Re: concerning lite, fast, large versions of OWL

Here we go again, I would almost say:-)

Large OWL (or: "OWL RDF-style") has classes as instances
  (any legal RDF combination of the constructs in the language is allowed,
   including "someclass type someclass)

Fast OWL (or: "OWL FOL-style") does not allow classes as instances
  (since this would go beyond the FOL framework)

Reasoners for OWL/RDF-style will be much harder to implement than reasoners 
for OWL/FOL-style (complete reasoners would be impossible to implement if 
OWL/RDF-style turns out to be an undecidable language, as it might well be)

In principle, we can decide to have OWL-Light "RDF style" (any 
RDF-combinations are allowed, include classes-as-instances), or we can decide 
to have OWL-Light "FOL style" (much more restricted).

Since OWL Light (still wanting a better name) should have ease of 
implementation as an important design criterion, it would make no sense to 
use the "RDF-style" for OWL Light, since this would make implementation of 
reasoners (and many other tools, e.g. editors, visualisers etc) much harder.

Furthermore, designing OWL light "FOL style" has the advantage that we have 
the following simple inclusion

	OWL Light < OWL/FOL-style < OWL/RDF-style		(1)

both syntactically and semantically.
Allowing classes as instances in OWL Light would break this chain; another 
reason for not having them in OWL Light.

It is often somehow suggested, either explicitly or implicitly, that a design 
according to (1) is somehow complicated. I disagree. I find such a strict 
inclusion of languages easy to explain to people whom I explain OWL to
(and that is a wide range of people varying from CS undergrads to industrial 
programmers and managers).

View (1) was discussed at the F2F (when a number of people were in another 
meeting discussing wine ontologies); there was widespread agreement among the 
people present at that discussion that (1) was a much better option than 
including classes-as-instances in OWL Light.


Christopher Welty wrote:

> Although I am strongly in favor of having "classes as instances" in some
> version of OWL, I am also very strongly in favor of as simple as possible
> a view of our language.
> Although consensus regarding the layering was a major accomplishment, it
> leaves us now with three versions of OWL: fast and large based on the RDF
> syntax/semantics, and of course the orthogonal "lite" version of the
> language.
> Having three versions of the language opens us up to some pretty obvious
> criticisms, in my view.  I think this would be even worse if OWL Lite,
> which is supposed to be a simplified version of OWL, is not a subset of
> Fast OWL, since Fast OWL is a subset of Large OWL.
> I was passionately ambivalent about OWL Lite in general, but I would
> strongly object to it as yet another subset of Large OWL.  Several people
> have expressed opinions that "classes as instances" should be in OWL Lite.
>  I'm not sure why - if it is allowed in Large OWL, then what difference
> does it make if it is in OWL Lite?
> -ChrisW
> Dr. Christopher A. Welty, Knowledge Structures Group
> IBM Watson Research Center, 19 Skyline Dr.
> Hawthorne, NY  10532     USA
> Voice: +1 914.784.7055,  IBM T/L: 863.7055
> Fax: +1 914.784.6078, Email:

Received on Thursday, 17 October 2002 05:01:18 UTC