The Spanish lexicon stores stems with theme vowels, not roots with inflectional class features

Abstract The choice of theme vowels in Spanish nouns and adjectives can be predicted neither from the phonological shape of roots nor from syntactic features like gender. However, this state of affairs does not require the postulation of inflectional class features. The alternative is for the Spanish lexicon to store stems with their theme vowels, instead of roots annotated with declension diacritics; default generalizations over the lexical entries of stems can be expressed by means of lexical redundancy rules. The hypothesis of stem storage is compatible with the failure of Spanish theme vowels to surface in certain environments: this is demonstrably caused by an entirely general and regular phonological process deleting unaccented stem-final vowels before suffixes beginning with another vowel. Stem storage receives further support from psycholinguistic data from recognition latencies. Additional new evidence comes from cyclic locality conditions on allomorph selection, as shown by an analysis of the well-known stress-driven alternation displayed by items like [kont-á-ɾ] ‘count.inf’ ∼ [kwént-a] ‘count. 3sg’. In derivatives like 〚N 〚V kont-a〛 ðóɾ-∅〛 ‘counter’, assuming that diphthongal allomorphy is a property of the root incorrectly predicts that the choice of allomorph will be determined in the first cycle: i.e. *[kwentaðóɾ]. The locality problem vanishes if /koNt-a/ and /kweNt-a/ are both listed in the lexicon as stem allomorphs. These data show that Stratal Optimality Theory allied to a stem-driven theory of morphology performs better than alternative approaches to allomorphic locality. Root-driven Distributed Morphology is too local: the domains for allomorph selection that it generates are too narrow. Conversely, noncyclic versions of Optimality Theory fail to predict allomorphic locality (even if they can describe its effects) because they endow allomorph selection with unrestricted access to the global environment.