AW: Peter's slides about the MOF metamodel

Hi Conrad,

I agree with you that it is possible to come up with multiple metamodels of
the same language, but this is (in my view) not so much due to the variety
of syntaxes, but rather due to design choices one makes.
The OMG made the design choice to provide a metamodel of OWL 1 that models
the OWL language as an extension of the RDF language (to be precise: of the
RDF data model, not a specific RDF syntax). A different approach was taken
in the metamodel for OWL DL we developed in our group [1], which rather
provides a Description Logic oriented view, independent of RDF (closer to
the model of the abstract syntax). 

In any case, the metamodel for OWL 2 is intended to provide the structural
specification of the OWL 2 language independent of a specific serialization
syntax. There may be many concrete syntaxes. For some syntaxes the mapping
will be trivial (XML Schema), for others more complex (RDF). 

Regards,
Peter

[1]
http://www.aifb.uni-karlsruhe.de/WBS/sbr/publications/ontology-metamodeling.
pdf



-----Ursprüngliche Nachricht-----
Von: public-owl-wg-request@w3.org im Auftrag von Conrad Bock
Gesendet: Mi 30.07.2008 15:25
An: 'Boris Motik'; public-owl-wg@w3.org
Betreff: RE: Peter's slides about the MOF metamodel
 

Peter, et al,

You might have discussed this already regarding metamodel alignment, but
languages with multiple concrete syntaxes like OWL might have multiple
metamodels.  This is because metamodels often just remove concrete
notational elements like punctuation and graphical shapes, and add
relations that are expressed with juxtapostion in the concrete syntax.
The OMG's OWL 1 metamodel is based on the RDF/XML syntax, the OWL 2
metamodel is based on the "W3C abstract" syntax (which is concrete in
OMG terms).

Since the RDF/XML and "W3C abstact" syntaxes are so different, there are
potentially two metamodels for each version of OWL.  A couple options to
reduce this to one metamodel per version might be:

  - mapping one metamodel to multiple concrete syntaxes.  These will be
    complicated for metamodels and concrete syntaxes that don't
    correspond to each other.  For example, mapping the OWL 1 metamodel
    to the OWL RDF/XML syntax would be more straightforward than to the
    OWL 1 W3C abstract syntax.

  - Choose one concrete syntax per version for the metamodel, preferably
    the same kind in each version. Since the OMG OWL 1 metamodel used
    the RDF/XML syntax, it would be easiest for migration if the OWL 2
    metamodel is derived from the RDF/XML syntax, rather than the
    W3C abstract syntax.

All the options I'm aware of have problems, just wanted to give my take
on the tradeoffs.

Conrad

Received on Thursday, 31 July 2008 00:37:01 UTC