RE: XG W3pm Scope, metasubclassing

Evan, 

 >  Technology moves, but one can't arbitrarily pick some non-DL
 >  construct and assume it will be available in some later version of
 >  DL.

In this case we're referring to a specific construct (specializing owl
classes and properties).

 >  Decidability creates trade-offs in the language design.  As some
 >  features are added, other potential features can become forbidden.
 >  Since the feature in question is not part of the current standard
 >  nor the proposed revision, and I have yet to hear any DL implementer
 >  voice support for it, this feature is unlikely to be in any near
 >  term revision.  

It works in existing tools, at least for classes, see
http://www.w3.org/2007/OWL/wiki/OWLMetamodelingExample1.  Boris and I
have talked about modifying the compliance language to make it clearer
that being outside DL still allows tools to reason with the DL portions.

 >  However, user input has played a strong role in the choices of new
 >  features in OWL2.  If this XG thinks that this is an important
 >  feature, it might get more serious consideration.

We can fake "meta" subclassing with punning, but judging from the
discussion with Boris I don't think we'll get full metasubclassing from
the current OWL standardization work.  To use punning, we'll need to
enforce the subclassing rules separately.  I checked that the proposed
OWL documents no longer claim punning is metamodeling.

 >  I was thinking about benefits of models in RDFS/OWL, that are
 >  essential for the effort spent by a member of this XG to be
 >  considered worthwhile.  These may of course vary from participant to
 >  participant, but I had the impression that there was some shared
 >  concerns among those who instigated the chartering of this XG.

The reason to be interested in metamodeling in the broad sense is to
distinguish the many kinds of classes and properties we'll be defining.
For example, we'll have classes of products (with serial numbers), which
divide into those expressing requirements, those expressing design,
etc. We'll also have classes expressing structure, classes expressing
behavior, etc.  Then there will be the various properties that have
these classes as domain and range, which will also need to be
categorized.

Metasubclassing (metamodeling) lets us categorize the above classes and
properties to distinguish their role in the language we're defining.
It's basically syntax, but defined in a way that's tied to the
semantics.  I have a prepub paper using this approach in product
modeling, and a number of slide sets applying it to process modeling
(eg, see http://doc.omg.org/omg/08-06-32, starting at slide 27).

Conrad

Received on Tuesday, 8 July 2008 14:10:24 UTC