Print

Print


On 2/17/15 7:38 PM, Thomas Baker wrote:
> On Tue, Feb 17, 2015 at 10:53:37AM +1000, Holger Knublauch wrote:
>> Constraints are inherited from superclasses, i.e. subclasses can
>> only narrow down superclasses. That aligns with typical RDFS/OWL
>> semantics and therefore inferencing is no problem. In fact the LDOM
>> engine will walk up the class hierarchy to collect constraints
>> anyway. I would not consider this a bug but a feature.
> I'm not aware that RDFS semantics includes any notion of 'inheritance'.
> What does an RDFS subclass 'inherit' from its superclass?

See the rule cax-sco of the OWL 2 RL spec. If you have an instance of a 
subclass then the system will infer that it is also an instance of the 
superclass. This means that when you run constraint checking after 
inferencing, the engine will evaluate the constraints for the subclass 
and then those for the superclass.

>
> OOP classes have inheritance, but they also allow names to be
> overridden.  Could 'constraints' be overridden in LDOM, as you see it,
> and what implications would that have for RDFS subclass relations?

Inheritance in OO usually only applies to functions/methods. There is 
nothing comparable in RDF - just fields.

You can however narrow down the constraints in a subclass, and thus have 
a light-weight form of overriding, e.g. to narrow down the value type of 
a property from Animal to Person.

>
>> Likewise, pretending that "Shapes" are something different than
>> classes unnecessarily alienates a potentially very large user base.
> To my way of thinking, saying that shapes are RDF classes, or that they
> can have inheritance relationships to each other, risks unnecessarily
> confusing the users.

Maybe, but this statement is subjective and others will have a different 
opinion. If we were able to make a vote, who would we ask? In the 
semantic web community, maybe some people will object because the term 
class is already used for RDFS/OWL. But if you ask outside of the 
semantic web field, maybe the majority of people would prefer to link to 
already established term. This is not an easy decision.

Holger