摘要
本文利用机器学习算法,基于自建《国富论》人机多译本平行语料库,通过分类、聚类、特征选择实验,考察人类专家译本与大语言模型机器译本在组间和组内的翻译风格差异。结果表明,人机译本组间分类效果最好,N元语法特征突显出机器译本受原文影响较大、直译较多,而人类译者能够发挥主体性;人工译本组内分类效果较好,助词、逗号比例、平均词长等特征表明郭大力译本行文简洁明了,语法显化及欧化程度相对较低;机器译本组内分类效果最差,机器译本之间的翻译风格差异并不明显。
This research applies machine learning(ML)algorithms,namely classification,clustering,and attribute selection experiments,to explore stylistic differences in human translations(HT)and large language model machine translations(MT)based on a parallel corpus of Chinese translations of The Wealth of Nations.The results show that the HT-MT inter-group comparison performs the best,where N-gram features reveal that MT translate the original in a literal and direct manner,while HT exhibit more adaptability.Comparison within HT group performs well,with features like auxiliary ratio,comma ratio,and average word length highlighting unique characteristics of Guo Dali's translation—clear,concise,and less explicit and Europeanized than the others in linguistic terms.Comparison within MT group performs relatively the worst,manifesting the least stylistic differences.
出处
《外语导刊》
北大核心
2025年第1期120-130,160,F0003,共13页
Foreign Languages Bimonthly
基金
国家社会科学基金项目“神经网络机器翻译的译后编辑量化系统模型研究”(19BYY128)
国家留学基金委员会2024年国家建设高水平大学公派研究生项目(202406260211)。
