机器学习算法在翻译风格研究中的应用

On Using Machine Learning Methods to Discriminate Translation Styles

支持向量机是文本分析领域应用最多的机器学习算法之一,但其在翻译风格研究中的应用却并不多见。本研究基于语言计量特征,运用机器学习的方法对比王科一、张经浩两译者《傲慢与偏见》译本的翻译风格。首先,用信息增益算法获取两译者具有差异性的语言特征,然后基于这些特征构建文本分类器即支持向量机(SVM),对两译者的翻译文本进行自动分类,最后用十折交叉法对分类器的有效性进行验证。研究证明,机器学习算法分类准确率高,且具有很强的预测功能,有助于判定或预测译本或者译者的风格。与传统方法相比,该方法省时省力,过程可重复,结果精确可靠,便于处理大量陌生语料和海量特征,能发现文本间更细微的特征差异,为大数据时代的探索性翻译研究提供了便利,是翻译研究科学化的一种新的尝试。另外,本研究结合统计数据对差异性语言特征进行分析,分析结果与前期研究和研究者对文本的感受高度吻合,具有很强的可解释性。

The present study is intended to use machine learning methods to discriminate translation styles. First, the informa- tion gain method is used to single out nine most distinctive stylistic markers; then the SVM classifier is employed to classify the translation texts; and at last the ten-fold cross-validation method is employed to test the validity and the precision rate of the classifier. The test proves that this machine learning method earl effectively detect delicate and subtle linguistic differences, rank them in the order of distinctiveness, and classify the translation texts with high precision rate. Further analysis shows that this classification result accords with statistical and qualitative analysis, in which there is a global difference between the two translations, with one comparatively closer to the source text and the other to the target text. This study suggests that SVM classifier is a reliable method in discriminating translation styles, and that its model has good predictive capacity, which could be used to predict the style of the translator of an unknown translated text when necessary.