摘要
作为深度学习语言模型的ChatGPT,在普通文本翻译中翻译领域表现卓越,但应用于特定背景和用途的文本,例如中央政治文献等,性能尚需深入研究。因此,本文以2023年政府工作报告为例,利用ChatGPT进行汉译英翻译,建设含有原文、官方译本和ChatGPT译本的平行语料库,从丰富性、可读性、规范性三个角度对比官方译本与ChatGPT译本,进而总结发现:ChatGPT译本所用词汇重复度较高、规范性较低,但具备一定的可读性。在大规模的预训练过程中,大语言模型输入得到广泛语言知识,具有通用性和灵活性;但预训练并非基于特定领域,因此在处理专业领域的文本时,缺乏对特定术语和专业性语言的准确理解。由此,本文进而提出人工翻译核心素养的培养,并对翻译工作尤其译后编辑提供建议。
ChatGPT,as one of the deep learning language models,reveals impressive performance in translating general text.However,further research should be conducted on its performance in translating texts for specific purposes,such as political literature.Therefore,this study takes the 2023 Government Work Report as an example,inputs translation prompt in ChatGPT,and constructs a parallel corpus of the original text,official translation,and ChatGPT translation.After a close examine in terms of lexicon,readability,and normativity,it is found that the ChatGPT translation exhibits higher word repetition and lower standardization but possesses certain readability.LLM(Large Language Mode)may acquire extensive linguistic knowledge after pretraining,but it lacks precision in specific fields.Consequently,this study discusses further about the cultivation of core competencies in human translation and provides recommendations for translation work,especially post-editing.
作者
于天博
Yu Tianbo(Beijing Normal University,Beijing,100875,China)
出处
《语言与文化研究》
2024年第4期154-158,共5页
Language and Culture Research