[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [olpc-software] Scaling and localization in package management

Jim Gettys wrote on Tue, 14 Mar 2006 19:56:58 -0500

> I think serious consideration of scale due to languages is needed, and
> that it will have serious effects on how we do packaging and
> localization.

Very good points and there is also the scale issue of human labor. If we
keep doing things the way they are now we will reach a point where
significant percentages of some country's populations will have to be
full time translators (remember the old "in the future every woman in
the US will have to be a telephone operator" problem?). And given the
economics of this project, we will need people willing to work for free.

The alternative is automation. Full automation is not practical at this
time (see http://babelfish.altavista.com or equivalent) but
semi-automatic translation could be a good match for our needs. Imagine
an author who speaks only language L1 and a user who only knows L2. A
program on the authors machine does the L1->IL (intermediate language)
translation semi-automatically asking for help from the author whenever
there are several options. The author runs a second program to do LI->L1
and checks if the result is acceptable (it won't match the source L1
text exactly). If not, he can change the source text and try again or
make different choices during L1->LI (or do both). When the result is
acceptable the LI file is shipped to the user. There a program does
LI->L2 translation. The results for this can vary considerably, but will
probably be acceptable if LI->L1 was also acceptable.

So we have:
- author's machine: L1 dictionaries and data files, two translations
- user's machine: L2 dictionaries and data files, one translation
- package: LI files (no L1 nor L2 data)

We don't need a person who knows both L1 and L2 and the author's (or
author plus a group of helpers/translators) is the same no matter how
many languages the users need.

All that I described above already exists:


At least a few years ago when I looked into this these guys were not at
all open source friendly. Which made no sense to me given that their
funding came from the UN and that most of their data came from
collaborators around the world funded by their local governments.
Perhaps they have changed recently.

I see no other option. Check out the project site and wiki pages in
Portuguese (and other languages) for a perfect example of what I am
talking about. You can put in some extra hours in the beginning to get
things started but it is impossible to keep up manually.

-- Jecel

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]