ARIA discussion of Math

31 Mar 2008


See also: IRC log


Al_Gilman, Gregory_Rosmaita, James_Nurthen, Jon_Gunderson, Lisa_Pappas, Michael_Cooper, Neil_Soiffer, Pete_Brunet, Rich_Schwerdtfeger, Kenny_Johar, Marc_Silbey, Matt_May
Janina_Sajka, Sally_Cain
oedipus, Gregory_Rosmaita




<Rich> scribe: oedipus

<Rich> last irc discussion thread on math: http://lists.w3.org/Archives/Member/w3c-wai-pf/2008JanMar/0453.html

<MichaelC> Agenda: http://lists.w3.org/Archives/Member/w3c-wai-pf/2008JanMar/0547.html (first item only)

<scribe> scribe: Gregory_Rosmaita

<scribe> scribeNick: oedipus

math role for ARIA 1.0

RS: trying to get your math request added to aria; think is great thing to add; want to 1) meet your needs and 2) provide best practices for embedding math
... last thread in IRC

<Rich> http://lists.w3.org/Archives/Member/w3c-wai-pf/2008JanMar/0453.html

AG: 19 March 2008 -- you replied, but bunch of technical details

RS: new math role -- what does it mean? how to use? how to reuse ALT text? how to handle images?

NS: brought up with MathML WG -- asked if had comments; some of things we discussed in PF email exchange came up -- restrictions on what type of math should be placed in a math role; MathML WG felt that a lot of stuff out, for past compatibility not good to abandon what is being done today and historically -- mix of TeX and MathML
... none of them are tagged identifying what they are; existing practice wide-open and that needs to stop
... sniffer trying to find math will be able to tell TeX, MathML, or similar syntax
... leave open-ended and not specify
... second: subtype "math-tex" -- discarded as bad idea -- to ambigous; too many TeX variants
... third issue: if math, might not be any alt text in suitable format, but down the road one could use OCR to pull equation from image -- already an OCR for math with acuracy in 97% range; again, not tagged, but open ended
... fourth: didn't care where it is -- BP can say "use alt text or use image" but one kind of math out there is JSMath -- uses TeX-like base to create javascripted (inaccessible) math
... don't want to force math into an image that has to be parsed
... no single practice -- math kind of a "wild west" frontier trying to get integrated into HTML

RS: what does having role="math" help with all formats?

NS: key thing which will allow anyone to do a11y -- 1) given cost of trying to discover math, having tag iding as math, keys a sniffer/handler to find and discover format; second) alerts AT that this is Math and calls a handler to come up with text to speak it, rather than gibberish

we are in #aria-math for math discussion

switch to #aria-math

NS: MathML WG felt that using attribute value "math" too general (plot, diagram, formula) want to restrict it more -- suggested "formula" or MSAAEquation

AG: MSAAEquation was the suggestion

RS: if prefer math, can in API mapping documents map to role="equation"

NS: not as picky as rest of wg, but thought "equation" restricted it and was more precise and maps well to what already exists

AG: a) have to channel simon p -- was pressuring us to have this fully specificied -- thinking that he is building into HTML5 and browser -- challanged that -- this is a flag AT uses to call helper -- it is the helper that decides how loose or strict it is going to be able to grok and that above 85% success would be BIG win
... where existing equations are in TeX can do that
... akward thing in that HTML has history of getting into trouble with putting sloppy definitions in specs which lead to differing/sloppy implementations; authos have been upset about what happens when their code hits the browser; strong desire in HTML WG to nail things down better and have more specific specificiation -- need to distinguish what role is here -- HTML and base UA not going to do everything for everyone

NS: math may be embedded in image -- may not even be perceptable to HTML parser

AG: SimonP would want to be quick -- if you are happy to sniff/ping inserts and give a sign "math handlers/sniffers" go find and process this; in accessibility also have problem -- in WCAG developed concept of accessibly supported technologies -- loose outer defninition (function performance) then "and furthermore if use this technique with this technology, value gets through to end user:


AG: value of "math" or "equation" is describing intent -- how you should understand this fragment of content;
... when tell authors what to do, need "and furthermore if you do this, it is supported" so both a loose and a stricter statement

<kenny> michael, I am using skype for this call today.

NS: should be suggestion in Best Practices that says "this technology will be exposed/found if you do x with y app" -- would not leave open ended if starting from scratch, but since ARIA attempting to capture meaning in current practice and those practices vary greately

RS: mark this as "equation" here are common BP to support interoperability today, but is there an optional parameter for "equation" you would want?

NS: 2 things: 1) where is the math going to be found; 2) give me a strong clue as to what type of math i need to deal with (self-discoverable)
... with TeX similar, but different -- one thing not well understood about TeX was written without logical parser -- doesn't tokenize things but tokenizes each character
... some variants of TeX tokenize elements; things like that make TeX an open ended thing
... would be nice to specify that, but 20 variantsof TeX and many implementations follow custom rules

AG: took first pass at approach: didn't address "casting to a dialect" but did point to where can find; put encoded math in ALT attribute one pattern has to be looked for (proably a search list); what is new in ARIA is "describedby" points to something for people or machines -- could point to a machine understandable -- could develop conventions for TeX -- ways to put into meta data (use SCRIPT that not displayed, ask what type of script, get return) -- hav

RS: describedby like a long desc for area in document
... can hide with CSS or have visible; gets mapped to accessible description in a11y APIs so one can extract text if needed;
... suggestion on how to process math object? OCR it, use description (describedby)
... can't specify TeX because too many variants

NS: right
... question i have which hasn't been answered by PF is "how important is compatibility with existing use cases?" -- is goal of ARIA to change and constrain use of math to make more accessible or just trying to describe current usage

RS: ARIA about interoperability with platform a11y APIs -- everything needs to be obtained through APIs; could say, if use ARIA, this is how you should use it for math -- if in best interest of industry -- use of ARIA might direct them to a more consistent use of math on web

NS: driving factor on some, but not all, putting math in HTML is terseness; some use TeX-like notation that uses javascript to "prettify" -- wikipedia/wikimedia has extension that creates TeX from ascii notation or to describe image containing math
... JSMath -- if javascript really doing writing, could put out more description
... maybe verbosity is not an issue on second thought -- tool will generate extra stuff

AG: wikipedia case where site doing patern driven transform is good likelihood of uptake thing -- getting JSMath library to generate more stuff would rate down

NS: working with us

AG: maybe will work with us like dojo
... everyone in end cares about character count -- not just mobile providers; wasted chars not good idea -- if put TeX in alt and mark with role then should work -- content devlopers who want to put in different flavors of TeX, that we have the "try harder" coding patterns we target for when more than just an image and alt text

JRG: ideally what is most accesible way to put math on web

NS: MathML -- works with XHTML no problem, but problem with HTML -- if you only use IE, can put in HTML document and IE extention mechanism will handle it, but nothing in firefox or safari

RS: role="equation" - source equals "URI for mathML" and let AT suck it out

NS: URIs have 2000 character limits?

AG: good question

RS: here is URI for source of that, don't need alt text -- if you understand MathML go get it

AG: problem when src used to point to math image

RS: can handle that

AG: HLink versus XLink wars

NS: essentially 1 thing saying let's make it explicit and put on image or embed tag -- MathML WG trying to make math more interoperable -- people carry around images, but leave alt text behind; embedding in image safe way to go -- always there
... extractable from image, not dependent on alt -- TeX part of image
... not bad idea to have attribute that make these things obvious, but worry about fragility of extraction method

RS: like to have 2 vehicles in BP: role="equation" -- in BP say "store equivalent text in image"

NS: here are 3 or 4 ways that are supported by most extraction techs
... 1 would be ALT text (current practice); other would be using "describedby" and other in image itself

RS: no URL to mathml equiv

NS: not sure

AG: should be supported

GJR: strong plus one

AG: if provider has equation in MathML, need to know -- want to use that if present

RS: mathML equiv to give URL (optional parameter)

NS: why just MathML? if to make discovery easy, shouldn't be limited to MathML -- just URL of equivalent

RS: server could tell media type when pull down TeX

NS: is a text format

MC: datatype or new mime-type property?

AG: has to be tested to make sure current implementations of OBJECT won't break; if put in OBJECT and make MathML first choice, that is the Best Practice possible; problem is whith mime-types for MathML

NS: persued and rejected by MathML WG

AG: problem: UA can't determine ahead of time if can process MathML -- doesn't know MathML is MathML until gets it

RS: right

NS: sigh, right
... though OBJECT has own problems

AG: matter of debate -- tantek told us it has been fixed; HTML5 faction against it (VIDEO, AUDIO, etc.) -- shouldn't lose cascading fallback idea; want to wait to have highest and best form avialable
... issues: need mime-type identifier; need extractor for TeX in image

<kenny> Do we know if any screen readers today support LaTex?

MC: putting mathML in attributes means using escape characters

AG: TeX in alt rather than MathML

<kenny> Thanks gregory.

AG: if load into IFRAME, how is communicated to UA that this thing contains math -- application/xml+mathml -- if can't process XML move onto something else

NS: not going to work right now

GJR: MC can WCAG2 address the author's best practice by advising that TeX be embeded in an image

AG: interim technique; also want to fix the web so highest and best math (MathML) can be documented as object and served to users

NS: MathML in HTML5 -- consensus that math will be in HTML5 -- probably MathML in some as yet unkown/specified form
... want to discourage hacks

AG: agree -- just trying to separate 2 issues: need role now so something can be processed and

GJR: need to take info to WCAG and ATAG

RS: if alt text not in document, are you going to pull alt text out of image and map to a11y api?

NS: don't think UA itself would do sniffing (although would be great if would) -- UA's not the ones that turn MathML into speech, AT vendors don't want to handle MathML directly, so would have to use Expert Handler (http://a11y.org/handlers) -- there is pllugin for IE that allows speech -- AT essential to getting braille support
... developed tech to go out and pull the alt text out, paste into equation editor; port that back into viewer -- take the TeX translate to MathML then run through MathML to Speech and MahtML to braille engines -- very doable; current limitation -- plugin calls for MathML and need to do javascript on imagesand scan of page to determine what math is there in what form
... if have role="math" could limit search

AG: summarize: if the user has a png format for their picture of equation and put embedded meta-data in png, not expect browser to put into API as accessible description -- can take alt text and processed as usual; if mathML in embedded, math helper might write good description into accessibility API

NS: ARIA exported to AT, javascript that binds will do translation and in case of UAs not serving up dom, can rewrite page AT should pick up
... alt text put out as value field in MSAA?

RS: no, AccessibleName and helpInformation -- MSAA has description API (equiv of help) IA2 has relationships that understand "describedby"; don't know what UIAutomation plans

NS: whatever written, write correctly, even if in alt

RS: wearing AT vendor hat, if TeX embedded in image, would want to know that if don't have TeX plugin

NS: what is loud and clear from AT vendors is "we're not dealing with math; you do that, tell us what to put out, and we'll support it" not going to search DOM, will speak if is there -- don't want to fetch image and parse it

RS: 2 things: 1) either can put a describedby as technique or a tool can fulfill that role; 2) HTML5 people will not like URL pointing

NS: HTML5 people will tell you that they will handle math as text/html -- dubious at this point if possible
... describedby field in TeX or MathML -- if had javascript that looks for math and outputs speech, is it still "describedby"?

AG: describedby takes list of IDs -- have to revisit what UAs do with concatonated methods
... can't do this on fly, but are things can write into DOM that can capture (SSML)

NS: with MSAA can't use SSML, have to use straight text
... as long as you come up with soulition that describedby is machine field, need to id field that needs priority to generate speech, braille, etc.

AG: hoemwork item, but this discussion has GREATLY cleared up the issues; thought going to take directoy to APIs -- if put back into DOM, have to rethink thorugh fully so AT ends up with it

NS: found varying levels of commitment from AT vendors: base level - have MSAA interface, but that is limited to string of text (can't put braille in that -- text you put in isn't braille you get out in math) -- that's why have to expose handler interface to all UAs all ATs all platforms; minimum base is ability to take math code and output speech

RS: braille involves fight with AT

NS: we dont' speak things, we return text for speech and braille engines

AG: take back to ATs -- knee-jerk reaction is that that is part of procedure call on dispatching expert handler -- should know if have braille device and follow user's preference; Fluid Project beating on W3C elsewhere, so will be a W3C user prefernces in context that can be exposed

RS: don't want future equivalent for MathML
... "equation" or "math"

LP: preference for math -- it's shorter

NS: MathML prefered equation, personally don't care

AG: can live with either -- math good idea now -- authors will relate better to math - will think equation doesn't always apply

RS: can say in description "this is a math equation"

NS: not equations, expressions, but...

RS: math expression in definition

<Al> +1


<Rich> +1

<kenny> +1

LP +1

<Jon> +1

raman gave +1 to "math" on list

RESOLUTION: ARIA will use math role type; in definition will state that this is a "math expression"

rssagent make minutes

RS: describedby a universal property, right?

MC: yes

RS: don't have to say antyhing explicit in spec about math, but a BP thing; have to compose best practices section 1) alternative text embedded in image, third party app extracts info and load "describedby" with that info, or may speak, or author may but in "describedby" himslef

NS: describedby -- words to speak or what should be written into DOM using existing API bindings

AG: need to get AaronL (mozilla) and SimonP (opera) in on this

RS: best way to speak it (as long as not changing voice dynamics) == that way AT doesn't have to think alot to deliver to TTS or braille device

NS: machine encoding shouldn't be part of "describedby"

AG: how does API binding respect cascade of acceptable forms

NS: someone could manually use field -- need to know if field exists

AG: action/homework item: UA best practice for AT helper (expert handler -- if writes back into DOM, how to do it)
... need browser brothers to address this

RS: if start putting other markup in there, going to go bonkers

AG: are we going to say if TeX in image and can figure how to say that can overwrite ALT and go with that -- haven't been considering that piece of problem

NS: 2 ways: 1) we say "override ALT or describedby" and put in description AT can pick up: 2) tighter binding -- id as Math -- call expert handler, and that will tell us what to say -- that fall back position is needed -- may be rewriting ALT or describedby -- should process, put back into DOM

AG: for PF, in terms of browsers binding DOM to API, much plainer if return to AT that called handler than figuring out override; don't want to overwrite something suitable to speak that is only handled by handler; handler api has to return best advoice on how to speak or braille expression

RS: what is best practice: how does an AT know that an expert handler/plugin (MathPlayer - works only in IE)

NS: have to work with each AT vendor to tell them "when you get to this node, use our interfaces" when you get to our interface, then speak, highlight, braille-output
... plug-in a dll that gets into registry
... want standard way that says "here is how you work with expert handler" rather than one-off solutions

AG: browsers look like going to implement ARIA, will be refelected in accessibleRole in API -- if use a11y API can bootstrap off of that, determine if math and then ask for help

RS: preferential way to look at describedby or alt -- that would be what is to be spoken

NS: JavaScript solution is a cross-platform base fallback -- can't give highlighting or braille currnetly; if could write SSML into math speech, would be much better, but if stuck with plain text, fake it with commas and periods and other klduges

AG: small SSML implementations not available yet

RS: if use describedby method should be processable in speech

NS: exactly what i did with MSAA/SAPI5 -- value field with plain text
... like RS' solution

RS: need to put something in best practices -- don't need to do anything special with UA devs

AG: have in our vocab "valuetext" -- if make "valuetext" apply to math role, have place to stuff it

RS: [ponders] --

NS: can it be a long string?

RS: yes
... currently valuetext associated with valuemax valuemin etc.

AG: document has tools to deal with it attribute by attribute -- math role has this property: "valuetext" defaults to how to say, can also put in how to braille

RS: in HTML what is maximum lenght of string

AG: have to look for it
... generic limit on strings 255

MC: anything directed towards human should be in an element

RS: between 2 tags can be as long as want

AG: that points to describedby over valuetext -- problem: have to work with AT so that role="math" goies straight to describedby and prcess it in its entirety

RS: think have enought to write something up

ACTION ARIA Editors: write up proposal for role="math" and best practices for its use

<kenny> bye

NS: happy to participate as needed

Summary of Action Items

[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.133 (CVS log)
$Date: 2008/03/31 21:54:52 $

Scribe.perl diagnostic output

[Delete this section before finalizing the minutes.]
This is scribe.perl Revision: 1.133  of Date: 2008/01/18 18:48:51  
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: RRSAgent_Text_Format (score 1.00)

FAILED: s/user;/user;/
Succeeded: s/hel/help/
Succeeded: s/soultiion/solution/
Found Scribe: oedipus
Inferring ScribeNick: oedipus
Found Scribe: Gregory_Rosmaita
Found ScribeNick: oedipus
Scribes: oedipus, Gregory_Rosmaita
Default Present: Gregory_Rosmaita, Cooper, Rich_Schwerdtfeger, Al, +1.512.238.aaaa, James_Nurthen, Pete_Brunet, Matt_May, SotosBarkas, Neil_Soiffer, Jon_Gunderson, marcsil, Kenny_Johar, +1.919.523.aabb, Lisa_Pappas
Present: Al_Gilman Gregory_Rosmaita James_Nurthen Jon_Gunderson Lisa_Pappas Michael_Cooper Neil_Soiffer Pete_Brunet Rich_Schwerdtfeger Kenny_Johar Marc_Silbey Matt_May
Regrets: Janina_Sajka Sally_Cain
Agenda: http://lists.w3.org/Archives/Member/w3c-wai-pf/2008JanMar/0547.html
Got date from IRC log name: 31 Mar 2008
Guessing minutes URL: http://www.w3.org/2008/03/31-aria-math-minutes.html
People with action items: 

WARNING: Input appears to use implicit continuation lines.
You may need the "-implicitContinuations" option.

[End of scribe.perl diagnostic output]