[Top] [All Lists]

[ontac-forum] What should be in an upper-level ontology

To: ONTAC-WG General Discussion <ontac-forum@xxxxxxxxxxxxxx>
Cc: "Lenat, Doug" <doug@xxxxxxx>
From: "John F. Sowa" <sowa@xxxxxxxxxxx>
Date: Sun, 21 May 2006 13:55:44 -0400
Message-id: <4470A9A0.8070003@xxxxxxxxxxx>
Folks,    (01)

I have been making some strong criticisms about upper-level
ontologies, even though I have been working on and writing
about such things for many years.    (02)

I haven't completely given up on the idea that an upper level
is useful, but over the years, I have come to the conclusion
that any upper level should be as *neutral* as possible --
i.e., it should not be biased for or against anybody's pet
theories, and it should not contain any axioms that contradict
any of the specialized axioms that anybody might need in any
application at any lower level.    (03)

These criteria imply that the upper levels should have very
few axioms.  Following are some kinds of axioms that should
*not* be in the upper levels:    (04)

  1. Any axioms that make empirical claims that might be
     falsified by future experiments or any claims that are
     known to be false in detail, but which may be useful
     approximations for many purposes.  For example, the upper
     levels should be neutral with respect to a Newtonian view
     vs. any more modern theory of physics because for many
     practical purposes a Newtonian description is accurate
     within the granularity of the usual measuring instruments.    (05)

  2. Any axioms that require, prefer, or rule out one kind of
     representation over another, such as a four dimensional
     vs. a (3+1) dimensional description of space and time.    (06)

  3. Any axioms that rule out exceptional cases that may be rare,
     but possible.  For example, it should not say that a tiger
     has four legs, because some tigers might be born with more
     than four and some might lose a leg.  In fact, there might
     be quadriplegic tigers that get around in some prosthetic
     device.    (07)

  4. Any axioms that imply a vase and the clay it consists of
     are or are not identical, because many respectable theories
     make different claims in that regard.  They should also
     avoid all claims about whether a child is identical or not
     identical to the adult at some later stage of life -- because
     some theories say yes, others say no, and other treat the
     question as context-dependent (i.e., identical for inheritance
     issues, but not identical for employment).  In fact, the
     entire issue of identity claims is so full of conflicting
     philosophical positions that the upper levels should *not*
     make any identity claims of any kind.    (08)

  5. Any axioms that imply physical objects and processes are
     disjoint.  Some theories say they must be disjoint, others
     say they may overlap, and others say that object and process
     descriptions are complementary ways of describing the same
     phenomena.    (09)

  6. Any axioms about artifacts that may be falsified by developments
     in technology.  For example, the attached phone.gif example is
     taken from a dictionary published in 1969, but very few of the
     features depicted are common in the telephones manufactured
     today.  However, the definition in that dictionary is still
     true:  "an instrument for reproducing sounds at a distance."    (010)

  7. Any axioms that distinguish essential properties from accidental
     properties.  This issue has been debated since the time of Plato
     and Aristotle.  The traditional definition of Human is Rational
     Animal, and the ability to laugh was considered an accidental
     property.  However, many philosophers have claimed that the
     ability to laugh is just as characteristic of humans and more
     easily defined than the ability to reason.  Today, genes are
     considered more fundamental to what is "essential", but that
     makes it harder to distinguish humans from chimps and bonobos.
     A truly neutral upper level should avoid any commitment to what
     is considered essential vs. what is considered accidental.    (011)

When you start to analyze the issue, the number of possible conflicts
becomes so large, that the safest position with regard to any axiom
in the upper levels is very short:  When in doubt, leave it out.    (012)

That definition of telephone from 1969, which is still true today,
suggests the kind of information that might be included in an upper
level:  a definition or axiom that is true because of the intended
function of an artifact.  Unfortunately, it is very difficult to
state such definitions in full generality and even more difficult
to find formally defined relations that can be used to state them.    (013)

For example, how can the words "instrument", "reproduce", "sound",
and "distance" be defined in a precise, but general way?  And how
could they be defined in a way that would distinguish a telephone
from a radio?  And would you want to distinguish telephones from
radios?  If you did, that would rule out cell phones, which are
in fact radios and can be used to reproduce radio programs and
even television programs.  In fact, cell phones are now being used
as cameras and even TV cameras that broadcast live events over the
Internet.    (014)

If you want the upper levels to be sufficiently general, you can't
put many axioms into them.  They become, in fact, just what I said
in my previous notes:  a cleaned-up version of WordNet.  It's not
an accident that WordNet is so widely used, because it performs a
needed function.  Anything that is more detailed than WordNet would
be too constrained and too inflexible to be useful for relating
one general-purpose ontology to another.    (015)

For deduction, however, inflexibility and constraints are necessary
to support detailed proofs.  But those constraints will inevitably
create contradictions with very important developments in technology.
If you define the word "telephone" in such a way that makes it disjoint
with radios, TVs, and cameras, then you rule out cell phones -- or
perhaps you may permit very simple cell phones, but you rule out any
kind of new technology without making a major category shift.    (016)

Summary:  When you examine all the conditions and applications that
an upper-level ontology must serve, you discover that it is very hard
to distinguish it from a cleaned up terminology.  It is clear that we
need detailed axioms in the microtheories, but it is not at all clear
whether we need an upper level that is distinct from a terminology.    (017)

John Sowa    (018)

GIF image

Message Archives: http://colab.cim3.net/forum/ontac-forum/
To Post: mailto:ontac-forum@xxxxxxxxxxxxxx
Shared Files: http://colab.cim3.net/file/work/SICoP/ontac/
Community Wiki: 
http://colab.cim3.net/cgi-bin/wiki.pl?SICoP/OntologyTaxonomyCoordinatingWG    (01)
<Prev in Thread] Current Thread [Next in Thread>