A conversation about website translation

[Ricky Zhou]

Hey,

Bart and I had an interesting conversation about website translations
last night (below), and I think it highlights some of our main
challenges with our current translator workflow.  I probably won't be
able to make the meeting tomorrow, but it'd be great to start up a
conversation about this (and i18n/l10n people as well).

Thanks,
Ricky

=====

02:17 < ricky> Argh, this is hard :-(
02:17 < ricky> I need to think a bit more before jumping into anything.
02:22 < couf> ricky: what's your plan with xml2po?
02:22  * couf just noticed a commit
02:24 < ricky> I've been experimenting with it, but there are downsides
that I didn't see earlier.
02:24 < couf> have you looked at how the docs team uses it?
02:24 < ricky> For example, since it'd be extracting text directly from
a generated site, it'd shut out any possibility of non-hacky "dynamic"
content.
02:24 < ricky> (As in, content that updates on every website build)
02:25 < ricky> The other downside is that it'd require serious po file
manipulation to migrate over.
02:25 < ricky> couf: Yeah, stickster_afk's the person that mentioned it
to me.
02:26 < couf> yeah, that's a serious downside
02:26 < ricky> Their needs are slightly different from ours, though.
02:26 < ricky> :-(
02:26 < ricky> Here's the easiest solution for me.
02:26 < ricky> Take the git repo, make a directory for each language,
completely break the workflow for translators and have them maintain
parallel versions of the website for each language.
02:27 < ricky> That's really hard on translators, though.
02:27 < ricky> (And there are issues such as - how does a translator
know when they need to update their translation?)
02:28 < ricky> I don't know...  I'm not sure if PO files work well for
fast-changing things like a website.
02:28 < ricky> I've been getting pain at every corner.
02:28 < ricky> Making the templating system substitute the right text...
sucks.
02:28 < ricky> String extraction...  sucks.
02:29 < ricky> Making sure that all translations are always in sync also
sucks.
02:29 < couf> right :-/
02:30 < ricky> So some general questions - how hard should I be trying
to maintain the standard translator workflow?
02:30 < ricky> Which of the two problems (PO files vs. maintaining
parallel versions) is more easily solvable?
02:31 < ricky> And I guess, what uses are we looking at with the website
(and how will our choice of handling translations affect them?)
02:32 < couf> well from the translators viewpoint having parallel
versions would erase all efforts L10N has been trying to do to create a
standard workflow for everything
02:32 < ricky> Yeah, and I really don't want to do that
02:33 < couf> OTOH I do get that this will lead to more bad then good in
the end
02:33 < couf> as PO-based work isn't really fitted for this stuff
02:34 < ricky> I'm not even 100% aware of a good alternative to PO-based
stuff.
02:34 < couf> neither am I
02:34 < ricky> With parallel versions, it's still a big challenge to
keep in sync with the original version.
02:36 < couf> hmm, we might want to cut down on translators a bit, have
the languages sync up at release time and let the original change
everytime we feel like it needs to
02:36 < couf> when it comes to users I'd say most users come to the site
to download Fedora, and for the rest they would divert to the
fedora-specific site in their locale like projectofedora.br and such
02:39 < ricky> Hm.
02:39 < ricky> Should that be our websites vision for the future?
02:40 < couf> I don't know, but we should think about it somehow
02:40 < ricky> And we have the exact same problem on other websites
(even if they aren't supposed to be updated as often)
02:40 < ricky> Yeah.
02:42  * ricky wonders what the rest of the l10n team has to say about
this.
02:43 < couf> they probably won't like it :)
02:43 < ricky> Well, not the parallel versions thing (which I'm not
considering quite yet)
02:43 < ricky> Just how we can handle translations for something like a
website.
02:44 < ricky> I wonder if anybody has experience doing something like
this in a job or anything.
02:44 < ricky> I guess I should also mention that wiki translations will
kind of break the workflow in the same way.
02:46 < couf> yeah, some have been talking of a PO-extraction layer on
top of MW-API to make it sort of work with the workflow, but I'm not
sure if that'll work in the end
02:48 < ricky> Yeah.  This would be something that updates even faster
than a website.
02:49 < couf> right, so in the end it looks like PO doesn't scale to
websites :)
02:50  * couf notes: personal view, not L10N's view :)
02:53 < ivazquez> Perhaps this is a conversation we should be having in
#fedora-i18n .
02:58 < ricky> Good point
02:58 < ricky> I'm actually going to get some sleep now
02:58 < ricky> couf: I think we touched over some important points in
this conversation - mind if I send it to websites-list?
02:59 < ricky> (I want to get it in the discussion in tomorrow's
meeting, even if I won't be there)
02:59 < couf> sure go ahead
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: not available
URL: <http://lists.fedoraproject.org/pipermail/websites/attachments/20080623/45ca5965/attachment.sig>