Research On Template-Based Machine Translation Technology In Cooperative Environment

With the rapid increase of net resource and the quick promotion of computer operating ability, machine translation has made great success. Although there are translation systems which can be used in some fields, due to the corpus size and some linguistic constraints, in other fields, these translation systems can not meet the practical requirements. Computer aided translation (CAT) has been a hot research topic. In the frame of CAT, human and machine assist each other in producing the correct translation together. Recently, with the rapid development of CAT, it is common that a number of individual translators be organized to accomplish a large-scale translation task. This translation mode can be called cooperative translation (CT).Compared with computer-aided translation, cooperative translation is characterized by the large user group. Different user has different education background, different level of translation, and different translation experience, therefore the users’translation habits and psychologies are different. Existing cooperative translation systems can not fully mine the translation habits and psychologies of various users, so they are not able to recommend the corresponding machine translation result according to user’s preferences for the translation.To solve the above problems, this paper made research on the template-based machine translation technology in cooperative environment, including individual template recommendation, word sense selection, and automatic post-editing on machine translation. The main work of this paper is as follows:Firstly, the paper proposed an individual template recommendation method based on relative entropy. If there were many translation templates matching the source sentence, the relative entropy based method is adopted to recommend the optimum template for the current user; to meet the user’s individual needs.Secondly, the paper proposed a word sense selection method based on user’s translation history. After determining the translation template, the variations in the template needed to be translated. If one word or phrase has many meanings, the paper determines which one should be used through counting the similarity between the current sentence and all the translation history which contain the word or the phrase.Thirdly, the paper proposed an automatic post-editing method based on user model. With the machine translation as the source language, and the correct translation which the user had revised as the target language, the Moses system is used to train statistical translation models. Before the machine translation result is recommended as aiding translation, the user’s own model is adopted to translate it into a translation which is closer to the correct translation. It can further reduce the users’post-editing time and improve the efficiency of cooperative translation.