Research On Term Automatic Translation Technology In English-Chinese Machine Translation System

With the development of technology, people pay more attention to patent documents translation. Because patent documents contain a large amount of terms, term translation has a significantly effect on the translation quality of patent documents. Terms concentrate the kernel knowledge in a specific field so that with the globalization of information and the development of Machine Translation, the research of term automatic translation technology becomes a focus to patent document translation.Based on the research on correlation technique at home and abroad, this thesis implements a term automatic translation system in patent documents translation of English-Chinese. Through the analysis of noun terms in a large number of bilingualism corpuses, the source language terms need to be reordered firstly. And then the n-best candidates can be obtained after being translated by Moses. At last, the best result is chosen according to scoring the n-best result again.The work of this thesis mainly includes:Firstly, the method of term automatic translation based on head is proposed aiming at the different characteristics of English and Chinese terms in linguistics. In order to provide the correct context information during the selection of the best word translation, the words sequence of terms is reordered before the translation because the orders of words in term of English and Chinese are different. According to the characteristics of head in term, the NP reordering pattern database is built to reorder the words sequence of source language. The terms those of which can not be reordered by NP reordering pattern are reordered using a unified approach.Secondly, during the analysis of the N candidates, we find that the best result of the candidates is not always the result of Moses choosed. So, the mutual information model is added to reorder the candidates. Concretely, the associations between the modifiers and the head of a translation result are calculated firstly. And then the results are rearranged by combining with the calculated result, language model of the translation and the scores of Moses. At last, the best translation is chosen.Combined with the tasks at the two aspects above, a term automatic translation system is constructed. The experimental results show that the method proposed by this thesis has good effect on the translation of noun term in the patent documents. And the translation results are better than that of other method of statistical machine translation at present.