Talk:Community Wishlist/Wishes/Wikidata recent changes patrolling per language

This page is for discussions related to the Community Wishlist/Wishes/Wikidata recent changes patrolling per language page.   Please remember to: Sign posts using the four tildes (~~~~) Remain civil and polite during discussions. Place new text under old text (start a new post).

More than one tool

In February, I made a summary of features supported by the three tools out there ([1][2][3]) that can assist in language-based Wikidata patrolling: d:Wikidata:Project_chat/Archive/2025/02#c-Matěj_Suchánek-20250202122800-M2k~dewiki-20250202102300. It would definitely be nice to have only one that rules them all. --Matěj Suchánek (talk) 11:52, 10 March 2025 (UTC)

Solution ideas

Feedback

the Wikidata recent changes pages would be good to link this to https://www.wikidata.org/wiki/Special:RecentChanges so people know what is meant. By this wish I found out that there is a help site for the recent changes for all the Wikimedia projects except for Wikidata (see d:Q3234746), shouldn't one be created for it as well?

I don't think it's feasible or efficient or effective for humans to check this page. Not once I have I think you meant to say "Not just once have I" or not? Or did you mean that the ways you have used so far did not enable you to discover such?

Broader solution concept

(at scale & effective); vandalized short descriptions that were lingering for long time periods Good subject and I think the solution to it is not relying or furthering the need for hugely time-intensive manual human labor but what I described here (and note that scarce volunteers can then use their time for more impactful other things at Wikidata and elsewhere and to improve upon what I suggested there like adjusting the short descriptions): Requested gadget: autotranslate aliases & desc. to spot vandalism. That's the early proposal – I think better instead or in addition to a gadget would be to have:

• machine translation (using latest MT tech which works near perfect for short texts like labels and short descriptions for many languages) set all missing labels and description for items which get a flag as being a machine translation that gets removed if users either click on its approve button or adjust the text
• using machine translation and NLP to identify labels and descriptions that are inconsistent (example: English label is bottle, German label is Flasche, machine translation for Spanish is botella, but the set label value for Spanish is taza (machine translated: cup) – these are then flagged for review, ultimately resulting in nearly all flaws and vandalism to be fixed

There are some problems or challenges here of course. Maybe the biggest one thereof is that when the source description or label changes, should the MT and the MT and then approved/adjusted ones be changed as well – maybe the newer version is much better and setting the MT text of a prior version would unduly favor early labels/descriptions? It would be a massive timer-saver, improve the multilingualism, and make Wikidata much more reliable.

Cause analysis

There are a lot of faulty and vandalized labels, aliases and descriptions in Wikidata (not just the descriptions). This is in part because there's so many items nobody can all continuously patrol effectively, there's fewer active editors, there's less motivation since items are not read a lot by people unlike Wikipedia, as with Wikipedia a 'since last seen'-diff button is missing in the Watchlist so people may only look at the edit summary or the latest edit, and because of course people can't read the label in another language to be able to verify (also see the linked proposal above which was initially mostly about having a machine translation displayed next to changes of languages one doesn't understand but that's probably not the efficient effective approach one could take here).

Narrow solution concept

Also relevant is this tool which one can use to have the Wikidata Watchlist hide changes to labels, aliases and descriptions not in the language(s) one can read: d:User:Lectrician1/filter-watchlist-languages.js. Considering how few people use that script, I think it would be good to make this some native watchlist filter functionality. It drastically improves efficiency and reduces the time-sink of WD Watchlist checking. Maybe it could be developed further so it also works on the recent changes page albeit a native filter would be better. Prototyperspective (talk) 13:49, 10 March 2025 (UTC)

@Prototyperspective: By "Not once I have discovered vandalized short descriptions that were lingering for long time periods." I meant that this happened multiple times.

I understand your concern concerning the inability of Wikidata patrollers to patrol edits in languages they do not know, but honestly, machine translation is usually bad for one-two words or other short word strings like short descriptions and aliases. I lean more towards providing native speakers with the necessary tools.

What irks me is that I cannot easily find and patrol edits in languages I know. The existing tools have their flaws: edits that show up on https://wdpd.toolforge.org/by_lang do not necessarily show up on https://wdvd.toolforge.org/, but https://wdpd.toolforge.org/by_lang does not provide "Entity title" and "Edit summary" as https://wdvd.toolforge.org/ does. --Paloi Sciurala (talk｜contribs) 14:35, 10 March 2025 (UTC)