Ende exhibits a distinctive pattern in which the nasal segments [m, n, ɳ, ŋ] seem to float through words depending on the number of consonants present, always surfacing immediately before the leftmost noninitial obstruent. Words containing these nasals may have them surface in root-medial or root-final position when non-prefixed; in contrast, when a prefix precedes that root, the nasal segment may appear in root-initial position (cf. √ɡu: [√ɡuŋ-ɡ=e] ‘to disturb bees’ and [da-√ŋɡu-ɡ-ən] ‘he disturbed the bees’). This floating segment pattern is predicted to exist by Zoll’s (1998) typology of ghost patterns but has been unattested until now. Moreover, these data provide further evidence of morpheme-specified nasalization, as posited for languages like Mixtec, Southern Barasano, Guarani, Tucano, Ikwere, and Máíhĩki. This paper analyzes the floating nasal pattern using theoretical concepts from autosegmental phonology and optimality theory in the context of other nasalization and floating ghost patterns.
