Re: SUO: Just wondering...
Bill Andersen a écrit :
> You're missing the point. While everyone argues, we (Ontology Works) build.
> We have a million lines of stable code and lots of customers and our tools
> are very effective in solving their problems.
>
> This is why I have so much trouble with what goes on in the group. Too much
> talking and not enough doing.
>
> Before you malign something, you ought to know what you're talking about
> first. If you want to find out what our stuff does - ask me and I will be
> happy to tell you. You're clearly a smart guy and if you would lead by
> example, then maybe you'd bring more (much needed) substance to the group.
> Why go out of your way to piss people off??
Sorry, this was not meant as a personal attack.
If your compagny strives it must obviously be doing something usefull.
I did meant, however, that an all purposes top-level ontology cast in
concrete with a definition negotiated between heterogeneous parties
is a really *loosy* idea.
I am quite close to Sowa on this matter.
Throwing together various available ontologies for the sole reason
that they were not subject to "licensing restrictions" is ludicrous.
It just gives an illusion of power, but this power is not additive,
about the same as trying to put 10 people on 1 man month project.
The adjustments needed to glue together the base ontologies are
obviously hand crafted and need to be redone, rechecked, maintained,
for each modification.
We are back to square one, software maintenance problems,
is not this (among other goals) that were supposed to be solved?
There is also the faith in the "one and only" truth or answer to
any formalisation.
It is nearly never the case that you can define a canonical form to
any encoded structure, not even a plain boolean algebra, and when you
can (boolean rings) it gets exponential or NP complete to normalize!
But if you don't have a target canonical form, what's the criterion
to choose the glorious "one and only" top level ontology?
Debating the "merits" of alternative shapes? For each change which
is nasty enough...
How long will you have to wait between "releases" of the top ontology?
What's the point having "category theory" or anything else of that kind
embedded in the top ontology?
How many applications will need this even indirectly?
That may open the door to something like the "DLL hell" in Windoze:
At some point you may need some packaged component or ontology from a
third party which makes use of the "category theory", but unfortunately
the revision level has been updated etc... and, you know what, I guess?
If your market strength allows you to use Microsoft like policies
that may be good revenue, but are you sure?
I do not trust either the consistency of any hand crafted mathematics.
Do you know of the HOL theorem prover and to which extreme paranoid
measures they go to ensure consistency:
http://lal.cs.byu.edu/lal/hol-desc.html
See "conservative extension". This product is *used* in the industry
to prove the correctness of chip design, not an University toy, but
due to the consistency requirements it is "extremely awkward".
I do not believe the merry members of the SUO WG will go to such
lengths of precautions.
Putting "ad libitum" mathematicals concepts in the ontology is also
a kind of a pandora box. Where will you stop?
This is a very good example to show the stupidity of an all purpose
ontology, mathematical theories grown from a small set of axioms are
potentially infinite in number and variety and they are not arbitrarily
domain specific.
Each one can claim to be "top-level"!
What will the SUO WG do when someone ask for "yet another" nice theory
to become top-level?
Even when some mathematical concepts are needed to define a sub-ontology
(whatever you call that) there is no guarantee that the needed concepts
will be already available. You may even have conflicts between design
approaches of two ontologies to be merged or related.
Suppose one of them use set theory and the other category theory as
the basis for defining elaborate concepts.
Do you pretend you will have already provided a builtin effective mapping
between equivalent concepts described under set theory versus category theory?
If not, you have gained almost nothing.
If you do pretend that, Good Luck!
I may be wrong on this next point for lack of more carefull examination,
but it seems to me that the concept names and the concept definitions are
tied together based only on the syntax of the names.
That is, if you need not just to rename a concept, but to swap part of
it's definition for a stronger or weaker one you cannot.
See my argument about 'penguins' in my message
http://suo.ieee.org/email/msg08062.html
If it ever happens that a concept has to be weakened or strengthened in
the top ontology the modification of existing applications will be much more
expensive than without this flexibility.
The whole task of matching two ontologies based on the same top SUO ontology
will only be marginaly improved. The real problems of growing an ontology
from scratch and matching concepts designed by differents sources have not
really been tackled. And this will be yet another layer of API to be learnt
by the developpers!
The last negative point is more subtle. Forcing every application into
the same straightjacket, will be detrimental not to anyone in particular
but to the overall capabilities offered to the customers.
Same effect that virus strikes on Outlook users.
So, yes, may be your software is wonderful, I wish you that.
But do not rely on SUO (except you probably *have* to for marketing purposes)
because you will be left in the dust like everybody else with respect to this
grand idea of interoperability based on a common top ontology.
*This* is the project that will not deliver, not yours.
Best.
-- Jean-Luc Delatre
-----------------------------------------------------------------------------
He who has injured thee was either stronger or weaker; if weaker, spare him;
if stronger, spare thyself. - Seneca.
-----------------------------------------------------------------------------
mailto:jld@club-internet.fr -- GSM: 06 11 24 06 29