Accessible Platform Architectures Working Group Teleconference

24 Jun 2020


janina, Matthew_Atkinson, MichaelC, jasonjgw, becky, Irfan, JF, NeilS, Joshue, CharlesL


<scribe> scribe: Matthew_Atkinson

Agenda Review & Announcements

<janina> akim, who's here?

Janina: Welcome, Nicolò!

Task Force Updates (including COGA CfC Update)

Janina: We're organising a meeting about different architectural approaches to pronounciation; need Joanie to ascertain Joanie's availability.

Irfan: As we're revisiting our technical approach document, we may need to revise the timeline.

Janina: Generally happy with the timeline. Need to ensure concerns are resolved and agreed by all, but shouldn't delay things too much.

<MichaelC> https://lists.w3.org/Archives/Public/public-apa-admin/2020Jun/0000.html

Janina: COGA CfC is out for "Making Content Usable for People With Cognitive and Learning Disabilities" https://w3c.github.io/coga/content-usable/
... Personalization CfC is approaching.
... Lots of activity in RQTF. The topic of Web of Things has been revisited. The UI can't be assumed to be accessible (sometimes this is not the case).

<Joshue108> Unified Smart Home Control and Status Interface

<Joshue108> https://github.com/w3c/wot-usecases/blob/master/USE-CASES/mmi-1-2_unified-smart-home-control-and-status.md

Janina: The Web of Things interest group has a couple of use cases that relate to accessibility and are asking for input on how critical they are. RQTF concluded that they are critical.

<Joshue108> Audiovisual Devices Acting as Smartphone Extensions

<Joshue108> https://github.com/w3c/wot-usecases/blob/master/USE-CASES/mmi-1-1_audiovisual-devices-as-smartphone-extensions.md

Janina: WoT has a survey out to which we'll need to respond.

<Joshue108> https://w3c.github.io/wot-usecases/

Janina: *introduces the idea of accessibility being "Essential for some and useful for all" (from the EO group originally)*

JF: What is the research group's perception of the intersection between WoT/XR and WCAG?

Jason: The focus is on understanding the scenarios provided by (in this case) the WoT group. If concerns about WCAG conformance arise, they're directed to the WCAG group.

MathML Update -- Neil

<NeilS> MathML Primer: https://mathml-refresh.github.io/mathml/docs/accessibility

Neil: Going to talk about Math (including science, technology, engineering, education, ...) - what it needs for accessibility and why it's different to accessibility for text.
... it's ubiquitous, so really needs accessibility consideration.
... Math is about conveying _concepts_, not just the words on a page. So translation to different modalities is non-trivial (and may differ).
... Also speech engines don't support math, so speak it poorly.
... e.g. inappropriate pauses/prosody; letters not being read as individuals (or announed incorrectly) when that's required.
... How this relates to the web: you may want to say "this is how I want this to be spoken" - bu that's not possible with aria-label nor live regions. MathML in a live region wouldn't be picked up.
... (nor SSML)
... It's also important to know "who's listening?" e.g. if someone is blind, they may need to know where things start or end (e.g. numerator/denominator) but someone looking at the equation would see this.
... Including "start" and "end" markers for parts of the equation could however really confuse someone who's dyslexic and doesn't see those markers on the screen.
... Level of expertise comes in too: experts tend to shorten things. E.g. in calculus, you may say "the derivative of y with respect to x" but someone familiar with the subject would say e.g. "dy dx".
... Meaning is conveyed by position _and_ value e.g. superscripts indicate powers. x^4 may be "x to the fourth" but x^2 would be "x squared".
... There's operator overloading too. (absolute value vs. determinant).
... A big expression is too much to take in at once. Navigation and the ability to have an overview (which maybe elides certain parts) is important.
... On the web you can't have nested interactive content, which presents a challenge.
... Also, for some learning disability tools, the words are highlighted as they are spoken; this is also useful for math.
... Now an overivew of MathML. Two parts: "Presentation" MathML which is about how things are laid out visually. "Content" MathML is about semantics e.g. "power" instead of "superscript".
... the generators/authoring tools generally produce content MathML. You can mix the two, such that they are linked, but that is extremely verbose and complicated, so content MathML is rare, and mixing both of them is very rare.
... There's a MathML refresh group that's working on encouraging browser support and imparting more semantics into MathML.
... Another approach is providing context such as the subject area. This can help resolve the ambiguity caused by operator overloading, for example.
... This doesn't always help though; e.g. things in parentheses could be coordinates, or intervals, ... so attributes are being added to allow people to indicate what is meant.
... Next topic: workflow. How does MathML get delivered and where do responsibilities lie?
... Authors may use WYSIWYG tools or TeX/LaTeX, these mostly describe what the math looks like. They may provide templates for different units of meaning (so the semantic attributes could be hooked in here). For TeX/LaTeX, the names of macros often impart some meaning, particularly for established sets of macros, so that's an avenue for including semantics too.
... The hope is to include more semantic info in the presentation MathML that is generated.
... this is passed to the AT. The AT doesn't know the subject area etc. so the author needs to specify that. The author doesn't know the user's needs (e.g. which type of math Braille they use, or other modality settings) so that falls to the AT.
... the primer gives a number of examples of issues around generating the speech.
... Ambiguity example: a horizontal line over something may mean: conjugate (in algebra); a line segment (geometry); mean (statistics); not (logic).
... this would present a big burden on AT to get right; if authors provide the info it's much simpler.
... AT (e.g. screen readers) support MathML, but with varying degrees of sophistication (e.g. they may say "square root" and "cube root" as special cases but not many other rules). NVDA uses MathPlayer which has many rules, and leads to more sophisticated output, but this requires a lot of community effort. Output improves dramatically when the subject area is known.
... In summary: text and math have different issues; authors need to provide semantics so AT can help; ATs have responsibility to generate the correct speech or Braille dialect as output.
... Please check out the primer for much more detail.

Matthew: *mentioned an experiment I head of where the structure of the equation is conveyed via prosody, but the details obscured*

Neil: We often think of conveying the structure as an overview as beneficial but users tend to not like it.
... Hence the inclusion of start/end cues; varying speech speed; auditory icons for markers (rather than speech).
... generally users preferred the words.

Nicolò: is there support in the browser for reading MathML with the web speech API?

Neil: Not much has been done in this area; most people would need math accessibility use AT. The web speech API would have limitations of haivng to make sure that things were proncouned correctly etc.

Nicolò: *Is aware of some work going on in this area; will send details*

Janina: Work is going on a W3C recommendation on how to pronounce certain things (including math) via the Pronounciation WG.

<Irfan> https://www.w3.org/WAI/APA/task-forces/pronunciation/

Neil: The MathML still needs to be turned into words, which is challenging due to the different ways people would speak math, and needing to know the semantics. You could take a very low-level approach and read things like "vertical bar x" but "absolute value of x" (or other meaning, depending on context) is easier to understand.
... the aspiration of our work is to provide tools that will help AT make that translation based on inputs such as subject domain, user expertise and requirements. Some experience on this via MathPlayer.

Janina: what can APA do to help the community group achieve its goals? (Next question, not for today, but wondering what sort of architectural support is going to be needed)
... *ACKs the inclusion of COGA*

Neil: We have Chemistry examples too (check out the primer).

Janina: we should get some of these examples into the pronounciation example (e.g. "doctor" and "drive" being mixed up is one thing, this is a bigger problem).

Charles: This is very interesting for the publishing WG (Chemistry work is going on in parallel within our TF).

Matthew: has a tree-based approach been useful?

Neil: for navigation it is (e.g. exploring a fraction); overviews in some systems elide parts of the tree.

Janina: thanks Neil!

Summary of Action Items

Summary of Resolutions

[End of minutes]

Minutes manually created (not a transcript), formatted by David Booth's scribe.perl version (CVS log)
$Date: 2020/06/24 17:03:04 $

Scribe.perl diagnostic output

[Delete this section before finalizing the minutes.]
This is scribe.perl Revision of Date 
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00)

Succeeded: s/udiovisual Devices Acting as Smartphone Extensions/Audiovisual Devices Acting as Smartphone Extensions/
Default Present: janina, Matthew_Atkinson, MichaelC, jasonjgw, becky, Irfan, JF, NeilS, Joshue, CharlesL
Present: janina Matthew_Atkinson MichaelC jasonjgw becky Irfan JF NeilS Joshue CharlesL
Regrets: Becky
Found Scribe: Matthew_Atkinson
Inferring ScribeNick: Matthew_Atkinson
Found Date: 24 Jun 2020
People with action items: 

WARNING: IRC log location not specified!  (You can ignore this 
warning if you do not want the generated minutes to contain 
a link to the original IRC log.)

[End of scribe.perl diagnostic output]