The Research on Mongolian and Chinese Machine Translation Based on CNN Numerals Analysis

Abstract

With the progress of science and technology and the development of artificial intelligence, the Machine Translation method based on neural network replaces statistical method better because of its translation, especially in the aspect of inter-translation among the major languages in the world. Recurrent neural network can extract more features when encoding the source language, which is vital to the quality of translation. In the aspect of translating Mongolian, it is difficult to obtain semantic relations sufficiently from the corpus due to lacking corpus. Therefore, a method of Mongolian-Chinese machine translation based on Convolutional Neural Network (CNN) is proposed. Analysis of Mongolian numerals is to improve the encoder and then selection out of vocabulary. In the process of encoding source language, through the pooling layer, the semantic relation and the series of key information of convolution neural network in the sentence can be obtained. Then, through the Gated Recurrent Unit adds to the global attention mechanism, the source language after encoding can be decoded into Chinese. The experimental result shows that the method takes advantage of the recurrent neural network (RNN) in the aspect of the accuracy and training speed of the translation.