The WMF section of the village pump is a community-managed page. Editors or Wikimedia Foundation staff may post and discuss information, proposals, feedback requests, or other matters of significance to both the community and the Foundation. It is intended to aid communication, understanding, and coordination between the community and the foundation, though Wikimedia Foundation currently does not consider this page to be a communication venue.
Discussions of proposals which do not require significant foundation attention or involvement belong at Village pump (proposals)
This page is not a place to appeal decisions about article content, which the WMF does not control (except in very rare cases); see Dispute resolution for that.
Issues that do not require project-wide attention should often be handled through Wikipedia:Contact us instead of here.
This board is not the place to report emergencies; go to Wikipedia:Emergency for that.
Threads may be automatically archived after 14 days of inactivity.
Behaviour on this page: This page is for engaging with and discussing the Wikimedia Foundation. Editors commenting here are required to act with appropriate decorum. While grievances, complaints, or criticism of the foundation are frequently posted here, you are expected to present them without being rude or hostile. Comments that are uncivil may be removed without warning. Personal attacks against other users, including employees of the Wikimedia Foundation, will be met with sanctions.
The extremely expensive, buggy and ill-thought out "Abstract Wikipedia" has gone live after many, many years, and already it hosts an "article" which not only go against the very purpose of it (putting a complete English article in html is not the way to get automated Wikidata-based translation to 300+ languages), but also creates an unattributed enwiki copy[1]. Not that I am able to get it to load completely, it only returns a few sentences and then nothing happens (which is better than the many errors other pages generate, usually either "Reached max retries. Try again later." or "Wikifunctions returned a failed response: Error in evaluation" or simply "Page not found" when going from "history" to "read" or "page"...). Why this pre-alpha thing has been released to the world is not clear, why the WMF would think it conceptually is a good or feasible idea even less so. Fram (talk) 10:05, 25 March 2026 (UTC)[reply]
Also see [2], it looks like it’s just going to be another Anglo-project, how tf are non-English speakers supposed to develop the wiki's policies when everything is only done in English. It’s a project that’s specifically not meant to be Anglo and is irrelevant to en.wiki. It should be put on ice until people can discuss via the functions or there’s some kind of multilingual support Kowal2701 (talk, contribs) 11:12, 25 March 2026 (UTC)[reply]
Abstract + wikifunctions is 2 sides of the same coin. And if wikifunctions has this many issues still, then abstract shouldn't have been launched as a "beta". The most basic things don't work yet: e.g. take a random page, click on "edit source" or "view history", then click on "page" or "read"... error, every single time. This is not some obscure thing, this is basic functionality for the whole site, and it doesn't work. Type a page you know exists in the search bar (e.g. Cape Verde), the first result in the dropdown is the Abstract Wikipedia page (see the AW at the end), click it, and again you get the "page not found".
This is probably a simple switch somewhere, but the total lack of care displayed by whoever decided that this was ready to go public is staggering.
As for the multi-language aspect, the core business of Abstract... Q143, Esperanto.
"Esperanto is the languages of internationality. An Esperanto is a languages." (sic!)
In French, this gives "espéranto langue international" plus an error (the title doesn't get translated..., international(e) should be female, the verb has disappeared, ...)
In Dutch, it becomes "taal internationaal". No error, but missing most of the poor original.
Q21, "England is a country in United Kingdom." (sic). In Spanish, this gives "Unable to render this fragment due to an unknown error. ", in Swedish "The rendering service is temporarily unavailable. Please, try again later. " in Italian "Wikifunctions returned a failed response: Reached time limit in orchestrator", in Portuguese "Wikifunctions returned a failed response: Error in evaluation", and in Thai "England is a country in United Kingdom." I'm sure it will work wonders in the small languages for which it is intended though!
I wonder how many employees have worked on this the past 10 years instead of on Phabricator and the wishlist... Fram (talk) 19:49, 25 March 2026 (UTC)[reply]
Most Western companies, and the WMF is no exception, would work better with half the people on twice the salary. But we would have to have a different attitude to work for that to succeed. Phil Bridger (talk) 20:07, 25 March 2026 (UTC)[reply]
Thanks for the link. The team bios seem somewhat telegraphic. Are any of them "hot shot" programmers? And they seem pretty happy. In the commercial world, best software is written by developers who are pushed to the limit. But not for me to dive into details of the team. Yesterday, all my dreams... (talk) 14:40, 26 March 2026 (UTC)[reply]
best software is written by developers who are pushed to the limit
Your word was "best", not most widespread. I heavily disagree that this is the Microsoft development process, and in any case, the market share of all Windows systems combined had been precipitously declining for years now because of how bad it is. The attempt by the WMF to do the development process you propose is Knowledge Engine (search engine) and the surrounding turnover. Aaron Liu (talk) 12:36, 30 March 2026 (UTC)[reply]
As a former programmer who's experienced severe, life-impeding burnout from overwork twice in her career, I can't disagree more strongly with the notion that "the best software is written by developers who are pushed to the limit". — Hex•talk11:37, 28 April 2026 (UTC)[reply]
best software is written by developers who are pushed to the limit. Assuming the limit here is a tight deadline, I would argue that software death marches do not result in great software. Tight deadlines result in insufficient time to write quality code. Quality is sacrificed to (try to) meet the deadline. Not to mention the impact on team morale and work-life balance caused by pressure and overtime. –Novem Linguae (talk) 22:51, 28 April 2026 (UTC)[reply]
I checked my browser console when viewing the OP's link and there were over 65 requests to the API! As of last week, the WMF (I'm presuming a different team which didn't talk to this one) has decided to limit unregistered users to 1000 requests per hour. Total, across all projects. That includes search suggestions, DiscussionTools previews, VisuaEditor, etc. So about 30 page views for you, and then every project breaks unless you have an account! Suffusion of Yellow (talk) 20:12, 25 March 2026 (UTC)[reply]
What I don't understand is how we are meant to make an "abstract Wikipedia" that is automatically translated to every language when the very design of the site is so centered around English. In abstract:Help:How to create an article I see a reference to a function called "Article-less instantiating fragment" which creates sentences like "Paris is a city". However, in some languages (e.g. Greek) such a sentence needs an article so the result will be ungrammatical unless a different function is used. Thankfully it seems like someone noticed this issue because the actual page for Paris, abstract:Q90, uses a different function called "defining role sentence" which doesn't have the same problem. But if basic stuff like this is wrong in the documentation, I don't have much faith in the project. In fact, pages like abstract:Q667 (south) still use this "article-less instantiating fragment" function. Warudo (talk) 11:36, 27 March 2026 (UTC)[reply]
Actually, I found abstract:Abstract Wikipedia:Useful functions for article composition which apparently says that "article-less instantiating fragment" is a good fit for sentences like "Nairobi is a city". It absolutely isn't. Again, it's article-less in English and several other languages but not all of them. While I'm not a linguist myself, it really feels like this system was designed without consulting experts. Warudo (talk) 11:40, 27 March 2026 (UTC)[reply]
It seems to me to be referring to the fact that it doesn't require an article in English. It may or may not require an article in any other language you care to mention, so this seems to be a very anglocentric point of view. Phil Bridger (talk) 17:32, 28 March 2026 (UTC)[reply]
No, that's definitely not it! Article-less and article-ful have different semantic meanings (the term article here seems to confuse a lot of people who do not understand the project, which I suppose means it needs a rename). All of these examples would use "article-less" (despite the fact that two of them have indefinite ones):
Golf is a sport
El golf es un deporte
The United States is a country
യുണൈറ്റഡ് സ്റ്റേറ്റ്സ് ഒരു രാജ്യമാണ്
And all of these examples would use "article-ful":
A bird is a dinosaur
Un ave es un dinosaurio
These are saying two fundamentally different things irrespective of language. Wikifunctions is already equiped to handle this distinction. Feeglgeef (talk) 18:28, 28 March 2026 (UTC)[reply]
I don't see any great difference between "Golf is a sport" and "A bird is a dinosaur" apart from the presence/absence of an article in English. Please explain. Phil Bridger (talk) 19:18, 28 March 2026 (UTC)[reply]
Because golf is a singular thing and there are many birds. Article-less is about one thing, article-ful is about a collection. Feeglgeef (talk) 19:57, 28 March 2026 (UTC)[reply]
Then the name of the function should refer to proper and non-proper nouns, rather than articles. The point was that the name of the function is based on English language thinking. TietoTeekkari (talk) 07:58, 4 June 2026 (UTC)[reply]
"Article-less and article-ful have different semantic meanings". But then why are these semantic meanings not mentioned in the documentation of f:Z26039? Instead the function's documentation says Makes a sentence of the form "X is a Y" e.g "Nairobi is a city.", i.e. it takes an entity (X) and its class (Y) and states that it is an entity of that class. What are the semantic differences with f:Z26095? In case you think this is a small problem, it really isn't. Look at the Japanese translation of this documentation (a language that does not use articles). The translator had to use an English example to explain what the function does because in their language the concept does not really exist. Indeed, as far as the users are concerned, these functions are in fact defined in terms of whether the sentences they generate require an article in English. The semantic difference behind that is hidden to them. Warudo (talk) 22:33, 28 March 2026 (UTC)[reply]
Also, by the way Abstract:Q30 currently says "United States is a country. United States is a republic. Washington, D.C. is the capital of United States." So, I guess the article-less function was not the correct one to use in this case. Warudo (talk) 23:36, 28 March 2026 (UTC)[reply]
I've already said this above, but the "article" being referred to here is a indefinite one (a/an) at the front. definite articles (the) have no semantic meaning and therefore are to be added on a language-by-language basis. Feeglgeef (talk) 00:40, 29 March 2026 (UTC)[reply]
Definite article = "the", Indefinite article = "a/an". Also that's not true. The definite article has semantic meaning. It means that we are referring to a specific member of a group and not to the concept in general. In this case, "United States" means any states that are united, which would also include e.g. the Mexican United States. The United States on the other hand are a specific set of states that are united, in this case, the United States of America. Warudo (talk) 00:49, 29 March 2026 (UTC)[reply]
We already have the distinction between the concept of a federation and the example in North America seperated by Wikidata. Some languages (like Malayalam, as in the example) do not use an article in front of the United States. Eventually, the function will be able to handle automatically adding a definite article. Feeglgeef (talk) 14:09, 29 March 2026 (UTC)[reply]
Ok, to be fair, you're right. The generated sentence is wrong but that can be attributed to an incomplete implementation of f:Z26039 rather than a logic error in abstract:Q30 itself.
Still, there are several questions that remain. First, why are the semantics of the two functions defined in terms of English grammar? You said, Article-less is about one thing, article-ful is about a collection. but as I've said above, that's not how it's described in the documentation. But more importantly, if that's the difference between them then why are two different functions for "article-less" and "article-ful instantiating fragment" given to the user in the first place? Why don't you expose a single "instantiating fragment" function that checks if its input has a subclass of (P279) statement in Wikidata? If yes, use an indefinite article, if not, don't use one.
Just FYI, editing your own comments after someone has already answered without indicating the changes is considered bad practice in the English Wikipedia. Not a big deal in this case as you were just fixing a mistake but keep it in mind in the future. Warudo (talk) 14:47, 29 March 2026 (UTC)[reply]
@Feeglgeef no, your statement is plain wrong. This is how 6 sentences above will be translated to Russian (browser machine translation, but it's a correct translation): As you can see, all 6 cases uses exactly the same grammar construction, because Slavic languages has no articles and they don't provide meanings, that is provided by articles in Germanic languages. We, native Russians, believe that this semantic meaning simply does not need to be conveyed. MBH (talk) 18:24, 21 May 2026 (UTC)[reply]
You have to accommodate the languages that don't, though. The names of the function have been thankfully changed now to "subject is instance of" vs "class is subclass of". These are different things, even if they look the same in Russian. Feeglgeef (talk) 18:30, 21 May 2026 (UTC)[reply]
The copy of enwiki is attributed now, and the project is not doomed because one user fails to abide by copyright law or to make a quality article.
As for the beta release, it's impossible to debug or improve without community content, so I'm not sure what you'd have the WMF do. Feeglgeef (talk) 17:03, 28 March 2026 (UTC)[reply]
"one user fails to abide by copyright law or to make a quality article." Not a single user has made a "quality article", which is hardly possible with the current setup. And the actual intention of the project, automatic translation to small languages, is just not happening.
Random "article"[4] translation to French: "Wikifunctions returned a failed response: No matching lexeme for item in language"
Random "article"[5], translation to Italian: "Wikifunctions returned a failed response: Number of arguments mismatch"
Random "article"[6] translation to Spanish: "(en) Bahrain is a country in Middle East." Hey, I can understand Spanish!
Random "article"[7] translation to Dutch: "Brussel is the hoofdstad of België." THE hoofdstad? Yep, clearly not a problem with the use of articles...
Random "article"[8] translation to French: "Paris est une ville. Wikifunctions returned a failed response: Error in evaluation" Translating two sentences was a challenge of course.
Random "article"[9], no translation: "Wikifunctions returned a failed response: Could not acquire WASI runner within time limit" and "Wikifunctions returned a failed response: No matching lexeme for item in language" and "Unable to render this fragment due to an unknown error."
You state "As for the beta release, it's impossible to debug or improve without community content, so I'm not sure what you'd have the WMF do." which is absolute nonsense. You don't do a public release of such extreleky buggy software, and you don't call it a beta either. The errors found so far are not edge cases where mass testing is necessary, but things a developers + QA team should easily have found. WMF should, for a $6 million + project, have some people on board who understand what this project is intended for and can test it before it is released as a beta. The release of a severely immature project where even the most basic fundamentals are being questioned is irresponsible. Then again, the decade-long development seems to have been irresponsible as well. Fram (talk) 10:19, 29 March 2026 (UTC)[reply]
Fram, you have more than enough examples now. Item 17 is disastrous and tells me that the entire system needs a rewrite following a redesign of the architecture. But that would be throwing good money after bad. I predict that there will be no remedy for this project anytime soon. Once the underlying software architecture has problems 1 million bandaids placed on it will be no cure. Yesterday, all my dreams... (talk) 12:32, 29 March 2026 (UTC)[reply]
None of these problems are architectural. All of these are the result of the work of a few (like less people than that participated in this topic) community contributors. Feeglgeef (talk) 14:17, 29 March 2026 (UTC)[reply]
So there is no Wikidata item for "is a" in French, but this thing will be able to handle translations to languages with very few editors somehow? And "it knows how to say the words in Dutch", er, no: "the" (or "is the") is English, not Dutch. Again, it can't even translate that most basic building block to a language with a large editing base. As for 19, I have no idea how "unknown error" and a failed time limit are the responsibility of Wikidata contributors, nor how a translation tool was ever tested by the developers if the most basic aspects are missing in French, Spanish, Dutch, ...
Someone at the WMF was aware that translation (and specifically translation to languages with very small user bases) was the intention of this tool, right? Because it sure doesn't look that way. I don't see how they can have tested this at an alpha-level to give it the green light to go to beta, if if can't even handle these basic things. Blaming it on Wikidata contributors is rather rude, the developers/testers should have added things like "is a" or "the" or ... in major languages (both the ones I just tested, but also completely different ones with other scripts and grammar) to Wikidata. I assume these people have some fluency in Wikifunctions and Wikidata editing, and in languages and translation? Seems a prerequisite for such a project... Fram (talk) 15:30, 29 March 2026 (UTC)[reply]
Question: How many people who work on Google Translate are linguists? Please answer to yourself before you read further. Answer: zero. We have all learned long ago that fiddling with linguistic constructs will end in one place: the shelf that holds the Aspirin bottle. So please do not assume that as a requirement. Their problems are much deeper and architectural in nature. They should have never used Wikifunctions, given that they are crowd sourced. Alas the key issue is that we can all huff and puff for ever but we have no control on what WMF does. So maybe we should all take an Aspirin and move on. Yesterday, all my dreams... (talk) 16:22, 29 March 2026 (UTC)[reply]
Google Translate is not entirely human-made though, but is based on computer-learning (statistics, deep learning and brute force basically). Abstract is based on "humans will build it all", which, while admirable, then requires humans with very specific skills. And "we can all huff and puff for ever but we have no control on what WMF does" is false, we have forced them to shelve things like Flow and Gather. Fram (talk) 16:49, 29 March 2026 (UTC)[reply]
I know exactly how G-translate works. Thanks. My point was/is that the crowd sourced paradigm works for text input but not for software. The world is moving towards automatic software generation now, so crowd sourcing will be inherently inefficient and error prone. But I think I have said enough now. No more comments from me here. You are right in objecting to the project but time will tell how much power you have over WFM. Cheers Yesterday, all my dreams... (talk) 22:57, 29 March 2026 (UTC)[reply]
How many people who work on Google Translate are linguists? Why are you using Google Translate as an example? Do you know how many mistakes Google translate makes when translating to and from languages other than English? For some reason English homographs throw it off completely. I've seen it do stuff like this where I asked it to translate a verb that means "to bear" and it came up with a word for the mammal. Even ignoring the fact that English is clearly used as an intermediate language here even though it isn't suited for this purpose (probably not by design but because of the way Google translate was trained), why on Earth is Google translate translating a verb as a noun in the first place? I think stuff like this shows how Frederick Jelinek's quote is outdated and is leading us astray at this point. Warudo (talk) 16:53, 29 March 2026 (UTC)[reply]
There is no implementation of the function in Dutch. The Abstract Wikipedia team has not added an implementation in Dutch because the community is responsible for creating the function in Dutch, and nobody has created the function in Dutch.
Essentially, how the current English implementation is to string together the first concept with "is the" with the second concept with "of" with the third concept. The function can get the Dutch terms for the concept, but it does not know how to string the words together because nobody who speaks Dutch has told it how. This is not something that the WMF can magically fix. Eventually, when the project is older than a week, somebody will implement it in that language, and in Malayalam, and Dagbani, and Massa, and Southern Altai, and Dusun. Feeglgeef (talk) 21:49, 29 March 2026 (UTC)[reply]
Please see my last response to Fram. Anyway, time to cool off and move on before someone busts an artery here. You will be glad to know that I shall make no further comments here. Now, in what language shall I say goodbye? Yesterday, all my dreams... (talk) 23:02, 29 March 2026 (UTC)[reply]
"when the project is older than a week"? It's a decade old or thereabouts, Wikifunctions specifically was created in 2020 and launched in 2023. But sure, some Southern Altai Wikipedian will go to Wikidata to translate everything that is needed, then go to Wikifunctions to translate all necessary functions into the grammatically correct version of their language (assuming naively that the used function can be one-on-one transformed to one in their language for every use of it), just so they can then autocreate stilted article stubs instead of either writing them directly, which would require a lot less effort and give a lot more satisfaction, or using an online translation tool to translate an existing Wikipedia article to give them much easier results. Totally realistic. Fram (talk) 07:32, 30 March 2026 (UTC)[reply]
When I go to this totally not-alpha project, to the page we are discussing[10], and click on "defining role sentence in English as string" (which should apparently go to abstract.wikipedia.org/view/en/Z28109), I am taken to the Abstract Wikipedia Main page[11]. Please explain to me again how this has been sufficiently tested and was ready to be opened to the wider public? Fram (talk) 08:39, 30 March 2026 (UTC)[reply]
"by putting several sentences into a single paragraph, the paragraph as a whole is being run, may cause time-outs, and will be cached. Instead, if, for now, you put one sentence into each fragment, caching and evaluation can be more spread out and should allow for more content. Eventually we want to fix that"[12] Gee, why would you fix the bug where putting more than one sentence into a paragraph makes it even more likely you will get a timeout error? Fram (talk) 15:55, 30 March 2026 (UTC)[reply]
Probably a few years of pointing out both the immediate and the fundamental issues. The work of developers, testers and product managers. Once the grant and endowment money stops flowing, and some quarterly or yearly goals can be checked on some paperwork, the drive to continue this will stop. At best/worst to keep whatever exists at the time running and let some volunteers play with it for a few more years before completely stopping it (see the soon to be closed down Wikinews). At least with Gather, Flow, ... we could point out that it was actively, directly negatively impacting enwiki (and other wikis): here it is only money and developer time disappearing down the drain. Fram (talk) 14:16, 31 March 2026 (UTC)[reply]
For what it's worth I would like to see them continue as the issues pointed out do not seem to be systematic (with the platform design) but an incredibly horrible implementation of it. I do agree that the state of the project seems at best alpha. Aaron Liu (talk) 14:37, 31 March 2026 (UTC)[reply]
The systemic issue is the belief that English grammar rules can be copied one-on-one to all other languages. We have e.g. functions for "use this the" and " use with a/an" and "use without either", which even within English is problematic (e.g. sometimes you need X is the capital of Y, and sometimes X is the capital of the Y, like with United Kingdom): but in other languages half the cases of a certain English function may use one construction, and the other half uses another construction, and this needs somehow to be built into the simply English function. I'm simplifying things here, but I hope yo get my drift.
Purely on a word level this whole construction works somewhat theoretically, but requires a massive amount of work which is exactly the problem for the small languages where this is supposedly built for. On a sentence/paragraph level though, I don't believe this will ever work (for simple cases for related languages, yes, but not in general). If this is pushed through regardless, we will probably end with new Scots and Greenlandic version catastrophes, but then on a larger scale. The setup and performace issues are just the icing on the cake. Fram (talk) 15:05, 31 March 2026 (UTC)[reply]
The ideal would be to store everything in context-free language, and then generate sentences in the target language by applying a set of rules. When I was a grad student in Linguistics lo these many years ago I wrote my dissertation on one aspect of Deep structure and surface structure. From my experience trying to figure out what some of the rules are in (my idiolect of) English for a limited subset of syntactical structure, it will take a very large and hard to maintain set of rules with extensive exceptions just for English. My mind boggles at the concept of doing that for all the currently spoken languages of the world. Donald Albury16:35, 31 March 2026 (UTC)[reply]
Indeed. At the moment, they produce the article [13] "New Jersey is an U.S. state." I presume the "an" comes from an "an before a-e-i-o-u" rule, which doesn't deal with the many exceptions to that rule. And this is a very simple example, in the main development language. Fram (talk) 16:45, 31 March 2026 (UTC)[reply]
The systemic issue is the belief that English grammar rules can be copied one-on-one to all other languages. You'd think that people would have learned from the failure of this approach when it was applied to Wikidata. Some editors wanted a property to express the relation "X is the mayor of Y". So they just made a property called "of". They then found out that the property was not only difficult to translate to certain languages but it also evolved into a monster that modeled many different, sometimes contradictory relations (which is kind of bad when the whole point of a database is to be machine readable) and it ultimately required a huge effort from the community to get rid of it.
Yet now, the abstract Wikipedia editors are doing the same things, defining their functions in terms of English grammar constructs and not the underlying logical relations those constructs represent. Warudo (talk) 17:21, 31 March 2026 (UTC)[reply]
All of these rules assume that an English grammar rule is also 1 rule in another language. "Article-less instantiating fragment" will sometimes need to be translated with, and sometimes without an article (even if the remainder of the grammar is the same). If this can be done with one function, then there was hardly any need to have different functions with or without article in English surely? And this is a very simple and basic example. So how is this solved? Fram (talk) 20:28, 1 April 2026 (UTC)[reply]
Article-less does not mean what you think it means here. The name is confusing and should probably be changed, but, for example "Golf is a sport" and "El golfo es un deporte" (both use article-less, despite the fact that the latter has an article) have the same meaning, but "A bird is a dinosaur" and "(The) bird is a dinosaur" have two very different meanings, even if, say, Bulgarian does not make the distinction. The distinction between article-less and article-ful is not actually articles. Feeglgeef (talk) 20:37, 1 April 2026 (UTC)[reply]
So if you start from an article which doesn't make a distinction, and go to a language that does make a distinction, you're screwed? If in your example the base language would have been Bulgarian, and you went to English, sometimes the Bulgarian function should give "the" in English, and sometimes "a", which depends on context. And all of this is still between very comparable languages basically. If the base article is for some sentence / meaning "article-less" and the target language "article-full" (or vice versa), you have a problem, no matter how you call these functions. Fram (talk) 20:48, 1 April 2026 (UTC)[reply]
I get the feeling that Feeglgeef is a little out of their depth here, but there's nobody with any decision-making authority at the WMF willing to rescue them. Phil Bridger (talk) 21:52, 1 April 2026 (UTC)[reply]
Based on your replies in this thread, it seems you are unwilling to assume good faith of either your conversational partners or the functionaries (pun intended) of the wiki in question. Why not just ignore it, if you feel it is consigned to failure? Arlo James Barnes19:09, 11 April 2026 (UTC)[reply]
You actually don't "start from an article which doesn't make a distinction", because all Abstract Wikipedia articles are supposed to be abstract. All articles are required to make the same distinction and all distinctions necessary for every single language (even if some languages ignore them) because they are all written in abstract language. There is no English or Bulgarian base, nor translating here. Feeglgeef (talk) 22:07, 1 April 2026 (UTC)[reply]
That, again, makes no sense. Because in English, "a bird" and "the bird" have two different meanings (but refer to the same Wikidata item), you need a different function than for "golf" (the sport), which has in English one meaning for that Wikidata item. But still you claim that the functions, the whole approach, are language-independent, as if these issues in English are the same across all languages for the same words. If Abstract Wikipeda were truly language-independent, you wouldn't need the article-less and article-full functions. And that's still only at the word level, and doesn't go into sentence- and paragraph structures and countless other quirks, irregularities, ... Anyway, it looks so dumb that after all this time, apparently there isn't a function yet to start sentences/articles with an article; we get things like "Bible is a religious text." for an article specifically about the Judeo-Christian Bible (not about the general word). And finding out how things actually work for Abstract articles is very opaque as well, I have no idea where the "suns" instead of "stars" comes from in [14] "Stars are sources of light. Stars contain metals. Suns shine."
Oh well, the WMF team doesn't respond here, but they seem to read it, as some of the most stupid errors get fixed after they are reported here at least. Fram (talk) 07:44, 2 April 2026 (UTC)[reply]
Wikipedia has articles about more than 20 million topics in more than 300 languages. But none of these languages alone allow access to the knowledge about these 20 million topics (...) Abstract Wikipedia does that without relying on AI. Each step of the way remains under human control, and is accessible and editable by the volunteers. I doubt volunteers will create articles about 20 million topics in "this thing" (and with 108 active editors, as there are now, it seems completely impossible to arrive at any place).
Looking at this article, it seems that a software-generated summary (yes, there is software other than AI) automatically combining some data from all language editions of Wikipedia, plus Wikidata and Commons, and using machine translation when needed (for machine translation, AI usage is usually not bad, and free and open source AI software also exists), would do a far better job.
In addition, if the point is to avoid AI, I don't understand why it says "Generated text" (who generated it?). Sorry if I am too critical (of course, I appretiate the efforts of the people who are working on the project), but I don't believe this adds anything new to the Wikimedia ecosystem: it looks as made for consumption by a machine, not by humans, so maybe it would be better if it was also mostly machine-generated.
The idea of joining the knowledge from all Wikipedia language editions (plus Wikidata and Commons, and, for some articles, also Wikisource, Wikivoyage, etc) is a very good one (and one I had thought about many times), but I think that a "global search" feature with integrated machine translation, and some way to access the combined data from multiple sources, would be the only way to achieve that. MGeog2022 (talk) 19:46, 8 June 2026 (UTC)[reply]
I doubt volunteers will create articles about 20 million topics. Yes, exactly. There's no chance we ever surpass enwiki or other major language Wikipedias, as abstract articles take much more time to write than normal articles, and there's much less chance for onboarding new contributors.
In addition, if the point is to avoid AI, I don't understand why it says "Generated text" (who generated it?). Sorry if I am too critical (of course, I appretiate the efforts of the people who are working on the project), but I don't believe this adds anything new to the Wikimedia ecosystem: it looks as made for consumption by a machine, not by humans, so maybe it would be better if it was also mostly machine-generated. It was made for the consumption of humans who speak only small-to-mid sized languages.
Machine translation might be worth consideration for a future project, it's been used, for example, on Caesar DePaço to circumvent a court order, but writing content in an abstract (instead of concrete) language means we don't have to deal with information loss (see the whole "Google translate 100 times" thing).
If you click through the endless errors and finally try to see a translated article, you get monstrosities like "Ein Äpfel ist eine Frucht."[15] or even worse " Bisexuelles lieben Fraus. Bisexuelles lieben Manns."[16]
People from the WMF, could you please clarify: before releasing this as a supposedly beta product, which tests did you run? Which articles have you created, with which functions, and tested for which languages? Fram (talk) 13:16, 8 April 2026 (UTC)[reply]
@Sannita (WMF): as someone who seems to be closely involved, but of course feel free to put this through to whoever is better placed to answer this. I also notice that most activity on Abstract seems to come from an editor who was indef blocked on enwiki for CIR / timewasting, and is now running some AI-generated tool to create non-working pages (like this 193K monstrosity[17]-) and to change working pages (no matter how bad they were) into non-working ones (e.g changing this into this ("Wikifunctions returned a failed response: Error in evaluation"), across a lot of pages. While I think the project should be abandoned as a waste of time and money, it shouldn't be done by mass-vandalizing the work of the editors there. Fram (talk) 16:30, 10 April 2026 (UTC)[reply]
Thanks for your comment. A discussion has been opened, and the bot runner decided to pause their edits. We'll sort this out the wiki way. Sannita (WMF) (talk) 21:59, 10 April 2026 (UTC)[reply]
@Fram We did some limited tests in a controlled environment, but doing so can only help you so much in identifying potential problems. We are learning a lot by releasing the beta project (because this is still a beta), and we'll improve from there. Sannita (WMF) (talk) 10:28, 11 April 2026 (UTC)[reply]
Thanks, but no, this is not a beta, this is barely an alpha, the thing is unworkable for all but the simplest sentences, terribly slow, had the most basic errors when released. Using volunteers as cheap/sheep testers on a $6 million+ project which hasn't been thought through, hasn't been tested, and has in its utopic, unrealistic ideals been overtaken by reality anyway, is old school WMF which I hoped had been left behind after previous such failures.
After the omnipresent "Reached max retries. Try again later.", you get a plethora of errors. A 2-sentence "article" gives "Wikifunctions returned a failed response: Reached time limit in orchestrator", basic English words give "Wikifunctions returned a failed response: No matching lexeme for item in language" (but this will work for languages with barely any editors somehow), other articles give "Wikifunctions returned a failed response: Error in evaluation", "Wikifunctions returned a failed response: Could not acquire WASI runner within time limit", "Wikifunctions returned a failed response: Reached rate limit in orchestrator"
And the things that do "work" give results like "Australian continent is a continent in the Earth. Australian continent contains Australia." (German translation of that last line: "Australien contains Australien.") Or the extremely basic issue that when you translate an article, you would expect the title of the article to be the first thing that gets translated, even in alpha-stage. No such luck. This thing is supposed to be used to create articles and translate them into manu languages, but not a single decent example has been produced so far. A "beta" product which simply can not produce an acceptable end product just isn't tested to even the most basic standards and should never have been released. And a project where the actual requirements don't seem to have been thought through, and where the results (if the wanted end result was ever reached) would probably make the Greenlandic and Scots disasters look like minor blips, should have been stopped much, much earlier, before so much money and time was wasted. Fram (talk) 17:44, 11 April 2026 (UTC)[reply]
I just don't get how they are going along and spending money on something so ill-thought out and poorly executed. Let's ignore for now the basic questions of "is this feasible, which competencies or expertises do we need to think this through, what is the best approach", the slight issue that reality (LLMs) has somewhat changed the whole environment this needs to be thought about, and the basic recurring problem that a completely untested, very buggy environment has been released as a beta for everyone to play with without any guidance. But why are they (WMF and editors) now going further with this in the most inefficient way possible? Everyone creates whatever they like, 99% of what is being made is absolute rubbish that serves no purpose at all. People are creating one- or two-sentence stubs for all countries which all have the same issues.
A logical, productive way of dealing with this project (apart from the most logical one of pulling the plug) would be to start with one article, take e.g. the lead from enwiki (or dewiki or whatever), and build the necessary structure (Abstract + Functions + Wikidata) to create this and translate this in 5 wildly different languages. Step by step, sentence by sentence, until you have a basic set of functions for this kind of article, and have an idea if it will work.
Second steps might then be either testing the same for a related article (say, test 1 was a country, take another country and see if the functions are all transferable or if there are things you missed), or doing the same "build from scratch" for a different topic (say, a biography).
That way, you build a structure, a set of reusable and needed blocks you can refine later on, and at the same time learn a lot of "don't do this" pitfalls and issues. And after you have done this for the 5 or 10 most common types of articles, you actually have a tested, usable, beta environment (or you have realised it won't work at all, or it will work in theory but the work to get things right for a more obscure language isn't worth the hassle).
Instead, we get articles[18] directly using the function "English plural"[19], really useful for an abstract, language-independent project. Or more commonly "articles" consisting of random repetitive sentences like "Reproduction is a biological process. Reproduction is a type of process. A reproduction is a biological process. A reproduction is a creation. Animal reproduction is the part of of reproduction. Plant reproduction is the part of of reproduction. Human reproduction is the part of of reproduction."[20]
Oh well, at least I learned that "An information is a knowledge."[21] or "biseksualiteit ∈ {seksuele oriëntatie}" in Dutch or still "Bisexuelles lieben Fraus. Bisexuelles lieben Manns." in German [22]. Note how here and on all other pages, the actual title doesn't get translated? This is not something the article creators can help, this is something which should have been included by the WMF as a basic element (assuming they realised what Abstract Wikipedia was intended for) but is missing.
Anyone minimally familiar with the history of Linguistics or Computer Science as disciplines should be aware that the core premise of the project was investigated and discarded decades ago (and anyone who has studied at least one language that isn't super closely related to their native language should be able to quickly come to similar conclusions). There is no universal deep structure across languages. Why was this ever greenlit? This is about as useful as funding new studies in Lysenkoismsigned, Rosguilltalk05:51, 3 June 2026 (UTC)[reply]
The WMF has launched a dashboard to follow the progress of Abstract Wikipedia [23]. It has e.g. a list of articles which work in any language! Well, most of them have no actual text translated through functions, just Wikidata items without any sentence-building, so yes, these work, they just aren't articles... The ones that do try to have actual sentences and supposedly work are e.g. Brussels[24], full text "Brussels is the capital of Belgium." This gets translated as "Brussel is the hoofdstad of België." in Dutch, which isn't correct Dutch. "Bruxelles is the capitale of Belgique." in French is equally wrong, as is the German "Brüssel is the Hauptstadt of Belgien." It doesn't work at all in e.g. Romanian, Moldavian. Anyway, I guess the function "defining role sentence in English as string" should perhaps not be used in Abstract Wikipedia, and items which use it should not be said to be working in any language, as the naturally don't. Something like "list of cities in Belgium"[25] gives a result in English, and keeps running endlessly when I try it in Dutch or French. So not really "working". It looks as if not a single actual article can be said to be working in all or even most languages, making the dashboard rather meaningless. Fram (talk) 11:49, 8 May 2026 (UTC)[reply]
From reading the Abstract Wiki, it also seems like there is no way to utilize the past tense, which um… seems a bit critical in an encyclopedia. ExtantRotations (talk) 18:34, 8 May 2026 (UTC)[reply]
On the plus side, they have introduced endless repetition, which makes for much better reading. At the moment, the article for New Year's Day reads in full: "New Year's Day is a public holiday. New Year's Day is part of the public holidays in Australia, public holidays in Australia, public holidays in Australia, public holidays in Australia, public holidays in Australia, public holidays in Australia, public holidays in Australia, public holidays in Australia, public holidays in Australia, public holidays in Australia, and public holidays in Australia." Fram (talk) 12:10, 22 May 2026 (UTC)[reply]
Somehow, they have made this dashboard worse. It now has a section for "Articles ready to publish in English". What this means is "Wikidata Qnumbers we have created in Abstract but don't exist in enwiki", not that these pages are in any way either wanted in enwiki, nor that they are actually "articles" which are "ready". This includes things like [26] "Evaluating the biological validity of European river typology systems with least disturbed benthic macroinvertebrate communities", ne of the thousands (millions) of scholarly articles with a Wikidata entry, but also [27] Blake Lemoine, apparently a software engineer, where the complete "article" exists of three identical links to research.google, but the link gives a "page not found". Ih yes, this is so ready to be published to enwiki. Other "articles" we are apparently missing is [28]"interaction", full text "An interaction is an effect.An interaction is an effect.An interaction is an effect." Oh look, a Picasso painting we are missing[29]. Actually, no, we have it at Three Musicians (Picasso)
I don't even want to try to guess what the next section on the dashboard, "Quick wins in English", is supposed to be about. "Add a label to California"? "Pending fixes in English", explanation "All items where English is the blocker", e.g. "photon" needs a label.
Mind you, that's not even the most bizarre or optimistic thing on that dashboard. At the very bottom, there is a section for "Almost there — English is the only thing missing
Already usable in the most other languages. One English edit gives each near-universal coverage." Well, perhaps not. [30]Giza is not one English edit away from anything, Giza (or any Abstract "article") shouldn't use the function "defining role sentence in English as string" because "in English" is not really what Abstract is about.
While for English these things won't be published here and we see the problems from miles away, for small languages (the intended target) this is another dramatic Scots Wikipedia scenario waiting to happen, but this time pushed directly by the WMF.
According to the latest newsletter about Abstract, all that is left is improving Abstract and integrating it into language Wikipedias (the horror), but as of now, "Abstract Wikipedia can compose articles from functions on Wikifunctions". Bizarrely, despite this claim, not one actual article has been composed, only a few collections of loose, short, often ungrammatical sentences, which in a few cases can even be translated in even worse sentences in very few languages. But oh well, what can one expect from a newsletter which claims "Abstract article pages now show the label of the Wikidata item alongside the QID in the page title. " when in reality, for a week or so now, what is actually shown next to the Wikidata label is the letters "ltr" in grey. When you click on them, you get the text "copied!", and you have literally copied the text "ltr". Brilliant! Even things that barely worked before have since become worse, the page "Paul Cézanne"[31] now shows "<a href="https://abstract.wikipedia.org/wiki/Q17277950">The Card Players</a>" instead of the links to Wikidata it had. Also errors like "Reached max retries. Try again later. Retry" have reappeared.
Let's be clear: "Franz Schubert is a composer in Austria. Franz Schubert is a pianist. Franz Schubert is a teacher." is not an article. "Wikifunctions returned a failed response: Invalid key" is not an article. "Organism is a part of of population. Organism is a part of of group of living things." is not an article. "Roman Empire is an empire." is not an article. "2 is a small number." is not an article. "List of French artists is a Wikimedia list article." is not an article (why would anyone think that this is the kind of text wanted in any language wikipedia?) "August is a calendar month. August is part of the Swedish calendar, Swedish calendar, and Swedish calendar." is more than an article, it's a mantra. "A Citrus × limon is a useful plant." is the full article about the Lemon.
And the translations... "Organisme is an onderdeel van of populatie." is not a Dutch sentence. "2 is een ." is not a Dutch sentence. "Franz Schubert est un enseignant ou enseignante." is what you get when you start from English as the norm for all languages. "August is a calendar month. August is part of the schwedischer Kalender, schwedischer Kalender, and schwedischer Kalender." is supposedly a German article. "{citroen} ⊆ {nuttige plant}" is supposedly a Dutch article.
I clicked on random article, the first one I saw was Homer:
Homer is a poet. Homer is an author. Homer is a writer. Homer is a human whose existence is disputed. Homer is a conceptual character. Homer is the part of of Greek mythology.
What I don't get is, they don't have support for past tense yet. Whatever, they'll add it eventually, I hope. So, why are they making abstract articles that need it? Warudo (talk) 20:35, 1 June 2026 (UTC)[reply]
Partially reminds me of caveman speak: If horse go walk, cart also go walk. You see, force from horse leg on ground push on rock, then normal force push horse leg in opposite direction, cause horse to move small amount, for each time horse leg push on rock. Rope tied to horse neck move with horse, and is connected to cart with wheels. Cart wheel spins if moved in same direction as wheel is facing, which glide across rock, but still exert normal force due to gravity. Therefore, if horse go walk, cart also go walk.
For what it's worth, the Homer abstract article was created by the editor who did this. As harsh as it may be to say, this is the level of quality we are used to seeing from this editor, both here and in the abstract Wikipedia. Warudo (talk) 19:04, 2 June 2026 (UTC)[reply]
They were blocked for CIR/wasting community time and (how do I put this politely) I couldn't agree more with that description. Anyhow, most of the slop problems Fram is pointing to were caused by them, so I don't see how that justifies shutting down the whole project. Feeglgeef (talk) 03:51, 3 June 2026 (UTC)[reply]
Not true though. Going over the issues in my above posts, the "ltr" instead of the Qnumber is obviously a WMF technical mistake (and a failure of whoever wrote the newsletter). The "articles" I randomly picked: Paul Cézanne "2", August, the Picasso painting, New Year's Day, Giza, ... all not edited by that blocked editor. And there are plenty of articles I could have picked, I don't think "Australian continent is a continent in the Earth.Australian continent is a component of Earth's surface.Australian continent contains Australia." or its Spanish translation "(en) Australian continent is a continent in the Earth.Australia is a componente of superficie terrestre." (note that the middle sentence simply disappeared, perhaps a blessing) is an indication that the issues are due to one bad editor and not to a terrible project. Bisexuality, in German, is first an error and then "Bisexuelles lieben Fraus. Bisexuelles lieben Manns." Not touched by the blocked editor. this is not created by the blocked editor. The "Italian" article "Holy Week is a liturgical season. Settimana Santa is part of the anno liturgico." is not edited by them. Obviously it would be best if you mass deleted the embarrassing creations by that editor (full article: "A current is a current.A current is an electricity."), but it wouldn't solve any of the fundamental issues. Fram (talk) 07:37, 3 June 2026 (UTC)[reply]
And it's not like this is the only editor who writes articles about dead people. "William Shakespeare is a playwright from Kingdom of England." Has not been touched by that this editor either. Warudo (talk) 11:22, 3 June 2026 (UTC)[reply]
@Fram Thank you for the continued updates, which are rather amusing, if in a morbid way... Would you be interested in starting an RfC on Meta on putting an end to this madness? If anyone has any better ideas, I'd be happy to hear them. Toadspike[Talk]11:49, 2 June 2026 (UTC)[reply]
I normally don't edit Meta, and I fear that many people who hang out around there are more inclined to let such projects continue no matter what. Fram (talk) 12:17, 2 June 2026 (UTC)[reply]
For what it's worth I agree with Toadspike and would like to see an RfC on meta. It might be true that this bias exists there, but Abstract is such a one-of-a-kind, unmitigated disaster that I believe it would not apply in this case. Choucas 🐦⬛12:29, 2 June 2026 (UTC)[reply]
Not the right process; you're thinking of m:Proposals for closing projects.For what it's worth I think there's something worthwhile to come out of this; to resolve the fundamental problem the articles look to be moving towards using functions that express specific meanings rather than grammar. I definitely agree that the WMF is hopelessly overstating its progress, as Fram has meticulously documented.I would support some kind of discussion to make WMF be accurate when assessing the level of progress on the project. An RfC might be able to do that but I'm not sure if purely WMF communication matters can be subject to an RfC. In solidarity, Aaron Liu (talk) 12:40, 2 June 2026 (UTC)[reply]
I have some strong opinions about the content created by a couple contributors. This one was the original unattributed enwiki cop[y] Fram had originally mentioned. In addition to Csisc, there's also User:Immanuelle, who was blocked here for WP:CIR. Generally I think they share a need to do something now, though at this stage these contributions are unhelpful and frankly actively disruptive. Feeglgeef (talk) 23:45, 8 June 2026 (UTC)[reply]
Coming from huwiki, I've been trying to understand the scope and progress of Abstract Wikipedia, and I want to say first that Frem's updates in this thread have been more informative than the project's own pages on Meta. Thank you for that.
That itself points to my concern. This is an enormous and complex undertaking with many moving parts. A project of this size depends on community buy-in, so it needs to communicate its scope, progress, and decision-making clearly and regularly. Right now I don't think it does. If you read meta:Abstract Wikipedia and its subpages, much of it appears out of date, and the current scope and state are not stated anywhere in a way an ordinary editor can quickly grasp. You could say they're rather abstract. Additionally the original project proposal did not lay out the alternative solutions considered for the individual problem statements, nor any cost-benefit analyses. From what I've been able to piece together there were more readily available approaches, that could have delivered usable results by now (not including the recent advancements of LLMs). Checking the pages and the Wikimedia Board minutes it is also unclear what changes of direction have been discussed or approved since the initial 2020 resolution, or under what criteria the project would be judged unfeasible to continue. This is important, cause this is a highly ambitious project that, without clear guardrails, could go on indefinitely.
In this thread serious concerns were raised, and it's shocking that so far no one from the project has responded to these. I can only add to those, I agree that the project cannot really be described as an "early beta": after reading through pages of Abstract Wikipedia documentation, I still have no idea how I, as an editor, could begin supporting the project and test Hungarian implementations.
Without a clear understanding of what is happening and why, community discussion is effectively blocked. So that this does not sink without a reply, I'd like to ask the project leads to respond:
Current state and scope: What is the precise, current scope and phase of the project? What is the progress on the tasks, which of these are done from the initial 30 months plan? What is still prototype or research?
Supporting a language: Is there a documented, step-by-step path for a language community to add or improve support for its language? Concretely, what could a non-English editor do today to help with testing?
Documentation: Given that the Meta pages appear outdated, is there another current, plain-language page stating scope, status, and roadmap? Could Meta be updated?
Regular review: Since the Board's approval (the resolution of 22 May 2020), what changes of direction have been discussed or approved, and by whom? Are there defined milestones or off-ramps, and what criteria would lead the Foundation to conclude the project is not feasible to continue? How does the Board assure itself that the project remains on track and worth its cost, and what would have to be true for it to decide otherwise?
Resources and partial wins: Approximately what has been spent on the project to date, and how is the continued resource commitment justified against other Foundation and community priorities? What concrete, independently useful results has the project delivered so far (with examples)?
And @NGunasena (WMF): if you could comment on, or connect us with someone who can say more about the Board's thinking and its insight into the project's actual progress (questions 4–5), I'd be grateful.
To be clear, I'm raising this in good faith, I want the project to succeed, and clear communication from the stakeholders is a precondition for the community to participate. Boro (talk) 11:32, 7 June 2026 (UTC)[reply]
Not WMF, but for #2, you could look at which functions are commonly used on abstractwiki (e.g. f:Z28016) and learn to contribute to them. I am aware this is a tall ask and if you want to, I could help you. In solidarity, Aaron Liu (talk) 14:17, 7 June 2026 (UTC)[reply]
Thank you for the offer, but it's a taller ask than it looks. Hungarian is an agglutinative language that uses vowel harmony. The simple "defining" structure you linked (X is the Y) does exist in Hungarian, but the neutral, encyclopedic version doesn't use a definite article at all, while the variant that does carries extra contextual meaning. So it's unclear which needs to be used here: match the intention or the definition? The correct rendering needs helper functions to get the morphemes right, and research groups in MT and linguistics are still working on tools to do this reliably for all possible cases. I'm not an expert in this field, so I don't feel comfortable writing semi-correct functions to make this work. I've already seen students pick up incorrect uses from AI output, so getting this wrong at scale could affect the language itself.
From the Meta page this seems to fall under "Task P2.9: Renderer for a language from another family". But I can't find any notes on whether that was evaluated, and with what results and to what extent, before moving forward with the project. As I understand it, the majority of its benefit is supposed to come precisely from supporting non-Indo-European languages, so checking more than one language family seems essential. Boro (talk) 20:10, 7 June 2026 (UTC)[reply]
My understanding of Hungarian isn't the best, but from what I remember, isn't just a rearranging of the words? So it'd be literally "Budapest Hungary's capital" (Budapest Magyarország fővárosa). Feeglgeef (talk) 21:09, 7 June 2026 (UTC)[reply]
Not quite, főváros is the base word, +a/+e is a suffix, that can change form based on the word: capital of → fővárosa (főváros), sibling of → testvére (testvér), older brother of → bátyja (báty), younger brother of → öccse (öcs), mother of → anyja (anya). Also all these sentences are valid, but subtly different in their meaning: (1) Budapest Magyarország fővárosa, (2) Magyarország fővárosa Budapest (3) Budapest Magyarországnak a fővárosa (with definitive article). Boro (talk) 22:03, 7 June 2026 (UTC)[reply]
If it doesn't use a definite article, then the output should not have a definite article. I think how this is supposed to work is that each function outputs the construct with the intended meaning no matter what the grammar is. In solidarity, Aaron Liu (talk) 18:15, 8 June 2026 (UTC)[reply]
I'm not WMF staff, but as a volunteer I'm one of the most active users on abstractwiki and a functioneer on Wikifunctions.
Currently, on the Abstract Wikipedia wiki, which is in beta, users are able to create basic stub articles that render in English and often other languages. Wikifunctions is no longer in beta and is generally usable, P1.1 through P1.13 and P1.18 though P1.19 are things I'm able to do on the wiki right now. The plan listed for Abstract Wikipedia is very different from what it seems is actually being implemented, so I can't comment on progress there.
Somewhat. I do believe we need much better documentation, but f:Help:Contents has a few pages that are good starting points. I have helped and am willing to help others with the process of creating localized versions of functions.
I don't believe so, unfortunately.
I'm not sure, as again I'm not WMF staff.
On results, we've seen some use by smaller-language Wikipedias and small-and-mid-sized Wiktionaries of using Wikifunctions like templates. As mentioned before, Wikifunctions is also useful on its own and has a large catalogue of functions.
"users are able to create basic stub articles" for extremely low bars of "basic" and "stub" (and "create" for that matter, many editors don't seem capable of this, the learning curve is quite steep).
" that render in English and often other languages. " Very rarely, actually. But feel free to give us a few examples of what you consider basic Abstract stubs and which languages they render in. Take for example Australia[32], which probably comes the closest to an actual article. When I try to translate it to German (and after a very long wait), I get 5 errors, and a mixture of German sentences, English ones, and mixed ones like "The Einwohnerzahl of Australien was 27614411 in 2025." Same when I try it in French, 6 errors (4 different ones), mixed languages. But at least in English you get an article that reads as if some dumb robot compiled it, with the first five (obviously very short) sentences starting "Australia is", and ending with varied but hradly correct ones like "Melbourne is the most population urban area in Australia." Or let's look at the most recent one you created, Labrador Retriever[33]. Full text: "A Labrador retriever is a dog." Capitalization is hard apparently. Dutch translation: "{labrador retriever} ⊆ {hond}". Another one you created, 6-7[34], doesn't even work in English ("Wikifunctions returned a failed response: Reached recursion limit in orchestrator") Mount Everest[35]: "Mount Everest is a mountain of China–Nepal border.". Frenglish: "Everest is a montagne of frontière entre la Chine et le Népal." Swedish: "Mount Everest is a berg of ." Spanish: "Everest is a montaña of Frontera entre China y Nepal."
So no, both creating decent stubs and rendering them in other languages doesn't really work, even when one of the most experienced Abstract editors creates them. Fram (talk) 09:32, 8 June 2026 (UTC)[reply]
It's crazy that the resources of the WMF are going into this hopeless and pointless black hole instead of fixing actual problems that the community have been highlighting for years. It would be fun if it wasn't tragic Ita140188 (talk) 09:42, 8 June 2026 (UTC)[reply]
Yes. It's time to pause development on Abstract (to avoid the embarrassment of cancellation) and redeploy the engineers to process community tech requests. That's not as good a result as retaining the staff with relevant experience, but it's a start. Certes (talk) 13:01, 8 June 2026 (UTC)[reply]
You're a one trick pony, Fram. We get it. Some of the articles on Abstract Wikipedia are not great. But if one billion useless articles were added to the English Wikipedia, I wouldn't call for it to be shut down. The fact that you can find articles that look bad doesn't mean the project needs to be shut down. As for your examples, they do not disprove (in fact, they prove) my intentionally conservative statement, users are able to create basic stub articles that render in English and often other languages (emphasis mine). Feeglgeef (talk) 15:07, 8 June 2026 (UTC)[reply]
If you would look at article after article on enwiki that looked like this, you would be right to call for it to be shut down though. And it's not a case of "I can find articles", I picked recent ones created (or in the case of Australia edited) by you specifically, as last time you complained that some examples came from an editor who is blocked on enwiki (but not on Abstract).
But I see that you consider these rather dreadful examples as "basic stubs" and the abysmal translations as evidence that they "often render in other languages" and provide no better examples, so I presume that this "one trick pony" (WP:NPA) somehow found representative examples and you won't point to some better Abstract articles instead. Fram (talk) 15:42, 8 June 2026 (UTC)[reply]
Firstly, I'm sorry for the "one trick pony" comment. I had meant it as a comment on the lack of variation in your evidence, and should have framed it as such.
Here's a decent article, abstract:Q1186. It's a basic stub (the information would be useful) and renders correctly in both English and Bangla, thus meeting all of the criteria I gave. Feeglgeef (talk) 16:25, 8 June 2026 (UTC)[reply]
Thanks. Why does it give the population in 1961? It's less than half of the current population, so hardly good information to include here. Fram (talk) 16:31, 8 June 2026 (UTC)[reply]
That's especially weird considering that not only does Wikidata have a figure from 2017 but it is also marked with preferred rank. Warudo (talk) 16:33, 8 June 2026 (UTC)[reply]
I've fixed it such that it will use preferred statements first. Unfortunately, despite the fact that I've requested it multiple times, there's no way to fully clear the cache, meaning the Kerala article is stuck with the old value. But it will work for new instances, like Illinois, which I just created. Feeglgeef (talk) 16:48, 8 June 2026 (UTC)[reply]
Seeing the working examples led me to think that, at the reader level, this is something that ArticlePlaceholder was already capable of: surfacing Wikidata facts about a topic with no local article. For topics that do have an article, inline Wikidata item calls can keep those facts current on its own (though the editing experience could be smoother). All in all, I don't see why sentence generation is worth all this effort, when easier paths were open to achieve essentially the same benefit. Boro (talk) 22:37, 8 June 2026 (UTC)[reply]
There're certain things only articles can do, like document the history of how something happened. I don't really see any advantages of ArticlePlaceholder over just referencing Wikidata when you do need to switch wikis. In solidarity, Aaron Liu (talk) 01:51, 9 June 2026 (UTC)[reply]
You described yourself as one of the most active users on abstractwiki and a functioneer on Wikifunctions, so I do not find it unreasonable for Fram to look at your articles and point at the glaring and massive issues there. Maybe being able to find articles that look bad means little, but are we even able to find articles that look good though? Or is the bar so low that "good" by Abstract Wiki standards just means "not broken in a thousand obvious ways"? You have to admit that from an exterior point of view, things really look dire, and the fact that there is not even more than one outdated meta page in terms of documentation does not bode well in terms of perspective. And that is even without getting into the immense issue that the basic premise of the project is based on disproved linguistics, as has been pointed out elsewhere. Choucas 🐦⬛16:04, 8 June 2026 (UTC)[reply]
I do agree that things might look dire from an external point of view, but, as I've been trying to show, Abstract Wikipedia is a viable project. It just needs more time, and, ideally, better technical support. I do agree with those above that Dr. Vrandečić is overstating our progress. Feeglgeef (talk) 16:56, 8 June 2026 (UTC)[reply]
Personally I think it's worthwhile to explore this avenue with the aim of making imperfect 'universal stubs' that serve as placeholders while small wikis are developing, but the design of abstract wiki is frankly amateurish, the team appear to be incredibly out of their depth. Why has wikifunctions been primarily based on English/Indo-European languages when there'll be practically no new wikis in languages closely related to English? Do we know if any professional linguists were consulted (the mastermind behind it sure as hell isn't one)? Why was this not done by experts? How on earth has $6 million been spent? On what? How has this gotten through so many checks and meetings? Kowal2701 (talk, contribs) 21:08, 8 June 2026 (UTC)[reply]
I'm open to the possibility that there could be a way to write articles using a lingua franca that is amenable to translation to many different languages. But I have strong doubts about Abstract Wikipedia being viable as a crowdsourced project using Wikifunctions as the lingua franca. isaacl (talk) 00:17, 9 June 2026 (UTC)[reply]
The lack of documentation means people have to (re)invent the wheel when they want to write things and, at least some of them are not good at it. Consider abstract:Q246463 where the editor who made it wanted to use a pronoun instead of repeating the name of the article subject, so they passed it (Q6091500) to the function that generates the second sentence. There are two problems with this. One, it leads to the awkward phrase "of it" instead of "its". But more importantly, it doesn't work for languages other that English because grammatical gender is different between languages. What is the equivalent of "it" in French, which doesn't have neuter? Or worse, what would happen in a language where neuter exists, but the word for "shrine" is not neuter? Warudo (talk) 17:08, 8 June 2026 (UTC)[reply]
This is a problem as well. Wikifunctions (and, by extention, Abstract Wikipedia) has a very steep learning curve, so we have a few users who understand it deeply and everyone else, who barely understand it at all, which makes it very hard to find contributors, whereas the English Wikipedia has a remarkably flat learning curve and a very high ability to specialize. Feeglgeef (talk) 17:19, 8 June 2026 (UTC)[reply]
This is the part that makes me ambivalent about the WMF's involvement. I think it's worthwhile doing research in understanding ways to capture information based on commonalities in language. It's not clear to me, though, that crowdsourcing is an effective way to generate an encyclopedia of articles written in a lingua franca with a programming-like syntax. Without the crowdsourcing aspect, I don't think the project is a good fit for the strengths of the Wikimedia Foundation. isaacl (talk) 17:42, 8 June 2026 (UTC)[reply]
I'm reading through this again and that "Dutch translation" in particular scares me because of how wrong it is on multiple levels. First of all, is returning math notation really helpful? Wouldn't an error be better here so there's no illusion of support? But more importantly, no matter how smart the person who came up with this workaround felt, it is still wrong! "{labrador retriever} ⊆ {hond}" means that the singleton set {labrador retriever} is a subset of the singleton set {hond} which is obviously false. It should be "labrador retriever ⊆ hond" or, more explicitly, "{x|x ∈ labrador retriever} ⊆ {x|x ∈ hond}". If after all this time, people who work on this stuff cannot handle what are essentially instance of (P31) (∈) and subclass of (P279) (⊆) correctly how can we expect this project to ever succeed? Warudo (talk) 10:35, 13 June 2026 (UTC)[reply]
Hello @Boro, and thank you for your good faith inquiry.
I agree with you that the project pages for Abstract Wikipedia could use an update. We will spend some time in the near future to do so, thank you for pointing that out. We have mostly focused our progress report through our weekly newsletter. We have been keeping the newsletter up for more than 250 editions now, and I am sure you would be able to find most of the answers to your questions in the updates. But I admit that sorting through so many newsletters to find the one answer you are looking for is not the best solution.
Here is the current status of the project, in a very short overview:
Wikifunctions launched almost three years ago (July 28, 2023). The project has a growing community, and there are now more than 4000 functions created. The speed of function creation is increasing.
Wikifunctions function calls can be embedded in almost all Wiktionaries, as well as Dagbani, Bengali, Igbo, Hausa and Malayalam Wikipedia. This allows editors to call a centralized store of functions to produce outputs, instead of having to create and maintain them locally in each wiki. An example is here, where the German declension table comes from Wikifunctions. If more Wikipedia language editions would like to have this functionality, please let us know. We are currently prioritizing article integrations and are planning to revisit embedded Wikifunctions at a later date, but let us know your feedback.
From Wikifunctions, we are able to access data from Wikidata. This includes most statements about items, and most lexicographic data. This allows the community to create functions using the rich lexicographic data in Wikidata, or using population numbers from Wikidata directly, etc.
Abstract Wikipedia is in early beta, and we made it available live to enable learning, testing, and iteration. We want to co-create this project with our communities, including on how NLG functions evolve. Here’s an example that was just working as I write this, a short article about the year 1234 in English and in German. We also want to see how the system performs with real traffic to make it more stable and responsive. We are listening to feedback and releasing improvements every week.
Now you might object that the Hungarian article about 1234 is better than our abstract article (for example, it doesn’t call 1234 a Gregorian year even though it’s a Julian one), and you’d be right. Our scope, however, isn't to replace existing articles, but to fill in gaps where they currently exist, particularly in language editions with a smaller community.
In a few months, this will be possible when language communities will be able to integrate Abstract Wikipedia articles into their Wikipedias. We hope by then such errors will be fixed, and more content will be possible. This will be the last major piece of the puzzle, and once in place, we will work closely with communities to discover missing capabilities, bottlenecks, and potential for improving the contributor interfaces. We will create feedback loops to ensure the product evolves to support community needs. We will assess how effective Abstract Wikipedia is at addressing multilingual content parity, and make further decisions on next steps from there.
We are fully aware that this is only addressing a small part of the questions here. Our suggestion is that we will summarize the constructive criticism and concerns to kick off a discussion page on Meta, and answer the questions there. This discussion here has surfaced great suggestions and criticisms which we want to address. We want to take this input seriously, but it will take some time, and we can then continue this conversation on Meta, where it is visible for all communities.
"making it much easier to start a local version of the article, instead of writing it from scratch or translating from another language. " That I doubt. It's not as if for these smaller languages, things will magically work if you have the right Abstract article and the right functions. Even completely ignoring the language structure issues (a pretty big issue in itself), people will need to navigate Abstract and Wikidata, add labels and lexemes, and who knows what else. Even a simple "article" like the rather useless 1234 linked here, gives 6 errors when translating to smaller languages like Nahuatl (and 2 errors for French), with the errors not really helping to knw what needs to be done.
Why you believe this can be integrated in any Wikipedia in a few months time is beyond me, sounds more like a KPI that needs to be met than an actual good plan. Publishing one or two symbolic "articles" from Abstract to some language and then celebrate this as a major milestone and a proof that it works, when in reality next to nothing has been achieved and questions about the fundamentals are completely ignored, does not give any confidence.
As for the Wikifunction, it looks to be wrong or at the very least oversimplified. A rather different table is given e.g. here or here or here.
Having basic information wrong - as in the 1234 example or several other examples listed here - is not filling in a gap on smaller projects it's spreading misinformation. I would not expect information from abstract to be 100% correct. Or even as correct as the average enwiki article or the average article on some other large project. But I would expect it to have basic information correct. And I would expect it to be reasonably intuitive for someone who spots an error to be able to correct it, which if I understand things correctly, would mean some kind of easy connection back to Wikidata. Best, Barkeep49 (talk) 15:59, 9 June 2026 (UTC)[reply]
How is that a easier way to populate articles for smaller languages compared to automatic translation from a bigger language and basic copyediting? It seems to me this would require a lot more work for a much worse result compared to the basic way that worked until now Ita140188 (talk) 07:59, 10 June 2026 (UTC)[reply]
They seem to have the wrong idea that when you have an abstract article and working functions, all you have to do is change the language and you have an article. Without even going into the grammar issues with many languages, it ignores the work needed in Wikidata to get everything you need for a certain language. Hence, in the 1234 example given, the 2 errors for France and the 6 for Nahuatl. It also ignores the complete instability of Abstract: if I go to 1234 (in English) now, instead of the text I saw yesterday I get "Wikifunctions returned a failed response: Invalid orchestrator result". Finally, it ignores that probably you will find more people interested in actually writing articles in their language, than people navigating 3 or 4 environments (Abstract, Functions, Wikidata, and Wikipedia) to copy someone else's creations, certainly when they can (for many, not all languages) much more easily copy creations to their language through online translation tools.
The only use case where I see this making sense is from a push-concept, where people from Abstract "push" translations (in languages they don't understand) to all Wikipedia languages where a certain topic doesn't exist yet. This obviously is a disaster waiting to happen, and should be opposed by all possible means. Fram (talk) 08:16, 10 June 2026 (UTC)[reply]
Even for the example at wikt:hr:Wort this is painfully useless at scale, since the construction and further editing of {{#function:Z29055|L2206|}} requires knowing that there is a Wikifunctions page called f:Z29055 and a Wikidata page called d:Lexeme:L2206. It is abhorrent that instead of doing something along the lines of mw:Multilingual Templates and Modules, millions of dollars from the Endowment (source here) are being spent on this website that’s extremely hard and annoying to use, and is impossible to edit without JavaScript. stjn15:05, 10 June 2026 (UTC)[reply]
@Sannita (WMF): can you give us clarity if there are future funds being planned to be sunk/invested into this project? Like many people above, I'm sceptical this can ever work at scale, and feel that small communities deserve better. In solidarity, —Femke (talk) 🐦 16:08, 10 June 2026 (UTC)[reply]
To be fair, there's no good reason to call {{#function:Z29055|L2206|}} directly from mainspace. Functions that editors will use often should be wrapped in templates precisely so editors don't have to memorize the ZID. As for the LID, you can open Wikidata on another tab and look up the lexeme to get its ID. It doesn't sound that bad. Warudo (talk) 16:14, 10 June 2026 (UTC)[reply]
I think if someone provides that as an example of the potential usefulness of Wikifunctions, I am going to judge it on its merits. And the merits here are entirely unconvincing (what you are describing in itself is extremely tedious), especially compared to investing in the less pie in the sky ideas that you can use the regular editing interface for (Wikifunctions and Abstract Wikipedia in themselves are a good example of how badly the website would be run if WMF developers ever forgot that to truly serve knowledge to everyone, we need to be HTML-first). stjn18:11, 10 June 2026 (UTC)[reply]
Clearly not. I can't believe that there is nobody that can say anything useful here, so can only conclude that the non-reply is because they have been ordered not to communicate with us plebs. What do we know anyway? Phil Bridger (talk) 16:42, 18 June 2026 (UTC)[reply]
@Phil Bridger, as of this message in this thread you have made nine comments. With them, you have managed to: - Imply every WMF employee answering about Abstract Wikipedia would be engaging in bad faith because their salary depends upon not fixing issues; - Suggest that the WMF workforce is so lazy that the place would run better after firing half of them; - Belittle (twice) an editor who made the error of trying to address your concerns in good faith; - Dare any of the hundreds of WMF employees to come spar with you; - Suggest their limited answers are actually due to a marching order because they despise the community. I want Abstract Wikipedia, a flawed project based on an erroneous assumption and no community to carry it, to be abandoned shortly as much as anyone here, including you presumably. I also think communicating with the community should be a bigger part of the responsibilities of WMF employees. Nevertheless, I do not believe dealing with your constant, pointless, and toxic abuse is part of that. Either engage in good faith to find solutions, or control yourself. If what you need is a place to vent, find another one, because aside from lashing out and poisoning this already heated conversation further, you have accomplishing nothing. Choucas 🐦⬛19:24, 18 June 2026 (UTC)[reply]
Thank you. I'm operating in good faith here. Everyone here wants what's best, and, even without AGF, we wouldn't have spent countless hours contributing to Wikimedia projects if we didn't. Feeglgeef (talk) 21:24, 18 June 2026 (UTC)[reply]
Thank you. However, I get the feeling that Feeglgeef is a little out of their depth here, but there's nobody with any decision-making authority at the WMF willing to rescue them wasn't particularly kind, nor did it have anything to do with the merit of my argument. I understand that this community has a poor relationship with the WMF (I don't particularly like them either!), but that doesn't mean that Abstract Wikipedia has zero potential in any form, and just because I believe in that potential doesn't make me a mindless WMF droid. Feeglgeef (talk) 21:59, 18 June 2026 (UTC)[reply]
@Feeglgeef, @Phil Bridger Can both of y'all take it down a notch. Calling folks mindless WMF droids or insinuating that they are lazy or bad faith assumption because they have been ordered not to communicate with us plebs is definitely not okay. Sohom (talk) 07:21, 19 June 2026 (UTC)[reply]
True, but the people in charge of the project posting halftruths at best and then ignoring all follow-up questions isn't really okay either, and causes the irritation that shines through in some ill-tempered remarks here. Fram (talk) 07:36, 19 June 2026 (UTC)[reply]
I have my own misgivings that roughly align with yours (I've tried contributing to Abstract Wikipedia and found it extremely hard to do so -- much harder than the cost of machine translation or even writing something from scratch in my native language). However, I don't think Sannita is talking in half-truths, I think WMF (as an org) does believe that they can create a system that can: "make it much easier to start a local version of the article, instead of writing it from scratch or translating from another language." and the approach they are taking is to first build out the whole pipeline first and only then measure whether operating Abstract Wikipedia causes production outages, whether or not editors find it easier to write articles in Abstract Wiki-lang (for the lack of a better term) and whether the effort of learning to write Abstract Wiki-lang is worth the trade-off of being able to write a few dozen linguistic rules (or whether that is even possible at scale). cc @Femke's call out of whether future funds will be invested, my understanding is that the team is still on grant runway and my understanding is that they do plan on measuring whether or not Abstract Wikipedia is viable this year (actually maybe @DVrandecic (WMF) or @ATsay-WMF would be better placed to answer that specific question?) Sohom (talk) 08:20, 19 June 2026 (UTC)[reply]
Comments like "We did some limited tests in a controlled environment, but doing so can only help you so much in identifying potential problems. We are learning a lot by releasing the beta project (because this is still a beta), and we'll improve from there. " are half-truths at best, with words like "limited" or "beta" doing an awful lot of heavy work here.
"Abstract Wikipedia is in early beta, and we made it available live to enable learning, testing, and iteration. " is also a dubious statement. Like has been said, it was an untested alpha, where the WMF really dropped the ball quite dramatically, with the most logical conclusion that some KPI or deadline had to be met. Even the examples given for the use of Functions and Abstract were and are very dubious. At the moment, 1234[36] gives me "Wikifunctions returned a failed response: Error in the function call API". In other major languages, for the same article, I get errors like "Wikifunctions returned a failed response: Invalid executor response", "Wikifunctions returned a failed response: No matching lexeme for item in language", and "Wikifunctions returned a failed response: Argument value error". Fram (talk) 09:00, 19 June 2026 (UTC)[reply]
But nobody called anyone a "mindless WMF droid". The only use of anything like that term was by Feeglgeef saying that they were not a "mindless WMF droid". But I see that my words here will be twisted beyond recognition. Phil Bridger (talk) 07:40, 19 June 2026 (UTC)[reply]
I know it's popular in wiki-land to hate on anything related to AI, and in particular LLMs, but I look at what Abstract does and I look at what the LLM-driven translation tools do, and I just don't see the point of what we're doing. Our goal is to be able to take an article written in one language and allow somebody who does not read that language to read the article in a language they do know. The LLM tools do that today. They're not perfect, but they're good enough for most purposes, and they'll only get better over time.
I don't mind investing WMF money in research projects. There is value in adding to our understanding of how information is stored and processed, and exploring better ways we can deliver that information to our readers. Abstract and Functions is about that, so I'm happy that we've explored what's possible. But given the progress LLMs have made over the past few years, and the current state of what Abstract appears to be able to deliver, I'm just not seeing how this has a chance of becoming a viable product. That's not to say the experiment was a failure. A failure would be "we haven't learned anything". But we have learned something, which is that "this approach to automatic translations doesn't appear to be the way to go". RoySmith(talk)13:00, 19 June 2026 (UTC)[reply]
AI translation has a tendency to be confidently incorrect, but AI is certainly more accurate than using buggy functions and blaming it on the community. ~2026-34629-56 (talk) 15:52, 19 June 2026 (UTC)[reply]
It's a record of data and not AI. If WMF steers this in the right direction and invests the resources and work in the right places (which they are surprisingly failing to do for a project that they so champion) then the articles have the capacity to be much better than AI. In solidarity, Aaron Liu (talk) 17:31, 19 June 2026 (UTC)[reply]
How? Why would WMF be able to translate into obscure languages without the help of one or two linguists of those languages? Someone will have to translate all lexemes for these languages, add all labels for these languages, check the output of Abstract, and then give the green light. We need to be certain that they are native speakers or very fluent in the language, to avoid a Scots scenario. And all that to translate an Abstract article which will probably be worse (language, completeness, ...) than enwiki, dewiki, jawiki... and not adapted to the target language (see e.g. the example of 1234, which doesn't take into account any other year system). LLMs won't do all of these things either, but for a fair number of languages it will already produce a result which is as good as can be expected from Abstract, but without all the additional effort described. How will Abstract ever be better at those things? Fram (talk) 18:15, 19 June 2026 (UTC)[reply]
Exactly, that's what they would need to do (the advantage is deterministic outputs). It's worrying that the infrastructure for what they need to do is in such an alpha-at-most state. In solidarity, Aaron Liu (talk) 18:59, 19 June 2026 (UTC)[reply]
"LLMs won't do all of these things either" is such an understatement by the way. They have no understanding of the things they translate at all, which is pretty bad for an encyclopedia because we use technical terms whose translation is not always straightforward. For example, in some languages a field is called a "body" (presumably the ones that borrowed the term from French). My native language, Greek, is one of them. Guess what Google translate does when I ask it to translate Field (mathematics) ([37]). It switches between the correct term (σώμα) and the literal translation of "field" (πεδίο) between paragraphs. To no one's surprise, it sticks to the wrong word most of the time and it also uses the wrong word in the title. "Field" is far from unique in this. While checking this and other articles, I saw words like function, matrix, ideal, differentiation, infinitesimal, domain etc. get mistranslated too. That is in addition to all the other things it butchers along the way ("general quintic equations" -> "general quantum equations"?? in the field article) as well as grammatical mistakes even a complete novice wouldn't make ("Εάν η συνάρτηση είναι διαφορίσιμο(??) στο a" in derivative[38]). It's not a google exclusive problem either. In my testing with DeepL it didn't do much better. Also, since I mentioned French earlier, here is a French example where Lie ring turns into "ring of lies" (I don't speak French so I verified this with Wiktonary): [39]. Warudo (talk) 20:04, 19 June 2026 (UTC)[reply]
To be clear the point of my comment is that the only translation that is worth considering is one done by humans. I understand why people want to compare abstract Wikipedia to LLMs but I think that comparison misses the point. Even if abstract Wikipedia reaches a point where it is able to produce LLM level output, which I feel would be nothing short of a miracle, it would not be anywhere near good enough to consider using. Warudo (talk) 20:10, 19 June 2026 (UTC)[reply]
Incorrect word choice is hardly limited to computers. People do it too. I used to work with a guy from Cuba who, while fluent in English, never quite mastered the difference between "in" and "on" (they're both "en" in Spanish). The phone would ring, he would answer it, and hand it to me: "Bob is in the phone for you". Oddly enough, "in the phone" actually makes more sense, but that's not the correct English idiom.
I read a lot of text related to checkuser actions in various languages. It didn't take long for me to figure out that "doll" was just Google Translate's way of saying "sockpuppet" in Spanish, and so on for other languages and other bits of wiki-jargon. In the end, it really isn't a big deal. It was obvious from the get-go that "doll" wasn't the right word but I quickly figured out what it meant and ultimately there was no real hit to my cross-wiki reading comprehension.
I expect the same is true of "field" for you. You recognized that πεδίο didn't make sense, but you figured it out and now that you know what it means, so what if it's not the correct word? Languages evolve. Hardly a day goes by on the internet without me being reminded that I'm no longer one of the cool kids because they're using words I've never heard of before and I have to resort to Urban Dictionary to figure out what they're talking about. Is it really fair that we hold the LLMs to a higher standard that we hold ourselves? RoySmith(talk)20:52, 19 June 2026 (UTC)[reply]
Let's assume I didn't speak English and elwiki didn't have an article on fields. I'd read the translated enwiki article, see that the term is called "πεδίο" (this is the term that appears most, the other terms are probably translation errors) and move on. Except what if I then wanted to go look for more information outside Wikipedia? I'd then pick up my mathematics textbook, check the index and only find results about scalar fields and vector fields (which are called πεδία in Greek) but nothing about the algebraic structure. What I'm missing is that I'm looking in the wrong place. If only I knew what it was actually called. Then I want to learn what an ideal is. Google translate said it's called "ιδανικό" [40]. Huh, there's nothing here? If only I'd known the word is "ιδεώδες". And on and on it goes. So, ultimately, you can't look things up on your own because you don't know what they are called so you're stuck asking the AI about them. You also can't really discuss these concepts with others because your idiolect is completely different from theirs. And before you call this hyperbole, I'll let you know that google translate also mistranslates "associative" in the field article. So even if you try to tell your interlocutor the definition of the thing you're talking about, you're still out of luck unless they speak English and guess what happened.
You also underestimate just how confusing things can get. In the translation of ideal (ring theory), the term "right ideal" (which is supposed to be δεξιό ιδεώδες) turns into something that basically means "correct ideal" (σωστό ιδανικό) and "proper ideal" (which is supposed to be γνήσιο ιδεώδες) into something that means "suitable ideal" (κατάλληλο ιδανικό). So someone who doesn't speak English and does not know why these error happened will try to figure out why a right ideal is correct and a left one isn't and what a proper ideal is suitable for. Warudo (talk) 22:38, 19 June 2026 (UTC)[reply]
I feel your pain. My particular peeve is symbols. If I read in an article: and don't know what that symbol is, how do I even begin to look it up? Claude actually does a good job if I copy-paste the markup, I get back "The notation: F \ {0} means "the set F, with the element 0 removed." which at least gets me pointed in the right direction. But to get back to my original point, "they're good enough for most purposes". Failing to correctly translate some terms in advanced math articles doesn't negate that. RoySmith(talk)23:01, 19 June 2026 (UTC)[reply]
If by "better" you mean produce better translations in theory, in essence this is pitting a rules-based translation system where the rules are explicitly programmed by humans versus a black box where a program adjusted its tuning parameters by itself that control how translation is done from input to output. One example is the world's strongest chess program, Stockfish, which originally had a human-crafted evaluation function, then introduced a trained neural network, and now has dropped the human-crafted evaluation function. However games have a very specific ultimate evaluation method – did the program win – which I suspect makes it easier for the network to evolve to a state where it wins more often. I think human language translation has more training challenges, and it's not as clear to me that a program self-training could in a theoretical perfect world do a well as an explicit set of rules.
From a practical standpoint, though, codifying rules to capture all nuances is very difficult. And even to the extent it's possible, it's not clear that there are enough people interested in working on it in a crowdsourced project. So while the black box may produce OK translations only some percent of the time, that might still offer a better cost-benefit ratio than getting Wikifunctions to meet the same level of quality. isaacl (talk) 02:51, 20 June 2026 (UTC)[reply]
We are planning to trial an "incident reporting" form here for the next two months, after some smaller-scale deployments on Portuguese Wikipedia and several other wikis. After discussing with functionaries, the trial will start small (visible to 5% of users) and we plan to coordinate with functionaries along the way to agree on the right points to expand visibility.
The incident reporting form is designed to do two things:
Help less experienced community members more easily report potentially bad behavior to the community-managed place that can best deal with it.
For more rare and severe cases, it provides a form to directly report imminent threats of harm to the WMF Trust and Safety team.
This trial will be primarily focused on calibrating the first use case: helping editors report potentially bad behavior, without overloading the system. Registered editors, who have at least one edit and are unblocked and have a verified email, can click a "Report" button on a discussion page. Eligible editors who have Discussion Tools enabled can click a "Report" button on a particular comment. The editor can then choose the category of behavior at issue, and then be directed to a location chosen by that wiki's community for that category. There are screenshots of these interactions at the bottom of this post.
For the last few weeks, we've been working with functionaries to figure out what categories are most needed for English Wikipedia, and to identify the right destinations to point users for those categories.
Based on those conversations, we've made some changes to the categories from what we have deployed elsewhere, to optimize this trial for English Wikipedia, given your community's longstanding policies and processes for dispute resolution. We've further supported functionaries for the last couple of weeks as they've configured the form's behavior for the community and prepared supporting material for end users. Special thanks go to CaptainEek in particular for spending significant time writing and editing material for this trial.
During the trial, we will focus on monitoring the volume of new reports, checking that reports are routed correctly, and identifying any immediate issues. We will be coordinating closely with all community members to fix bugs if they arise, and to otherwise streamline the process. For example, we are exploring some ways to tighten the user experience and help people more directly submit their reports, which we may deploy and measure during the trial as well. We expect to gradually increase user visibility over the next two months.
We welcome your thoughts. If you'd like to talk to us off-wiki, the easiest way to reach us is on Discord.
1. The "Report" button, as shown in the Tools menu (see arrow).
2. The "Report" button, as shown in the overflow menu in a comment thread (for users with the Discussion Tools beta feature enabled).
3. The full set of categories of potentially unwanted behavior that an editor can pick from.
4. Support information for users reporting potential bullying.
5. Support information for users reporting potential sockpuppetry.
Thanks for that folks! For those who are interested, the system is locally customizable in two ways, which I spent the last month or so working on. Special:CommunityConfiguration/ReportIncident allows us to configure various options and links. This query shows all mediawiki pages that we may locally edit to localize text. Typo and link fixes are welcome; all other edits should obviously be discussed as these are full protected media wiki pages with wide transclusion. You may test out the incident reporting system at Event talk:Sandbox (it is not currently enabled in any other namespace); be cognizant that the system is live and unless you are listed as an end-to-end tester, your button clicks may be logged/your emergency reports may be sent. Eric and his team are working on building the tool out more, and they tell us that having test data will help build the tool out more effectively; the next step is hopefully to include a baked in reporting form or pre-filled form templates. CaptainEekEdits Ho Cap'n!⚓18:22, 29 April 2026 (UTC)[reply]
Additionally, what is the best place to discuss and ask questions? For example, I see that AIV is not listed in "How to report obvious spam", but is listed lower for vandalism – is there a reason for that? It could be very helpful to have a documentation page and a talk page for this feature, if that isn't already the case.Other question I'm having (sorry!), for reporting hateful content, Open a new thread on the appropriate noticeboard for routine and public incidents may give the impression to the reporter that their concerns are dismissed as "routine", is there a better wording or is my fear unfounded? Chaotic Enby (talk · contribs) 15:41, 30 April 2026 (UTC)[reply]
First, yes, I agree having a centralized project/talk page would be good; I can set one up this evening. Second, the trouble with that line is that the wording isn't customizable per category, so every option that has a report to the appropriate noticeboard link has to have the same wording (for now at least). But I agree I had a hard time thinking of wording that worked in most general scenarios; we could probably just remove "routine". CaptainEekEdits Ho Cap'n!⚓15:58, 30 April 2026 (UTC)[reply]
This seems like a cool idea, but if I understand correctly the "Report" button doesn't actually create a report? That seems misleading. I was expecting it to ask the user to fill out a form which would automatically post to ANI when submitted. Toadspike[Talk]12:13, 30 April 2026 (UTC)[reply]
You do understand correctly. This mismatch has been a source of feedback from the U4C and Steawrds from the start of the project. Eventually that is the goal but that won't be happening soon. However, there is now work being done to allow communities to choose email as an option for some report types. So if we sent people to say the OS queue for doxxing that would generate an actual report. And work has been completed on allowing communities to use a URL with parameters meaning we could do some prefill work at least to generate an actual notice out of the system. But yes an Incident Reporting System that only generated a report in one very narrow case (emergency) for a long time has been and continues to be a weakness of the system. Best, Barkeep49 (talk) 14:24, 30 April 2026 (UTC)[reply]
In that case, as a temporary measure, is there an option to bold the links to where editors can report? Given how the text can easily fill a page, it can be helpful to prevent editors from getting lost in the sea of links. Chaotic Enby (talk · contribs) 15:50, 30 April 2026 (UTC)[reply]
When I last tried on test wiki, bolding did not work (using markup or html); however some changes to allow code to work in some places have been made so I'll test it again and see about bolding links CaptainEekEdits Ho Cap'n!⚓19:20, 30 April 2026 (UTC)[reply]
@EMill-WMF, @MAna (WMF), is it possible to add horizontal rule in between the sections? It was not apparent that there are multiple forms for the different reporting/noticeboard venues because visually the noticeboard header and the form labels are similar. i.e. – robertsky (talk) 15:37, 30 April 2026 (UTC)[reply]
Hi, I'm not sure I understand what you're looking when mentioning multiple forms for different reporting/noticeboard venues, are you referring to how these are configured via Community Configuration? MAna (WMF) (talk) 17:42, 30 April 2026 (UTC)[reply]
Just to update - this incident reporting form is now live on English Wikipedia, though for now, the report link is only visible to 5% of users. EMill-WMF (talk) 19:30, 8 May 2026 (UTC)[reply]
Hi, we have a new update. The trial has been running in 5% for a week now, and the numbers are quite small, so we're planning to raise this to 20% on Friday.
Soon afterward, we'll be adding a new feature to the tool to make things more direct for users. For categories where the community wants reports to go to an email address, we're incorporating a web form into the tool that will let the user submit the report from there and it will send it to that email address on their behalf. MAna (WMF) (talk) 17:35, 14 May 2026 (UTC)[reply]
No, it's not general purpose - if the community specifies that reports of a certain type should go to a specific email, this form will appear and will send the contents to that email. It's more similar to how emergency reporting already works in the tool, which sends the contents to a specified (WMF) email address. EMill-WMF (talk) 18:47, 14 May 2026 (UTC)[reply]
Hi all, we have now enabled direct reporting to email destinations. For categories where the community has configured an email address as the reporting destination, users can now submit their report directly through a web form.
Now that we’ve deployed this change, after discussion with functionaries, we’re planning to increase visibility to 60% of enwiki users. MAna (WMF) (talk) 16:13, 3 June 2026 (UTC)[reply]
Thank you to everyone for their work to make this happen! Can we get a better default text at MediaWiki:Reportincident-unacceptable-behavior-footer? The first sentence is a tautology—of course unacceptable behavior is not tolerated. Even removing is not tolerated would be a substantial improvement: {{SITENAME}} is a community where [[Foundation:Special:MyLanguage/Policy:Universal Code of Conduct#3 – Unacceptable behaviour|unacceptable behavior]] is managed by the community of experienced users. (I know we can customize this locally, but this should be a global change.) Best, HouseBlaster (talk • he/they)02:33, 7 June 2026 (UTC)[reply]
Hi all, the trial has now been running at 60% visibility, and we have continued monitoring it closely. Based on our conversations with functionaries, we have enabled IRS in the Wikipedia namespace, and are planning to increase visibility to 100% of eligible users this week. MAna (WMF) (talk) 08:11, 17 June 2026 (UTC)[reply]
Hello, not sure if this is the best place to post this so pls lmk if elsewhere is better.
I really like the recent social media campaign WMF is running lately, where interesting things about/on Wikipedia are being introduced. While these are all great, they're very readership focused and a little surface level.
Imo what would be really helpful is if WMF made entertaining explainers about how Wikipedia works. E.g. this kind of format: [42]
For example, a really wide misunderstanding is whether Wikipedia has "mods"; many incorrectly assume Wikipedia admins are basically like Reddit mods: all-powerful and with strong control over Wikipedia's content. At least for the English Wikipedia, that's not true. Vids could be made to address this misconception. Would need multiple vids on this topic, spread out over time; it's so widespread and information takes time to disseminate.
Other possible topics include:
Explaining how article deletions/deletion nominations work (also really misunderstood)
Introducing the various parts of Wikipedia, like talk pages, the Teahouse, etc
Introducing various interesting tools Wikipedians use, like anti-vandal stuff, Twinkle, and AutoWikiBrowser.
The WMF is currently running banners with the text:
17 June: Wikipedia still can't be sold We're sorry we've made several attempts to reach you, but it's Wednesday, 17 June, and we still need some help. If Wikipedia is useful to you, please take a minute to protect its future with a $2.75 donation today. Most readers don't donate; only 2% do. Wikipedia is run by a nonprofit, not by a billionaire. So if Wikipedia has given you $2.75 worth of knowledge, please give. Any contribution helps, whether it's $2.75 or $25.
In the past, we produced a consensus that banners that state or imply any of the following are not considered appropriate on the English Wikipedia:
Wikipedia's existence or independence is under threat or dependent on donations
Donated funds are used primarily to support Wikipedia and/or its volunteer editors
Readers should feel obliged to donate regardless of their means ("guilt tripping")
I don't believe that these current banners comply with this consensus. I'm hopeful the WMF can resolve this quickly; I will post to the talk pages of a few relevant employees, asking them to comment. BilledMammal (talk) 07:25, 17 June 2026 (UTC)[reply]
Could you expand on why you believe this is against community consensus? Compared to the ads of old, I think this passes muster but maybe I'm missing something. In solidarity, —Femke (talk) 🐦 07:56, 17 June 2026 (UTC)[reply]
The most problematic text seems to be "protect its future", where "it" clearly means "Wikipedia". That could be read as suggesting firstly that a substantial part of any donation would go to Wikipedia, and secondly that Wikipedia is $2.75 short of being able to survive without it. However, the implication is indirect and less deceptive than previous banners. Certes (talk) 10:16, 17 June 2026 (UTC)[reply]
@Certes: A substantial part does go towards Wikipedia, cc the Annual Plan breakup. Regarding the original question, I personally align with Femke and RoySmith in that I don't think the banner breaks consensus given the context of the guidelines (which were primarily made for a different category of much more egregious banners) Sohom (talk) 16:42, 17 June 2026 (UTC)[reply]
Complaining about a banner is a different bar from "banner violates past consensus, needs to be resolved quickly" (which is the tone of this discussion). One is "lets have a dialog" another implies "do this now, or else". Sohom (talk) 19:41, 17 June 2026 (UTC)[reply]
@Black Kite Could you explain what part of this is deceptive? (When you answer, one thing to keep explicitly keep in mind that the funding situation has significantly changed since 2022 when this consensus was obtained, WMF does spend 77% of it's budget on volunteers, a majority of which goes to English Wikipedia, and unlike in 2022, donations are actually on a downward trend and WMF is actually now increasingly turning to long-term donors to continue funding itself -- while it still has a fair bit of money around, it is not infinite and the lack of any donations will indeed lead to insecure future at some point, especially in a climate where we are competing with billionare led ventures like Grokipedia). Sohom (talk) 20:15, 17 June 2026 (UTC)[reply]
"...take a minute to protect its future with a $2.75 donation today". Nothing is going to happen if you don't donate $2.75, and neither is it if no-one else does either. That was the reason the first clause was agreed the last time we had this discussion. Black Kite (talk)20:26, 17 June 2026 (UTC)[reply]
Collectively if no-one else does either is a problem. I remember doing some back of the napkin math, at our current spend rate, I think we have roughly 1.5 to 2 years assuming everyone immediately stops donating. This is obviously not a dooms-day, we pack up tomorrow scenario, but it's not "if nobody donates we will be around for 10 more years" scenario either and the "nobody really has to worry about donating" scenario that you are laying out. Sohom (talk) 20:49, 17 June 2026 (UTC)[reply]
Nothing is going to happen if you don't donate $2.75, and neither is it if no-one else does either. How exactly do you think WMF pays its employees and its bills for running its data centres? – SD0001 (talk) 16:47, 18 June 2026 (UTC)[reply]
I'm sorry, SD, you're just plain misinformed about where the money comes from. The way it works is the good wiki fairy comes along and waves her magic wand. That keeps the disks spinning and the bits flowing and enlightens repressive regimes around the world about freedom of information so we don't have to waste any effort defending our editors in court. It also makes hotel rooms and airline tickets drop from the sky so volunteers can get scholarships to wikimania. And as the wiki fairy walks along the beach, gold coins grow out of her footprints in the sand; these go to funding local affiliates to keep their education and outreach programs afloat.
All we need to do is join hands and think happy thoughts in a big group hug so the wiki fairy will come and save us. Until then, I'm willing to put up with a few banners. RoySmith(talk)17:26, 18 June 2026 (UTC)[reply]
Do we have any detail on this 77% figure? It's one I've not seen before, and I am really struggling to see where $150 million is being spent on volunteers this year. It feels much, much less. Certes (talk) 20:41, 17 June 2026 (UTC)[reply]
Thank you. That's interesting. I'm particularly heartened to see that the Annual Plan actually mentions Wikipedia – an improvement on the last one I read. It's always difficult to categorise spending. "Investment in Technology" (48%) includes issues such as management of scraper bots, which is a very welcome function but not one directly aimed at volunteers. Presumably it also covers developments such as Abstract Wikipedia which fewer volunteers might prioritise. We certainly appreciate the work of the community tech team which this item has funded in previous years. Certes (talk) 21:05, 17 June 2026 (UTC)[reply]
Presumably it also covers developments such as Abstract Wikipedia which fewer volunteers might prioritise., it does, I had written up a guide about the current annual plan to contextualize what is planned for next year. It's a bit out of date, but it should be relatively accurate and Abstract Wikipedia is a fairly small portion of the pie. A lot of WMF's current spending is on making it easier for intermediate editors learn to become competent editors, make readers stay around for longer and a chunk to help admins fight vandalism effectively (outside fighting scrapers itself and making sure we have high availability) Alongside that you have work being done to make Toolforge more user friendly and making existing APIs better so that folks can reuse content without taking down Wikipedia itself. Sohom (talk) 04:18, 18 June 2026 (UTC)[reply]
Most readers don't donate; only 2% do was previously rejected as "guilt tripping"
I also consider Wikipedia still can't be sold to be suggesting that Wikipedia's independence is under threat, although it isn't as blatant as previous examples.
I think people would also be frustrated by the fact they're seeing something several times in the first place. I can't guarantee how other people would perceive it, but more than three times feels like a lot, especially if the answer has been "sorry, no, don't have the money to spare" the first time. If I recall correctly (please tell me if I'm incorrect!) the banners also have a "maybe" option if you want to be reminded, so I don't think it's simply a matter of people that might actually want to be reminded. When I worked at McDonald's, we had lots of people who would donate to the Ronald McDonald house the first day or two of fundraising and then got quickly annoyed later on as we kept going throughout the week ("I already donated"; "no, I just want my coffee"). Clovermoss🍀(talk)01:08, 24 June 2026 (UTC)[reply]
No one is going to donate when prompted the first time, regardless of how much they have in the bank. The fundraising team gets it. Once someone has donated, I believe cookies are used to track it and they stop seeing the banners. – SD0001 (talk) 07:53, 24 June 2026 (UTC)[reply]
As I recall, it's been previously said that no association is made between donors and on-wiki user names or any session information for a given browser (linked via cookies or I suppose local browser storage). From a privacy perspective, personally I think this is sensible. isaacl (talk) 17:26, 24 June 2026 (UTC)[reply]
I support the banner ad program and the current language. It is a reasonable, unobtrusive, and voluntary option for meeting the organization's financial needs... ensuring that Wikipedia can persist long into the future as an independent source of information about the world. They tried letting the community run the ship on the fundraising a few years back and it imploded. Clearly this is not something that the "consensus" is equipped to handle. Ice Vest (talk) 22:57, 17 June 2026 (UTC)[reply]
Is no one else irked by We're sorry we've made several attempts to reach you? That's the kind of nagging tone that turns me off any appeal from a charity or non-profit if I get it in a phone call/voicemail or in the post. It sure feels like "guilt-tripping" to me. —In solidarity with Wiki Workers United · ClaudineChionh (she/her · talk · email) 23:00, 17 June 2026 (UTC)[reply]
But are they truly sorry? Wouldn't someone who was genuinely apologetic about having contacted me refrain from doing so again? Certes (talk) 11:11, 18 June 2026 (UTC)[reply]
I want to give some context about the phrase "protect Wikipedia's future;" the intent is not to suggest that Wikipedia's existence is in immediate danger or that any individual donation determines whether Wikipedia survives. In an era of falloffs in referral traffic and readership decline, protecting Wikipedia's future means ensuring that Wikipedia remains accessible, visible, and sustainable in a changing world.
I also recognize that people may interpret fundraising language differently – which this conversation shows. We value your input and will think of other ways to convey this point, and we'll share them back with you. Feedback on different interpretations is something we look at closely and frequently incorporate into banner language. Every year we workshop fundraising language with communities. You can see Fundraising/2025 banners for community-WMF banner discussions about last year's campaign. Next week, we will be publishing our regular collaboration on fundraising banner content at the Fundraising Hub. We welcome feedback on particular messages there, where we can get more in depth around wording and the considerations behind it.
These banners that are asking for donations and making false claims about the website going bankrupt is "misleading and unacceptable" and is being used as a scare tactic to force editors to donate. These banners represent nothing but corporate greed and lies.
The platform's fundraising strategy faces considerable public and internal critique:
Substantial Reserves: Financial statements reveal the Wikimedia Foundation holds hundreds of millions of dollars in cash and investments, meaning it is not in danger of shutting down.
Aggressive Language: Critics, and even some veteran Wikipedia editors, argue that the banners misleadingly imply the site is on the verge of going offline if readers do not donate immediately.
Organizational Growth: A large portion of the budget goes toward expanding the Foundation’s bureaucratic footprint and funding adjacent non-profit grants rather than just maintaining the core website.
Hello @~2026-35699-89: could you share what you see in those banners? They should not imply bankruptcy is near.
Just so you know, the Foundation has stopped its grant programme for knowledge equity. Having reserves is a good thing given how tumultuous the world is now I'd say. In solidarity, —Femke (talk) 🐦 18:26, 18 June 2026 (UTC)[reply]
This is exactly what i saw: 18 June: Wikipedia still can't be sold
We're sorry we've made several attempts to reach you, but it's Thursday, 18 June, and we still need some help. If Wikipedia is useful to you, please take a minute to protect its future with a $2.75 donation today. Most readers don't donate; only 2% do. Wikipedia is run by a nonprofit, not by a billionaire. So if Wikipedia has given you $2.75 worth of knowledge, please give. Any contribution helps, whether it's $2.75 or $25. ~2026-35699-89 (talk) 05:43, 19 June 2026 (UTC)[reply]
Anyone who knows me will be well aware I won't hesitate to criticize the WMF when warranted, but I don't see anything there that states or implies imminent bankruptcy. SeraphimbladeTalk to me14:02, 25 June 2026 (UTC)[reply]
I am disappointed to report thatThe user interface icon library will be updated later this week or next week. Most of the ~300 icons have been slightly refined and ~30 new icons have been added. These changes improve the icons to make them more consistent and comprehensible, and provide more visual balance when they are used in groups.
Despite the claim that these updates will make the interface more consistent and comprehensible, I think quite a few of the designs are poorly thought out, overly simplified, and missing some of the clarity from the icons they are replacing. Frustratingly we're losing a couple of beloved icons like the classic UserTalk into a version which is.. well, honestly just awful in comparison.
I know it feels like I am an old grump just hating on change. I genuinely don't mind refinements to UX when it's thought out well, but I do see a lot of the new icons as a real downgrade. It's also mildly frustrating that there wasn't much community involvement in this process, considering it'll impact the look and feel of the site we use, however small. qcne(talk)17:29, 17 June 2026 (UTC)[reply]
+1 but also I feel like people engage in copyright pedantry when they say that can be present in both editors no issue (well, not any more with this new, worse icon version, but you get my point) but as soon as you include it on a random page, then it needs the full MIT licence linked somewhere to be fine to use. stjn18:59, 17 June 2026 (UTC)[reply]
In one place it's part of the software and can be presumed to be covered by the software's licensing. In the other it's part of the content, not covered by the software's licensing. Anomie⚔22:38, 17 June 2026 (UTC)[reply]
It is pretty questionable that images that are entirely template-generated like that case are an open and shut ‘part of the content’ just depending on their position in the interface (or the method of addition? does a JS-added image need to also adhere to MIT licence by linking? would something added via OOUI methods need to adhere to that?). stjn22:46, 17 June 2026 (UTC)[reply]
"Position in the interface" doesn't really matter. "Method of addition" is more relevant. An image added by content JS or CSS should be attributed when the image's license requires it, while images that are part of the OOUI software would be covered by that software's license. Probably a good rule of thumb (i.e. let's not get into nitpicking edge cases) would be that if the image is coming from Commons or a local enwiki upload then it should likely be linked if the license requires attribution, while if it's loaded from some MediaWiki path then it's covered by being part of the software. Yes, even if it's the same underlying image data. Anomie⚔12:12, 18 June 2026 (UTC)[reply]
Some of the icons were genuinely good changes (I liked the overall idea of combining filled and outlined icons to establish importance through weight), but the overall set feels very hit-or-miss. I'd be happy with keeping most of the current ones and adapting those that could do well with an outline/fill pair without having to redo the whole thing.Best example that comes to mind is the message icon set: adding an outlined version could easily be done without changing the whole thing and creating whichever cursed abomination the new "thanks" (UserTalk) button is.In general, I would say the focus on having a consistent outline width hurts the icon set more than it helps it, leading to awkward icons (such as the Article, HalfBright or SpecialPages ones, to name a few) or sometimes just straight-up overflows (ArticleNotFound, or Keyboard being all keys and no board). Icon symbolism also gets lost: SpecialPages previously had a star and UserRights a cogwheel, but both now get a more obscure asterisk, despite PageSettings keeping the cogwheel.These (and many others) really makes this feel like a downgrade in terms of having a consistent scheme. Most of it does feel caused by the focus on keeping every icon as an outline with a consistent thickness, which itself wasn't really necessary, but some are certainly odd choices beyond that. Chaotic Enby (in solidarity · talk · contribs) 18:39, 17 June 2026 (UTC)[reply]
Anyone know of a way to undo the new icon changes and restore the old icons? Can this be done with an alternate skin or a bit of CSS? Levivich (talk) 01:19, 21 June 2026 (UTC)[reply]
Hi everyone -- I posted the below update on the separate page about the wishlist, but also wanted to copy it over here to increase its visibility for people who are interested. I think that the separate page might be better for discussion than here on VPWMF.
Hi all -- following up from June 11, I just posted an update on Meta about how we're thinking about the wishlist program going forward, after discussing ideas and concerns with some users with extended rights (thank you!) and hearing the comments from the many volunteers who have been writing on pages like this one and elsewhere. This update proposes some of the broad strokes for the future, and it includes open questions that I'm hoping volunteers will weigh in on. We'll keep discussing and posting updates to clarify details. Please speak up on here or on the talk page on Meta with your thoughts -- we want to shape a wishlist program that will deliver what volunteers need. -- MMiller (WMF) (talk) 20:34, 17 June 2026 (UTC)[reply]
Here is a quick overview of highlights from the Wikimedia Foundation since our last issue on June 5. Please help translate.
Simplifying account creation: The Foundation is working on improving the account creation process to reduce potential friction for newcomers to create an account. Improvements include making "Create Account" icon more prominent on mobile, simplifying the registration form, and introducing real-time username validation.
New U4C members elected: The Universal Code of Conduct Coordinating Committee (U4C) has new members and has two remaining vacancies in Middle East & North Africa and Sub-Saharan Africa.
Wikimania conference program: The Wikimania 2026 program is now live! Take a moment to review the program, and, if you are logged in, you can mark your "must see” sessions with a star to start building your personal schedule. Register for a virtual ticket here, if you haven't signed up yet.
Screenshot showing the Explore Feed refresh in the Wikipedia app (from the Community tab entry point)
App Explore feed: The redesigned App Explore feed, now called "Home", has been released to all Android users. The update introduces a refreshed feed experience along with the first set of new content modules, including Did You Know, Places of Interest, Random Article, and a new end-of-feed experience. Additional content and improvements are planned in future releases.
Wikipedia games: The Which came first? daily trivia game is now available in the beta version of the Wikipedia iOS app in English, German, French, Portuguese, Russian, Spanish, Arabic, Chinese, and Turkish. The game uses historical events from Wikipedia’s “On This Day” content and challenges readers to guess which of two events happened first.
Reusing references: Sub-referencing, a new MediaWiki feature that allows editors to reuse references with different details, has been rolled out to Group 1 wikis and French Wikipedia following a successful pilot phase.
Article guidance: The Article guidance feature is being tested with some editors creating new articles on the Simple English, French, and Turkish Wikipedias. The experiment will soon begin on the Arabic and Bangla Wikipedias as well. The outlines guide less experienced editors in creating high-quality articles. A quick guide to markups used in outlines can be found on this page. Example outlines that can be adapted and instructions for how to adapt them are on this section of the project page.
Mobile Page Previews: The Page Previews experiment on mobile web has concluded with the decision not to roll out the feature. The results showed no statistically significant impact on reader retention – the primary success metric. Page Previews, which are already available on desktop and in the apps, display a thumbnail, lead paragraph, and link to the full article when readers tap a blue link.
Wikifunctions: You can now add images to Abstract Wikipedia and the loading and display of test results when viewing Functions has been improved.
Wikidata: The Foundation is migrating the Wikidata Query Service (WDQS) away from the Blazegraph backend since it no longer scales efficiently with Wikidata’s growth. The migration will take place in several phases. Here is the timeline.
Growth features: Growth features are now available at Wikidata! Wikidata administrators are still configuring the features through Community Configuration, but this update enables access to Mentorship (if configured), Impact, the Help Panel, and a simplified Newcomer Homepage (without Suggested Edits).
Mentors' management: The Growth team will soon provide a system to automatically suspend or remove inactive mentors from the list of mentors. Communities can already start configuring the process in the Community Configuration.
CollaborativeContributions: If you need help setting up the Collaborative Contributions and the Goal setting features, check out these video guides. These features allow you to view which edits are made during an event and allows the group to track progress against a goal with a public progress bar. Learn more.
Latest experiments: An upcoming experiment is testing whether we can serve readers better when a footnote click in read mode shows the full bibliographic information rather than flying them to the reference list. See all live, upcoming, and completed experiments in Product & Technology.
Tech News: The latest highlights from Tech News week 24 and 25 include how the user interface icon library is being updated. Most of the ~300 icons have been slightly refined and ~30 new icons have been added. See also the 62 community submitted tasks that were resolved over the last two weeks.
Digital Safety: Join a conversation about Using AI Safely. It'll explore risks & concerns of using AI tools in personal and organisational contexts and practical strategies to reduce those risks. It will take place at 03:30 UTC & 14:30 UTC on June 26. This session is not about using AI to edit Wikipedia. It's focused entirely on safe personal and organisational use.
Don't Blink: Read the latest developments from around the world about protecting the Wikimedia model, its people and its values. Highlights include exploring the implications that new child safety regulations have on privacy online.
Grantmaking strategy & Affiliate model: Members of the Global Resources Distribution Committee (GRDC) and the Affiliations Committee (AffCom) met to advance two key movement initiatives: the development of a new Grantmaking strategy and a refreshed Affiliate Model. They produced initial proposals and advanced work on both initiatives ahead of broader conversations planned for Wikimania 2026.
Is it true they've voted? I am unfamiliar with UK union processes. But I know in the US this step would be after member cards had been signed by a majority of workers, with no vote having yet occurred. The employer would then have the option of recognizing the union based on the membership cards or requiring a secret vote to confirm that a majority of workers support the union (and this is where many anti-union measures then take place). Either way I hope the WMF recognizes WWU-UK. Best, Barkeep49 (talk) 15:18, 24 June 2026 (UTC)[reply]
Unlike unions in the US, individuals can join a trade union and pay membership dues. This means before a vote even happens, British unions have a strong sense of their levels of support before seeking recognition. There were some changes in April of this year too, that made union recognition easier both voluntary and via arbitration.
Users whose account email addresses are unconfirmed are at higher risk of getting locked out of their account, as they cannot reset their password, and the Foundation's support desk may not be able to help them recover their account. More broadly, unconfirmed email addresses are unreliable, and may not even belong to the user who controls the account. We are considering further interventions like this to ensure that the email addresses users provide are more consistently confirmed.
We recently ran an experiment to measure whether this banner is effective at convincing users to confirm the email address. There was a significant increase in people confirming an email address and negligible evidence of users removing their email address entirely.
I know it's unfashionable to use this page for anything other than "I hate the WMF and everything they do" screeds, but I'd like to point out a recent win. As of a month or so ago, there's a "Preview" button in the wiki editor that breaks the screen vertically, letting you edit the wikitext on the left and see the rendered output on the right. This is a really nice feature. Not perfect (I wish the refresh happened quicker), but more convenient than the "Show preview" button. I use it all the time. Thank you to whatever team made this happen. On a related note, I do wish T384947 could be addressed, but this is sort of a partial solution to that problem, so a good move in the right direction. RoySmith(talk)12:43, 26 June 2026 (UTC)[reply]
Good point. I've not tried that Preview function but I see the button now that you mention it -- it's right in my current editing window. It helps in that case that it has the word "preview" to identify and explain what it does. Next to that button is an unlabelled magnifying glass icon. This has a tooltip saying "find and replace" which is an editing function I have long wanted. I haven't tried this one yet but will get to it. One obstacle to doing this is that there are multiple editing modes on Wikipedia including: Visual editor, Text editor for articles, Text editor for talk page replies. I'm not sure how identical that last two are. I'm using the main text editor now I think but it's hard to be sure because there's a lack of context to tell me which editor I'm using or what its name is.
Two other new functions that I've noticed lately are cards in the iOS app. One is editing suggestions which usually suggests that I add a picture to an article. This seems to work reasonably well and be reasonably useful and interesting. The other is a Game card which usually offers the chance to guess which of two events came first. This doesn't interest me because, either I know and it's obvious or I don't know and it's a guess.
Another new thing I noticed lately is the baby globe which I dislike.
So, the WMF is putting out a variety of new bells and whistles but it is rather hard to spot and understand them because there are so many modes and options. What helps me is having labels and tooltips and titles to provide a clue as to what these things are and what mode I'm in. Does the WMF have a style guide for its interface(s) like we have the Manual of Style for articles?
The find-and-replace button has always been there, actually. There's a very simple reason lots never knew about it: It was under the "Advanced" dropdown for some reason until like last year or something. (The normal/article text editor is called WikiEditor, or sometimes the "2010 wikitext editor". It is indeed different from the quick reply text editor.)Editing suggestions are also here, on the website; they've been here since 2020. See mw:Growth/Feature_summary or WP:GROWTHTEAM.
Does the WMF have a style guide for its interface(s)
I've been using this for a while, although I forget it sometimes. I'm always messing up something with formatting, so it's helpful. Katzrockso (talk) 02:54, 27 June 2026 (UTC)[reply]
Two days ago, Wiki Workers United, the global WMF staff union, formally requested recognition as a bargaining unit in the UK. Bernadette Meehan, the CEO of the WMF, stated a month ago that the Wikimedia Foundation respects the rights of all eligible employees to determine for themselves whether to pursue union representation. We are committed to ensuring that any process related to union conversations respects the rights and perspectives of all staff. We aren’t only doing this because the law requires it; I want the Foundation to operate this way because I think it’s the right approach.
I have a simple pair of questions for Bernadette, or Jimbo, or anyone else who can speak for the WMF. And I don't want these questions to sound adversarial, and would ask anyone who also wants them answered to not turn this into something adversarial. The WMF has repeatedly reassured both the community and WMF staff that it is not anti-union and that it supports whatever outcome may come. At the same time, a lot of the WMF's messaging has given off the impression that fixing community–WMF relations is a lost cause, that some of us can't be pleased. So, Bernadette, Jimbo, anybody, please understand: I and—I can confidently say—the vast majority of the other 1,123 editors willing to go on strike in solidarity with WWU would love to see this be a turning point where the WMF makes good on what it's committed to and starts to rebuild the trust it's lost in the community and its workforce alike. We're a month out from Wikimania and it would be great for us all to be able to have conversations there in a spirit of mutual trust.
According to Wikipedia:Copyrights, the Wikimedia Foundation does not own Wikipedia's article text or images and cannot grant permission to reuse these contents outside of the Wikipedia project. However the Wikimedia Foundation has recently partnered with major Generative AI companies to allow them to train their Large Language Models on Wikipedia content, according to multiple news articles[43][44][45][46] and the WMF themselves.
Generative AI and LLMs usually train off of other people's work, often without letting them know or giving proper attribution, which infringes on the rights of copyright holders (in this case the editors) and the terms of the Creative Commons Attribution-ShareAlike 4.0 license.
For the above reasons I would like to ask to the Wikimedia Foundation what steps are you doing to ensure that Wikipedia gets properly attributed when Large Language Models get trained on it and generate an output that contains (entire or parts of) Wikipedia article text and images, and additionally I'd like to ask for clarification on whether you own or not such content.
As long as the generative AIs and LLMs train on Wikipedia content, there is no violation of the CC license. Anybody (or any program) can read Wikipedia without notifying the copyright holders. Anyone can download the database without notifying copyright holders. The CC license requirements only apply if the AIs and LLMs quote or closely paraphrase Wikipedia content. If an AI or LLM is using quotes or close paraphrases from Wikipedia in responses that are made public, then the copyright holders could attempt to stop it, but the Foundation would not have any legal standing in such a matter. What other actions do you think the Foundation could take to prevent AIs/LLMs from quoting Wikipedia content without proper attribution? I might add that this problem is not confined to AIs/LLMs. I have seen (years ago) content that I added to Wikipedia copied word-for-word (multiple whole paragraphs) in Facebook posts without any attribution. In one case, I complained to the admin of the Facebook group about the license violation. The admin removed the post, but that was a minor victory. Donald Albury14:52, 29 June 2026 (UTC)[reply]
.png)
can be present in both editors no issue (well, not any more with this new, worse icon version, but you get my point) but as soon as you include it on a random page, then it needs the full MIT licence linked somewhere to be fine to use. stjn 18:59, 17 June 2026 (UTC)[reply]
