Neural Machine Translation Method for Low-resource Scenarios
Neural machine translation requires a large-scale bilingual parallel corpus to build a translation model using deep learning methods.In low-resource scenarios,unsupervised neural machine translation is usually applied due to lack of large-scale bilingual parallel data.This paper proposes a method of fusing syntactic knowledge in unsuper-vised neural machine translation,so that the model can fully learn the syntactic information of sentences.At the same time,a small amount of bilingual parallel corpus is introduced to assist unsupervised neural machine translation training,so that the model can directly learn the mapping between source language and target language words.Compared with the baseline system,the proposed method imporves 1.65 to 1.79 BLEU score on the English-French and German-English tasks,respectively.
万方数据
维普
