Re: owl:allValuesFrom and rdfs:domain

Bernard Vatant wrote:

> 
> I need some help from experts in logic ...
> 
> I've sent a few days ago a message about "TexasThings" example in OWL Guide
> (see below), where the interpretation of allValuesFrom seems wrong to me.
> Without answer so far from there, I push the question here.
> 
> Suppose I have the following - more enlightening to me at least than Texas
> Things :))
> 
> <owl:Class rdf:ID="BlueThing">
>   <owl:equivalentClass>
>     <owl:Restriction>
>       <owl:onProperty rdf:resource="#color" />
>       <owl:allValuesFrom rdf:resource="#ShadeOfBlue" />
>     </owl:Restriction>
>   </owl:equivalentClass>
> </owl:Class>
> 
> <owl:ObjectProperty rdf:ID="color">
>   <rdfs:domain rdf:resource="#VisibleThing"/>
> </owl:ObjectProperty>
> 
> What can be asserted between the classes BlueThing and VisibleThing ?
> 
> Of course, if some BlueThing X has a value for "#color", then X is in the
> domain of "#color". But using "allValuesFrom" means that some other
> BlueThing Y may not have any value at all for this property.
> How can this happen? Quite naturally, if Y is not a VisibleThing one might
> not be able to specify any value for its color. One knows somehow Y is a
> BlueThing, without being able to specify any ShadeOfBlue.
> 
> So a BlueThing is not necessarily a VisibleThing.
> 

Right.  The first block says that any color property of a BlueThing has 
to come from the collection ShadeOfBlue, but it does not say that a 
BlueThing has to have a color property at all.  So - with the help of 
the second block - a BlueThing also has to be a VisibleThing *if* it has 
been assigned a color.  Then it would have to be a VisibleThing. 
Otherwise, no constraint has been levied.

> There is more tricky.
> 
> <owl:Class rdf:ID="InvisibleThing">
>   <owl:complementOf>
>     <owl:Class rdf:about="#VisibleObject"/>
>   </owl:complementOf>
> </owl:Class>
> 
> Does the following triple hold?
> 
> InvisibleThing  rdfs:subClassOf  BlueThing
> 
> Sounds weird ... but I can't find any solid argument against it.
> 

I do not see anything that could require that conclusion.  Some 
BlueThings maybe VisibleThings - we would know for sure if one has a 
color assigned - and some may not.  The ones that have no color gove us 
no way to know about their status as VisibleThings one way or the other. 
  So there is no constraint that compels the subClassOf triple.

Or to put it in more RDF model-theoretic terms, there are 
interpretations in which some BlueThings are VisibleThings, and some in 
which some are InvisibleThings, and possibly some in which all 
BlueThings are Invisible Things.  To have an entailment, the conclusion 
has to exist in all possible interpretations, and it clearly does not.

Cheers,

Tom P

Received on Thursday, 18 December 2003 19:49:54 UTC