One way to approach the Metadata task

At PAWS.SIS.Pitt Lab (AKA Peter's lab) we did an excersice aimed at semantic description of the data models of some of our applications.
While not all of this excercise is about metadata, I do feel like the way we did it might suggest a way to do it for Ensemble.
Here you can find more details (http://adapt2.sis.pitt.edu/wiki/ADAPT2_RDF_Binding).
We have split the problem into systems/applications that have data models expressed in terms of objects and realtions (ER model). Then we have looked at several "vocabularies" available out there - Dublin Core one of them - and listed all of the respective objects and properties. We created links between vocabularies and ER models. The work is half-done/in-the-process, but part of it is already implemented and works.