[00:26:57] yes. As Mahir says, I am hesitant with putting an ML model unsupervised into the pipeline, but I do imagine a workflow that allows to put in natural language text and then the model tries to find possible Abstract contents that might capture that natural language. I have the following UX mock up of that: https://commons.wikimedia.org/wiki/File:Abstract_wikipedia_mockup.png [00:27:27] that could be either in Abstract Wikipedia itself, or, as you suggest, in Toolforge. The important think here is that nothing here happens automatically, but always has a human in the loop. [00:29:15] The way Jan mentioned is a parallel way to create stubs quickly for categories of articles. I think that's an interesting possibility, but will be in parallel to hand-written individual articles, likely using the code for the stubs for parts of its own content. [18:46:54] Interesting concept but we'd need some way to tag these articles as autogenerated and make it opt in imho. This would have to be from the start, as communities don't necessarily like autogenerated stubs and we'd poison the well of adoption if things are done w/o local consensus [18:48:14] That is already the plan. (re @Spook: Interesting concept but we'd need some way to tag these articles as autogenerated and make it opt in imho. This would have to be...) [18:48:46] Maybe per subject area as well? (re @Jan_ainali: That is already the plan.) [18:49:10] I think it has to per article even. (re @Spook: Maybe per subject area as well?) [18:49:35] That might be tough since it would take a while to manually approve entire areas. [18:50:24] Though with the fact articles make statements in structured data it might be possible to write arbitrary conditionals on what kinds of articles to include [18:50:35] A local community could make a decision to approve all in one subject area (or perhaps per generation template) and then bulk import batches (re @Spook: That might be tough since it would take a while to manually approve entire areas.) [18:52:55] that'd be my guess. Though there might be a grey line around "auto generated" given that everything is code anyways and uses NLG. [18:53:10] NLG? (re @Spook: that'd be my guess. Though there might be a grey line around "auto generated" given that everything is code anyways and uses NLG...) [18:53:19] "natural language generation" [18:53:19] natural language generation [18:53:41] I don't understan what is a grey line? (re @Spook: that'd be my guess. Though there might be a grey line around "auto generated" given that everything is code anyways and uses NLG...) [18:53:44] What if I created a work saving template of some kind that made a bunch of statements at the same time? [18:54:08] Grey area I should've said. Mixing my metaphors up. [18:54:21] I still don't understand :) (re @Spook: Grey area I should've said. Mixing my metaphors up.) [18:55:48] They are clearly auto generated so if your local community is willing to approve such, it is white, and if not it is black. [18:57:21] let's say I'm writing an article on a topic that generally has the same structure among multiple articles [18:58:06] And rather than making the same series of statements for a bunch of articles, I cheat and create a function that creates a bunch of statements at once. [18:58:08] Yes, that is totally doable! And me, for one, encourage it (re @Spook: let's say I'm writing an article on a topic that generally has the same structure among multiple articles) [18:58:48] At what point does that become "auto generated"? [18:59:38] It becomes auto generated form even in the simplest form in my opinion, since it will be able in any language with enough high quality lexemes to render it [19:01:36] but would that count as a "bot created stub" for community purposes? [19:01:47] Depends on your community (re @Spook: but would that count as a "bot created stub" for community purposes?) [19:02:34] And even if it does, some communities accepts those. Embraces even. [19:02:45] Definitely, but how would one be able to make that distinction in a policy? [19:02:52] . (re @Jan_ainali: Depends on your community) [19:04:28] I'd imagine a community might agree on what kind of articles to accept and have people in charge of importing those articles. [19:04:37] I think it will become easier for a community to discuss these questions when they are not hypothetical but instead you can look at real examples. [19:04:47] Probably [19:05:21] That's why you should help out with gitlab.com/mahir256/ninai so that we can better generate such real examples! [19:06:12] Sounds fun. I will check it out and try writing something. [21:00:31] A new development in Ninai: world-modeling/context-building! : https://tools-static.wmflabs.org/bridgebot/203e2cf5/screenshot_2022_02_03_14_57_43_587436577.png [21:08:00] The example sentences in that image would have had @Jan_ainali as a "schwedischer Lektor" and Nikki as a "britische Vorbild", but three of those lexemes weren't available so I opted to describe নিকোলা and @WMYupik instead 😊 [21:29:05] Are you using P1549? [21:33:44] Yup! [21:34:08] (the "GeneralDemonym" constructors) [21:34:41] sorry, misread the property; I'm using "demonym of" [21:36:37] since ideally all "demonym" values should go on lexemes instead [21:39:39] yeah, that sounds like a better way to model it [22:44:52] You folks are great at answering questions! [22:48:31] So the plan is at least to have three options: opt in to each individual article, or opt in to all articles that have no article already and are not explicitly opt out. In addition, any article can incorporate parts of Abstract content. That should be consistent with the answers provided here. Thanks all!