大语言模型下科技文本译文质量比较研究——以四种典型文本为例

A Comparative Study of the Quality of Translations of Scientific and Technical Texts under Large Language Models—Taking Four Typical Texts as Examples

摘要: 本文主要以ChatGPT和文心一言两个大语言模型为主要研究对象，着眼于大语言模型和传统机器翻译软件在四个不同类型的科技文本中的翻译对比，选取具有代表性的文本作为典型案例，将大语言模型翻译、传统机器翻译软件以及人工翻译三者进行对比，对译文的准确性和通顺程度进行评价。发现大语言模型在各方面展现出更高的准确性和流畅性，传统机器翻译尽管在速度和可获取性上具有优势，但其翻译结果往往存在流畅性不足和细节处理欠佳的问题。大语言模型在语言处理能力上虽有进步，但在精确性、用词的准确性、情感处理以及语境的把握上无法超越人工翻译，未来还有很大的发展潜能。

Abstract: This paper takes two large language models, ChatGPT and Wenxin Yiyan, as the main research objects and focuses on the comparison between large language models and traditional machine translation software in the translation across four distinct types of scientific and technological texts. Representative texts are selected as typical cases, and translations from large language models, traditional machine translation software, and human translators are compared to evaluate accuracy and fluency. It is found that large language models demonstrate higher levels of accuracy and fluency overall. While traditional machine translation offers advantages in speed and accessibility, its results often exhibit shortcomings in fluency and detail handling. Although large language models show progress in language processing capabilities, they still fall short of human translation in aspects such as precision, word choice, emotional tone, and contextual understanding, indicating significant potential for future improvement.