A linguistic taxonomy of human-AI interaction: discursive strategies in prompt engineering for large language models

Abstract

This article examines prompt engineering for large language models as a linguistically structured form of human-AI interaction rather than as a purely technical procedure. The study is based on the assumption that every prompt functions as a communicative act in which the user defines intention, context, role distribution, genre expectations, semantic limits, evaluative criteria, and ethical boundaries. Special attention is paid to the realities of modern Ukraine and the russian-Ukrainian war of aggression, communication has become a crucial instrument of resilience, verification, civic responsibility, and resistance to manipulation. By drawing an analogy between wartime information literacy and responsible AI use, the article argues that the quality of interaction with large language models depends not only on the technological capacity of the system, but also on the user\"s ability to formulate precise, contextually grounded, and ethically conscious prompts. The paper proposes a linguistic taxonomy of discursive strategies used in prompt engineering, including instruction, contextualisation, role framing, genre modelling, constraint setting, exemplification, evaluation, correction, and dialogic refinement. These strategies are analysed as mechanisms that help users reduce ambiguity, guide machine-generated responses, maintain control over meaning, and critically assess the reliability of the output. The study demonstrates that prompt engineering can be productively interpreted within applied linguistics, discourse analysis, pragmatics, genre theory, and academic communication. It also shows that prompt literacy is becoming an important component of digital competence, particularly in higher education, ESP instruction, translation, media literacy, and professional communication. The findings suggest that students and specialists should be trained not merely to use AI tools, but to interact with them as critical language users who understand how wording, context, constraints, and feedback influence generated texts. The article concludes that a linguistic taxonomy of human-AI interaction can support more responsible, transparent, and pedagogically meaningful use of large language models in academic and professional contexts.