Some observations from accessibility testing

[jordanc...@gmail.com]

Recently the distance learning staff where I work at Portland Community College conducted accessibility testing of WeBWorK, which uses MathJax. The blind user assisting with testing was very impressed with MathJax accessibility features, and for anyone who would like to hear his praise firsthand, you can watch the full video linked below.

There were a couple of things that didn't go perfectly that I'd like to report here. I'm not able to say if they are issues that MathJax could address, or if they are issues that would fall under the domain of MathPlayer or NVDA ("after" MathJax) or maybe even Firefox, Desire2Learn, or WeBWorK ("before" MathJax). I'm hoping that someone here would be willing to watch these clips and offer an initial diagnosis of which technology is the issue.

1. https://www.youtube.com/watch?v=g_D4kdyAWmg&feature=youtu.be&start=1725
   Starting at about 29 minutes for a minute or two, our tester (John, who is blind) finds that if he is "zoomed all the way in" then there is an open paren that is not reported but the close paren is. Is this a bug at some level or expected behavior?
   
2. https://www.youtube.com/watch?v=g_D4kdyAWmg&feature=youtu.be&start=2830
   Starting at about 47:30 John encounters a multiline equation that was originally coded as an aligned environment with LaTeX input. There is about 15 minutes of video here and roughly the first 5 is John probing and trying to understand what he's looking at. I think that ultimately there are two issues he encounters, maybe related.
   
   1. "cells" in the aligned environment that are empty do not indicate they are empty. The screen reader just says nothing, which could confuse a user. It takes him a while to recognize the empty cells.
      
   2. he has trouble navigating vertically through the mtable into and out of empty cells.

The testing here was for a desktop computer with Windows 7 or greater, Firefox 48, NVDA 2016.3. With MathPlayer loaded on the computer (which helps NVDA read math properly). This note was also in the write-up of the testing (which is internal and I can't link to):

What AT was used, include version? Any special settings on AT? NVDA (Originally we right clicked on math problem to render Math as MathML, but then we returned to render it as HTML and CSS and got the same results) - MathPlayer installed and NVDA in math mode (Alt + NVDA Key + M; ESC takes you out of Math Mode)

Personally I think my item 2 is an NVDA/MathPlayer issue and my item 1 may be a bug with zooming in on MathJax, or may also be an NVDA/MathPlayer issue. Seeking expert opinions.

[Davide Cervone]

Alex:

Thanks for passing on the results of your testing of MathJax within WeBWorK.  The efforts you all are putting into the accessibility testing are amazing.

I'm not as familiar with the different browser/reader combinations as others in the MathJax team, so perhaps others will have more details than I do.  The usual way that MathJax works with a screen reader is that MathJax inserts MathML into the page that is hidden visually but available to the screen reader in place of the visual representation generated by MathJax.  The screen reader then reads the MathML (for those screen readers that understand MathML).  

As an aside, version 2.7 of MathJax, which is currently in beta test, has the ability to translate math into text strings that can be read by screen readers that don't understand MathML.  See

https://www.mathjax.org/mathjax-accessibility-extensions-v1-now-available/

for details.

NVDA does understand MathML, and so it should be reading the MathML and allowing you to navigate the MathML tree.  Once the MathML is created, MathJax is no longer involved (the zooming is part of NVDA, not MathJax at this point).  So I suspect that your issue 1 is really an NVDA issue, not a MathJax one.  If I recall correctly, however, I think NVDA also is aware when MathJax is on the page, and it may interact directly with MathJax rather than using the hidden MathML.  So it may have to do with that.

For issue 2, you have fallen prey to one of the serious issues with how multi-line equations often are encoded in MathML: as tables.  These are not semantically the correct structure, and does cause confusion for screen readers.  [We faced the same issue when working with our own math-to-speech-string translation, and have plans for deconstructing the tables, but have had to put that off to the second phase of the project.]  I believe you are correct that this is (again) an NVDA navigation issue, as MathJax is not involved once the MathML is generated, and if it doesn't have a way to identify and move through empty cells, that is something that should probably be reported to them.

MathML does have other means of alignment than tables, but MathJax doesn't currently implement them (I don't think Forefox's native MathML does either).  In the long run, having MathJax's TeX-to-MathML conversion use those rather than tables would make accessibility easier, but that is not possible until those alignment features are implemented, which won't be soon.

In any case, we're very glad to hear of your results.  I hope that Peter or Volker will be able to say something more authoritative about your specific complaints.

Davide

--
You received this message because you are subscribed to the Google Groups "MathJax Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mathjax-user...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Peter Krautzberger]

Hi Alex,

Just in addition to Davide's comments:

* NVDA understands MathML only if MathPlayer 4beta is installed (though James recently received funding to integrate Volker's SRE as well, which MathJax uses in its a11y extensions)

* NVDA does not try to detect MathJax and does nothing special with MathJax output (unlike the horror of JAWS messing around in IE).

Like Davide wrote, it would be great if your users had any feedback on the (still relatively new) MathJax accessibility extensions, which do not require a MathML-aware screenreader.

Best wishes,

Peter.

To unsubscribe from this group and stop receiving emails from it, send an email to mathjax-users+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "MathJax Users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to mathjax-users+unsubscribe@googlegroups.com.

[Rachael Verbruggen]

I wanted to thank you for posting this. We've had trouble finding users to test our content for us for accessibility. Although this is only one user and one software, I found this video very informative.

Thanks,

Rachael

[jordanc...@gmail.com]

Thank you Davide and Peter for your replies. (Sorry it's taken me so long to acknowledge these; I've been swamped.)

My takeaway is that the issues are NVDA issues, and even then they are issues we can reasonably expect to be overcome someday. And probably other screen reader setups have similar issues.

I expect that over winter break we will upgrade WeBWorK to 2.13 (if that version is released by then) and upgrade MathJax to 2.7. At that time (or perhaps sooner using our development server) I will try to work with our distance learning office to conduct some testing of MathJax's various new accessibility features. If either of you would like to send me a specific list of things to try, I'd use it. Right now I'm just thinking about the Explorer tool and the speech strings.

[jordanc...@gmail.com]

Rachael, I will pass on your thanks to our distance learning office that conducts this testing. If you check out Karen Sorensen's YouTube channel (https://www.youtube.com/channel/UCPmCILYxA7AuvaYcscJourQ) you will probably find an abundance of end-user testing videos like this one. Karen works in our distance learning department as the accessibility advocate.