[A12n-forum] Re: Fwd: open-source machine translation toolbox

["Don Osborn" <dzo@xxxxxxxxxxxx>]

Last year I forwarded this item and I want to return to it quickly as it 
came up again on another list (AfrophoneWikis). The project Mikel mentions, 
OpenTrad, actually works with 2 open source engines: Apertium, mentioned in 
the original mail, which is useful for similar languages; and Maxtin, for 
languages that are not closely related. Links to all of these are on the 
updated MT in Africa page, at http://www.bisharat.net/Trans/#res .
As we know, there are many groups of languages that are closely related on 
the one hand, but wide variations in the structure etc. of languages across 
the continent on the other. Something like Apertium might be useful within 
groups of very similar languages.
It occurs that the situation in South Africa, and southern Africa more 
broadly, where there are widely spoken languages that fall into at least a 
couple of major groupings (thinking here of Zulu/Xhosa/Ndebele and 
Sotho/Tswana), might be ideal for something like Apertium. The idea being 
that where production of materials for, say schools, public awareness 
campaigns, extension, etc, need to be in each official language, such a 
translation software might reduce costs and speed things up. I don't know 
how the results of the OpenTrad project have been used in Spain, but perhaps 
the object was similar?
In any event, I would hasten to add that I don't see such MT software 
putting human translators out of work. Since MT is not perfect, and because 
its use would hopefully increase the amount of material getting translated, 
there would still be lots of work, though with an emphasis on perfecting 
translations rather than doing them from scratch.
Eventually, such translation efforts will also probably make use of 
translation memory programs, which in effect use records of previous 
translations in new tasks. (Such is already common in Europe.)
(On the other hand, some research suggests that the level of computer skills 
and cost of translation memory might be obstacles. See: 
http://www.localisation.ie/reader/localisation_in_south_africa.php . One 
might hope that an open source MT would be cost effective even in the face 
of need for training and assuring connectivity for translators involved in 
cleaning up MT products.)
It may be that there are folks in South Africa already well ahead on all of 
this, and if so it would be great to hear more about it and what they are 
learning. Their results could inform similar efforts to come elsewhere in 
the continent.
Don

Don Osborn
Bisharat.net
PanAfrican Localisation project


----- Original Message ----- 
From: "Donald Z. Osborn" <dzo@xxxxxxxxxxxx>
To: <a12n-forum@xxxxxxxxxxxx>; <loc-dev@xxxxxxxxxxxxxxxxxxxxxxxxx>
Sent: Monday, September 19, 2005 2:23 AM
Subject: [A12n-forum] Fwd: open-source machine translation toolbox


This item may be of interest, although for the moment it is apparently 
useful
only for Romance languages and does not yet support Unicode.

Don Osborn
Bisharat.net



Dear Bisharat representative,
this letter is just to let you know that very recently my group has released 
an
open-source machine translation toolbox called Apertium through
http://apertium.sourceforge.net. Documentation is still scarce, but the main
use of this toolbox would be to produce machine translation systems for
languages that are not syntactically very different (we offer data for
Spanish-Catalan). You might be interested in posting a link to Apertium in 
your
http://www.bisharat.net/Trans/ , just in case some of the people working in
language technologies for African languages find it interesting. The kit 
still
uses only the Latin-1 or Latin 9 encodings, but we are planning to extend it 
so
that it can work with Unicode.

Sincerely,

Mikel Forcada

--
Mikel L. Forcada                    E-mail: mlf@xxxxxxxxxx
Departament de Llenguatges          Phone: +34-96-590-9776
i Sistemes Informàtics                also +34-96-590-3772.
UNIVERSITAT D'ALACANT               Fax:   +34-96-590-9326, -3464
E-03071 ALACANT, Spain.
.................................
URL: http://www.dlsi.ua.es/~mlf/
---------------------------------------------------------------
[!] Please avoid sending me Word, Excel or PowerPoint attachments.
Send them to me as plain text, HTML, CSV, Postscript or PDF instead.
...............................................................
[!] Per favor, no m'envieu fitxers Word, Excel o PowerPoint adjunts.
Envieu-me'ls com text pelat, HTML, CSV, Postscript o PDF.
..............................................................
[Info? http://www.fsf.org/philosophy/no-word-attachments.html]



----- End forwarded message -----


_______________________________________________
A12n-forum mailing list
A12n-forum@xxxxxxxxxxxx
http://lists.kabissa.org/mailman/listinfo/a12n-forum