Chinese-English Tense Translation Using Tree Conditional Random Fields

Human society has entered the information age, cross-regional exchanges and cooperation have been strengthened, but the language barrier is becoming obstacles to communication, machine translation research has important social value and economic value. With the growth of China’s overall national strength, Chinese is being used more and more frequent in the international community, and English is the most popular international language, so Chinese-English machine translation research is particularly important.In fact, our country started machine translation studies between Chinese to other languages very early, there have been practical English-Chinese Machine Translation System, but due to the uniqueness of the Chinese language, the Chinese-English machine translation research work has been slow. There is a certain distance between existing Chinese-English Machine Translation System from the practical level. English has a strong regularity of organizational form, English shows its’tense form auxiliary verbs and the morphology of verbs clearly. Chinese is a kind of hieroglyph, and expresses the tense of verb by lexical semantic, this have got the Chinese-English machine translation tense processing in difficulties. The Chinese-English machine translation must dealing with the conversion of tense in Chinese and English, otherwise it will affect the accuracy and the reading fluency of translation.This paper starts from the essence of Chinese and English, and analyzes the Chinese expression method and the specific factors in expressing verb tense. In this paper, by constructing Chinese sentence tenses tree, the issue of Chinese-English tense translation is changed into the issue of tagging a tense tree. And then we use tree-CRF to tag nodes of the incomplete tense tree with English tenses.The tree-CRFs model takes the hierarchy relationship between nodes into consideration, and we can get an optimal tagging from a global dimension, it alleviates the long-distance dependency problems of verbs tense. Templates of feature functions are suitable for the need of model inference, and experimental results show that the method of tree-based tense translation is much better than linear-based tense translation, and show that tense trees can better express tense dependencies between clauses.