Xref: utzoo comp.sys.next:1134 comp.sys.mac:24672 comp.cog-eng:811
Path: utzoo!attcan!uunet!lll-winken!lll-lcc!ames!ucsd!cogsci!norman
From: norman@cogsci.ucsd.EDU (Donald A Norman-UCSD Cog Sci Dept)
Newsgroups: comp.sys.next,comp.sys.mac,comp.cog-eng
Subject: Sorry, no PDP (or any other) speech recognition yet
Message-ID: <679@cogsci.ucsd.EDU>
Date: 4 Jan 89 13:16:31 GMT
References: <263@gloom.UUCP> <908@quintus.UUCP> <4524@xenna.Encore.COM> <23040@apple.Apple.COM> <397@laic.UUCP> <23287@apple.Apple.COM>
Reply-To: norman@cogsci.UUCP (Donald A Norman-UCSD Cog Sci Dept)
Organization: UC San Diego Department of Cognitive Science
Lines: 36

In an article in this news distribution, it was stated that:

> ...  I saw a setup awhile back at UCSD that did speach
>recognition using neural networks (P.D.P. for you purists).  Although I
>never actually saw it run (only graphics output), it was supposed to be
>able to 'decode' sentences in roughly 1/4 real time.  Presumably, this
>was with ideal conditions, short sentences, etc.  

Sorry, not true..  The work that is being referred to is probably that
of Jeff Elman (who has worked with David Zipser and Jay McClelland).
Both Elman and Zipser are faculty in UCSD's new Department of
Cognitive Science. McClelland is faculty at Carnegie-Mellon.

Yes, the work is exciting and may lead to great progress someday.  But
we are very far from speech recognition of connected speech.  Note
that more than simple decoding of sound waves is required: a knowledge
of the context, semantics, pragmatics, etc., will be required.  Too
many speech strings are ambiguous without this information.  In fact,
not even the boundaries between words can be detected without knowing
the intended meaning. (My favorite example because it is a real one
that once happened to me is to think of the distinction between the
sounds of the phrases "new display" and "nudist play": the acoustical
waveforms can be identical.)

The Elman et al work is very important.  But it does NOT do real
speech processing.  Nor do they claim that it does.

Don Norman

Donald A. Norman	[ danorman@ucsd.edu   BITNET: danorman@ucsd ]
Department of Cognitive Science C-015
University of California, San Diego
La Jolla, California 92093 USA

UNIX:  {gatech,rutgers,ucbvax,uunet}!ucsd!danorman
[e-mail paths often fail: please give postal address and all e-mail addresses.]