Exploring the Role of Translation Brief Elements in Prompts to Large Language Models

Unlike conventional machine translation (MT), large language models (LLMs) are unique in producing various translations based on different prompts. Researchers have suggested designing detailed translation prompts to improve the translation output of artificial intelligence (AI)-powered LLMs. This study investigates how LLMs respond to prompts that include extratextual information typically included in a translation brief to a human translator. The underlying goal is to probe LLM models’ potential in human translator work and training. To answer the research question, three passages from tourist brochures were submitted to ChatGPT with five prompts, which included information on the text function, translation purpose, and target audience. To identify any effect on the output, cultural references and promotional devices (e.g., imagery, personal deixis, and emotive words) were identified in the passages, followed by a comparison of their translations across the five texts. The study’s significance lies in applying concepts in translation theories such as the translation brief to the novel field of prompt design, moving beyond the usual linguistic accuracy metrics of MT assessment. The analysis showed that ChatGPT did not recognize cultural references as elements that might require explication for the target audience. It also rendered textual persuasive devices faithfully regardless of any changes in the prompts. These faithful translations were not always successful in preserving the textual functions of persuasion. The results suggest that more specific and detailed prompts are needed to bring MT closer to human translation.