On Thursday 27 July 2006 00:22, John F. Sowa wrote:
>
> >
> > We've been running around for 50 years or more finding incomplete
> > compressions. You would think we'd get the hint.
>
> I don't know what hint you're suggesting. That no rule-based
> system can ever be complete? I think that's obvious.
I'm glad you see this, John. It is the point I was trying to make.
Chaitin's work helps us understand why it should be so.
For the rest it I am not so much saying that incomplete grammars are useless,
as that we must understand incomplete or uncertain is all they will ever be.
With this insight I believe we should now be able to move forward and find
solutions to all those problems we haven't been able to solve. The key will
be to approach them as problems of how to fit particular generalizations to
their purpose. I think with this understanding we will do better than by
futilely trying to find a better global fit all the time as we have been
doing.
Just to summarize:
No single grammar of natural language can ever be complete. This is because
natural language text is at some level Kolmogorov complex. That means we
should stop looking for a single logical basis in rules, and try to use the
experimental observations, the corpus, to make specific, subjective, context
dependent, generalizations. The way we should make those generalizations is
the same (functional contrast/distributional analysis.) Only our goal
(specific generalizations or "local grammar", not global grammar, -- a
different "grammar" for each analysis decision, I think) need change.
But that must change. Unless we want to spend another 50 years running around
finding fuzzy categories and asking ourselves why.
Mike - I'm not sure what you are saying, other than that linguists have been
careless about fitting theory to the data.
-Rob
This archive was generated by hypermail 2b29 : Thu Jul 27 2006 - 12:59:21 MET DST