Aymara is an isolate suffix-only SOV language spoken in parts of Peru, Bolivia, Chile and Argentinia. Its person/tense agreement paradigm comprises four person categories (1st, 2nd, 3rd, 1st inclusive) and four tenses (future, simple, recent past, remote past). Although Coler (2015) claims the portmanteau morphemes encoding these categories cannot be systematically separated into tense and person markers, the patterns one can see in the paradigm suggest there is a way of breaking them down into tense, subject and, in some cases, object markers. However, this raises a couple of issues for theoretical morphology:
- The linearisation of tense and person markers depends on the subject>object constellation: In the 3>3, 1incl>3, 2>1, 3>1 and 3>1incl forms, person precedes tense while in all other contexts, tense precedes person. In Distributed Morphology, according to the Mirror Principle (Baker 1985), this differential linearisation must already be there in the syntactic derivation, i.e. in the adjunction of v to T. For predefined templates in Paradigm Function Morphology, this is a serious challenge.
- In the 2>1 future as well as in the 2>1, 3>1 and 3>1incl remote past forms tense is not expressed by a CV suffix but by partial reduplication. Assuming Marantz' (1982) theory of reduplication, this can be seen as suffixation of melody carrying CV skeletons and copying melodic elements of the preceding subject agreement suffix. In the same way, the property of some suffixes of deleting the preceding vowel can be analysed as a melody carrying C skeleton at the left edge of the suffix and a high ranked DEP constraint in phonology.
- Some of the markers appear in contexts that do not form any natural classes. This can be easily handled in Paradigm Function Morphology by postulating arbitrary rules of referral. In Distributed Morphology, as it seems, we have a choice between expanding the feature decomposition by adding features, postulating impoverishment rules and assuming arbitrary homonymy.
