This paper is a critique of two foundational assumptions of generative work culminating in the Minimalist Program: the assumption that, as a matter of conceptual necessity, language has a ‘double-interface property’ and the related assumption that phonology has a realizational function with respect to syntax-semantics. The issues are broached through a critique of Holmberg's (2000) analysis of Stylistic Fronting in Icelandic. We show that, although empirically motivated, and although based on the double-interface assumption, this analysis is incompatible with that assumption and with the notion of (phonological) realization. Independently of Stylistic Fronting, we argue that the double-interface assumption is a problematic legacy of Saussure's conception of the linguistic sign and that, conceptually, it is neither explanatory nor necessary. The Representational Hypothesis (e.g. Burton-Roberts 2000) develops a Peircian conception of the relation between sound and meaning that breaks with the Saussurian tradition, though in a way consistent with minimalist goals. Other superficially similar approaches (Lexeme–Morpheme Base Morphology, Distributed Morphology, Jackendoff's Parallel Architecture) are discussed; it is argued that they, too, perpetuate aspects of Saussurian thought.