Re: Semantics of rdfs:subPropertyOf

Michel Klein wrote:
> 
> "McBride, Brian" wrote:
> >
> > I had assumed the interpretation was that if resource R has to properties
> > sp1 with value v1 and sp2  with value v2, where sp1 and sp2 are both
> > subproperties of p, then has R has two properties P, one with value v1 and
> > the other with value v2.
> 
> We agree with this interpretation of the current RDFS specification, but
> our main concern is that the inheritance of *values* will make the reasoning
> nonmonotonic. It seems to us that this is not desirable for a basic technique
> like RDFS. All techniques that will build upon it, will have to deal with
> these problems.
> 
> If the definition of subPropertyOf is changed according to our proposal
> (only inheritance of domain and range), this problem does not exist.

As I understand it, subPropertyOf is meant to allow you to build
taxonomies of properties. If I say childOf(A,B), I also mean
relatedTo(A,B), and it is nice to be able to specify these semantics by
saying subPropertyOf(childOf, relatedTo). This could be easily expressed
in a Horn logic which is clearly monotonic.

Now if you add things like cardinality constraints to a language built
on RDF, then yes you might have conflicting information that must be
resolved. However, this is a problem of adding cardinality constraints
in general and is not due to the definition of subPropertyOf. Even
without subPropertyOf, you still have to specify means of resolving
conflicts between statements made by multiple individuals. For example:

if president has cardinality 1, someone said president(US, Bush) and
someone said president(US, Gore) then we have a conflict. If we
automatically accept the first president statement, and reject all
others then our logic is still monotonic. However, the "whatever I heard
first must be the truth" rule seems pretty silly in practice. More
likely than not, you'll have to add some sort of rules for comparing
conflicting statements and choosing one based on a series of
argumentation. This would be nonmonotonic, because you may remove a
previously made assertion based on new information.

Jeff

Received on Tuesday, 12 December 2000 11:17:20 UTC