AI翻译与人工翻译可读性对比研究——以第三十七届韩素音国际翻译大赛材料为例

A Comparative Study of Readability between AI Translation and Human Translation—A Case Study of the 37th Han Suyin International Translation Contest Material

摘要: 随着大语言模型持续迭代并加速进入翻译实践场景，围绕AI翻译与人工翻译质量差异的讨论在学界日益升温。本研究以第三十七届韩素音国际翻译大赛的英译汉散文节选为案例，对大型语言模型(以DeepSeek为代表)生成的AI译文与大赛官方发布的参考(人工)译文进行了文本可读性对比分析。研究采用量化方法，使用“AlphaReadabilityChinese”可读性分析工具，从词汇、句法和语义三个维度下的九项具体指标进行测量。研究发现，AI译文的整体阅读难度高于人工译文。在细分维度上，二者各有优劣：人工译文在语义精确度和清晰度上表现更佳，词汇选择更简约，话题推进更集中；而AI译文则在句法丰富度和语义丰富度上具有优势，语言形式更多样，话题覆盖更广。研究旨在从量化视角为AI翻译与人工翻译的质量对比提供新的实证数据，并为翻译质量提升和译后编辑实践提供参考。

Abstract: As large language models continue to evolve and become more frequently used in translation, scholarly debate on the quality differences between AI and human translation has intensified. This study takes an English-to-Chinese prose excerpt from the 37th Han Suyin International Translation Contest as a case to conduct a comparative readability analysis between AI-generated translation (exemplified by DeepSeek) and the officially released reference (human) translation. It applies the AlphaReadabilityChinese tool to measure the two translations across nine specific indicators under the three dimensions: lexis, syntax, and semantics. Results show that the AI-generated translation exhibits a higher overall reading difficulty than the human translation. At the sub-dimensional level, each translation has its own strengths: the human translation performs better in semantic accuracy and semantic clarity, as reflected in simpler word choices and more focused topic development. The AI-generated translation shows relative advantages in syntactic richness and semantic richness, featuring greater linguistic variability and broader topic coverage. This study aims to offer new empirical data from a quantitative perspective for the quality comparison between AI-generated translation and human translation, and provide suggestions for improving translation quality and post-editing practices.