NetNews Usenet Archive 1992 #26/NN_1992_26.iso/spool/comp/ai/4253

Internet Message Format | 1992-11-11 | 3.6 KB

Path: sparky!uunet!charon.amdahl.com!pacbell.com!ames!sun-barr!cs.utexas.edu!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!emory!ogicse!das-news.harvard.edu!calliope!kehler
From: [e-mail] (Andy Kehler)
Newsgroups: comp.ai
Subject: Re: Loebner Turing test for Human like conversation
Keywords: Loebner, turing, natural language
Message-ID: <[e-mail]>
Date: 12 Nov 92 02:32:54 GMT
Article-I.D.: das.1992Nov12.023254.2176
References: <[e-mail]> <[e-mail]> <[e-mail]>
Sender: [e-mail] (Network News)
Organization: Aiken Computation Lab, Harvard University
Lines: 59

In article <[e-mail]> [e-mail] (Kenneth Colby) writes:
>
>Kehler at Harvard mistakenly thinks from second-hand information
>that all the programs in the Loebner competition were ELIZA-like
>"or worse".
>
>Much depends on the level of description being used for likeness
>comparison.  All computer programs can be said to be ELIZA-like
>in the sense that at the register transfer level they all involve
>machine language instructions operating on bit vectors.  But at
>the top semantic-interpretive level they are quite different in the
>way they function.
>
>Our Loebner prize conversational program translates the input into
>a semantic database of conceptual primitives, constructs sets of
>interpretations, selects one for response by various criteria, and
>produces a response that does not use the surface words of the input
>as part of the output.  Any one who thinks this is the way ELIZA
>functions is unfamiliar with ELIZA's code as well as with the new
>ideas in conversational language processing - AI/CLP.
>
> [flames at Raja at Microsoft deleted]
>

Wow!  Mr. Colby asks for comments, and this is what you get for
answering!  Please don't get so defensive -- I didn't mean it
personally!

Thank you oh so very much for your lesson concerning the fact that all
programs come down to 'machine language instructions operating on bit
vectors' -- this little story being only a bit less ridiculous than
your analogy with chemistry and atoms in your response to Mr. Raja
(and I'll let him respond to those comments if he so chooses).  And
yes, I understand how ELIZA works, thank you.  Did you perhaps
consider that perhaps it is *your* program that I am not familiar
with?  Your response is especially unfair and underhanded considering
that you didn't identify yourself as a participant in your earlier
message (and in fact word it as if the AI Magazine was your first
introduction to the competition), so how would I know how your system
works?

I was not at the competition, and furthermore I don't believe those at
the competition got to see the inner workings of any of the programs
-- so obviously I was commenting on the *output* of the programs.  I
did see some excerpts of output, and frankly they did look ELIZA-like.
I have no idea whether I was looking at *your* transcripts or someone
else's.  Since, thanks to your last post, this discussion has gone
from a general discussion of the Loebner competition to a defense of
your particular system architecture, why don't you post all the
transcripts of your system to the net?  Then people can judge for
themselves if your worried now about the reputation of your system.

Anyway, enough already, sorry to spend so much bandwidth on this but
the above arguments deserved response.  The point of my original
response was that I feel the Loebner competition hurts AI/NLP, a
position which I stand by.  Please feel free to disagree, but please
don't respond in the tone that Mr. Colby chose to.

  -- Andy