Abstract
Due to the lack of domain classification theory for domain-specific machine translation, the quality of translation in this area is low. We propose a domain classification system based on HNC and design a new method that can enhance domain-specific machine translation by jointly using this system with alarge language models. We propose a multi-agent system for domain-specific machine translation and a prompt generation method guided by the domain classification system. Tests of cross-lingual translation in the domains of science and technology, health and culture on open-data test sets and English-Chinese translation in the domains of politic, economy, military, and culture on human-generated test sets show our method successfully improves the capability of domain-specific machine translation of LLM. Finally, a real case is provided to demonstrate the workflow of the proposed method.