Volkova Ya.A. Modeling emotivity in generated media texts

The article addresses the problem of modeling emotivity in generated media texts. The study objective is to identify which components of the emotive organization of an authorial text are reproduced by large language models and which undergo transformation in the process of generation. The material includes Grigory Tumanov’s journalistic text “Dreams Are Bad for Your Health” and three texts generated by ChatGPT, DeepSeek, and GigaChat using the same prompt. The study employs comparative linguostylistic analysis, focusing on linguistic markers of emotivity, the nature and dynamics of the emotional-evaluative position, the degree of explicitness of evaluation, imagery, and compositional organization. The results show that all three models are capable of reproducing complex emotive dynamics, although in different ways. Generated texts tend to make changes in the subject position more explicit, display greater logical organization, and reduce evaluative contradictions, whereas the authorial text preserves a higher degree of emotional and evaluative ambiguity.