The systematic relationship
between noun class marking in heterogeneous morpho-syntactic environments
suggests that the surface forms are derived from the same underlying morphemes.
While it is certainly possible that derivational processes change the
morpho-syntactic feature structure, I opt for the simplest possible assumption:
the surface forms differ, but the morpho-syntactic features are exactly the same
in all environments. The burden of proof is on claims to the contrary.
Since the function of NCs is to identify the grammatical gender of nominals, I assume that the nominal paradigm provides the best clues about the underlying characteristics of the featural and/or categorial make-up of NCs. I thus follow Carstens 1993 in assuming that noun class markers have the status of functional heads (of category D) which select for an NP argument. The existence of absolute pronouns shows that the [+N] complement can be phonologically null, i.e. pro. The general structure of nominals with class markers is thus:
 Aronoff (1994: 94) makes a similar proposal for Arapesh, a language from Papua New Guinea: "pronouns in this language are actually determiners of phonologically null nouns". He limits this structure to third person pronouns, because neither the verbal morphology nor 1st and 2nd person pronouns participate in gender agreement in Arapesh.