The subject of the study is the peculiarities of the use of hedging in English-language texts generated by neural networks. Hedging refers to language forms that allow for the expression of uncertainty or assumption. The author analyzes how often such forms are used, which of them are predominant, and what role they play in the text. Special attention is paid to comparing neural network texts with authentic academic English texts to determine whether they differ in the expression of uncertainty and how closely such texts resemble real scientific discourse. Additionally, the contexts in which hedging is most frequently used are examined, and whether its use is related to attempts to reproduce a scientific style is explored. The relevance of this research is determined by the need for a systematic description of the pragmatic features of English-language discourse generated by neural networks, as well as the insufficient degree of development of this issue in modern linguistics. Despite the growing number of works dedicated to the analysis of texts generated by artificial intelligence, the aspect of hedging as a separate pragmatic mechanism remains insufficiently studied. The study employs corpus and comparative methods of analysis. The material for the research consists of texts generated by a neural network model, compared with data from a contemporary English language corpus. Within the framework of the research, a quantitative analysis of the frequency of hedging usage was conducted, its distribution in the text, and functions in various communicative contexts were studied, which allowed for the identification of the main patterns of its use. The scientific novelty of the research lies in the identification of the peculiarities of hedging usage in texts generated by neural network models as an independent type of discursive practice. It was established that in such texts, hedging is used more frequently but is characterized by less variability compared to authentic English academic discourse. The forms expressing assumption are most actively used, while other ways of expressing uncertainty are represented more limitedly. A tendency towards the repetition of the same constructions, as well as their excessive combination within a single statement, was identified. It was shown that, unlike scientific texts, where hedging is related to argumentation and data interpretation, in neural network texts, it often serves a formal function. The results obtained allow for viewing neural network discourse as more standardized and predictable.
Maria Pavlovna Kolesnikova (Fri,) studied this question.