It is well-known that Ethiopian Semitic languages have different verb classes which determine root shape in different paradigms, and cannot be reduced to syntactic, semantic or phonological features of the respective roots. In current analyses of these languages, it is implicitly assumed (e.g. Buckley, to appear, for Tigrinya) that verb classes are atomic (feature) specifications associated with roots.
In a line with Mueller's (2003) analysis of noun classes in Russian, I argue that verb classes in Amharic must be actually decomposed in different, more basic diacritic features. These features correspond roughly to properties like "gemination in the perfect aspect" or "a after the penultimate root consonant", which characterize together traditional verb classes (e.g. "A,B,C" for triradical and "1,2" for quadriradical verbs). Departing from Mueller's approach, I assume that class features are organized in a feature-geometric tree as argued for pronominal features in Harley and Ritter (2003).
This analysis allows to account for the fact that only certain gemination and vowel patterns coocur. Moreover, there is morphologically conditioned neutralization of verb classes, i.e. in some derivational patterns, verbs "join" a different inflectional class or a pattern which corresponds to an "intermediate stage" in the feature geometry, not instantiated by underived verb forms. These changes are unidirectional: there is neutralization of class A to B and of A and B to C, but not the other way around. Unidirectionality follows from the feature-geometric account under the assumption that gemination and vowel quality are the effect of Distributed-Morphology-style spellout operations, class neutralization is impoverishment (zero spellout), and spellout operations are always feature-deleting (Trommer, 2003).
Crucially, neutralization of contrasts between verb classes seems to be problematic for approaches which exploit paradigms based on lexemes (e.g. McCarthy, 2002), since the Amharic data would imply "attraction" among forms belonging to different roots. But even allowing paradigms corresponding to specific derivational affixes would not explain the unidirectionality of neutralization which follows straightforwardly from the feature-geometric account.
Buckley, E. (to appear) Emergent vowels in Tigrinya templates. Research in Afroasiatic Grammar II (Proceedings of the 5th Conference on Afro-Asiatic Linguistics), ed. Jacqueline Lecarme. John Benjamins.
Harley, H. and Ritter, E. (2003) A feature-geometric analysis of person and number. Language, 2003.
McCarthy, J. (2002) "Optimal Paradigms." Ms. ROA.
Mueller, G. (2003) A Distributed Morphology Approach to Syncretism in Russian Noun Inflection. In: Olga Arnaudova, Wayles Browne, Maria Luisa Rivero, & Danijela Stojanovic (eds.), Proceedings of FASL 12.
Trommer, J. (2003). Feature (non-)insertion in a minimalist approach to spellout. Proceedings of CLS 39: Main Session.Jochen Trommer