15.636 handwriting recognition

From: Humanist Discussion Group (by way of Willard McCarty (w.mccarty@btinternet.com)
Date: Sun May 05 2002 - 02:24:33 EDT

  • Next message: Humanist Discussion Group (by way of Willard McCarty : "15.637 Urgent Business Proposals"

                   Humanist Discussion Group, Vol. 15, No. 636.
           Centre for Computing in the Humanities, King's College London
                   <http://www.princeton.edu/~mccarty/humanist/>
                  <http://www.kcl.ac.uk/humanities/cch/humanist/>

       [1] From: "C. Perry Willett" <pwillett@indiana.edu> (21)
             Subject: Re: 15.635 OCR for handwriting?

       [2] From: "Al Magary" <al@magary.com> (10)
             Subject: Re: 15.635 OCR for handwriting?

       [3] From: Willard McCarty <w.mccarty@btinternet.com> (39)
             Subject: handwriting recognition

    --[1]------------------------------------------------------------------
             Date: Sun, 05 May 2002 06:43:52 +0100
             From: "C. Perry Willett" <pwillett@indiana.edu>
             Subject: Re: 15.635 OCR for handwriting?

    Things may not be as hopeful as they appear. If you follow the
    link up to the cover page of the report that Willard notes,
    you'll see that it's from 1996. In fact, research on OCR has
    dropped off a good deal since then. If you look at the Information
    Science Research Institute at Univ of Nevada, Las Vegas
    <http://www.isri.unlv.edu> or the Center of Excellence for
    Document Analysis and Recognition at Univ of Buffalo
    <http://www.cedar.buffalo.edu> you'll note that their publications
    have tailed off in recent years. The Institute at UNLV used to produce
    an annual report on OCR accuracy, but hasn't done one since 1996. I
    don't think OCR for either printed or handwritten texts has advanced
    much in the past few years.

    Perry Willett
    Main Library
    Indiana University
    pwillett@indiana.edu

    On Sat, 4 May 2002, Humanist Discussion Group (by way of Willard McCarty
                    <w.mccarty@btinternet.com>) wrote:

    > A brief look around the Web turns up, for example, a report from the Center
    > for Spoken Language Understanding, Oregon Graduate Institute of Science and
    > Technology, "OCR: handwriting",
    > <http://cslu.cse.ogi.edu/HLTsurvey/ch2node6.html>, which makes one hopeful.

    --[2]------------------------------------------------------------------
             Date: Sun, 05 May 2002 06:44:15 +0100
             From: "Al Magary" <al@magary.com>
             Subject: Re: 15.635 OCR for handwriting?

    > ...at the moment, I am strongly considering committing myself to
    > producing an authorised edition of ca 1,000 unpublished letters by
    Huxley.
    > However, I have no funding set up, and since many of the letters are
    > handwritten, I was wondering if you have any more recent information
    about
    > OCR software that recognizes handwriting?

    A Google search for material about Apple's Newton would probably turn up
    quite a bit about the problems of handwriting recognition, and maybe some
    solutions that evaded Apple.

    Al Magary

    --[3]------------------------------------------------------------------
             Date: Sun, 05 May 2002 07:10:07 +0100
             From: Willard McCarty <w.mccarty@btinternet.com>
             Subject: handwriting recognition

    Two points, neither of which will be comforting to the questioner, I'm afraid.

    (1) Hand-transcription, i.e. typing from the mss., has much to recommend it
    in comparison to OCR. For mss. of course no off-shore company would be able
    to deliver the well-known favourable rates of accuracy at lost cost for
    printed material -- which is, I understand, why so many projects elect to
    have their printed material digitized in that way. But hand-transcription
    is one way of becoming very familiar with the material, and with sufficient
    preparation can be done simultaneously with some amount of encoding. Any
    comments on this?

    (2) As a friend commented to me, the brilliance of the
    handwriting-recognition scheme used by the Palm and similar devices, in
    contrast to the Newton's, is that it assigns to the human user what he or
    she is good at by nature -- learning an arbitrary (but intelligently
    designed) set of gestures -- and to the machine what it is, at this stage
    of development, as much as such a thing can handle. I use the "Graffiti"
    system almost daily on my Palm, while making the trip into Central London
    by tube, for taking notes on the books I must devour for my research. I can
    write quite rapidly and accurately that way -- the Palm filters out the
    irregularities that were all too visible in the handwritten notes I used to
    take on 3x5 cards, allows me to write much more in each note and delivers
    the lot to my larger machines already in electronic form, so no labourious
    transcription -- which means that I actually do use the notes, rather than
    assign them to a graveyard-archive of curious artefacts which rapidly
    become unintelligible because each note was perforce so brief.

    Things like the Palm are becoming rapidly more powerful, of course. (As
    Bill Wulf once calculated, a musical greeting card chip has approximately
    the same amount of computing power as the ENIAC did; what a hand the Palm
    would once have required!) But for us I think it's rather important to see
    in the very clever combination of human intelligence with machine power a
    paradigm that won't change. The apparent goal of at least some work in CS
    is to get the human out of the picture. Which, I think, is one thing that
    Winograd and Flores, in Understanding Computers and Cognition, are strongly
    arguing against.

    Yours,
    WM

    Dr Willard McCarty, Senior Lecturer,
    Centre for Computing in the Humanities, King's College London,
    Strand, London WC2R 2LS, U.K.,
    +44 (0)20 7848-2784, ilex.cc.kcl.ac.uk/wlm/,
    willard.mccarty@kcl.ac.uk, w.mccarty@btinternet.com



    This archive was generated by hypermail 2b30 : Sun May 05 2002 - 02:36:46 EDT