Many browsers include user preferences to select preferred colours for text, backgrounds and links. This is intended as an accessibility feature so that people with limited vision, dyslexia or other disabilities can choose a colour scheme that makes it easier for them to read web pages. In theory it’s a great idea. In practice however it seems to be nearly useless.

There are two “modes” these colour preferences can be used in:

1. As preferences that web sites can override.
2. As forced settings, where website specified colours and background images are completely ignored.

Preference Mode

In the first case, there are a vast number of sites which become unusable because they specify a specific background colour without also specifying a foreground colour. So if you happen to set your preferences to have light coloured text on a black background, then try to log into Google you get:

Choosing white text on a black background would make half the text on the internet disappear.

Clearly this is poorly coded websites causing accessibility programs and they should get with the program and specify foreground and background colours in pairs. Which of course makes that dialog box look like:

Much better. Except now there is no sign that I ever changed my browser preferences rendering them completely useless. So now designers have a choice between respecting browser preferences or largely designing a black and white site.

The net result is that changing your browser colour preferences will have no effect on well coded sites and break a large number of sites. So why bother?

Forced Mode

So we’re left with forcing our preferred colours on web sites. This actually gives us high contrast and is most likely very useful for people with limited visibility. However, for people with dyslexia who wanted more readable colours1 rather than just high contrast now tend to struggle because so much visual context is lost from the page. Sure, on a well coded site all the information is there, but different background colours is a really useful indicator for the various sections. That extra context and readability is now lost. That’s a disaster for someone with good vision who was trying to make the page more readable.

Compare the very well coded BBC website with preferred colours vs with forced colours.

It’s important to note that the BBC website is actually quite well coded – all the information is still available when you force the preferred colours, the designers simply have a significantly reduced ability to provide context and delineation, so inevitably the page becomes harder to read.

What to do About it?

Frankly, I have very little idea. The never properly implemented and now deprecated CSS system colours would have helped a lot in some cases. Design heavy websites would have still wanted their own colour set but web applications could have at least used them. Perhaps some way to specify a colour along with a fallback system colour could have worked. Then in the forced colour mode, the system colours would be used.

Beyond that it would be really nice if browsers were smarter about the options they provide. Instead of just a default colour setting, have a minimum contrast setting – the browser automatically increases the contrast between background and foreground colours to a specified point. Then you could combine that with a preference that text is always blue – if the website specifies black text on a white background it comes out blue text on a white background. If it’s white on black then it comes out as blue on suitably contrasting gray.

I don’t doubt that there are huge challenges here and I certainly can’t suggest the answers, but it seems like it’s about time we started putting pressure on user agents to do more to support accessibility rather than expecting website designers to do the impossible and create great designs that work with such limited accessibility tools.

1 – blue on white is actually much more readable than black on white ↩

In ARIA, there are two ways that you can attach a label to an element, aria-label and aria-labelledby. In theory, you should use aria-labelledby if the text is visually on-screen somewhere and this form is preferable. You should only use aria-label when it’s not possible to have the label visible on screen. This makes a lot of sense because generally sighted users like to be able to see the label as well and it prevents the visual and non-visual cues from getting out of sync.

However, with current support, you should always use aria-labelledby, even if you need the label to be hidden. While at least JAWS 12 and VoiceOver both read the aria-label as expected, it turns out that VoiceOver doesn’t correctly read the hierarchy if aria-label is in use. For example:

<p id="group1">Outer group</p>
<p id="item1">First item</p>
<div role="group" aria-labelledby="group1">
<a href="javascript:" role="button" aria-labelledby="item1">Item Content</a>
</div>

everything here is using aria-labelledby and VoiceOver will read the button as “First item Outer group button”. In other words, the button label, the group label and then the type of object.

However, if you change any of the elements to use aria-label, for example:

<p id="item1">First item</p>
<div role="group" aria-label="Outer group">
<a href="javascript:" role="button" aria-labelledby="item1">Item Content</a>
</div>

VoiceOver will now read the button as simple “First item button”. It doesn’t seem to matter which item uses aria-label, if it’s anywhere in the hierarchy, only the label for the button itself will be read out.

Brad Neuberg has a post up about ARIA on the iPad and some of the tricks he’s used to bend it to his will. Blogger won’t accept my OpenID to comment on the post for some reason so I’ll add some thoughts here.

tabindex

As Brad notes, you can set tabindex=”-1” to prevent an item in HTML from appearing in the tab order. Also as Brad notes, this won’t stop the VoiceOver cursor from moving to that element. It’s important to remember that the VoiceOver cursor is not linked to the keyboard focus, it’s linked to what is being read out to the user. This can be very confusing but it’s an important concept, allowing you to review parts of a document without losing the current caret position where you want to continue editing. Most screen readers seem to have this distinction between the text input focus and the screen reader cursor.

tabindex=”-1” also won’t prevent the user from moving keyboard focus to an element by means other than the tab key. For example, clicking on a textarea will move the keyboard focus to it even if it has a tabindex of -1.

aria-hidden

As a work-around to try and prevent VoiceOver from focusing an element, Brad then added aria-hidden=”true”. To his credit, Brad notes this isn’t a good situation and indeed it’s not. Now we have an element which is available to sighted users, but completely hidden to the screen reader. This is almost never what you want.

If something shouldn’t be a landmark, it probably shouldn’t be a heading element and if it shouldn’t be read by a screen reader it should generally be either not there at all or both aria-hidden=”true” and display:none; That way the element disappears for everyone, not just those using a screen reader.

I’m not quite sure what Brad was trying to achieve so I can’t really suggest the right workaround for his use case but it sounds like there may be a need to go back and get the basic semantics of the HTML right.

aria-labelledby

I’m not entirely sure what Brad’s aiming for here again but it’s a bit unusual to try and add extra audible information to a heading which isn’t available to sighted users. Normally labelledby would be used for form elements to link them with the text that describes what they’re for. My first thought here is to adjust the HTML to make it more semantic rather than trying to fix it up with ARIA. For the example HTML:

<div aria-labelledby="testH1 testH2 testDIV1">
<h1 id="testH1" tabindex="-1">
This is an H1
</h1>
<h2 id="testH2" tabindex="-1">
This is an H2
</h2>
<div id="testDIV1" tabindex="-1">
This is a div
</div>
</div>

I would say the right thing is to not have any ARIA attributes (or tabindex) at all. h1,h2 and div all default to the equivalent of tabindex=”-1” anyway and if a sighted user would see this as two separate headings and a normal paragraph, a screen ready probably should too. If it’s intended to be read as a single sentence then as a signted user, I’d appreciate it being laid out that way.

aria-label

As a side note, aria-lablledby is only one option for adding information and it’s useful if the extra information you need to present is already available in a textual form somewhere on the page. However, if the content isn’t available (for example because it’s using an icon instead of text) you can use aria-label=”The label you want” to directly specify the label text rather than referring to some other HTML element.

More Info?

It would be great to get some better background information on what’s being attempted here so that better solutions could be proposed. Right now it looks a bit like trying to use ARIA to hammer a square peg into a round hole which not surprisingly isn’t a lot of fun and doesn’t work very well.

The bottom line is that ARIA isn’t the answer to accessibility, it’s just a smattering of extra information to help out in complex cases where HTML and JavaScript are being used to create custom widgets. The first and most important tool for creating accessible web pages is still to use appropriate semantic HTML – especially for the “document” parts of a page as opposed to interactive widgets. If that foundation isn’t set, ARIA won’t save you.

The Wall Street Journal has an article on a new bill just passed by US Congress that updates accessibility requirements.

The legislation, called the 21st Century Communications and Video Accessibility Act, covers things like captions on Internet video, Internet phone services that work with hearing aids, television menus that can be seen by people with vision loss and even touch screens that the blind can use. The bill, which passed by voice vote and will now go to President Obama for approval, updates existing regulations to bring them in line with the Internet age.

I haven’t taken the time to decipher the details from the actual legislation text as yet, so it’s hard to say exactly what implications this will have. It is at least good to see improvements to US laws – to date it has really been civil threats, such as the suit against Target, which have been the main legal avenue to ensure accessibility. This is generally under the American's with Disabilities Act which was written in terms of physical accessibility for public places, but has been deemed to also apply to the web. As I understand it, this new legislation will provide some more teeth on the criminal1 side.

1 – as in, Government-led as opposed to civil. I’m not sure I’m using the term right. ↩

Over the weekend I went to JSConf EU and every time I met someone and told them I was working on TinyMCE the conversation rapidly veered off into why contentEditable is too hard/buggy to work with and how it would be so much better if it was all built on top of canvas or just used the DOM with a custom selection API. This matched the thinking throughout the conference that it doesn’t really matter how messed up browsers are because JavaScript can be used to work around it.

What disturbs me is that incredibly few people realise just how much work is involved in implementing caret and selection handling correctly, let alone most of the other infrastructure provided by contentEditable. I wound up talking to Mark Wubben who actually does have a pretty good idea – he implemented a DOM based selection system for the Xopus editor. Needless to say he had quite the war story to tell, but despite doing what sounded like a very impressive bit of work, the selection code is still very broken in a couple of key ways: accessibility and international inputs1.

Accessibility

Accessibility support was my first response to anyone suggesting using canvas as an editor – no-one had considered it. If you play around with a screen reader for a while, you start to realise just how important the selection is to accessibility. Selecting text generally causes it to be read out for example. It’s not just screen readers either, screen magnifiers will move the screen along to keep up with the caret, ensuring the user can see what they’re typing. If the caret is fake, none of these features will work and the editor becomes inaccessible.

International Inputs

Even I’d forgotten about this one, but it’s likely an even bigger problem to solve than accessibility. The array of input devices, keyboard shortcuts and other ways that people input characters is pretty amazing when you start exploring it. People in English speaking countries don’t tend to notice this because they have quite a small number of characters in use which all fit on the keyboard. Even Western European countries have the vast majority of characters required available with a direct key sequence. However there are far more complex types of inputs around. The simplest of these is the use of “dead keys” such as acute on a US or British keyboard. To type é actually takes multiple keystrokes on a UK Mac keyboard:

1. Option-e. This is a dead-key so only a placeholder appears in the output ().
2. e. Now the e-acute character actually appears and is inserted into the document content.

Not to pick on Xopus but at least in Firefox on Mac, I can’t type é at all in the editor because it doesn’t correctly support dead-keys2.

Asian languages can have even more keystrokes required to type a character and much more complex input methods that to be usable, require changes to the caret, underlining content etc. Plus the methods and UI is different on different OS’s. While users can handle some variation, they shouldn’t have to learn a whole new input method to be able to use your application3.

The Way Forward

Instead of trying to work around the limitations in contentEditable we need to be pushing browser vendors to fix it and make it usable. Where the current APIs aren’t good enough, standardise better ones – we’re dealing with multiple incompatible APIs already so the backwards compatibility problems are limited. This needs to be a focus so that we don’t wind up in a world where online editors only work for fully-abled, English speaking users. Most if not all of what’s required is already there, it just needs some love and care to remove the bugs and better handle corner-cases.

1 – There is some good discussion of the complexities involved here from the W3C canvas list as they struggle with the accessibility implications of canvas.↩

2 – and to be completely fair, EditLive! fails to paint the placeholder, though the character does type correctly. ↩

3 – So no, Google Virtual Keyboard is not a real solution, however good an idea it is for people using things like internet kiosks with foreign keyboard layouts. Besides, would you like to be stuck typing only with your mouse? ↩