basis for CCAT ontologies (Fritz Lehmann)
Date: Fri, 18 Nov 94 00:40:57 CST
From: (Fritz Lehmann)
Message-id: <>
Subject: basis for CCAT ontologies
Cc:, anquetil@IRO.UMontreal.CA,,,, billrich@VNET.IBM.COM,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,
Precedence: bulk

     Brandon Buteau ( wrote:

>What are "Norvig-names," Fritz?  I've got a copy of WordNet and its
>documentation, and I didn't see that term anywhere.  I thought that it
>identified concepts as word clusters (e.g., {beat, strike, hit} vs. {beat,
>defeat} vs. {beat, circumvent}, etc.)

     Wordnet originally assigned numbers to the meanings (clusters of words)
in English.  This is a bit unwieldy for human users.  Someone (Ed Hovy?)
recently told me that Peter Norvig has written English-like "names" for the
meanings in Wordnet, and that these are known as "Norvig-names".  I assume
that they can be obtained either from the Wordnet site or from Norvig.

>By the way, I am impressed with the
>breadth and depth of WordNet

     I'm also impressed with the length.

>and think it would be a useful starting point for
>many ontologies.  It is comprehensive (over 70,000 distinct word meanings or
>distinct concepts), and although it has a strongly lexical flavor (concepts are
>grouped into the four fundamental categories noun, verb, adjective, and
>adverb), it still contains a great deal of support for ontology development.

     Agreed.  Kevin Knight has exploited this by adding the Wordnet
categories to the Pangloss Ontology Base.  The diversity of CCAT members is a
bit of a challenge; some deal with very low-level, precise mathematical
axiomatizations of the "core" subjects (like Pat Hayes' treatments of TIME)
whereas others are interested in higher-level and everyday concepts which are
sometimes closer to natural language and usually lack precise definitions
(like Pat Cassidy's and Paul Doudna's work on GENERAL THESAURI).  I'm in the
middle, since I'm particularly interested in how the higher-level concepts
can be founded upon the "core" concepts.  (To you "core" axiomatizers: TV ES

>For example, WordNet includes the following concept metarelations in its
>concept descriptions:
>1) Synonymy / antonymy
>2) Hyponymy / hypernymy
>3) Meronymy / holonymy
>4) Entailment based upon:
>        Temporal Inclusion (Synchronymy?)
>            Troponymy (Co-extension)
>            Nontroponymy (Proper inclusion)
>        Temporal Sequence (Prechronymy?)
>            Presupposition
>            Causality
>[ Synchronymy & prechronymy are terms I coined to cover the equivalent WordNet

    All these "nymy" words!  (Making up words is fun but jargon is pretty
burdensome on students and readers.)   CCAT could use some formal definitions
of these (second-order) relations between _words_, and between _predicates_
if possible.  These can be used in ontological definitions and constraints.
> ...
>WordNet is available on the net right now.

     What is the current net source for Wordnet?
> ...
>And I am highly attracted to any
>approach to CCAT ontology development that is strongly rooted in common
>language usage.  Otherwise, it will be too difficult and/or error-prone for a
>potential ontology user (i.e., a person) to map their *own* concepts into those
>defined by CCAT contributors. ...

     I agree.  I think the vast majority of CCAT concepts at the higher levels
will correspond to a word (or phrase) in natural language.  But, the important
thing in adding a concept or relation to the hierarchy is "What follows from
it?", that is, what is inherited from it, what constraints does it have, etc.
While natural language is a good guide to natural (useful) bundles of
qualities -- associated with words -- at the higher level, it's not so good at
resolving what Lenat & Guha call "thorny representation issues".  TIME, SPACE,
CAUSALITY etc. have to be formulated in a certain way in order to get the
right set of automatic conclusions to flow, and natural language is of only
limited help in this.  Those automatic conclusions include many that are
normally considered too obvious to bother saying at all (like: if A is before
B then B is after A).  Also, these are the areas where natural language is the
_most_ ambiguous.  For engineering applications, there may be many concepts
which have no corresponding words or standard natural language expressions.

>As far as suggestions for CCAT ontologies go, I would like to see groups
>focused on the domains of COMMUNICATION (the passing of information between
>agents) and COOPERATION (the coordination of agent activities to achieve
>goals).  Although these would draw upon or contribute to some of the existing
>they would also cover some important areas that might otherwise fall through
>the cracks.  For example, it would help provide a sound ontological basis for
>KQML and other evolving techniques/tools/languages/standards for distributed
>agent interactions.
>-- Brandon

     A COMMUNICATION subgroup seems like a good idea.  If you'll start the
ball rolling (say, by sending us a tentative suggested list of needed concepts
and relations), we'll add COMMUNICATION to the list of subgroups.  For deep
definitions I assume it will need to presuppose all of the "core" ontologies.
I'm less keen on making COOPERATION an ontology by itself.
Cooperation/collaboration is a "hot" subject in AI now, as a technique, but I
don't see that it has a rich or basic ontology.  (The CCAT "subgroups" are
totally informal -- just a way of focusing attention.)

                          Yours truly,   Fritz Lehmann
GRANDAI Software, 4282 Sandburg Way, Irvine, CA 92715, U.S.A.
Tel:(714)-733-0566  Fax:(714)-733-0506