大模型技术驱动的国防科技情报智能翻译生产线实践与评估  

Large language models-driven translation:Practice and evaluation of defense science and technology intelligence translation workflow

作　　者：岳圣雅 吴思 YUE Shengya;WU Si(Beijing HIWING Scientific and Technological Information Institute,Beijing 100074,China)

机构地区：[1]北京海鹰科技情报研究所,北京100074

出　　处：《国防科技》2025年第1期37-44,共8页National Defense Technology

摘　　要：大语言模型技术的迅速发展为国防科技情报翻译领域带来了前所未有的机遇与挑战。针对大模型技术赋能国防科技情报翻译的需求特性和技术路径,提出一套以大模型技术驱动的定制化智能翻译生产线方案,包括大模型术语抽取、检索增强生成、大模型融合翻译及自动润色,并通过人工评估的方式,深入分析该方案在国防科技情报领域的翻译质量和效率提升情况。探讨国防科技领域翻译的未来展望,探索进一步提升翻译效率和质量的方案及其可能性。大模型技术在国防科技情报翻译领域展现出巨大的应用潜力,但仍需持续优化和完善。构建一个可持续的大模型翻译生产线系统,对于支撑未来精准、高适应性的智能化翻译体系至关重要。The rapid development of large language models(LLMs)presents unprecedented challenges and opportunities for the translation of defense science and technology intelligence.In this paper,a customized intelligent translation workflow driven by LLMs technology is proposed to enhance the quality and efficiency of defense science and technology intelligence translation,incorporating terminology extraction,retrieval-augmented generation,large language model-integrated translation,and automatic polishing,with quality and efficiency assessed through manual evaluation.The prospects of translation in the field of defense science and technology are discussed at the end of the paper,exploring potential solutions and possibilities for further improvement.LLMs technology demonstrates significant potential for application in defense science and technology translation,but it still requires continuous refinement.Building an enduring LLMs translation workflow is crucial for supporting a future intelligent translation framework that is precise and highly adaptable.

关 键 词：大模型技术 术语抽取 检索增强生成 翻译质量评估 

分 类 号：V19[航空宇航科学与技术—人机与环境工程] N03[自然科学总论—科学技术哲学]