There are very interesting observations about this relationship on the final pages (pp. 313-314 sic!) of the current revision of the submission by IBM and others to OMG of an Ontology Definition Metamodel (ODM) http://www.omg.org/docs/ad/05-08-01.pdf

Metamodels are mechanisms from Model Directed Architecture which, among other things, are designed to promote interoperability of models, not merely interoperability of the systems which implement them (which then comes for free in rigorous MDA, because implementations are a special kind of model). Users of modeling tools which convert from one model to another, e.g. code generation from UML, are often the beneficiaries of formal or informal metamodels in the conversion framework.

[Here I must confess to having read only the TOC and these two pages of the ODM submission. But from those, I am prepared to speculate that the rest of it has a lot to say about the topics at hand. Indeed, if it is a successful metamodel proposal, it will have identified every single issue raised here in the last few weeks, and more....].

Bob

Roderic Page wrote:

Dear Gregor,

Various comments:

Relationship to ER

In many ways RDF is very like ER-modelling (e.g., http://bit.csc.lsu.edu/~chen/pdf/Chen_Pioneers.pdf), indeed the W3C states that "RDF is a member of the Entity-Relationship modelling family." One could think of the modelling the link between two entities as a triple.

There's also a comment on this on Danny Ayers' blog (http://dannyayers.com/2005/03/16/xml-andor-rdf/) - embedded in some other extraneous stuff.

You might also want to look at D2R Map (http://www.wiwiss.fu-berlin.de/suhl/bizer/d2rmap/D2Rmap.htm) for mapping between database schema and RDF.

One way I like to think of this is that I think RDMS are geared towards locally held data, hence the emphasis on primary keys and normalisation. If, however, much of the data is elsewhere (as I believe it is in biodiversity informatics) then you need ways to refer to remotely held objects (i.e., URIs, which is another way of saying GUIDs). Hence, instead of locally defined primary keys you start to store GUIDs (which, by definition, are unique). If you then start to think about relationships between the objects identified by GUIDs, then you've pretty much got RDF. I would also say that in terms of managing local data, RDBMS are not going away. RDF, in my opinion, only really becomes relevant once you want to make the data available to others, and once you want to enable people to be able to aggregate that data.

[...]