Re: translation automation?

[chaddadun at bluewin dot ch]

>-- Message original --
>From: Arafat Medini <lumina at silverpen dot de>
>To: Documentation and Translation <doc at arabeyes dot org>
>Date: Sun, 17 Oct 2004 10:12:52 +0200
>Subject: Re: translation automation?
>Reply-To: Documentation and Translation <doc at arabeyes dot org>
>
>
>Am Samstag, den 16.10.2004, 22:59 +0200 schrieb chaddadun at bluewin dot ch:
>> Salam,
>> 
>> Experience proved that "machine translation" is still far from being a
>perfect
>> solution, as texts translated this way require heavy editing and usually
>> more time. While we might gain 10%, editing the remaining 90% will take
>> more time and effort.
>> 
>
>I don't agree really, I installed Kbabel (last week?) and I was blown
>away by the improvements done to it's database, the automatic
>translation is really VERY good, sure it is far from perfect, and I hope
>it will ever stay like that ;) But today editing an automatically
>translated file usually does not take more time then beginning from
>scratch.
>Especially if your prior work has a solid foundation and you're not
>building your database from pos that are not well and/or completely
>translated.

Salam,

I might not have been clear enough about what I was trying to say. Machine
Translation software are different than other translation tools. Machine
translation tools translate full documents and not only sentences or strings.
They have their own built-in dictionaries, as well as morphological, structural,
semantic and other rules to construct complete sentences (i.e. in Arabic)
based on what the software "understands" from its analysis of the foreign
language text submitted for translation (i.e. in English). Examples could
be Ajeeb, Al-Mutarjim Al-Arabi, etc. Machine Translation is still far from
being a perfect tool.

Translation memory tools, such as Trados do not translate. They just build
a "Translation Memory" composed of a set of segments (mostly sentences) and
their translation (i.e. English-Arabic), based on previously translated documents.
When you have a new document to translate (i.e. from English), each sentence
is compared to those stored in the TM and when a match (in Arabic) is found,
it is proposed as a translation. I think that Kbabel can be associated with
TM, maybe except that it is exclusively designed to translate strings in
.po files and not full documents such as HOWTOs or computer manuals (please
correct me if I am wrong). These are better tools.

Corpus-based translation tools, such as MultiTrans, not only build a kind
of "Translation Memory", but also keep track of, and display, the full text
(in both languages) from which translation of a certain sentence or smaller
segment is found. These are, to me, near-perfect tools for translating full
documents.

Salam

Yasser