Data Augmentation for Low resource Neural Machine Translation Based on Se mantic Related Word Replacement and Grammatical Error Correction

Firstly, the low-resource language is data-enhanced through the method of semantically related word replacement; secondly, the data-enhanced bilingual parallel corpus is grammatically corrected to make it conform to linguistic syntax and common sense reasoning.

The results show that the method proposed in this paper not only guarantees the quantity of training corpus, but also improves the quality of training corpus, realizes effective data enhancement for low-resource languages, and further improves the effect of neural machine translation for low-resource languages.