[math-on-web] a11y TF meeting minutes, 2018/08/13

Hi everyone,

Here are the minutes from the meeting of the accessibility task force this
week.

Best,
Peter.


# Accessibility TF meeting, 2018/08/13

* Present: Dani, Peter, Charles, Volker
* Charles: after last meeting woes, I have a zoom set up
  * Peter: does it need an owner present?
    * Charles: doesn't need to
    * Peter: great!
  * let's start with a11y TF and then see
  * ACTION Charles to send out invite
* Peter: Joanie had begun to work out a space on the aria end for the
proposed states/roles/properties work
   * that will allow us to work collaboratively in their space
* Dani: let's start the work
  * Charles: how do we start? Do we go through simple math and list out the
things we see?
  * Dani: that's one way but we need a basic strategy to work it out
    * e.g., plus sign. What does it mean? It can change a lot from context.
    * we could look at content mathml, open math dictionaries
    * but looking at 1+3-2
    * do you need at tree or flatten it?
  * Charles: but if you want to provide overview, then you'll need the tree
    * to let AT give overview
  * Charles: unless you want to provide an overview of an equation, which
may be relevant but perhaps start with content mathml / openmath and then
go from there
* Volker: we provide an overview with MathJax and don't need any content
mathml etc
  * why do we need this level of semantics?
  * Dani: last meeting Joanie proposed to collect roles/state/properties,
we thought we start with previous work
* Peter: just for the record and transparency, I'm not convinced this is
the right approach but I want to support those who pursue it
  * I think we need to tackle this from the other end: what does AT do
today? Mostly it's heuristics and mostly it's terrible.  How can content
providers help them do a better job? how can specs make their existing
process easier?
* Volker: imho I think we need a rational reconstruction
  * dictionaries have been done many times and always fail because we don't
have the content
  * I think we should look at the content that we have and how to get more
out of it
  * we should work on the bridge of the representation
  * the dictionary can come out of what we can currently get out of
heuristic recognition
  * then we can re-write it to something that we alreayd have in the
dictionary or write something else
* Dani: I'm thinking about tagging this manually but you think more about
the automatic generation
  * Volker: right
* Peter: I'm much closer to Volker. I want to work with real world tools
and thus the real world content they digest.
  * help devs of AT with equation support to do a better job
* Dani: heuristics work but it's not perfect and that is one problem
  * those are two different problems
  * setting a closed set of properties or helping heuristics
* Volker: it's mathematics; it canot be closed.
* Charles: re Peter. why do, e.g., aria-labels or extra roles, not work on
MathML? Is it just a a spec gap?
  * Peter: it's because the heuristics cannot make sense of a label or a
role. E.g., on an mfrac, some random role will throw a wrench into any
heuristics that try to identify if it's a fraction, a binomial, the
legendre symbol etc.
  * Charles: so roles could clarify that
* Peter: another example is a role on the + in 3+2-4 ; if we make it a role
of something random (say, integral) what will happen? How would AT know how
to voice or navigate?
  * how will the navigation work? how could AT decide?
  * currently, they can only understand MathML, ignore everything else.
  * I often come across such cases in real world content production.
* Volker: subject area will influence this. E.g., the walk on 2+3-5 may be
distinctly different in different areas
* Dani: would roles be enough?
  * Volker: you'd need more than roles
    * e.g., the light walker we built uses aria-owns etc
* Peter: states and properties are ARIA
  * but not exactly "mathy" or STEM, more general
  * Dani: but if we need it, we need to talk about it
  * Peter: sure. Just cautioning that it may feel off topic to ARIA when
the starting point was feature parity with MathML

* Peter: Dani & Charles: how can we get you started with your approach?
* Charles: uncertain. Whenever you try a single expression, you get
different results in different AT
  * unclear if its ambiguity or bugs
* Volker: happy to try things out that you throw at SRE
  * e.g., change operator precendence
* Dani: I think we should start with concrete examples
  * basic examples
  * be optimistic, make suggestion how to make htem better
  * Charles: I like that, coming up with example that are ambiguous
* Peter: do you need anything beyond https://github.com/w3c/mathonwebpages/
?
  * Charles: is this on github.io?
  * Peter: yes. Just make PRs