- From: Jeremy Carroll <jjc@hpl.hp.com>
- Date: Thu, 29 Nov 2007 14:42:00 +0000
- To: Alan Ruttenberg <alanruttenberg@gmail.com>
- CC: Pat Hayes <phayes@ihmc.us>, Michael Schneider <schneid@fzi.de>, Owl Dev <public-owl-dev@w3.org>, Ian Horrocks <horrocks@cs.man.ac.uk>, bmotik@cs.man.ac.uk
Rather than talk about universal role, i.e. an object property; I perhaps should have talk about a universal attribute, i.e. a data property; or even the union of the two). In RDF, since all literals are resources, a universal role is a universal attribute. A worked example is given in: http://www.hpl.hp.com/techreports/2003/HPL-2003-142.html Briefly, to make RDF canonicalization work effectively, I needed a source of meaningless triples, that I could add or delete freely without changing the semantics of the graph. For simplicity I sufficed with a near-universal attribute, with range xsd:long. This can be achieved by defining the appropriate semantics of a new property as a semantic extension to RDF, or, less formally, with: <rdf:RDF xml:base=”&c14n;” xmlns:c14n=”&c14n;#”> <rdfs:Property rdf:ID=”true”> <rdfs:description>This property is true whatever resource is its subject, and whatever literal is its object. Thus triples with literal objects, and c14n:true as predicate, can arbitrarily be added to and deleted from an RDF graph without changing its meaning. </rdfs:description> </rdfs:Property> <rdf:RDF> Or within OWL 1.0, with the defn class(rdfs:Resource complete restriction(c14n:true cardinality=2^64)) dataValuedProperty( c14n:true range( xsd:long ) ) In OWL 1.1, we have already had the universal role defined by: http://lists.w3.org/Archives/Public/public-owl-dev/2007OctDec/0074 (modified) > (A0) ClassAssertion(w owl:Thing) > (A1) SubClassOf(owl:Thing ObjectHasValue(pA w)) > (A3) SubObjectPropertyOf( > SubObjectPropertyChain(pA InverseObjectProperty(pA)) > pU ) > (A4) ObjectPropertyDomain(pU owl:Thing) > (A5) ObjectPropertyRange(pU owl:Thing) A4 and A5 seem unnecessary The universal attribute cannot be defined directly in OWL 1.1 DL, mapping this construct to use DataProperty's hits various limits. The universal attribute can simply be defined by fiat as the product of owl:Thing by rdfs:Literal (mapped into the appropriate domain of interpretation). Then it would be possible to have every annotation property defined as having such an extension. Then they would be explicitly meaningless, in that knowing that such an annotation held would tell you nothing. But consider an axiom that is legally actionable: e.g. an axiom that contains within one or more of: a) a libellous statement (e.g. classifying somebody within an offensive class) b) a culpably negligent proposition e.g. subClassOf( owl:Thing, eg:KnownToBeSafeToConsume ) (which could well result in poisoning) Suppose that annotations are semantic free, then for either of these axioms, attribution information, such as who wrote them, can be freely added or deleted, and the culpability involved, would be ascribed mistakenly. Annotations have semantics. The question is what semantics. Jeremy
Received on Thursday, 29 November 2007 14:42:36 UTC