You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Simplified Chinese Translation is PROBLEMATIC #399
It seems that the Simplified Chinese translation of this website was automatically generated by AI and added to the localization files without being proofread. Some of the translations sound a bit unnatural or too casual, which takes away from the site's professional tone and makes it harder to effectively convey its intended message to Simplified Chinese users.
I'm currently working on a revised translation and will open a Pull Request later before 2026/5/17 24:00 GT+8.
Thanks for letting us know. This is a peril of accepting PRs "blindly", which we've been doing thus far when we don't have a reviewer that can read the language with sufficient fluency to understand nuances and style.
It might be interesting to review the recent closed PRs around zh-CN to examine the differences to identify if major regressions happened with any of the individual contributions. I am sometimes concerned that many of the large-scale translation PRs are just copied from the output of an LLM, despite the caution against doing that.
Regardless, we will appreciate any assistance with making the text more natural and feel less machine-translated!
(It might also bear mentioning that since the impending lockdown affects only Android Certified Devices, which are generally unavailable in mainland China due to the blocking of Google Mobile Services, that this issue won't directly affect many users in mainland China since they tend to be running other AOSP-derived Android variants; but I will leave it to the translator's discretion to determine how best to highlight this caveat, if at all)
Actually, if the translation is made by LLM without any prompt, it will be better than the current version. From my perspective the translation is probably made by LLM with an "informal tone" prompt, which mixs plenty of words unsuitable for such a website, such as “为啥” (a casual form of "why", with some dialect tone) and “点头” (casual synonym of "approve", but is not accurate).
Most Chinese softwares provide free service, but at the cost of massive annoying ADs. Therefore, some users will get Root access to install another AOSP OS or install some community adblockers, both are gradually blocked by long user tests and software blacklists. We may face different exact issues, but we are all fighting for a free operating system.
It's over 00:00 AM here and my dormitory is during blackout in weekday nights. I have finished my translation before, but it's in my PC and not pushed to my github repo, and I'm sleepy now. I'll push the git to cloud and check the PRs carefully the next morning.
I scanned the latest PR. The version prior to the current one was even more casual-style, or from my perspective, not suitable. But the latest PR tweeked the translation instead of overhaul, thus not fixing the problem. I'll check the PRs carefully tomorrow morning.
I am the contributor of #386. I am commenting here because I believe some of the statements in the current discussion have problems.
Let me be clear first: I am also against using such a punchy tone without careful consideration. I already stated this in the description of my pull request.
The reason I still followed that tone is as follows:
Fully re-translating the whole project would be a very large amount of work. I came here because I noticed multiple mistranslations, which I already listed in a previous pull request.
The "professional" version is not necessarily better than the "punchy-toned" version, especially in its current form.
Previous contributors also followed a similar tone, and I am not sure what the project's official attitude toward that tone is.
TL;DR:
You are not simply restoring a "professional" tone. You are replacing the existing tone with your own preferred tone.
Your statement does not accurately reflect what happened. It amplifies the perceived damage while skipping necessary fact-checking.
The real issue is whether the translation is accurate, natural, consistent, and properly reviewed.
As we all know, China is, at least for now, a country where public discussion is often shaped by infighting, cynicism, and quiet quitting. People living in China are often more likely to attack each other than to calmly engage with a neutral message. A general, normally toned article may weaken the intended urgency. So if your point is that the translation should be "harder to effectively convey its intended message to Simplified Chinese users," then I do not agree.
Don’t get me wrong, I’m not attached to that tone. I am only disagreeing your point.
Besides that, your edited version is not professional as you claimed, nor is it even natural. Generally speaking, it reads like translationese.
Example:
Original: "You bought an Android phone because Google told you it was open. You could install what you wanted, and that was the deal."
Translation: "你购买 Android 手机,是因为 Google 曾声明它会开放。你可以任意安装软件,这曾是说明好的。"
Back-translation: <The translation is technically understandable.>
However, there are several problems.
"这曾是说明好的" is unnatural Chinese. "说明好的" is not a natural collocation here.
The structure copies English too directly. English can say, "You could do X, and that was the deal." But Chinese would usually express this as "我们当初说好的,你可以……"
"这" feels vague and awkward. In Chinese, "这曾是……" sounds stiff because it points back to the whole previous clause in a very English-like way.
Another example:
Original: "This is bigger than Android:"
Translation: "Android 只是起始"
Back-translation: "Android is only the starting point."
From a localization perspective, this is not necessarily wrong. The problem is that it does not sound natural in daily article/life. Same problem as previous one.
And another example:
Original: "But wait, isn't this..."
Translation: "但是稍等,这不就是……"
Back-translation: <Correct translation>
But do we actually say "但是稍等" in daily life? It sounds unnatural. Still same problem.
You also made disruptive translations here. For example:
Original: "every hardware manufacturer on the planet is watching."
Your suggested translation: "全世界其他硬件厂商都可能效仿。"
Back-translation: "Every other hardware manufacturer in the world may imitate it."
The original phrase "is watching" echoes "The whole world is watching," especially considering the broader context: "tell them the whole world is watching and they'll never get away with it again." Translating it as "都可能效仿" changes the meaning from "watching" to "imitating." I think this should be considered over-translation.
I also do not think you are strictly making the translation more professional.
For example:
Original: "Worse: this flow..."
Translation: "祸不单行:这一流程..."
Back-translation: "Misfortunes never come singly. This flow..."
Strictly speaking, this also alters the translation, just as my pull request did, only to a lesser extent. "祸不单行" is not equivalent to "Worse." I would call this a "clever" translation if we are not talking about "professional." This is also a stylistic localization choice, not a strictly neutral or professional rendering.
(And don't you find "祸不单行" a bit awkward in this context?)
By the way, I noticed you rejected my similar edit, "Who this hurts" ("谁会遭殃"), even though you used the same localization approach like above.
Another example:
Original: "Nine steps, 24-hour wait"
Translation: "然而需要九步操作、等待二十四小时强制冷静期。"
Back-translation: "However, it requires nine steps and waiting for a 24-hour mandatory cooling-off period."
The original has a speech-like tone. You also modified it, yet called it a "professional tone."
From my perspective, at least based on the edits you made, you are not trying to restore a "professional" tone. You are trying to impose your own tone, because you dislike the punchy tone. I also dislike that tone, but that does not mean your version is automatically more professional or more accurate.
I also digged the initial copy (aa3f951) when I was wrting #386. As far as I can tell, the original author may have used an LLM to adjust the tone, but probably did not use an LLM for the actual translation. For example:
Original: "free and open-source"
Translation: "免费开源"
Back-translation: "non-paid and open-source"
This is not the kind of mistake expecting from an LLM. Since LLMs are generally trained on large amounts of public web text, including common phrases like "free and open-source software," they are unlikely to mistranslate such a standard term so literally.
So when discussing the risks of using AI tools, I think we should first make sure everyone is talking about the same thing. There is a difference between using AI to translate, using AI to adjust tone, and manually translating text that may later be edited with AI assistance. Those cases should not be treated as the same issue. Obviously, your current statement is misleading the maintainer/owner of this project.
It is easy to say that we should "help the site improve the translation," but the question is: Who will do that? Who maintains the standard? Who is responsible for the final result? And how long?
Before my previous pull request, I honestly could not believe that no one had fixed obvious mistranslations such as "evidence of your private signing key," which had been translated as "private singing key." That incorrect translation was even tweeted into different versions. Is that really something that can be explained simply as "the damage caused by AI"? An LLM could have easily identified that mistake.
"if the translation is made by LLM without any prompt, it will be better than the current version."
Unfortunately, I also do not see your edits as clearly better than what an LLM could produce. With proper prompts, an LLM can generate a version that sounds more natural for everyday Chinese readers.
So I think the issue should not be simplified as "AI translation caused damage" versus "human editing improves quality." The real issue is whether the translation is accurate, natural, consistent, and properly reviewed. Right now, I do not think your proposed edits fully meet that standard either.
But in any case, this project does not seem likely to have a dedicated reviewer. In that situation, the only realistic solution may be to add an automated review step to the workflow, for example by using Copilot to flag obvious mistranslations, terminology inconsistencies, and formatting issues before changes are merged. @marcprux
Activity
marcprux commentedon May 17, 2026
Thanks for letting us know. This is a peril of accepting PRs "blindly", which we've been doing thus far when we don't have a reviewer that can read the language with sufficient fluency to understand nuances and style.
It might be interesting to review the recent closed PRs around
zh-CNto examine the differences to identify if major regressions happened with any of the individual contributions. I am sometimes concerned that many of the large-scale translation PRs are just copied from the output of an LLM, despite the caution against doing that.Regardless, we will appreciate any assistance with making the text more natural and feel less machine-translated!
(It might also bear mentioning that since the impending lockdown affects only Android Certified Devices, which are generally unavailable in mainland China due to the blocking of Google Mobile Services, that this issue won't directly affect many users in mainland China since they tend to be running other AOSP-derived Android variants; but I will leave it to the translator's discretion to determine how best to highlight this caveat, if at all)
QZCrane7021 commentedon May 17, 2026
Actually, if the translation is made by LLM without any prompt, it will be better than the current version. From my perspective the translation is probably made by LLM with an "informal tone" prompt, which mixs plenty of words unsuitable for such a website, such as “为啥” (a casual form of "why", with some dialect tone) and “点头” (casual synonym of "approve", but is not accurate).
Most Chinese softwares provide free service, but at the cost of massive annoying ADs. Therefore, some users will get Root access to install another AOSP OS or install some community adblockers, both are gradually blocked by long user tests and software blacklists. We may face different exact issues, but we are all fighting for a free operating system.
It's over 00:00 AM here and my dormitory is during blackout in weekday nights. I have finished my translation before, but it's in my PC and not pushed to my github repo, and I'm sleepy now. I'll push the git to cloud and check the PRs carefully the next morning.
QZCrane7021 commentedon May 17, 2026
I scanned the latest PR. The version prior to the current one was even more casual-style, or from my perspective, not suitable. But the latest PR tweeked the translation instead of overhaul, thus not fixing the problem. I'll check the PRs carefully tomorrow morning.
JoeyC-Dev commentedon May 19, 2026
I am the contributor of #386. I am commenting here because I believe some of the statements in the current discussion have problems.
Let me be clear first: I am also against using such a punchy tone without careful consideration. I already stated this in the description of my pull request.
The reason I still followed that tone is as follows:
TL;DR:
As we all know, China is, at least for now, a country where public discussion is often shaped by infighting, cynicism, and quiet quitting. People living in China are often more likely to attack each other than to calmly engage with a neutral message. A general, normally toned article may weaken the intended urgency. So if your point is that the translation should be "harder to effectively convey its intended message to Simplified Chinese users," then I do not agree.
Don’t get me wrong, I’m not attached to that tone. I am only disagreeing your point.
Besides that, your edited version is not professional as you claimed, nor is it even natural. Generally speaking, it reads like translationese.
Example:
Original: "You bought an Android phone because Google told you it was open. You could install what you wanted, and that was the deal."
Translation: "你购买 Android 手机,是因为 Google 曾声明它会开放。你可以任意安装软件,这曾是说明好的。"
Back-translation: <The translation is technically understandable.>
However, there are several problems.
Another example:
Original: "This is bigger than Android:"
Translation: "Android 只是起始"
Back-translation: "Android is only the starting point."
From a localization perspective, this is not necessarily wrong. The problem is that it does not sound natural in daily article/life. Same problem as previous one.
And another example:
Original: "But wait, isn't this..."
Translation: "但是稍等,这不就是……"
Back-translation: <Correct translation>
But do we actually say "但是稍等" in daily life? It sounds unnatural. Still same problem.
You also made disruptive translations here. For example:
Original: "every hardware manufacturer on the planet is watching."
Your suggested translation: "全世界其他硬件厂商都可能效仿。"
Back-translation: "Every other hardware manufacturer in the world may imitate it."
The original phrase "is watching" echoes "The whole world is watching," especially considering the broader context: "tell them the whole world is watching and they'll never get away with it again." Translating it as "都可能效仿" changes the meaning from "watching" to "imitating." I think this should be considered over-translation.
I also do not think you are strictly making the translation more professional.
For example:
Original: "Worse: this flow..."
Translation: "祸不单行:这一流程..."
Back-translation: "Misfortunes never come singly. This flow..."
Strictly speaking, this also alters the translation, just as my pull request did, only to a lesser extent. "祸不单行" is not equivalent to "Worse." I would call this a "clever" translation if we are not talking about "professional." This is also a stylistic localization choice, not a strictly neutral or professional rendering.
(And don't you find "祸不单行" a bit awkward in this context?)
By the way, I noticed you rejected my similar edit, "Who this hurts" ("谁会遭殃"), even though you used the same localization approach like above.
Another example:
Original: "Nine steps, 24-hour wait"
Translation: "然而需要九步操作、等待二十四小时强制冷静期。"
Back-translation: "However, it requires nine steps and waiting for a 24-hour mandatory cooling-off period."
The original has a speech-like tone. You also modified it, yet called it a "professional tone."
From my perspective, at least based on the edits you made, you are not trying to restore a "professional" tone. You are trying to impose your own tone, because you dislike the punchy tone. I also dislike that tone, but that does not mean your version is automatically more professional or more accurate.
I also digged the initial copy (aa3f951) when I was wrting #386. As far as I can tell, the original author may have used an LLM to adjust the tone, but probably did not use an LLM for the actual translation. For example:
Original: "free and open-source"
Translation: "免费开源"
Back-translation: "non-paid and open-source"
This is not the kind of mistake expecting from an LLM. Since LLMs are generally trained on large amounts of public web text, including common phrases like "free and open-source software," they are unlikely to mistranslate such a standard term so literally.
So when discussing the risks of using AI tools, I think we should first make sure everyone is talking about the same thing. There is a difference between using AI to translate, using AI to adjust tone, and manually translating text that may later be edited with AI assistance. Those cases should not be treated as the same issue. Obviously, your current statement is misleading the maintainer/owner of this project.
It is easy to say that we should "help the site improve the translation," but the question is: Who will do that? Who maintains the standard? Who is responsible for the final result? And how long?
Before my previous pull request, I honestly could not believe that no one had fixed obvious mistranslations such as "evidence of your private signing key," which had been translated as "private singing key." That incorrect translation was even tweeted into different versions. Is that really something that can be explained simply as "the damage caused by AI"? An LLM could have easily identified that mistake.
Unfortunately, I also do not see your edits as clearly better than what an LLM could produce. With proper prompts, an LLM can generate a version that sounds more natural for everyday Chinese readers.
So I think the issue should not be simplified as "AI translation caused damage" versus "human editing improves quality." The real issue is whether the translation is accurate, natural, consistent, and properly reviewed. Right now, I do not think your proposed edits fully meet that standard either.
But in any case, this project does not seem likely to have a dedicated reviewer. In that situation, the only realistic solution may be to add an automated review step to the workflow, for example by using Copilot to flag obvious mistranslations, terminology inconsistencies, and formatting issues before changes are merged. @marcprux