RE: [BioRDF] UML/RDF [Was: Meeting Notes Feb 27, 2006]

--Michael,

> For me, my interest was in the transformation of an RDF XML 
> document to the equivalent (to some measure of equivalent) 
> UML XML document and back to describe/define an ontology.

What is the purpose of this? I am a bit fazed.
 
> The practical problem with the RDF XML notation we ran into 
> for our use is that many of the XML element names are 
> dependent on the class and/or property names in the ontology, 
> so every ontology would need to have its own XML Schema to 
> validate the document. 

At least in its current state, you cannot validate RDF/XML. Not with XML
schema language.  Someone say it can be done with Relax-NG, which I don't
know anything about.
But what is the point of validation?  Part of the reason that XML needs
validation is due to its tree-based data-structure.  So, a simple alteration
could have serious effects on an application.  RDF/XML uses directed labeled
graph, adding or changing subgraph doesn't break the model.  In other words,
you process what you know and ignore the rest.
 
> Our use case for MAGE-ML (generated from the MAGE-OM UML 
> model) is to simply allow creating and/or referencing 
> individuals based on arbitrary ontology classes 
> (http://mged.sourceforge.net/).  Not arbitrary to the user 
> but arbitrary as far as applications based on MAGE.  So we 
> needed a formulation in our model that was not dependent on 
> RDF specific XML schemas.

You can NOT selectively import a part of the ontology without some other
engineer artifacts like named graph or C-OWL. In straight up OWL, even if
you want to reference a single URI from MGED, you must take the entire
ontology with you.  This is my very first problem with MGED.  The design has
no modularization whatsoever, which will ultimately make it unsharable to
anyone outside of its community.

> Since we have no interest in incorporating entire ontologies, 
> only referencing, our needs are a bit simpler than probably a 
> lot of other folks.  In fact, it was only the Individual 
> diagram from an earlier draft, which has been deprecated but 
> still described in "17.2.4 InstanceSpecification", that is 
> the basis for what will be used in the FuGE-OM UML model 
> follow-up to MAGE-OM (http://fuge.sourceforge.net/).
> 
> The specification is also useful for thinking how arbitrary 
> ontologies can be referenced from relational or object 
> databases without needing foreknowledge of particular ontologies.

If everyone developed "arbiturary" semantics to SW ontology, I wonder why
bother going to RDF.  Any existing technologies will do.  The key and - the
very key IMHO - of the entire web thing is to share.  To share resource with
URI, to share our comprehension about world in RDF.  Now, if every ontology
is always "arbituarily" used, what can we possibly achieve with SW
technologies?

Xiaoshu

Received on Thursday, 2 March 2006 19:29:55 UTC