Analysis of Machine Translation of Verbose Constructions

Machine Translation provides automatic translation of the source text from the original natural language to another, by preserving the basic idea of input and rephrasing it into the target language. Although the vast majority of types of MT systems use translation models that divide text into units, none of them has sufficient accuracy for all compositional, syntactic and semantic representations of word sequences in long phrases. This article describes step-by-step comparison of word-based and N-gram based models for translating word pairs, and suggests a language model that combines these methods to optimize the translation quality and the word sense disambiguation. Experimental results on the OPUS News-Commentary translation task for parallel Russian-English corpus demonstrate the benefits and drawbacks of already existing methods and show significant improvements in the translation and detection of dependencies for phrases in the proposed model.