TPAC 2016 F2F Minutes

Hi math-on-web CG,

Below are the minutes to the F2F at TPAC.

Given the late notice, the group was very small and we ended up focusing
entirely on accessibility related problems.

@Dani I noticed I didn't represent your example from the whiteboard. Can
you reconstruct it?

Best regards,
Peter.


# 2016-09-20 math-on-webpages CG, TPAC 2016

* Introductions / Present
  * Manuel (Igalia): browser implementations, CSSWG
  * Volker (ProgressiveAccess, U of Birmingham): accessibility interest,
math specifically, also working on MathJax
  * Dani (WIRIS): known for web-based math editor, also handwriting
recognition, putting formulas (any which way), also a11y (but more
rendering).
  * Peter (krautzource): works on MathJax, works on publishing workflow
chainend
* Peter: summarized the activity of the Group
  * still getting to know each other, working on who can contribute on
which areas
  * not about anything like MathML but interests so far are: practical
accessibility, procedural/computational, and incremental layout
improvemenets

## topic: accessibility

* Dani: starting point: we want to go beyond image+alttext
  * exploration, alternative renderings etc
  * if you have MathML you have a tree you can attach to; if you have HTML
or SVG, you also have a tree.
* Volker: our solution applies an "orthogonal" tree to the visual rendering
tree
  * abstracting away the visual layout
* Dani: can we assume a static version of the equation?
  * Volker: we do that to some degree (embed the data structure that
enhances)
    * have done initial exploration for ARIA structure to expose such
information
  * Dani: do you respect the visual output tree hierarchy?
    * Volker: try to but sometimes not possible
    * Dani: do you think it's possible?
      * Peter: not sure it can be. Human understanding is too complicated;
cf. how you visually explore an image/SVG structure. but limitation of the
platform is: tree.
    * Volker: demonstrated some examples from MathJax's accessibility
extensions
      * the data structure is added to the MathML and is preserved in HTML
and SVG rendering
      * while the semantic tree structure is ad-hoc, it seems this wheel
has been invented several times  (conversations with Murray Sargent, Neil
Soiffer etc)
* Dani: I like the idea of a tree that is both semantic and visual
  * can we find something simpler/less powerful but where semantic = visual?
  * [Peter made unhelpful comment until he realized he misunderstood]
* Dani: example: 2+xy
  * <span role="math"> x+2y</span>?
  * but AT might read "ecks too whee", so slap on some aria-labels to fix
that
  * maybe you can add "role=variable" etc.?
    * Peter: cf our DEIMS paper
* Volker: should we have a structure that is independent of the actual DOM
representation
  * e.g., commutative diagrams, chemistry etc have circular semantics that
cannot be mapped to a tree structure
  * Dani: the descendants should be the default descendants
* Peter: there probably is (should be) a connection to SVG a11y
  * e.g., an SVG for a clockface should need some circular component
* Volker: have not followed SVG a11y TF well enough
  * asked Chaals, who said you can't do ProgressiveAccess's style chemistry
exploration with current SVG a11y TF work
* Dani: can we use hmtl trees and references
  * [Dani's example, needs to be reconstructed]
  * Peter: like aria-describedby?
* Peter: what if the visual tree does not match the semantic tree.
  * e.g., \overbrace{a+...+a}{n-times}
  * need to indicate to start at "a"
  * Dani: pointers handle this
  * Volker: so like aria-labledby/describedby
* Peter: what is in the elements in the second tree?
  * Dani if referenced, nothing, but more than that
  * Volker: why do labels and describeby not suffice?
* Peter: I suspect ARIA people will not accept a secondary tree.
  * aria attributes are on the unique tree and, via accessible name
calculation, modify the accessible trees (potentially changing /
overwriting the original content)
  * Dani: how much can you modify this way? E.g., order of the tree
  * Peter: I think you can (vague memory from building menus, specifying
children and, I think their order)
  * Volker: and obviously hacks (Google knowledge cards, taborder, flexbox
direction)
* Volker: have done test implementations to do both two trees and one.
  * with a second tree, you might open up using any other formar (chemml)
  * Peter: this comes very close to custom elements and shadowDOM
* ACTIONS: try to schedule meeting with aria folks