RE: A proposal for the fragments document

Hi Jim!

(since I am cc'ed...)

I remember that you have already put a related question in our last week's telco w.r.t. pD*. I admit, pD* suddenly entered the discussion without any announcement and without any deeper explanation on why it was brought into play. And now, there even exists a second proposal for a rule language (OWL-R-Full), which is apparently pretty different from Zhe's and your original suggestion, which was OWL-Prime. So I will give it a try and present my own view (which might be biased, I suppose, so be warned! :)).

Remember, a few weeks ago I have sent a mail [1], in which I discussed the following preliminary idea: Similar to the approach followed by the RDFS semantics spec, we might try to define a triple rule set, which is semantically equivalent (in a certain sense) to at least some sub semantics of OWL-Full. At that time, I was pretty certain that it is *not* possible for whole OWL-Full, but that it *might* be possible at least for OWL-Prime (I knew that it *was* possible for RDFS). Later, Jeremy answered to this mail that it was probably Hermann ter Horst's initial idea to make his language pD* into such a kind of language. So I skimmed through the respective paper [2], and I came to the conclusion that pD* has several interesting features for our purpose:

  * It is an extention of RDFS model-theoretic semantics, and semantically a sub language of OWL-Full. This was intended for OWL-Prime to hold, too.

  * It covers syntactically all of the proposed OWL-Prime language features.

  * It is a triple-rule language, with a focus on rulebased reasoning on instance data.

  * Both, the model-theoretic semantic conditions and the rules are IMO pretty easy and straightforwardly defined. I don't see any significant "costs" in exchange for more generality in this respect. So there seems to be no advantage from this pov to use a smaller language such as OWL-Prime.

  * pD* is "exactly as fast as" RDFS regarding the two reasoning tasks consistency-checking (polynomial), and checking whether a certain entailment holds or not (polynomial when bNodes are regarded as skolems, which will be typical for implementations; NP-complete in the case that bNodes are seen as existential variables, but even then an important subset of entailment queries is still polynomially decidable). Again, from this pov, OWL-Prime in the original proposal wouldn't bring any advantages, since OWL-Prime is an upper language of RDFS.

  * The ruleset is complete in a certain sense: One can, based alone on the rules, always find out whether a pD* ontology is model-theoretically satisfiable or not, and if it is satisfiable, then one can "simulate" every entailment query in a "reasonable way" (ok, that's vague, but might suffice for this mail). (And @Peter: I disagree that "completeness" does necessarily mean that a rule system has to provide Ex Falso Quodlibet. :))    

  * pD* "exists", i.e. it is well defined and analyzed, both its model-theoretic semantics and its ruleset. It was one of the "soft" requirements in an earlier telco that we, the WG, should not spend too much time on research on a new language. And pD* can just be used "out of the box".
 
Ok, this is why I am currently in favour of pD*: It seems to me that I get from pD* all you want for OWL-Prime, and more than this, and nearly without having to pay for this "more". And I can get this all right now! (I promised that I will give a biased view, didn't I? ;-)) 

But if you really want OWL-Prime, then, AFAICS, it will be easy to restrict both the set of model-theoretic semantic conditions and the set of triple rules to just match the OWL-Prime features. Note, however, that pD* is not very much larger than OWL-Prime, so I would believe that this would be a questionable step.

This brings us to the newly proposed language OWL-R-Full, which has an obvious advantage over pD*: It covers significantly more language features, providing for example boolean constructs (intersection and union), sub property chains, and the new OWL-1.1 property characteristics. It shares with pD* the property that it is semantically a language somewhere between RDFS and OWL-Full, and it is an upper language of OWL-Prime.

But I did not yet find much time to look deeper into this new proposal, although I am very interested. As a next step I plan to do a closer comparison between pD* and OWL-R-Full for my personal understanding. However, I will not be able to check for all the important things ter Horst has checked for pD*, i.e. questions about computational complexity, completeness of the rule system w.r.t. to the model-theoretic semantics, giving a satisfiability criterion for the language, etc. (Because, to recall what was said in that former telco, I, as a WG participant, should not spend too much time on language research. ;-) -- no, honestly, I don't think that I would be able to do this.)

Ok, fun aside, I hope my explanation will at least help you to understand why some people are talking about pD* or OWL-R-Full, when they rather should talk about OWL-Prime. :)

Cheers,
Michael

[1] <http://lists.w3.org/Archives/Public/public-owl-wg/2008Feb/0077.html>
[2] <http://linkinghub.elsevier.com/retrieve/pii/S1570826805000144>

>-----Original Message-----
>From: Jim Hendler [mailto:hendler@cs.rpi.edu] 
>Sent: Thursday, March 13, 2008 1:55 AM
>To: Boris Motik
>Cc: Michael Schneider; public-owl-wg@w3.org
>Subject: Re: A proposal for the fragments document
>
>I actually have some questions about the choices of features in the  
>fragment, and some things left out (most of the non-semantic features  
>- like imports and annotations, which I think are important to the  
>community).  I want to make clear that the proposal that I've been  
>backing from the beginning is not to create the maximal set of OWL  
>that can be expressed in rules, but rather a set designed with many  
>different factors in mind (including ease of implementability 
>with the  
>syntax, so I'd prefer not to include restrictions).  The 
>original RDFS  
>submissions, and the rule set we submitted are what we've proposed.
>  At this point I'm not trying to have this disucssion, but to 
>ask how/ 
>where we should have it.  We decided not to have the TF, but then I  
>discovered there'd been some extensions to the first version I'd seen  
>(my fault, not yours) and thus I do want to have some design  
>discussion - may not need the whole WG .
>  -JH

--
Dipl.-Inform. Michael Schneider
FZI Forschungszentrum Informatik Karlsruhe
Abtl. Information Process Engineering (IPE)
Tel  : +49-721-9654-726
Fax  : +49-721-9654-727
Email: Michael.Schneider@fzi.de
Web  : http://www.fzi.de/ipe/eng/mitarbeiter.php?id=555

FZI Forschungszentrum Informatik an der Universität Karlsruhe
Haid-und-Neu-Str. 10-14, D-76131 Karlsruhe
Tel.: +49-721-9654-0, Fax: +49-721-9654-959
Stiftung des bürgerlichen Rechts
Az: 14-0563.1 Regierungspräsidium Karlsruhe
Vorstand: Rüdiger Dillmann, Michael Flor, Jivka Ovtcharova, Rudi Studer
Vorsitzender des Kuratoriums: Ministerialdirigent Günther Leßnerkraus