基于多语言预训练的越南语机器翻译优化技术

Vietnamese Machine Translation Optimization Technology Based on Multilingual Pre-training

传统越南语机器翻译系统只在源语言和目标语言之间构建一个独立的双语翻译模型,无法学习到双语之外的其他信息。通过引入多语言机器翻译模型,并得益于富资源语言知识的正向迁移,越南语机器翻译可以从其他富资源语言方向中获取到丰富的知识信息,翻译质量得到显著的提升。在此基础上,采用双语调优、知识蒸馏、模型压缩等方法进行调优,可以使越南语的翻译效果得到进一步提升。

Traditional Vietnamese machine translation systems only build an independent bilingual translation model between the source language and the target language,and cannot learn information in other languages.This paper introduces a multilingual machine translation model.Due to the posi-tive transfer of resource rich language knowledge,Vietnamese machine translation can obtain rich knowledge information from other resource rich languages,and the translation quality has been sig-nificantly improved.On this basis,the translation effect of Vietnamese can be further improved by bilingual optimization,knowledge distillation,model compression and other methods.