Re: Fw: Intro to natural language processing
John and Jean-Luc,
I completely agree with your criticisms
of Rob F's comments and proposal. John's
"grain of salt" comment is sufficient.
But your comments are worth some comments:
JLD> Not formally defining our purposes may be
> an enhancement to our "creativity" in our research
> quests UNLESS we fall prey to "scholastic thinking"
> which will stymies us into barren tracks.
Please do *not* use the adjective "scholastic"
in a derogatory sense. The work on logic and
the semantics of natural language (i.e., Latin)
was outstanding, and 20th-century linguists,
philosophers, and logicians would have profited
enormously from studying it.
I frequently cite William of Ockham's Summa Logicae
because it was the last of the great Summae, which
incorporated and systematized over 200 years of
research and debate during the 12th to 14th centuries,
which built on earlier work by Aristotle and his
followers and commentators in the Hellenistic and
Arabic world. (Citations below).
And by the way, they most definitely had a purpose
for that analysis: the clarification and correction
of the reasoning processes used in philosophy and
especially theology. In today's business world,
the goal of developing a great system of logic for
the purpose of doing philosophy or theology would
not get much funding, but in those days, theology
was where the money (or at least subsistence wages)
could be found. The Church was the DARPA of the
middle ages.
By the way, I like the one-line review of the PhD
dissertation by Karol Woytyla (AKA Pope John Paul II):
"scholastic steel meets phenomenological mush".
JLD> The assumption looks like there must be some "essence"
> semantics in some platonistic realm and that if we were
> smart enough, wise enough ot lucky enough to "discover"
> it then all "problems" with semantics, NLP, AI and what
> have you will be solved or at least within easy reach.
The scholastics adopted Aristotle's categories, but
they did try to minimize the differences between Plato
and Aristotle by considering A to be a "correction"
to P, and their theology to be a "correction" to A.
I believe that they had a better designed upper ontology
than Cyc or WordNet (but that's not saying much).
JLD> I am fairly sure this is the WRONG approach, and
> if we knowingly or unknowingly lean toward that,
> there will be no significant progress.
I wouldn't say it's completely wrong, but I do say that
it's only one piece of the puzzle. I recommend Ockham's
Theory of Propositions (Part II of his Summa) as an
example of how a model-theoretic analysis of a natural
language can clarify a large number of issues. (And I
use the term "model-theoretic" advisedly, because the
main differences between Ockham and Tarski are that
(1) Tarski used an algebraic notation, but (2) Ockham
had a much broader coverage -- certainly much bigger
than Montague's 37-word "fragment" of English.)
Following is my view of how model-theoretic semantics
fits into the puzzle:
1. Natural languages are capable of being used with
as much precision as any artificial language, and
the usage in Euclid, Ockham, or any textbook on
mathematics demonstrates that point.
2. Model-theoretic semantics is only applicable when
a single consistent formal ontology has been
adopted and used unequivocally. That is exactly
what Euclid, Ockham, and most math textbooks do.
3. The main difficulties in understanding NL are
caused by the enormous complexities of the world,
the enormous number of possible points of view on
any aspect of the world, and the enormous flexibility
of NLs to adapt to any and every point of view.
4. The artificial languages of logic and most programming
languages can only be used *after* a single consistent
ontology has been adopted. The great strength of NLs
is that they can be used both in the early stages of
developing an ontology and in the later stages of using
the ontology (as well as every intermediate mixed stage).
5. The greatest problem of NL semantics is trying to find
a way of preserving the flexibility of NLs at the early
stages that also allows for the use of NLs in a precise
way during the later stages.
I agree with John B's comments, and I just wanted to applaud
his concluding sentences:
JB> If you want to seriously get into NLP, then starting
> points have to involve *all* of the approaches mentioned.
> No one of those by itself is in a position to deliver.
The Aristotle-Ockham-Tarski-Montague-Kamp style of model-
theoretic semantics is an important piece of the puzzle,
but we still don't know how many boxes of pieces haven't
yet been found.
John Sowa
__________________________________________________________
For a quick overview of Ockham's work and how it relates
to his contemporaries and predecessors, I recommend the
article by Paul Spade in the Stanford encyclopedia:
http://plato.stanford.edu/entries/ockham/
For more, see Spade's download page (including his 410-page
book on medieval logic and translations of various sources):
http://pvspade.com/Logic/noframes/download.html
Mediaeval Downloads
Following are three books with translations of all of Part I,
all of Part II, and some selections from Part III of Ockham's
Summa Logicae:
_Ockham's Theory of Terms_, Part I of Summa Logicae translated
by M. J. Loux, St. Augustine's Press, South Bend, IN.
For selections from Part I, translated by P. V. Spade, see
http://pvspade.com/Logic/docs/ockham.pdf
_Ockham's Theory of Propositions_, Part II of Summa Logicae
translated by A. J. Freddoso & H. Schuurman, St. Augustine's
Press, South Bend, IN.
Ockham, William of, _Philosophical Writings_, ed. and translated
by P. Boehner, revised by S. F. Brown, Hackett Publishing Co.,
Indianapolis. (Latin and English selections from Summa Logicae
and other works.)