The OCP in the perception grammar

An important concept in Optimality Theory (OT) is input-output faithfulness, the phonological similarity between underlying forms and surface forms.1 McCarthy & Prince (1995) distinguish three main types of families of input-output faithfulness constraints: MAX-IO, which states that an element (e.g. segment or feature) that is present in the underlying form should have a corresponding element in the surface form; DEP-IO, which states that an element present in the surface form should have a correspondent in the underlying form; and IDENT-IO, which states that if an element in the underlying form and an element in the surface form correspond to each other, then these two elements should be identical. A remarkable (though inconspicuous) property of these constraint families is their symmetry: MAX-IO and DEP-IO are one another’s mirror images, and the formulation of IDENT-IO requires the commensurability of underlying form and surface form, that is, in order that the underlying and the surface element can be assessed as identical or not, they should have a common basis for comparison, which means that they should be expressed in the same kind of units. If the underlying form and the surface form shall be expressed in comparable units, the question arises what the nature of these units is. I will share here the common view that the underlying phonological form of an utterance is basically an aggregate of lexical forms, augmented with information about morphological boundaries. These lexical forms are economical representations, which means that they are encoded in abstract features without much redundant information, disregarding most of what can be added by rule, such as positional variants, syllabification, or metrical structure. This lexical economy has been the one of the drives after proposals for contrastive underspecification of phonological features (Chomsky & Halle 1968, Steriade 1987), and in its extreme form of lexical minimality, it has inspired proposals for privative features (Trubetzkoy 1939), feature-tree underspecification (Jakobson, Cherry & Halle 1953), and radical underspecification (Archangeli 1984, 1988). But while there has been a heated debate about its extent, the consensus was that some sort of economy existed, implying at least, for instance, that predictable positional variants were not encoded in the lexicon. This situation appeared to change with the Optimality-Theoretic maxim of richness of the base (Prince & Smolensky 1993:191), which states that the input to the grammar is not restricted and that the only formal restrictions are imposed by grammars on their outputs. At first sight, this contradicts lexical economy, but in the three-grammar model defended in the current article, the lexicon is itself the output of the perception and recognition grammars, and the restrictions imposed by these grammars will be directly reflected in lexical forms (§3.6). Thus, lexical economy and richness of the base are not incompatible,