中国文化术语翻译的AI译后编辑研究——以中医术语英译为例

AI Post-Editing of the Translation of Chinese Cultural Terms—A Case Study of English Translation of TCM Terminology

摘要: 随着生成式人工智能和大语言模型(Large Language Models, LLMs)逐渐进入翻译生产流程，AI翻译已经由传统的词句转换工具发展为能够结合上下文生成译文、解释译法并参与术语决策的语言技术。然而，中医术语并不是普通的词汇集合，而是由哲学概念、医学知识、诊疗方法和文化观念共同构成的专业术语系统。针对AI在中医术语翻译中出现的概念泛化、术语不规范、文化内涵缺失和上下文适配不足等问题，本文以中医术语英译为研究对象，结合功能对等理论、目的论以及归化与异化理论，对AI生成译文与权威人工译文进行对比分析。在此基础上，本文构建“问题–策略”对应框架，将不同类型的AI翻译偏差分别对应到音译保留与意译补充、专业词汇规范化、语境补充与体系化、分级译后编辑四类处理策略。研究发现，AI在表层语义转换和一般性语言生成方面具有较高效率，但在中医核心概念的边界识别、术语系统的一致性维护以及文化知识的准确传递方面仍需人工介入。本文的主要贡献不在于笼统评价AI翻译的优劣，而在于将传统翻译理论具体转化为面向中医术语后编辑的判断标准，并建立“偏差类型–理论解释–编辑策略”的对应关系，为文化负载型专业术语的AI辅助翻译提供可执行的分析框架。

Abstract: With the rapid development of generative artificial intelligence and large language models (LLMs), AI-based translation has evolved from a tool for sentence-level conversion into a language technology capable of generating context-sensitive translations, explaining translation choices, and assisting terminology-related decisions. However, Traditional Chinese Medicine (TCM) terminology is not merely a collection of lexical units. It constitutes a specialized conceptual system involving philosophical categories, medical knowledge, diagnostic principles, and cultural values. Focusing on the English translation of TCM terminology, this study compares AI-generated translations with authoritative human translations through the perspectives of Functional Equivalence Theory, Skopos Theory, and domestication/foreignization. On this basis, it establishes a problem-strategy correspondence framework, which links four major types of AI-related problems to four post-editing strategies: transliteration plus explanatory translation, professional terminology standardization, contextual supplementation and systematization, and graded post-editing. The findings indicate that AI performs efficiently in surface-level semantic transfer and general language generation, but still requires human intervention in identifying the conceptual boundaries of TCM terms, maintaining terminological consistency, and conveying culture-specific meanings. The contribution of this study lies not in making a general judgment on the quality of AI translation, but in transforming classical translation theories into operational criteria for post-editing TCM terminology and establishing a correspondence between translation problems, theoretical explanations, and editing strategies. The proposed framework may provide a practical reference for AI-assisted translation of culturally loaded specialized terminology.