High-quality Mongolian-Chinese unsupervised neural machine translation method

Abstract

The invention provides a high-quality Mongolian-Chinese unsupervised neural machine translation method which comprises the steps that large-scale Mongolian-Chinese monolingual corpora are pre-segmented, Bert is used for carrying out monolingual language model pre-training on the segmented corpora to obtain Mongolian and Chinese language models, and an unsupervised Mongolian-Chinese word segmentation device is trained in combination with a correlation matrix generation method fusing sub-word segments; the correlation of any two sub-words in the to-be-segmented Mongolian-Chinese sentence is scored to complete word segmentation, the segmented Mongolian and Chinese are embedded into a shared potential space, Mongolian-Chinese bilingual word vector space is optimally aligned by using an unsupervised adversarial autonomous learning method, a Mongolian-Chinese language model is trained for the segmented Mongolian-Chinese monolingual corpus in the space, nearest neighbor search is performed byusing a CSLS method to obtain a Mongolian-Chinese bilingual dictionary based on a GAS framework, an initial Mongolian-Chinese translation model is trained in combination with the Mongolian-Chinese language model generated by a pre-training model, and a high-quality Mongolian-Chinese-Mongolian bidirectional dual unsupervised translation model is jointly trained by using an unsupervised back-translation method in combination with a dual learning strategy.

Description

technical field [0001] The invention belongs to the technical field of neural machine translation, and in particular relates to a high-quality Mongolian-Chinese unsupervised neural machine translation method. Background technique [0002] Machine translation has flourished in recent years, and machine translation tasks for rare resource languages ​​and minority languages ​​have also gained more attention. Mongolian is a language that is widely used across multiple countries and regions. The Mongolian language is the official language of the Inner Mongolia Autonomous Region. On the one hand, the research on Mongolian-Chinese machine translation is of great significance to the promotion of national culture dissemination and the communication of multi-ethnic people. On the other hand, it plays a positive role in promoting the development of rare resources and minority language machine translation research. However, due to the relatively late start of research on Mongolian natu...

AI Technical Summary

Problems solved by technology

[0003] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide a high-quality Mongolian-Chinese unsupervised neural machine translation method, and apply high-quality unsupervised learning strategies to Mongolian-Chinese neural machine translation to make full use of the existing large number of Unlabeled monolingual data to alleviate the lack of Mongolian-Chinese parallel corpus resources

Images