Chinese-Mongolian Machine Translation Combining Sentence Structure Information
Aijun Zhang, Weijian Hu, Mingxing Luo, Lingfang Li · 2023
Neural machine translation have achieved state-of-the-art accuracy on widespread used language pairs, but falls short on low-resource or complex-morphological language pairs. Mongolian is a typical agglutinative language which of complex morphological changes lead to the misunderstand of machine translation. In this work, we improve the quality of Chinese-Mongolia translation by combining double syntactic structure information. Convolutional neural network follows with interest is clause structure in the sequence, and extract the clause features through window sliding. The dependency structure of sentences affects the distribution of attention between words, we construct the dependency score to upgrade the attention mechanism for deeper encoding. We show the efficacy of our model not only in Chinese-Mongolian but also English-German, and the results prove that the fusion of double syntactic structure information helps machine translation.