- How will this change affect articles on philosophy, interpretive social sciences, and the humanities? Many articles in those areas could be improved with better citations, but some articles have fairly solid coverage without many citations beyond the few books that are being discussed. This article says that "opinions" need citations while "book plots" do not. Where do interpretive synopses of complex works fall into this? Often, insufficient book reviews exist for coverage of a humanities/social science text for the article to focus on quoting reviews (though this is usually recognized as best practice), so interpretation is necessary for much of these works, which may look a lot like opinions.
I guess I'm just worried that half of humanities wikipedia will become "unverified," or worse, "unverifiable," overnight, if the ML algorithms' sensitivity is set just a little too high, or it never gets trained on H/SS articles.- - mathmitch7 (talk/contribs) 19:04, 30 April 2019 (UTC)[reply]
- Mathmitch7 This is an interesting point. Although there's no cause to be worried because there has been no "change"—no new products have been developed, no community policies have been changed. If a wiki decided to adopt this technology (for example, in a new kind of CitationBot), then they would have control over implementation. If WMF decided to incorporation citation need predictions into a MediaWiki feature or something (and there are currently no plans to do that), then they would be suggestions, not mandates. Individual wikis still determine what notability and verifiability mean. Cheers, Jmorgan (WMF) (talk) 15:29, 2 May 2019 (UTC)[reply]
- Other comments have mentioned the humanities, but I work extensively in the biology projects and you might be surprised at just how many unsourced or poorly-sourced claims there are. I don't generally work with our best articles, to be fair, but this is a problem for all of enwiki--NOT just the humanities. Honestly, while I do think that this kind of tool would be extremely helpful for us in figuring out what needs citations, I also think it is the cart leading the horse. The problem with lacking citations has NEVER been that people don't understand when we need them. Sure, there are plenty of inexperienced editors who just cobble an idea together without sources because they don't know better. But I would argue strongly that the bottleneck has always been that finding and citing refs is a pain in the rear. There, I said it. What I really think that WMF should focus on to solve the WP:V problem is making research and citation easier. Look at what Microsoft and Google have been doing with their word-processing software. I think we should be BOLD and consider ideas like having research tools built into the editing interface, including (for starters) links to other WMF resources. Many journals and other sources actually create citations for you automatically, now, but they are in different formats. What about machine-learning tools to reformat citations semi-automatically (reviewed by a person) and strip information out of them? What about redesigning the code editor to improve syntax highlighting to make it easier to see refs or easier to see article text, depending on need? Those are all ways which I think would directly support the goals of WP:V very well. Don't get me wrong. I like the citation checking concept. But it isn't any good without a way to dramatically increase the amount of sourcing by editors, especially the less-dedicated. Prometheus720 (talk) 02:45, 3 May 2019 (UTC)[reply]
- This! I think that's a great idea. Sometimes articles already point in a direction of a citation but don't connect all the dots ... I think ML could definitely help us out with that!
My only potential concern (which could be mitigated!) is definitely about citational politics: it seems that an ML system would likely point us toward the already over-cited resources, instead of new resources that could substantially contribute to an article. I don't think that's a problem per se, just a new technical/political challenge to consider. How do we point people toward quality resources that aren't widely used? How do we know they're quality if they're not widely used? Maybe there's a cultural reason they're not used (i.e., pseudoscience that has all the packaging of legit science but supports totally bogus claims that most people already know are to be avoided). Just a thought! - - mathmitch7 (talk/contribs) 03:11, 3 May 2019 (UTC)[reply]
- @Mathmitch7: I actually was referring to just reformatting citations. Throw at it some citations with all the info needed of various types of sources, and then say, "Here, clean these up and make them look the same." Even better, it could actually follow the doi or other link and collect any additional information which might be needed, or perhaps even go out and find a doi link. I would not want to use machine learning to find new citations. That would be dicey as you pointed out. Prometheus720 (talk) 16:09, 4 May 2019 (UTC)[reply]
- --Guy Macon (talk) 15:49, 4 May 2019 (UTC)[reply]
Why are the images in this article... images? For someone using a screen reader, or with images turned off, they provide no information. The image at the top of the article is decorative, fine. The distribution of reason labels might be difficult to turn into a text explanation. Understandable.
But 'Reasons for adding a citation', 'Reasons for not adding a citation', and 'Examples of sentences that need citations according to our model, with key words highlighted': Why are these images and not text? All three could be communicated as effectively in text, without the accessibility failure. The first two are especially bad. There's no good reason for these to be images and not text. If you (the people who wrote the article, the people who created or added the images, the Signpost editors) thought about this and made the decision to use images rather than text, why did you not add alt text?
I would fix it myself were I more expert in the use of Commons and editing of image files here. That wouldn't, however, change the copies of Signpost that are on talk pages or in other locations.
Please read the section on images on the Accessibility page of the Manual of Style, and please, don't do this again. BlackcurrantTea (talk) 05:28, 13 May 2019 (UTC)[reply]
← Back to News from the WMF