[Systems] translate.sl.o

[Chris Leonard]

On Mon, Apr 18, 2011 at 11:48 PM, Bernie Innocenti <bernie at sugarlabs.org>wrote:

> On Mon, 2011-04-18 at 11:54 -0500, Rafael Ortiz wrote:
> >
> > Although this off-line workflow  would save us a lot of headaches,
> > imo is a work overhead  both for translators and activity devs, we
> > must  make the work needed to do activities on  sugar, the most
> > pleasant one.
>
> Perhaps it's because I'm a geek, but if I were translateing Sugar to
> Italian I'd appreciate using my favorite text editor, without any
> network delay and with the ability to test the new strings immediately
> in Sugar.
>
> I can see how humans may prefer a simple web interface over a text
> editor, but how do they test their work when they work in Pootle?
>
> Let me guess it: they don't test at all! :-)
>
> This would also explain why so many translations look awkward or out of
> context.
>
> To check if Pootle is really necessary, I'd like to see the list of the
> 10 most active translators and see how many of them wouldn't be able to
> work directly with the po file.
>


Please forgive me if my rhetoric seems impassioned and for the length of my
reply, but I do feel quite strongly about this issue.  I will try to
distinguish my logical and verifiable arguments from those that come from my
gut.

The top five submitters are listed at the bottom of the Pootle home page:

Submissions
cjl              25532
readmanr   6055 (mostly from en_GB, hardly counts as L10n)
aputsiaq     3402 (a heroic effort in lang-da)
subbu         2665 (lang-kn, mostly an eToys localizer)
rubina         2545 (another heroic effort in lang-hy at special request of
OLPC)

Key Pootle functions:

Being able to easily access statistics like those above is just one of many
functions that Pootle performs above and beyond holding POT files. These
stats are broken down by language and project and serve as my primary means
of identifying potential language admins.

Pootle is absolutely essential to my own work as a L10n community
organizer.  The translation memory function of the Terminology project is
important for consistency and productivity, the built-in pofilter review tab
is critical for enabling quality control processes.  Suggestions are only
possible in the context of a Pootle server, they are not contained within
the PO file format itself, which only has "fuzzy" flagging.  No such
advantages are available in an unhosted solution.

Disadvantages of unhosted localisation workflows:

There are a few projects that use a CVS/repo for POT files and a mailing
list for submissions.  Their localizations are typically in fewer languages
than web-hosted localization workflows (Pootle, Transifex, Damned Lies) and
also suffer from significant quality issues.

I've compiled a good number of links to upstream projects (separating hosted
from unhosted L10n workflows) on this page:

http://wiki.sugarlabs.org/go/Translation_Team/Upstream_localization

I've also looked over the results of these unhosted efforts quite closely.
Consider these statistics for Spanish (the most commonly localized second
language and one quite important to Sugar Labs / OLPC) on a number of
well-known packages that use a repo/mailing list L10n workflow.  On hosted
solutions lang-es is typically 99-100% complete.

Abiword is only about 90% complete in Spanish.

Inkscape is a little over 80% translated in Spanish.

Gnash is about 25% complete in Spanish and it's translation quality is
frankly pathetic.

Community Impact:

A far less radical shift (Fedora going from an internally hosted Transifex
instance to an externally hosted Transifex instance) caused L10n community
unrest in Feb/Mar of this year as it was done without adequate community
consultation and was seen as a breach of trust between the Infrastructure
Team and the Localization Team.  We cannot come to any conclusion without
adequate consultation with the L10n community itself.

It should also be considered that "dropping" Pootle could be seen as
breaking faith with the eToys community and in my opinion would be a
betrayal of our localization community.  No such decision can be taken on
this private systems list, it would require input from the localizers and
should be taken to the Oversight Board.  I would see a conversion to an
unhosted L10n workflow as a disastrous failure on the part of Sugar Labs to
support it's mission and it would almost certainly cause me to reconsider
whether the time and effort I have contributed over the past several years
to that mission was valued highly enough for it to be worth continuing my
involvement with the project.

There are other hosted L10n workflow tools that could be investigated, but
they will all come with their own unique problems.  I am not unsympathetic
to the maintenance costs of Pootle, but I do not think we have reasonably
exhausted the appropriate avenues to stabilize our instance, nor am I aware
of it being an inordinate maintenance burden compared to the
mission-critical function(s) it provides.  I have not seen any posts on the
upstream Pootle list documenting the issues (which I don't fully understand)
or asking for assistance.  They are fairly responsive to request for help.
I filed a ticket requesting an enhancement (not even a bug) in Virtaal and
it was rolled out in trunk in three weeks.

In essence, what you are suggesting in going to an unhosted L10n workflow is
the equivalent of saying, we will not maintain a wiki, but we will give you
an FTP site.  It is my opinion that we do not have the strength or depth in
our cadre of language admins to make this switch succeed in the long run,
nor do we have the resources t odo a thorough job of assessing and migrating
to a different hosted L10n workflow solution, even if one were identified
that accomplished as much as Pootle does.

Respectfully Yours,

cjl
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sugarlabs.org/private/systems/attachments/20110419/e7822a27/attachment.html>