Personally, I found the dissertation highly enlightening and helpful.
Yes, it addresses ambiguity problems with rules (though I debate both of
Matt's descriptors -- the term huge and the term complicated) without
specifying how these rules might be machine-learned -- but doing so is still
a tremendously useful first step. Further, I think that requiring a model
that learns *all* language rules from unlabeled text (and starting from
scratch) is making the problem far harder than it needs to be. You didn't
do that as a child so why should you insist upon it for a system?
Thank you very much for the reference, Lukasz!
Mark
----- Original Message -----
From: "Lukasz Stafiniak" <[EMAIL PROTECTED]>
To: <agi@v2.listbox.com>
Sent: Tuesday, May 08, 2007 3:57 PM
Subject: **SPAM** Re: [agi] What would motivate you to put work into an AGI
project?
This work is a progress along a clearly stated line of research. It is
about ways of managing ambiguity. The dissertation is not about a
complete AGI system, it does not go into machine learning. It is not
about discovering meaning, but analysing meaning: the interpretation
of output is fully specified by a concrete application of the system
(I would call this a symbolic approach).
On 5/8/07, Matt Mahoney <[EMAIL PROTECTED]> wrote:
--- Lukasz Stafiniak <[EMAIL PROTECTED]> wrote:
> Iddo Lev has a more practical answer:
> http://www.stanford.edu/~iddolev/pulc/current_work.html
Just looking at it briefly, it appears to clearly present the many
problems
with natural language understanding (i.e. various forms of ambiguity).
Then
it addresses these problems with a huge, complicated set of language
rules
that have to be hand coded. Am I correct? If so, this approach is not
really
new. I would be interested in models that can learn language rules from
unlabeled text.
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?&
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=231415&user_secret=fabd7936