17

We all know that adding an image to a post would give something like this [![Enter image description here][1]][1]. Most people will just press the ask question button without adding it, even though there is a warning notification begging for users asking users to add it. The only issue is, is that people are in areas where internet decides to be terrible and thus loading images will give 'Enter image description here'. This however is not good because we don't know what the image is about if there is no description.

For example: if I upload an image of some LEDs that don't work onto electronics.stackexchange.com, and there was someone who had bad WiFi connection, for example, and the image doesn't load, the first thing they are going to expect that the image description says: 'Broken LED string' or something similar. However all that they are going to see is the default 'enter image description here'.

According to this answer, it is an accessibility issue and I agree on that. Also, nothing has been done since 2018 when this question has been asked

So why not just make it so that when you upload an image, no matter if you try to avoid adding an image description because you're lazy, it is going to "pressure" you into adding the description. I feel that this feature can really help people who have bad internet and for those who are blind etc. An alternative, if making this feature required doesn't work well, is to use Machine Learning and create automatic image descriptions (Like Microsoft Word or PowerPoint).

Improve this question
19
  • 19
    People will end up hammering nonsense into the field until character requirements are met. Commented May 2 at 0:10
  • 4
    The word problem is banned in question titles. Did that fix anything? No, people write probelm in question titles again and again and again and again. What they absolutely don't do is completely rewrite their title not to use the banned word. Commented May 2 at 0:29
  • 1
    That is because some of those people are new and don't know about the word problem being banned in question titles. Commented May 2 at 0:30
  • 1
    It has nothing to do with bad internet and showing the image description and everything to do with accessibility guidelines and screen-readers. Commented May 2 at 0:48
  • 2
    @Otaku, I had an issue when I was at a building whose wifi connection was far from the main building where the router was. I was loading a question to see if I could answer it, the images failed to load and gave me the image description Commented May 2 at 0:55
  • 9
    @muru "People will end up hammering nonsense into the field" Even "Blue thing on white background" beats "Enter image description here" by a mile. Low effort beats no effort every time. Commented May 2 at 2:14
  • 5
    @galacticninja I'm thinking more on the lines of "asjeh23iuehreqwiorhnw3h8or hiwefnoi24hrweiofhweoirfhwiofkqwejdiflqw" rather than "Blue thing on white background" Commented May 2 at 2:22
  • 1
    Might there be any way we can make it pop-up on hover so curators can see it? It bothers me that we don't see it without pressing the edit button to check at present. Commented May 2 at 2:49
  • 2
    @muru That's not what typically happens with required fields, though. Look at question titles - they're mandatory too, but people write actual (if sometimes vague) titles rather than gibberish. I think that's unlikely to happen in practice. I've never actually seen anyone keyboard-smash an image description - they'll be minimal perhaps, but rarely gibberish. Commented May 2 at 7:00
  • 3
    @galacticninja People generally tend to care a bit more for titles since they know that's what other users will first see, yet we still see plenty of low effort titles too - often just error messages pasted in. Another mandatory field is tagging, and users here have to spent a lot of time fixing tags because posters often put random tags, sometimes paste errors into tags, etc. I don't expect any higher effort for fields that they don't actually care about, so in this case, I do expect low effort to become gibberish, which is actually worse than no-effort leave-the-default-in status quo. Commented May 2 at 7:05
  • 1
    Related: Please catch no alt text in the automated question review process Commented May 2 at 8:04
  • 2
    @muru But gibberish would be no worse than what we already have. That's not a practical downside if it provides the exact same amount of information as what we already have, especially if requiring it caused an uplift in the amount of users which did fill them out with something valid. Commented May 2 at 16:56
  • 3
    @zcoop98 if you sincerely believe gibberish is no worse, try listening to meta.stackexchange.com/questions/408696/… using a screen reader a few times. Commented May 2 at 20:20
  • 1
    @muru Hand out a suspension every time you see "asjeh23iuehreqwiorhnw3h8or hiwefnoi24hrweiofhweoirfhwiofkqwejdiflqw" and we will get to "blue thing on white background" I'd say. Commented May 2 at 20:36
  • 2
    Getting the desired result here would require an incentive. However, the rep system as designed clearly cares more about new questions and answers than about improvements to existing ones. Commented May 2 at 21:17

5 Answers 5

Reset to default
9

I think adding better guidance around writing good alt text on the site might be even more important than adding a requirement, and needs to happen first, or in tandem with, this change if it were to be built.

There seems to be a lot of concern around the addition of this requirement leading to authors adding gibberish or bad alt text to satisfy the requirement– but I raise that this is what we already have. Gibberish is no worse than, conveys no more information than, "enter image description here" for screen-reader users.

But that goes both ways– there's no reason to make such a change if it will just lead to differently bad captions. I think we need better messaging and guidance around alt text broadly on site, both for post authors and for curators, if it was going to be made required.

Better messaging, even outside of the proposed requirement, could alone lead to better outcomes here– it's hard to know from the outside how many people would fill it out today if they had some sense of what should even go there.

As far as I can tell, the Help Center doesn't even mention alt text at all– the syntax is demonstrated on the post formatting page, but the only direct reference I found is in the images section of the buried advanced Markdown help page. I couldn't even find anywhere we directly call it out in an FAQ post here on MSE; it's not mentioned in the image upload FAQ, nor the image syntax FAQ.

The post quality check step isn't really helpful either– it simply offers "Images should have specific alt text briefly explaining what is shown", which might be the best guidance on the site... which is a joke.

Ultimately, it feels woefully unsurprising that we get little alt text on our sites when we give little to no material information on how it should be used, or why it's beneficial beyond handy-wavy remarks like "Be sure to include meaningful alt text for screen-reading software" (that's from the advanced Markdown help page).

I'm not convinced that adding a hard requirement will do anything meaningfully beneficial unless we also accompany it with useful guidance in the right places for users to actually see and use it. Otherwise, I see no reason to expect anyone other than the highly motivated to know what it is, why it's important, who stands to benefit, and how it impacts them.

Improve this answer
2
  • I definitly agree that we should a help guide designated to writing good captions Commented May 6 at 17:34
  • 1
    Not coincidentally, adding good alt text guidance is the main thrust of my suggestion here (which predates the generic warning about alt text that exists today). Commented May 19 at 17:23
17

Unfortunately, requiring people to add an image description might do more harm than good.

Writing alt text is a skill. You have to know both what's important to include in alt text generally and what's important in this particular case; the exact same image can have different alt text depending on the context, since the alt text needs to convey only what is relevant in this context and nothing else. While there are guides to writing accessible posts available, it's a bit much to ask for everyone to read a guide and become familiar with how to correctly write alt text before posting their question or answer.

If we make it required, people will do what they can to meet the technical limits of the requirement - which can perhaps be more harmful than helpful.

Let's say you have a post. The post includes a screenshot of an error, and you write something along the lines of "Here's a screenshot of the error:". You then include the image, and since you have to include alt text, the alt text is simply "screenshot".

That alt text is now completely useless. The fact that it's a screenshot has already been mentioned in the text of the post, and the important information - what the error says, or what it looks like, or whatever the important bit is here - isn't in the alt text. And now I can't find that image to fix the alt text later.

I have some data.SE queries that were written for me to find images that have either blank or the default alt text. As soon as you require that people insert their own text, it becomes impossible to automatically detect images with poor alt text and fix them, meaning that both the issue remains and it becomes harder to fix it later.

I agree that there should be more awareness around accessibility and how to write accessible posts; for instance, in the question here, I noticed you included two links with the same nondescript link text of "this", which is also bad from an accessibility point of view. SE finally added a prompt for alt text when writing questions, after years of my pushing for that step; and while prompting will hopefully inspire users to actually think about it, forcing them to add alt text will only result in poor alt text and lower discoverability for posts that need to be fixed.

Improve this answer
3
  • 1
    I just don't agree with you that the tradeoff here is a poor one. "Detecting poor alt text" shouldn't be something that superstar curators need to comb through SEDE to find; posts should be fixed on a case-by-case basis by editors in the normal, organic flow of improving posts on a site, and guidance for editors should be enhanced to support it. As far as I can tell, aside from the recognition issue, "bad alt text" is what we already have, so if requiring it indeed led to an uplift in the amount of good alt text being written by OPs, then I think that's better than what we have right now. Commented May 2 at 17:04
  • @zcoop98 we have enough community developed scripts that the company could build tooling to effectively detect the issue. The problems are a) the company isn't motivated to invest in this area beyond lip service, and b) what do you do with the detections that doesn't trigger a cobra effect. At least the friction means that people who are tracking these down are sufficiently 8nvested to fix them properly. Commented May 2 at 18:20
  • 1
    Perhaps, as an alternative to mandating alt-text, the site could simply remind users to add descriptive text to their images if they haven't, with a link to a community guide on writing good descriptive text and a "post anyway" button if they can't (e.g. if they themselves have vision impairment, a language barrier, or any number of other reasons) or don't want to. That way you have the benefit of reminding users to supply the text, providing education on how to do so, and avoiding malicious compliance. A similar thing has been deployed on many Mastodon instances to great effect Commented May 3 at 0:31
3

Totally with you on this. The default "Enter image description here" is basically useless, especially when images don’t load (slow internet, restricted devices, or accessibility tech like screen readers). It’s frustrating both for people trying to help and for users who rely on alt text to understand content.

Honestly, it shouldn’t even be optional at this point. If you're uploading an image, the platform could at least strongly nudge you to add a description — maybe even block the post until you do, or make it super obvious you're missing something. Right now, people skip it because it’s easy to ignore.

And yeah, using machine learning to generate a basic alt text would be a great fallback. It wouldn’t be perfect, but it’s way better than the current placeholder. Even something like "photo of a circuit board with LEDs" is 100x more helpful than nothing.

It’s wild that this has been discussed since 2018 and still hasn’t changed. Improving alt text isn’t just about accessibility anymore — it’s about usability for everyone. Hope this gets more attention soon.

Improve this answer
-15

A picture is worth a thousand words. Sometimes, if you can't see the picture, there's no point explaining it. For example, if I have a post of how do I fix this, I could label the picture as picture of broken object, but if you can't see the picture, the label won't make a difference. I could give a longer description in the body of the post such as what type of glue I tried, but without actually seeing the picture you won't understand why it didn't work.

Improve this answer
3
  • 1
    Won't the label be shown on the post though? Commented May 2 at 0:34
  • 10
    Image descriptions are good, always. If nothing else, they let people using screen readers or with slow connections read the post at least somewhat properly. Accessibility is important, even if the description isn't perfect. Commented May 2 at 1:09
  • 24
    If a picture is worth a thousand words, then an image description is worth a thousand thank-yous from people who depend on them: users of screen readers, those with visual impairments, and anyone whose images won't load for whatever reason. Commented May 2 at 2:07
-21

I'm against that proposal of making image description (a.k.a. alt texts) required. Reasons:

  • It will decrease the number of uploaded images.
  • It will increase the number of linked, non-embedded images, which I believe don't support alt texts. (How can one add an alt text for a linked image?)
  • Some users will write random image descriptions. See What percentage of images on SE have a meaningful alt-text? (e.g., from Aaron Bertrand: "I casually scanned through the first, oh, thousand or so entries, and found very few that seemed like meaningful text."). Useful alt texts require more thoughts.
  • AI can do a decent job at it in many cases so I'd prefer SE to draft the description, that'd be more efficient than having each users call themselves an image caption model or always write from scratch.
  • The number of questions have decreased by a factor of 10 already, so it's probably not an ideal time to make posting even harder.
Improve this answer
10
  • 3
    The Site Analytics Priviledge page specifically says "Since this is a restricted privilege, we'd prefer you not share the raw data." Commented May 2 at 20:43
  • 1
    @StarshipRemembersShadow image != raw data Commented May 2 at 20:44
  • It is the raw data though, or at least all the data you have Commented May 2 at 20:48
  • @StarshipRemembersShadow no eg not daily Commented May 2 at 20:49
  • 6
    I deeply disagree with the assertion that "AI" can generate sufficiently useful alt-text, and I think many in the blind community would too. An automated system that confidently provides plausible but incorrect information is deeply frustrating and serves to put up more barriers than it breaks down. Not to mention the energy consumption at this scale. Commented May 3 at 0:36
  • 1
    @Polynomial I've looked at image captioning but not specifically alt texts. If you see some nice studies about alt texts, let me know. There are some solutions for it e.g. chatgpt.com/g/g-bqbZtTpHb-alt-text-generator-assistant but I don't know how good it is beyond eye balling a few results. Commented May 3 at 0:49
  • @FranckDernoncourt I don't have any studies to hand, just the general opinions of the blind folks I know. In short: advancements in forced alignment and similar assistive technologies for sighted creatives are the star of the show, generative TTS occasionally offers mild improvements in inteligibility, auto-captioning and voice transcription have barely improved and are still barrel-scraping for everyone except normative American English users, and automated image description is a deeply begrudged sticking plaster in lieu of a society that actually cares enough to write descriptive text. Commented May 3 at 1:06
  • @Polynomial Thanks for the feedback. I am surprised to hear about voice transcription, I use it myself in two languages that don't include much American accent and it works great, which concur with ASR benchmark progress. Regarding alt texts, I think that one key issue for AI is guessing the user intent. E.g., if the image is a full-screen screenshot, the AI needs to guess what in the screenshot needs to be mentioned in the alt text. But in some cases, e.g. the screenshot of an error message, the AI will do great (e.g., mere OCR). Commented May 3 at 1:15
  • 2
    @FranckDernoncourt Certainly possible that my social group is skewed to folks with unusual accents or dialects, but I suppose that's rather the point being made. And yes, context is very much the problem for the image description stuff. For straight OCR, anything not involving a large language model layer would be preferable due to the insidious tendency to gravitate towards outputting token sequences that fit their training set instead of what's actually written in the image. Given that issues and questions here are often novel, it's all the more problematic. Commented May 3 at 1:23
  • 1
    As someone who's been occasionally called on to write alt text for illustrations of the sort that get uploaded to SE, my experience is that AI flatly can't do it. It can sometimes describe the contents of an image, but it doesn't know what parts to call out as important and what parts to ignore. Commented May 20 at 3:36

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.