汉译英自动评分系统中的语义相似度比较研究  

A comparative study of semantic similarity in the model for the computer-assisted scoring of Chinese EFL learners' Chinese-English translation

作　　者：王金铨[1] 

机构地区：[1]扬州大学外国语学院,江苏扬州225009

出　　处：《外语测试与教学》2011年第4期8-17,共10页Foreign Language Testing and Teaching

基　　金：扬州大学校级教改课题"基于计算机辅助翻译平台的应用型翻译人才培养模式研究"部分成果;国家社科基金青年项目"中国学生汉译英机助评分模型的研究与构建"(项目编号09CYY042)的部分成果

摘　　要：本研究围绕汉译英自动评分系统中的语义相似度,考察了N-gram、潜在语义分析和语义点与学生译文语义成绩之间的相关关系以及它们在评分模型中的预测力,数据表明:这三大语义变量能够较好地预测译文语义质量的优劣,其中语义点的预测力最强,SVD最弱;总语义模型的相关系数和决定系数R2最高,分别达到0.891和0.794,SVD语义模型的相关系数和决定系数R2最低;三大语义变量之间的关系是互补的,由它们组成的总语义模型的机器评分与人工评分之间的相关系数与语义点相差无几,但内部一致性略好于单个语义点模型。This study, focusing on the semantic similarity measures in the computer-assisted scoring model of Chinese EFL learners＇ CE translation, compares the correlation relallonship between human rater score and the three semantic variables： N-gram, latent semantic analysis and semantic points. Furthermore, the study investigates the predicting power of the three variables. Findings of the research indicate that the three semantic variables can satisfactorily pre- dict the semantic quality of the Chinese EFL students＇ CE translation with the semantic point variable ranking first, SVD variable last; the multiple R of the overall semantic scoring model is 0. 891, accounting for 79.4% of the vari- ance, which is the highest in all the four scoring models; the relationship among the three semantic variables is com- plementary. The computer-human meaning score correlation coefficient of the overall semantic scoring model and semantic point model is almost the same, but the Cronbach＇s coefficient alpha of the former is a bit higher than the lat- ter, showing better intra-rater reliability.

关 键 词：汉译英 自动评分 语义相似度 

分 类 号：H319[语言文字—英语]