[SCL] Re: [CL] Dagstuhl semantic integration paper

[John F. Sowa]

Chris,

I like the general approach, and I believe that a
revised version could be very important for a wide
audience.

Some comments:

I agree with your point about

 > Joseph Goguen ([6], [7]), Robert Kent ([8]), Marco Schorlemmer
 > and Yiannis Kalfoglou ([11], [12]), and others...

I think they have some important ideas, which they have
been trying to sell for the past 20 years or so without
making a dent on anybody who might need to use them.

I remember a remark many years ago about the difference
between the "primitives" (most programmers) and the
"space cadets" (people who publish in the J. of the ACM,
SL, etc.).  And those people are among the spaciest of
the space cadets.

 > The approach I’ll discuss is quite similar in certain
 > respects to the one outlined by Ciocoiu and Nau in their
 > short paper [2]. I have myself in the past been somewhat
 > critical of the approach for being a bit too model
 > theoretic in orientation...

Minor point:  Citeseer has a lot of useful info, but I'd
rather have a pointer directly to the paper.  If I want
to see the Citeseer stuff, I can go there myself.

I'm not sure exactly what you mean by "a bit too model
theoretic", but I am certainly tired of seeing paper after
paper restating Tarski in a less comprehensible way than
he did 40 years ago.

I also like Quine's criticism that models prove consistency,
but they don't, by themselves, tell you anything about the
intuitions behind the symbols.

If nothing else, I would hope that SCL/CL could provide a
model theory that is general enough so that most of these
people can just adopt the SCL/CL model theory, show the
mapping, and drop their upteenth hundred rehash of Tarski.

But there is also a much more fundamental criticism of all
extensional semantics:  It cannot distinguish two predicates
that just happen to have the same extension in the current
model.  Ciociu and Nau use the example (mother ?x ?y), but
if all the women in your model just happen to be mothers,
you're going to get a mapping that is, in general, wrong.

Re Gruber: I agree that "specification of a conceptualization"
is popular, but it is a totally useless definition.  Some
people think they understand "conceptualization" because it
suggests something about "concepts", but that move just dumps
you in the conceptual swamp instead of the ontological swamp.

Section 2 gets to the meat, but (a) these issues need to be
recognized and understood by the primitives, yet (b) your
usual writing style appeals only to the space cadets, who
don't need the tutorial.

I would also suggest that you include lots of terms (and
examples) that the primitives know and love.  For example,
the primitives represent the "signature" of a language L
by entity-relationship diagrams, which are familiar to
everybody who has been exposed to the DB literature and
which have become one of the six basic notations of UML.
I suggest that you say that front and center, and give
an example of an E-R diagram that shows how each of your
words and symbols map to parts of the E-R diagram.

By the way, are you aware of the paper that for many years
had the highest citation index of any publication in the
computer science literature?  It was Peter P. Chen's paper
on E-R diagrams in TODS (Transactions on Database Systems)
of 1976.  And E-R diagrams, by the way, are just a prettier
version of the "Bachman diagrams" published by Charlie B.
in 1969.  The first version of Chen's paper was so badly
written that one reviewer, Ted Codd, practically rewrote
it in his review before it was accepted for publication.

The moral is that P. P. Chen hit a "sweet spot", which
few of us have been able to find.  A more recent example
is Tim Berners-Lee, who got knighted for his sweet spot.
You have to be in the right place at the right time with
the right product packaged with the right ribbons in order
to hit those sweet spots.  I don't know what all the right
ribbons are, but notations with lots of Greek letters are
the wrong ones.

And by the way, examples from Peano and Zermelo-Fraenkel
are not "vivid" for the audience we need to address.

John