Surface Constraints versus Underlying Representation

The shift from derivational rules to representational constraints that characterizes much contemporary work in phonology calls for a serious reevaluation of the notion of Underlying Representation (UR). The reason is that the ability of constraints to apply "in parallel" systematically obviates the need for a derivation. The purpose of this paper is to argue, consistently with my (1994) Principles of English Stress (PES), that the notion of UR is neither conceptually necessary nor empirically supported, and should be dispensed with. The paper is structured as follows. In the next two sections I consider the two main motivations for postulating a UR: calculation of regularities of sound structure such as prosody, and calculation of allomorphy. I argue that neither calculation makes UR necessary in a constraint-based framework, and outline a fully "parallel" account of both. I then turn to the empirical evidence supporting the proposed account of allomorphy, an account which links surface forms directly to other surface forms, without UR. I argue in section 4 that the phenomena traditionally attributed to the "cycle" are just this sort of evidence, since they follow directly from our approach, while the need for the "cycle" is an unexpected complication for the derivational alternative. In sections 5-8 I then discuss further evidence that also follows from surface-to-surface links, but fails to reduce to any "cyclic," or even derivational, account, and conclude in section 9.