Cross-Lingual Argumentation Mining for Russian Texts

Argumentation mining refers to automatic extraction of arguments and their relations from texts. This field has been evolving rapidly in recent years, but there is almost no research for the Russian language. The present study is an attempt to overcome this gap. Firstly, we create the first argument-annotated corpus of Russian based on Argumentative Microtext Corpus and make it publicly available. Secondly, we study the importance of various feature types. Contextual and lexical features turn out to be the most significant. Thirdly, we evaluate the performance of various classifiers for argumentation mining. Bagging and XGBoost classifiers give the best results. Fourthly, we assess the possibility of using several machine translation systems (Google Translate, Yandex.Translate and Promt) for automatic creating of argument-annotated corpora. Google Translate appears to be the best system to reach this goal.