Research On Automatic English-chinese Term Translation Technology Based On Multi-strategy

Patent documentations translation plays an important role for the spreading andexchanging of scientific information. As an important component of patent documentations,the translation of terms has a significant effect on the translation quality of patentdocumentations.At present, machine translation method has been fully developed, various kinds oftranslation systems emerged one after another. When translators translate the terms in thepatent documentations, they often translate them with the help of these translation systems.At the same time, in the process of translation for patent documentations, translators willproduce large translation examples stored in the term example library and can use it tooptimize the results from the translation systems, thereby provide better reference for thetranslators to assist them translate the term.This paper presents an automatic English-Chinese term translation method based onmulti-strategy. The work of this thesis mainly consists of two parts as follows:Firstly, through analyzing the characteristics of terms, this paper introduced three kindsof features that adequately describe term translation fluency, compactness and loyalty, namelyterm word formation feature, mutual information feature and term example library feature.These three features are combined with basic features, and regression algorithm in machinelearning is used to predict the translation quality so as to select the best result.Secondly, this paper optimized the term translation results with low quality scores by theterm example library. String similarity algorithm is used to find several term translationexamples that are highly similar to the source language of the term to be optimized in theterm example library. According to the target language of these examples and the translationof the term to be optimized, a translation grid is constructed. With translation model featureand language model feature, the Viterbi decoding algorithm is adopted to search a top scoringpath in the grid, thereby generate a new term translation. On the basis of the above work, an English-Chinese term translation system based onmulti-strategy is constructed. Experimental results show that the method achieves goodperformance in term translation quality estimation and term translation optimization. Thesystem can help translator to translate the term of patent documentations.