[00:40:47] Feeglgeef , you've created https://www.wikifunctions.org/view/ru/Z22714 , and I wanted to check something: you know that this is _not_ how "an orange orange" is written in Russian, right? :) [00:41:21] It google translated to something along those lines [00:42:41] Not a chance. [00:49:21] Needless to say, you are not supposed to ever, ever use Google Translate for anything serious without actually knowing the language. But I'm pretty that this is not how you got that string in this case. [00:49:56] i translated the string back [00:50:09] And it seemed reasonable enough [00:50:25] Well, now you know that you are not supposed to do that either 🤷🏻‍♂️ [00:50:42] What does it mean in Russian? The output? [00:51:05] "a carrot-colored orange" [00:51:33] Oh [00:51:42] So basically the same thing? [00:51:55] It's not horribly wrong, but it's not something that anyone is likely to say. [00:52:06] So it's not a useful output. [00:52:34] Neither is "an orange orange" 😏 [00:52:48] True 🤷🏻‍♂️ [00:53:04] And this is arguable Wikidata's fault anyway [00:53:36] I'd argue "carrot-colored" and "orange" are different concepts [00:54:03] That probably happened because there are (at least) two Russian lexeme senses that link to the "orange" item. One is the word that people are likely to use, and the other literally means "carrot-colored". It's a correct word, but it's not so common either. [00:54:41] Also true. (re @Feeglgeef: I'd argue "carrot-colored" and "orange" are different concepts) [00:57:07] Wait, are they basically the same thing, or different concepts?? The English versions sound like synonyms to me. But the general issue of prioritising the more common lexeme when two point to the same thing seems worth exploring further. [00:58:30] Perhaps statement rank? The new function would then have to order them by this rank? (re @u99of9: Wait, are they basically the same thing, or different concepts?? The English versions sound like synonyms to me. But the general...) [00:59:36] I thought the link came from the lexeme to the QID. So at both of these lexemes it is likely to be normal rank. [01:00:21] Well, the rank of the statement is still weaker if it's a worse lexeme [01:01:21] Choosing something with colors as one of the first demos for natural language generation is probably too ambitious, given that color naming is one of the most prominent examples of semantic diversity across languages. [01:01:55] Define "worse" :) (re @Feeglgeef: Well, the rank of the statement is still weaker if it's a worse lexeme) [01:04:39] So, is this like saying "a peach nectarine"? [01:07:58] I don't think we should seek to change the Wikidata definitions of statement ranks. Hopefully we can work it out some other way. (re @Feeglgeef: Well, the rank of the statement is still weaker if it's a worse lexeme) [01:08:36] Yes, kind of. (re @u99of9: Is it just that the word is less common, or is it like saying "a peach nectarine", or can you give us a sense of why people are ...) [01:09:57] That kind of error sound hard to avoid without a statistical language model... (re @amire80: Yes, kind of.) [01:13:13] But I actually started this whole discussion for a different reason. [01:14:03] I was trying to check if I can do "а white horse" in Russian, and it didn't work. [01:14:50] It works in English. There are Russian lexemes, senses, and forms. Russian is one of the languages that explicitly appear in the function configuration. And yet, it returns void. [01:15:49] https://www.wikifunctions.org/view/en/Z23296 [01:27:17] I think that I never got a response about this. @vrandecic , David , perhaps you know? (re @amire80: There's a new translatable message: "Pages with calls to Wikifunctions that result in a systems error") [03:35:08] It looks like there are some problems with the implementations. I'll look into this. (re @amire80: It works in English. There are Russian lexemes, senses, and forms. Russian is one of the languages that explicitly appear in the...) [03:45:52] I'm not sure why the code implementations of Z19295 aren't working. I've disabled them. [03:54:01] Can you put some appropriate tests in Z22710? It is failing because L1418847 does not have any masculine forms, but the grammatical gender of the noun L144878 is masculine. (re @amire80: It works in English. There are Russian lexemes, senses, and forms. Russian is one of the languages that explicitly appear in the...) [05:27:32] By the way, for "horse", I think Q726 is better than Q27022 but at the moment they are both linked from L1124, so all good for now until we get a lexeme for Equus. [05:28:10] L1123 I mean! [06:44:44] Second column in Nature correspondence: https://www.nature.com/articles/d41586-025-00715-x.epdf?sharing_token=U338r03QHWCofNpjb1INE9RgN0jAjWel9jnR3ZoTv0NPTFDjB5wCGCTUGo2ItO05_W6A_Hg1spWt6ondsfuMeE-_5Ime8GSTPSk3lMvu6YGz6pA3aSitWaswYWUDAKAMEK3m9D4a97ygwMJEPNepLaVCZ-fBitXeUH32KFMpN6s%3D [09:21:44] I’m inclined to agree, but… the Wikidata community might consider linking sense to sense before linking sense to item. This would mean that when “hound” (L6419) means “dog” (L1122) the sense of “hound” (L6419-S2) links to the sense of “dog” (L1122-S1 or L1122-S2) that links to the item for dog (Q144). This would not only reduce the discovery of variants when message> [09:21:44] moving from item to lexeme(s) but would also support the discovery of variants when beginning at a core lexeme sense (realising “fast hunting dog”, say, as “hound”). (re @u99of9: I don't think we should seek to change the Wikidata definitions of statement ranks. Hopefully we can work it out some other way.) [09:35:24] Well, so here we have a problem that is a similar problem to the "carrot-color" that I mentioned above. The word at L144878 (конёк) means something like "small male horse": "male horse" with a diminutive suffix. I cannot think of good English word or Wikidata item as a translation for that Russian word. (And no, "stallion" or "colt" wouldn't be a good translation; [09:35:24] there's an [09:35:25] other Russian word for that.) It's really not the first word that Russian speakers would think about as a translation for horse, but it's the first word that Wikifunctions' current algorithm finds. Most Russian speakers would think of L1418261 (лошадь) or maybe L1418932 (конь). (re @u99of9: Can you put some appropriate tests in Z22710? It is failing because L1418847 [09:35:25] doe [09:35:26] s not have any masculine forms, but the grammati...) [09:50:16] I think that’s another good example of what I was saying. It makes more sense to link a diminutive to the more basic sense than to a more general item. (re @amire80: Well, so here we have a problem that is a similar problem to the "carrot-color" that I mentioned above. The word at L144878 (кон...) [09:51:12] I think the lexeme community already are in consensus that sense to item is preferred and sense to sense is only a fallback when there is no item. The problem with sense to sense is that we essentially are creating the old style interwiki all to all mesh (mess) again. (re @Al: I’m inclined to agree, but… the Wikidata community might consider linking sense to sense [09:51:12] before link [09:51:13] ing sense to item. This woul...) [10:01:23] It’s hard to model a mess without making a mess! The case here is that Q144 is the item for a more general sense than L6419-S2, which should probably link instead to Q38923. But when it comes to diminutives or sex/gender variants, it seems more appropriate to link the senses to the core (neutral) sense and link that sense to the item. (re @Jan_ainali: I think the lexeme [10:01:23] communi [10:01:23] ty already are in consensus that sense to item is preferred and sense to sense is only a fallback when...) [10:18:10] Possibly an easier question: [10:18:34] How do you know that this is the reason? (re @u99of9: Can you put some appropriate tests in Z22710? It is failing because L1418847 does not have any masculine forms, but the grammati...) [12:08:14] I looked into Z22713. In particular this bit. It gets the adjective string from the set of representations that are singular *and* have the first grammatical gender listed in the noun. The first listed in the noun is masculine. The adjective does not have any masculines. Hence this bit fails to return. : [12:08:14] https://tools-static.wmflabs.org/bridgebot/b2314c76/file_69133.jpg [18:22:49] I agree that "system error" is preferable to "systems error", but only slightly. I feel like the difference is so minimal that changing it in the code base is unwarranted (but I don't have much experience regarding translation procedures/expectations). (re @amire80: I think that I never got a response about this. @vrandecic , David , perhaps you know?) [19:48:23] There are nine occurrences of "system error" in messages related to Wikimedia, and one occurrence of "systems error". If they mean the same and "system error" is at least slightly preferable (and I think that it's preferable, too), then a change is definitely warranted. (re @David: I agree that "system error" is preferable to "systems error", but only slightly. I [19:48:23] feel like the d [19:48:23] ifference is so minimal that ...) [19:49:16] Now that I look at it a bit more, it should probably also be consistent with `"wikilambda-functioncall-error-unclear": "Function call error: something went wrong with the system"` [20:22:32] okay, I've just submitted that change ( systems -> system) in the WikiLambda extension code base [20:25:02] Thanks! [23:53:02] Why do we have one "Type" called Error and another Type called "Error Type"? This seems bound to cause confusion. Z5 Z50.