Re: WOWG: compliance levels on next teleconf

I don't believe that features interacting in a way that makes them
difficult to implement is the problem. What are these interactions and
difficulties exactly?

The real problem is that it is difficult/impossible to define a total
ordering on features based on their utility. This means that simply by
taking everyone's "single most important feature" we end up with the
whole language (more or less) - this is roughly what happened when
this issue was discussed in Amsterdam.

The solution proposed by the sub-group who met in Toulouse is to
define a minimal level 1 so as to encourage implementors to get on
board by making it easy for them to claim level 1 compliance. It was
assumed that various implementors would choose different points in the
lattice of possible features in accordance with their perception of
the requirements of their target market (and that to be
useful/competitive, most would implement something more that level 1).

Regards, Ian



On May 1, Frank van Harmelen writes:
> The discussion on how to define the compliance levels of OWL will be an
> important issue in the next teleconf.  
> 
> This email is as a preparation for the teleconf.
> 
> We should at least decide on how to reach consensus (although I would
> prefer to actually cut some knots during the teleconf).
> 
> Frank.
>    ---
> 
> 1. The strategy for defining compliance levels at the Amsterdam F2F and
> immediately afterwards was more or less to include any item in level 1
> for which somebody could make a convincing "frequent/common use" case.
> This lead to top-heavy proposals which were hard to justify because
> they  differed only little from full OWL. 
> 
> 2. A subgroup met at KR'02 in Toulouse, and explored an alternative
> strategy.
> As explained in messages by me, Deb, Ian a.o., the core of the Toulouse
> proposal was
> to identify a language that would be 
> 
> - easy to implement (thus encouraging toolbuilders)
> - a sufficient step up from RDF Schema (to justify existence)
> - not prejudice the difficulty of any extentions (partly or full) with
> other elements of OWL
> 
> This lead to the proposal of "RDF Schema on steroids" as a compliance
> level 1 for OWL (see [1] for what this includes). 
> 
> 3. Subsequent discussion revealed that various people would like to have
> additional items in level 1 (e.g. cardinalities, local range
> restrictions). Although none of these by themselves would be hard to
> implement, they are all judged to make it hard to add additional
> features because of the way the features interact.
> 
> EXAMPLE: an example is local range restrictions. When interpreted as
> existential ("slot S must have at least one value of type T") they are
> easy to implement, similar when interpreted as universal ("all values of
> slot S must be of type T"), but the combination of these two is hard to
> implement. Thus, including either in level 1 will make it harder to
> extend beyond level 1 and include the other. (Different communities seem
> to have different need for either extension). 
> 
> 4. The minimal required result at the teleconf is that we agree whether
> the "Toulouse approach" is the right one to take. 
> If yes, we can argue the details of "RDF on steroids" as level 1
> If no, we must formulate another strategy for defining level 1 (keeping
> in mind that the "Amsterdam F2F approach" has already shown to lead to
> failure). 
> 
> Please spend some time thinking all this over before the teleconf.
> 
> Frank.
>    ----
> [1] http://lists.w3.org/Archives/Public/www-webont-wg/2002Apr/0329.html
>

Received on Wednesday, 1 May 2002 20:38:02 UTC