On 9/7/05, Jim Self <[EMAIL PROTECTED]> wrote:
> Holloway, Thomas wrote:
> >Unfortunately our GW is not that original:
> >
> > http://en.wikipedia.org/wiki/Colorless_green_ideas_sleep_furiously
> 
> Wikipedia has become a marvelous resource :)
> 
> It's been quite a long time since I have read Chomsky. It is good to be 
> reminded.
> I hope to shift my work in the not too distant future more in the direction 
> of natural
> language processing.

I also have been interested in natural language processing.  There is
something about it that makes one think that it should be easy to
create a parser etc, and from there make a step towards understanding.

I played with the code for a language parser called "Link Grammar".  
http://www.link.cs.cmu.edu/cgi-bin/link/construct-page-4.cgi#submit

I created an interface for it and was quite happy with it at first. 
It could take almost any sentence and isolate the verbs, nouns etc.  I
even extended it slightly by converting all verbs into their infinite
form, and adding a tag indicating their tense (past, present, present
perfect progressive etc.)

Then realized the emptiness of words to a computer.  There is no link
between the word "dog" and something with a wagging tail to a
computer.  So if I were to pass to my computer a command "give me a
cookie", the computer needs more information about "giving" and
"cookies"

So then I looked into ontology software, which is a heirarchy of
concepts.  The best I found then was
http://www.opencyc.org/

But even this seemed lacking of real world connection to me.  And
ontology could verify the truth that a "cookie" is type of food, and
object, something round etc etc.  But this is just defining words in
terms of other words.  I bit of circular reasoning.

About this time, by children were young and rapidly acquiring language
skills.  By watching them, I realized that a child knows what a bottle
of milk is all about long before they get their mouths working enough
to call it a name, or even to recognize the word "bottle" when mommy
says it.  So there has to be a lower level of understaning, and the
words of a sentences are just window-dressing.

So I decided to try to model 3d world and see if I could get a simple
AI agent to wander around, bump into things and notice (i.e. setup
associations) what things satisfied low-level drives (pleasure,
pain-avoidance).  And then perhaps later I could layer a parsed
sentence on top of this and finally create some connection between
words and reality.

Well, all the above was the result of about 2 yrs. of experimentation.
 I basically got in over my head and couldn't get neural nets working
properly etc.  I hit the wall and decided to take a break from it.

But it is still a fascinating area of study that I wish I could have
done more in.

Hmmm. Its hard to have too many irons in the fire (aka Jack of all,
master of none.)

:-)

Kevin

> 
> The attempt to construct an inherently meaningless expression like the one 
> below,
> immediately confronts our drive to find meaning in virtually everything. One 
> thing left
> out of the discussion on wikipedia is data loss and error correction. For 
> instance, it is
> extremely common in email to find that words have been left out of phrases 
> (like NOT) or
> that inappropriate words have been substituted or that words have been 
> written out of
> order. Rather than treating such messages as meaningless, we generally find 
> ways to
> transform them into something reasonable.
> 
> Thus the phrase below reforms in my mind to "Colorless notions of green ideas 
> filled her
> sleep as she lay there furiously dreaming." Perhaps a reference to Chomsky 
> himself as he
> searched for the canonical nonsense phrase.
> 
>


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf
_______________________________________________
Hardhats-members mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/hardhats-members

Reply via email to