SUO: Re: ontology as science
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
John,
For instance, here are some of the critical questions
that are going to have to be addressed in a much more
critical way sooner or later, if are going to leave
behind the realm of pleasant dreams and soothing
blandisments and tackle a few real problems.
JS: The selection of words that people have used to express
their thoughts have a strong correlation with the concepts
that are important for science, engineering, business, law,
medicine, agriculture, manufacturing, and everyday life.
This sounds very nice. Indeed, it's such a beautiful notion that I wish
I could still believe in it, in quite the simplistic way that I once did.
Probably I will go on believing that some sort of "critical common sense"
is still possible, someday, probably not anytime soon, and that there are
deep underlying continuities between everyday coping skills and scientific
inquiry. But I must wake up from that dream, because that is not at all
what we are faced with here. We are not talking about science and common
sense, we are talking about OpenCyc and SUMO, neither of which is either.
And under these limitations there remain deep and irreconcilable differences
that divide these two aproaches to ontology: (1) the methods of science and
the sorts of knowledge frameworks that I could even start a non-laughable out
of the room discussion with if I tried to use them for any sort of research
support purpose, and (2) the methods and content that are part and parcel
of these current ontologies. In politics, it's nice to compromise, and not
be too "skeptical", but there is just no way of mushing togther a research
oriented ontology with a general info, person in the street, journalistic
ontology, without generating more mush, and there is no way of deriving
scientific knowledge from popular (mis-)conceptions without radically
changing common sense concepts in the process. That's just history.
A research support ontology, at a non-negotiable minimum,
would start from a stock of "basic undergrad knowledge"
and go from there. So far, it been like having teeth
pulled (by an amateur dentist) trying to get across
even the most basic standard textbook definitions
of concepts that nobody would argue with for
a second in any of the research communities
that I have ever worked with.
There seems to be some bizarre idea in certain segments
of the AI world that expert systems failed because of
the annoying expertise of experts rather than the
inadequacy of the AI methodology of the times,
so they threw out the expertise and kept
the bathwater.
Jon Awbrey
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
> John,
>
> My question is: Is anybody here ready to quit bkue-skying
> this and get serious about the sort of things you say below?
>
> Jon
>
> o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
>
> John F. Sowa wrote:
> >
> > Bill,
> >
> > WB> Okay. I agree that lexicographers can treat language use as
> > > data and be as objective about its data analysis as any other
> > > scientist. I'll also accept that lexicographers can account
> > > for and predict semantic drift. But what are the ramifications
> > > of this on ontology design or engineering? (Specifically of
> > > the SUO.)
> >
> > The major implication is that there exist testable methods and
> > criteria for guiding the selection of categories and the axioms
> > and definitions that define them. Following is the kind of
> > data that ontologists should take into consideration:
> >
> > 1. The selection of words that people have used to express
> > their thoughts have a strong correlation with the concepts
> > that are important for science, engineering, business, law,
> > medicine, agriculture, manufacturing, and everyday life.
> >
> > 2. The sounds that people assign to those concepts may vary
> > from one language to another, but the common semantic
> > features across different languages are important sources
> > of semantic distinctions that an ontology, especially an
> > upper-level ontology should be able to represent.
> >
> > 3. The concepts and theories of every scientific discipline
> > represent the best thoughts of highly trained people who
> > have examined the phenomena (and the existing things
> > those phenomena indicate) in many important fields.
> >
> > These are sources of data that a good ontology should be able
> > to represent and organize in a coherent way. The IF framework
> > represents a collection of tools that can be used to analyze
> > such data, derive categories for classifying the entities,
> > and define those categories in more formal notations than
> > the usual natural language descriptions found in dictionaries.
> >
> > The basic point is that the choice of categories is not
> > arbitrary and there are ways of evaluating the categories
> > and their definitions by objective criteria.
> >
> > John
>
> o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o
o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o~~~~~~~~~o