Re: motivations for owl 1.1 design choices

On 12 Feb 2007, at 15:46, Jeremy Carroll wrote:

>
>
> Hi
>
> I am sorry that I am somewhat behind on owl 1.1; I am trying to  
> catch up at the moment.
>
> One area that I could do with some help is understanding the  
> motivations for the choice of the new features. In particular I read:
>
> http://owl1_1.cs.manchester.ac.uk/overview.html#1
> [[
> designed to provide simple extensions to OWL DL that
>
>    1. have been requested by major users of OWL DL;
>    2. have effective reasoning methods, as evidenced by theoretical  
> results; and
>    3. are expected to be implemented by the developers of OWL DL  
> reasoners.
>
> ]]
>
> It's unclear whether this is a conjunction or a disjunction.
>
> i.e. are all new features ones requested by major users of OWL DL,  
> or only some of them?
>
> are all the features ones for which effective reasoning methods are  
> known to exist, are they all expected to be implemented?

It is a conjunction.

>
> ====
>
> I am aware that QCR are required by some OWL DL users.
>
> How much of SROIQ is similarly in response to user request, and how  
> much is it merely a statement of the possible; without being a  
> comment on the desirability of these features?

More expressivity w.r.t. roles, in particular being able to express  
propagation of one role across another, is crucial in many  
applications, e.g., in the life sciences. If you look at some of the  
examples in the SWRL proposal you will see that they concern exactly  
this requirement, e.g., the "uncle" example, and the "style of an art  
object is the same as the style of the creator" example. I recall  
that the lack of this feature was also a serious impediment to the  
use of OWL in web services applications.

A simple and elegant way to address this requirement is to extend the  
set of role constructors to include composition. This allows for  
complex role inclusion axioms of the form [hasLocation o isPartOf -->  
hasLocation] or [parent o brother --> uncle]. This is exactly what  
SROIQ provides (with some restrictions needed in order to retain  
decidability). It also provides for reflexive, irreflexive,  
symmetric, antisymmetric and disjoint properties. These are an easy  
addition once we have complex roles, and there are convincing use  
cases for them (see recent discussions on this list).


>
> ====
>
> we are trying to determine which parts of OWL 1.1 are the most  
> critical, so that we can prioritise our attention and support for  
> those features. This seems to be about understanding the 'pull'  
> rather than the 'push'. The SROIQ paper gives the push; are there  
> clear indications for the pull somewhere; maybe in the OWL  
> experiences and directions workshop?

Exactly. The design of the language really did come out of extensive  
discussions between logicians, ontologists and systems developers,  
including several sessions at the OWLED workshops in Galway and Athens.

Regards,

Ian



>
> Jeremy
>
>
>
>

Received on Monday, 12 February 2007 16:44:02 UTC