RE: OWL Full reasoning

Hans:
Many of the problems you are dealing with stem from peculiarities of OWL as
an ontology language; they are not a problem of using ontologies in general.
OWL might be suited for small and expressive ontologies, but has significant
disadvantages when it comes to deriving ontologies from huge, existing data
schemes, because it does not provide sufficient flexibility to capture the
original semantics of industrial standards (e.g. taxonomies) in a
straightforward manner.

I recommend the following paper (you should easily find it using Google):
Jos de Bruijn et al.: OWL DL vs. OWL Flight: Conceptual Modeling and
Reasoning for the Semantic Web, WWW 2005 conference.

Especially section 4 (pitfalls of owl) might be helpful.

I am currently working on a transformation of eCl@ss into OWL ontologies,
and some of the findings might be useful for your case, too. You can find
related information and papers at http://www.heppnetz.de/eclassowl.html

The following paper summarizes preliminary findings:
Martin Hepp: "A Methodology for Deriving OWL Ontologies from Products and
Services Categorization Standards",
Proceedings of the 13th European Conference on Information Systems
(ECIS2005), May 26 - May 28, 2005, Regensburg, Germany, pp. 1-12. You can
get it from http://ssrn.com/abstract=689184.

Basically I assume that in your case, you do not need the full expressivity
provided by OWL DL on one hand, but would want some more flexibility on the
other hand that OWL DL does not provide (for example the fact that
relationships between classes can only be annotation properties in OWL; you
cannot axiomatize them). Also, even for OWL DL, you will likely experience
performance problems with any existing reasoner when you go beyond say 1000
concepts, because I only know in-memory implementations that load the
complete ontology first. Also be careful with the non-intuitive usage of
domain and range in OWL and RDF-S; its usage implies class membership
instead of restricting the respective range, as somebody from the data
modeling community might suspect.

As an alternative, you may want to have a look at the WSML family of
ontology languages, http://www.wsmo.org/wsml. WSML Core should provide
everything you need.
 
Best wishes,
Martin
---------------------------
martin hepp
digital enterprise research institute (deri)
university of innsbruck
urls:
http://www.deri.org       deri
http://sebis.deri.org     sebis research cluster
http://www.heppnetz.de    personal page


________________________________

From: semantic-web-request@w3.org [mailto:semantic-web-request@w3.org] On
Behalf Of Hans Teijgeler
Sent: Donnerstag, 18. August 2005 22:36
To: 'Danny Ayers'
Cc: semantic-web@w3.org
Subject: RE: OWL Full reasoning



Danny,

See my responses below.

Regards,

Hans

 

_______________________ 

Hans Teijgeler

ISO 15926 specialist

www.InfowebML.ws <http://www.InfowebML.ws> 

hans.teijgeler@quicknet.nl <mailto:hans.teijgeler@quicknet.nl> 

phone +31-72-509 2005      

 

-----Original Message-----
From: semantic-web-request@w3.org [mailto:semantic-web-request@w3.org] On
Behalf Of Danny Ayers
Sent: Thursday, August 18, 2005 5:06 PM
To: Hans Teijgeler
Cc: semantic-web@w3.org
Subject: Re: OWL Full reasoning

 

 

On 8/18/05, Hans Teijgeler <hans.teijgeler@quicknet.nl> wrote:

 

It's good to see a practical problem as a use case. Ok, IANAL, so I

have more questions than answers here -

 

> 1.      Given the fact that this is clearly an implementation of OWL Full,

> can any OWL reasoner handle this at present? 

 

First question - have you determined the inferences you are looking

for are actually available through OWL Full semantics? 

[HT] To be brutally honest: I don't have a clue. Where can I read about that
in non-logician terminology?

 

My guess is that whether or not this is the case then the reasoning is
likely

possible with existing tools using a subset of OWL Full + rules  (e.g.

using  cwm, Euler or Pychinko).

[HT] I am not one of the OWLites, and keep having problems to find what to
learn and what to do. I tried every OWL Validator that I could find on the
W3C site, but the only thing these do is telling me that they have a parsing
problem, without telling me why that is. You guys still have a LOT to do in
order to make OWL a bit more userfriendly and accessible for the lesser
initiated. I'll try the three you mention (if I can find them).

 

Next questions - are you tied to this representation? Cannot the same

information be refactored to be OWL DL?

[HT] In our little world that is being debated. I am against it, because
that would mean denormalization, and we would lose too much of the
expressivity of ISO 15926, and lifetime stability (see below).

 

For example, you have "2" 150/300#RF Flanged Connection" as subClassOf

"ClassOfDirectConnection", yet an instance of "Valid Connection per

ANSI B16.5". Might it not be reasonable to make it a subClass of the

latter too?

[HT] Not if I want to stay in compliance with ISO 15926. A
ClassOfDirectConnection may only be used for things like "electrical
connection", "hydraulic connection", etc. In case you want to express things
like "flexible connection" then you have to use a metametaclass (that
ClassOfClass) (the data model is close to 5th Normal Form).

 

> 2.      If not yet, may we realistically expect such a capability to be

> available by the year 2010? 

 

I'll let a logician answer that. My guess is it's either "yes, now" or

"never" depending on whether sound & complete reasoning is possible

for the bits of OWL Full used.

My own deadlines tend to be closer than 2010 ;-)

[HT]  Any logician in the room?

 

> 3.      And if not, why is there OWL Full? 

 

My personal take is that RDF has proven its utility as a

representation language for the Web irrespective of the availability

of general-purpose reasoners. Going back to my earlier question - why

have you chosen this representation?

[HT] In the process industries (and in other industries as well) there is a
need to be able to:

-        share complex technical data-based information about everything
(from nuts and bolts to complete 20MM$ three-stage centrifugal compressors)
with everybody, to a great extent due to the globalization of the economy,
and the ever-stricter rules and regulations;

-        integrate lifecycle data, meaning all data for all stages of a
refinery or so (conceptual design to demolition), and for its entire
lifetime (which can be, for an oil well, 100 years)

-        hand over data from subcontractor and supplier to contractor, and
from contractor to plant owner/operator (the Exxons, Duponts, and Shells of
this world ) in order to fill their lifecycle data warehouses.

This means that the data model must be very solid, generic, and rigorous,
and that the medium (here OWL) may (and will) disappear before the end of
that lifetime. The data model has 201 entity data types only, and is
data-driven by means of what we call a Reference Data Library (the "Part 4"
balloons are in that library). Right now we have some 20,000 classes, and
when we are ready in 20 years or so we may have reached 100,000.

We worked on that data model since the late eighties (I joined in 1992), and
it became IS (ISO International Standard) in December 2003.

Since 2000 we worked on an implementation standard and based it on XML
Schema, for lack of something better at that time. I want to make the move
to OWL, because for the most it fits hand-in-glove.But I tell you, it ain't
easy for a data modeller to grasp the intricacies of RDF and OWL.

I'll get off my soap box..

 

Looking at RDF/S through OWL spectacles provides a formal basis which

can help answer questions like whether or not there will be a solution

to a problem by 2010.  But in the scenario you present it's not hard

to imagine an engine with hard-wired flange logic working on the data.

[HT] I figured that we will need to write code ourselves indeed. That still
leaves me with the question: What's the use of OWL?

 

One final question - do you happen to have this available as RDF/XML

or other serialization?

[HT] You can find the Ontology for Data Mode :under:

http://www.infowebml.ws/links/ontology-for-data-model.rdf

It is still not OK. The main things I still have to do is the disjoints and
at some places the LISTs (where OWL isn't very good in, I found). And then I
need to seek help in tweaking it in such a way that an OWL Validator accepts
it (Wonderweb accepted it as OWL Full, not as DL, because I'd have to do
away with the many blank nodes we have).

 

I will write OWL code for the problem I painted, and send it later (mind
you, until now I haven't found an OWL IDE, so I write the whole stuff by
hand, check it for XML well-formedness with XMLSpy (a leftover from my XML
Schema days), validate RDF with the W3C RDF Validation Service, and then I
get stuck. The Pellet guys, for example, want me to be a member (of what
they didn't tell on their screen), and just tell me that their moderator
will look at the error report that I filed. He/she is probably on vacation..

 

Cheers,

Danny.

 

-- 

 

http://dannyayers.com

Received on Friday, 19 August 2005 10:19:11 UTC