N元组和翻译单位对英译汉自动评分作用的比较研究  

A comparative study of Ngram and translation unit alignment in automated scoring of students’English-Chinese translation

作　　者：江进林[1] 文秋芳[1] 

机构地区：[1]北京外国语大学

出　　处：《现代外语》2010年第2期177-184,共8页Modern Foreign Languages

基　　金：教育部人文社科重点研究基地重大项目"大规模考试主观题(英汉互译)自动评分系统的研制"的资助(批准号07JJD740070)

摘　　要：N元组匹配和翻译单位对齐各有优劣。本文采用320篇学生英译汉译文探讨了它们在自动评分中的取舍问题。研究比较了N元组匹配数量和翻译单位对齐数量与人工对译文语义、形式、总评分的相关性,并采用多元回归考察了它们对译文质量的解释力。结果表明:(i)翻译单位对齐数量与人工评分的相关度高于绝大多数词-和字-N元组匹配数量;(ii)与N元组匹配数量的整体作用相比,翻译单位对齐数量对语义评分的解释力稍高,对形式和总评分的解释力稍低;(iii)与仅以N元组匹配数量或翻译单位对齐数量为自变量的模型相比,词-一元组和翻译单位对齐数量结合产生的模型对人工评分的解释力更强,模型评分与人工评分的相关度和一致性也更高。这表明词-一元组匹配与翻译单位对齐互为补充,两者结合对译文质量的预测效果最佳。Ngram is an important quality predictor in machine translation evaluation,but it does not take context into full consideration.When evaluating human translation,it ignores the translating process.This study investigates the automated scoring of 320 students’English- Chinese translations.In order to match translating practice,it adopts'translation unit'(TU)and makes TU alignment based on self-made dictionary.Then it compares the correlations between Ngram and human scorings of meaning,form,and overall quality of translations with those between aligned TU number and scorings.It further explores the predicting power of Ngram and aligned TU number with multiple regression analysis.The research indicates that: (i)aligned TU number is more correlated to scorings than most word-and character-based Ngram;(ii)aligned TU number has greater explanatory power for meaning scoring than overall Ngram,but lower power for form and overall scorings;(iii)models with word-based unigram and aligned TU number as independent variables explain more scorings than those with Ngram, and their calculated scores are more correlated to and consistent with human scorings.Therefore, the combination of word-based unigram and aligned TU number has the best predicting effect on translation quality.

关 键 词：英译汉 自动评分 N元组 翻译单位 

分 类 号：H319[语言文字—英语]