RUDN Journal of Language Studies, Semiotics and Semantics

-

Background. Internet discourse, traditionally studied within the paradigm of computer-mediated communication as a hybrid formation combining features of oral and written speech, is entering a qualitatively new phase of its development. The public launch of large language models (GPT-4 and equivalents) in 2022-2023 marks an algorithmic turn: generative artificial intelligence ceases to be a neutral channel for message transmission and transforms into an active participant in communication, capable of independently producing content, engaging in dialogue, and influencing the linguistic practices of millions of users. However, linguistic theory still operates within a conceptual apparatus oriented toward an anthropocentric model of discourse, which does not allow us to adequately describe the new communicative reality in which the boundaries between author, tool, and message are becoming blurred. Objective. To identify and systematize the fundamental transformations of internet discourse in the era of generative artificial intelligence, proposing a theoretical framework for their analysis within the context of digital anthropolinguistics. Materials and Methods. The study employs an interdisciplinary design combining methods of corpus linguistics and critical discourse analysis. The material comprises three original contrasting corpora with a total volume of 1.5 million word usages in Russian and English: (1) a corpus of human-to-human communication (H2H), representing the “pre-algorithmic” era (up to 2021); (2) a corpus of prompts (H2A), capturing the new genre reality of human-algorithm communication; and (3) a corpus of AI-generated texts (A2H), reflecting the specificity of algorithmic discourse. Results. Quantitative analysis, for the first time on representative material, confirmed the hypothesis of discourse homogenization: the A2H corpus demonstrates a 37.6% reduction in lexical diversity and a 45.3% reduction in syntactic diversity compared to the human communication corpus. The genre specificity of the prompt as a new type of discursive practice has been identified - a directive-exploratory genre characterized by imperativeness, query parameterization, and the user’s metalinguistic reflection. The phenomenon of the algorithmic norm has been discovered - the stable reproduction of clichéd bookish register constructions (“it is important to note,” “thus”), forming an averaged, stylistically homogenized register. Hybrid discursive acts have been recorded and described for the first time - communicative units arising from the iterative interaction between human and algorithm, the authorship of which cannot be unambiguously attributed to either participant. Conclusion. In response to the challenges of the new era of internet discourse, theoretical concepts are introduced that allow for the description of the transformed communicative reality: the parameter “Degree of Algorithmic Mediation” (DAM) and the concept of the “hybrid discursive act.” A comparative analysis of the Russian and English parts of the corpora demonstrates the universality of the identified transformations, suggesting not local linguistic changes but a civilizational shift - an algorithmic turn that requires a reconsideration of the fundamental categories of linguistics (authorship, text, discourse, linguistic persona).

Prompt engineering, a phenomenon arising from human interaction with large language models (LLMs), is the subject of this linguistic analysis. The study argues that a prompt represents a qualitatively new type of controlling text constructed in natural language. The research aims to demonstrate that the analysis of prompts requires the full range of modern linguistic tools: semantics, pragmatics, discourse analysis, cognitive linguistics, and the philosophy of language. The article examines the semantic features of prompts, highlighting the need to handle fuzzy categories and manage reference given the model's lack of human experience. Pragmatic analysis reveals the specific nature of the prompt as a directive speech act, in which felicity conditions must be explicitly verbalized. Within a discursive approach, it is demonstrated how a prompt establishes the macrostructure of the future text and constructs professional roles and social registers. Special attention is paid to the cognitive foundations of prompt engineering, where the formulation of a request serves as an external metacognitive practice that structures human thinking. The article also addresses cross-linguistic aspects related to the dominance of English-language corpora and the specifics of working with the Russian language. It concludes that prompt engineering marks a profound transformation of linguistic activity and necessitates the formation of a new interdisciplinary field: the linguistics of human-model communication.

This article examines media transposition involving generative neural networks and the specifics of transferring verbal content into visual form. The relevance of the research is determined by the trends in using “artificial intelligence” to create an infosphere in modern communication and to transmit cultural memory, involving its remediation and reinterpretation. The aim of this article is to examine the characteristics of transferring printed poetic text into a media format using generative neural networks, to identify specific techniques, and to outline the features of the original text’s remediation that occurs in this process. The research material consists of works (6 in total) by neuro-illustrator Nikita Glukhov (pseudonym DJ БлокNote). Generative neural networks, guided by a human operator and trained on a dataset of human-generated content, represent a mediated form of social interaction. In this regard, this study implements an integrative approach and examines the stated phenomena from the perspectives of interactionism, hermeneutics, psychological and visual anthropology, and linguosemiotics. The main research methods are qualitative: interpretation, analysis, synthesis, descriptive method, functional-pragmatic and contextual analysis, and elements of psychoanalytic method for extracting hidden meanings. A three-level analysis of semiotically complex text, developed by R. Barthes, and adapted to the needs of the research, is applied. The following was discovered. During the transposition of printed text into a screen medium with the involvement of generative neural networks, its textual structure is recoded into visual and auditory matter. The constructed work consists of intermedial images, while dynamics, montage, composition, color, and light replace rhythm and metaphoricity. In addition to the audio-visual, a plot-narrative expressive layer is also identified - the narrativization of the lyrical. Prompts for the generative model are created by the human operator based on key semantic points of the poem - specific lexical units (LUs), with the addition of extra LUs that are absent in the original text but necessary for its visualization. Following the prompt, the neural network materializes the poem’s context, “frames” the symbols embedded within it, and “translates” emotional expression into a dynamic whole. This article is aimed at students, graduate students, teachers, linguists, philologists, and anyone interested in general and linguistic semiotics.