Re: [SUO] Re: Key questions about common upper ontologies
Danny Ayers brings up an important point that I didn't
address in my earlier note on this topic. John Sowa
agrees, and so do I, that almost no one would have to
us all of a standard upper ontology for internal purposes,
and probably not even for communication between systems.
I will respond to some of John's other points in a
separate note on this thread.
[DA]
>
> A small clarification here - I'm very much suggesting the cherry-picking
> usage for inter-organizational communication. If two organizations were
> in the business of feline seating arrangements and decided to
> communicate in Warlpiri, they would need only enough of the language to
> be able to say "cat", "sat on" and "mat". This would of course require
> some common (abstract) language, but I think it's commonly agreed around
> here that for most purposes that will be (a subset of) FOL. The
> communication will also need the concrete terms - two nouns and a verb.
> Ok, terms like "cat" are unlikely to be in the upper scope, but I
> believe the principle is the same. Successful communication does not
> require the whole abstract language or concrete terminology. This is
> advantageous to the adoption of any SUO, as people are more likely to be
> willing to augment their existing languages/vocabularies than replace
> them with a whole new language infrastructure.
>
I agree that as a practical matter, it will be essential
for wide adoption of a common upper ontology that it have a mechanism
for selecting from among its many concepts only those that are
required to specify unambiguously (see below) those concepts that
need to be treated in the local domain of use. As a first approximation,
that might be only a **base set** of the classes that are in the line of
inheritance up to the top concept, for all classes of interest in the
domain -- to which one will add all relations defined on those
classes, plus the classes that are related to the base classes by
one relational link. This would give one a good set of concepts
to reason with, while restricting the need to deal with concepts
that appear irrelevant to one's interests. It will be necessary
to increase or reduce that first approximation depending o
how well it serves one's needs. But this can be done by selecting
from exactly the same one common upper ontology.
"Unambigously" here is used to mean "as precisely as possible using
the concepts in the ontology" -- which may not be enough to
avoid real-world ambiguity in certain circumstances.
[DA]
> If two organizations were
> in the business of feline seating arrangements and decided to
> communicate in Warlpiri, they would need only enough of the language to
> be able to say "cat", "sat on" and "mat". . . .
> Successful communication does not require the whole abstract
> language or concrete terminology.
How much they would need depends on what they want the computer to
do for them. Accurate communication requires that the sender
and receiver use the same upper ontology. What is not contained in
the computer must be supplied by the humans that create and interpret
the communication. If the humans are expected to do most of the interpretation,
very simple tactics like providing a translation table of the three
terms of interest will serve, no ontology is needed. If we expect the
computer to do a lot of reasoning with the transmitted information,
much more detail will be required, and both systems must agree on how
the concepts are to be interpreted -- i.e. what kinds of inferences
are permitted. Reasoning in ontologies is extremely sensitive
to the numbers and definitions of the relations, and small
variations can cause devastating differences in inference.
That is why I can't imagine a serious project to allow
computers to **interpret** communications to any extent
without agreement on a common upper ontology. The use of,
say, a sales-order schema between vendor and purchaser
could be viewed as a primitive common ontology sufficient
for the amount of computer reasoning required for that narrow task.
"Interoperability" is sometimes used as an all-or-nothing
property of interacting computer systems, but I think it is much
more reasonable to view "Semantic Interoperability" as
the **proportion** of reasoning in any activity that can be
done by the computer without human intervention. For the
foreseeable future, humans will always be in the loop somewhere.
What I assume we are trying to do is to make the computer capable
of substituting for humans in a wider range and larger fraction of
information-processing tasks. When the task requires communication,
any part of the communication that requires computer interpretation
(changing the computer response depending on the content of
the communication) will require the use of the same semantic
definitions by sender and receiver -- i.e. that specification of
reasoning processes which tells the computer **how** to vary
its response depending on content. Otherwise miscommunication
will result. And that would be absolutely unnecessary and wasteful,
because it is technically feasible to develop an upper ontology
that will fit the **needs** of all users -- though not necessarily
their personal preferences.
Pat
=============================================
Patrick Cassidy
MICRA, Inc. || (908) 561-3416
735 Belvidere Ave. || (908) 668-5252 (if no answer above)
Plainfield, NJ 07062-2054
internet: cassidy@micra.com
=============================================