How complex are complex words? Evidence from linearization

There is considerable disagreement between theories of morphology concerning the complexity attached to words consisting of more than one morpheme. While, e.g., Distributed Morphology views complex words as a hierarchical structure of individual pieces associated with morpho-syntactic features, inferential frameworks such as A-Morphous Morphology and Paradigm Function Morphology treat complex words as morphologically simplex, consisting merely of a phonological string without any morphological constituent structure. Based on evidence from the Bolivian language Baure this paper argues that the restrictiveness of the latter view prevents an elegant analysis of certain syncretism patterns. The pervasive property of the Baure paradigm is that all agreement markers may appear in word-initial and work-final positions. This pattern can only be directly expressed in the analysis if complex words actually have more than just phonological structure. The argument thus challenges rule-based frameworks of morphology.