Usage of Morphotranslation in Data Augmentation of the Training Dataset for “TurkLang-7” Project on MT Systems Creation

Abstract

The paper discusses an approach to solving the problem of low-resource languages in the development of neural network machine translation systems for Turkic language pairs by artificially increasing training data in form of parallel corpora. The basis of this approach lies in usage of toolset of the multilingual portal “Turkic Morpheme” for morphological analysis and generation of word forms. The issues of creating machine translation systems for Turkic languages are considered, a hypothesis about possibility of using the presented approach to improve the results of machine learning is formulated, and the rationale for using morphological analysis methods is given. The developed algorithm for artificial augmentation of training data is presented.