Research On Mongolian-Chinese Machine Translation Based On End To End Neural Network

With the development of Natural Language processing and Machine Translation technology,the use of Artificial Intelligence technology to improve the efficiency and accuracy of translation between different languages has gradually become a research hotspot.Among them,the E2E model structure of Machine Translation neural network has been the focus of research due to its high translation accuracy and strong semantic meaning for translation.However,there are still some problems such as limited vocabulary and low loyalty of translation.In order to solve the above problems,this paper does research focusing on the segmentation and corpus annotation in Machine Translation preprocessing,to alleviate the corpus and vocabulary constraint.At the same time the construction process of the E2E model will also be studied to improve the quality of translation,In the preprocessing stage of the E2E Machine Translation model,we study the segmentation algorithm of corpus,and give a segmentation model based on HMM to mark the sequence segmentation.At the same time,we use discriminant method and CRF model for Mongolian stems and affixes segmentation and tagging.For the expression of corpus,a low dimensional distributed expression is applied to deal with the words vectorization of Mongolian-Chinese bilingual corpus,so as to adapt to the input and output of E2E model.For the translation loyalty issues in Mongolian-Chinese Machine Translation,it combines with the mainstream Machine Translation nerve E2E frame to construct a coding and decoding model based on the combination of CNN and GRU.CNN building encoder takes advantages of overlapping layers characteristics of convolution and parallel computing to acquire long-time information of the source language sentence,and expresses the semantics in vector mode to realize the coding of the model,which can speed up the encoding speed and quality.According to the encoder situation,it uses GRU to do the target language decoding.In bilingual word alignment process,it combines with a global attention model to obtain bilingual word alignment information,then predicts and outputs target language.After encoding and decoding,it achieves mapping on target language from source language.In the end,this paper does translation experiments on Mongolian-Chinese Machine Translation system on the basis of E2E,compares and analyzes the experimental results,evaluates the quality of the translation in model by using the BLEU value evaluation index.The experimental results show that Mongolian-Chinese Neural Machine Translation model based on E2E is much better than the traditional statistical methods and the Machine Translation Benchmark model based on recurrent neural network in terms of the quality of translation and the degree of semantic confusion.