Text complexity and linguistic features: Their correlation in English and Russian

Text complexity assessment is a challenging task requiring various linguistic aspects to be taken into consideration. The level of the text should correspond reader’s competence. A too complicated could incomprehensible, whereas simple one boring. For many years, features were used assess readability, e.g. average length words and sentences or vocabulary variety. Thanks development natural language processing methods, set parameters for evaluating readability has expanded significantly. In recent articles have been published authors which investigated contribution lexical, morphological, syntactic level. Nevertheless, as methods corpora are quite diverse, it may hard draw general conclusions effectiveness information due diversity corpora. Moreover, cross-lingual impact different on datasets not investigated. purpose this study conduct large-scale comparison nature. We experimentally assessed seven commonly feature types (readability, traditional features, morphological punctuation, syntax frequency, topic modeling) six in English Russian employing four common machine learning models: logistic regression, random forest, convolutional neural network feedforward network. One corpora, corpus fiction literature read by school students, was constructed experiment using survey ensure objectivity labeling. showed can significantly improve performance analyzed their according dataset characteristics, language, data source.