Re: ontology as science (was: Re: SUO: Re: (ELP's summary of MRW's standards experience))
I would suggest another interpretation of the science of #2 and #3 - I think
there are at least two levels of objective tests for these.
1. "using my knowledge of some concept, does an ontology classify to the
correct concept I expect" (the most obvious failure would be it not
containing some property restriction that you thought was necessary and
sufficient.)
2. "what can we hypothesize about what someone(or everyone if you combine
ontologies) says about some concept" - I guess this should be broken into
single ontologies and shared(combined) ontologies.
In #1 we could perhaps test formally against our own ontology, in #2 it
would seem like a few different methods. For single ontologies we can test
for logical consistency, for shared ontologies, we could build statistical
like hypotheses about what people say about something, or, if a shared
ontology is logically consistent, that #1 kinds of test pass or fail, or,
that all people, each with their own mini-ontologies surrounding a concept,
agree or disagree with the shared ontology for that particular concept.
In the end, an ontology is simply a model, just like other models. Various
representations give us different tools to help test the knowledge says what
we think it does.
While the interpretation of labels for concepts can be subjective, ontology
gives you a formal and unambiguous way to represent that interpretation, so
that indeed we can ask questions of this interpretation, and get the same
answer - no matter the person or the time a question is asked.
Matt
----- Original Message -----
From: "Burkett, Bill" <WBurkett@modulant.com>
To: <sowa@bestweb.net>; "John F. Sowa" <sowa@bestweb.net>;
<standard-upper-ontology@ieee.org>
Sent: Friday, June 20, 2003 1:09 AM
Subject: RE: ontology as science (was: Re: SUO: Re: (ELP's summary of MRW's
standards experience))
>
> Thanks, John:
>
> Perhaps I should have inserted the word "objectively" and said
"objectively testable".
>
> >Every axiom and definition of ontology makes a testable claim
> >about one or more of the following three kinds of things:
> >
> > 1. What is in the world,
> >
> > 2. What do people say about what is in the world,
> >
> > 3. What are the relationships among the things in #1 and #2.
>
> I agree that physical science has developed methods for objectively
determining "what is in the world" independently of individual perspective
or time. (And is still refining these methods.) However, #2 and #3 will
always be subjective and therefore of questionable testability from a
scientific perspective.
>
> >By this criterion, lexicography is a science. It gathers data
> >consisting of word occurrences and the surrounding context,
> >and formulates definitions that cover the data. Those
> >definitions can be tested by anyone else who cares to gather
> >similar data and check whether the definitions apply.
>
> Yes, it would be considered a science because it's dealing with lexical
symbols as the data under objective analysis. With respect to semantics,
however, it cannot take into account the semantic drift of words in a
language as a language evolves over time. Nor can it take into account
different languages except as a distinct data sets.
>
> >That kind of science satisfies the point #2 above, which deals
> >with things that people say about the world. It is true if and
> >only if people do indeed say such things about the world.
>
> I don't think any here is objectively testable. It is all subjective and
will always be subjective - hence the problem.
>
> Can you describe any kind of objective test that will succeed or fail
(unambiguously given any tester at any point in time) that will prove that
when I say "tree" that there are trees in the world? Sense perception
might be one such test, but then I'm left questioning what is a "tree" and
what isn't, and where do the boundaries of "treeness" end.
>
> >The kind of science stated in point #1 above is physical
> >science, which has an enormous overlap with ontology. Just
> >look at the information in Cyc, SUMO, and other ontologies.
> >They are filled with testable information about scientific
> >topics.
>
> I agree that point #1 is about physical science, but disagree about the
enormous overlap with ontology except insofar as ontology is an example of
#2. Ontologies are not testable (setting aside syntactic validity, but
we're not talking about that.) The only "test" of an ontology that is of
any value is whether or not in represents the information that I want my
information processing machines to deal with and whether that representation
can be communicated effectively with other information processing machines.
(And note that the "test" is about value to a person, not about whether the
tested hypothesis succeeds or fails.)
>
> I still stand by the claim that if it involves semantics - perceivable
symbols that "mean" something - that it cannot be a science. (At least as
we're dealing with it here.)
>
> Bill
>
> -----Original Message-----
> From: John F. Sowa [mailto:sowa@bestweb.net]
> Sent: Wed 6/18/2003 6:26 PM
> To: Burkett, Bill; Mike Pool; John F. Sowa;
standard-upper-ontology@ieee.org
> Cc:
> Subject: RE: ontology as science (was: Re: SUO: Re: (ELP's summary of
MRW's standards experience))
>
>
>
> Bill,
>
> Every axiom and definition of ontology makes a testable claim
> about one or more of the following three kinds of things:
>
> 1. What is in the world,
>
> 2. What do people say about what is in the world,
>
> 3. What are the relationships among the things in #1 and #2.
>
> WB> To be "science", it must be falsifiable and it must be
> > possible to devise experiments to test hypotheses indepedently
> > of any particular individual or any particular point in time
> > and achieve the same results....
>
> By this criterion, lexicography is a science. It gathers data
> consisting of word occurrences and the surrounding context,
> and formulates definitions that cover the data. Those
> definitions can be tested by anyone else who cares to gather
> similar data and check whether the definitions apply.
>
> That kind of science satisfies the point #2 above, which deals
> with things that people say about the world. It is true if and
> only if people do indeed say such things about the world.
>
> The kind of science stated in point #1 above is physical
> science, which has an enormous overlap with ontology. Just
> look at the information in Cyc, SUMO, and other ontologies.
> They are filled with testable information about scientific
> topics.
>
> The kind of science stated in point #3 is mathematics. It is
> not an empirical science, but a very hard-edged discipline for
> counting, analyzing, and relating everything and anything in
> points #2 and #3.
>
> I agree that many things in social sciences are "soft" and
> hard to falsify. But if the ontology attempts to define and
> axiomatize what people say about those subjects, then its claims
> can be tested by the same methods used by lexicographers.
>
> John Sowa
>
>
>
>