COOPERATION WITHOUT A PARTNER: THE MIGRATION OF PRAGMATIC LABOUR IN USER–AI DISCOURSE

Abstract

This article describes the communicative situation created by the spread of conversational interfaces built on large language models, a situation for which linguistic theory was not designed: users mitigate directives, thank the system and apologise for their own imprecision, although a language model holds no beliefs, assumes no obligations and possesses no face requiring protection. Pragmatic description of this behaviour is entirely successful, yet the frameworks themselves presuppose communicative intention on both sides. The article examines what becomes of pragmatic competence once that presupposition is withdrawn. It argues that established frameworks retain their descriptive adequacy while surrendering their reciprocity: they no longer model an exchange between interlocutors but the sustained pragmatic work of a single participant, directed at an addressee that can neither endorse nor return it. The theoretical groundwork is laid by J. Austin, J. Searle, H. P. Grice, S. Levinson, D. Sperber and D. Wilson, G. Leech, P. Brown and J. Habermas; the survival of social norms in machine-directed speech is explained by B. Reeves, C. Nass and Y. Moon, A. Gambino, J. Fox and R. Ratan; empirical refinement comes from L. Panfili, R. Bowman, Z. Yin and E. Miehling; the applied dimension from J. White and J. D. Zamfirescu-Pereira, the metapragmatic from C. Caffi and J. Verschueren, the sociolinguistic from P. Joshi, A. Radford, L. Bilaniuk and S. Sokolova. The study employs critical analysis and synthesis of sources, conceptual analysis of pragmatic categories with respect to their reciprocity requirement, comparison of the status of the maxims in human dialogue and in user–AI discourse, and the modelling of a two-level account of the artificial addressee. Established frameworks are shown to retain their descriptive adequacy while surrendering their reciprocity: interpretive labour migrates to the user, and the Gricean maxims are re-encoded from expectations into design specifications within the prompt. The contradiction between automatic and strategic politeness is resolved through a distinction between an agent model and a resource model of the system, and the term operational politeness is proposed. Assistant-directed code-switching is reclassified as compensation for asymmetric channel availability, so that speakers of thinly resourced languages carry an inferential and a channel burden at once. Prospects for further study lie in eliciting a corpus of machine-directed speech from Ukrainian users, testing the distinction between inferential and channel burden against languages of comparable resourcing status, and reconceiving prompt design within AI-literacy curricula as a linguistic rather than a technical skill.