Morphologically-conditioned tonotactics in multilevel Maximum Entropy grammar*

Phonological sensitivity to lexical class has been addressed in the literature; at least two sets of hypotheses have been advanced regarding its codification in grammar. The first type of approach holds that lexical class differences are limited by restrictions in the design of grammar itself (or more specifically, Universal Grammar): for example, it has been hypothesized that only faithfulness, but not markedness, constraints can be ranked in a lexically class-specific way, or can be class-specific (e.g., Itô & Mester 1995a, 1999, 2009; Alderete 2001; Smith 2011); or that there certain preferential classes (e.g., nouns) demonstrate more contrasting phonological patterns than do other classes (e.g. verbs, with adjectives somewhere in between; Smith 2011). The second type of approach to lexical class specific phonology allows each lexical class to have its own completely independent phonological profile, without a priori restrictions or specifications on the structure of the grammar (Anttila 2002; Inkelas & Zoll 2007; Pater 2009; Becker & Gouskova, to appear; see also Itô & Mester 1995b).