Abstract
Human communication is typically modeled as semantically grounded, with pragmatics supplying context-sensitive modifications. This paper reverses that order. I develop an inferential–compression account on which speakers intentionally underdetermine their signals, and listeners recover meaning through relevance-guided inference, contextual integration, and recognition of public intentions. Pragmatics is therefore the primary cognitive and evolutionary engine of communication, while semantics is a later cultural mechanism for stabilizing interpretive outcomes. Communicative variation supports this architecture: autistic communication exemplifies a semantic-dominant, low-compression profile that minimizes inferential load, while narcissistic hyper-pragmatics illustrates high-compression strategies that shift interpretive work onto the listener; neurotypical profiles form a stable midpoint. These patterns reveal compression as a core dimension of communicative cognition and supply a mechanistic explanation for the emergence of systemizing phenotypes. The model unifies research across pragmatics, cognitive science, and evolutionary anthropology and reframes meaning as coordinated inference rather than semantic encoding.