新闻语体汉维机器翻译译文质量评测

Quality Evaluation of News Stylistics Chinese-Uyghur Machine Translation Outputs

摘要: 近年来，机器翻译利用人工智能技术和专业人工翻译经验有机结合，在跨语言信息传递领域实现了快速发展与广泛应用。为了测评不同翻译软件在新闻语体汉维互译中的译文质量，帮助新闻领域的译者更有效地利用翻译系统，从而提升汉维译文质量，本文选取了必应翻译、izdax翻译、国语助手、DeepSeek-V3翻译、谷歌翻译以及小牛翻译六款机器翻译软件作为评测对象。评测以新闻语体为测试材料，涵盖了汉语译维吾尔语与维吾尔语译汉语的双向翻译任务。评测方法采用人工打分，依据忠实度和流畅度两项关键指标分别进行评分，并通过汇总计算得出各软件的综合平均得分。此外，研究还结合了笔者实际使用体验的反馈，并对新闻标题这一特定语体进行了专项测评。基于以上评测结果，本文从核心功能优化、操作体验提升、付费模式与功能匹配调整三个主要方面提出了改进建议，旨在为新闻语体译者在MTPE (Machine Translation Post‑Editing)工作流下选择合适的翻译工具提供参考，同时也为汉维机器翻译系统的针对性优化提供实证依据。

Abstract: In recent years, machine translation, enabled by the coupling of artificial intelligence technology and professional human translation techniques, has achieved rapid development and wide application in the field of cross-language information transmission. To evaluate the translation quality of different translation software in Chinese-Uyghur mutual translation of journalistic texts and assist translators in the journalism sector in utilizing translation systems more effectively to improve the quality of Chinese-Uyghur translations, this paper selects six machine translation tools, namely Bing Translator, izdax Translator, Guoyu Zhushou, DeepSeek-V3 Translator, Google Translator, and Xiaoniu Translation, as evaluation objects. The evaluation adopts journalistic texts as test materials, covering two-way translation tasks: Chinese to Uyghur and Uyghur to Chinese. Human rating is employed as the evaluation method, with scoring conducted separately based on two key indicators, namely fidelity and fluency, and the overall average score of each software is obtained through aggregated calculation. In addition, this study incorporates feedback from the author’s practical usage experience and conducts a special evaluation on the specific genre of news headlines. Based on the above evaluation results, this paper puts forward improvement suggestions from three main aspects: optimization of core functions, improvement of operation experience, and adjustment of the matching between paid modes and functions. The study aims to provide a reference for translators of journalistic texts to select appropriate translation tools under the MTPE workflow, and also offer empirical evidence for the targeted optimization of Chinese-Uyghur machine translation systems.