Utilizing Translation Memory-based Prompting to Enhance GPT-4’s Translation Performance: A Case Study

Abstract This study investigates the integration of Translation Memory (TM) with GPT-4 to enhance its translation capabilities. It employs a TM-based prompting strategy that draws on the 2023 Chinese Government Work Report to inform the translation of the 2024 version of the Report. Methodologically, the research compared TM-based prompts with non-TM-based prompts and official translations, evaluating translation quality through Bilingual Evaluation Understudy (BLEU) scores and human assessments. Contrary to expectations, the results indicate that TM-based prompting did not significantly improve the accuracy or contextual relevance of translations, compared with the baseline methods. This suggests that, while TM can provide context, it may not always translate with higher quality, without further prompting or optimizations.