[OWL] Modularize it!

[ewallace at cme.nist.gov]

Denny Vrandecic wrote:

>Modules mean you can buy what you want. A single OWL 1.1 means you have 
>to take it all. It's like with the XML DOM API, standardized by the W3C 
>as well, and the seperation in different modules (and even levels).

Detractors will say that modules mean that tools can claim to be OWL
compliant even though they do not support the same features and thus
are not interoperable.  

The reason that I am wary about this idea is because even the limited
options provided in OWL were controversial.  For some of those with loud
voices in the Semantic Web activity this profile idea is anathema.  This
is the case, even though conformance points, conformance classes, and even
conformance building blocks (which are similar to what Denny has proposed) 
have been used successfully in other standards efforts over many years. 
However, I didn't know that such features had been included in other W3C
standards.  That is interesting.

>The point is, that most existing tools do not support all of OWL DL, but 
>more than OWL Lite. I don't think this will get exactly better by 
>extending the possibilities of OWL with things like Units, extended 
>Annotations, RDF-incompatible comments, Metamodelling through Punning, 
>Axioms with URIs, or whatever comes in eventually in some future OWL 
>version.
>
>With releasing well-named modules, tools can exactly claim what they 
>support and what they don't, just using a well defined label instead of 
>saying "Yes, we support all of OWL DL reasoning, just without nominals." 
>Not everyone is aware of what nominals are, and it's not obvious from 
>the specification. But by giving names to the different modules you can 
>just tell the sales person in your company "Get me a package that 
>supports this "OWL XA + Meta", or whatever the names.
>
>My guess would be that this will help the creation of OWL based tools, 
>as you don't have to implement all the features at once and still be 
>standard-compliant. It makes the whole framework more flexible.

This last bit is exactly what some don't like about such a flexible 
framework as part of a standard.  However, I am sold on the idea and
would be happy to see it included as part of a proposal for a revised 
OWL recommendation, provided it was kept as a separate goal from the OWL
1.1 feature enhancements.  As part of this, we could demote OWL Lite 
to being just one profile among many that could be described with the 
mechanism we define to characterize feature sets.  I doubt anyone would
complain.

...

>> I prefer an extensions/modules approach for things that are more 
>> substantively potentially in conflict, or just mark clear boundaries. I 
>> don't think any of the OWL 1.1 things are like that.
>
>Breaking RDF-semantics?

Indeed.  This is the one bit of the 1.1 feature discussion that I find 
worrisome.   This could cause much more resistance to formalizing an OWL
revision than the profiling capability discussed above.

-Evan