Pragmatics beyond humans: meaning, communication, and LLMs

Abstract

The paper argues that the emergence of large language models (LLMs) requires a conceptual and methodological reconsideration of pragmatics. It first challenges the traditional positioning of pragmatics within the semiotic trichotomy of syntax, semantics, and pragmatics, suggesting that the Human–Machine Communication (HMC) framework provides a more suitable perspective for analysing meaning-making in hybrid (human–LLM) interactions. The paper then identifies three interrelated tensions that characterise contemporary pragmatic research on LLMs. First, it argues that prevailing approaches remain grounded in anthropocentric pragmatic theories and proposes probabilistic pragmatics (particularly the Rational Speech Act framework) as a more appropriate account of the connectionist nature of LLMs. Second, it introduces the notion of substitutionalism to describe methodological practices that conflate specific models with AI in general, particular languages with language as such, and (most importantly) replace human communicative participants with LLMs while neglecting changes in human communicative behaviour. Third, it examines the divergence between technical and pragmatic conceptions of context, arguing that the simultaneous expansion of computational context and the collapse of shared communicative context give rise to (what I call) context frustration.