RE: [OWLWG-COMMENT] Defining the universal property in OWL-1.1

Hi, Carsten!

Carsten Lutz wrote on Sunday, November 11, 2007:

>Hi Michael,
>
>> This week in the OWL-WG mailing list, there was some discussion about
>> introducing the "universial property" into OWL-1.1.
>[...]
>
>I agree with what you say. Your encoding trick is a variation of what
>is often called a "spypoint" (an element that sees all other elements),
>see e.g. the paper "A Road-map on Complexity for Hybrid Logic" by 
>Areces, Blackburn, and Marx.

Oh, this idea even has an "official" name? :) In my previous mail about the
definition of cross products I called this individual a "glue instance",
because it "glues together" the properties in the sub property chain. :)

>Just one remark, namely that one should be a bit careful here. First,
>different ontologies using your encoding may use different names
>for the universal role. Second, even if they use the same name, they
>may use different names for the individual (spypoint) in the encoding.
>For this reason, you are actually not *defining* the universal role in
>a strict logical sense. I would maybe call it a projective definition.
>
>However, the point is that there are subtle differences. In 
>particular, 
>if you are interested in the reasoning task of ontology 
>entailment ("does
>one ontology entail another?") you run into trouble if the two 
>ontologies
>use different role/individual name (whereas this is no problem for
>satisfiability, subsumption, classification, etc). Regarding 
>the discussion
>on OWL-WG, this means that your construction cannot be used to 
>define true
>anonymous individuals.
>
>An obvious remedy is to have a convention for the name of the role
>which is defined to be universal. But then, that's the same as
>explicitly introducing a universal role into OWL, i.e., just what I
>have proposed.

Yes, of course.

One thing which concerns me a bit about introducing the universal property
'uprop' into OWL-1.1 is that it cannot be used in a straight forward way. As
you pointed out in one of your mails in the WG list, there are several
situations in which the universal property could not be used. And the
conditions, under which uprop can be used, are certainly pretty hard to
understand for "normal" ontology modelling people. For an ontology language,
which is intended to become a mainstream SemWeb language, this might be a
considerable practical problem. Because /if/ a language feature is available
officially, then people will start to use it in ways you would never expect.
:)

But this does /not/ mean that I speak against introducing uprop. Rather the
question here is: Will it be possible to /efficiently/ decide for a given
ontology, which uses uprop, if uprop is used in a proper way or not? So
reasoners, or better ontology editors, could give feedback to people if they
use uprop in a wrong way. Is this possible?

Otherwise, what would be the consequences if someone uses uprop in a
disallowed way, but does not get warned by his tools? As far as I understand
you, allowing unconstrained usage of uprop would make OWL-1.1 undecidable.
So it would be interesting to see what consequences this would have in
/practice/. 

In general, I am not very afraid of undecidability. To know that some
algorithmic problem is decidable just tells me that there is some algorithm
which terminates on /every/ input AND gives the correct answer in every
case. Well, it would certainly be not acceptable to get wrong answers, but
producing inputs on which my algorithm simply doesn't terminate doesn't
concern me much. Personally, I tell every process "non-terminating" which
makes me waiting on a result longer than drinking a cup of coffee. ;-) And,
of course, in practice I do not create /every/ possible ontology, but I try
to create my ontologies according to design patterns which have turned out
to be useful and working in the past. This reduces the input space
dramatically.

So an idea would perhaps really be to let people just play around with
uprop, and see if this leads to any serious problems in every day ontology
engineering. I know, this is a non-popular argument, but perhaps it is about
time to put the requirement for decidability into question - not already for
OWL-1.1, but perhaps for future versions of OWL. Just an idea. :)


Cheers,
Michael

--
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

Received on Monday, 12 November 2007 08:57:48 UTC