Applying Multi-Dimensional Analysis to a Russian Webcorpus: Searching for Evidence of Genres

The paper presents an application of Multidimensional (MD) analysis initially developed for the analysis of register variation in English (Biber, 1988) to the investigation of a genre diverse corpus, which was built from modern texts of the Russian Web. The analysis is based on the idea that each linguistic feature has different frequencies in different registers, and statistically stable co-occurrence of linguistic features across texts can be used for automatic identification of texts with similar communicative functions. By using a software tool which counts a set of linguistic features in texts in Russian and by performing factor analysis in R, we identified six dimensions of variation. These dimensions show significant similarities with Biber's original dimensions of variation. We studied the distribution of texts in the space of the dimensions of our factors and investigated their link to 17 externally defined Functional Text Dimensions (Forsyth and Sharoff, 2014), which were assigned to each text of the corpus by a group of annotators. The results show that dimensions of linguistic feature variation can be used for better understanding of the genre structure of the Russian Web.