SUO: RE: RE: Re: General Design

["Fowler, Julian" <JFowler@modulant.com>]

Matthew, John et al

I agree fully that constrained models restrict their utility (outside their original design intent - sometimes, though, such constraints are important to understanding and making use of the meaning of data governed by/described by those models).

I take something else, however, from John's comment about the inherent danger of imposing an axiomatic structure.  Not only are the structures likely to become restrictive over time (implying a need for change, or for "fudging" of the ontology to make newly-discovered or newly-interesting information fit with it), but the intermediate levels between "thing/top/universal/whatever" and useful terms/concepts may themselves be problematic.  From a somewhat oversimplified perspective, (some) top-down ontologies and generic data models tend to have the form:

thing
philosophical-stuff
useful-stuff

and for many analysts, programmers, and subject/domain experts trying to use such a framework, the philsophical-stuff is seen as confusing, unnecessary, and (as John points out) introducing a new vocabulary in which to express what should be familiar and well understood concepts.  As someone pointed out in a previous thread, try telling engineers at GM that their products are anything other than automobiles!

There is, of course, a key role for the philosophical-stuff components of such an ontology: without it we may not be able to establish correspondences (and therefore enable communication) between my-useful-stuff and your-useful-stuff.    In many cases, though, the necessary abstractions to enable communication need not be at the highly generic (thing, philosophical-stuff) level.  Therefore, an approach that recognizes this and can, within an evolutionary/automated system capture and exploit the natures of:

- my-useful-stuff context
- your-useful-stuff context
- my-useful-stuff-and-your-useful-stuff context

and their inter-relationships is likely to be more fruitful (and comprehensible to a wider user community) than seeking/imposing a stable axiomatic framework for absolutely-everybody's-useful-stuff.

Julian

-----Original Message-----
From: West, Matthew R SITI-ITPSIE [mailto:matthew.west@shell.com]
Sent: 2003-08-12 09:43
To: John F. Sowa; Jon Awbrey
Cc: Kenneth Fields; Ontology; protege-discussion@SMI.Stanford.EDU; SUO;
cg@cs.uah.edu
Subject: SUO: RE: Re: General Design



Dear John,

A word of support.

> Of course, not.  The main reason why WordNet is more flexible than
> Cyc, Sumo, Dolce, or any other axiomatized ontology is simple;
> 
>      The axioms get in the way.
>
I have spent many years trying to improve the quality of data models
in Shell and elsewhere. By far the biggest problem has been that 
data models impose constraints that simply aren't true (except in
some limited set of circumstances). As a result I have become very
conservative about constraints or axioms.


Matthew West
Principal Consultant
Shell Information Technology International Limited
Shell Centre, London SE1 7NA, United Kingdom

Tel: +44 20 7934 4490 Other Tel: +44 7796 336538
Email: matthew.west@shell.com
Internet: http://www.shell.com


> -----Original Message-----
> From: John F. Sowa [mailto:sowa@bestweb.net]
> Sent: 11 August 2003 21:11
> To: Jon Awbrey
> Cc: Kenneth Fields; Ontology; 
> protege-discussion@SMI.Stanford.EDU; SUO;
> cg@cs.uah.edu
> Subject: SUO: Re: General Design
> 
> 
> 
> Jon,
> 
> I am on my way to California tomorrow, and I have a million things
> to finish up before I leave.  So I'll have to wait until later,
> perhaps this weekend, to send more info.
> 
> > Perhaps you can explain to us how "'knowledge soup' -- the 
> loosely organized,
> > semi-structured mix of whatever people have in their heads" 
> can be canned in
> > FOL cans, because this is a problem that has exercised me 
> for way too many
> > years now, and though I may say "I think I can" till I can 
> not say it any
> > more, I just can't make myself believe that can be done any 
> more, not
> > in that "direct and literal" (DAL) fashion with which ontologicists
> > keep on trying to curry favor with the geists of atomists past.
> 
> Of course, not.  The main reason why WordNet is more flexible than
> Cyc, Sumo, Dolce, or any other axiomatized ontology is simple;
> 
>      The axioms get in the way.
> 
> That simple observation is heresy, which will bring down more threats
> of excommunication than a gay bishop.   But it is a fact:  the reason
> why anybody wants axioms is to make their ontology more precise and
> better adapted to whatever little niche they currently inhabit.  But
> if you want an ontology that is general enough to apply to anything
> and everything, you can't live with a hot-house pansy that can only
> survive when every weed in sight has been extirpated.
> 
> > Failing that, and GOL knows any sensate person ought to 
> know that it's failed by now,
> > the only still "logical" alternative seems to be something 
> like the indirect approach,
> > via the explicit recognition that our models are 
> abductively approximate analogues of
> > the real thing that is ever out there, beyond us, and thus 
> that we have no choice but
> > to begin in more amorphous, proto-formal settings like sign 
> relations, taking serious
> > Peirce's notion of "logic as (a specialized form of) semiotics".
> 
> My proposed solution is based on some principles that I have been
> preaching for as long as anybody has been willing to listen (which
> isn't very long, for most of them):
> 
>   1. You cannot build a hierarchy by drawing trees (or lattices)
>      by hand -- you must have tools that can derive the hierarchy
>      automatically from whatever set of distinctions are embodied
>      in the ontology.  (One example of such a tool is the Toscana
>      implementation(s) of Formal Concept Analysis (FCA), but there
>      are also other similar tools that could be used.)
> 
>   2. You cannot assume that any given hierarchy is fixed for all
>      time -- it is only fixed until somebody adds another axiom.
>      If their axiom makes a new distinction, they will push a
>      button to run the hierarchy-derivation program to redraw the
>      lattice.  If you're lucky, the new axiom won't change very
>      much.  But if you're unlucky, it might reorganize everything.
> 
>   3. If you need to preserve your axioms as long as anybody is
>      using some application that depends on them, then you can't
>      let people push buttons that will change your hierarchy just
>      because they want to change theirs.  The net result is that
>      the total hierarchy bifurcates, trifurcates, or multifurcates
>      into incompatible contexts, modules, microtheories, or
>      whatever you want to call them.
> 
>   4. Meanwhile, nobody wants to learn a new language with a totally
>      different vocabulary just because somebody decided to make
>      a new distinction.  Therefore, they just recycle their old
>      words and force the lexicographers or WordNetters to add more
>      word senses or synsets.   And then you have to add more ropes
>      (or pointers) to align the terminological hierarchies with
>      the multipicity of axiomatized hierarchies.
> 
> And as you pointed out, Peirce recognized these problems a century
> ago.  He explained, as you said, that any theory of logic is merely
> a subset of the broader theory of semeiotic, which he designed to
> accommodate all these ways that people and other sentient beings
> organize their ways of making sense of what they experience.
> 
> So I really shouldn't complain too much that people have been
> ignoring what I've been preaching since 1987, since they have
> been ignoring what Peirce was preaching for at least a century
> or more before that.
> 
> John
> 
>