Re: SEM DESIDERATA: my initial desiderata list

>		Potential Desiderata for the Web Ontology Language
>
>NB:  These are not all my desiderata.
>
>Syntax:
>
>1/ Syntax is (equivalent to) n-triples (i.e., RDF syntax).

I would rather say, syntax can be *encoded* in RDF. I think that the 
key operational requirement here is simply that it is possible to 
send OWL through an RDF engine without causing a syntax error and it 
be possible to reconstruct the OWL afterwards. "Carried" is a good 
word. But note that this does not require that all RDF graphs are 
legal OWL.

>    All syntax (except, maybe, datatypes) is carried in triples.
>    Triple elements are URIs or literals (object only) or blank ids.

Note however that OWL might place a special interpretation on the 
urirefs that occur in some triples, particularly unasserted triples. 
We could take advantage of this.

>Semantics:
>
>1/ There is a model-theoretic semantics, compatible with the RDF(S)
>    model-theoretic semantics.
>    All triples are assertions.

I disagree with that one. I don't want that to be the case even in RDF.

>    All URIs denote elements of the domain of discourse.

OK, though I would rather say that all URIs *in asserted triples* 
denote elements.

>    All classes are elements of the domain of discourse.

Er...what does 'all classes' mean here? All classes in the domain of 
discourse are in the domain of discourse, obviously. But which 
classes are in the domain of discourse depends on what assumptions we 
make in the semantics. RDFS for example can always be given an 
interpretation in which the only classes (and properties) in the 
domain are those which are mentioned in the graph, ie there are no 
comprehension principles built into the semantics. I gather that OWL 
will not be so conservative.

>    All properties are elements of the domain of discourse.
>    rdf:type is the instance relationship, and a property.

It could consistently be a subproperty  of the OWL instance 
relationship. We might expect that OWL would have more ways for 
something to be member of a class than RDFS does.

>    All classes (including restrictions) belong to rdfs:Class.

Actually I would be quite happy to allow some OWL classes that were 
not in rdfs:Class, provided there was no inference path that would 
lead to that conclusion, if there was a pressing need to make this 
distinction. That might well be true for restrictions, for example.

The acid criterion for membership in rdfs:Class is being a value of 
rdf:type. If you can somehow prove a triple of the form

xxx rdf:type yyy .

then yyy must be in rdfs:Class. Otherwise, not necessarily. RDFS is 
really the RDF theory of rdf:type; all the RDFS notions of class 
membership hang on that one property. (The fact that CEXT is 
definable in terms of the EXT of rdf:type is a give-away in the MT.)

>2/ Inference is standard entailment in the model theory.
>    There is no special treatment of either side of the entailment.
>Classes:
>
>1/ A way to create / query complex properties is via a defined class, e.g.,
>    the intersection of a and b, the union of a and b, objects who have at
>    least one child.

A way, but not the only way. There is no need to create intersection 
classes to express conjunctive queries, for example.

I would add a semantic desiderata: Any OWL interpretation of some OWL 
syntax which includes some asserted RDF(S) should be an extension of 
an RDF(S) interpretation of that RDF(S). That is, OWL should preserve 
the meaning of all embedded asserted RDF(S) triples. This seems like 
a minimal criterion for OWL to be 'layered' on RDF.

Pat

-- 
---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes

Received on Monday, 15 April 2002 10:57:32 UTC