Machine Translation Methods for Creating a Corpus of Simplified Texts in Russian

Today natural language processing is rapidly increasing in popularity. Neural networks solve many routine tasks that were previously performed manually. Machine translation is one of the most promising areas in which significant progress has already been achieved. Machine translation methods can also be successfully applied to the problem of sentence simplification. The relevance of this research is based on the fact that sentence simplification for Russian has not been sufficiently investigated. This work aims to study the role of machine translation in training sentence simplification models and overcoming the problem of scarce data. To fulfill the purpose, the English parallel corpus WikiLarge was automatically translated into Russian and used to train a neural machine translation model to adapt it to sentence simplification for Russian. The fine-tuned model achieved good quality and showed that the translated data can be applied for training. Its combination with the corpus of paraphrases proved to be especially effective. This is an incentive for further studies of the sentence simplification problem for Russian, including the application of existing improvements and the development of new methods. The translated corpus WikiLarge can be used to study the simplification problem for the Russian in general, as well as to teach and evaluate new, more complex models in the future.