Which ontologies; how to compare them

fritz@rodin.wustl.edu (Fritz Lehmann)
Date: Thu, 17 Feb 94 06:37:46 CST
From: fritz@rodin.wustl.edu (Fritz Lehmann)
Message-id: <9402171237.AA10497@rodin.wustl.edu>
To: poli@risc1.gelso.unitn.it
Subject: Which ontologies; how to compare them
Cc: cg@cs.umn.edu, interlingua@ISI.EDU, srkb-list@ISI.EDU

Dear Roberto Poli:

>Dear Lehmann,
>Thank you for your clarifications. They give me the occasion to present you
>some new remarks.
>It seems to me that there is something unclear in your methodology, even if
>your distinction among 'very casual', 'fairly casual' and 'serious
>classifications' is clear enough.

     To say my list of concept-systems has a "methodology" at all
is charitable.  Doing things right takes work; I was more
concerned that this be done at all rather than that it be done
right.  (There is a time to do things wrong.)

>... on the other side you make exclusions that look quite
>unjustified. For instance, how can you judge that Brentano shouldn't be
>You have listed a number of his pupils (Husserl, Meinong) and some
>of his contemporary followers (Simons, Smith). Looking at their writings
>you soon realize that there are important points in their positions that
>are almost unintelligible without a reference to Brentano's ideas and

     What points?  If you can show me a system of useful concepts
of Brentano's which is not already covered better in the systems
of his progeny Husserl, Meinong, Simons, B. Smith/Mulligan,
Guarino et al., then I'll add him to the list.  Remember, this is
not a history-of-philosophy list but rather a list of potential
concept-system sources to be used by engineers, AI people,
knowledge-base builders and integrators, etc.  If anything, I have
too many philosophers and pasigraphers and too few currently
available (FTP-able!) machine-ontologies.

>     Your criteria of (1) potentially useful ontologies as machine-encoded
>concept-systems for AI and for knowledge base integration, and (2) not too
>specialized ontologies, could be a good starting point.
>In any case, I would interpret (1) in a quite loose way. As a matter of fact
>we do not have sharp criteria for distinguishing ontologies that are
>encodable in machines from ontologies that cannot be encoded. Moreover, the
>borders of encodability vary in time. Today we (better: you) can encode
>things that were out of any possibility some decades ago. I would like to
>know if there is some good criterium for a distinction between what (at
>least in principle) can be encoded and what cannot be encoded.

     "Machine-encoded" was not meant to be my main point.  I'm
assuming that any concept-system can be machine-encoded in a
semantic network.  (Or in some other form like order-sorted n-adic
n-order predicate calculus.)  My main point was the "potentially
useful".  No doubt many are intrigued by Thales or Derrida or
Machiavelli or Hegel or Shirley Maclaine but I consider them
useless for practical AI until I'm convinced otherwise.  I can't
think of a good criterion at the moment.

>I am thinking that your idea to collect ontologies and to organize their
>network is a very important one. For this reason, I would like to suggest you
>to integrate your work with the following procedure:
>(A) Ask people to propose (1) definitions of ontologies and (2) criteria for
>the classification of ontologies.
>(B) Choose or elaborate a sufficiently wide definition of ontology and build
>up the tree(s) of ontologies.
>(C) Verify, using the items collected in your list, that (B) is good enough.
>(D) Choose the group of ontologies that are of some interest for your
>purposes (say, possibly encodable and not too specific ontologies).
>(E) Elaborate, compare, develop, etc. the items under (D).

     Yes, this would be great.  My personal problem is that I am
not in an institutional position to undertake such a major
collaboration and spend a lot of time on it.  I am "unfunded".  You
are in a most interesting position because of your intimacy with
the important continental philosophers while at the same time being
oriented to current industrial AI (as in your collaboration with
Nicola Guarino).  I think Guarino has in a sense already made "the
call" for people to come forward with AI/philosophical ontological
proposals (originally due maybe to his frustration with the KL-ONE
community's parched preoccupation with model-theoretic semantics?). 
Someone should be responsible for organizing and comparing them.

     A minor structural point: I would say your "tree(s)" in (B)
should be "ordered sets".  See my "Combining Ontologies" submitted
to you last year.

>As to Nino Cocchiarella and Jean Petitot. I think that their work is to play
>a very deep role in any serious research in ontology.
>Cocchiarella studies the general framework in which an ontology whatever
>can be realized. He was able to prove that some frameworks are intrinsically
>unadequate. To put things in a nutshell, his work shows stat we need second
>order logic and that the usual second order predicate calculus is precisely
>one of the above mentioned intrinsically inadequate framework. These are
>results that anyone should consider. His work is precious because make us
>consciuos that there is something wrong (or at least inadequate) in some of
>the 'usual' systems.

     Good!  I told Sowa and the KIF people to consider using
strongly higher-order logic as a basis instead of what you so
rightly call "the usual systems" (like the commandant's "Round up
the usual suspects!" in the movie Casablanca.).  The conventional,
plodding, crowd-following thing to do is automatically to make all
logic First-Order.  For this reason (only) I'll post a copy of this
to the interlingua list, which is about KIF, -- in case
Cocchiarella buttresses my argument.  But is "the usual _second_
order predicate calculus ... intrinsically inadequate..."?   I'm
not sure what you mean.  Also, this kind of logic insight (like
Montague's) doesn't necessarily mean Cocchiarella has a useful
_concept-system_  to offer AI.

>Petitot is relevant for some different reasons. Here I am trying to explain
>only one point of his general perspective. Even if many of us (myself
>included) are unable to understand his mathematics, the starting point of his
>reflections is quite easy to understand. The strategy is as follows. The
>first step is to build up the space of observators of some phenomenon F.
>Let's call U the space of observators. For any element u of U we associate a
>new space S(u) in which the phenomenon F is represented by the object O(u).
>At this object we can associate new spaces representing its qualities, etc.
>It is obvious that when we move from the observator u to the observator
>u + du the object changes from O(u) to O(u) + dv. We have objective
>phenomena when there are some connections between du and dv (that is
>between the variation of the observator and the variation of the object).
>The above looks appealing, even if the mathematical framework uses
>sheaves theories, Lie algebras, fibrations and others misteries that are
>almost unintelligible to non-mathematicians.

     This strongly reminds me of the unusual article by Robert
Marty, "Foliated Semantic Networks," in the "Semantic Networks"
collection which I edited (Pergamon Press, 1992).  Both use
topological/category-theoretic notions to specify what formal
relations are preserved in the conceptual systems of multiple
observers, like Husserl's shared "eidos".  These might provide
formal bases for inter-relating shared ontologies based on shared
experience/phenomena.  In your Padova Ontology Workshop, logician
Godehard Link got up and asked whether all that fancy mathematics
used by Petitot is really needed. I don't remember the answer.

>>I don't understand 'internal relations'.
>It is an old distinction. It could perhaps be rewritten saying that an
>internal relation is a relation of mutual existential dependence. Consider
>the connection teacher/pupil. There is no teacher if there is no pupil (and
>viceversa). It is important to note the following aspect: we can say that
>someone is a teacher even if s/he doen't have any pupil, but in this case we
>say that it is a potential teacher, not an actual one. Consider now the
>lenght/breadth opposition. No thing has lenght without having also some
>breadth (and viceversa). We can build up things that have only lenght and
>no breadth, but such things are ideal things, not real ones (if you prefer:
>abstract, not concrete things). You see from the
>above example that at least some internal relations are useful in
>distinguishing real things from ideal things and real actual things from
>things that are only potential. From these skratched examples you see that
>internal relations (or double existential dependence relations) play a deep
>role in the organization of a conceptual network. What is quite strange to
>our scientific taste is that the topic of internal relation was the pi!ce de
>r!sistance of the so-called idealists philosophers. For this reason, the
>choice to exclude idealists thinkers (say Hegel) from your list could be
>(at least in part) unjustified.

     This symmetric internal/external "foundation"-like thing
sounds useful  -- OK, so maybe my gut feelings are wrong about
Hegel, but please, don't ask people to read him!  I hear he's as
bad as late Husserl.  For these Tedeschi we need interpreters like
you who are willing to give simple examples.  (So do their fellow-
Tedeschi, I bet.)

>At the end of your mail you did a reference to the SRKB list. Could you give
>me some information for subscribing it? Thank you.

     You can try sending the message "subscribe srkb-list Roberto
Poli" to isd-serv@isi.edu, but I'm confused myself about how to
subscribe to SRKB.  I thought I subscribed months ago, but got no
email from it, so I subscribed again and this time got a
confirmation, but there has been no email from it since (including
copying of my own earlier message to you).  I also tried sending
an INDEX command to the listserv to see the archive of past mail,
but got no results.  (I was able to see some using World Wide
Web.)  Maybe the list administrator will see this and tell us what
to do.

     The SRKB list is the "shared ontologies" list for the ARPA
Knowledge Sharing Effort (as opposed to the interlingua list for
KIF, the ontolingua list,  the KQML list for query protocols, and
maybe a KRSS list somewhere for the KL-ONEites.).  I don't know
who subscribes to SRKB-list or how active it is.  I would really
like to see the serious practical mechanized-ontology discussions,
where ever it is that they are taking place.

>Yours sincerely,
>Roberto Poli

                          Yours truly,   Fritz Lehmann
4282 Sandburg Way, Irvine, California 92715, U.S.A.
Tel.: (714)-733-0566  Fax: (714)-733-0506  fritz@rodin.wustl.edu