Beyond Machine Translation: Studying the Role of Prompts in AI-Assisted Translation Quality Across Levels of Textual Complexity

The growing use of large language models (LLMs) in translation has drawn increasing attention to the extent to which prompting can influence the quality and characteristics of AI-generated translations. While previous research has demonstrated that different prompting strategies can affect translation performance, less is known about the nature of the textual changes produced by prompt variation. The present study examines whether changes in translation prompts lead to substantive syntactic and stylistic transformation or primarily result in lexical and surface-level variation. To investigate this issue, eight passages from Tintin: The Complete Companion by Michael Farr were initially selected through purposive sampling and translated into Persian using ChatGPT under five different prompting conditions, with five information-rich cases subsequently examined in detail. The passages were purposively selected for differences in lexical and structural complexity and for the presence of semantic voids, culture-specific elements, and idiomatic expressions. The AI-generated translations were compared with a translation produced by a qualified human translator. The analysis focuses on syntactic stability, lexical variation, naturalness and readability, and pragmatic and cultural equivalence. By examining these dimensions, the study provides a more fine-grained account of how prompting shapes AI-assisted translation and of the extent to which prompt variation influences the reconstruction of source-text meaning and style in the target language. The findings suggest that prompting primarily affects lexical choice, phrasing, and register, while syntactic structure remains relatively stable across prompting conditions; deeper pragmatic and cultural adaptation occurs less consistently, particularly in passages involving idiomatic expressions, cultural references, humor, and character-specific language.

کلیدواژه ها:

AI-assisted translation ، large language models ، ChatGPT ، prompting ، textual complexity ، literary translation