Multi-level OT: An argument from speech pathology

Since the early 1990s, Optimality Theory (OT) has quickly gained ground in phonology. Its main appeal lies in two characteristics: the focus on wellformedness of the output and the softness of constraints, where a constraint can be violated in order to satisfy more important requirements. A conspicuous characteristic of classic OT is that all constraints on the output should compete with each other at all times. This basically means that the ‘construction’ of the output occurs in one step. OT is therefore minimally derivational, the only derivation being that from the input to the output. Mainly on the basis of data from aphasic speakers, I argue that the ‘one-step’ variant of OT lacks psychological validity and that it is better to assume that the OT algorithm plays a role in determining (phonological) structure at different cognitive levels of processing, at least in production.