From CCNET-L@uga.cc.uga.edu Mon Jun 22 14:08:57 1992
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.67/5.10) with SMTP id AA19931; Mon, 22 Jun 92 14:08:53 -0400
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 1828; Mon, 22 Jun 92 14:06:32 EDT
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.07) with BSMTP id 3823;
 Mon, 22 Jun 92 14:06:30 EDT
Date:         Mon, 22 Jun 1992 14:05:55 EDT
Reply-To: Charles W Tang <cwt2@CUNIXA.CC.COLUMBIA.EDU>
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: Charles W Tang <cwt2@CUNIXA.CC.COLUMBIA.EDU>
Subject:      OCR system for Chinese?
Comments: To: CCNET-L@uga.cc.uga.edu
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
Status: RO

Anyone know anything about some of the most popular OCR systems for
Chinese used in Taiwan?

Thanks,

Charles


  **** Om Ma-Ni Pe-Me Hung **** | "All the world's phenomena and ideas
  ==============================|  Are like a dream, like magic,
      Charles Wei-Ting Tang     |  and like a reflected image."
   cwt2@cunixa.cc.columbia.edu  |                 -- The Diamond Sutra

From cwt2@cunixa.cc.columbia.edu Sun Jun 28 21:38:27 1992
Received: from cunixa.cc.columbia.edu by sirius.ctr.columbia.edu (5.67/5.10) with SMTP id AA10947; Sun, 28 Jun 92 21:38:25 -0400
Received: by cunixa.cc.columbia.edu (5.59/FCB)
	id AA08201; Sun, 28 Jun 92 21:38:23 EDT
Date: Sun, 28 Jun 92 21:38:15 EDT
From: Charles W Tang <cwt2@cunixa.cc.columbia.edu>
To: Yuan Jiang <yjj@ctr.columbia.edu>
Subject: Re: OCR system for Chinese?
In-Reply-To: Your message of Mon, 22 Jun 92 14:38:29 EDT
Message-Id: <CMM.0.90.0.709781895.cwt2@cunixa.cc.columbia.edu>
Status: R

Hi!  I only got two replies about this.  One from Hui Zhao Wang:
------------------------
I don't know such kind of software in Taiwan, but I know in china, there is
program call Tesing chinese ocr developed by e.e. dept of Tesing univ.
the developer claimed it could recognize up to 98% multi-fonds pure chinese
document(no western character). I have found it weak in the document
which has digit, or chinese charater mix with western character. the ocr
result is gb text. no self-learing process inside(terrible).
 
hz
------------------------

The other from Avron Boretz, saying
------------------------
I would like to know about this too -- but according to the latest
info (a bit out of date, admittedly) that I have, a big consortium (with
government backing) was working on CHinese OCR, and it was not in a
very advanced state, probably nowhere near ready for commercial
release.  On the other hand, the Japanese have already come up with
commercial OCR softwa (but based on the proprietary NEC platform).
Avron
---------------------

Good luck!

-Charles


  **** Om Ma-Ni Pe-Me Hung **** | "All the world's phenomena and ideas
  ==============================|  Are like a dream, like magic,
      Charles Wei-Ting Tang     |  and like a reflected image."
   cwt2@cunixa.cc.columbia.edu  |                 -- The Diamond Sutra

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Thu Apr 26 10:56:08 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA01464; Thu, 26 Apr 90 10:56:03 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 2638; Thu, 26 Apr 90 10:53:58 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 6967; Thu, 26 Apr 90 10:53:57 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07B) with BSMTP id
 8723; Thu, 26 Apr 90 10:55:42 EDT
Date:         Thu, 26 Apr 90 09:55:36 EDT
Reply-To: Tin Kam Ho <hotin%CS.BUFFALO.EDU@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Tin Kam Ho <hotin%CS.BUFFALO.EDU@cuvmb.cc.columbia.edu>
Subject:      pattern recognition
Comments: To: CCNET-L@uga.cc.uga.edu
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

  For those who are interested chinese character recognition:
  Please refer to

  Shunji Mori, Kazuhiko Yamamoto, and Michio Yasuda,
  "Research on Machine Recognition of Handprinted Characters",
  IEEE Transactions on Pattern Analysis and Machine Intelligence,
  Vol. PAMI-6, No. 4, July 1984, pp 386-405.

  This paper contains a very good survey on methods for recognizing
  handprinted chinese characters.

  As a worker on a closely related topic,   I'll also be happy to
  discuss with you if you'd like to know more or want to share some ideas.

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Thu Apr 26 12:15:36 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA01971; Thu, 26 Apr 90 12:15:31 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 2730; Thu, 26 Apr 90 12:13:24 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 7209; Thu, 26 Apr 90 12:13:23 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07B) with BSMTP id
 3677; Thu, 26 Apr 90 12:13:55 EDT
Date:         Thu, 26 Apr 90 11:36:17 EST
Reply-To: "LI Jialin, UGA Physics" <LI%USCN@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: "LI Jialin, UGA Physics" <LI%USCN.BITNET@cuvmb.cc.columbia.edu>
Subject:      Re: patter recognition
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Talking about Chinese text input method, there is someone, Dong qiqi, working
on "speaking" recongnition. I was told, it's very "promising".

Scott, why didn't you include this into your wish list? :-)

Jialin

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Thu Apr 26 14:20:08 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA02642; Thu, 26 Apr 90 14:20:02 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 2878; Thu, 26 Apr 90 14:17:55 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 7647; Thu, 26 Apr 90 14:17:54 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07B) with BSMTP id
 6176; Thu, 26 Apr 90 14:19:36 EDT
Date:         Thu, 26 Apr 90 12:51:14 CDT
Reply-To: Sluggo <UD154342%NDSUVM1@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Sluggo <UD154342%NDSUVM1.BITNET@cuvmb.cc.columbia.edu>
Subject:      Chinese Spoken-input method
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Hi,

I recently heard that Chinese is the EASIEST to develop a spoken-input
method for, because Chinese has only 4,000 sounds.

(English has about 10,000 sounds).

Some computer labratory has invented a 'robot'(?) that responds to
commands in Chinese (I assume Mandarin). Supposedly the CHinese
is easiest...

Of course on a PC, you'd need a speech-input board of some kind.
And, hardware runs up the expense..  ugh!

                                       Sluggo

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Thu Apr 26 16:15:23 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA03265; Thu, 26 Apr 90 16:15:13 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 3003; Thu, 26 Apr 90 16:13:03 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 8037; Thu, 26 Apr 90 16:13:02 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07B) with BSMTP id
 3035; Thu, 26 Apr 90 16:14:47 EDT
Date:         Thu, 26 Apr 90 15:58:50 EDT
Reply-To: Scott Horne <horne-scott%YALECS@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Scott Horne <horne-scott%YALECS.BITNET@cuvmb.cc.columbia.edu>
Subject:      Re:  Chinese Spoken-input method
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

> because Chinese has only 4,000 sounds....  (English has about 10,000 sounds).

You mean syllables--in which case, by your argument (which, incidentally, is
not valid), it would be still easier to develop a speech-recognition system
for Japanese, as Japanese has only a couple hundred syllables.

                                        --Scott

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Thu Apr 26 16:36:31 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA03357; Thu, 26 Apr 90 16:36:27 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 3035; Thu, 26 Apr 90 16:34:20 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 8148; Thu, 26 Apr 90 16:34:19 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07B) with BSMTP id
 7281; Thu, 26 Apr 90 16:36:03 EDT
Date:         Thu, 26 Apr 90 13:52:20 -0400
Reply-To: Jay Sekora <js%PHOENIX.PRINCETON.EDU@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Jay Sekora <js%PHOENIX.PRINCETON.EDU@cuvmb.cc.columbia.edu>
Subject:      Re: patter recognition
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Li Jialin <LI@USCN.BITNET> quoth:
>Talking about Chinese text input method, there is someone, Dong qiqi, working
>on "speaking" recongnition. I was told, it's very "promising".

I would expect Chinese voice input to be *MUCH* easier than, say, English or
French voice recognition, because
(1) Chinese has relatively few consonants, and
(2) they are acoustically easy to distinguish (e.g., aspirated vs. unaspirated;
  cf. English ``voiced'' and ``voiceless'', which are really cluster concepts--
  `p'-sounds are often unaspirated and partly voiced, while `b'-sounds are
  often unvoiced),
(3) Chinese has vastly fewer consonant clusters, and vastly-fewer cross-
  syllable contiguous consonants, and, most importantly,
(4) Chinese is tonal.  It's a lot of trouble to get a computer to distinguish
  between a `t'-sound and a `p'-sound.  It's fairly easy to get a computer to
  distinguish between a falling pitch contour and a rising pitch contour.  Of
  course, Chinese uses both distinctions, but the point is that you can get,
  say, 80% of the phonetic information in a Chinese word (`syllable' would be
  an unfair comparison) far more easily than you can get 80% of the phonetic
  information in an English word.

I'd be interested in hearing more about Mr. Dong's work.  Is there a paper
reference?

-j.

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Wed Apr 25 11:44:35 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA26273; Wed, 25 Apr 90 11:44:31 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 1463; Wed, 25 Apr 90 11:42:27 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 3255; Wed, 25 Apr 90 11:42:26 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07B) with BSMTP id
 0270; Wed, 25 Apr 90 11:44:07 EDT
Date:         Wed, 25 Apr 90 11:27:00 EDT
Reply-To: "J.B. WANG" <SPO_JWANG2%STEVENS@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: "J.B. WANG" <SPO_JWANG2%STEVENS.BITNET@cuvmb.cc.columbia.edu>
Subject:      Re: Pattern Recognition
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

It really depends on what you would define as pattern recoginition. If you
ask if anybody has developed a algorithm that can recoganize a Chinese
character according to its image or strokes, I would say yes.

Pattern recogonition of Chinese characters may have formed its own way
un-intentionally due to some effort of creating stroke input schemes.
It does not follow the traditonal momentum of insertia, and center of gravity
concept. That concept has also brought English characters pattern recogonition
into a stagnation point since proportion fonts have all different
characteristics although same character is being scanned. This scheme will
never work for Chinese fonts and handwritten fonts. Some institutions of
mainland China have successfully produced pressure-sensitive boards on
which you write Chinese characters that will appear on the screen. I
would also believe that it is not too far away from recoganing characters
through a scanner.

J.B. Wang

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Fri Apr 27 18:12:13 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA04288; Fri, 27 Apr 90 18:12:08 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 4503; Fri, 27 Apr 90 18:10:00 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 2522; Fri, 27 Apr 90 18:09:58 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 4771; Fri, 27 Apr 90 18:11:40 EDT
Date:         Fri, 27 Apr 90 17:11:20 EDT
Reply-To: BUCKBD%MOREKYPR@cuvmb.cc.columbia.edu
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: BUCKBD%MOREKYPR.BITNET@cuvmb.cc.columbia.edu
Subject:      Signing off/Orthogonal functions?
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

     Oh yeah, and the serious bit.  I haven't kept up with character & speech
recognition methods much lately.  Does anyone know of anything that's being
done with systems of othogonal functions (like Walsh and Fourier) in those
applications?  (Since we're on the topic, and it does have big ramifications in
Chinese computing.)

     Ben Buckner   BUCKBD@MOREKYPR


Received: from TCSVM.BITNET by NDSUVM1.BITNET (Mailer R2.03B) with BSMTP id
 2045; Thu, 26 Apr 90 17:38:22 CDT
Received: by TCSVM (Mailer R2.05) id 0771; Thu, 26 Apr 90 17:38:42 CDT
Date:         Thu, 26 Apr 90 17:22:58 CDT
From:         Andrew Nulman <PL6AAIU@VM.TCS.TULANE.EDU>
Subject:      Re: Chinese Spoken-input method
To:           Sluggo <UD154342@NDSUVM1>
In-Reply-To:  Your message of Thu, 26 Apr 90 12:51:14 CDT


I believe spoken-input method is the easiest. However there is one problem
which must be overcome, and that is the standardization of the Mandarin
accents.
   I have heard that people in Taiwan and mainland have begun communicating
by telephone now that there are direct links available, but Beijing and
Taipei people are having trouble communicating with each other in Mandarin!

The advantage of Chinese in developing a phoneme-based system is the fact
that Mandarin only has 4 tones, plus the neutral tone.

Compare this with English.  The standard "newcaster's" accent is pretty much
the same all over the United States and Candada.  I guess a spoken-input
method would have to be localized for England, Ireland, Australia, N.Z, India,
.  Oh , yes, and Southern U.S.!


Japanese will be easier to develop spoken input method more than Chinese or
English.  It has very few phonemes, and is a more monotonal language.

The most difficult language to develop spoken-input method for is English,
because we borrow foreign sounds into our language.  Chinese and Japanese
borrow foreign words too, but don't absorb the foreign phonemes as well.
In this sense, these languages are more pure.

It isn't that urgent to develop spoken input for English, because keyboards are
designed for Roman-based languages.
   But the difficulty of computerizing symbolic languages, especially
ones without an alphabet like Chinese, is what has made technology like
FAX so popular in Asia.

      My prediction is that the first language which will have a usable,
cost-effective voice interpreter will be Japanese.

  Three reasons:  1) R&D funding is there, 2) Demand is strong, 3) the
pronunciation is very standardized, and there aren't a lot of phonemes.

                                 -Andy Nulman

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Mon Apr 30 17:19:19 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA25067; Mon, 30 Apr 90 17:19:15 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 6582; Mon, 30 Apr 90 17:17:01 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 8732; Mon, 30 Apr 90 17:17:00 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 7671; Mon, 30 Apr 90 17:17:36 EDT
Date:         Fri, 27 Apr 90 21:48:10 EDT
Reply-To: Nobody is allowed to log on to this machine
              <yjj@sirius.ctr.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Nobody is allowed to log on to this machine
              <yjj@sirius.ctr.columbia.edu>
Subject:      Stroke Recognition (Pattern Recognition)
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Different from pattern recognition, stroke recognition is very
easy for Chinese input.  There are many products, writing pads,
on the market.

Office Plus in New York City, the first distributor for BrushWriter,
developed its writing pad and tend to pursuade BrushWriter as the
standard input method.  I'm not sure about the current status.

Dr. Yan YongXin, the creater of Byx, once told me that one of his
friend wrote a compact and powerful driver (about tens of KB)
for writing pad input.  It was very fast too.

Stroke recognition may be good for well educated Chinese, who
learned the standard stroke sequences in elementary school.
But it's hard for others.

Yuan

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Mon Apr 30 19:26:31 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA25602; Mon, 30 Apr 90 19:26:27 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 6708; Mon, 30 Apr 90 19:24:13 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 9149; Mon, 30 Apr 90 19:24:12 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 5014; Mon, 30 Apr 90 19:26:05 EDT
Date:         Mon, 30 Apr 90 17:57:48 EDT
Reply-To: Scott Horne <horne-scott%YALE.ARPA@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Scott Horne <horne-scott%YALE.ARPA@cuvmb.cc.columbia.edu>
Subject:      Re:  Stroke Recognition (Pattern Recognition)
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Well, Yuan, even well-educated Chinese people don't all follow the same stroke
orders.  Compare the way most mainlanders write `xian1sheng' de `sheng1' to
the way most Taiwanese write it.  And how do you write `ao1' `concave' and
`tu1' `convex'?

                                        --Scott

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Mon Apr 30 21:27:45 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA26208; Mon, 30 Apr 90 21:27:40 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 6781; Mon, 30 Apr 90 21:25:26 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 9394; Mon, 30 Apr 90 21:25:25 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 3011; Mon, 30 Apr 90 21:27:16 EDT
Date:         Mon, 30 Apr 90 21:22:36 EDT
Reply-To: BUCKBD%MOREKYPR@cuvmb.cc.columbia.edu
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: BUCKBD%MOREKYPR.BITNET@cuvmb.cc.columbia.edu
Subject:      Stroke-order Input Systems
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

     With this stroke order system, is the idea that you input standard strokes
and the program uses this sort of "stroke vector" to characterize a character?
I can see where that could be a problem, but people who can't spell are in a
similar position with alphabetic languages.  Can the average Chinese-literate
person write faster than they can input stroke order?  And which can trained
stroke-order input users do faster?

BTW: Where my last post said "othogonal functions", it should have said
"orthogonal functions."  I must have been thinking in some r-drop dialect.
And doesn't anyone know how to use a Listserver?

Ben Buckner    BUCKBD@MOREKYPR

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Mon Apr 30 23:32:14 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA26756; Mon, 30 Apr 90 23:32:09 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 6836; Mon, 30 Apr 90 23:29:55 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 9600; Mon, 30 Apr 90 23:29:53 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 3285; Mon, 30 Apr 90 23:31:46 EDT
Date:         Mon, 30 Apr 90 17:18:26 -0700
Reply-To: Ed Lai <lai%APPLE.COM@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Ed Lai <lai%APPLE.COM@cuvmb.cc.columbia.edu>
Subject:      Re:  Stroke Recognition (Pattern Recognition)
Comments: To: CCNET-L%UGA.BITNET@CUNYVM.CUNY.EDU
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

A stroke order input method does not imply that a character must be written
in one and only one way, otherwise it would not be very useful. For one thing
it is possible to have multiple entries for the same character denoting the
different writing order. Of course this can quickly get out of hand,d then the
trick is, when the writing order is not exactly the same, how to get the still
be able to make an educated guess. BrushWriter is already doing that, Apple's
stroke input method is based on BrushWriter is even better, and I am quite
sure it can be further improved along the same line.

/* Disclaimer: All statments and opinions expressed are my own */
/* Edmund K. Lai                                               */
/* Apple Computer, MS77-E                                      */
/* 20525 Mariani Ave,                                          */
/* Cupertino, CA 95014                                         */
/* (408)974-6272                                               */
zW@h9cOi

From @cuvmb.cc.columbia.edu:horne-scott@YALE.ARPA Mon Apr 30 23:34:30 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA26763; Mon, 30 Apr 90 23:34:26 EDT
Received: from SUNED.ZOO.CS.YALE.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with TCP; Mon, 30 Apr 90 23:32:09 EDT
Received: by SUNED.ZOO.CS.YALE.EDU; Mon, 30 Apr 90 23:33:10 EDT
Date: Mon, 30 Apr 90 23:33:10 EDT
From: Scott Horne <horne-scott@YALE.EDU>
Message-Id: <9005010333.AA13364@SUNED.ZOO.CS.YALE.EDU>
Received: by yale-zoo-suned (jaguar) 
          via WIMP-MAIL (Version 1.3/1.7) ; Mon Apr 30 23:33:15
To: yjj%ctr.columbia.edu@cuvmb.cc.columbia.edu
Subject: Re:  Stroke Recognition (Pattern Recognition)
Status: RO

I said that I don't care much about input methods because other features of
a word processor are more important to me.  Give me a big, expandable character
set containing both fan2ti3zi4 and jian3ti3zi4, and I'll be happy.  Give me
my favourite input method but a small character set or one with only fan2ti3zi4
(or only jian3ti3zi4), and I won't be.

Another good example of a character with various stroke orders is `tian2'
`field'.  In the mainland, it's  -  7  -  |  -   ; in Taiwan and Japan, it's
-  7  |  -  -    .

                                        Best,
                                        Scott



From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Tue May  1 00:02:04 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA26895; Tue, 1 May 90 00:02:00 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 6856; Mon, 30 Apr 90 23:59:46 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 9655; Mon, 30 Apr 90 23:59:44 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 7825; Tue, 01 May 90 00:01:18 EDT
Date:         Mon, 30 Apr 90 23:49:40 EDT
Reply-To: Scott Horne <horne-scott%YALE.ARPA@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Scott Horne <horne-scott%YALE.ARPA@cuvmb.cc.columbia.edu>
Subject:      Re:  Stroke Recognition (Pattern Recognition)
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Sure, you could have a system with multiple stroke orders accepted.  But,
as you said, that can quickly get out of hand.  It's trivial to accept the
two usual stroke orders of `tian2' `field', but a system which attempts
to recognise just any old stroke order (and some people [especially those
from Hong Kong, I've noticed] use very strange stroke orders for certain
characters) will 1) be inefficient; and 2) make lots of mistakes.

                                        --Scott

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Tue May  1 00:29:58 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA27103; Tue, 1 May 90 00:29:37 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 6875; Tue, 01 May 90 00:27:16 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 9719; Tue, 01 May 90 00:27:15 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 4339; Tue, 01 May 90 00:29:01 EDT
Date:         Mon, 30 Apr 90 20:40:00 EDT
Reply-To: "J.B. WANG" <SPO_JWANG2%STEVENS@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: "J.B. WANG" <SPO_JWANG2%STEVENS.BITNET@cuvmb.cc.columbia.edu>
Subject:      RE: Stroke Recognition (Pattern Recognition)
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Actually in stroke recognition the sequence of strokes can be come
non-crucial. In Mr. Wang, Yongmin's five-stroke input method, for instance,
the sequence is really not that important. A friend of mine at University
of Pittsburgh used a mouse to stroke over the pattern on the screen to
create an index dictionary for stroke input of Chinese; even in this fashion
it is possible to make sequence unimportant.

In fact it does not need good or advanced education to follow strokes correctly
since it is taught in elementary schools only. But it does need one to have
a good writing habit, and a fundamental understanding of essential strokes.
In ao1 and tu1, as Scott mentioned, it would not be difficult to stroke it
correctly if you know how to write kou3 (mouth), i.e., vertical down first,
horizontal left to right closepath last, and the rest in the middle.

Here I would like to brief a piece of news appeared on p21 of April 27,
1990 issue of World Journal, which reported that a mainland Chinese who
now has his bussiness in Silicon Valley, is trying to sell his Chinese
pattern recognition technology to Taiwan manufacturers. Reportedly,
a reported research result of China's Telecommication Ministry had a
pattern recognition scheme employing laser beaming using rotational momentum
of inertial and another crition to allow 100% of acurracy on recognizing
6000 Chinese characters. A research result of Harbin Polytech showed
98% accuracy in speech recognition. It also hinted that research in such
aspects in mainland China lead Taiwan ('s technology) a great deal, which
is not a supprise to us.

In USA and Europe, I think we should take advantage of the advance and heat
of desktop publishing and TeX/LaTeX, as well as the success of Adobe PostScript
in standardizing output description. No matter how cheap the labor in mainland
China is, a laser printer will still remain a luxury. To be more accurate,
if we concentrate on Chinese text formatting and natural keyboard inputting,
the only ones who have conditions to compete with us are Singapore scholars
and commercial circles in the west.


JB Wang

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Fri May  4 03:03:57 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA19028; Fri, 4 May 90 03:03:53 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 0979; Fri, 04 May 90 03:01:33 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 1453; Fri, 04 May 90 03:01:32 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 5953; Fri, 04 May 90 03:03:23 EDT
Date:         Thu, 3 May 90 20:10:40 EST
Reply-To: Qiru Zhou <ZHOUQI%DUPR@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Qiru Zhou <ZHOUQI%DUPR.BITNET@cuvmb.cc.columbia.edu>
Subject:      Stroke Recognition (Pattern Recognition)
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Stroke recognition input for Chinese (or Japanese for the
Ch character part) is practical for the small amount
of information inputs, like take information or fill
up spreadsheet by computer novices. I saw some of these
devices (like a small tablet or digitizer) market available
several years ago with supporting software in PRC. They
are not depend on any writing habit or orders. The problem
is that kind of input device is not good for vast, fast input,
like typing a paper, etc. The same situation as in voice
recognition (this is more difficult problem for any language).

The stroke recognition algorithm is faster than voice
recognition. But still it need to speed up. I try to think
to do something on it by using so called "neural network"
structure. BTW, our lab. now have a project in speech
recognition by "neural network" systems. Although it
deal with English, the idea can be used in any other languages.

Qiru Zhou,
Dept. of ECE
Drexel University
Phila. PA

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Mon May 21 18:12:24 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA22910; Mon, 21 May 90 18:12:19 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 7807; Mon, 21 May 90 18:09:25 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 6153; Mon, 21 May 90 18:09:23 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 4859; Mon, 21 May 90 18:11:14 EDT
Date:         Mon, 21 May 90 18:09:58 EDT
Reply-To: Huangxin Wang <wang%pennmess.physics%UPENN.EDU@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Huangxin Wang <wang%pennmess.physics%UPENN.EDU@cuvmb.cc.columbia.edu>
Subject:      Sound Input for Chinese OS on Mac
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

I wonder what's the qualification about the sound input method for Chinese,
as discussed on the net.  I think there are two cases:  (1) computer is able
to choice the character by recognizing the voice, and through context.  This,
I think is pretty difficult, though my impression is most discussions are
on this aspect.   (2)  Computer recognizes the voice, then list all the
possible charcters corresponding to this voice.  The user will make the final
choice about which charcter is the right one.   This kind is quite similar
to the Chinese OS on Mac:  you type in the pin-yin, even part of it,  the
computer will list all the possiblity corresponding to that part (or whole
word) of pin-yin.  Say, even when you type in a "f" you will get a list.

        My conjecture is that for the current technology, it should not
be too difficult to achieve the second case.  Is a voice recognition device
able to tell the difference between "fa" and "pa"?  The list corresponding
to the sound may be more strict, or more liberal, say, it may or may not
distinguish the tones, or "s" and "sh",  "ng" and "n", etc.  Since
on Mac Chinese OS, one can mixing the input methods of pin-yin and bi3-hua4,
then voice might be another method which can be used to mix with other common
input methods to accelerate the input.

        Do I under-estimate the difficulties of the second case?  Or are
there anything available already?   I see the Mac a more suitable machine
for this method, since its liberal input method make it very natural to extend
to this kind of voice recognization.

                Huangxin Wang of University of Pennsylvania

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Mon May 28 10:05:06 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA18602; Mon, 28 May 90 10:05:02 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 2116; Mon, 28 May 90 10:04:58 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 1643; Mon, 28 May 90 10:04:57 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 3854; Mon, 28 May 90 10:04:31 EDT
Date:         Sat, 26 May 90 07:18:24 -0700
Reply-To: Ed Lai <lai%APPLE.COM@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Ed Lai <lai%APPLE.COM@cuvmb.cc.columbia.edu>
Subject:      Re: Sound Input for Chinese OS on Mac
Comments: To: CCNET-L%UGA.BITNET@cunyvm.cuny.edu
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

I agree that two next step to take in using voice input technology is to use
it as a keyboard replacement (when you say "pin", p i n is typed) and not
complete translation to Chinese. This is certainly doable today. The Mac is
probably not any better than other machine in this aspect. The concept of the
Chinese Mac is probably closer in spirit to such a scheme. The example given
in the pinyin input in the Chinese Mac would not contribute directly to voice
input. However, the very concept that when you typing in a character, you are
not trying to encode the Chinese character, but rather you are trying to narrow
down the range of possible character, is an important concept for the new
technology. Actually we have far more to learn from Chinese mac's stroke input
method than pinyin input method which is relevant here. The stroke input method
would prefer you to type in the strokes in the order they are normally written,
but if that is wrong (either because the order is wrong, some stroke is wrong,
some is missing or extra), it still try its best to guest what you mean.
If we translate the concept back to a pinyin input method. What it means is
this: today the best pinyin input can only do something like this, if you
type pinyin (with tones optional), it give you the phrase and save you from
choosing from numerous homonyms. But if you type "pinyn" or "pinying" you
could not get the right phrase in any existing pinyin input method. When
such input method exists, we are very clode to our goal.

/* Disclaimer: All statments and opinions expressed are my own */
/* Edmund K. Lai                                               */
/* Apple Computer, MS77-E                                      */
/* 20525 Mariani Ave,                                          */
/* Cupertino, CA 95014                                         */
/* (408)974-6272                                               */
zW@h9cOi

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Tue May 29 23:10:26 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA28174; Tue, 29 May 90 23:10:23 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 3531; Tue, 29 May 90 23:10:17 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 5976; Tue, 29 May 90 23:10:16 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 0804; Tue, 29 May 90 23:04:36 EDT
Date:         Tue, 29 May 90 17:10:46 PDT
Reply-To: portal!cup.portal.com!James_T_Caldwell%SUN.COM@cuvmb.cc.columbia.edu
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: portal!cup.portal.com!James_T_Caldwell%SUN.COM@cuvmb.cc.columbia.edu
Subject:      sound
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

Sound input for Chinese is not a new concept and has come a long way based
on the phrase input methods of TianMa and Sindex (Oxford University). Since
Chinese has only 400+ basic sounds, and since phrased input can distinguish
among characters based on context without tone designators, it is not a big
leap to sound input.  Probably easier than English sound input.  The
software would still have to be trained for each speaker, but only 400+
syllables (which would have to be spoken clearly, one term or phrase at a
time).  Nevertheless, it will be expensive until mass produced.  The SINDEX
people have spent years reasearching the characteristics of sound input for
Chinese.  It still has to be commercialized: packaged for the market in a
debugged format with appropriate hardware.   If someone is interested in
financing that project, I am sure they would be eager to go ahead.  They
have been looking for funding for at least two years now since I talked
with them in London; TianMa could not come up with the money at that time.
Jim Caldwell

From @cuvmb.cc.columbia.edu:CCNET-L@UGA.BITNET Tue May 15 00:36:40 1990
Received: from CUVMB.COLUMBIA.EDU (cuvmb.cc.columbia.edu) by sirius.ctr.columbia.edu (4.1/5.10) id AA17274; Tue, 15 May 90 00:36:36 EDT
Received: from CUVMB.COLUMBIA.EDU by CUVMB.COLUMBIA.EDU (IBM VM SMTP R1.2.1) with BSMTP id 1531; Tue, 15 May 90 00:33:55 EDT
Received: from PUCC.PRINCETON.EDU by CUVMB.COLUMBIA.EDU (Mailer R2.05) with
 BSMTP id 6470; Tue, 15 May 90 00:33:54 EDT
Received: from PUCC.BITNET by PUCC.PRINCETON.EDU (Mailer R2.07) with BSMTP id
 1715; Tue, 15 May 90 00:36:09 EDT
Date:         Mon, 14 May 90 21:30:57 PDT
Reply-To: Qiang Wan <wan%pollux.usc.edu%USC.EDU@cuvmb.cc.columbia.edu>
Sender: Chinese Computing Network <CCNET-L%UGA@cuvmb.cc.columbia.edu>
From: Qiang Wan <wan%pollux.usc.edu%USC.EDU@cuvmb.cc.columbia.edu>
Subject:      Re: Any info on speech recognition in China?
Comments: To: ccnet-l@uga.cc.uga.edu
To: Multiple recipients of list CCNET-L <CCNET-L@uga>
Status: RO

When I was working on Recognition of Printed Chinese Characters in the
CS&E Dept of Qinghua University, another group was working on
recognition of spoken Chinese, and yet one other grad in the
Wu2Xian4Dian4 Dept was working on Chinese speaking system. The speech
recognition group was using Apple IIe computer. The project later
passed a "Ministry Level Appraisal" and, according to official news
broadcast, was adopted in the Beijing 114 telephone number index
assistance system. I am not sure if 114 operators in Beijing are still
using this system though.
The guy in the Radio Engr Dept assembled a system using a PC clone
(if I remember correctly). When I was paying a visit, the machine
was reading Li3 Bai2's poems. ("The moonlight was shining at the
bedside...")
The speech recognition system suffered high failure rate. The fact
is, you'll have to "train" it virtually every time you want to
use it. The joke that the group had was that system would recognize
your voice in the morning and it would reject you in the afternoon;
it would recognize your happy voice, but would be confused the
minute you change your mood. I would not be surprised if the newly
introduced system behaved the same way.
FYI, the first similar news I heard about "wonderful speech recog-
nition system" was back in the late 70's, when a computer could
recognize its developer's voice input of math formula. The news
was carried on TV. The designer had a heavy accent. The computer
printed out a "+" sign when the gentleman read Jia4-Rang. I am
sure that if anybody said Jia1-Shang4, the system would crash.

Based on my limited experience, the stage of current speech
recognition systems in China are similar to those that are supposed
to predict earthquakes. Rumor has it that China predicted the
San Francisco earthquake, and informed the U.S. government.

wan qiang

From CCNET-L@uga.cc.uga.edu Wed Jul 17 16:33:36 1991
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.61/5.10) with SMTP id AA28906; Wed, 17 Jul 91 16:33:17 -0400
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP R1.2.2MX) with BSMTP id 5576; Wed, 17 Jul 91 16:33:20 EDT
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.07) with BSMTP id 0900;
 Wed, 17 Jul 91 16:33:18 EDT
Date:         Wed, 17 Jul 1991 13:26:53 PDT
Reply-To: terryk%MICROSOFT.COM@uga.cc.uga.edu
Sender: Chinese Computing Network <CCNET-L%UGA.BITNET@uga.cc.uga.edu>
From: terryk%MICROSOFT.COM@uga.cc.uga.edu
Subject:      References for Chinese OCR R&D
Comments: To: ccnet-l@uga.cc.uga.edu
To: Multiple recipients of list CCNET-L <CCNET-L@UGA>
Status: RO

Here is a list of references for Chinese OCR ( Optical Character Recognition )
R&D.

1.  Kenichi Mori and Isao Masuda, "Advances in recognition of Chinese
    characters," Proc. 5th Int. Conf. on Pattern Recognition, Miami,
    pp 692-702, 1980

2.  R.Casey and G.Nagy, "Recognition of printed Chinese character,"
    IEEE Trans. on Electric Computer. Vol EC-15. No. 1, pp.91-101, 1966.

3.  P.Wang and R.Shian, "Machine recognition of printed Chinese characters
    via transformation algorithms," Pattern Recognition, Vol.5, pp.303-321,
    1973.

4.  S.Yamamoto, A. Nakajima, K.Nakata, "Chinese character recognition by
    hierarchical pattern matching," Proc. 1th Int. Conf. on Pattern
    Recognition, pp.187-196, 1973.

5.  M.Umeda, "Recognition of multi-font printed Chinese characters," Proc.
    6th Int. Conf. on Pattern Recognition, pp. 793-796, 1982.

6.  P.Chen, Y.Chen, W.Hsu, "Stroke relation coding -- a new approach to the
    recognition of multi-font printed Chinese characters," International
    Journal of Pattern Recognition and Artifical Integlligence,
    Vol 2.No.1 pp 149-160, 1988.

7.  S.Zhang, B.Taconet, A.Faure, "A Complexity Measure based Algorithm for
    multi-font Chinese character recognition," Proc. 10th, Int. Conf. on
    Pattern Recognition, VOL.1, pp.573-577, 1990.

From CCNET-L@uga.cc.uga.edu Wed Jul 17 19:14:52 1991
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.61/5.10) with SMTP id AA29548; Wed, 17 Jul 91 19:14:44 -0400
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP R1.2.2MX) with BSMTP id 6096; Wed, 17 Jul 91 19:14:47 EDT
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.07) with BSMTP id 5283;
 Wed, 17 Jul 91 19:14:43 EDT
Date:         Wed, 17 Jul 1991 19:08:10 EDT
Reply-To: Yuan Jiang <yjj%CTR.COLUMBIA.EDU@uga.cc.uga.edu>
Sender: Chinese Computing Network <CCNET-L%UGA.BITNET@uga.cc.uga.edu>
From: Yuan Jiang <yjj%CTR.COLUMBIA.EDU@uga.cc.uga.edu>
Subject:      Voice recognition from Taiwan
To: Multiple recipients of list CCNET-L <CCNET-L@UGA>
Status: RO

According to a piece of new on Tue 7/16 World Journal
(in the middle of the first section),
a company in Taiwan, AiSiDe Limited (bad translation, please correct me)
announced a voice input card (for IBM PC?)  It claims 99% accuracy.  The
company is at
Taiwan, TaiBei County, Wu1Gu3 Vallege, ZhongXin Road, Section 1,
12 88th Alley.  The phone number is (02)982 7430.

From CCNET-L@uga.cc.uga.edu Fri Nov 13 13:10:52 1992
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.67/5.10) with SMTP id AA18043; Fri, 13 Nov 92 13:10:36 -0500
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 1699; Fri, 13 Nov 92 13:06:53 EST
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.08 PTF008) with BSMTP id
 0073; Fri, 13 Nov 92 13:06:52 EST
Date:         Fri, 13 Nov 1992 12:21:00 EST
Reply-To: tkh@RESEARCH.ATT.COM
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: tkh@RESEARCH.ATT.COM
Comments: To: ccnet-l@uga.cc.uga.edu
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
Status: R

Dear Guojin,  I see your posting on CCNET about the PH corpus.
We are a group of OCR researchers who would like to use such a corpus.
Could you give me further information on how to obtain a copy of it?
Please email me at
tkh@research.att.com

or fax to me at  (908) 582-5857
or mail to me at

Tin Kam Ho, AT&T Bell Laboratories, 600 Mountain Ave, Murray Hill, NJ07974,USA

I'll appreciate any help you could provide,
Tin.

From CCNET-L@uga.cc.uga.edu Wed Nov 18 17:52:16 1992
Received: from uga.cc.uga.edu by sirius.ctr.columbia.edu (5.67/5.10) with SMTP id AA10880; Wed, 18 Nov 92 17:51:45 -0500
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 2257; Wed, 18 Nov 92 17:46:43 EST
Received: from UGA.BITNET by UGA.CC.UGA.EDU (Mailer R2.08 PTF008) with BSMTP id
 7349; Wed, 18 Nov 92 17:46:41 EST
Date:         Wed, 18 Nov 1992 17:44:01 -0500
Reply-To: John McRae <jrm@CRUX2.CIT.CORNELL.EDU>
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: John McRae <jrm@CRUX2.CIT.CORNELL.EDU>
Subject:      Re: Q: Optical Character Recognition for Chinese?
Comments: To: Roger Hart <hartr@nicco.sscnet.ucla.edu>
Comments: cc: ccnet-l <ccnet-l@uga.cc.uga.edu>
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
In-Reply-To:  Your message of "Wed, 18 Nov 92 08:28:06 PST."
Status: RO

>I was wondering if anyone had information about optical character recognition
>for Chinese (or Japanese) character, especially cost, accuracy and
>possible future developments?
>
>
>Thanks,
>
>
>Roger Hart
>UCLA History
============================
The only hardware/software I've heard of that works successfully
is NEC 9801 and Saijichoo [the kanji mean "select-character-notebook,"
or caizizhang in Chinese].  This is being used by colleagues in Japan
to input large sections of the Chinese Buddhist canon, which has the
virtue of being a humongous amount of text in a consistent font.

I can't really tell you much more about this software, except that
the colleagues referred to train students to run it, and once they're
familiar with it they work two computers at a time.  Each machine
buzzes along until it can't recognize a character and then displays
that character's bitmapped image on the screen, with a beep and a
message "kono ji ga wakaranai" [wo bu zhidao zhege zi].  The
operator looks it up, the machine adds it to its shape dictionary,
and goes back to buzzing.

I won't give you the rates claimed for time and accuracy.  However,
I'll be visiting these folks in a couple of weeks, so perhaps I'll
be able to send an update.  In the meantime, if there's anyone out
there with contacts in Japan, could you try to find out if this
program will be ported over from the isolated NEC standard to
JDOS/V?  I think they would sell a whole bunch of copies of it
around the world if they did!  (That's a none-too-subtle hint to
be conveyed to the software makers!)

Happy computing.

-- John M.

==================================================================
John R. McRae
Assoc. Professor of East Asian Religions
Department of Asian Studies
Cornell University
Ithaca, NY  14853-2502
607/255-1328
fax:   -1686
e-mail:  jrm@crux2.cit.cornell.edu  (Internet)
         jrm@cornella.bitnet          (Bitnet)

P.S.  Mail sent to the jrm@cornella address will
automatically be forwarded to crux2.
==================================================================

From @UGA.CC.UGA.EDU:owner-ccnet-l@UGA.CC.UGA.EDU Fri Jul 23 10:43:02 1993
Return-Path: <@UGA.CC.UGA.EDU:owner-ccnet-l@UGA.CC.UGA.EDU>
Received: from uga.cc.uga.edu by ifcss.org (4.1/IFCSS-Mailer)
	id AA05828; Fri, 23 Jul 93 10:42:53 CDT
Message-Id: <9307231542.AA05828@ifcss.org>
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 2013; Fri, 23 Jul 93 11:41:49 EDT
Received: from UGA.CC.UGA.EDU (NJE origin LISTSERV@UGA) by UGA.CC.UGA.EDU (LMail V1.1d/1.7f) with BSMTP id 6577; Fri, 23 Jul 1993 11:41:47 -0400
Date:         Fri, 23 Jul 1993 10:40:47 -0500
Reply-To: Peter V Henstock <henstock@ECN.PURDUE.EDU>
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: Peter V Henstock <henstock@ECN.PURDUE.EDU>
Subject:      Re: Chinese OCR on UNIX (Summary)
Comments: To: CCNET-L@UGA.BITNET, g53150@SAKURA.KUDPC.KYOTO-U.AC.JP
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
Status: R

There is one package I'm aware of for Japanese text which will run
on a Macintosh.  It is called MRJ (MacReader Japanese) or something
like that.  They had a character-by-character version out a few
years ago for US$2000 or so which worked very fast on a regular
Macintosh system with the Kanjitalk operating system.  They now
have a new version which uses a dictionary to obtain 98% accuracy.
It would be beneficial to probably convert the JIS to a Chinese
standard and use that software, which would more than likely be
much faster than just typing.

--Peter

From @UGA.CC.UGA.EDU:owner-ccnet-l@UGA.CC.UGA.EDU Wed Jul 28 23:10:54 1993
Return-Path: <@UGA.CC.UGA.EDU:owner-ccnet-l@UGA.CC.UGA.EDU>
Received: from uga.cc.uga.edu by ifcss.org (4.1/IFCSS-Mailer)
	id AA18822; Wed, 28 Jul 93 23:10:42 CDT
Message-Id: <9307290410.AA18822@ifcss.org>
Received: from UGA.CC.UGA.EDU by uga.cc.uga.edu (IBM VM SMTP V2R2)
   with BSMTP id 5131; Thu, 29 Jul 93 00:09:28 EDT
Received: from UGA.CC.UGA.EDU (NJE origin LISTSERV@UGA) by UGA.CC.UGA.EDU (LMail V1.1d/1.7f) with BSMTP id 5694; Thu, 29 Jul 1993 00:09:26 -0400
Date:         Wed, 28 Jul 1993 23:09:44 CDT
Reply-To: Yuan Jiang <jiang@ifcss.org>
Sender: Chinese Computing Network <CCNET-L@uga.cc.uga.edu>
From: Yuan Jiang <jiang@ifcss.org>
Subject:      OCR software from Singerpore
Comments: To: CCNET-L@uga.cc.uga.edu
To: Multiple recipients of list CCNET-L <CCNET-L@uga.cc.uga.edu>
Status: R

I got a little information about an OCR software from a netter.
=======
(The ~{~} stuff is HZ coded Chinese for your reference)

Brand name: AUTOSTYLE Chinese OCR ~{0B6`K<4oS!K"::WVJ61pO5M3~}
by ~{?*7"IL~}: ~{PG<SFB~} ~{1vA*FsR5~}
        PIN LIANG ENTERPRISE
author:~{QPVFU_~}: ~{A;PqCy~}
        LIANG XUMING
Address:~{A*O55XV7~}:
BLOCK 92 ZION ROAD #01-197
SINGAPORE 0316

Tel:~{5g;0~}: (65) 7336433, 7341717
Fax:~{4+Uf~}: (64) 2356748

From guojin@iss.nus.sg Tue Aug  3 05:38:34 1993
Return-Path: <guojin@iss.nus.sg>
Received: from bochap.iss.nus.sg by ifcss.org (4.1/IFCSS-Mailer)
	id AA19297; Tue, 3 Aug 93 05:38:28 CDT
Received: from badger.iss.nus.sg by bochap.iss.nus.sg (4.1/SMI-4.1)
	id AA12153; Tue, 3 Aug 93 18:39:24+080
Message-Id: <9308031039.AA12153@bochap.iss.nus.sg>
To: Edith Wu <Edith@ULI.MSMAIL.CUHK.HK>, Yuan Jiang <jiang@ifcss.org>,
        Bob Chalmers <earth@NANGUO.XANADU.DIALIX.OZ.AU>,
        M H Foo <foo@MHFOO.PC.MY>, Peter V Henstock <henstock@ECN.PURDUE.EDU>,
        christian wittern <g53150@SAKURA.KUDPC.KYOTO-U.AC.JP>,
        rulst@lis.pitt.edu (Ruqing Li)
Cc: guojin@iss.nus.sg
Subject: OCR Software for Chinese
Date: Tue, 03 Aug 93 18:39:22 +0800
From: guojin@iss.nus.sg
Status: RO

Hi,

If you are interested in the Chinese OCR package below, you can ask for a demo
version by sending an application email to cc-ocr@brahms.iss.nus.sg.

> (The ~{~} stuff is HZ coded Chinese for your reference)

Guo Jin
-----------

~{!!!!!!!!!!!!!!!!PBPGS!K"::WVJ61pHm<~#(PBPGVPND#O#C#R#)<r=i~}


~{!!!!1>NDOr6AU_=iI\R;VVD\WT6/1fHOSIM<NDI(ChRG#(~}Scanner~{#)I(ChJdHk5DM<Oq~}~
~{ND<~VP8wVVS!K"::WV5D5gDTHm<~!!#-#-!!PBPGS!K"::WVJ61pHm<~#(R`3FPBPGVPND#O#C~}~
~{#R#)!#~}

~{!!!!PBPGVPND#O#C#R?IRT8_>+6H5XJ61pS!K"::WV5D<r!"710K4s1jW<WVLe!!#-#-!!<rKN~}~
~{Le!"<r:ZLe!"<r?,Le!"<r7BKNLe#,<071CwLe!"71T2Le!"71?,Le:M71:ZLe!#!!K|6TVP5H~}~
~{S!K"VJA?RTIO5DVPNDND<~#,F=>yJ61pU}H7BJ4o#9#7#%!#!!K|TZ~} IBM AT/386-33 ~
~{IO5DJ61pKY6H4o#1#0WV#/Ck!#!!K|D\4s4sLa8_SC;'5D9$WwP'BJ#,J971VXN^N65DS!K"::~}~
~{WV5D5gDTTYJdHk9$Ww1d5CGaKI!"Sd?l!"8_P'~}!~{!!~}

~{!!!!PBPGVPND#O#C#R6AHkSIM<NDI(ChRG4S1(V=!"Ji?/!"TSV>;rFdK|S!K"::WVND<~IO2I~}~
~{</OB5DM<OqND<~#,R@4N4S8CM<OqND<~VP7VNv3v8w::WVM<Oq?i!"TY6T8w::WVM<Oq=xPPM<~}~
~{Oq1fJ6#,2"=+C?8vJ61p=a9{SCR;8v1jW<::WVDZBk1mJ>#,4S6x=+U{8vM<OqND<~W*;/N*?I~}~
~{RT1`<-P^8D5D1jW<5D::WV5gDT4zBkND<~#(?IQ!4sB=9z1jBk~}(GB)~{;rL(MeNe4sBk~}(Big~
5)~{#)!#!!PBPGVPND#O#C#R8=SP::WV1`<-3LPrRT6T4K4zBkND<~=xPP1`<-P^8D!#!!8C::WV~}~
~{5gDT4zBkND<~R2?I1;>x4s6`J}5DVPND2YWwO5M3!"VPNDNDWV4&@mO5M3<0VPNDW@CfEE0fO5~}~
~{M3Ky=SJ\#,RT=xPP=xR;2=5D1`<-!"EE0f9$Ww#,Wn:s04SC;'R*Gs5D8wVVWVPN!"8qJ=4rS!~}~
~{Jd3v!#~}

~{!!!!PBPGVPND#O#C#R6T5gDT5DTKPP;7>3R*Gs:\5M#:~} ~{K|?ITZ>_SP~} 640K ~{DZ4f~}~
~{<0~} EGA~{!"~}VGA ~{M<OqJJEdFw5D~} IBM/AT ~{O5APN"5gDT<0Fd<fH];zIOTKPP!#!!~}~
~{K|?ITZ~} MS-DOS ~{;r~} DR-DOS ~{NwND2YWwO5M3IOV1=STKPP#,2;PhR*OH=xHkVPND2Y~}~
~{WwO5M3;7>3!#!!4KHm<~M(9}~} TIFF ~{M<Os1jW<8qJ=V'3V8wVVI(ChRG#(Hg~} HP~{!"~}M~
ICROTEK~{!"~}UMAX~{!"~}MITSUBISHI~{!"~}AGFA~{!"~}OASCAN ~{F=L(J=I(ChRG:M~} LOG~
ITECH~{!"~}GENIUS~{!"~}MASTEK~{!"~}PROLAB ~{JV3VJ=I(ChRG#)#,K|;9?IV1=S?XVF<[~}~
~{A.5D~} GENIUS 4500 ~{JV3VJ=I(ChRG!#~}

~{!!!!PBPGVPND#O#C#R>_SP!0TZO_!15DHK9$P#U}9&D\:MJ61p3LPr5D!0WTQ'O0!19&D\!#!!~}~
~{SC;'?IRT8y>]J61p9}3LVP2zIz5DPm6`SPSCPEO"#(HgT-J<M<OqPEO"#,A*OkWV</:MO`KFWV~}~
~{</PEO"#)7=1c5XP#U}R;P)J61p4mNs#,2"?IRT!0=L5<!1R;P)J61pHm<~IP2;HOJ65DWV7{!#~}~
~{!!AmMb#,PBPGVPND#O#C#R;9EdSPH}8v8(Vz3LPr#,7V1pSCSZ#:I>8D!"N,;$Q'O0?b#;F4:O~}~
~{SC=OU-5DJV3VJ=I(ChRG7VA=4NI(Ch!"J61p=O?mND<~Ky5C5=5DWs!"SRA=8v4zBkND<~#;RT~}~
~{<01`<-!"W*;;J61p=a9{4zBkND<~!#~}

~{!!!!PBPGVPND#O#C#R>_SP9c7:5DS&SCG0>0#,>Y@}HgOB#:~}

~{!!!!#(#1#)!!1(V=!"TSV>!"Ji?/3v0fIg!"S!Ji9]:M4rWVPP#:~}

~{!!!!1(V=!"TSV>3#U*B<1p5D3v0fNo5DNDUB#,Ji?/Tr3#PhIT<SP^8D:sTY0f!#RTMyPhSC:\~}~
~{6`HKA&!"J1<d=+Ub4sA?5DS!K"WJAOVXPBB<Hk5gDT!#!!2ISCPBPGVPND#O#C#R:s#,S!K"5D~}~
~{VPNDWJAO>M?ISI5gDTWT6/!"8_KY!"8_>+6H5XKMHk5gDT!#~}

~{!!!!#(#2#)!!Gi1(<lKw2?CE!"0l9+JRWT6/;/#:~}

~{!!!!RTG05DGi1(<lKw7=7(4s6<JG=+WJAORTM<Oq5DPNJ=4f4"!"<lKw!#!!UbVV<lKw7=J=Ph~}~
~{HK9$6TWJAO=xPPSPO^9X<|4J5DKwR}7V@`9$Ww#,6x2iQ/WJAOU_R2V;D\8y>]O^6(5DIYJ}9X~}~
~{<|4J=xPP<lKw#,Fd<lKwP'BJ5M#,GRM<OqND<~5D4f4"?U<d<+4s!#!!VASZ0l9+JR#,C?Ll6<~}~
~{SPPm6`ND<~PhR*4&@m!"1#4f!#!!RTMyR;0c6<SIHK9$=xPP7V@`4f55#,V=UEND<~U<SC:\4s~}~
~{5D0l9+?U<d#,2iURWJAOR2<+2;7=1c!#!!2ISCPBPGVPND#O#C#R:s#,>M?ID\=+ND<~RT5gWS~}~
~{WJAO5DPNJ=4fSZ5gDT!#!!HtTYEd:OR;8v4zBkND<~<lKw3LPr#,>M?IJ5OVHNRb9X<|WV<06`~}~
~{8v9X<|WV5D?lKY!"Ai;n!"8_P'5D5gWSWJAO<lKw!#!!Ub?IJ!H%HK9$KwR}7V@`5DJ1<d#,2"~}~
~{4s4s=ZJ!WJAO5D4f4"?U<d!#!!?IP^8D5D4zBkND<~5D4f4"?U<d;a1HO`S&5DM<OqND<~<uIY~}~
 1000 ~{16WsSR!#~}

~{!!!!#(#3#)!!=LJ&3v?<>m!"WwU_P4NDUB#:~}

~{!!!!=LJ&3v?<>m3#PhW[:O1pHK5DJTLb#,LX1pJGSoND=LJ&3vTD6ASk@m=bJTLbJ1Ph4sA?3-~}~
~{Hk1pHK5DNDUB!"JTLb!#!!6xWwU_;rR;0c8vHKP4NDUBJ1#,R23#PhU*B<P)1pHK5DNDUB6NBd~}~
~{#,4KJ12ISCPBPGVPND#O#C#RJGTY:C2;9}AK!#~}


~{!!!!MF9c:MFU<0PBPGVPND#O#C#R#:~}

~{!!!!N*AKHC8|6`5DSC;'AK=bPBPGVPND#O#C#R#,J9K|D\N*8|6`5DSC;'7~Nq#,8CHm<~5DQP~}~
~{VFU_M,RbTZ~}1993~{Dj~}8~{TBVA~}10~{TBFZ<dSC5gDTMxBgN*2?7VSC;'Cb7QLa9)Q]J>~}~
~{Hm<~!#!!Q]J>Hm<~1#AtAKU}0fHm<~5DVwR*9&D\5+N^7(1#4fJ61p=a9{!#!!SC;'?IM(9}W"~}~
~{2a5C5=W"2a?ZAnRT=+Q]J>Hm<~1dN*U}0fHm<~!#!!TZ~}1993~{Dj~}8~{TBVA~}10~{TBFZ~}~
~{<d#,>-<C0fW"2a7Q#:~}US$149~{#(U}3#<[#:~}US$399~{#)#,!!FUM(0fW"2a7Q#:~}US$296~
~{#(U}3#<[#:~}US$699~{#)!#!!>-<C0fV;D\J61pR;VVSC;'Q!Tq5DWVLe!#!!SPPKH$U_?I7"~}~
~{5gWSSJ<~#(~}email~{#)5=~} cc-ocr@brahms.iss.nus.sg



From guojin@iss.nus.sg Tue Aug 17 04:16:09 1993
Return-Path: <guojin@iss.nus.sg>
Received: from bochap.iss.nus.sg by ifcss.org (4.1/IFCSS-Mailer)
	id AA23864; Tue, 17 Aug 93 04:16:04 CDT
Received: from coyote.iss.nus.sg by bochap.iss.nus.sg (4.1/SMI-4.1)
	id AA14581; Tue, 17 Aug 93 17:18:21+080
Message-Id: <9308170918.AA14581@bochap.iss.nus.sg>
To: ftp-admin@ifcss.org
Cc: guojin@iss.nus.sg
Subject: StarOCR 
Date: Tue, 17 Aug 93 17:18:21 +0800
From: guojin@iss.nus.sg
Status: R

software:  StarOCR
version:   1.0 (demo)
function:  Printed Chinese character recognition system (Chinese OCR).
           Both for simplified and traditional Hanzi.  Support 8 fonts.
           GB and big5 selectable.  PC/MS-DOS environment.
ftp site:  ifcss.org [129.107.1.155]: /software/dos/input
filename:  cc-ocr.exe (self-extract), cc-ocr.doc
author:    Liang Xu Ming, (c/o. cc-ocr@brahms.iss.nus.sg)

GUO Jin
-------
Institute of Systems Science             Tel: (65) 772 2587
National University of Singapore         Fax: (65) 778 2571
Heng Mui Keng Terrace, Kent Ridge        Internet: guojin@iss.nus.sg
Singapore 0511                           Tel (H): (65) 779 0400

From asiandoc-request@magnus.acs.ohio-state.edu Wed Oct 13 13:58:44 1993
Return-Path: <asiandoc-request@magnus.acs.ohio-state.edu>
Received: from quark.magnus.acs.ohio-state.edu by ifcss.org (4.1/IFCSS-Mailer)
	id AA06617; Wed, 13 Oct 93 13:58:42 CDT
Errors-To: asiandoc-request@magnus.acs.ohio-state.edu
Received: by quark.magnus.acs.ohio-state.edu (5.65/3.910213)
	id AA22290; Wed, 13 Oct 93 14:56:24 -0400
Reply-To: asiandoc@magnus.acs.ohio-state.edu
Errors-To: asiandoc-request@magnus.acs.ohio-state.edu
Sender: asiandoc-request@magnus.acs.ohio-state.edu
Received: from netcom6.netcom.com by quark.magnus.acs.ohio-state.edu (5.65/3.910213)
	id AA22269; Wed, 13 Oct 93 14:56:18 -0400
Received: by netcom6.netcom.com (5.65/SMI-4.1/Netcom)
	id AA03801; Wed, 13 Oct 93 11:56:50 -0700
From: mcevilly@netcom.com (Carlos McEvilly)
Message-Id: <9310131856.AA03801@netcom6.netcom.com>
Subject: Re: Proofreading Chinese:  a pattern comparison software question
To: asiandoc@magnus.acs.ohio-state.edu
Date: Wed, 13 Oct 93 11:56:49 PDT
In-Reply-To: <9310131807.AA26351@crux2.cit.cornell.edu>; from "John McRae" at Oct 13, 93 2:07 pm
X-Mailer: ELM [version 2.3 PL11]
Status: R

This does not directly answer your question about pattern comparison
software, but may still be of interest.  In Taipei I saw a demonstration
of software that reads a Big5-coded text file on a computer, and scans
for incorrect phrases.  For example (if memory serves) the 'xu' in
'xuyao' (xu1yao4; need) is written with either one of two characters;
I have been told that one of these ways is incorrect, although both
phrases are in the dictionary.  Note that we are talking about
different characters 'xu', not different forms of the same character.

Writers often confuse characters in idioms they may have learned by
ear but have never seen on paper.  This can twist meaning in sometimes
ironic ways, and we are all familiar with examples in English such
as 'tow the line.'  I should be able to think up some examples in
Chinese quickly, but I'm drawing a blank and most of my books are
somewhere in the middle of the Pacific right now. 

This software keeps a list of incorrect phrases, and when they
are encountered, flags them so a human can decide whether a change
is necessary.  Of course this doesn't catch all errors, but it
helps greatly with a significant class of errors.

Unfortunately I don't have the name of the company in Taipei that
was selling this.  And of course it might be of dubious value when
applied to classical texts, unless it was highly customized for
the appropriate vocabulary--difficult since it is much easier to
note and anticipate the common errors in modern usage than in
data entry of classical text.

It would be interesting to hear more about your ideas for OCR 
pattern comparison as a proofreading tool.

- Carlos McEvilly
  mcevilly@netcom.com
. 

e

-- 

From asiandoc-request@magnus.acs.ohio-state.edu Fri Oct 15 09:25:26 1993
Return-Path: <asiandoc-request@magnus.acs.ohio-state.edu>
Received: from quark.magnus.acs.ohio-state.edu by ifcss.org (4.1/IFCSS-Mailer)
	id AA29006; Fri, 15 Oct 93 09:25:22 CDT
Errors-To: asiandoc-request@magnus.acs.ohio-state.edu
Received: by quark.magnus.acs.ohio-state.edu (5.65/3.910213)
	id AA24526; Fri, 15 Oct 93 10:23:23 -0400
Reply-To: asiandoc@magnus.acs.ohio-state.edu
Errors-To: asiandoc-request@magnus.acs.ohio-state.edu
Sender: asiandoc-request@magnus.acs.ohio-state.edu
Received: from phem3.acs.ohio-state.edu by quark.magnus.acs.ohio-state.edu (5.65/3.910213)
	id AA24516; Fri, 15 Oct 93 10:23:20 -0400
Received: from PUCC.PRINCETON.EDU (VMMAIL@PUCC) by phem3.acs.ohio-state.edu
 (PMDF V4.2-13 #3557) id <01H44ZF2KB4W8X3YUD@phem3.acs.ohio-state.edu>; Fri,
 15 Oct 1993 10:23:12 EDT
Received: by PUCC (Mailer R2.10 ptf008) id 1438; Fri, 15 Oct 93 10:22:25 EDT
Date: Fri, 15 Oct 1993 10:12:52 -0400 (EDT)
From: GEST LIBRARY <GEST%PUCC.BITNET@phem3.acs.ohio-state.edu>
Subject: Re: status
In-Reply-To: Message of Thu,
 14 Oct 1993 18:41:55 -0400 (EDT) from <mdonovan@magnus.acs.ohio-state.edu>
To: asiandoc@magnus.acs.ohio-state.edu
Message-Id: <01H44ZF2KB4Y8X3YUD@phem3.acs.ohio-state.edu>
Content-Transfer-Encoding: 7BIT
Status: R

About Chinese OCR:

A few months ago the Taiwanese version of PC Magazine
had a critical, up-to-date article on Chinese OCR. Basically, it cast doubts
on the claims of the available programs (three so far) of recognizing 95%
of the characters in at least one font. What was new to me is that those
claims are RESTRICTED to some 5,000 characters; there is NO program which can
do more than that. Program's claims that they support the Big-5 set had to
be read that yes, the additional 8000 could be entered by normal ways, but
not by OCR... And that while for practical purposes, even 95% is probably
too low to be useful: it means 5 characters wrong on every three lines of text!
There obviously is a long, long time to go before Chinese OCR is even remo-
tely useful.

As a sidenote: it was interesting to see that this is a field in which PRC-
Taiwan cooperation is quite advanced; in fact all or most Taiwanese research
depends on the PRC, which seems way ahead in OCR recognition.

Martin Heijdra/Gest/Princeton

From asiandoc-request@magnus.acs.ohio-state.edu Fri Oct 15 10:48:37 1993
Return-Path: <asiandoc-request@magnus.acs.ohio-state.edu>
Received: from quark.magnus.acs.ohio-state.edu by ifcss.org (4.1/IFCSS-Mailer)
	id AA00443; Fri, 15 Oct 93 10:48:33 CDT
Errors-To: asiandoc-request@magnus.acs.ohio-state.edu
Received: by quark.magnus.acs.ohio-state.edu (5.65/3.910213)
	id AA01876; Fri, 15 Oct 93 11:46:07 -0400
Reply-To: asiandoc@magnus.acs.ohio-state.edu
Errors-To: asiandoc-request@magnus.acs.ohio-state.edu
Sender: asiandoc-request@magnus.acs.ohio-state.edu
Received: from garnet.Berkeley.EDU by quark.magnus.acs.ohio-state.edu (5.65/3.910213)
	id AA01867; Fri, 15 Oct 93 11:46:04 -0400
Received: by garnet.berkeley.edu (5.64/1.33r)
	id AA06605; Fri, 15 Oct 93 08:46:02 -0700
Date: Fri, 15 Oct 93 08:46:02 -0700
From: buddhst@garnet.berkeley.edu (Lewis Lancaster)
Message-Id: <9310151546.AA06605@garnet.berkeley.edu>
To: asiandoc@magnus.acs.ohio-state.edu
Subject: Re: status
Status: R

From: Lewis Lancaster
	In the Buddhist text projects, we have tried a
number of OCR approaches.  Twice, I saw demonstrations
in the national computer laboratory in Taiwan.  At the
moment, typists can input 30-60 characters a minute in
China with an accuracy rate that far exceeds the OCR.
The speed of input and final proofing is far better
with typing in the characters.  Unless the OCR reads
in the 99% level, the corrections take much longer
than doing manual input.  
	With roman letters in the Pali canon input, we
found that double typing was the answer.  By matching
the two type examples on the screen, we eliminate most
of the errors, except for a few that are made in common.
When Academia Sinica in Taipei did the Chinese dynastic
history data base (40 million characters), they experimented
with double typing the characters.  In the case of Chinese
up to 40% of the errors were made in common by both
typists, since both were trained to expect a certain form
they both accepted that they were in fact seeing that
form.  With Roman letters or the Greek alphabet, we have
many examples of input by people who cannot read the
language, but in the case of Chinese we have no program
that allows input by someone who does not know how to
read the character.

From easug-error@guvax.acc.georgetown.edu Mon Mar  7 19:38:08 1994
Received: from axp1.acc.georgetown.edu by ifcss.org (4.1/IFCSS-Mailer)
	id AA28017; Mon, 7 Mar 94 19:37:56 CST
Resent-Date: Tue, 08 Mar 1994 10:29:20 +0900 (JST)
Resent-From: easug-error@guvax.acc.georgetown.edu
Errors-To: easug-error@guvax.acc.georgetown.edu
Return-Path: mike@allegro.uxp.bs2.mt.nec.co.jp
Received: from guvax.acc.georgetown.edu by guvax.acc.georgetown.edu (PMDF
 V4.2-10 #5850) id <01H9PC80RYB490N5PY@guvax.acc.georgetown.edu>; Mon,
 7 Mar 1994 20:28:02 EST
Received: from CUNYVM.CUNY.EDU (MAILER@CUNYVMV2) by guvax.acc.georgetown.edu
 (PMDF V4.2-11 #5850) id <01H9PC7O60U88YG6CJ@guvax.acc.georgetown.edu>; Mon,
 7 Mar 1994 20:27:32 EST
Received: from CUNYVM (NJE origin SMTP@CUNYVM) by CUNYVM.CUNY.EDU (LMail
 V1.1d/1.7f) with BSMTP id 0731; Mon, 7 Mar 1994 20:26:08 -0500
Received: from TYO.gate.nec.co.jp by CUNYVM.CUNY.EDU (IBM VM SMTP V2R2) with
 TCP; Mon, 07 Mar 94 20:25:52 EST
Received: from mailsv.nec.co.jp by TYO.gate.nec.co.jp (8.6.5/6.4J.6-TYO_gate)
 id KAA16492; Tue, 8 Mar 1994 10:26:00 +0900
Received: from minnie.d1.bs2.mt.nec.co.jp by mailsv.nec.co.jp (5.65c/6.4JAIN)
 id AA04047; Tue, 8 Mar 1994 10:25:57 +0900
Received: from necbs6.uxp.bs2.mt.nec.co.jp by minnie.d1.bs2.mt.nec.co.jp
 (5.65c/6.4JAIN-1.0) id AA25554; Tue, 8 Mar 1994 10:28:19 +0900
Received: from allegro.uxp.bs2.mt.nec.co.jp by necbs6.uxp.bs2.mt.nec.co.jp
 (5.67+1.6W/6.4J.6) id AA02218; Tue, 8 Mar 94 10:25:52 JST
Received: by allegro.uxp.bs2.mt.nec.co.jp (4.1/6.4J.6) id AA01336; Tue,
 8 Mar 94 10:29:20 JST
Date: Tue, 08 Mar 1994 10:29:20 +0900 (JST)
From: mike@allegro.uxp.bs2.mt.nec.co.jp (Mike Collinson)
Subject: Japanese OCR
In-Reply-To: ZAVELOFF@delphi.com's message of Fri,
 04 Mar 1994 19:15:11 -0500 (EST) <01H9L2PV1HCI8ZJVLQ@delphi.com>
To: ZAVELOFF@delphi.com
Cc: easug@guvax.acc.georgetown.edu
Errors-To: easug-error@guvax.acc.georgetown.edu
Resent-Message-Id: <01H9PC80W7TU90N5PY@guvax.acc.georgetown.edu>
Message-Id: <9403080129.AA01336@allegro.uxp.bs2.mt.nec.co.jp>
X-Vms-To: IN%"ZAVELOFF@delphi.com"
X-Vms-Cc: IN%"easug@guvax.acc.georgetown.edu"
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Content-Transfer-Encoding: 7BIT
Status: RO

Steven,
 
	Sorry to here of your data loss, I know the feeling.  If I
made a copy of reply and can find it I will send it on. If I recall the
gist was that products which do Japanese OCR including Kanji only came
on the market in the last year here so they may be rare on the ground
in the US.  The only place I know that might help is Kinokoniya a big
Japanese bookstore chain with shop(s) in the US which do Japanese
related software.  Here in Japan I have brochures for a
hardware/software system put out by Oki (of fax machine etc fame) and
a software package by a small company called Birds.  The latter runs
on NEC PC9801 PCs only (non-IBM compatible) but I dimly recall them
saying they would bring a DOS/V version out for IBM compatibles.
Email me if you want contact tel/fax/address for either of these.
 
--------------------------------------------------------------------------
Mike Collinson
Assistant Manager, Product Engineering Dept., UX Software Development Div.,
NEC Corporation, Daito Tamachi Building, 14-22 Shibaura 4-Chome,
Minato-ku, Tokyo 108, JAPAN
Email: mike@uxp.bs2.mt.nec.co.jp               Fax:+81-3-3456-6675
				               Tel:+81-3-3456-7451
--------------------------------------------------------------------------

From easug-error@guvax.acc.georgetown.edu Mon Mar  7 21:28:58 1994
Return-Path: <easug-error@guvax.acc.georgetown.edu>
Received: from axp1.acc.georgetown.edu by ifcss.org (4.1/IFCSS-Mailer)
	id AA01880; Mon, 7 Mar 94 21:28:52 CST
Resent-Date: Mon, 07 Mar 1994 22:07:15 -0500 (EST)
Resent-From: easug-error@guvax.acc.georgetown.edu
Errors-To: easug-error@guvax.acc.georgetown.edu
Received: from guvax.acc.georgetown.edu by guvax.acc.georgetown.edu (PMDF
 V4.2-10 #5850) id <01H9PFOPJP0090NLJC@guvax.acc.georgetown.edu>; Mon,
 7 Mar 1994 22:07:41 EST
Received: from guvax.acc.georgetown.edu by guvax.acc.georgetown.edu (PMDF
 V4.2-11 #5850) id <01H9PFD1H7LS8YFMUO@guvax.acc.georgetown.edu>; Mon,
 7 Mar 1994 22:07:15 EST
Date: Mon, 07 Mar 1994 22:07:15 -0500 (EST)
From: Patricia O'Neill-Brown <ONEILLP@guvax.acc.georgetown.edu>
Subject: RE: Japanese OCR
To: easug@guvax.acc.georgetown.edu
Errors-To: easug-error@guvax.acc.georgetown.edu
Resent-Message-Id: <01H9PFOPJVLE90NLJC@guvax.acc.georgetown.edu>
Message-Id: <01H9PFD1O6DE8YFMUO@guvax.acc.georgetown.edu>
X-Vms-To: IN%"easug"
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII
Content-Transfer-Encoding: 7BIT
Status: R

I found a reference to an OCR program in Vol. IV Issue 1 of 
Multilingual Computing. Description follows:

SOCR Symbolic OCR
For Unix, Sun

Languages: Chinese, Hebrew, Japanese, Russian

SOCR is an omnifont optical character recognition program which 
converts page images of Japanese text (including kanji, kana and 
romaji) to machine readable Shift-JIS files. SOCR runs on Sun 
workstations; ports to other UNIX workstations are available upon 
request. A MS Windows version is under development.

UNIX versions of SOCR are based on X Window. PCs and Macs can share a 
UNIX based OCR engine across a network via X Windows. TIFF image files 
are supported for input. Output files are in shift-JIS. SOCR produces 
output files which are compatible with machine translation programs.

SOCR uses English language menus. The user does not need to know 
Japaense to process Japanese text pages. Other language character sets 
(e.g., Cyrillic, Chinese, Hebrew, etc.) are available upon request.

SOCR is available in a variety of formats, from software only through 
complete turn-key systems. A variety of options are available, 
including scanners, machine translation programs, word processors, and 
text search databases.

Disk formats: 3 1/2" diskette, Tape
Documentation: Printed, On-Line

MRJ Inc.
10455 White Granite Drive
Oakton, VA 22124
703-385-0700
703-385-4637 Fax
No internet address provided


Patricia O'Neill-Brown

