基于机器学习的译者风格研究——以《红字》两中译本为例

A ML-Based Study on Translator’s Style—A Case Study of Two Chinese Translations of The Scarlet Letter

摘要: 基于机器学习的方法，采用卡方评估法提取有效特征集，使用支持向量机(SVM)分类器对《红字》傅东华与潘庆舲译本进行分类。研究发现，SVM能够有效区分两个译本，分类正确率达到98.0%。通过分析文本，揭示了两译者风格的差异。傅东华在翻译中偏好使用复杂定语、显化原文时态，词汇选择相对简洁，并频繁运用中文特殊句式，更多地使用连词和破折号作为衔接手段。潘庆舲的翻译风格更贴近现代汉语的表达习惯，倾向于隐化助词，偏好使用三字词语和四字成语来装点译文，常以冒号作为衔接手段，更加注重译文的流畅性和可读性。通过研究验证，机器学习方法在译者风格研究中具有可行性。

Abstract: Based on the machine learning approach, the distinctions was revealed between the Chinese translations of The Scarlet Letter by Fu Donghua and Pan Qingling. Utilizing the Chi-squared Attribute Eval feature selection method to extract significant style features, a Support Vector Machine (SVM) classifier was trained, achieving an accuracy of 98.0%. The analysis reveals marked stylistic divergences between the two translators. Fu’s translation exhibits a preference for complex attributive structures, explicit temporal markers, and simplified lexical choices, alongside frequent use of Chinese-specific syntactic patterns and cohesive devices such as conjunctions and dashes. In contrast, Pan’s translation aligns closely with modern Mandarin conventions, characterized by the omission of auxiliary particles, the incorporation of tri-syllabic expressions and quadri-syllabic idioms for stylistic embellishment, and a reliance on colons as cohesive markers, prioritizing textual fluency and readability. These findings underscore the efficacy of computational methods in identifying translator-specific stylistic fingerprints.