Always fun to read #hackerNews and their rather ... let's call it ... qualitatively exceedingly disappointing mentality towards #accessibility (stroopwafel to the dutchies who see what i did there). But some points do make me think to what degree, say, #screenReaders could be innovating more.
One thing that stuck by me was a comment on an article regarding the use of "click here" that basically suggested some kind of heuristic for, say, reading the entire line of text when that's encountered, rather than the link text. Another one could be including the previous sibling's accessible name on unlabeled form fields which I ...think? JAWS might already be doing?
Like obviously these should be toggleable to not screw up audits but I think after 30 years we can safely say people often can't be forked to learn #HTML basics and it might make some things smoother?
in reply to Florian

I have no idea how computationally difficult this would be or to what degree imperfect implementations of things like UIA would mess this up, but yeah... I overall wonder to what degree we're innovating, currently. I know, again, JAWS is doing things like AI-generated labels for unlabeled controls now but that might be a little too far in the wrong direction. It's also one element at a time, and a manual process, that users have to learn about. Ideally I'd love for screen readers to do more little things the user doesn't need to explicitly make happen
in reply to Florian

It's really sad how we went from "there's no accessibility API so let's just hook some interrupt handlers" to "not our problem, please file an issue upstream"

Screen reader developers these days seem to be doing very little in therms of "forcing" 3rd party apps to be accessible, even if their developers don't care.

If we want accessibility, we should be doing as much of the legwork as possible, not depend on somebody else's generosity and charity.

in reply to Florian

Some ideas off the top of my head:

- Addons/scripts not for applications or websites, but for specific UI toolkits (QT/the 50 different .NET ones/WX etc.) so the screen reader can work out better labeling for unlabeled, or badly labeled controls. Yes, .NET developers should toString() their list items. No, they probably aren't going to do so, so please just concat the text of the child(ren) of the listItem and echo that instead.
- When "click here", "Learn more" etc. are encountered check for more text in the current line. If exists, read. If not, depending on a config settings with loads of "you might not want this" disclaimers, read the previous line instead.
- I'd be curious to what degree we can infer state/roles for "clickables" based on what comes back from the browser's accessibility tree... CSS classes? Common heuristic patterns? I doubt it, but I'd be curious.
- Form field with no associated label? Check the previous non-<i> sibling, it's ALMOST always the right label.
etc.