Re: A possible way of going forward with OWL-R unification (ISSUE-131)

Hi Ivan,

Let me first say that I have some sympathies for both perspectives. I  
think we've inherited a fairly unhappy situation and it seems that we  
are (sensibly) trying to figure out "the best" coping strategy. I  
don't know that "the best" is actually discernable or even exists,  
which makes it extra tricky.

On 14 Jul 2008, at 10:05, Ivan Herman wrote:

> Hi Boris,
>
> thanks for reacting that quickly:-)
>
> Boris Motik wrote:
[snip useful distinction between documents and implementations]

> I must admit I do not think we have ever defined a profile as being  
> necessarily a _syntactic_ fragment.

We have not, but it's also pretty clear that there is a lot of  
pressure that way. 1) It's standard (in a general sense, i.e., how  
most logics are distinguished), 2) it's what OWL has at least strived  
for (see OWL Lite), 3) it's how the other profiles work, and 4) there  
are some inherent benefits (most of which Boris has articulated).  
Note that some people (e.g., Sandro) regard the lack of syntactic  
subsetting as a serious bug.

The bit I would add is that without syntactic subsetting in OWL-R  
(i.e., if the legal syntax is every RDF graph) we make it very hard  
to coherently and detectably *extend* OWL-R, even to cover larger  
fragments of OWL Full. I think this is especially critical given the  
"constructor" approach that many people have when characterizing  
languages (we already face this with DL-Lite btw). Part of what led  
people down the rathole in OWL Lite was thinking that merely  
forbidding constructs (without considering how the rest were used)  
was sufficent to significantly constrain the expressivity.

If we *don't* make OWL-R some syntactic subset, then I think we'll  
have a very hard time characterizing the "active" bits of the  
permitted graphs, esp. in a user understandable way. At least, I  
don't feel very confident about it.

(I think, point in fact, that people *do* treat RDFS as a syntactic  
superset of RDF in that they generally expect *something* to happen.  
Given that not much happens with RDFS :), it's easier to elide the  
difference.)
[snip]

> I did not comment on owl:intendedProfile because I did not really  
> think this through yet. I can understand it is a gross hack but,  
> frankly, I think it is much more important for us to have a proper  
> 'story' for the community, including the RDF community, than to be  
> too picky about the 'hackness' of a particular property. If we can  
> accommodate both then, well, it is fine, but, well...

As a proposer of this, I've become convinced that the hackiness isn't  
the primary problem, but that I don't see that we know enough to get  
it right. Just how to deal with "conflicting" statements is hard to  
figure out. If we make it *just* a hint or a very attenuated hint  
(that is, no firm semantics) then we've just introduced some more  
unconstrained vocabulary (a la deprecatedClass). Better *not* to do  
that and let the wider community invent what's needed.
[snip]

I think, in general, as is clear from the datatype discussion, we  
need to talk about conformance (whether of implementation or of  
documents). This is a regular problem (in my experience) and I don't  
think punting entirely (as webont aimed to do :)) is a good idea.  
Particularly if we expect certain knotty problems to be "solved" by  
punting :(

Cheers,
Bijan.

Received on Monday, 14 July 2008 09:46:50 UTC