Re: plain rules, please [was: Semantic Web Rule Language (SWRL) 0.5 released]

On Wednesday, November 26, 2003, at 07:59  AM, Sandro Hawke wrote:

>> On Wednesday, November 26, 2003, at 01:41 AM, Sandro Hawke wrote:
>> [snip]
>>> I can't speak for everyone on the committee, but I went along with
>>> this because (1) it's important that the rule language be usable in
>>> concert with OWL, and (2) several members of the JC were skeptical
>>> that a rule language can be specified orthogonal to OWL and then
>>> combined with it, while maintaining the proper semantics and desired
>>> reasoner performance characteristics.
>>
>> There is a sense, I think, in which OWL rules is orthogonal to OW (I
>> take it's combinability to be clear). There's no difficulty in using
>> OWL Rules with "plain" RDF,
>
> Good point.    Some of the expressiveness (eg existentials in the
> consequent) require OWL constructs, though, as I understand it.

Yes, the orthogonal rules language might be less expressive than you 
want. You might be able to recover some of that by letting in only bits 
of OWL (though, you'd be wise to be extra careful not to repeat the OWL 
Lite mistake of making syntactic restrictions that don't restrict 
expressive power).

>  i.e., atoms restricted to using primitive
>> classes with no defined TBox or role hierarchy. It'd still be darned
>> expressive (e.g., you wouldn't really be able to eliminate various 
>> TBox
>> axioms and role hierarchy axioms since they are trivially expressible
>> as OWL Rules).
>>
>> The is probably not the orthogonality people wanted :)
>
> Yeah -- even if the vocabulary can be used by itself, the
> specification is not orthogonal.

Erhm? You mean that the current OWL rules proposal document builds on 
the S&AS document? This is, I would think, a radically non-harmful form 
of non-orthogonality. Actually, I think it's a very good one. I'm 
really quite astonished that you think the desired form of 
orthogonality is *specification* orthogonality. It should be trivial to 
achieve specification orthogonality! (Why would you want to? And if you 
did, you'd be adding annoying problems for combining them.)

(Ok, I just blew a comprehension fuse. I *must* not be understanding 
you!)

>>>  I think the sense was that such
>>> orthogonality "might be nice" or even "would be nice", but until the
>>> details can be worked out, this 0.5 version seemed like a good
>>> waypoint.  A future language with orthogonality should be able to be
>>> compatible with this one, so letting folks begin coding to 0.5 seems
>>> reasonable.
>>
>> Compatible in what sense? I mean, "well, the orthogonal language 
>> needed
>> to be entirely disjoint with OWL, and heck, it's also disjoint with 
>> OWL
>> rules!" or "Once we figure out how to be usefully orthogonal and yet
>> combinable with OWL, we'll find that we've subsumed OWL rules or at
>> least fit in nicely with them"?
>
> In the latter sense.

Ouch, there goes another fuse :) (Why not just extend OWL rules?)

>> Personally, I don't yet have even a vague sense of this :(
>
> I have a strong sense of it, but as I'm sometimes reminded, that
> doesn't mean I'm right.  I think of RDF-based logics (RDFS, OWL, SWRL,
> ...) in terms of FOL axioms on sentences like "rdf(s,p,o)".  I think
> this approach is suggested by LBase, and it's what Surnia (my quick
> hack OWL implementation) does.  From this angle, orthogonality and
> combinations of logics are fairly straightforward.

I didn't believe that even before Peter's post :)

> The downsides to this approach, I think, are (1) it doesn't really
> tell you how to make practical implementations, just experimental ones
> (eg throw it at a theorem prover); (2) the axioms are software, and so
> unlikely to be perfectly correct.   (I'm not sure I buy #2, but I hear
> it a lot.)

Hmm. On the one hand, (1) seems pretty nasty. OTOH, that's one thing 
said for OWL rules :) I think the stronger point is that such 
combinations are likely to be resistant to practical implementations. 
Of course, the evidence for this stronger point is rather thinner :)

As for 2, forget that the axioms are software (though, yes, let's 
hammer that at some point), the challenge is showing that the axioms, 
especially in aggregate, express the semantics of the logics, 
especially in combination. I'd love to see some example proofs for 
existing semantic web languages (this should be somewhat easier as 
they've been heavily designed, nipped, and tucked to align).

Cheers,
Bijan Parsia.

Received on Wednesday, 26 November 2003 16:02:16 UTC