Translation Freeze – 2026-02-26

[github-actions[bot]]

Translation Freeze for Upcoming Release

The branch has been created or updated. Translatable strings are now frozen for this release.

0 string(s) added, 350 string(s) removed since the previous template.

Removed strings:

---

Translators: Please update your files against the branch using .
This notification is sent once per release, when strings are considered stable.

Commit: 3218de3

[trypsynth]

wow, so this template is still broken. Translators, I'd like to have new Paperback at some point this weekend, so please have your strings ot me in the next 48-72 hours if you want them included. @schulle4u @nidza07 @Ruslan-Gulmagomedov @qt06

[nidza07]

Also CC @tareh7z for Bosnian.

@trypsynth Don't worry about the template, Poedit already handles displaying the newly added/removed strings, or concretely it automatically removes what needs removing and has an option to sort the file so untranslated strings appear on top. You're already handling the important part extremely well, which is notifying us when a new release is approaching.

[trypsynth]

Thanks @nidza07, knew I was missing someone. Also thanks to everyone who's submitted translations so far!
I'm considering switching to an automated translation system in the future, where the translations for many languages are automatically generated from the English strings by DeepL, and then I'll maybe even leave the freeze in place if any translators who know a given language find errors in it. Thoughts? Should I just leave it how it is?

[tareh7z]

Hi @trypsynth!

I wouldn't like to see automated translations in Paperback for the simple reason that machine translations very often miss context, making the translation completely unclear and useless. I believe we, the translators, are still irreplaceable by AI. It could also affect others who might give up on translating if they saw that translations are automatically generated, that their work is being replaced by a machine. Besides that, I think it goes without saying that correcting someone else's mistakes is much more tedious than translating from scratch.

Thanks.

[trypsynth]

@tareh7z thanks for the reply! As far as I know, DeepL isn't using an LLM for their translations. I could be wrong, but I recall using it in Spanish class in 2021, before the release of ChatGPT, and it had super solid translations. Not perfect, of course, the phrase "lost in translation" doesn't exist for no reason, but it could make me full, contextually aware sentences. It would be sad, and somewhat although not entirely surprising, if they ditched all that infrastructure and are now using an LLM. The main reason I thought of this approach is because software at my dayjob uses it, and AFAIK we've gotten no complaints, but I take your point too, hence why I asked here.

[qt06]

I think automatic translation is feasible. Because most automatic translation is fine. We can proofread and correct machine translation.

[tareh7z]

Hi @trypsynth!

Thank you for the clarification regarding DeepL. You are right that even before ChatGPT it was using neural machine translation and producing very good results. And yes, for full sentences with clear context it can be impressively good.

My main concern is not actually the general quality of translation in regular text, but specifically software localization.

UI strings are mostly short and without context, for example words like "Open", "Apply", "Continue", "Issue". Without additional explanation. And now, even a strong system can choose the wrong meaning, grammatical gender, or case in many languages. A human translator will usually ask: "what exactly does this button do?", or will immediately know because they can recognize the context, or because they already use the program. A machine cannot do that.

Another problem is terminology consistency. In software it is very important that the same term is translated the same way everywhere. And a machine can translate one word in three different ways. Then the translation becomes inconsistent and unclear.

And if translations are automatically generated and then frozen, realistically most contributors will not actively look for subtle errors. I think people are usually more motivated to create than to correct machine text. Over time this can reduce the overall quality, without anyone directly complaining.

As for the fact that there are no complaints at the company, that makes sense, but users rarely report translation issues unless they are truly catastrophic.

I understand everything, but I think that localization quality and community motivation are important factors that should be considered alongside efficiency.

Thanks again for opening this discussion.

[trypsynth]

Hi @tareh7z and thanks for your detailed information. My goal here isn't to replace human translators, you guys are incredible and fast. That said, there are some languages in Paperback that are just flat out incomplete, such as French and Spanish. I had DeepL and Claude update the French translation and used my shitty Spanish knowledge to update the Spanish one also with some DeepL help pre 0.8, but I don't like that. How does a hybrid solution sound? I.e. if you know a language and want to maintain a Paperback translation for it, you take "ownership" of that language. Basically how it works now, but actually properly codified somewhere. For any languages that have no owner, or an inactive one, we'll have DeepL run over the strings right at the start of the translation freeze, noting which languages have no maintainer in the description. Then, if someone knows that language and wants to make sure it didn't fuck it up, they can look at it, fix some things, and if they want maybe even become the maintainer for that language. The reason I propose this is because while yes, it's true that DeepL can miss context, grammar, etc., having at least a fully Spanish UI with maybe some questionable grammar is preferable to half Spanish, half English, at least I'd think.
Thanks again for all your input on this!

[tareh7z]

Hi @trypsynth!

The ownership model you are proposing actually sounds like a good step forward to me. Clearly defining who maintains which language makes sense.

As for the hybrid solution for languages without a maintainer, I can understand the logic behind it. It is better to have the interface fully translated than to have it half in English and half in the target language.
However, if DeepL is used strictly as an initial draft for languages without an active maintainer, and if it is clearly indicated that the translation is machine-generated and currently has no maintainer, that sounds more acceptable. It is important that such translations are not treated as final, but remain open to serious revision.

Also, it might make sense to clearly mark somewhere that a specific language "needs maintainer", so that people who know the language can more easily take responsibility and improve the quality.

Thanks again for involving translators in the decision-making process and for being open to discussion.

[trypsynth]

@tareh7z Yes, I'd definitely put something in the UI, probably next to the language name in the combo box, indicating that it is likely imperfect and encouraging those who use it to contribute. I don't think I'd use DeepL to add new languages, at least not often, but for languages like French, for example, where someone did the initial work and then stepped away, it could be quite helpful, I think.