Wikipedia talk:Translation

Implementing the above RFC

Hi @Chaotic Enby: and others. Following the above RFC closure, which found consensus for Option B and the adoption of a new guideline, I have created a new page WP:LLMTRANSLATE which contains the text of the guideline as agreed.

I implemented it this way because the existing project page WP:TRANSLATE already contains a substantial amount of background information and advice that has not been designated as guideline, and the adopted guidance is specific to LLM-assisted translation.

If you think there is a better way to arrange it, or I've missed anything, then please do comment here. Cheers  — Amakuru (talk) 12:57, 10 February 2026 (UTC)

The original proposal was to use the section parameter of the guideline template, and incorporate this text here. I cannot agree that there should be a separate page. This simply makes things more confusing for editors seeking to translate articles. The text of LLMTRANSLATE should be merged into this page. Yours, &c. RGloucester — ☎ 13:16, 10 February 2026 (UTC)

Thanks a lot! Regarding @RGloucester's proposal, I think that we could use something like {{guideline|section=y}} to mark only that specific section as a guideline when incorporating it in WP:TRANSLATE? Chaotic Enby (talk · contribs) 13:27, 10 February 2026 (UTC)

Hi both, unfortunately I don't agree with the suggestion above. It would be confusing for readers to have a guideline hidden amongst what is otherwise an unvetted information page, and in fact the guidance at {{Guideline}} explicitly recommends against this. It says: "[section=y] is only used when an guideline of a particular sort (e.g. a naming convention) is a section in a larger combined guideline" and "The template with or without this parameter should not be used on wikiproject advice pages; they are essays not guidelines." Therefore I think a separate page for the new guideline is required here, as long as the rest of the Translation page does not enjoy guideline status. Cheers  — Amakuru (talk) 17:35, 10 February 2026 (UTC)

Thanks! That's on me for forgetting to read the documentation before posting! Chaotic Enby (talk · contribs) 18:01, 10 February 2026 (UTC)

No worries at all! (I only found out about that piece of guidance myself when I followed the link to the template in question...) That being the case, I suspect the separate page is the only way to go, unless it's folded into some other page that's already a guideline. We can make sure it's prominently linked from all the relevant places.  — Amakuru (talk) 18:52, 10 February 2026 (UTC)

Just added a "see also" at WP:MACHINETRANSLATION! Chaotic Enby (talk · contribs) 19:09, 10 February 2026 (UTC)

This is very unfortunate, Chaotic Enby. I specifically asked you about this point prior to the RfC, and you said that this guidance would be incorporated into this page as a section. We've only just gone through the process of merging all translation-related guidance into this page – now we are back to taxing editors with having to consult multiple pages. Much as with the user right proposal, this problem has arisen because of the slapdash manner in which you launched this RfC. I hope you will take this as a lesson to clearly define what precisely will happen if and when a proposal is adopted in future RfCs. Yours, &c. RGloucester — ☎ 22:59, 10 February 2026 (UTC)

Template/category for unreviewed LLM translations

One may note that the new guidance specifies: You may put the raw LLM-translated text into your userspace (preferred) or draftspace while you work on it. If you do, you must immediately tag it as automatically translated text needing review from a human who is skilled in both languages. It would be helpful if someone more proficient than myself could create a template/categories to facilitate the implementation of this guidance. One may see the existing Category:Pages with unreviewed translations, but I expect a new one should be created specifically for pages with unreviewed LLM-based translations. Yours, &c. RGloucester — ☎ 23:42, 10 February 2026 (UTC)

Hello RGloucester. I modified the existing Template:Cleanup translation and updated its documentation accordingly so that a message informing the reader about WP:LLMTRANSLATE appears in the message box when the type parameter has the check-llm value instead of just check. Otherwise, it behaves exactly as with the check value. That is, it adds the article to Category:Wikipedia articles to be checked after translation. This is probably the minimum number of changes needed in order to adhere the RFC's decision. No new template, no new category.

If you think this is the wrong way to go, let me know. I can modify my changes or just revert them. AntinatalistLiberationFront (talk) 23:54, 28 February 2026 (UTC)

@AntinatalistLiberationFront: I appreciate that you have attempted to address this problem, but please pay attention to the actual text of the guideline. LLM-translated content that needs cleanup should never be placed in the article space. We need a template/category for user and draftspace content, not for articles! Yours, &c. RGloucester — ☎ 23:57, 28 February 2026 (UTC)

I see. So, the new template will essentially be a temporary tag for the duration the translated article is in draft/userspace. Once the LLM-assisted translation has been reviewed, the template can be removed and the article can be moved to mainspace.

I will revert my changes and see if I can create a new template using either the category you provided above or a new one. Is it important to have a separate category just for pages translated with LLM-assistance? AntinatalistLiberationFront (talk) 00:11, 1 March 2026 (UTC)

Yes, the new template would serve as a temporary tag for draft/user space content in need of review by someone able to check it. In an ideal circumstance, this template would populate a category that potential reviewers could use to find articles that require checking. Given that this requirement only exists for LLM-generated content, I do think that the category should be separate. Yours, &c. RGloucester — ☎ 00:15, 1 March 2026 (UTC)

Second attempt: I have created a new template in my userspace and you can see it in use in one of my drafts. Please let me know what you think.

When the calling page is either in draftspace or in userspace, the template adds that page to the test category Category:X3, using it as a placeholder for now. An issue with creating a new category for this scenario is that WP:USERNOCAT doesn't make any exceptions for translations or any other case of draft/userspace categories. We could use a hidden category but I'm not sure that it wouldn't be a violation of WP:USERNOCAT. AntinatalistLiberationFront (talk) 10:48, 1 March 2026 (UTC)

Thank you for your work. This seems like a good start to me. I see no violation of WP:USERNOCAT, as this will not be a 'content' category, but a maintenance one – see the exception for project categories at WP:USERNOCAT. 'Project category' is defined at WP:PROJCATS. Yours, &c. RGloucester — ☎ 11:20, 1 March 2026 (UTC)

• New template: Template:LLM-assisted translation
• New category: Category:LLM-assisted translations in need of review
• Example draft: Draft:Elena Knows
• Entry in template index: Wikipedia:Template index/Translation

RGloucester, and anyone else interested, please let me know what you think. AntinatalistLiberationFront (talk) 18:01, 1 March 2026 (UTC)

Excellent work! Thank you so much. Now, at least, the guideline can operate as was intended. If I were to suggest one minor improvement, I notice that the default template language still uses the term 'article'. Would it be possible to change this to 'page'? Yours, &c. RGloucester — ☎ 22:18, 1 March 2026 (UTC)

Great work, thanks a lot! Following RGloucester's advice, I set the default as "draft" for drafts and "page" otherwise in the second parameter. Hope that it is okay with you! Chaotic Enby (talk · contribs) 22:49, 1 March 2026 (UTC)

Absolutely, your addition makes sense. Thank you both. AntinatalistLiberationFront (talk) 06:17, 2 March 2026 (UTC)

Gibberish

My usual protocol for identifying gibberish in G1 is to check a machine translation tool for gibberish output, and then checking against a different machine translation tool. This is important for Daisuke Roppungunai (talk · contribs) SPI's, who often create gibberish drafts in non-Latin scripts. This works because neural-network machine translation tools are known to hallucinate if prompted with gibberish, but when this happens, two different tools often give completely different results. Sometimes, the output will be gibberish as well (empirically, this happens often in Devanagari) and LLMs may also refuse to translate (e.g. The sentence is a meaningless combination of characters.) –LaundryPizza03 (dc̄) 18:34, 16 February 2026 (UTC)

Guidance on sourcing

Could we have some clear guidance on sourcing? Can a (human) translator translate text and its supporting citation without checking the text of the source? The reason I ask is this: At the moment, we have two statements, one ambiguous, and potentially conflicting with the other. The original guidance on content states:

"Meet the English requirements: Many foreign language articles do not meet the levels of verifiability that is expected in English Wikipedia articles. If an article has sections with no citations then follow the guidance in UNSOURCED and either provide additional inline sources, or add appropriate citation needed templates as described in UNSOURCED. Content that cannot be verified must not be imported into English Wikipedia.

The difficulty with this is how to interpret the last sentence. Does it mean "Unverifiable content must not be imported" or does it mean "content that has not been verified by the translator must not be imported"? In the context of a paragraph on standards of content, and given that the preceding sentence was:

"All material in a Wikipedia article, including everything in lists, tables, and captions, must also be verifiable. This means that people are able to check that information comes from a reliable source."

(my bold) ...I would interpret the requirement as "it must be possible for someone ("people", the reader, not specifically the translator) to check that the information in this translation came from somewhere reliable". But the new machine translation guidance states:

"c) You've checked the sources in the origin language article and you're sure the translated text reflects them fairly, and each fact or claim likely to be challenged is supported by an inline citation to a reliable source"

This is quite clear that the translator must go beyond merely checking that the translation is a true and accurate translation of the French (or whatever) Wikipedia, but must in person check the text of the original sources to make sure that the French Wikipedian didn't misrepresent them when he/she wrote the French article.

This is clearly a very specific interpretation of the previous more general guideline on verifiability, but the new guideline is limited in scope:

"Scope: These rules apply to machine translation tools that include a large language model ("LLM"). Assume these rules apply to any online translation tool unless you've confirmed there's no LLM element."

(1) If we're going to require translators to personally verify or omit every source used in the original text, then logically we should do so whether LLM was used or not.

(2) My take on it is that there's a big difference between including information that is not in any way verified, and including information that's been verified by an editor of a foreign Wikipedia (and could potentially be verified by readers). We already take it on trust that creators of English articles have read the sources they use; it feels xenophobic not to extend the same trust to creators of French and German articles. We don't expect an AfC reviewer to remove all text from an article he/she is reviewing if he/she can't obtain the book that the original creator used... the reviewer is allowed to assess whether on balance the creator seems to have done a decent job, and release their article into main-space based on a bit of checking and an assumption of good faith extended to the uncheckable bits. A translator is exactly what it says on the tin: a translator. You're translating a page, not creating one. When you obtain an official translation of your birth certificate, the translator is certifying that their translation is accurate and good, not certifying that you were born in Manchester in 1974, the checking of which goes beyond their role. Wikipedia-translators are in the same boat: their academic honesty is that their translation is accurate, and that should be the hard limit of their obligations. Elemimele (talk) 09:59, 10 April 2026 (UTC)

My take is much the same as yours: the minimum requirement ought to be that translations from a foreign wiki are faithful to the original. The translator should be allowed to assume good faith on the part of the source-language article's editors, and not be required to re-verify everything. That minimum requirement should not change for LLM-assisted translations, provided that the original sourcing is retained in full.

If any sources not cited in the original have been added or substituted (by the translator or by an LLM), the translator must ensure that they verify the claims. In this case it is perfectly reasonable to have a stronger requirement with regard to any new sources introduced by an LLM, which will of course require particular scrutiny. Rosbif73 (talk) 11:49, 10 April 2026 (UTC)

I whole heartedly disagree with part of this. Editors should be required to verify everything used, regardless if it's new material or a translation from frwiki, dewiki, etc. I think it's poor to class it even remotely close to calling it xenophobic. The English language project has different standards than these projects and others, and I know from first-hand experience that other projects can sometimes have shortfalls when it comes to reliable sources. Whether it's a translation or a brand new article, verifiably is everything. • Quinn (talk) 12:12, 10 April 2026 (UTC)

Different standards is a red herring in this context. Different wikipedias have different overall requirements about verifiability, so citations may be missing in another language, which is unacceptable here. But if a citation is present, it either supports the statement to which it is attached, or it doesn't. There is no intermediate, no lower-but-acceptable standard. No Wikipedia willingly accepts citations that are contrary to the text to which they're attached. So if a citation is present, the only reason not to accept it is a belief that the author of the foreign-language article actually misrepresented the source they used. Yes, we can be that suspicious in pursuit of reliability, but if we pursue that belief to its logical conclusion, we certainly shouldn't be accepting articles at AfC unless the AfC reviewer has actually read every source to confirm that the English wikipedian hasn't misrepresented their source. My comment about xenophobia is that we are expecting that foreign-language editors are incompetent or dishonest, while assuming that English-language editors are not, and that strikes me as being unjustified.

Another way to look at it is the balance of reliability versus usability. What we gain in reliability by assuming foreign-language editors don't read their sources is very small, while what we lose in verifiable information by refusing to give our readers information about sources that we don't ourselves have is potentially very large. Elemimele (talk) 13:51, 10 April 2026 (UTC)

Hmmmm... but they actually might also have different standards about what sources are allowed to support a statement. Perhaps we should be clearer in pointing out that the source used in the other languiage-wiki not only needs to support a given statement, but also be a reliable source in EN-Wiki. Lectonar (talk) 14:04, 10 April 2026 (UTC)

This was more along the lines of what I probably didn't phrase correctly. A specific instance I had in mind was, I personally work on football articles. Other projects allow certain websites as citations, such as Transfermarkt—a user-generated website with no oversight. It may support a statement and may even be correct, but enwiki rightly classifies this as unreliable. I'm sure there are similarly related websites that other projects allow, but aren't RS here. • Quinn (talk) 14:09, 10 April 2026 (UTC)

A translator is exactly what it says on the tin: a translator. You're translating a page, not creating one. When you obtain an official translation of your birth certificate, the translator is certifying that their translation is accurate and good, not certifying that you were born in Manchester in 1974, the checking of which goes beyond their role. Wikipedia-translators are in the same boat: their academic honesty is that their translation is accurate, and that should be the hard limit of their obligations. As I understand it, you are correct that this is the standard for human translations. For LLM-assisted translations, there is a stricter standard for verification, as explained in WP:LLMTRANSLATE. You can see the arguments in the RFC above to understand why this is the case, but it basically comes down to the community discovering many examples where an LLM-assisted translation resulted in big changes in meaning/insertion of new claims that went beyond what was said in the original-language wiki article. We aren't re-checking citations because we don't trust the human editor on the other wiki, we are re-checking them because we don't trust the LLM to not add in additional claims that the original human editor didn't write. -- LWG ^{talk (VOPOV)} 14:37, 10 April 2026 (UTC)

LWG That's a valid point, and a good explanation of why we might have different standards for LLM translations and human translations - thank you! Would it be worth including the explanation somewhere in the text? Quinn, yes, I've thought this over, and your football example is a really good one. I do think that a translator should do an assessment of the likely quality of the source. I translate a lot of stuff on old organ builders, so I was thinking about books: if they come from a respectable publisher, and particularly if they have a widely-respected author, I'm prepared to take it on trust that the source is good, even if it's now out-of-print and hard to obtain. But I agree it's a difficult balancing act. Elemimele (talk) 15:06, 11 April 2026 (UTC)

I’m experiencing a problem

Whenever I switch into an Arabic text, I get logged out. If I go back to English, I am still logged out. Wikipedian12512 (talk) 00:26, 6 May 2026 (UTC)

Try asking at our Wikipedia:Help desk please. Lectonar (talk) 07:43, 6 May 2026 (UTC)

Question:

Is there any variation between the requirements for an article between the different language wikipedias? Wikipedian12512 (talk) 01:34, 11 May 2026 (UTC)

Yes, every Wikipedia and Wikimedia project has its own policies and guidelines, which you need to familiarise yourself with before trying to translate content. I believe that English Wikipedia has the strictest policies of all, especially concerning Notability (what can be included in the first place) and Verifiability (referencing). ClaudineChionh (she/her · talk · email · global) 13:49, 11 May 2026 (UTC)

Okay. I’ll just go with a rule of “don’t translate stubs”. Wikipedian12512 (Talking is fine | contribs) 01:32, 17 May 2026 (UTC)