MathJax 4.0 possible issue

[glarose]

I have been working with the staff in my accessibility office, and they had the following observation about changes made in MathJax v4. They note that it has many improvements and useful accessibility features, but expressed some concern about its default configuration.

The underlying change that concerned them is that by default it does not work with MathCAT (in NVDA or JAWS) or other assistive technologies. In particular, the default configurations of MathJax v4 no longer include the underlying hidden MathML for the math expressions it renders. The change seems to be that in MathJax v2 and v3, the assistive-mml extension was loaded and enabled by default, but in v4, the explorer component is the default assistive tool. The MathJax documentation says that the explorer can be disabled and the assistive MathML re-enabled using the MathJax contextual menu on any typeset expression, and page
author’s can override the defaults in their MathJax configuration objects, if they so desire. (As an aside, the staff at my end say that actually one has to turn off the MathJax semantic enrichment option, not the the explorer on its own; turning off semantic enrichment also disables the MathJax explorer.)

I don't think this changes WeBWorK's accessibility, but may change how we want to present it to users. The people I'm working with think that the default really should be to have MathJax set to allow the user's screen reader to handle accessibility, and to allow them to go over to MathJax instead if desired. If we think that's the right strategy, we might just change the default MathJax configuration in WeBWorK. If we go with the default in MathJax v4, we may want to update the orientation set.

[drgrice1]

Note that we are not using MathJax version 4 yet. We are still using version 3. Is this preemptive for when we switch to version 4?

[glarose]

That's correct. I just wanted to make sure that it was on the radar for when that switch gets made.

[drgrice1]

Okay. Thanks for clarifying. We will make sure to consider this when the upgrade is done.

[drgrice1]

I am not sure about this. In my testing with Gnome Orca, the MathJax explorer works far better than including the hidden MathML. I don't have a proprietary screen reader such as NVDA or JAWS to test with, but the MathJax developers claim to have tested with several screen readers including NVDA and JAWS, and various browser and configuration combinations. I am not saying that your accessibility office is wrong, but it seems that if the MathJax developers have tested as claimed and have chosen the MathJax explorer to be the default, then it is probably best to leave that as the default.

I think that it would be good to see what @dpvc has to say on this.

The mathInteraction.pg problem in the student orientation set certainly will need to be updated. Several of the things it says are not correct for MathJax 4.

[dpvc]

The accessibility office is correct when they say "it does not work with MathCAT" because MathJax does not need MathCAT to work with screen readers like NVDA and JAWS. MathCAT is Neil Soiffer's successor to the MathPlayer plugin for Internet Explorer (When IE stopped allowing plugins like this, he had to find other routes for handling MathML in web pages in Windows).

MathCAT only works with MathML, and in v2 and v3, MathJax would insert visually hidden MathML for screen readers that understand MathML to read. As mentioned above, MathJax can still be configured to do that (though one need not go so far as to disable semantic enrichment entirely, as turning off speech and Braille generation will disable the explorer while still enriching the underlying MathML).

There are a number of problems with using the hidden MathML, however. The first is that the screen reader needs to be able to understand MathML, and not all do. For example, NVDA needs the MathCAT (or similar) plugin to do that, and that means extra downloads and setup for the user who needs it. Second, the hidden-MathML approach was fragile, and didn't work reliably across browser/os/screen-reader combinations. For example, Orca in unix environments was only sometimes able to read the MathML, and VoiceOver on MacOS ignored it completely, despite their claim of handling MathML (which I could never get to work, even outside of MathJax).

MathCAT works as a plugin for some screen readers, like NVDA, to convert MathML in the page into speech strings that it passes to the screen reader. MathJax v4 (and v3 and v2 if you enable it) does this conversion to speech strings directly itself and makes those string available to the screen reader through various ARIA standard attributes. So they both do similar things, but MathJax will work with NVDA, JAWS, Orca, and VoiceOver out of the box, with no special plugins needed. So they are right that MathJax doesn't work with MathCAT, but it doesn't have to work with MathCAT as it works with NVDA and JAWS directly with no need for it. It may be that they are used to how MathCAT Works and aren't interested in changing how they operate, but to say that MathJax doesn't work with those screen readers is not correct.

When redesigning the assistive support for v4, we had the following goals. It should:

• Work across the major browser/OS/screen-reader combinations.
• Properly read the full expression when reading the whole page or stepping through the page sentence-by-sentence (or by other units).
• Automatically enter “focus mode” when the expression is focused via tabbing or clicking, in screen readers that have separate focus/browser modes, and start the expression explorer.
• Allow control over the description of the math (e.g., “clickable math”) used during reading and focusing (via a menu option).
• Voice the phrase “press H for help” when the math is first focused (but this can be turned off by a menu preference).

The first three are the most important, and were not being met by the hidden-MathML approach. The last two are to help the user understand that the math is interactive, and to get help on the explorer, if needed (but they can turn off those features once they know how the explorer works).

You might ask the accessibility office what their criterion for success is (what constitutes a "working" assistive solution). I don't think that "working with MathCAT" is a reasonable answer for that, as the answer should be based on meeting user needs, not supporting a particular software technology. If MathJax doesn't meet those needs, we would be happy to hear about how we can improve it.

I know people don't like to change things that they have been using in the past (and I count myself among them), but the hidden-MathML technique was not working and needed to be replaced by a better one. Users who prefer the old experience can still get that by switching the preferences in MathJax (which are sticky site-wide, so will stay in effect when the come back to the site again). But I would discourage you from switching the default settings for WeBWorK just for MathCAT, which only runs on Windows, as that will mean that the unix and MacOS experiences will be degraded, if not outright broken, and the default settings work fine on Windows without MathCAT.

[glarose]

Thanks, Davide. I appreciate greatly your (as usual) complete evaluation of the issues here. I'll follow up with my accessibility folks. My sense is that they are testing with the perspective that they have a working screen reader that effectively manages math and expect that the default should be that MathJax defers to that. If I'm correct, it may be a question of whether we (or they) expect a user who is relying on the screen reader to have a configuration that allows them to read MathML already in place. My general feeling is that the "issue" is more of ensuring that the orientation articulates for users how MathJax is configured and how to change that configuration if it doesn't match a user's need.

[dpvc]

I should say that I think MathJax will work with NVDA even with MathCAT (but MathCAT will not be used). I don't have MathCAT to test with, but I don't see why it would interfere with the way MathJax works. It's just that MathCAT won't be doing the math reading for them, NVDA will.