A Mongolian-Chinese neural machine translation method based on a triangular architecture

Abstract

In order to change the status quo that machine translation of low-resource languages is relatively backward, the invention discloses a Mongolian-Chinese neural machine translation method based on a triangular structure. Compared with the existing pure end-to-end neural machine translation method, the method of the invention fully considers the problem of limited parallel corpus in small languages,especially the problem of scarcity of Mongolian-Chinese parallel corpus, and improves the quality of Mongolian-Chinese translation under the premise of scarcity of parallel corpus. Secondly, a unified bi-directional EM algorithm is used to jointly optimize the Mongolian translation model. Finally, the pseudo-samples generated from the model x to z or z to y and the real bilingual samples are mixed in the same small batch at a ratio of 1: 1 to stabilize the training process.

Description

technical field [0001] The invention belongs to the technical field of machine translation, and in particular relates to a Mongolian-Chinese neural machine translation method based on a triangular framework. Background technique [0002] Machine translation, which can use computers to automatically translate one language into another, is one of the most powerful means to solve language barriers. In recent years, many large search companies and service centers such as Google and Baidu have conducted large-scale research on machine translation, making important contributions to obtaining high-quality translations of machine translation, so the translation between major languages ​​is close to that of human beings. At the translation level, millions of people communicate across language barriers using online translation systems and mobile apps. In the wave of deep learning in recent years, machine translation has become the top priority and has become an important part of prom...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the object of the present invention is to provide a Mongolian-Chinese neural machine translation method based on a triangular framework. This method is mainly aimed at the problem of limited parallel corpora in small languages, especially the scarcity of Mongolian-Chinese parallel corpora. Mongolian (z) is introduced into the translation between English (x) and Chinese (y) as an intermediate hidden variable, and the translation between English and Chinese is decomposed into two steps via Mongolian

Images