机器翻译系统跨领域移植方法的研究  

Research on Adapting Machine Translation System to New Domain

作　　者：张春祥[1] 赵铁军[1] 杨沐昀[1] 李生[1] 时晓升[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《计算机工程与应用》2005年第3期10-11,66,共3页Computer Engineering and Applications

基　　金：国家自然科学基金(编号:60375019);国家863高技术研究发展计划(编号:2002AA117010-09)资助

摘　　要：机器翻译系统跨领域移植是降低系统开发代价的关键性因素。论文以翻译模板对训练语料库机器译文评测分数的贡献为依据,对其进行评价。从模板库中过滤出适应目标领域的翻译知识,以实现系统向新领域的快速移植。利用论文所提出的方法将一个通用领域的机器翻译系统向旅游领域进行移植,并使用来自旅游领域的2469句汉语句子作为测试语料进行开放测试。实验结果表明在开放测试中,机器译文的5元Nist评分提高了2.2807,上升了71.1%。Adapting machine translation system to new domain is key to decrease the cost of system development.In this paper,translation pattern is evaluated based on its contribution to machine translation assessment score of training corpus.Translation knowledge unsuitable to target domain is deleted from pattern base,which leads MT system transfered to new domain quickly.The proposed method is applied to adapt a general domain MT system to traveling field,and2469Chinese sentences from traveling field are used for open test.Experiment result shows that in open test,automatic assessment score of machine translation increases2.2807under5gram Nist metric and the assessment score is improved by71.1%.

关 键 词：译文评测分数 机器翻译系统 Nist评分 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]