Re: Proposal to close ISSUE-69

Jeremy,

I'm afraid your reply is based on several misunderstandings.

First, it just seems false that users expect URI to "mean one thing"  
or are widely taught that. I know the web arch document would like  
that, but it just doesn't seem all that prevalent, mostly because it  
isn't operational.

Second, it's evidently the case, just as with "local names" instead  
of variables for bnodes, any philosophical worries are swamped by the  
pragmatic of not being able to load various RDF garphs "as" OWL  
graphs. From a user persepective, the latter is highly visible but  
the former is invisible. And harmless.

Third, nothing requires "separate objects" in implementations.  
Mapping to distinct terms is a convenient way to go for some  
implementations, but it is not required at all. It's not required in  
the semantics (you just need parameterized CEXT relations). It's not  
even detectable unless 1) you have equality and 2) you know what to  
look for in the entailments. It's perfectly clear that people just  
don't know and don't care and often don't want the a=b => A equiv B  
(which *itself* turns on a recognition of different roles)  
entailment. I remember last spring you were surprised by it (or to be  
reminded of it). Similarly, last spring other OWL Full people I  
talked with were both surprised and several said that it must be a  
bug in the semantics.

(And, in the RDF case, this entailment, necessarily restricted to  
atomic classes and having no class expressions can be easily  
recovered with a rule. So there is no *need* to impose it at the OWL  
level.)

(I imagine that if someone had no classes in my document, and had A  
sameAs B that that person would be very surprised to find a A equiv B  
*added to the RDFS+sameAS closure* of their document.)

Finally, as with the forcing owl:Thing to be infinite, you don't seem  
to be sensitive to any of the points I make, preferring instead some  
highly tendentious "ideological" considerations which simply *do not*  
have the technical impact you claim for them. You also ignore the  
very obvious user issues (i.e., graphs they can't process) which  
*dominate* the landscape. Also, you ignore that Pellet has punned for  
*years* and you never noticed or cared as a Jena developer...as you  
shouldn't.

To make it a bit easier on you, let's start simply: given that it's  
perfectly possible to have a semantics wherein a URI has "one thing"  
as its value (with a tweak to how EXT relations work), does that  
semantics address your philosophical concern? If not, why not? If  
not, what are the practical consequences?

Cheers,
Bijan.

Received on Wednesday, 23 January 2008 14:55:57 UTC