[00:14:17] Perhaps lexemes should be overridden by values from properties e.g. P1705 (re @u99of9: That may be the best lexeme specifically for the role, because the (better) label is made of regular lexemes...) [01:03:25] Hot news: Z33068 may now be usable! Apparently memoization fixed T432021 [01:31:45] If we are looking to add an important city as a demonstration that is missing in all three languages, I'd nominate *Q30340* (re @u99of9: Maybe my misunderstanding, but I thought we were just offering the system demonstration items in the pilot? Q319, Q42, Q84, Q52,...) [01:57:19] wouldn't it be nice to have one from each of the pilot communities? [03:03:11] Yes that would be good, but they likely have an article in their own language about their own city, so this would be for authoring it to provide to the other two. IMO the pilot communities should prioritise configuring functions/labels/lexemes into their languages to receive AW articles, rather than learning to create AW articles. It's hard for them to do it all at [03:03:11] once. (re @wmt [03:03:12] elegram_bot: wouldn't it be nice to have one from each of the pilot communities?) [05:50:17] But then we we need to add all forms as values on the item too? (re @u99of9: Perhaps lexemes should be overridden by values from properties e.g. P1705) [06:02:47] *Abstract Wikipedia Newsletter 267* [06:02:48] * Refreshing results: Cache usage reporting and "Run with latest data" [06:02:49] * Working with our first pilot communities [06:02:51] * Recent Changes in the software [06:02:52] * Volunteers’ Corner on October 12th 2026 [06:02:54] * News in Types: Change in anonymous functions [06:02:55] * Fresh Functions weekly: 166 new Functions [06:02:57] https://www.wikifunctions.org/wiki/Wikifunctions:Status_updates/2026-10-02 [06:02:58] _(as always, you get the preview, I plan to send out the email and mastodon news later tonight or tomorrow)_ [06:07:11] That sounds hard. Maybe only when we need to override lexemes like "POTUS"? In this particular case we do already have Q11696#P2521 and Q11696#3321. What's your recommendation? Should we override by making competing lexemes? (re @Jan_ainali: But then we we need to add all forms as values on the item too?) [06:13:41] This entire sentence is tricky. For Swedish, "George Washington was the first president of the United States" could be something like "George Washington var den förste amerikanske presidenten." (Most natural.) or "George Washintgon var den förste presidenten i USA." (Works as well, but sounds more technical.) Both these cahnge the order and have a determined form. [06:13:41] To squeeze in [06:13:42] the label without changing the form, we have to change it to something like "George Washington var den första personen som var USA:s president." (= George Washington was the first person who was president of the United States.) (re @u99of9: That sounds hard. Maybe only when we need to override lexemes like "POTUS"? In this particular case we do already have [06:13:42] Q11696#P2...) [06:23:13] Comparing the English translations of the three, I actually find the third the most specific, especially if linking [[POTUS]]: (re @Jan_ainali: This entire sentence is tricky. For Swedish, "George Washington was the first president of the United States" could be something...) [06:23:14] George Washington was the first American president. [06:23:15] George Washington was the first president of the United States. [06:23:16] George Washington was the first person to serve as President of the United States. [06:25:59] We're back at the "items that are not proper nouns and have labels that are phrases consisting of several lexemes" again :) [06:26:59] Radical idea: scrap labels made up of strings and have them be sequences of lexeme forms instead. [06:34:03] I think that [07:52:03] I'd like to contribute to abstract wikipedia. Where should i request help when i think a function should be improved? [07:54:10] Here. Or WF project chat. Or if it needs a language specialist, I'd go straight to their talk page. [07:56:25] Or if you want to learn how to fix it yourself, we could have a go at that too. [07:57:31] As @u99of9 says, but if it’s a particular problem with a particular function, I’d raise it on the function’s talk page. (re @Shi: I'd like to contribute to abstract wikipedia. Where should i request help when i think a function should be improved?) [08:02:57] I tried that and got no reaction so far (re @Al: As @u99of9 says, but if it’s a particular problem with a particular function, I’d raise it on the function’s talk page.) [08:03:42] Which function? [08:05:28] Z38757 requires a person to have a date of birth. In my case i couldn't find it out. Can we make the birth day part optional? (re @u99of9: Here. Or WF project chat. Or if it needs a language specialist, I'd go straight to their talk page.) [08:06:10] The sentence should work without i suppose [08:06:22] Another even slower method which doesn't ask anything of anyone in particular but is pretty powerful in the long term is to add a test case demonstrating the problem. [08:07:10] Yes that is a good idea (re @Shi: The sentence should work without i suppose) [08:09:39] As with short descriptions and other natural-language content in Wikidata, I think Abstract Wikipedia might be the appropriate location for their construction, with Wikidata exposing a particular value served by an AW API (label, short description, image, section, article…). (re @vrandecic: I think that's a great idea. Once we have a ZObject-type for properties in [08:09:39] Wikidata, I w [08:09:40] ould hope there will be a property for e...) [08:18:03] Sounds good. Did that 👍 (re @u99of9: Another even slower method which doesn't ask anything of anyone in particular but is pretty powerful in the long term is to add ...) [08:29:21] We should think about the architecture of functions like these. The NLG function should operate over the required/optional values (as arguments), with language-neutral functions fetching the required data before dispatch by language. Please also see [[:abstract:Abstract Wikipedia:Project chat#How should Abstract Wikipedia work?]]. (re @u99of9: Yes that is a good idea) [08:34:27] Is it clear how that should be done without optional arguments? (re @Al: We should think about the architecture of functions like these. The NLG function should operate over the required/optional value...) [08:35:21] Not yet 😏 (re @u99of9: Is it clear how that should be done without optional arguments?) [08:36:43] Are we sure all language MOS want the same inputs? In addition to the dates many want to dig the gender out of the QID. (re @Al: We should think about the architecture of functions like these. The NLG function should operate over the required/optional value...) [08:38:04] I find that supplying only the QID to the config protects me against linguistic and grammatical assumptions. (re @u99of9: Are we sure all language MOS want the same inputs? In addition to the dates many want to dig the gender out of the QID.) [08:48:58] Hmm… I do think there should be clarity about what information a function call conveys, or is capable of conveying. There is a grey area around how different languages frame their referents (gender, animacy, definiteness etc), but if we’re saying that some Wikipedias will include the date of birth while others won’t… let’s just say I have misgivings. (re @u99of9: I find [08:48:58] [08:48:58] that supplying only the QID to the config protects me against linguistic and grammatical assumptions.) [09:19:31] I don't anticipate any language ditching the dates (?unless they have no long term calendar concept). But I suppose some lead sentences could be augmented, e.g. by titles, postnominals, ... (re @Al: Hmm… I do think there should be clarity about what information a function call conveys, or is capable of conveying. There is a g...) [09:31:16] ... Aliases, native language names, pronunciations... [09:55:17] Indeed… but my first instinct is to suggest that the “augmentations” are standard and omission is an option. For example, name in native language could be suppressed for content in that language or variants. In any event, contributors to AW should be able to see what actual “facts” are asserted across all languages, not just the small number they can speak. Despite the [09:55:17] [09:55:18] grammatical grey areas, I would be inclined extend that to the information underlying {gender, number, animacy…}, even if that information is conveyed in only a few languages. (re @u99of9: I don't anticipate any language ditching the dates (?unless they have no long term calendar concept). But I suppose some lead se...) [10:03:11] How would you expect the AW editor to see this? (re @Al: Indeed… but my first instinct is to suggest that the “augmentations” are standard and omission is an option. For example, name i...) [10:12:55] Probably in the “mul” view, initially. But it would be nice if explicit omissions could be surfaced within any language’s content. Otherwise, the content should still be represented in the HTML (heroic assumption). (re @u99of9: How would you expect the AW editor to see this?) [10:14:47] And you want the top level html function to pull all of those bits out of the WD item and pass them all to the NLG config functions? [10:31:51] Yes. Have you read [[:abstract:Abstract Wikipedia talk:About#How does Abstract Wikipedia work?]]? It doesn’t answer all questions, but it might be a useful starting point. (re @u99of9: And you want the top level html function to pull all of those bits out of the WD item and pass them all to the NLG config functi...) [10:33:36] I have. It was too abstract for me, but I'll have another read now. [10:34:48] Me too 😏 But we have to start somewhere! 😎 (re @u99of9: I have. It was too abstract for me, but I'll have another read now.) [10:42:28] Okay I've read again, but didn't get much more this time, sorry. If I could see actual functions designed your way, I'd happily critique them. Otherwise I just have to anticipate and ask clarifiers... [10:45:37] One difficulty I see with this is that it puts a big burden on the initial function writers to anticipate exactly what will be wanted. Every "..." we wrote when listing possible augmentations would have to be exhaustively anticipated when the config was first set up. It would be very hard to change later. (re @u99of9: And you want the top level html function to pull [10:45:37] all of those [10:45:37] bits out of the WD item and pass them all to the NLG config functi...) [10:48:41] Perhaps you could pass a list of typed pairs with all the sub-arguments which would allow for later extension... But this would be ugly for the receiving NLG functions to have to deal with. [11:03:29] No worries either way. There are limits to how far we can get with our current capabilities. Conceptually, we just construct a map per referent and pass that down, which is more “variadic” than the current approach, and naturally allows for optional arguments and later extensions. The key simplifier, at this stage, is that the context for the KR-map is limited to the [11:03:29] particul [11:03:30] ar domain of the NLG function it supports. In the [person lead sentence…] case, it includes a person’s dates, nationality, notable occupations, natural gender… whether these are retrieved from Wikidata or supplied by the AW call (or evaluated with respect to both). (re @u99of9: Okay I've read again, but didn't get much more this time, sorry. If I could see actual [11:03:30] functions [11:03:31] designed your way, I'd happily ...) [11:08:35] Yes I can see that working structurally if it's all passed as a single map. And yes the domain limit helps. So, technically it sounds okay, but every NLG function then has to extract all its ~arguments from the map, which sounds both tricky for newcomers and laborious compared to having them already available as arguments. (re @Al: No worries either way. There are [11:08:35] limits to how f [11:08:36] ar we can get with our current capabilities. Conceptually, we just construct a m...) [11:15:58] Maybe. You could still call a [birth date] or [property of person/QID] function; you would just be able to pass the map as well as the QID, relying on the called function to encapsulate the map access, falling back to fetching from Wikidata if the map lacks the required kvp. (re @u99of9: Yes I can see that working structurally if it's all passed as a single map. And yes [11:15:58] the domai [11:15:58] n limit helps. So, technically it s...) [11:27:42] Would you be interested in building an example of your ideal? (re @Al: Maybe. You could still call a [birth date] or [property of person/QID] function; you would just be able to pass the map as well ...) [11:32:34] I’ll certainly bear it in mind when considering how to handle the Z43962 case. (re @u99of9: Would you be interested in building an example of your ideal?) [11:49:19] Parallel implementations with distinct configurations could cause trouble for the implementation selector until one was definitively better. (re @Al: I’ll certainly bear it in mind when considering how to handle the Z43962 case.) [11:59:45] Yes, but the function signatures would be different, so it would be a physically distinct function. (re @u99of9: Parallel implementations with distinct configurations could cause trouble for the implementation selector until one was definiti...) [12:01:44] Oh, I thought the top level html call was still the simple QID one? (Then it's implementation had to extract everything and pass it to the NLG configs.) (re @Al: Yes, but the function signatures would be different, so it would be a physically distinct function.) [12:02:31] This? (re @u99of9: And you want the top level html function to pull all of those bits out of the WD item and pass them all to the NLG config functi...) [12:14:53] Well, it could be… but in this case it already isn’t, because the countries and professions are arguments to the call whereas the dates are not. 🤔 I may reach a different conclusion once I get started, but {QID, map, language} is what I’m currently thinking. (re @u99of9: Oh, I thought the top level html call was still the simple QID one? (Then its implementation had to e [12:14:53] [12:14:54] xtract everything and pass ...) [12:23:28] Okay fair enough. It's certainly simpler if they're totally separate. Although if you're writing something from scratch, perhaps it could cover a whole new domain? (re @Al: Well, it could be… but in this case it already isn’t, because the countries and professions are arguments to the call whereas th...) [12:25:35] Such as…? (re @u99of9: Okay fair enough. It's certainly simpler if they're totally separate. Although if you're writing something from scratch, perhaps...) [12:25:37] The split of occupations and countries was because I think it's an editorial decision how many and which occupations (and countries to some degree) should be in the first line of an article, whereas the editor shouldn't get to pick the birth date of their choice. (re @Al: Well, it could be… but in this case it already isn’t, because the countries and professions are [12:25:37] arguments [12:25:37] to the call whereas th...) [12:30:17] A similar but distinct one would be: short description for person. Returning a string/monolingual, so it dodges some of your HTML concerns. (re @Al: Such as…?) [12:35:55] Or one for generic populated places: "*Khollar* (Persian (https://en.wikipedia.org/wiki/Persian_language): خلار, also Romanized (https://en.wikipedia.org/wiki/Romanize) as *Khollār*)[1] (https://en.wikipedia.org/wiki/Khollar#cite_note-1) is a village in Hamaijan Rural District (https://en.wikipedia.org/wiki/Hamaijan_Rural_District), Hamaijan District (https [12:35:55] //en.wikipedia.or [12:35:55] g/wiki/Hamaijan_District), Sepidan County (https://en.wikipedia.org/wiki/Sepidan_County), Fars province (https://en.wikipedia.org/wiki/Fars_province), Iran (https://en.wikipedia.org/wiki/Iran)." from [[:w:Khollar]] (re @Al: Such as…?) [12:37:23] Yes, I’m not criticizing that approach. It should be possible for all the data to be provided by contributors even if the community prefers to avoid it when the data is available on Wikidata. It’s probably preferable to provide distinct arguments where the editorial view is expected. 🤔 I’m not starting today in any event. (re @u99of9: The split of occupations and countri [12:37:23] [12:37:24] es was because I think it's an editorial decision how many and which occupations (and countr...) [12:41:22] "It should be possible for all the data to be provided by contributors even if the community prefers to avoid it when the data is available on Wikidata." I'm not sure. I feel that allowing custom dates of birth etc is inviting difficult vandalism to spot. When it's a question of fact, I think we are best directing the arguments to Wikidata. Similar to the way [12:41:22] copyright questions [12:41:22] are all sent to Commons. [12:43:10] No rush of course, but I'll certainly be interested. I was amused by suzuneu's description of his method as "trial and error". (re @Al: Yes, I’m not criticizing that approach. It should be possible for all the data to be provided by contributors even if the commun...) [13:18:10] I’m not sure how hard it would be to spot. The AW community could lock it down, I suppose, or the function could expose the provenance (or its lack) more violently. Not this week’s problem, I feel, but “completing the map” could include some validation of supplied values, such as disallowing P569, maybe just for a Q5. (re @u99of9: "It should be possible for all the data t [13:18:10] [13:18:10] o be provided by contributors even if the community prefers to avoid it when the data i...) [13:20:25] …or the intended article as a test case within test-driven development? (re @u99of9: No rush of course, but I'll certainly be interested. I was amused by suzuneu's description of his method as "trial and error".) [15:57:11] Great suggestions; thanks! I'll be happy to have a look at that (but cannot promise that it will become a top priority right away). I will create a ticket with initial thoughts. (re @u99of9: Yes, I agree that's making the best of the current state. Lexemes are unlikely to ever be as long as the longest items (although...) [15:58:43] Great suggestions; thanks! I'll be happy to have a look at that (but cannot promise that it will become a top priority right away). I will create a ticket with initial thoughts. (re @u99of9: Yes, I agree that's making the best of the current state. Lexemes are unlikely to ever be as long as the longest items (although...) [18:19:48] not to bring mood down, but maybe someone can tell me what the timeframe for 'The future of Abstract Wikipedia'? I'm sure I'm blind, but I don't see any date for when a decision should have been made. [19:00:26] Fingers crossed for at least a couple of decades. (re @wmtelegram_bot: not to bring mood down, but maybe someone can tell me what the timeframe for 'The future of Abstract Wikipedia'? I'm...)