On Sep 17, 2008, at 1:49 PM, John F. Sowa wrote:
As an "academic" myself, I certainly do not want to belittle the
importance of academic work. But I do *not* consider the academic
work to be intellectually superior to good engineering work that
is forced to get "down and dirty".
Neither do I. Who said anything about 'superior'? But I reject the implicit 'ivory tower' criticism inherent in your contrast with "down and dirty". In my experience, good academic work is often downer and dirtier than a great deal of applied "practical" real-world work, which tends (for good reason) to use known and established ideas and systems and to work around hard problems rather than analyze then.
There is always a tension
between theorists and practitioners, and the best work is usually
done by people who have a solid appreciation for both sides.
I have a very high regard for formal logic and model theory. But
the number of people who study and use those subjects is a tiny
fraction of the number who use languages that are logic-like,
but much less formal, such as SQL or UML.
I'm sure. I don't find this either interesting or important. The number of people who can drive a car greatly exceeds the number who can build a car. Nevertheless, were it not for the latter, the former would have nothing to drive.
The 19th century goal was to use set theory as the foundation for
mathematics and thereby extend logic to all of science. But in
practice, working mathematicians *ignore* the foundational work.
It may be a required course, but it doesn't help them solve problems.
This is wholly irrelevant. Ontology isn't concerned with foundations of mathematics.
PH> In fact, we all manage extremely well with a single-layered
semantics called model theory. This two-level idea is a
chimera, and an intellectual dead end.
What group of "we" are you talking about and for what purpose?
People doing semantics of the formal languages used to write ontologies. Which topic, you may recall, is supposed to be the main concern of this forum, and was the original topic of this thread.
People were using integers for thousands of years before Frege
defined them in terms of sets, so you can't count a child who
learns to count as using logic, set theory, or model theory.
An academic publication that uses model theory to analyze the
foundations of OWL or SQL has little or no influence on the
people who use those languages for any practical application.
That has not been my experience with RDF, OWL and SPARQL development. On the contrary, practical issues of usage, and model theoretic semantics, are constantly involved with one another. Admittedly, often to the mutual frustration of people on both sides, but still they find themselves obliged to pay close attention to one another. I know, because I was often the guy who had to do the mutual translations of terminology.
PH> You have argued for the idea, but do you have even an outline
or a sketch of what the second, model-to-reality, semantic theory
Yes, indeed. For categories #2 and #3, it is called systems analysis,
database administration, and various fad terms.
You cannot be serious. Systems analysis as a semantic theory? I find this hilarious.
Compare the Google
counts for "systems analysis" to "Tarski" or "model theory".
PH> [Model theory] is the foundation of ontology engineering, for
a start, as well as such topics as decidable logics (such as DLs).
I agree that academics who design formal logics (as well as SQL)
know model theory. But just look at the Google counts: for
"ontology", 15,400,000 hits; for "model theory", 532,000 hits.
and 17 million for 'semantics'. These figures mean nothing.
JFS>> How are you going to form n-tuples of such things [people,
tables, and cabbages] when you can't even state clear criteria
for identifying individuals of those types?
PH> Easily. You don't NEED to give such individuation criteria in
order to talk clearly and precisely about n-tuples.
That kind of very precise talk is done by the academics who publish
papers about model theory.
It is done, perhaps implicitly, by anyone who uses mathematics to talk about the real world. Which is a lot of people.
But when they ignore the model-to-reality
stage, the people in the trenches who use SQL, RDFS, etc., are left
with no guidelines for relating those neat n-tuples to the very
messy world in which borderlines are often vague or invisible.
Take someone using your favorite 'logic', SQL, to query a database. Must they be concerned with identity criteria for cabbages? Of course not. They tend to accept the same robust assumptions about identity that are also used by everyone else doing any kind of applied mathematics (which includes databases.) In fact, this concern with vagueness or blurriness of boundaries is itself a purely academic issue, verging at all times on the philosophical. Where are the exact "edges" of mount Everest? Nobody knows; yet everyone knows that Edmund Hilary was the first to climb Everest, and is not immediately thrown into a morass of identity confusion by such a statement. It is possible to be perfectly clear, even when referring to things whose edges, and even identity criteria, are not perfectly clear or even fully defined by the best human minds. This is an academic point, of course, but it is one that (unlike your obsession with the largely imaginary problems of boundary line determination) is firmly based in the down-and-dirty nitty-gritty real-world way that we all use everyday language quite successfully.
BTW, the people who know most about boundary-line issues are, of course, the people who specialize in this topic, which includes surveyors, lawyers (of a certain specialization), cartographers and others. I've worked with some of these people, and its quite possible to ontologize their expertise, just as it with other experts.
PH> But in any case, its fairly easy to state individuation criteria
for people, tables and cabbages: you can do it using spatiotemporal
contiguity in all three cases.
That is inadequate for distinguishing an egg from a person or
cabbage from cauliflower, broccoli, wild mustard, and many other
variations of the species Brassica oleracea.
Ah, you have changed the ground here. Previously we were talking about individuation criteria (this cabbage vs. that cabbage), now you are talking about how to distinguish among genera (cabbage vs. broccoli). I agree, simple contiguity doesn't do the latter. My own view on questions like this is, consult an expert: for Brassica Oleracea, a botanist or gardener, and so on. (Whatever you do, don't ask a philosopher :-)
JFS>> And those are the simplest "real world" examples you can find.
Everything else is immensely more difficult.
PH> Well, in fact it often isn't all that difficult, which is why
ontology-writing is possible at all.
The writing is easy. Most of it is done by the 97% who heard the
word "ontology", but know nothing about model theory. Even for the
3% who know model theory, the task of ensuring that those models
reflect reality won't get done by itself.
What is this 'task' that you keep worrying about? I think it is a chimera, that there is in fact no work to be done here. Which is quite likely why nobody manages to ever do it.
PH> By and large, its the people who build the best robots or the
best NL-comprehension software who also tend to write the best
I agree. But that is because the people who build the best robots
and NLP systems are forced to understand the distinctions and
*synthesize* them. The ones who compartmentalize those topics
(logic, ontology, and epistemology) in separate academic fields
are rarely the ones who make significant breakthroughs.
PH> But in any case, building ontologies and building robots are two
different activities. And understanding natural language is yet
I agree that they are different. But note that, in practice, the
labels used on those ontologies are nothing more than slightly
disguised English words and phrases. Even if an RDFS+OWL spec
is proved consistent, the dirty little secret is that the people
who write tags on web pages or choose them from menus will still
be thinking in terms of informal English semantics.
Quite likely; but...
that their usage will be unrelated to the formal definitions.
...it does not mean this. It means only that they may fall into the mistake of 'reading more into' the RDFS than it can in fact carry. And this is quite common, and is widely recognized as a practical issue, and down-to-dirty-earth user guides warn against it.
JFS>> But their attitudes [of Frege, Russell, & Carnap] toward
language, logic, and their interrelationships have polished
formal semantics to an elegant system that ignores problems
instead of solving them.
PH> This is such rubbish that I won't even try to sketch a response.
That is my summary of points made by Hao Wang (1986), a student
of Quine's who implemented ...
Yes, I know who Hao Wang was. And although he wrote that elegant early theorem-prover and was an important minor mathematician, he was also notorious for his loud views on a number of topics, on many of which he was proven wrong. In fact, he was wrong about his own main invention (Wang tiles.)
the first theorem prover (in 1959) that
proved all 378 theorems in propositional and first-order logic in
the _Principia Mathematica_. Wang also worked as an assistant
to Kurt Gödel and published many books and papers on logic.
PH> Which *semantic* problems are being ignored?
From Hao Wang, _Beyond Analytic Philosophy: Doing Justice to What
We Know_, MIT Press, Cambridge, MA, 1986:
Quine merrily reduces mind to body, physical objects to (some
of) the place-times, place-times to sets of sets of numbers,
and numbers to sets. Hence, we arrive at a purified ontology
which consists of sets only....
This is absolute nonsense as a summary of Quine's views and methodology, even if Wang was his student. Just read Quine - he is one of the clearest writers in English, after George Orwell - and see what he says. But in any case, even if Wang were right on Quine, what has this got to do with what we are talking about? Model theory does not "reduce mind to body" nor "place-times to sets of numbers". If anyone does that, it is probably the extremely down-to-earth US Information Mapping Agency, whose main basic ontology is based on piecewise-polygonal descriptions of space.
I believe I am not alone in
feeling uncomfortable about these reductions. What common and
garden consequences can we draw from such grand reductions?
What hitherto concealed information do we get from them?
Rather than being overwhelmed by the result, one is inclined
to question the significance of the enterprise itself.
Or the mental alignment of the reviewer.
In support of this view, Wang quoted a personal letter from another
of his former teachers, Clarence Irving Lewis, about the state of
philosophy in 1960:
It is so easy... to get impressive 'results' by replacing the
vaguer concepts which convey real meaning by virtue of common
usage by pseudo precise concepts which are manipulable by
'exact' methods -- the trouble being that nobody any longer
knows whether anything actual or of practical import is being
discussed. (p. 116)
Wang also observed that Carnap was "willing to exclude an
exceptionally large range of things on the grounds that they are
'not clear,' or sometimes that 'everything he says is poetry.'"
And he traced the sources of the problem back to the very
negative attitudes toward language by Frege and Russell.
Critiques like this have traction only when the critic is able to offer an alternative to clarity. Otherwise this is just carping. To slightly paraphrase Wittgenstein: whereof one cannot speak, thereof one should be silent, especially when criticizing people who do actually manage to say something.
If you don't believe Wang or me, just look at Carnap's famous
"Logische Aufbau der Welt".
I know it very well, with great affection: this is the book that I read as an undergraduate and which made me decide to go looking for AI (rather than mathematics!) at a time when there was no such subject yet. It is a masterpiece, and should be required reading for all ontology-builders. Early, and of course incomplete, and easy now to critique in details; but the very first attempt at what we would now call a comprehensive upper ontology. By contrast, Wang achieved nothing of comparable consequence.
It starts off with very precise and
formal definitions and axioms, but it get sketchier and less
detailed in each chapter. By the end, it's all hand waving for
all the serious problems. One of the terms he admits is "very
difficult" to define by his methodology is 'sign'. That makes
it hard to define 'word' or 'language' etc.
These are comparatively straightforward. What is hard to do is to define the general
notion of 'sign', which transcends language. And Carnap was right: it is, indeed, very hard to do that. To the best of my knowledge, nobody has yet succeeded in doing it in generally satisfactory way. Attempts like Korzybski and Ogden&Richards are clearly inadequate, and collapse into vacuity when put against modern cognitive science. Goodman's "Languages of Art' is a nice early survey of some of the issues, but it doesn't get very far.
PH> In fact I would go so far as to say that the whole idea of
treating NL as similar to a logic, which reached an apogee
with Montague, is now dead in the water. But that is NL
comprehension, and this forum about ontologies. Different
I agree that Montague's project is dead. But just look at what
all those "ontological engineers" are doing: using English
words and phrases for their tags.
I'm glad you put in the scare quotes. Surely you don't claim that tagging is building an ontology?
Pretending that there is
anything other than informal English semantics to support 97%
of those tags is wishful thinking.
Tags don't even have English semantics in many cases. (I've been studying image tagging in particular, which admittedly seems to be in a worse state than other content-tagging.)
IHMC (850)434 8903 or (650)494 3973
40 South Alcaniz St. (850)202 4416 office
Pensacola (850)202 4440 fax
FL 32502 (850)291 0667 mobile