The paper argues that prompt engineering is a significant type of human intervention in AI-assisted translation, establishing the active role of a human translator in shaping the translational intent rather than just a post-editor.