基于多特征的英汉术语译文质量自动评价  

Automatic qualityestimationfor English-Chinese term translation based on multiple features

作　　者：林康静 李楠[1] 叶娜[1] 蔡东风[1] 

机构地区：[1]沈阳航空航天大学知识工程研究中心.沈阳110136

出　　处：《沈阳航空航天大学学报》2014年第6期59-65,共7页Journal of Shenyang Aerospace University

基　　金：国家"十二五"科技支撑计划项目(项目编号:2012BAH14F00)

摘　　要：该文通过深入分析术语的语言学特点和领域特点,引入了能充分刻画术语译文流利度、紧密度和忠实度的三类特征,即术语构词、互信息和术语实例库,将其与基础特征进行组合,采用机器学习方法中的排序学习算法将列表排序问题分解成二分类问题,最后利用最大熵分类器预测多个翻译系统的术语译文质量。实验结果表明,在评价英汉术语翻译任务上的多个系统的输出译文质量时,该文引入的特征,可以进一步提高术语译文质量评价结果和人工评价结果的一致性。By analyzing term＇ s linguistic and domain characteristics, three kinds of features, namely, term word-formation, mutual information and term example library, whichcan adequately describe term- translation fluency, compactness and loyalty, were introduced. Then those three kinds of featureswere- combined with basic features. List ranking has been transformed into a binary classification problemwith the application of ranking algorithm. The quality of term-translation from multiple translation systems was predictedover the maximum entropy classifier. The results show that, the proposed features, when assessing the translation quality of the submitted systems outputs of English-Chinese term-translation tasks, canfurther improve the consistency between the results of the automatic estimation and human judgments.

关 键 词：术语 特征 质量评价 最大熵分类器 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]