Research Technology Of Translation In Assisted Tranlsation Based On User Model

At present, statistical machine translation has made great success. Even there are translation systems which can be used in some fields, but in respect of quality, these translation systems can not meet the use requirements, and it is increasingly difficult to improve the quality of translation. Most of the researchers make the focus on the computer assisted translation because assisted translation can integrate various factors into the process of human-computer interactions. However, it is very difficult to research how to mine and use translators’implicit knowledge in the process of assisted translation.By analyzing the process of modifying the assisted translated text into the final translated text by the translator, and combining with the application of the user model in some fields, this paper provides a qualitative quality assessment method based on the user model without reference translation, and applies the method to select the assisted translated text. Experimental results show that the method is more effective than the traditional method based on linguistic features, and simultaneously, the method has certain correlation with manual assessment on the selection of the translated text. The main work of this paper is as follows:First, by tracking and recording user behavior in the process of assisted translation, the knowledge in the process of translation is automatically learnt, and an incorrect instance template library is formed to be used as the user model in this paper. On this basis, a qualitative quality assessment method based on the user model without reference translation is provided. The method uses a machine learning method to construct a prediction model of the assisted translated text. Except the linguistic features (such as the information of words and part of speech, language model, etc.), user behavior features are also added into the model, and the assisted translated text is assessed from the view of the user model.Second, in the research of the assisted translated text generation technology based on the user model, the score is put into a reasonable range by scoring and mapping the assisted translated text so as to score the same translated text in multiple systems and select the best translated text, and the scoring results has certain correlation with manual assessment.