Keywords: phonologically conditioned morphology, architecture of grammar, affix order, Huave
<1> 1 IntroductionThe phonological conditioning of mobile affix placement in Huave, as analyzed in Noyer (1994) and Kim (2010), appears to be problematic for a modular feed-forward architecture of grammar in which all morphological operations precede all phonological operations. Following McCarthy & Prince's (1993) approach to phonologically conditioned morphology, the analyses implement a "P >> M" strategy within Optimality Theory where phonological (P) and morphological (M) constraints are evaluated in parallel. Mobile affix positioning then results from a global optimization of phonological structure, at the expense of morphological preferences and defaults. In other words, the outcome of phonological processes such as epenthesis and syllabification must be evaluated in order to determine whether the mobile affix surfaces as a prefix or a suffix, and this type of analysis is incompatible with an ordered separation of the two modules.On the other hand, Paster (2009) argues that "true" cases of phonological affix order do not exist. In an extensive survey of putative cases of phonological affix order, she finds that nearly all of them are better analyzed as cases of purely phonological operations taking place after affixation, of phonologically conditioned suppletive allomorphy, or of phonological subcategorization (where phonological underlying forms are visible to the morphology, but not touched by it; cf. Bermúdez-Otero's 2012 "insect trapped in amber" analogy), all of which are compatible with a modular feed-forward architecture. Huave is the only case that is not reanalyzed (although a potential direction for reanalysis is suggested), and so the question remains of whether any strictly modular analysis is possible. If so, this is consistent with Paster (2006aPaster ( , 2009 and Yu (2007)'s arguments against the P >> M model of phonologically conditioned morphology. If not, we are left in the uncomfortable position of having very few counterexamples to the claim that phonologically driven affix ordering does not exist. Aside from Huave, these include Athabaskan (Rice 2011:183) and Moro (Jenks and Rose 2013).In this chapter I make three points. The first is that Paster's (2009: 34) tentative reanalysis of Huave mobile affix placement, as a phonologically predictable procedure of associating floating features to skeletal positions (cf. Rose 1995), is not compatible with the data. The second is that Huave mobile affixation is compatible with a modular feed-forward architecture and the subcategorization-based approach, if we pursue an alternative generalization that allows us to follow Paster's (2009: 35) reanalysis of Afar mobile affixes and view the Huave case as a type of phonologically conditioned suppletive allomorphy. The P >> M architecture is thus no longer strictly needed. The third point, however, is that the suppletion analysis is potentially powerful enough to generate the type of data predicted b...