[MathOnWeb] a11y task force minutes 2018-08-27

Hi everyone,

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

Best,
Peter.


# MathOnWeb Accessibility Task Force, 2018-08-27

* Present: Volker, Charles, Neil, Steve, Kevin, Peter
* Peter: last time we basically discussed dictionary-based approaches vs
heuristics
  * Charles: and leaned towards moving away from dictionary since they
failed
* Volker: to rehash my argument: this was attempted several times
  * they did not succeed
  * I think we should challenge our assumption
* Neil: do you have another idea in mind?
* Volker: start with what you currently can do in terms of pattern
recognition and start from there
  * instead of working on better input formats, start with existing input
and see what we can do with that
* Neil: have a program that can do the interference and share that
information with other programs
  * how would you share this without having a dictionary
* Volker: you need to have an automatic extension mechanism
* Neil: Content MathML said 135 and then maybe extend, OpenMath was
infinitely extensible
  * what's the difference?
  * how does any other program take advantage of that?
  * [scribe missed the response]
* Neil: peter could you rehash yours?
  * Peter: much like Volker, but I'm more focused on the other end, the UX,
getting the behavior we want / making it easier for tools to get it
  * Peter: the example from the AIM workshop: server-side generated, highly
enriched content but needs a few lines of JS to do what is a very standard
tree exploration
    * hoping ARIA may be open to discussing this, help eliminate the need
for JS in easy cases (cf aria owns nested lists example, aria tables)
* Charles: aren't there a finite amount of concepts to manage?
  * Neil: to me, the problem is that humans speak the same structure,
depending on content, e.g., people change how they read fractions depending
on the complexity of the numerator & denominator.
* Volker: the bigger problem I see is that the same structure is used for
many things
  * last time we discussed the typical problem of "fractions" which are
used for fractions, binomial, Legendre, logical inference etc
  * Neil: so do we want to build up a table of notation examples?
  * Volker: more: when I come from fractions, can detect it, how do I get
beyond that, build the extension
  * Peter: from my end, this also shows that we don't need a whole lot
there in terms of e.g. exploration features
  * Neil: but isn't there very little variation?
  * Peter: yes, exactly. I was trying to explain how much overlap there's
between my focus and Volker's. I don't think a lot needs to be added to
ARIA to get the little bits of different exploration. The major work is in
what Volker said, identifying the parts.
* Charles: so AT currently would speak a derivative as fraction?
  * Neil: the smart ones won't
    * I don't expect AT to do this but we could provide them with
information
* Charles: so would expressions be thought of as MathML(-derived)
  * Neil: not necessarily
  * Volker: shouldn't depend on representation
* Volker: example: VO voices mfrac linethickness 0 always as ""choose"
  * SRE at least checks for parentheses
    * Neil: right, MathPlayer too
  * Charles: so pattern matching?
  * Volker: e.g., Legendre. It seems impossible to distinguish
    * but still: if you can, how would it fit in into a world where we have
fraction, binomial?
  * Charles: is that the point where we can then insert an ARIA role to
clarify?
    * Volker: not necessarily the author
* Neil: document-level metadata (subject area etc) was always very helpful.
Back in the day, MathType could insert this into HTML exports from Word
  * Volker: yes but again e.g. Legendre won't work because the document
will likely have fractions as well
* Charles: so then we may need something for something as extreme as
Legendre? Fill in the rest with pattern matching
* Steve: if we look at common use cases from literary texts
  * screenreade users read text all the time, mostly fine but every once in
a while, screenreader reads ambiguously, then user goes into char-by-char
mode to work it out
  * if we could give them that, that might be sufficient
  * Peter: trouble is that the information isn't there for the visual user
either, e.g., Legendre can't be identified visually either
  * Neil: in English, 'red/read' is a typical example for regular words
    * looking at Wikipedia for Legendre, there's examples that just can't
be made sense of without context
* Peter: should we gather some next tasks?
  * Neil: maybe focus on example of fraction
    * collect all kinds of expressions with fraction-like constructions
  * Volker: yes, if only to get a feeling for what we're coming for
  * Charles: let's not try wiki tables.
* ACTION: peter to start page and share link in minutes
  * Charles: we can also link to more complex docs if we need them
  * =>
https://github.com/w3c/mathonwebpages/wiki/%5Ba11y-TF%5D-ambiguous-notation