From my thesis:
Models are represented as typed, directed graphs. So, if the metamodel of the modeling language under question is represented by the type graph M, any model can be described as a couple <G,t> , where G is a graph and t a homomorphism t : G -> M, also called a typing map.
In this thesis, the EMOF metamodel will serve as our type graph. So, for example, for a class c of a metamodel L=<G,t> it will be t(c)=EMOF::Class. Likewise, for an instance o of the class c in a model <Gi,ti>, it will be ti(o)=L::c.
(Granted, it looks better in Greek…)
Michalis, it is important to separate what something is, logically, to how something is implemented (clearly, EMOF is a how, not a what)