Ontological EDI & Wittgenstein
fritz@rodin.wustl.edu (Fritz Lehmann)
Date: Wed, 28 Sep 94 07:03:49 CDT
From: fritz@rodin.wustl.edu (Fritz Lehmann)
Message-id: <9409281203.AA08521@rodin.wustl.edu>
To: cg@cs.umn.edu, edi-new@tegsun.harvard.edu, srkb@cs.umbc.edu
Subject: Ontological EDI & Wittgenstein
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk
Nitin Borwankar (nitin@sybase.com) wrote about "ontology-based"
Electronic Data Interchange on the edi-new@tegsun.harvard.edu list:
>The discussion about ontologies and attempting to make meaning precise
>at a deep level brings to my mind similar attempts to make language precise,
>by a philosopher ( Ludwid van ? ) Wittgenstein.
>( I am quoting loosely from memory of stuff I studied during a minor track in
> philosophy during my EE undergraduate studies - it was quite a while back )
>He wrote a book "Tractatus Logico Philosophicus" a hierarchically arranged
>explication of the universe starting with
>1 The world is all that is the case
>1.1 ...
>1.2 ...
>...
>2 ...
>2.1 ...
>etc. with each statement building on those that had gone before, starting from
>first principles and assuming nothing, attempting once and for all to rid
>Philosophy of confusions due to ambiguities in language.
>He was attempting to unambiguously bind meaning to each word.
>Much later in his life he came to the conclusion that the meaning of a word was
>defined by its use ( ie the multiple contexts in which it was used )
>and could not unambiguosly
>be bound to a "value".
Borwankar is right about Wittgenstein. This is an apt criticism,
one which we who advocate ontology-based EDI should address. The answer
is that Wittgenstein shifted from one extreme to an opposite extreme.
He started by thinking that each concept has a precise definition, and
ended up with the idea that concepts are based only on "language games"
and that every concept is arbitrarily loose and any word can almost mean
anything (the latter is only a slight caricature of his later musings).
What is needed, and is feasible, is a reasonable middle ground. In the
EDI world it is true that there cannot be a dogmatic definition of
precisely what is and is not a "suburb" (X12 Element 309:SB), but we
know that Manhattan is not a suburb and that Webster Groves is; we can
define some qualities that suburbs certainly should have and other
qualities that suburbs certainly should not have -- even though this
leaves a large "gray area" for human beings to judge the borderline
cases. That is, we can state _necessary_ conditions, even though they
are not _sufficient_ conditions for classification. These are also
called constraints. And we can agree completely that no barge (X12
Element 211:BRG) is a suburb, despite the natural imprecision of both
concepts.
>It would be interesting to see how far the "dogmatic" definitions at the deep
>level will hold in the face of real world market dynamics and multiple rapidly
>evolving contexts.
Well, the same applies now, to "old EDI". The current EDI
"definitions" depend on human understanding of the field-names (keeping
in mind that most data elements in X12 and EDIFACT have only one-line or
one-word English phrases). So the meanings (when understood at all) are
as stable as the English words. The meaning of a word can't be
stretched beyond recognition, so humans periodically have to update the
standard. We would need to update "ontology-based EDI" too (except that
it would be more work, and maybe more interesting work). The benefit
would be that once the ontology is updated, machines all over the world
would respond automatically to the changed formats without the current
horrendous integration and maintenance expenses. A one-time-effort.
Based on this foundation, two machines can negotiate the descriptions of
the concepts their forms will be about, and the format of the forms.
That goes a long way towards automatically handling "rapidly evolving
contexts".
>Does the ontological approach have a built-in framework for updating the
ontologies
>automatically in the face of fresh real world evidence ? Or is it a static
model
>updated periodically by humans?
Automatic ontology updates are the subject of some promising machine-
learning research, but for now you need to update the base-ontology by hand
periodically (as far as I'm aware). In EDI it's also possible that
experienced EDI users could broadcast their ontology corrections and
enhancements on an ongoing basis, so that others could freely update as
desired.
>On an unrelated subject but relating to an earlier comment of Nick Szabo,
>it would be an interesting statistic if someone were to tally how many of the
>creators of the ontological approach to EDI had actually run a company involved
in
>trade, or even run a part of a company especially the purchasing department.
>I am referring now to large companies or organizations - the kinds in which
>ontological models may be viable in the future.
(It might also be interesting to know how many EDI people, period,
have run companies, etc. I suspect the answer is "Lots -- lots of EDI
consulting companies, that is!") The people who are moving toward the
ontology-based EDI idea have experience in database integration. In
this field it is known that you can only go so far in automatically
integrating schemas or records by common field-names and structures --
then you hit the Semantic Wall: in order to integrate dissimilar
(heterogeneous) databases, you have to understand what the data is
_about_. That's why governments and corporations pay billions every
year to human systems analysts and programmers to translate data and
metadata from one machine format into another. Years of study have gone
into automated database (and enterprise) integration. Without built-in
real-world semantics (ontologies, not data-structures) a human being
still has to do the vast bulk of the work. My position is that the EDI
challenge is really the same as the database/enterprise integration
challenge: one system, set up a certain way for certain purposes, has to
communicate information correctly to another system, set up a different
way with different data-names and structures (and maybe for different
purposes), but with subject-matter common to both systems. For the deep
semantics needed, we turn to Knowledge Representation people in the AI
community (and any others who like to ponder these things). People who
actually run companies and purchasing departments can come in with a
description of their practices and a stack of business forms they have
been using (that's how X12 and EDIFACT arose); I don't expect them to be
the advanced EDI theorists.
>This would give us a reality check.
I doubt it, for the reason just stated, although it's always good to get
feedback and subject-matter expertise from users. (The "check" I would like
to see is a nice fat refund of the billions spent on unnecessary, tedious
human translation and maintenance of intercommunicating data systems,
including EDI.)
>Nitin Borwankar,
>nitin@sybase.com
Besides answering Borwankar's message, I want to say that some of
the conflict on the edi-new list between the "advanced EDI" faction
(like Ken Steel and me) and the "forms EDI" and "smart contracts"
faction (like Nitin Borwankar and Nick Szabo) seems unjustified. These
are _complimentary_ problem areas. If Ken Steel or I or others
accomplish our more advanced goals, we will depend heavily on the more
immediate results for "forms" etc. Transactions based on "open-EDI"
negotiations or "ontologies" will not be too impressive if they are
grounded in outdated "old-EDI" forms-processing and communications
technology. I appreciated everything Borwankar wrote on the forms
issues; I guess I don't care much about the philosophical "what does EDI
really mean?" question (since I consider "EDI" a misnomer anyway).
Different people have different interests and talents and I look forward
to exploiting every advance made by the "short-term results" people.
Yours truly, Fritz Lehmann
GRANDAI Software, 4282 Sandburg Way, Irvine, CA 92715, U.S.A.
Tel:(714)-733-0566 Fax:(714)-733-0506 fritz@rodin.wustl.edu
=============================================================