In defence of the skeletal tier

This paper investigates empirically the viability of a hypothesis (advanced by Lowenstamm & Kaye [1986]) that morphological classes (in particular, those typical in languages with nonconcatenative morphologies) are fully definable in terms of syllabic structure. This hypothesis has theoretical significance, for, if correct, the skeletal tier-generally regarded as a core object in autosegmental phonology--becomes a derivative and, consequently, redundant entity. Data from four Ethiopian Semitic languages are presented as evidence that it is not always possible to posit unique syllabifications for morphological classes and that underlyingly many such classes are only partially syllabified, full sy11abifiabi1ity being secured by processes occurring later in the derivation. Analyses are proposed for the data, which demonstrate the necessity for a non-derivative skeletal tier.