[wg11-owl] Goal of the review

[Ed Barkmeyer]

David Price wrote:

> As a start, there are multiple ways in which OWL might be used by SC4. As
> OWL is built on RDFS and RDF, it may also be appropriate to consider some
> use of those languages as well.

We do need to talk about RDF, not because SC4 uses it, but because OWL 
does, and because certain SC4 applications might use it in lieu of, or 
in addition to, OWL.  RDF is essentially a general-purpose knowledge 
representation format with very few pre-defined semantics.  Knowledge is 
a set of triples (Subject, Verb, Object), and there are a handful of 
predefined relationships (verbs), which include is-an-instance-of (IN), 
is-a-kind-of (SUBTYPE), and IMPLIES, AND, and OR.  The rest is 
roll-your-own.

The emerging W3C SDAI-like stuff is aimed at RDF, with possible OWL 
specializations.

> Seems like there are at least four uses:
> 
> 1 - As a language for defining general purpose, universal semantic
> repositories similar to what 15926 provides for the Oil and Gas industry.

The intent of OWL, as I understood it, is to be a language for defining 
the information content of a Web resource (which is what RDF was for). 
If "defining universal semantic repositories" means defining a model of 
all the objects and properties that are interesting in a particular 
industry area from a particular perspective, then, yes.  OWL, EXPRESS 
and SQL are languages in which you can do that, and each has its own 
strengths and weaknesses.

> 2 - As a language for defining taxonomies to be used with the External_class
> modules in STEP, etc.

Yes!  Properly used, OWL provides a means of *defining* taxonomies in a 
way that lets one reason about the classification of an object from its 
known properties.  One can represent these ideas in EXPRESS, but nothing 
in Part 11 defines that kind of interpretation.  Unfortunately, the 
limitations of OWL as a "rules language" will greatly limit the possible 
specification of classification rules.

This approach could also be used to revamp the IRMs, or to "define" SC4 
"module" MRMs as well.

> 3 - As a modeling language for domain-specific ontologies (i.e. schemas) as
> done in STEP today, Mandate, etc.

Yes, but in that regard, OWL is more abstract than EXPRESS and generally 
less competent.  Its primary advantage is the one mentioned in (2) 
above.  Most of the EXPRESS rules language has no rendering into OWL. 
Some careful RDF extensions to OWL, particularly in the area of 
mathematical functions, would allow us to capture the principal ideas 
that EXPRESS rules state, although rarely in the way they are 
conceptualized in EXPRESS.  (But I hold no hope that SC4 would have the 
competence to develop those extensions.)

> 4 - As an implementation language with the concepts of Class, Property and
> Individual, standard XML and other encodings, industry query languages, and
> industry APIs.

The standard "implementation" ("exchange form") of OWL is a collection 
of RDF triples.  I assume that is what David means.  Query languages and 
interfaces like KQML are not OWL-specific, and SPARQL is RDF-specific, 
not OWL-specific.

Yes, if you render EXPRESS entities into OWL classes, and EXPRESS 
(explicit) attributes into OWL properties, then you can do data exchange 
with OWL-defined RDF triples.  And some knowledge engines will be better 
prepared to consume that form than Part 21 or XMI.  And if you are 
building knowledge repositories whose interface is KQML or SPARQL 
instead of SDAI or SQL, you will want the OWL-RDF format.  Now as to 
"industry" query languages and "industry" APIs, the current state is 
competing draft standards implemented as academic hacks fronting 
academic engines, with a few fledgling EAI-like products emerging.

But from the point-of-view of moving information into and out of 
manufacturing software applications as we know them, OWL-RDF is just yet 
another ugly data form.

I don't deny the potential here, but the state of the practice in 
"implementation" in OWL land is SC4:1992.  In 5 years, half of the 
standards may be supported by niche products.  But there will be more 
tools than industrial applications.

> 1, 3 and maybe some cases of 4 enable the use of reasoners, etc. too.

Well, I don't really see how (3) differs from (1).  All useful 
ontologies are domain-specific; the question is only the breadth of the 
domain and the hubris of the modeler(s).

2 and 4 enable the use of certain kinds of reasoners.  Translating 
EXPRESS to other logic languages (e.g. CL) would enable the use of 
different (and less predictable) kinds of reasoners.  What kinds of 
reasoning does SC4 need to enable?

> Anyone think of any other uses?

None of the above is a "use".  A use is when you know what you will 
populate an OWL model with and what you will do with the results.
And as David says:

> To my knowledge, OWL is not "deployed" anywhere in SC4.

I.e. there is no known use, or intended use.

> It is being used in OASIS standardiziation for use 2 with AP239.

And when some other organization defines an AP239 specialization using 
OWL models consistent with the registered taxonomies, and builds tools 
that perform real "industry" application tasks using data exchanged via 
those models (or front-ends for existing applications that convert such 
data), then we will have a "use".  And then we will (maybe) know whether 
an EXPRESS/Part21 exchange, or a UML/XMI exchange, or an XML schema/XML 
exchange would have had exactly the same "use", or whether, in fact, the 
"use" actually involves reasoning about classifications.

Frankly, I think OWL/XML will be used in a lot of industry exchanges 
over the next 5-7 years, but as a 1-for-1 replacement for 
EXPRESS+Part21, i.e. concept model + data exchange representation.  So 
the major concerns for SC4 are:

(1) Is OWL/XML a better choice of migration path to "mainstream" than 
XML Schema/XML?  Alternatively, is it another necessary choice, like UML 
for presenting SC4 models?

(2) Are there any current, immediate SC4 projects that will develop real 
classification systems and want real classification reasoners?  In 
addition to 15926 and AP239, I would look at PLIB, eOTD, and the like.

If the answer to both is No, then we are done.  I would guess from 
David's rhapsody that he believes the answer to (1) is Maybe and yes, 
and the answer to (2) is "PLCS is using it, but primarily because of (1)."

I strongly believe that the technical answer to (1) is Yes, but the 
socio-economic answer is not so clear.  I hope it is also Yes, because 
if XML Schema was the right way, SC4 has spent 3 years making it harder 
for us to go there.

-Ed

-- 
Edward J. Barkmeyer                        Email: edbark at nist.gov
National Institute of Standards & Technology
Manufacturing Systems Integration Division
100 Bureau Drive, Stop 8264                Tel: +1 301-975-3528
Gaithersburg, MD 20899-8264                FAX: +1 301-975-4694

"The opinions expressed above do not reflect consensus of NIST,
  and have not been reviewed by any Government authority."