Design and Evaluation of Prompts for Large Language Model Translations in Specialized Domains

In recent years, Large Language Models (LLMs) have made significant breakthroughs in natural language processing. However, effectively applying them in domain-specific translation remains a significant challenge. This study aims to enhance the translation quality of LLMs in specialized domains by investigating the impact of different prompt strategies. We constructed four domain-specific corpora and assessed the efficacy of four prompting strategies—code-based, natural language-based, CRISPE mode, and comprehensive mode—using automated metrics such as BLEU, ChrF, and TER on the ChatGPT 3.5 model. Results indicate that different prompting strategies significantly affect LLM translation quality. The natural language-based and CRISPE mode prompts showed the best performance, providing richer contextual information and clearer task instructions, effectively enhancing the performance of LLMs in specialized domain translations. This study offers crucial empirical evidence to improve translation performance in specific domains and propels new directions for LLM-based translation practice and research.