Indicators of Texts Generated by Neural Networks in the Aspect of Authorship Examination (Based on Experimental Study)

Diagnostics of indicators of texts generated by neural networks in the aspect of forensic authorship is a relevant and complex task, the solution of which requires new skills from the expert. The article provides an overview of the recent publications in the field of identifying the indicators of generation of Russian-language texts at the current stage of development of science and technology. The article presents the results of a study conducted at the Forensic Center of the Ministry of Internal Affairs of Russia as part of an All-Russian training workshop among experts in the field of forensic authorship examination of forensic units of the Ministry of Internal Affairs of Russia. For the first time, using the material of colloquial texts within the law enforcement discourse, generated using freely available programs (GPT4, DeepSeek, Grok 2, Hailuo), through expert analysis, the signs indicating the use of these programs are summarized. They were formed on the basis of their reproducibility in the conclusions obtained by forensic authorship experts with a subsequent assessment of their relevance. The identified indicators expand the list proposed in existing scientific works. Attention is also paid to the relationship between the conditions for creating texts set by programs and the quality of implementation of each of the specified parameters in the generated text. The article outlines the feasibility of developing an interdisciplinary approach to expert analysis of such texts, including methods of computer and classical linguistics. An “expert experiment” traditionally used in the production of certain types of forensic examinations is proposed as a new method for authorship examination. In the context of intensive development of technologies for creating human-quality texts, the need for dynamic improvement of the scientific and methodological base and further research is outlined.