Re: OWL Lite vs OWL DL-Lite

a few background points here and then details below:

I am historically clearly a DL person although I am also someone who has put a lot
of applications in practice (both dl and frame-inspired).  I see great value in DLs
and continue to support work on dls both theoretically and in practical
applications. I am also aware of the issues that arise when putting DLs into
practice.
The applications work is what makes me stand VERY strongly on the lite soapbox.
ultimately as long as we have one lite version i will survive.   OWL DLite is the
lite that i think is more important of the two proposed.
the point below is that i think there is value in owl flite as well but i will not
hold up our process by trying to argue this point beyond this message.
I just wanted to go on record that i think some will find value in fLite.
some support below:

Jim Hendler wrote:

> Ian -
>    It becomes more clear to me with every message that your world view
> and mine are so diverse that we will nevereven  be able to
> communicate about some of these issues, let along agree.  At the same
> time, I am encouraged by the fact that we have a language design that
> we both find livable - it leads me to optimism that if we can ever
> get this thing out the door, lots of people will find it usable.
>   So, you can believe me or not, but building a powerful tool which
> does useful web tasks using the Lite vocabulary, and provides useful
> answers much of the time, is easier than building one that has the
> same performance against the full OWL Full vocabulary.

This is the position that I would support as well.
In my original email embedded below where i state that i support ian's examples as
cases where owl dlite is useful i tried to say that the above case i have seen
being of use as well.
I was thinking in particular of cases such as the work on ontobuilder by vertical
net.
that was a new project that wanted to provide a simple modeling environment for
building ontologies to be used in initially lightweight applications.
If OWL flite was available when that project started i would have considered that
as the language to support.  we took a stripped down version of xol for the
project.
the project did not expect to require class subsumption determination and while
complete reasoning would have been convenient, in reality it was deployed without
complete reasoning and was found to be valuable.  (thus i claim it fits the
characterization of providing useful answers most of the time).

similarly i was thinking of ontolingua (without its connection to jtp).  Ontolingua
includes fairly expressive representational capability and makes no claim about
complete reasoning.  It does some reasoning by itself (and connects to jtp when it
wants more complete reasoning).  It makes no claims about completeness.
It has been widely used for a number of years.  When I looked at some of the
ontologies built in ontolingua previously i noticed a number of applications that
used very simple representational power.  Thus i believe that there are some number
of users who are interested in modeling fairly simple class and individual
descriptions with some simple checking.

Similarly although i am less expert in protege, i have interacted with the protege
group on numerous occasions.  My understanding of protege is that it also does some
checking on its input but i do not believe that it makes claims about completeness
of its reasoning.  Protege has a very large user base so presumably one can assume
that a significant number of people find value in it as a tool.  My belief is that
one of protege's perceived strengths is that it is fairly easy to use.  The protege
team appears to keep this as a high priority.  They specifically ask questions
about "how will i explain that concept to my naive users" when faced with a
decision to include something new.
Thus, I expect that they would prefer not to have to explain things like the
limitation on max number restrictions on transitive roles that is in place on owl
dl.
I presume that a migration path for them that makes sense would be to start with
supporting the ability to input the constructors in lite without having to explain
the limitations required by owl dl.

i was also thinking about an ontology evolution environment that i helped a large
company build internally.  for political reasons, they wanted to build their own
system.  They had similar needs in that initially they had fairly simple
representational needs and could get away with some reasoning.  If they were
starting today they might have chosen a path that would start with owl flite and
move to owl full if they had more expressive needs or introduce the additional
restrictions of OWL Dlite and OWL DL when connected to a more complete reasoning
component.
It is not atypical to allow input of terms without complete checking and then put
in the complete (or just more complete) checking in another module.

one other note worth considering is loom.  it has existed for many years and has
had a significant user base.  I not seen any claims about it having a complete
reasoner and at best one could hope to get a mostly complete operational
specification of the reasoning that it performed.  Still it has been used in a
number of applications over many years.  I have less experience with it so I am not
sure how many of those applications use a very expressive language but it may be
considered to be another example of a system that gets useful answers most of the
time.

> That said, I
> don't think we need to give this a specific name, and I'm satisfied
> with how it is currently dealt with in our documents, so I can live
> with the status quo and only address this if we need to during the
> comment period.  Thus, I see no point in us arguing for the sake of
> arguing.
>   -JH
>
> At 18:49 +0000 2/17/03, Ian Horrocks wrote:
> >On February 16, Jim Hendler writes:
> >>
> >>  At 23:01 +0000 2/16/03, Ian Horrocks wrote:
> >>  >On February 13, Evan Wallace writes:
> >>  >>
> >>  >>
> >>  >>  Deb McGuinness wrote regarding OWL F Lite and OWL DL Lite
> >>  >>
> >>  >>  >I think ian's examples are valid real world examples of usefulness
> >>  >>  >of OWL Lite DL.
> >>  >>  >
> >>  >>  >Essentially they are characterized by an application being able to
> >>  >>  >take advantage of a reasoner's ability to classify descriptions
> >>  >>  >correctly.  this requires iff semantics.
> >>  >>  >
> >>  >>  >Similarly I think there are users who come more from a modeling
> >>  >>  >orientation who would like a simple transition path up from rdfs
> >>  >>  >and would benefit from an OWL lite that does not require them to
> >>  >>  >understand the limitations imposed by DL.
> >>  >>
> >>  >>  Speaking as someone who comes from the world of modeling (which I
> >>  >>  take to mean data modeling and object modeling), I don't see this
> >>  >>  at all.  People who are currently modeling using NIAM/ORM, EXPRESS,
> >>  >>  or UML who are considering using Semantic Web languages will go
> >>  >>  directly to OWL DL Lite or OWL DL (or perhaps to DAML).  At OMG,
> >>  >>  we are specifically asking for a mapping to OWL DL in our RFP for
> >>  >>  Ontology Definition (an OMG version of a UML Presentation syntax
> >>  >>  for OWL).  This is because the motivation for moving models to
> >>  >>  ontology languages is to attain a practical capability for reasoning
> >>  >>  about those models.  I am not denying that there is a constituency
> >>  >>  for a migration path from RDFS to OWL F Lite, just that I haven't
> >>  >>  encountered it in OMG, ISO STEP, or other communities doing
> >>  >>  manufacturing, business, or systems modeling.
> >>  >
> >>  >Good to hear you say that. This has also been my experience in the
> >>  >bio, medical and e-science communities.
> >>  >
> >>  >>
> >>  >>  Speaking for Two Dimensions in Lite
> >>  >>
> >>  >>  One public relations issue that I have encountered regarding OWL is
> >>  >>  the perception that it is not worth looking into the language because
> >>  >>  it contains a union of the problems, limitations, and disliked features
> >>  >>  of RDF(S) and DLs (many of which are actually mis-perceptions, outdated
> >>  >>  understandings, and/or prejudice).  This suggests to me a good reason
> >>  >>  for adopting a two dimensional structure for the OWL sublanguages.  It
> >>  >>  would show clearly that there is a partial decoupling in the language
> >>  >>  between RDF and Description Logics that allows users to pick the
> >>  >>  sublanguage with the features they need, rather than being limited to
> >>  >>  one design compromise in merging the two.
> >>  >
> >>  >The trouble with this is that it is VERY hard to see the justification
> >>  >for the existence of OWL F Lite - it is in no sense Lite, and the loss
> >>  >of a couple of constructors w.r.t. OWL full is of trivial significance
> >>  >compared to the semantic and computational complexities you get by
> >  > >mixing RDF with a standard logic.
> >>  >
> >>  >Ian
> >>
> >>  [probably chair hat off so as not to offend anyone by my having an opinion]
> >>
> >>  Very Hard??? We heard several implementors get up at the f2f and say
> >>  they would like to do Full semantics limited to the Lite vocabulary,
> >>  and we've seen similar on our public comments list and in the
> >>  feedback we've gotten from RDF Core people -- so maybe as Pat has
> >>  said there are two different communities here.  For what it is worth,
> >>  not only do I find it not hard at all to see the justification but,
> >>  in fact, the data seems to agree -- that is, many of the things using
> >>  DAML or OWL right now are tools which add some Lite constructs to
> >>  RDF-Schemas, thus "enhancing" the schema with more of our
> >>  expressivity.
> >
> >None of what you just said seems to me to be a justification for
> >having OWL Flite. What it does suggest is that some people may want to
> >use OWL full, but would be happy with (very) incomplete
> >implementations.
> >
> >>   And, btw, the implementation simplicity argument
> >>  (which you are on record as saying is the only one that matters)
> >>  still holds - it is much easier to implement the Lite constructs than
> >>  the whole kit and kaboodle.
> >
> >In what sense is it easier? I presume you don't know of a decision
> >procedure or even if one is possible. If we therefore assume we are
> >talking about incomplete reasoners, then an OWL Flite reasoner would
> >be an OWL full reasoner, so it wouldn't be any easier to build one!
> >
> >So, from a users point of view, they should be perfectly content just
> >to use (part of) OWL full, and from an implementors point of view, why
> >would they want to claim that they had implemented an OWL Flite
> >reasoner when they could, with zero additional effort, claim to have
> >implemented an OWL full reasoner?
> >
> >Similarly for editing tools etc. OWL full and OWL Flite documents
> >would be completely interchangeable - the difference is only in the
> >semantic interpretations placed on some of the RDF syntax. So, an OWL
> >Flite tool might as well claim to be an OWL full tool.
> >
> >>    In fact, however, we need not really argue, because we have already
> >>  created Owl Flite (it is simply using the vocabulary for Lite, as
> >>  defined in the Overview, without the extra cosntraints of DL).  Thus,
> >>  we are only talking about doing something expository.
> >>    We can do it now, or wait until we get asked to do it in the
> >>  requests from our LC -- it is an "editorial" change, not a technical
> >>  one, so I'm not too worried about it
> >
> >The question is whether we want to have the added confusion of
> >additional officially labelled languages in a non-linear family
> >relationship. It is hard to see why we should do this in order to
> >introduce a new language that is to all intents and purposes
> >indistinguishable from OWL full.
> >
> >Ian
> >
> >
> >>    -JH
> >>
> >>
> >>  >  >
> >>  >>  -Evan
> >>  >>
> >>  >>  Evan K. Wallace
> >>  >>  Manufacturing Systems Integration Division
> >>  >>  NIST
> >>  >>  ewallace@nist.gov
> >>  >>
> >>
> >>
> >>  --
> >>  Professor James Hendler                               hendler@cs.umd.edu
> >>  Director, Semantic Web and Agent Technologies         301-405-2696
> >>  Maryland Information and Network Dynamics Lab.        301-405-6707 (Fax)
> >>  Univ of Maryland, College Park, MD 20742      240-731-3822 (Cell)
> >>  http://www.cs.umd.edu/users/hendler
> >>
>
> --
> Professor James Hendler                           hendler@cs.umd.edu
> Director, Semantic Web and Agent Technologies     301-405-2696
> Maryland Information and Network Dynamics Lab.    301-405-6707 (Fax)
> Univ of Maryland, College Park, MD 20742          240-731-3822 (Cell)
> http://www.cs.umd.edu/users/hendler

--
 Deborah L. McGuinness
 Knowledge Systems Laboratory
 Gates Computer Science Building, 2A Room 241
 Stanford University, Stanford, CA 94305-9020
 email: dlm@ksl.stanford.edu
 URL: http://ksl.stanford.edu/people/dlm
 (voice) 650 723 9770    (stanford fax) 650 725 5850   (computer fax)  801 705 0941

Received on Monday, 17 February 2003 17:30:27 UTC