[04:32:55] I can confidently say that Umar_Punjabi is the best investment coach out there. With his mentorship, I earned $50K — and I couldn’t be more grateful! [04:34:00] Reported [11:45:16] Hello everyone. [11:45:17] Concerning Multilingual Punctuations, I have made a full analysis of the scholarly publications I found using ChatGPT and other LLMs and I compiled this list representing all the concepts, reflecting punctuation usages. [11:45:59] https://docs.google.com/document/d/1Lbh0cWD6CjDBKgax0YUMXmX2TG6Yj_Gh7DqV5w2wm3s/edit?usp=sharing is the link. [11:47:23] According to my research, the way I propose to do it is accurate. Places do not differ. Just the fact that a usage exists in a language or the used char differs from a language to another one. [11:47:56] When a usage does not exist, the function can be void. [11:50:17] Concerning the point about that using a function of multilingual punctuations with spacing can alter how references are represented, as this means that a reference can come after a space, I can say that a reference can come before punctuations and this solves all the problem. [11:50:29] In fact, this is applied in French Wikipedia. [11:53:30] However, I think that implementing this would cause a lot of conflict with the developments that are ongoing. I leave all this to NLG SIG so that they can make a final decision. It is up to them to adjust my playground (abstract:Q15433043) and make the implementation for Z33449. [11:55:59] They can also adjust the examples for Z33449 for a better final output. [11:57:28] References should not come before punctuation in English. (re @Csisc1994: Concerning the point about that using a function of multilingual punctuations with spacing can alter how references are represen...) [11:58:35] We are talking about a multilingual project. We should see what all languages do and decide. (re @u99of9: References should not come before punctuation in English.) [11:59:19] I think we should make output that suits all languages. Moving a full stop either before or after a reference is easy BTW. (re @Csisc1994: We are talking about a multilingual project. We should see what all languages do and decide.) [12:02:56] French: Before [12:02:56] English: After [12:02:58] Arabic: After [12:02:59] Spanish: After [12:03:01] Italian: Before [12:03:02] Portuguese: After [12:03:04] Dutch: After [12:03:05] Japanese: Before [12:03:07] German: After [12:03:08] Russian: Before [12:03:10] Danish: After [12:03:11] Turkish: After. [12:05:11] Yes, so in the long run we should be able to provide either (based on language configuration). (re @Csisc1994: French: Before [12:05:13] English: After [12:05:14] Arabic: After [12:05:16] Spanish: After [12:05:17] Italian: Before [12:05:19] Portuguese: After [12:05:20] Dutch: After [12:05:22] Japanese: Before [12:05:23] Germa...) [12:05:29] That will add to the complexity of the thing. Instead of always inserting references before and implementing punctuations with spacing, the references should be embedded inside the function so that it can be put before or after the punctuation and before the spacing. This requires a lot of compilation time. (re @u99of9: I think we should make outputs that suit all [12:05:29] languages. Movi [12:05:29] ng a full stop either before or after a reference is easy BTW.) [12:05:36] Since you've looked this up, maybe make a function that returns either before or after based on langauge. [12:06:46] There are other ways this could work. For example, the punctuation could be moved after production of the sentence+ref. (re @Csisc1994: That will add to the complexity of the thing. Instead of always inserting references before and implementing punctuations with s...) [12:08:25] Sounds legit. As I said, I leave it to NLG SIG. It is beyond my taste. It needs a consensus. If they judge my work as fair, they can resume it. (re @u99of9: There are other ways this could work. For example, the punctuation could be moved after production of the sentence+ref.) [12:08:34] I will implement this. (re @u99of9: Since you've looked this up, maybe make a function that returns either before or after based on langauge.) [12:09:17] I'm not sure what the return type should be. Maybe a Boolean would be simplest? (re @Csisc1994: I will implement this.) [12:10:22] Boolean. After is FALSE. Before is TRUE. (re @u99of9: I'm not sure what the return type should be. Maybe a Boolean would be simplest? "Are references before punctuation?") [12:13:12] On svwiki it depends if the reference support the entire sentence or just the last clause. (re @Csisc1994: Concerning the point about that using a function of multilingual punctuations with spacing can alter how references are represen...) [12:13:31] With the new gradual test-running, I'm concerned that WikiLambdaBot is not doing its job as well as it used to. I just added a superior implementation to Z29825, and after running all the tests to make clear that it was better, it still didn't get promoted, so I've ended up disabling the previous implementation. [12:13:58] That is interesting. (re @Jan_ainali: On svwiki it depends if the reference support the entire sentence or just the last clause.) [12:14:15] can you show us an example of when it is just referencing the last clause? (re @Jan_ainali: On svwiki it depends if the reference support the entire sentence or just the last clause.) [12:20:30] Another point I have to share again is that I compiled a list of concepts corresponding to multilingual prepositions at https://docs.google.com/document/d/1OmmuE5z55biIg5r5W-K6Vi5fPeJnTdQYQYtzLNBH9UQ/edit?usp=sharing. It is up to NLG SIG to decide what to keep and to adjust my playground (abstract:Q15433043) as well as the examples of Z33451. They can also adjust [12:20:30] the examples at [12:20:31] Z33451 for a better final output of this work. I found that I could not implement this by myself, as well, because I can do it and it renders something that is controversial. So, I leave my progress to NLG SIG. [12:22:42] I think that my work ends here. All the best on developing Abstract Wikipedia, as it should be. I will implement the "reference after or before punctuation" function. [12:23:44] ?? I was hoping your work would never end here! (re @Csisc1994: I think that my work ends here. All the best on developing Abstract Wikipedia, as it should be. I will implement the "reference ...) [12:25:25] I need to take a sabbatical after Wikimania. I will probably be back but not in the coming months. Health concerns. Thank you. (re @u99of9: ?? I was hoping your work would never end here!) [12:25:58] With respect, the NLG SIG is not an “authority”. The community proceeds by some general notion of consensus, as in any Wikimedia project. 😎 (re @Csisc1994: However, I think that implementing this would cause a lot of conflict with the developments that are ongoing. I leave all this t...) [12:26:42] I know. But, at least they are not one person. They deeply think on the matter. (re @Al: With respect, the NLG SIG is not an “authority”. The community proceeds by some general notion of consensus, as in any Wikimedia...) [12:27:06] Let us see if they can play with my input. [12:35:56] Done, Z36365. (re @u99of9: Since you've looked this up, maybe make a function that returns either before or after based on langauge.) [12:37:21] Lest we forget… [12:37:22] “Reference tags are placed _before_ dashes, not after. If a footnote applies only to material within parentheses, the tags belong just before the closing parenthesis.” [12:37:23] [[en:mos:refpunct]] [12:37:25] I suspect the first part of the exception applies more generally to any punctuation that is preceded by a space, and the second part applies to closing brackets of all kinds. (re @u99of9: References should not come before punctuation in English.) [12:38:46] [[en:MOS:REFPUNCT]] (re @wikilinksbot: [[en:mos:refpunct]]) [12:39:34] maybe this is specifically about the end of a sentence relative to the period? (re @Csisc1994: Done, Z36365.) [12:40:15] Not really. Works also with comma, semicolon, and colon. (re @u99of9: maybe this is specifically about the end of a sentence relative to the period?) [12:40:47] Sometimes, closing parenthesis. [12:41:53] Yes, it will often be transferrable. But to make a clear judgement for every language, we need to ask just about one location? (re @Csisc1994: Not really. Works also with comma, semicolon, and colon.) [12:42:48] I will change it to full stop. (re @u99of9: Yes, it will often be transferrable. But to make a clear judgement for every language, we need to ask just about one location?) [12:42:58] Reference before full stop. [12:43:25] Then, if someone would like to replicate it, it is up to him. [12:46:07] Done. (re @Csisc1994: I will change it to full stop.) [12:48:50] Al were you at or have you listened to the VC recording? Your answers to the Eiffel Tower question sound a bit different to Denny's (personal capacity). [12:56:30] Technically, I think language is just a proxy for “style-guide” here, so “en” means [[en:MOS]]. In that style, the explicit general rule is that references always follow punctuation (and, implicitly, punctuation clusters) unless an exception applies. For any language tag corresponding to a Wikipedia edition, I would expect that edition’s style to be followed, whether or [12:56:31] [12:56:31] not it is generally applicable for that language. That is tentative pragmatics, however! (re @u99of9: Yes, it will often be transferrable. But to make a clear judgement for every language, we need to ask just about one location?) [12:57:04] /delete@wikilinksbot [12:57:19] [[en:WP:MOS]] (re @Al: Technically, I think language is just a proxy for “style-guide” here, so “en” means [[en:MOS]]. In that style, the explicit gene...) [12:58:09] Neither. I aim to do so before Tuesday’s NLG SIG. (re @u99of9: Al were you at or have you listened to the VC recording? Your answers to the Eiffel Tower question sound a bit different to Denn...) [12:59:05] Absolutely. If AW articles get included in a language wiki, they need to match its MOS. (re @Al: Technically, I think language is just a proxy for “style-guide” here, so “en” means [[en:MOS]]. In that style, the explicit gene...) [12:59:59] Please feel free to raise my work as well as the function I have just created to NLG SIG. I probably missed some languages for Z36365. They can do a better work identifying all languages. (re @Al: Neither. I aim to do so before Tuesday’s NLG SIG.) [13:02:08] Yes, I’m sure that will come up. (re @Csisc1994: Please feel free to raise my work as well as the function I have just created to NLG SIG. I probably missed some languages for Z...) [13:02:21] I envision [[abstract:Abstract:MOS]] as for content creators, and then function authors should be mindful of their relevant language's style guide. (re @u99of9: Absolutely. If AW articles get included in a language wiki, they need to match its MOS.) [13:06:41] I read most of https://en.wikipedia.org/wiki/Wikipedia:Village_pump_(WMF)#Abstract_Wikipedia_goes_somewhat_live_and_already_hosts_unattributed_enwiki_copies... today. Quite cynical, but also plenty of valid criticisms. Although we are in beta, and are testing all kinds of things out, we ought to be measured, and aim for quantity over quality. [13:09:18] we did intervene and said that we will take action on some of the criticism [13:09:31] we are in fact analysing it and seeing what we can do [13:10:31] that also gives me the possibility to say my cue which is.,, [13:11:56] Hi all! I wanted to share that the *next meeting of the NLG SIG (Natural Language Generation Special Interest Group)* will be on June 16, at 16:00 UTC (https://zonestamp.toolforge.org/1781625600) on Google Meet (https://meet.google.com/xzn-kqer-mah). [13:11:56] Denny will lead a discussion on the new NLG types (https://www.wikifunctions.org/wiki/Special:MyLanguage/Wikifunctions:Status_updates/2026-05-15#News_in_Types:_way_forward_for_natural_language_generation_types). Everyone is invited to join! [13:12:00] (re @Sannita: Hi all! I wanted to share that the next meeting of the NLG SIG (Natural Language Generation Special Interest Group) will be on J...) [13:12:16] Thanks, I thought your input was helpful there. Many of the problems they identify are just because our functions are not sophisticated enough yet (by far), so it's not a tech issue, but a community speed issue. Which is my cue to say: [13:15:06] My request: Once you have the tech ready to host AW articles on local language wikis, please don't enable it until we have had a thorough discussion on AW/WF. We need time to build good content before it is shared widely at a substandard level, because otherwise it will backfire for the reputation of AW. (re @u99of9: Thanks, I thought your input was helpful there. [13:15:06] Many of the pr [13:15:07] oblems they identify are just because our functions are not sophis...) [13:16:18] This, please! There was a bit of a mention in the monthly WMF update, please wait for a community consensus. (re @u99of9: My request: Once you have the tech ready to host AW articles on local language wikis, please don't enable it until we have had a...) [13:21:49] If tech does end up far ahead of us in this regard, please focus on any tech that is stopping us from making good fast functions. For example, T427644 causes problems for functions I rely on, and many times a week I dig down from an error message, and find it messing me up. But it is marked medium. [13:23:26] And if this is right, this new issue will cause many more problems, because we will often fail to promote the most accurate and fastest implemetations. (re @u99of9: With the new gradual test-running, I'm concerned that WikiLambdaBot is not doing its job as well as it used to. I just added a s...) [13:27:27] This has been a problem I've noticed, a focus on moving on instead of fixing critical issues with existing stuff. Of course, there are KPIs or whatever that the team need to meet, but I'd rather have a complete product before moving on, rather than many buggy features. (re @u99of9: If tech does end up far ahead of us in this regard, please focus on any tech that is [13:27:27] stopping us fr [13:27:28] om making good fast functions...) [13:46:43] I think I broadly agree with both Richard and Denny, but I am more optimistic than Richard in believing that we can “generally” identify the [tower] or [battle] item as a semantic (language-neutral) head. In any event, we must have an approach that works for the case where there is no lexeme, rather than imagining that we shall have a lexeme for each item. Conceptually, I [13:46:43] ref [13:46:44] er to that as constructing lexemes rather than fetching Wikidata lexemes. This has the desirable consequence that functions can operate in the same way, whether the lexeme representations are fetched (and probably projected) or constructed. (re @u99of9: Al were you at or have you listened to the VC recording? Your answers to the Eiffel Tower question sound a bit different [13:46:44] to Denn [13:46:46] ...) [13:48:00] Why does https://abstract-data.toolforge.org/article/Q3130 think that Sydney Opera House doesn't have a label in English?? : https://tools-static.wmflabs.org/bridgebot/9263a791/file_80976.jpg [13:48:55] Probably due to being generated by a slop-machine or something (re @u99of9: Why does https://abstract-data.toolforge.org/article/Q3130 think that Sydney Opera House doesn't have a label in English??) [13:56:42] See for example note 14 and 15 in https://sv.wikipedia.org/wiki/Klimatf%C3%B6r%C3%A4ndring_i_Sverige?wprov=sfla1 (re @u99of9: can you show us an example of when it is just referencing the last clause?) [13:59:39] Thanks. 14 is where I would expect. I understand the reasoning for 15, but it still looks strange to me! Anyway, I think we should have a "move reference before fullstop" postprocessor function available for when languages want it. (re @Jan_ainali: See for example note 14 and 15 in https://sv.wikipedia.org/wiki/Klimatf%C3%B6r%C3%A4ndring_i_Sverige?wprov=sfla1) [14:02:06] I've been wondering about other postprocessors too. Perhaps we should generate HTML with links to all referred QIDs, then have a function which strips the ones we nominate. Or we should generate all these sentences starting with the subject "Australia is X. Australia is Y. Australia is Z." then have a postprocessor wrapping the second two sentences which replaces with [14:02:06] a pronoun i [14:02:07] f possible in the language. [14:04:22] We have so much more to do at the ball tonight, but I am about to turn into a 🎃 [14:08:06] Oh, one last thing. Why do I sometimes get "ZID not found"? It works here : https://tools-static.wmflabs.org/bridgebot/88d8837c/file_80977.jpg [14:08:09] Yes, it’s almost irresistible… and bear in mind that different sources may be preferred according to the language, which is a good argument for mediating references via Wikidata, especially where the sources are properly secondary (generally accepted facts at school textbook level). (re @u99of9: I've been wondering about other postprocessors too. Perhaps we should [14:08:09] generate HT [14:08:10] ML with links to all referred QIDs, then have a...) [14:42:15] We do it to be clear that 15 doesn't support the claim 14 supports. (re @u99of9: Thanks. 14 is where I would expect. I understand the reasoning for 15, but it still looks strange to me! Anyway, I think we shou...) [15:04:27] I wouldn't replace pronouns at post-process level. Sometimes pronouns are very weird, and it is necessary to know them at early stages of languages generation (re @u99of9: I've been wondering about other postprocessors too. Perhaps we should generate HTML with links to all referred QIDs, then have a...) [23:14:23] why is wikifunctions so cyclic, it's messing my brain up [23:16:56] Can you show us an example you're confused by? (re @wmtelegram_bot: why is wikifunctions so cyclic, it's messing my brain up) [23:17:33] Z4 [23:17:48] it's a type whose type is type and whose first key is a type?? [23:34:17] I'm not the best to explain this! (re @wmtelegram_bot: it's a type whose type is type and whose first key is a type??) [23:50:11] Every object’s first key is its type. Some objects are types; their type is Z4. Z4 is a type, so its type is Z4. For a full explanation, please see [[Wikifunctions:Function model]] (re @wmtelegram_bot: it's a type whose type is type and whose first key is a type??) [23:51:14] Function model has become my favorite page lately [23:51:59] Random question but how do you get a Z523? I tried many different things but it eithers fails to normalize or just refuses to evaluate... [23:52:39] Please feel free to raise detailed questions on its talk page. (re @wmtelegram_bot: Function model has become my favorite page lately) [23:57:10] You’d have to do something illegal. It’s not an error I’ve ever encountered and it might well be technically impossible. (re @wmtelegram_bot: Random question but how do you get a Z523? I tried many different things but it eithers fails to normalize or jus...) [23:59:38] checked both the repo for evaluator and orchestrator, no mention of Z523. guessing it was created early in the development process and later replaced with other, more generic errors.