RE: Antwort: RE: Semantic web article in Nature Biotechnology

Sorry, I was out of town for a few days.  Just came back and very glad to
see my "request" can trigger so many response.  I didn't even know there are
many people actually pay attention to this list.  Hope we can keep this up.
So, here will be my responses.

> The namespace in GO is a URI not a URL. Should there be any 
> expectation that it does anything other than 404? 

Yes, the RDF/OWL file should be accessed from the ontologies' namespace.
RDF's namespace differs from XML's.  The latter only promises a unique
string (that is why XML has schemaLocation tag but not in RDF) but the
former promise a set of axioms. 

> There have been many considerations of practicality made. Monolithic
> development is a problem, but both OBO and the MGED ontologies have 
> been attemping to ensure consistency between ontologies as well as 
> minimise overlap. 

Consistency is only the nesessary condition for an ontology but not
sufficient to make the ontology reusable. For instance, if I want to borrow
the "experiment" concept from MGED, I must import all other classes and
properties.  It is a tramandous waste.  Besides, what if I don't subscribe
to the view of the rest concepts?  Ontology should be build orthogonal to
each other. It needs to be carefully pacakaged by the use of namespace. I am
currently writing something about it but may take a while to finish it
because I am still chewing on some problems.   

> We routinely reason over ontologies much larger than the wine 
> ontology.

You can doesn't means you should.  The "wine ontology" issue is just a
figure of speech. What I want to imply is we need carefully designed
ontology to minimize unintended "imports".  

> Semantic web technology much scale to the size of the "encyclopedic" 
> ontologies; otherwise, what use is it? 

But in a distributed but not monolithic manner, don't you agree?  
 
> The boss ontology looks, at a quick glance, fairly similar to the 
> parts of the provenance ontology from mygrid, or perhaps the 
> experiment ontology by Larisa Soldatova and Ross King, or even the 
> work on Hybrow. MGED also has an experimental ontology somewhere
> within it, or, at least, I think that this is the case. 

The "parts" is the key.  I put BOSS up in such a trivial manner is for a
reason.  I can simple put up a hiearchy as the subClass of boss:Study,
boss:Data, etc.  We should, when designing ontology, give users a certain
degree of control on the granuality.  For instance, if I only do data
analysis, why I care about the boss:Protocol? If I don't need it, why I
bother to import it?

Though I do have some thoughts, I don't think I have all the answers.  I
hope from this, you can see why I post that request.

Received on Friday, 30 September 2005 02:19:35 UTC