Artificial Intelligence in media discourse: the classification of linguistic and pragmatic features of the generated text

The purpose of this scientific article is to study the distinctive features of information and analytical texts generated by a neural network in a special issue of the daily business newspaper RBC (РБК) (the edition is claimed to be partially generated using GigaChat and Kandinsky neural networks) and to put these features into a classification. Procedure and methods. A comparative method was used in the work (the author of the article attempted to identify and describe not only the common features (markers) of the generated text, but also the features of generation in various types and genres of media discourse, the difference between linguistic and pragmatic features in different languages is also described). Among others, a quantitative method was used (content analysis of specially selected text units – features of textual neural network generation). According to the results of the study, the previously identified classification of linguistic and pragmatic features of generated texts in media discourse was confirmed; statistical data were provided illustrating the presence of certain features of generation in the text (160) and proving their artificial origin: punctuation errors (10); morphological errors (7); syntactic errors (7); lexical errors (8); lexical repetition (13); spelling errors (7); graphic errors (8); logical errors (7); factual errors (11); clichés (71); redundancy (6); template structure (6). The text of the article provides an example of the analysis of one of the data analytics articles of the special issue of RBC (РБК), containing the largest number of linguistic and pragmatic features of neural network generation (55 items). Theoretical significance of the research results lies in the necessity to isolate a new separate area of linguistic knowledge – media linguistics of texts generated by a neural network. The practical significance contains the idea that the classification of linguistic and pragmatic features of a neural network media text identified by the author can be used to examine content for its artificial origin and intellectual fraud.