Talk:Como

Suggestions for Features

Here you can list some ideas for features you want to see. Kristbaum (talk) 20:33, 16 June 2024 (UTC)

Gnom's test

Hi, I recently tested the tool, and I can say the following:

• Usability was super easy (but I am also a longtime Wikimedian, so maybe that does not say much).
• As someone who as worked as a translator, I was a bit worried about being prompted to work on some translations: How do we translate the German-language term 'Stolperstein'? Is an 'Unterfamilie' really a 'sub-family'? I could already here the biologists yelling at me for creating additional work for me to fix my errors...
• Have you been in touch with Magnus Manske about this tool? It reminded me a lot of Mix'n'Match.

Gnom (talk) 07:41, 12 September 2024 (UTC)

Feedback from User:Spinster

Thanks for creating this tool! I am an art historian and long-time Wikimedian and an active Wikidata contributor. This will influence my input here.

• I really like the flexibility of the tool, and its basic design to allow people to create smallish games. It reminds me of the Wikidata game, but as I understand it, it also allows to work with structured data on Wikimedia Commons, which is great! Does it allow editing other Wiimedia wikis too?
• In that regard, I think it would be awesome if the creation of new games would still be made simpler, i.e. not require knowledge of JSON or other programming languages.
• Your three test games (Art describer, Deepl mistakes, ArtHist triplets) are all about checking AI input, and I will comment about that below. I would also include examples for games that don't work on AI but simply on Wikidata or SDC queries. I currently have one idea for that which I will list below.
• To confidently make decisions in a game like this, a bit more context is usually needed, for instance metadata about the work you are looking at, the description of a Wikidata item, the option to click through to an image, or the option to read an entire text. It will make the game interface more complex but will encourage better quality contributions.
• In yes/no/skip types of games, it would be good to have an option that actually means "skip and never show this suggestion again because it's impossible to solve"

Game-specific feedback

In general I am not super fond of the AI games you show as a test, for various reasons. I think AI can help Wikimedians but it needs to be well chosen. Let's think about better cases. In general I think Como games should aim at already high-quality starting data. With other Wikidata and Commons gamification and crowdsourcing platforms, there have been numerous (way too many) cases where less experienced users use the game platform blindly and do approve sometimes questionable or incorrect statements, resulting in a lot of wrong and dirty data. This should very much be avoided.

• ArtHist Triplets checker:
  • We should preferably not use Wikipedia as a source for Wikidata statements, so I'm not a fan of this game. It would be another thing entirely if other reliable datasets would be used as a source of Wikidata statements, of course.
  • The quality of the data to be approved isn't great. Many of the triplets I get are very generic (woman instance of human) and don't pertain to the work specifically.
  • Context is missing to be able to check if the statement is indeed correct and if the article actually mentions the statement.
    • Only a snippet of the Wikipedia article is shown and it's often impossible to read if the statement is indeed present in the article.
    • Wikidata items in B) and D) should be clickable and should be shown with descriptions in order to be able to verify that one is indeed working with the right items.

• Deepl translation checker:
  • Fully agree with Gnom's assessment above. I would not use this game for translations that require field knowledge, e.g. names and terminology of taxons. And I was also very unsure about the correct translation for Stolpersteine. I think translation of Wikidata descriptions can quite easily be done otherwise - by someone knowledgeable who uses a tool like QuickStatements or OpenRefine for large batch translations instead.

• Art describer:
  • I assume this game produces captions for Wikimedia Commons?
    • The game mixes up visual description (which rather belongs in Alt text) and basic metadata about the work (the creator). I am not the biggest fan of that.
    • As an art historian, I would actually use captions to write the title, year, creator, possibly collection. Something like: Jan Both: Landscape with Morra Players (The Small Limekiln) (circa 1637), collection Museum of Fine Arts, Budapest
    • And visual description ("a landscape with trees and a shepherd with five sheep") should go in an Alt text SDC statement. Alt text is very context-dependent by the way; this is a tricky subject that has already been discussed a lot on Wikimedia Commons and (AFAIK) not with a very solid outcome.
  • In order to be able to confidently approve a description, I would need to see the basic metadata about the work (at least its creator, title, creation date). I would also like to click through to the Commons file page to read more metadata.

Spinster (talk) 07:53, 28 September 2024 (UTC)