Improving the translation process

Improving the translation process

[Julien Lepiller]

Hi guix!

Our current translation process involves sending a source tarball to the TP. Updating a translation is a manual process that can't be done too often, and I don't think the TP is very user friendly. I'd like to set up another service, such as weblate.

The Fedora people have recently switched to it and are inviting other free software projects, so it could be a nice way to setup this new process, before we can make our own service and host it. I've talked with someone responsible for weblate at Fedora, and they would be pleased to help us. We could also apply for the main instance at hosted.weblate.org.

As I see it, we don't want any of these instances to have push rights on our repos. The idea is to have a separately hosted repo for translations, used only by weblate. We would host a bot to regularly  push up-to-date pot files to that intermediate repo, which will update weblate's data. Weblate would push to that repo and any guix commiter would be able to download, check and commit new po files. No commiter would need to have write access to that repo: only the bot and weblate would use it. It could be on savannah or fedora's inhrastructure or somewhere else, but not in the guix namespace at savannah, because access rights are per-namespace iiuc.

This new process would allow for daily or weekly updates of translations, which fits better with our rolling-release model and will help translators of the manual do work on a regular basis instead of per-release. The interface is way more understandable for newcomers who will quickly understand how to contribute. The downside is, as always, more fragmentation in the l10n community esp. if we go for our own instance.

WDYT?

Re: Improving the translation process

[Julien Lepiller]

Le 6 février 2020 12:01:51 GMT-05:00, Julien Lepiller <julien@lepiller.eu> a écrit :
>Hi guix!
>
>Our current translation process involves sending a source tarball to
>the TP. Updating a translation is a manual process that can't be done
>too often, and I don't think the TP is very user friendly. I'd like to
>set up another service, such as weblate.
>
>The Fedora people have recently switched to it and are inviting other
>free software projects, so it could be a nice way to setup this new
>process, before we can make our own service and host it. I've talked
>with someone responsible for weblate at Fedora, and they would be
>pleased to help us. We could also apply for the main instance at
>hosted.weblate.org.
>
>As I see it, we don't want any of these instances to have push rights
>on our repos. The idea is to have a separately hosted repo for
>translations, used only by weblate. We would host a bot to regularly 
>push up-to-date pot files to that intermediate repo, which will update
>weblate's data. Weblate would push to that repo and any guix commiter
>would be able to download, check and commit new po files. No commiter
>would need to have write access to that repo: only the bot and weblate
>would use it. It could be on savannah or fedora's inhrastructure or

Actually, not on fedora's infrastructure :) but hosting a repo on a server or on a separate namespace at savannah shouldn't be too complicated.

>somewhere else, but not in the guix namespace at savannah, because
>access rights are per-namespace iiuc.
>
>This new process would allow for daily or weekly updates of
>translations, which fits better with our rolling-release model and will
>help translators of the manual do work on a regular basis instead of
>per-release. The interface is way more understandable for newcomers who
>will quickly understand how to contribute. The downside is, as always,
>more fragmentation in the l10n community esp. if we go for our own
>instance.
>
>WDYT?

Re: Improving the translation process

[pelzflorian (Florian Pelz)]

I think the Translation Project mailing lists are a good place to
discuss translations and they should be recommended.  I am thankful to
the German TP team’s Mario Blättermann for introducing me to the TP’s
way of working and giving help.  Also it is good to look at the PO
files collection at the TP to see how some word was translated in
other software.

However, this does not mean Guix needs to use the TP services for
translation.  To me, the need to upload a tarball to the TP seems
pointless.  Automated, regular upload of POT files would be very
welcome.

If Guix or Savannah decides on Weblate, the reason should be easier
maintenance for the maintainers, which I cannot comment on.  Opinions
on that would be interesting.

I had never heard of Weblate, so I watched:

https://www.youtube.com/watch?v=IVlXt6QdgdA

Most things I don’t care about.  The speaker talks about non-technical
people not downloading PO files.  I believe Guix translators need
technical knowledge anyway to understand what they translate.  I do
not believe sending emails is a problem for translators.  In my
opinion, downloading PO files is the best way to search, diff and
proofread.  I do not like that Javascript seems to be required to
download and upload PO files via Weblate.  Adding automated checking
for valid Texinfo would in a few cases be helpful, but is really
needed only if Weblate were to commit automatically.

I don’t think full machine translation is useful, in contrast to
machine-aided dictionaries like:

https://www.linguee.com/english-german/translation/proofread.html


Thank you for all your efforts in improving translations!

Regards,
Florian

Re: Improving the translation process

[Julien Lepiller]

Le 6 février 2020 16:41:04 GMT-05:00, "pelzflorian (Florian Pelz)" <pelzflorian@pelzflorian.de> a écrit :
>I think the Translation Project mailing lists are a good place to
>discuss translations and they should be recommended.  I am thankful to
>the German TP team’s Mario Blättermann for introducing me to the TP’s
>way of working and giving help.  Also it is good to look at the PO
>files collection at the TP to see how some word was translated in
>other software.
>
>However, this does not mean Guix needs to use the TP services for
>translation.  To me, the need to upload a tarball to the TP seems
>pointless.  Automated, regular upload of POT files would be very
>welcome.
>
>If Guix or Savannah decides on Weblate, the reason should be easier
>maintenance for the maintainers, which I cannot comment on.  Opinions
>on that would be interesting.
>
>I had never heard of Weblate, so I watched:
>
>https://www.youtube.com/watch?v=IVlXt6QdgdA
>
>Most things I don’t care about.  The speaker talks about non-technical
>people not downloading PO files.  I believe Guix translators need
>technical knowledge anyway to understand what they translate.  I do
>not believe sending emails is a problem for translators.  In my
>opinion, downloading PO files is the best way to search, diff and
>proofread.  I do not like that Javascript seems to be required to
>download and upload PO files via Weblate.  Adding automated checking
>for valid Texinfo would in a few cases be helpful, but is really
>needed only if Weblate were to commit automatically.
>
>I don’t think full machine translation is useful, in contrast to
>machine-aided dictionaries like:
>
>https://www.linguee.com/english-german/translation/proofread.html
>
>
>Thank you for all your efforts in improving translations!
>
>Regards,
>Florian

For maintainers: once in place, we won't need to manually generate tarballs and will be able to have more frequent updates.

Even if you are technically inclined, there is a cost to pay in learning the process to register at the TP, dodge the "do I need to sign a waiver?" question, get the file, know what to do with it, carefully read the page about the bot to know how to send it. Compare to weblate, where you need to register in a more familiar way, can edit online or offline (with a download and an upload button).

And I disagree you need to be technically inclined to translate guix: you don't need to when translating package descriptions (if you translate music software, you'll need to know what a fourrier transform is, and knowing computers won't help you there). You could also simply want to fix a typo (or a bunch of them ^^").

Maybe we don't need all these features, but it would also give translators a way to communicate and coordinate (more than one person on a project and language pair). Conversations are archived and linked to a specific string so others can find it when they want to translate.

Re: Improving the translation process

[pelzflorian (Florian Pelz)]

On Fri, Feb 07, 2020 at 08:00:31AM -0500, Julien Lepiller wrote:
> For maintainers: once in place, we won't need to manually generate tarballs and will be able to have more frequent updates.

Please go forward with Weblate then unless someone opposes.  Should we
ask the TP team what they think about hosting such a setup?  I’m
afraid I won’t help with the setup, but I would be thankful for more
current translations and am waiting to upload an almost complete (but
not almost finished) German translation.  Proofreading and fixing
errors in the English manual still needs to be done.


> Even if you are technically inclined, there is a cost to pay in learning the process to register at the TP, dodge the "do I need to sign a waiver?" question, get the file, know what to do with it, carefully read the page about the bot to know how to send it. Compare to weblate, where you need to register in a more familiar way, can edit online or offline (with a download and an upload button).
>

Well, the cost is mostly sending an email to the TP team to be told
what to do.  Although yes, Web uploads are more sensible than email.
Javascript isn’t sensible, but it’s not important.


> And I disagree you need to be technically inclined to translate guix: you don't need to when translating package descriptions (if you translate music software, you'll need to know what a fourrier transform is, and knowing computers won't help you there). You could also simply want to fix a typo (or a bunch of them ^^").

Yes, you are right.  I suppose people will find the
download button if they want it.



> 
> Maybe we don't need all these features, but it would also give translators a way to communicate and coordinate (more than one person on a project and language pair). Conversations are archived and linked to a specific string so others can find it when they want to translate.

I agree, this is an advantage.



> Le 6 février 2020 16:41:04 GMT-05:00, "pelzflorian (Florian Pelz)" <pelzflorian@pelzflorian.de> a écrit :
> >I think the Translation Project mailing lists are a good place to
> >discuss translations and they should be recommended.  I am thankful to
> >the German TP team’s Mario Blättermann for introducing me to the TP’s
> >way of working and giving help.  Also it is good to look at the PO
> >files collection at the TP to see how some word was translated in
> >other software.
> >

I believe the TP mailing list could still be mentioned in the manual
or on the Weblate server as a place to discuss with other teams
outside Guix.  Unless the TP team decides to adopt Weblate or similar.

Regards,
Florian

Re: Improving the translation process

[Ludovic Courtès]

Hello!

Julien Lepiller <julien@lepiller.eu> skribis:

> This new process would allow for daily or weekly updates of
> translations, which fits better with our rolling-release model and
> will help translators of the manual do work on a regular basis instead
> of per-release. The interface is way more understandable for newcomers
> who will quickly understand how to contribute. The downside is, as
> always, more fragmentation in the l10n community esp. if we go for our
> own instance.

The first two arguments are very compelling to me.

The counter-argument, I don’t know; in practice, it seems that Guix
translators are mostly involved with Guix, no?

I don’t know Weblate but if it works for you translators and doesn’t
require us maintainers (well, me!) to open a JS-enabled browser and
click through, I’m all for it.  :-)

BTW, in the meantime, would you feel like uploading a tarball to the TP
so we can give translators some time to start updating everything?

Thanks,
Ludo’.

Re: Improving the translation process

[zimoun]

Hi Julien,

This proposal to use hosted.weblate.org seems nice. But I think we
should be able to use the both. I mean, I feel more comfortable with
Emacs to edit some text than some webapp. To be concrete, often my
"window" is split into several pieces and I compare one part with
another, etc. And I do not know how to do that with my browser (since
the Next browser is not ready yet).

However, this will totally help people with less technical skills.
I think the translation/proofread is a really good door to come in a
project. And the TP machinery is not obvious the first time you deal
with it; even the second time. ;-)


All the best,
simon

Re: Improving the translation process

[Julien Lepiller]

Le 9 février 2020 18:55:11 GMT-05:00, zimoun <zimon.toutoune@gmail.com> a écrit :
>Hi Julien,
>
>This proposal to use hosted.weblate.org seems nice. But I think we
>should be able to use the both. I mean, I feel more comfortable with
>Emacs to edit some text than some webapp. To be concrete, often my
>"window" is split into several pieces and I compare one part with
>another, etc. And I do not know how to do that with my browser (since
>the Next browser is not ready yet).
>
>However, this will totally help people with less technical skills.
>I think the translation/proofread is a really good door to come in a
>project. And the TP machinery is not obvious the first time you deal
>with it; even the second time. ;-)
>
>
>All the best,
>simon

So if we agree to follow that plan, we need a new repository and a script running somewhere. I can take care of adding the script, but I don't know how to create a repo that could be accessible to weblate (it needs commit access, so it can't be the same namespace as guix on savannah). Ludo?

My proposal was to use fedora's weblate, since I already know Jean-Baptiste who is resppnsible for it. Also, it should be faster to create a project with them. But I don't really care, maybe hosted has a less confusing domain name ^^'

Re: Improving the translation process

[Ludovic Courtès]

Hi, Julien!

Julien Lepiller <julien@lepiller.eu> skribis:

> So if we agree to follow that plan, we need a new repository and a
> script running somewhere. I can take care of adding the script, but I
> don't know how to create a repo that could be accessible to weblate
> (it needs commit access, so it can't be the same namespace as guix on
> savannah). Ludo?

Sure!  Could you file a “support request” on Savannah to ask for a new
repo in the guix name space?  You can put guix-devel@gnu.org in copy so
we receive notifications.

> My proposal was to use fedora's weblate, since I already know
> Jean-Baptiste who is resppnsible for it. Also, it should be faster to
> create a project with them. But I don't really care, maybe hosted has
> a less confusing domain name ^^'

If it works for you, go for it!

Earlier I also wrote:

> BTW, in the meantime, would you feel like uploading a tarball to the TP
> so we can give translators some time to start updating everything?

WDYT?

Thanks,
Ludo’.

Re: Improving the translation process

[Julien Lepiller]

Le 11 février 2020 09:23:20 GMT-05:00, "Ludovic Courtès" <ludo@gnu.org> a écrit :
>Hi, Julien!
>
>Julien Lepiller <julien@lepiller.eu> skribis:
>
>> So if we agree to follow that plan, we need a new repository and a
>> script running somewhere. I can take care of adding the script, but I
>> don't know how to create a repo that could be accessible to weblate
>> (it needs commit access, so it can't be the same namespace as guix on
>> savannah). Ludo?
>
>Sure!  Could you file a “support request” on Savannah to ask for a new
>repo in the guix name space?  You can put guix-devel@gnu.org in copy so
>we receive notifications.

Yeah, but as I understand it, it means weblate would have access to every other repo in the guix namespace, which is probably not what we want. Do we want to give a private ssh key to weblate or fedora people that could allow them to push arbitrary commits to guix? :p

>
>> My proposal was to use fedora's weblate, since I already know
>> Jean-Baptiste who is resppnsible for it. Also, it should be faster to
>> create a project with them. But I don't really care, maybe hosted has
>> a less confusing domain name ^^'
>
>If it works for you, go for it!
>
>Earlier I also wrote:
>
>> BTW, in the meantime, would you feel like uploading a tarball to the
>TP
>> so we can give translators some time to start updating everything?
>
>WDYT?

I'll do that tonight!

>
>Thanks,
>Ludo’.

Re: Improving the translation process

[Ludovic Courtès]

Hi!

Julien Lepiller <julien@lepiller.eu> skribis:

> Le 11 février 2020 09:23:20 GMT-05:00, "Ludovic Courtès" <ludo@gnu.org> a écrit :
>>Hi, Julien!
>>
>>Julien Lepiller <julien@lepiller.eu> skribis:
>>
>>> So if we agree to follow that plan, we need a new repository and a
>>> script running somewhere. I can take care of adding the script, but I
>>> don't know how to create a repo that could be accessible to weblate
>>> (it needs commit access, so it can't be the same namespace as guix on
>>> savannah). Ludo?
>>
>>Sure!  Could you file a “support request” on Savannah to ask for a new
>>repo in the guix name space?  You can put guix-devel@gnu.org in copy so
>>we receive notifications.
>
> Yeah, but as I understand it, it means weblate would have access to every other repo in the guix namespace, which is probably not what we want. Do we want to give a private ssh key to weblate or fedora people that could allow them to push arbitrary commits to guix? :p

Ah yes, good point.  :-)  Let’s have a separate repo, then.

>>> BTW, in the meantime, would you feel like uploading a tarball to the
>>TP
>>> so we can give translators some time to start updating everything?
>>
>>WDYT?
>
> I'll do that tonight!

Thanks for doing that!

Ludo’.