ChatGPT译文风格生成:评析与思考  

The Analysis and Reflection on Text Style Generation of ChatGPT Translation

作　　者：张政[1] 王赟 

机构地区：[1]北京师范大学外国语言文学学院,北京100875 [2]太原学院外语系,太原030032

出　　处：《翻译研究与教学》2024年第1期10-15,共6页Translation Research and Teaching

基　　金：教育部人文社科项目“中国古代科技典籍英译数据库构建与传播研究”(项目编号:22YJC740081);国家社会科学基金青年项目“The Formulation of Database and Spreading of Chinese Ancient Classics on Science and Technology Translated in English”(项目编号:22YJC740081)的阶段性研究成果

摘　　要：OpenAI于2022年11月发布ChatGPT,即聊天生成型预训练变换模型,将机器翻译的发展带入全新模式,但其生成的译文在风格上与原文之间依然存在隔膜,这是自然语言处理领域不容回避的重要问题。本文以弗吉尼亚·伍尔夫短篇散文“How Should One Read a Book?”(《一个人应该怎样读书?》)的汉译为研究对象,聚焦正式性、文学性及古典性三种风格生成能力考察项,对比、分析、阐释人工译者与ChatGPT译文风格扞格的缘由,为大语言模型时代下翻译技术的发展提供思考。Since the OpenAI released the ChatGPT(Chat Generative Pre⁃trained Transformer)in November 2022,it has brought the machine translation into a new development mode.However,the gap between the generated translation style and that of the original text still exists,which is an important issue that cannot be avoided in natural language processing(NLP).Taking the Chinese translations of Virginia Woolf's short essay“How Should One Read a Book?”as the research object,this paper focuses on three style generation elements:formality,literariness,and classicality.It compares,analyzes,and comments on the reasons for the disparity in styles between human translation and ChatGPT translation,in order to provide prospects for the development of translation technology in the era of large language models(LLMs).

关 键 词：ChatGPT 英汉翻译 译文风格 生成 

分 类 号：H315.9[语言文字—英语] I046[文学—文学理论] H085[自动化与计算机技术—控制理论与控制工程] TP18[自动化与计算机技术—控制科学与工程]