英译汉语言质量自动量化研究

An automatic approach to evaluating the linguistic quality of English-Chinese translations

本研究在总结前人经验的基础上,针对中国学生英译汉的特点,在三种文体、近1000篇译文中提取了20多个能够反映语言形式质量的文本特征。研究进一步采用多元线性回归方法考察了这些量化指标对每种文体内一半译文语言形式分数的预测力。研究结果表明,回归方程对同一题目另一半译文的评分取得了理想效果,人机评分的相关度和一致性良好。因此,本文提取的特征可用于构建自动评分系统。

Drawing on previous research,this study investigates how to automatically evaluate the linguistic quality of English-Chinese translations for Chinese EFL learners.Over 20 textual features which can predict the linguistic quality are extracted from about 1000 translated texts of three different types.Multiple linear regression analyses are performed for half of the translated texts of each type to establish equations by which the features can best predict the human-assigned scores.Then the formulas are used to compute scores for the other half of the texts.The results indicate that computer-and human-assigned scores are highly correlated and consistent.Therefore,the proposed features can be used to construct an automatic evaluation system.