CAN LARGE LANGUAGE MODELS UNDERSTAND PRESUPPOSITION? A SEMANTIC–PRAGMATIC INTERFACE ANALYSIS

Abstract

Presupposition constitutes one of the most complex categories at the intersection of semantics and pragmatics, as it encodes implicit meaning that is linguistically triggered but contextually validated. With the increasing use of large language models (LLMs) in academic and communicative settings, questions concerning their capacity for linguistic understanding have intensified. This paper examines whether LLMs can be said to understand presupposition or whether they merely reproduce presuppositional patterns without access to the underlying semantic–pragmatic mechanisms. Adopting a semantic–pragmatic interface approach, the study analyzes presupposition as a linguistically encoded phenomenon whose interpretation depends on contextual assumptions, discourse continuity, and shared background knowledge. Drawing on theoretical linguistics and comparative insights from English and Uzbek, the analysis evaluates LLM-generated texts with respect to presupposition triggering, projection, accommodation, and failure. The findings indicate that while LLMs successfully generate structures containing presuppositional triggers, they systematically fail to manage presuppositions across contextual shifts and culturally embedded assumptions. This demonstrates that LLMs operate at the level of formal pattern simulation rather than genuine presuppositional understanding. The study contributes to AI linguistics by clarifying the limits of machine-generated meaning from a strictly linguistic perspective.