Re: Web-based translation for GNOME

[danilo gnome org (Danilo Åegan)]

Hi Dwayne,

Thanks for sharing your thoughts!

Today at 14:49, Dwayne Bailey wrote:

> 1) Pootle has the potential to bring much new fresh blood into your
> communities, blood that is not all CVS, KBabel, vi inclined.

Indeed, that's why we already have http://prevod.org/ for Serbian
translations, though I've been lazy in updating it for recent Gnome
releases (I need to run a script to fetch recent
translation-status.xml and update my database). 

Basically, we have tracking (who did what), reservation ("I'm working
on this, don't touch"), glossary and restrictions (you can't update PO
file which someone is working on) and (static web pages) rules.
E-mail announcements about updated translations go to the mailing list.

The big feature we're missing is per-message translation (we have
per-domain translation, but without web editor).

The biggest problem though is that my PHP code sucks there :)

> 2) Pootle has the potential to improve the quality of translations
> generally, through structured review cycles.

Indeed, that's one of the best features I could see in Pootle, or any
such system.  That's why I said web-based translation system would be
nice to have for Gnome.

> 3) Pootle has the potential to make coordinating a language much easier
> as you know whose assigned what could set goals, etc

I agree.  But I probably have too much experience with my system, that
I'm unwilling to sacrifice advantages it brings for advantages Pootle
would bring, so basically,... 

> I'm biased :)

...I'm biased too :)

My problem with Pootle is only that: I have my own biases, and it's
hard for me to overcome them, especially since I have my share of code
which overlaps with what Pootle tries to do. :)

>> While I think having web-based translation tool would be excellent for
>> official GTP, I find Pootle quite lacking in features.
>
> I met a guy last week who refuses to use a cellphone: "I used a car
> phone in the 80s they were big and impractical, been there done that, so
> I'm not bothered with getting a cellphone"

>From my sentence "I don't want us to make cellphones an official
(meaning: promoted and recommended) way to talk to your mom", you
deduce that I'm saying "car phones were crap back then, so I think
cellphones are crap as well, and you won't be able to talk to your mom
with them".  Does this really compare? :)

> At Translate.org.za we're actively using it. Yip there are many things
> that could be added and they are appearing every day.  I've run it with
> 40 people translating Firefox into Zulu.  We completed Xhosa in 2 days.

So, I should be more constructive, I agree :)

> I guess what I'm saying is don't judge a moving target on your
> experience from the first releases.

Well, one can judge something just by what it is, not by what it may
become :)

> What can work now - or with a few changes - is to download the .zip file
> from Pootle and extract that over your checked out Gnome.

That's going to be very problematic (think: CVS conflicts).  I doubt
most translators have broadband and ability to "cvs up" their entire
Gnome checkouts prior to extraction step (they'd want only to cvs up
what needs cvs up-ing).

> Email sounds good but do you want 100 emails?  We need some better ideas
> on what change notification needs to come from Pootle.

That's a problem with per-message translation systems.  Basically,
what we want here is to have maintainers give forward announcements:
"in three days we'll be releasing, so translators, please finalize your
translations."

Though, that's far from perfect as well.  "Notification granularity"
is what translation coordinators would probably need (e.g. each
message edit could be marked "critical", "normal" and "enhancement",
with the default being "normal"; 1 "critical", 10 "normal" and 100
"enhancement" updates might trigger an e-mail notification).

> How would be best to represent what changed? diffs on PO files are often
> hideous and unhelpful.  

(Shameless plug: I'm actually working on sort of a podiff which will
ignore differences in source references, translator's comments, etc.
one slightly more complicated issue is actually emitting a minimal
"diff -u" compatible patch, at least it's more complicated with my
current PO file reader which ignores meaningless PO file data, such as 
differences between

  msgid ""
   "blah"

and

  msgid "blah"

)

> How would you reject translations?  I'm not sure
> if it would work to create a complex email parser.

E-mails referring to web pages are just fine, IMO.  I'm too lazy to go
to the same web pages and discover there have been no changes, while I
regularly check my e-mail (basically, fetchmail does it automatically
for me :).

> With the Translate Toolkit we can already merge changes into an existing
> PO file.  Maybe one option is to email those changes and the coordinator
> can edit those changes and merge them offline?

That seems too complicated.  I think we need e-mail just for
notification purposes.  Translation coordinator could then log in and
enter CVS account info (which will not be stored, except *maybe* for
username) to commit.  HTTPS would be quite beneficial for that (btw,
is there any free software community based CA, which is also
recognized by free software?).


Things I have in my TODO list, which I suppose would be very useful
for Pootle as well:

 - msgmerge reimplementation in Python (with my PO reader class), and
   ability to use any fuzzy matching algorithm
 - word diff for msgmerge-d entries (I need this for xml2po as well)
 - glossary and translator integration
 - web translation editor should colour untranslated or fuzzy entries
   background differently (when you _want_ to know context, so you're
   showing all messages, these messages should be easily
   distinguished), support search
 - using base language other than what's in msgid's (eg. if there's a
   complete Croatian translation, and a Serbian translator is not very
   well versed in English, she might want msgid's to be in Croatian;
   this is easy to implement, though, but should be a big gain)
 - support full names in UTF-8 (heeey!) :)

For reference, I've put my (yet unfinished) PO file reader class at

  http://kvota.net/hacks/zmijsko-gnezdo/gettext_po.py  

(it seems to be structured similar to Pootle's storage/po.py)

Also note that I really love some features of Pootle web editor!


Cheers,
Danilo