Multilingual Document Images and Corpora
<BR>
            ----------------------------------------
<BR>

<BR>
                       A. Lawrence SPITZ
<BR>

<BR>
               Document Recognition Technologies, Inc.
<BR>
          616 Ramona St, Suite 20, Palo Alto, CA 94301 USA
<BR>
      spitz@docrec.com Tel. +1-650-688-0842 FAX +1-650-688-0841
<BR>

<BR>

<BR>
Images
<BR>
 
<BR>
  We have over 700 page images in the following languages:  Africaans,
<BR>
Chinese, Danish, Dutch, English, French, German, Icelandic, Italian,
<BR>
Japanese, Korean, Norwegian, Portuguese, Romanian, Spanish, Swedish.
<BR>
 
<BR>
 
<BR>
Truthed images
<BR>
 
<BR>
  We have about 600 "truthed" pages in the following languages:
<BR>
Africaans, Danish, Dutch, English, French, German, Icelandic, Italian,
<BR>
Norse, Norwegian, Portuguese, Romamian, Spanish and Swedish. However
<BR>
these should be considered only approximations to the truth since the
<BR>
data have been entered only once. We are seeking sponsorship for
<BR>
completion of the truthing process.
<BR>
 
<BR>
 
<BR>
Corpora
<BR>
 
<BR>
  We have corpora ranging in size from 600KB to 91 MB (averaging 16MB) in
<BR>
the following languages:  Africaans, Danish, Dutch, English, French,
<BR>
German, Icelandic, Italian, Norwegian, Portuguese, Romanian, Spanish.
<BR>
These corpora have been cleaned up in terms of removing special
<BR>
characters and any markup.
<BR>
 
<BR>
  We would be willing to consider cooperative efforts where DocRec and
<BR>
other entities combine data and analysis effort in return for shared
<BR>
rights to use and distribute the data.
<BR>

<BR>

<BR>
Software
<BR>
 
<BR>
  We have a suite of software that runs on UNIX or WindowsNT that is capable
<BR>
of deteriniming the script and language of the document. It currently
<BR>
handles 23 languages in Roman script plus Chinese, Japanese and Korean. We
<BR>
are seeking sponsorship for expanding this capability to a broader range
<BR>
of scripts and languages.
<BR>