Re: Inferencing more specific roles

Hi Denny,

Thanks for pointing this out. Yes, with your asserted data, you can 
inference a more specific role, with the addition of just one more axiom 
to your list that asserts or infers that Betty is a Person - one of:

ClassAssertion (Person Betty)

or

PropertyDomain (parent Person)

Also, you might be interested that you can get this inference without 
the disjoint property axiom OR the closure axiom on the domains of those 
properties because they are mutually redundant (it was this sort of 
redundancy I was investigating).

Either way, my 'challenge' was to identify different (minimal) data 
models as for this very small domain, and then see how many of them 
could be made directly comparable to each other with inferences from an 
OWL model. My original example is one for which I would like to be able 
to add axioms into an OWL ontology so that the more specific 
relationship can be inferred.

Here, your data model that represents relationships from the Parental 
perspective is another possible data model (from third theoretical 
datasource in the context of my original example). As you have 
suggested, we can get interesting inferences out of a few assertions 
with a nice rich model, and so OWL expressivity does well there.

Cheers
Andrew


Denny Vrandečić wrote:
> Funnily enough it seems that the other way (not kids, but parents) it is
>  indeed possible.
> 
> Given the TBox:
> DisjointUnion( Person Male Female )
> PropertyRange( mother Female )
> PropertyRange( father Male )
> SubPropertyOf( mother parent )
> SubPropertyOf( father parent )
> DisjointProperties( mother father )
> SubClassOf( Person ExactCardinality( 2 parent ))
> SubClassOf( Person ExactCardinality( 1 mother ))
> SubClassOf( Person ExactCardinality( 1 father ))
> 
> From the ABox:
> PropertyAssertion( parent Betty Adam )
> ClassAssertion( Male Adam )
> 
> Now the following does follow:
> PropertyAssertion( father Betty Adam )
> 
> Using the OWL2 Syntax from here:
> <http://www.w3.org/2007/OWL/wiki/Semantics>
> 
> This is due to the interplay of cardinalities and ranges.
> Or am I wrong?
> 
> denny
> 
> (thanks to Sebastian Rudolph who checked the claim briefly, since my DL
> is not so great)
> 
> Bijan Parsia wrote:
>> On 11 Aug 2008, at 14:35, Andrew Gibson wrote:
>> [snip]
>>> In this case, the Property hierarchy looks like:
>>>
>>> hasChild    (range Person)
>>>     hasDaughter    (subPropertyOf hasChild, range FemalePerson)
>>>     hasSon        (subPropertyOf hasChild, range Maleperson)
>> [snip]
>>> In this case I was investigating how Theoretical Datasource 1, that
>>> states for each individual:
>> My first, reflexive response is "DL Safe Rules", since you are aligning
>> data.
>>
>>> Gender         (as MalePerson or FemalePerson)
>>> Children    (as hasChild)
>>>
>>> could be integrated with another Theoretical Datasource 2, that uses:
>>>
>>> Gender        (not asserted, only Person)
>>> Children    (as hasSon and hasDaughter – Genders inferred)
>>>
>>> Simply put, the problem with directly integrating these is that in the
>>> Datasource 1, given the Class assertions:
>>>
>>> Vera instanceOf FemalePerson
>>> Jane instanceOf FemalePerson
>>> Vera hasChild Jane
>>>
>>> It does not seem possible to infer the more specific relationship:
>>>
>>> Vera hasDaughter Jane
>>>
>>> that would make the data directly comparable to the representation of
>>> Datasource 2.
>> [snip]
>>
>> You might be able to use the following trick:
>>     http://www.w3.org/2007/OWL/tracker/issues/22
>> Which would give you a stronger result (i.e., class subsumptions) than
>> the corresponding DL Safe rule.
>>
>> Cheers,
>> Bijan.
>>
>>


-- 
Dr Andrew Gibson
Universiteit van Amsterdam

Received on Tuesday, 12 August 2008 09:22:59 UTC