Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!uwm.edu!psuvax1!rutgers!mcnc!rti!ntpdvp1!sandyz
From: sandyz@ntpdvp1.UUCP (Sandy Zinn)
Newsgroups: comp.ai
Subject: Re: Language Acquisition (was: Machines *CAN* think!)
Summary: lingo
Keywords: Language processing, Language Understanding, Language Acquisition
Message-ID: <370@ntpdvp1.UUCP>
Date: 9 Apr 90 22:15:45 GMT
Organization: Northern Telecom DMS-10 Div., Raleigh, NC
Lines: 44

> (Paul Steven Mccarthy) writes:
> When you think, don't you "talk
> >to yourself" -- primarily in _English_?  But enough of that.
> 
(Ken Presting) writes:
> When I think in words, it's
> usually with some particular listener in mind.  I seldom think in terms
> of text arranged on a page or screen, or in the form of a conversation
> with myself.  But I do think that *any* public activity can serve as
> a model for internal representations.  How (potential) listeners fit
> into the picture is puzzling.

Perhaps not so puzzling if we propose that the internal representations
of language use (modeled on remembered conversations) also contain context
markers, and if we consider that "style" is part of language context (which
I think it must be).  I'm using "style" to denote characterizable sets of
relationships in language usage such as word choice, length of sentences,
inclusion or lack of extended metaphors, etc.

The Balinese have a very elaborate system of social interchange which
regulates word usage based on whom the person will be talking to.  Long
silences can ensue when two strangers meet if they can't discern from
context what "rank" of person they've got there.

Sentences as words-strung-together-according-to-syntax is too simplistic
a conception of language, in my view.  Communication is entangled in many
levels in human speech, and I have no problem with the idea that more
than just the words-and-syntax are represented in language memory -- or
IMO, represented in the processing functions (input or output) themselves. 

I heard Jerry Fodor give a seminar today advocating a modular theory of
speech recognition.  He was selling a sentence parser which structured
input based only on grammatical rules; he radically isolated his parser
from contextual inferences -- a rash stance, I think (but if you see
connectionism as the plague...) -- Nature, who's a very old hand at
maintaining viable organizations, is a profligate user of redundancy,
and makes good use of context as valuable info.....

An illuminating point, Ken -- glad you brought it up.

@ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @ @
    Sandra Zinn              |   "The squirming facts
    (yep these are my ideas  |      exceed the squamous mind"
     they only own my kybd)  |         -- Wallace Stevens