From: nybakken@world.std.com (Kris Nybakken)
Subject: Re: APPS: VR word proccesing?
Date: Mon, 11 Jan 1993 02:51:03 GMT
Organization: The World Public Access UNIX, Brookline, MA



In article <1hp3f5INNcd4@shelley.u.washington.edu> Andy Kossowsky
<AKOSSOWSKY@TURBO.kean.edu> writes:

>Thanks for the info, but what I was really wondering about is a PURELY
>virtual keyboard, ie, to an observer, it just looks like you're typing
>in mid air!
[...]
>The reason I bring this VR wordprocessing subject up at all, is a few
>weeks ago, an InfoWorld columnist wrote an article saying pocket
>computers would be useless because people can't type on them.
>
>Apparently, he hasn't heard that computers are occationally used for
>things other than word processing, but even so, I'm betting that soon
>even word processing won't be dependant on a full-size qwerty
>keyboard.

Well, in the overall VR picture, I think that keyboards are extremely
overrated.  The direction that I believe that we should all be looking
is speech recognition.  This person's word processing app would let
them babble on for a couple pages, then they could stop and say
"Review", and see the whole text like it was going to look in the
magazine.  They could then point to the word _blemish_ [show visual
feedback] and say "replace with scar"; "move paragraph down";
"reference to synonyms"; whatever.

IBM has a couple of working, commercial systems that (IMHO) provide
satisfactory performance; they are demoing them in medical and legal
transcribing applications and are getting GREAT accuracy rates (I believe its
over 90%).  The big vocabulary system runs on AIX/RS6000 stuff (on a wrist
near you in a matter of months - :^)     The small vocab system (which is at
least over 1000 speaker independent words, maybe more, I don't have lit
with me) runs onder OS/2 - and i486 computers the size of a Sharp Wizard
are on their way here...

Any who, I think that there needs to be some thought about speech
recognition technologies in VR.  The RECOGNITION part is here now - don't
become bogged down in the lexical/natural language parsing stuff right
now.  All the transciption stuff that IBM has now, for example.  The
system doesn't interpret AT ALL;  it does, however, translate and
re-represent character/word/sentence data very accurately.  Enough for your
word processing (the system even does context checking to know when you
mean 'our' instead of 'hour' and the like).  And Enough to string together
commands like 'copy this [pointing gesture] to this [another pointing gesture]
container option replace older'.

Now if I could only find room for an RS600 in my loft.... :)

>What do y'all think?  ApK

Just my ideas - You had to ask.

kristof-
nybakken@world.std.com
