C-E Translation Technology For Product Names Based On The Structure And Examples

With the development of economy,as a great power of manufacturingindustry,our country produce a large number of products every year.And it stillgrows at a very fast way.With the deepening of international trade and economiccooperation,There are a large number of old and new products are exported overseasevery year.So the translation of the name of products is a very worthwhilearea.There are some differences between the name of products and traditional namedentities, we can not use the method of named entity translation directly. In order toimprove the accuracy of translation, we analyzes the structure characteristics of thenames of products.In combination with the other machine translation technology,weput forward a method of translation specific to the names of products.Main contentsof this paper involves the following:(1) Combined with the structure features of the name of product and the name ofthe enterprise,some different methods are proposed in this paper to mining the nameof brands in products.We get a large number of high precision names of brands,witch has a high value in the rapid economic development today.(2) Put forward a kind of method for brand name recognition based on CRF.Weused the name of brand and the name of product tagging the corpus,and got ahigh-precision CRF model.It maked up for the error of the segment of the name fobrand in the product,and improved the quality of translation of name of products.(3) In this paper, we got the statistical dictionary from bilingual corpus and thetranslation patterns from bilingual product names using the technique of wordalignment based on statistics. By the dictionary,we can translate the word outsidethe place name and the brand name.By the patterns, we can adjust the order of thetranslation of word outside the palce name and the brand name.It maked thetranslation result easy to understand.Thereby we implement the preliminarytranslation of the product name.(4) Puts forward a rapid and effective method for classification of productnames.As a result,the product name is divided into hundreds of classes.So we cangreatly improve the performance of the template thinning, and at the same time,doesnot reduce the amount of template.Results show that,the classification result andaccuracy meet the requirements of this system basically.(5) On the basis of the classification,combining with the example based machinetranslation technology.We refinement of the templates using the bilingual productname.Template thinning has brought significant effect for the improvement oftranslation quality.