[xtm-wg] A Proposal (was Re: Knowledge management claims re XTM)

["Patrick Stickler" <patrick.stickler@nokia.com>]

On 24 Aug, EXT Hans Holger Rath wrote:
> Hi,
> 
> My 5 cent (of a Euro :^) about this topic:
> 
> Bryan Thompson wrote:
>> Yes, I agree.  However I feel we could address much of this
>> continually re-emergent confusion if the XTM AG made specific
>> recommendations concerning *how* the XTM (or the ISO standard)
could
>> be used, without modification to the specification, to encode
>> knowledge modeling languages.
> 
> I already proposed such an approach a year ago at GCA's
Metastructures 99,
> Montreal. The ISO WG call them TM templates. A template (or
application profile
> or schema) contains all ontology declarations, transitivity
information,
> type hierarchy information, explicit inference rules, and validation
> constraints.
> 
> Last week I presented the final results at GCA's Extreme Markup
> Languages 2000, Montreal. The concept is based on PSIs (Public 
> Subject Identifiers) which have to be publicly registered by ISO or 
> OASIS or whoever. This registration is everything what is needed,
> no standard has to be changed, but TM tools can refer to the PSIs
and
> support their semantic which is clearly defined together with the
> registration. This method allows definiton of various
> application dependent profiles (e.g. knowledge representation,
> subject classification, and other use cases). 
> 
> My paper is in the Extreme ML proceedings and the appropriate
> XTM WG should have access to it. If requested I can upload a 
> PDF to the egroups server.
> 
> Having such technique at hand knowledge representation can be done
> using TMs. Reading Sowa's book will convince you (he is talking
about
> conceptual graphs which could be expressed as TMs without any
problems).
> 
> Regards,
> --Holger
> 

(A lurker emerges ;-)

I promised Michel in Montreal to post my thoughts on this topic, but
rather than doing so in a more formal manner as I had intended, since
there clearly has been much discussion already about this, I will
throw
out a few informal thoughts and see what the rest of you think.

It seems to me that the issue of encoding particular ontologies and
semantics using TMs is not so much whether the standard is expanded
itself to support them, but rather that there be *some* kind of
standardized mechanisms and definitions so that the portability and
long term usability of data and tools is surer. I.e., whether the
TM spec defines e.g. transitivity of associations or whether there
is a standard template (or other mechanism) for doing so, is not so
critical as there actually *being* such a standard mechanism which
most if not all tools and data employ.

One can encode essentially any ontology using TMs, but one cannot
garuntee that anyone else (particularly any system) will understand
the vocabulary or semantics of that ontology. That is the crux of
the problem. How to e.g. define things such as transitivity in
a way that is standard, portable, and valid over the long term.
One can use structured names, roles, facets, etc. to do this, but
any such approach is limited in utility because it is non-standard.

In my opinion, TMs and XTM will not and cannot reach critical mass
in the marketplace without a core set of standardized ontologies
based on a standardized specification mechanism, because without
that, there is no information interchange -- and it seems to me that
information interchange (or more correctly, information integration)
is the very heart of TMs. To move past "toy" systems, we must have
standard ontologies and means to validate instances to ensure
conformity. This is IMO even more critical than having common PSIs.

These core ontologies should not, IMO (and it seems also in many
others),
be on the same level as say a DTD such as DocBook or TEI, but
would have some degree of official endorsement and promotion in
association with the XTM standard. The idea of defining a core
standard, compatible with the ISO TM standard, and then a set of
official templates or extensions included in or referenced directly
by the standard seems a reasonable approach. It allows the core
to remain concise and "pure" while giving official weight to common
ontologies.

That said, let me offer one concern (and a proposed solution) about
defining templates as sets of PSIs, which is similar to problems
associated with RDF Schemas, and that is rigorous syntactic validation
of instances, from the perspective of information interchange and
tool compatibility.

By simply defining PSIs, inference rules, etc. one cannot ensure
completeness of specification in a TM instance at the level of 
resolution that one (or at least I ;-) would like. One can ensure that 
*only* the defined/licensed PSIs are used, or that conflicting
assertions 
are not made, etc. but one cannot, eg. ensure that e.g. *every*
instance 
of a given element type is associated with a particular classification
in a given ontology, and thus, a software application that would
depend
on an exhaustive classification of topics, associations, association
roles,
etc. could not be sure, by validation prior to processing, that the TM
instance is acceptable input; and must then resort to its own
validation,
which is inefficient and costly. It is no different than having to
process
non-validated but merely well formed XML instances, and equally 
unacceptable where high-volume automated processing and/or data
integrity
are concerned. 

I believe the idea of templates is the way to go, but fear the
definition
of yet another schema language, particularly one which limits itself
to the higher semantic and KR level and not addressing the
nuts-n-bolts
issues of day-to-day data interchange. It also means that vocabularies
are 
defined in two places, one defining syntax, such as a DTD, and another
defining auxilliary semantics, either as a TM or some other encoding.

I would like to thus propose that XML Schemas would serve as an
excellent
solution to this problem (and certainly others). The XTM standard
could
define a core XML Schema which defines the basic syntactic encoding of
TMs in general. This core XML Schema could then be extended and
augmented
with additional Schemas which define both general and more specific 
ontologies, including rigorous constraints for obligatory
classifications,
data formats, etc. The mechanisms provided by XML Schemas for
"subclassing"
would enable the core TM specification to remain free of further
semantics,
while providing a framework for defining official supersets which meet
the needs of various communities.

The beauty of this approach is that instances which happen to conform
to a highly specialized schema can still be processed in a meaningful
way by tools which only understand the core TM schema, as the Infoset
will provide information about the complete "inheritance path" of 
any given element, even if the GI differs from that in the core
Schema. 
This maximizes the utility and benefit of general TM tools while
allowing 
for highly specialized processing, possibly as extensions to general
tools 
analogous to the subclassing of XTM schemas.

Granted, the XML Schema standard is not finished, and that may
certainly
be an obstacle in adopting it in some official manner by the XTM
working
group; however, the timing seems workable, if concensus is reached and
there is cooperation between the XTM and XML Schema working groups.

I will leave the issue of what those "core standard ontologies" might
be to
another thread, though I have a number of thoughts about that too, as
I think it is more critical *how* we define and validate against
those standard ontologies in an explicit and reliable fashion than
what those ontologies are (but only just ;-)

Adopting XML Schemas for XTM both allows the core XTM spec to remain
pure of additional semantics while allowing the standardization of
common ontologies, all within the same framework, and with a clear
hierarchy of relations between general and more specialized
ontologies.

Cheers,

Patrick

--

Patrick Stickler
Senior Specialist
Customer Documentation Applications
Nokia Information Management
Nokia Networks

Hatanpaankatu 1A
33101 Tampere Finland

phone +358 3 25 74808
fax   +358 3 25 75700

patrick.stickler@nokia.com




-------------------------- eGroups Sponsor -------------------------~-~>
GET A NEXTCARD VISA, in 30 seconds!  Get rates
of 2.9% Intro or 9.9% Ongoing APR* and no annual fee!
Apply NOW!
http://click.egroups.com/1/7872/4/_/337252/_/967540625/
---------------------------------------------------------------------_->

To Post a message, send it to:   xtm-wg@eGroups.com

To Unsubscribe, send a blank message to: xtm-wg-unsubscribe@eGroups.com