Re: [SWC] RIF & OWL compatibility

On 12 Jan 2006, at 17:06, Michael Kifer wrote:
> From talking to a number of people with real-world RDF experience,
> I get the impression that b-nodes are used to refer to individuals  
> when it
> is too inconvenient to give them explicit names. This is just  
> Scolemization.
>
> Does anybody have *real-world* RDF experience with cases where truly
> existential semantics of b-nodes is used?

You don't need to go very far.
It is needed in order to capture pure basic SPARQL with RDF entailment.

For example, given the data
     age(:john, "25"^^xsd:decimal)
corresponding to the RDF triple
     :john :age "25"^^xsd:decimal .
then the following is RDF entailed
     rdf:XMLLiteral(_:b)
corresponding to the RDF triple
     _:b rdf:type rdf:XMLLiteral .
where _:b is a bnode, with a true existential semantics for _:b.
Note that the triple
     "25"^^xsd:decimal rdf:type rdf:XMLLiteral .
is not legal RDF, since literals can not appear in subject position.  
In fact, the RDF triple
     _:b rdf:type rdf:XMLLiteral .
is really the strictest (and only) entailment that can be done from  
the starting data.
So, the SPARQL query (allow us some liberty in the syntax for the  
sake of clarity)
     Q() :- rdf:XMLLiteral(_:b)
should return TRUE.

Now, let's play this game in RIF, that should *at least* capture the  
basics of SPARQL.
In this example we have chosen to represent triples within RIF with a  
predicate "triple/3".
Given the data
     triple(:john, :age, "25"^^xsd:decimal)
the above query can be rewritten as
     :- triple(B, rdf:type, rdf:XMLLiteral)
In order to correctly answer the above basic query in RIF, you need  
an additional (unsafe) rule of the type:
     triple(B, rdf:type, rdf:XMLLiteral) :- triple(X, Y, Z), isLiteral 
(Z)
where the variable B in the head is a true existential.
Note that the safe rule
     triple(Z, rdf:type, rdf:XMLLiteral) :- triple(X, Y, Z), isLiteral 
(Z)
would be wrong, since it would entail the fact
     triple("25"^^xsd:decimal, rdf:type, rdf:XMLLiteral)
which is not a legal RDF triple.

The moral is: if RIF has to at least capture (and extend with  
recursion, negation, etc) basic SPARQL with RDF entailment, then true  
existential in the heads of rules are required.

cheers
--enrico+sergio

Received on Saturday, 14 January 2006 16:55:53 UTC