The talk investigates spatial case marking in Tsez. Comrie and Polinsky (1998) argue for the decomposition of these forms into at least two morphemes (roughly Path and Place), and optionally others, like the distal marker. The talk points out that the bi-morphemic analysis leaves several puzzles unanswered. To resolve them, I argue that a tri-componential underlying structure is needed, augmenting Path and Place with Svenonius’ AxPart. Despite the tri-componential structure, the marking of some cases is indeed bi-componential on the surface, i.e., some of the expected markers are missing in some of the cells. The talk argues that this is because of portmanteau realisation: three underlying positions are present but realised by two markers only. The specific conditions under which this happens provide us with some general insights into the process of lexicalisation, arguing against context-sensitive rules as a tool for modelling allomorphy.