Lexical and syntactic features of academic Russian texts: a discriminant analysis

Abstract

This article presents three mathematical models to differentiate academic texts from three subject discourses written in Russian (i.e., Philological, Mathematical, and Natural Sciences) which further enable design and automated profiling of corresponding typologies. Our models include 5 indices, one at surface level (i.e., sentence length) and 4 syntax features (i.e., mean verbs per sentence, mean adjectives per sentence, local noun overlap, and global argument overlap). We identified and validated the five statistically significant features out of 45 linguistic features extracted from our research corpus consisting of 91.185 tokens. The shortest sentence length is found in Russian language textbooks while the longest sentences are identified in Natural Science texts. The mean number of verbs, nouns, and adjectives per sentence is higher in Natural Science textbooks, whereas Mathematics discourse is characterized by the shortest word length, highest local noun overlap, and highest global argument overlap. We assign the metric differences between the three discourses to their functions: Natural Science texts are characterized by descriptions and narrative passages in contrast to Philology that is associated with opinions. Mathematical discourse operates with precise definitions, explanations and justifications thus exercising numerous overlaps. The discriminant analysis built on top of the features supports the development of text profilers targeting parametric analyses. The automation of these features and the provided formulas for classification enable the design and development of text profilers required for textbook writing and editing. Our findings are useful for professional linguists, technologists, and academic writers to select and modify texts for their target audience.