Re: Web Rule Language - WRL vs SWRL

> 
> On 1 Jul 2005, at 02:51, Michael Kifer wrote:
> 
> >
> >
> > Ian Horrocks wrote:
> >>
> >> One reason seems to be the following (potentially infinite) sequence  
>> of
> >> argumentation:
> >>
> >> Ian: DLP (with FO semantics as per the paper) and DLP-Datalog (i.e.,
> >> with Datalog semantics) are not equivalent (they have different
> >> models), but are indistinguishable w.r.t. entailment of ground DLP
> >> atoms. The can, however, be distinguished if we consider other kinds  
> >> of
> >> entailment (i.e., richer query languages).
> >>
> >> **
> >>
> >> Michael:  DLP and DLP-Datalog are equivalent if we restrict our
> >> attention to entailment of ground DLP atoms.
> >>
> >> Ian: Yes, they are *indistinguishable* if we restrict our attention to
> >> entailment of ground DLP atoms, but not if we consider other kinds of
> >> entailment (i.e., richer query languages).
> >>
> >> Repeat from **
> >
> > Good. At least we agree on where the cycle is.
> >
> >>> 2. The claim is that both stacks extend DLP *as languages* --
> >>> semantically
> >>>    and syntactically (the latter after some mapping, of course).
> >>>    We use the standard definition of language extension here (which I
> >>>    have spelled out in a previous message).
> >>
> >> Your "standard definition" assumes a common underlying semantic
> >> framework, which is not the case here.
> >
> > Good. So now we agree that LP does extend DLP modulo the "common  
> > underlying
> > semantic framework"; see below.
> 
> It extends DLP-Datalog, which has the same syntax as DLP, but a  
> different semantics!
> 
> Repeat from **
> 
> >
> >> E.g., consider DLP, OWL-Lite,
> >> OWL-DL, SWRL and FOL. All share the same set of models; the only
> >> difference is the richness of the language that is available for
> >> restricting the set of admissible models. A DLP theory (set of rules)
> >> would have the same set of modes if treated as OWL, SWRL or FOL. As we
> >> have seen, this is *not* the case for DLP and Datalog.
> >
> > But you agree that there are embedding homomorphisms (both over models  
> > and
> > syntax), which form a commutative diagram. This is good enough for me.
> > I did explain this in my email where I defined what we mean by  
> > extension.
> >
> > As far as I am concerned, this difference is a trivial matter. Embedded
> > objects are commonly identified with their images in mathematics.
> > If you think that this embedding is a big deal, you are entitled to  
> > disagree.
> 
> This seems to me to be an example of "Proof by obfuscation" (see  
> http://www.maths.uwa.edu.au/~berwin/humour/ 
> invalid.proofs.html#1.7Proofbyobfuscation). For my part, I have  
> provided a counter example to the equivalence theory. I rest my case.

Your "counterexample" query uses language constructs that are not in DLP.
I gave a precise definition, while you are obfuscating things by bringing up
fallacious arguments. 


> >>>>> People think databases but use OWL.
> >>>>> Now, this is a real semantic mismatch: people mean (and want!) one
> >>>>> thing,
> >>>>> but get a completely different thing.
> >>>>
> >>>> You need to get out more. I meet and interact with many users who  
> >>>> are
> >>>> building large and complex ontologies, and who have a very good  
> >>>> grasp
> >>>> of the semantics of OWL. Hopefully you caught the presentation from
> >>>> Christine Golbreich at the rules workshop and have been keeping up
> >>>> with
> >>>> the work being done in the SWBP working group. You should also look  
> >>>> at
> >>>> what is being done by NCI, SNOMED and the Gene Ontology Consortium
> >>>> amongst many others.
> >>>
> >>> Thanks for the advise.  I am trying to get out whenever I can. And  
> >>> you
> >>> should also try to visit places outside of your immediate  
> >>> neighborhood.
> >>>
> >>> I am not involved in any of the aforesaid great projects, but
> >>> occasionally
> >>> I do run into interesting articles
> >>>     http://ontology.buffalo.edu/medo/NCIT.pdf
> >>>
> >>> http://www.ipsi.fraunhofer.de/orion/pubFulltexts/NCIReview18Feb04.pdf
> >>> which raise questions about the use of OWL for NCIT.
> >>
> >> Of course there will always be questions and criticism aimed at any
> >> large project of this kind, but the fact is that many people *are*
> >> using OWL, and that many of them have a very good understanding of the
> >> semantics of OWL. It is condescending, not to say insulting, to  
> >> suggest
> >> otherwise.
> >
> > I didn't realize that those papers are insulting.
> 
> Perhaps you find it amusing, but I don't believe that much is  
> contributed to the discussion by your pretending I said something which  
> I clearly didn't. If you read the above text, you will see that what I  
> say is that it is "condescending, not to say insulting" to suggest that  
> most/all users of OWL do not understand the semantics of the language.


But who said that "most/all users of OWL do not understand the semantics of
the language"? Many -- perhaps. But I didn't even say that!


> > I thought that they are
> > quite thoughtful and point to real problems. What those papers say is  
> > that
> > people do make mistakes even if they understand the semantics. It is  
> > just
> > too hard to state certain things correctly, and some things are  
> > possibly
> > missing.  All this is even more telling given that the NCIT project  
> > enjoys
> > the benefit of substantial involvement of outside KR experts.
> >
> > When I "go out," I sometimes run into people who don't enjoy those  
> > benefits
> > and see that they are grappling with much more basic things (typically
> > thinking and wanting database-style semantics---read CWA---while using  
> > OWL).
> > Cardinality restrictions is usually the misunderstanding #1.
> 
> Of course such people exist too - I claimed the existence of people who  
> use and understand OWL, not the non-existence of any other kinds of  
> people.


Who claimed that there are NO people who use and understand OWL?
(Someone named "Ian" would have been an obvious counterexample.)
You are making bizarre statements tonight...


> >>>>> SWRL is complex in a wrong way. How many people use it in a way
> >>>>> where programs come out substantially non-Datalog? (I.e., where one
> >>>>> can't accomplish the same thing in an LP-style language in a  
> >>>>> simpler
> >>>>> way?)
> >>>>
> >>>> See my comment above - more than are drempt of in your philosophy.
> >>>
> >>> We are talking about SWRL here, not OWL, are we?
> >>> Can you please point me to the use of SWRL in NCI, SNOMED, etc.,  
> >>> which
> >>> requires the complexity of SWRL? (Something that goes beyond, say,  
> >>> the
> >>> simple AL-Log?)
> >>
> >> Now it is *my* turn to suggest that *you* haven't been paying
> >> attention.
> >
> >
> > Appreciate your help with keeping my attention focused.
> >
> >
> >> The presentation by Christine Golbreich at the rules
> >> workshop included several such examples. In medicine (and in many  
> >> other
> >> domains where complex structures must be modelled) it is often
> >> important to capture situations such as "a fracture of the shaft of  
> >> the
> >> femur is a kind of a fracture of the femur". This can be expressed in
> >> SWRL (using of axioms asserting complex relationships between binary
> >> predicates) but not in OWL.
> >
> > I didn't say "show me the benefit of SWRL with respect to OWL."  I  
> > said,
> > show me a use case "where one can't accomplish the same thing in an
> > LP-style language in a simpler way" and "point me to the use of SWRL  
> > ...
> > which requires the complexity of SWRL? (Admittedly, by mentioning  
> > AL-Log I
> > was aiming too low.)
> >
> > With no intention to detract from Christine's excellent project, I  
> > must say
> > that the use of the rules there is LP-trivial. My aforesaid request was
> > to see a realistic use case where people would need to use the truly
> > complex parts of SWRL, such as descriptions in the head.
> 
> I don't know what you mean by "LP-trivial". Do you mean that you could  
> express the same thing in Datalog?
> I don't think so.

LP is not Datalog. Datalog is a *trivial* part of LP. It is certainly easy
to express the axiom below in LP (together with the rules sitting on top).
Of course, the semantics will be nonmonotonic and so some answers will be
different. For instance, if you ask something like in your favorite example
(whether certain cardinality = 1) then you might get a "yes" because this
is so in the current state, while you will get a "no" in SWRL. This may or
may not be important or desirable. In cases where this is important and
undesirable, e.g., when you want to query the schema, there are ways to
deal with this in some versions of LP. For instance, in LP based on F-logic.

> E.g., consider the  
> class Hemisphere defined in the paper using the following axiom:
> 
> Hemisphere := Anatomical Entity ∩ (∀ hasDirectAnatomicalPart Lobe)
> ∩ (= 1 hasDirectAnatomicalPart FrontalLobe) ∩ (= 1  
> HasDirectAnatomicalPart
> ParietalLobe) ∩ (= 1 hasDirectAnatomicalPart OccipitalLobe) ∩ (= 1
> hasDirectAnatomicalPart LimbicLobe) ∩ (= 1 hasDirectAnatomicalPart
> TemporalLobe)
> 
> Regarding "the truly complex parts of SWRL, such as descriptions in the  
> head", I'm not sure what you mean by this. The SWRL rules given in the  
> paper are used *in addition* to axioms such as the one above, and  
> capture additional property relationships that cannot be expressed in  
> OWL. In the general case, such rules lead to undecidability when  
> combined even with OWL-Lite. Is that complex enough?

Ian, you are not following. The discussion was about SWRL, not OWL.
I was asking if you have examples of interesting uses of the more advanced
features of SWRL (not OWL) where complex rules are used. You pointed me to
this paper, but the rules there are very simple. So, this didn't satisfy my
curiosity. (Again, nothing against Christine's great project.)


	--michael

Received on Friday, 1 July 2005 22:31:39 UTC