Re: declaredAs

On 15 Aug 2007, at 11:44, Turner, David wrote:

>> -----Original Message-----
>> From: Bijan Parsia [mailto:bparsia@cs.man.ac.uk]
>> Sent: 14 August 2007 12:22
>>
>> On 14 Aug 2007, at 11:52, Turner, David wrote:
>>
>>> I realise that this point has been elaborated on later in
>> this thread,
>>> but I wanted to express my general agreement with it anyway.
>>> Punning is
>>> rather different from what OWL Full does,
>>
>> I don't know what you mean by "rather different". For me, for
>> example, "rather different" suggests *incompatibility*.
>
> Roughly speaking, yes. A user well-versed in OWL-1.0 (DL or Full)

As we are defining a future language (i.e., a revision of OWL) that  
we hope will become the dominant dialect, I think we have to label  
this a transition cost. To measure the *actual* transition cost, we  
have to measure things like how many existing ontologies this would  
change, what changes to tools would be needed to support the Full  
entailment, etc.

> coming
> across the example I gave might

"might"

I think in the existing userbase this is not at all a problem  
because, to my knowledge, there are no ontologies which depend on the  
additional Hilog behavior. Indeed, when I explain the difference  
between punning and the hilog part of OWL Full to various people  
(including, I believe, you), they are *surprised* by the Hilog  
behavior. In fact, I had one user, once they understood the Hilog  
semantics, said that *that* was a bug!

I don't believe we came across any examples of classes being sameAsed  
in our survey, but I don't recall specifically looking for that.  
Something worth doing.

> be surprised by its meaning in OWL-1.1.
> I realise that the model theories are compatible from a strictly
> mathematical POV, but surprising users by changing the meaning

...which is in common use...

> of
> already-defined syntax is likely to be painful.

...if it's not clearly marked, etc. etc.

And I don't know how likely that would be, esp. given how surprising  
that entailment can be.

I have trouble coming up with actual use cases as well, which is at  
least suggestive that it's not all that useful a feature (though,  
perhaps I'm just not creative enough).

>> Punning is, as far as I know, strictly weaker than Hilog or
>> OWL Full semantics.
>
> Right, but it was a design goal in 1.0 to have the DL semantics as  
> close
> to the Full semantics as possible.

One design goal. And I think it's derivative on wanting the  
*languages* to coincide as much as possible.

> Punning is a further divergence,
> because it's strictly weaker.

Any OWL ontology with syntactic higher order use is *not* a legal OWL  
DL ontology. So, all the cases we'd have to consider are  
*meaningless* (since ill formed) in OWL DL, but meaningful in OWL  
Full. This is a rather large, user impacting divergence as witnessed  
by the rise of (often used) repair tools.

In current OWL 1.1, those ontologies are syntactically legal and  
meaningful and their semantics (as far as I can reasonably tell)  
coincides with the OWL Full meaning. Those semantics were also  
implemented prior to the OWL 1.1 language design (i.e., in Pellet).  
No surprises have been reported to me in the, oh, probably 3 years,  
that Pellet has had that behavior. Hilog semantics have never been a  
feature request either. From anyone. No one's ever mentioned them at  
an OWLED.

Now, we can *add* that behavior (OWL 1.1 hilog-full) and we *would*  
get a divergence: OWL 1.1 punning would be sound but incomplete wrt  
the OWL 1.1 hilog-full semantics. This seems quite understandable.

> AIUI a Hilog-like semantics would be
> closer to what Full does.

Yes, but OWL 1.1 is *overall* closer to OWL Full and punning  
semantics are very easy to implement robustly. Hilog semantics are  
not. So if we want to improve the *effective* state of the art for  
users, we need to consider whether requiring this feature in all OWL  
1.1 profiles/species/what have you is worth doing and whether  
forbidding the slight weaker semantics is also worth doing. I'm not a  
huge fan of punning per se, myself, but it has some design advantages.

>> There are many graphs that are consistent in under simple
>> interpretations and inconsistent under RDFS semanatics. So
>> either this sort of being rather different is relatively
>> innocuous (at least ceteris paribus), or there's something
>> else that distinguishes them.
>
> The difference was the observation that it is desirable to have a  
> close
> relationship between DL and Full semantics, unlike simple vs RDFS
> semantics.

I don't see why it's undesirable in the simple vs. RDFS semantics.  
Syntactically identical graphs (you don't need to introduce any new  
vocabularly!) and under one reading consistent, under another,  
inconsistent. Pretty close to what we're proposing here. In fact,  
it's much more motivated here because that extra feature (hilog  
semantics) is 1) not used, 2) not wildly known, 3) when explained is  
not requested, and 4) has unknown implementation consequences.

Note that I'm not proposing to remove it altogether. OWL 1.1 Full can  
retain it. Nor am I arguing in favor of punning here (though I've  
raised some pro and con points wrt punning in this message). I'm just  
saying that this divergence is not sufficient to justify not having  
punning. It certainly isn't sufficient based on *a priori* grounds  
and I don't see you've presented any empirical evidence either.

Cheers,
Bijan.

Received on Wednesday, 15 August 2007 11:40:15 UTC