Combination of Traditional Methods for Ukrainian-Vietnamese Deep Translation Models
摘要
The conflict in Ukraine attracts attention from us. For many of us, it is the first time we hear about the country in Eastern Europe. As a way to promote understanding of the Ukrainian people, Ukrainian culture we have done a survey on the machine translation from Ukrainian to our native Vietnamese. We are surprised to find that there has not been any publication on this topic yet. As for our initial goal, as well as to enrich the literature on the machine translation field, we propose the first Ukrainian → Vietnamese machine translation model. Our model is derived from the state-of-the-art Transformer deep model. Moreover, the proposed model inherits the model with source-word decomposition and the model guided by alignment, which have proved their effectiveness for translation from other Slavic languages such as Russian. The composite model surpasses the base models by 0.71 and 4.12 BLEU, respectively. The proposal is further confirmed by a limited semantic analysis of the translation results. The analysis shows that the proposed model constructs more fluent, natural Vietnamese translations with established collocations, standard sentence structures. Since all Slavic languages have much in common in terms of grammar as well as vocabulary, we recommend our model for the construction of translation systems from a Slavic language of all branches, including East Slavic branch (Russian, Belarusian) or West Slavic branch (Polish, Czech) and South Slavic branch (Bulgarian, Serbian, and Croatian).