Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!dali.cs.montana.edu!milton!hlab
From: billk@sco.COM (William K. Karwin)
Newsgroups: sci.virtual-worlds
Subject: Re: Recognizing sign language
Message-ID: <1991May2.041620.15724@milton.u.washington.edu>
Date: 1 May 91 21:24:47 GMT
References: <1991May1.015034.16232@milton.u.washington.edu>
Sender: hlab@milton.u.washington.edu (Human Int. Technology Lab)
Organization: The Santa Cruz Operation
Lines: 36
Approved: cyberoid@milton.u.washington.edu




growf@ucscl.ucsc.edu (Purple Dragons!  EVERYWHERE!!!) writes:

>
>In article <> kovach@rtc.atk.com (Pete Kovach) writes:
>
>[N.B. I don't really know what I'm talking about here, but the little
> lights in my head lit up when I read about this application, so
> I got wild and posted a reply.]
>
>>The sifficult area is recognizing sign language as a whole.
>
>Hm.  Given that I'm not an expert on sign language, it seems to me
>the two things you'd be looking to recognize would be a) hand positioning
>and b) arm positioning.

There's also:
        c) Facial expression, eyebrow movement, facial tension.
        d) Body motion; leaning forward while raising eyebrows
           is a common way to indicate an interrogative.
        e) Exaggeration of signs for emphasis.
        f) Many signers perform specific signs the way they
           first learned them.  Signs for "mommy" and "daddy"
           often appear clumsy and childlike, even for clear,
           adult signers.  How do you program a computer to
           recognize signs even if they are "slurred" like this?

Not to criticize you, growf (you did acknowledge that you weren't
an expert on ASL), but signing is a full-body language.  For this
reason, I think true sign translation machines will be a long way
off.  There's a reason most work has been done in fingerspelling only.
-- 
William Karwin                               ARPA: billk@sco.COM
C code.  C code run.  Run, code, run!        UUCP: ...!uunet!sco!billk