Re: ISSUE-5: n-ary datatypes - decidability on merging

Alan Ruttenberg wrote:

> I'm not sure how to characterize the interactions. For instance there 
> are issues around combining cardinality restrictions with transitive 
> roles that can be triggered on a merge.

These can only be triggered when the two ontologies being merged each 
contain one or more axioms mentioning a proeprty in common, and those 
axioms violate some condition.

And ditto with your other examples.

In the Pan and Horrocks version of the n-ary datatypes, the datatypes 
used in an ontology, as a whole, form a datatype group, and so if one 
ontology mentions decimal-multiplication, another decimal-addition, and 
a third integer (as a subtype of decimal), the merge of all three is 
non-conforming - even though the only common thing they have is the 
builtin decimal - which is not modified by any of the ontologies.

This has the potential to give a much worse user experience.

On the other hand, the approach used by racer is likely to be somewhat 
betetr behaved, but the formal properties seem significantly less clear.


> 
> Speaking as a user, I don't currently have the expectation that the 
> merge of two OWL-DL ontologies will yield yield an OWL-DL ontology.

No - but what about two almost completely independent  OWL-DL ontologies?


> My 
> gut is that we don't want to make this a restriction as we would land up 
> over constraining our ability to add useful features.
> 

I am not suggesting it as a general restriction - more as a pertinent 
restriction in this case - largely because there is a need for more 
theoretical work.

> More merge issues come up with subproperty chains.
> 
> Regarding N-ary datatypes, for the first part of this I think we need to 
> start talking about this in a way that people might understand. Vipul, 
> when he raised the question of working with linear inequations in OWL, 
> did not recognize the connection to N-ary datatypes.  To me, this 
> adequately demonstrates that we need to rephrase how we discuss this 
> feature in the specification and in our documentation.
> 
> I'm pretty sure that most application developers would gladly trade some 
> risk on merge for the ability to have  linear inequations in OWL, but we 
> could certainly test that with a survey.

Agreed. The problem is quatifying the 'some' in 'some risk'.

Also the n-ary datatypes design is much much more general than linear 
inequations - the member submission design is not a proposal to 
integrate linear programming with OWL

Jeremy

Received on Tuesday, 13 November 2007 15:52:32 UTC