W3C home > Mailing lists > Public > public-webont-comments@w3.org > January 2004

Re: OWL hole

From: Jeff Lansing <jeff@polexis.com>
Date: Wed, 28 Jan 2004 13:38:46 -0800
Message-ID: <40182BE6.1000604@polexis.com>
Cc: public-webont-comments@w3.org
Ok, the hole is getting smaller (or the owl is getting bigger, depending 
on your perspective).

The suggestion to reason about the optional 'A has C' from the range 
definition is a good one.

But this only works well when the domain of the property is just A. If 
the domain is also a union of classes (e.g. in the WSA case already 
mentioned) then a lot of other unintended inferences pop out.

Could I use something like maxCardinality = 0 to rule the unintended 
inferences out?

More specifically, given classes 'A', 'B', 'C', 'D', and 'E', a property 
'has', and the "reality" that A has B and optionally has C, and that B 
is distinct from C, and that D has E, how do I say that A doesn't have 
E, and that C doesn't have B or C?

Jeff


Benjamin Nowack wrote:

>Hi Jeff,
>
>(this is probably the wrong ML for a reply but I'll give
>it a try..)
>why don't you just remove the "someValuesFrom C" from your
>ontology. then "has" can relate individuals of type A to 
>either individuals of type B or type C. keeping 
>"someValuesFrom B" ensures that A has at least one B. 
>if C is optional, the range definition is sufficient.
>
>did I understand your question? or did you mean that it's
>not possible to attach cardinalities to value constraints
>(e.g. A 3x has B)? DAML+OIL has this type of qualified
>constraints, IIRC, but I'm not sure. If they are not 
>available in OWL now, then this is probably not a hole in
>the spec but has been well thought-through by the webont wg...
>
>hope that helps..
>benjamin
>
>
>On 28.01.2004 09:38:12, Jeff Lansing wrote:
>  
>
>>There seems to be a hole in the coverage of OWL.
>>
>>More specifically, given classes 'A', 'B', and 'C', a property 'has',  
>>and the "reality" that A has B and optionally has C, and that B is 
>>distinct from C, it does not seem possible to model this situation in OWL.
>>
>>Some considerations:
>>
>>Renaming 'has' to 'has-an-A' and 'has-a-B' appears to be a non-starter. 
>>The property has already been named in the "reality" that is being 
>>modelled. Perhaps I am translating from an E-R model, or from UML. Where 
>>is the information about this gratuitous renaming going to go? Perhaps I 
>>don't own the namespace of the property.
>>
>>If the example seems contrived, look at the WSA, where there appear to 
>>be cases just like this. (See e.g.: 
>>http://dev.w3.org/cvsweb/~checkout~/2002/ws/arch/wsa/images/ServiceModel.png)
>>
>>The best that I can do seems to be this:
>>
>>Ontology(
>>Class(A
>> restriction(has someValuesFrom (B))
>> restriction(has someValuesFrom (C)))
>>Class(B)
>>Class(C)
>>ObjectProperty(has
>> domain(A)
>> range(unionOf(B C)))
>>)
>>
>>which contradicts the fact that having C is optional.
>>
>>Jeff
>>
>>
>>
>>
>>    
>>
>
>
>  
>
Received on Wednesday, 28 January 2004 16:39:07 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:43:29 GMT