Re: WCAG

Further to my last message, Wendy and I agree that the log of our
IRC session yesterday should be of interest here,  It deals with
WCAG, EARL, and dealing with heterogenous sources of information.
Among other things, it expands on the thoughts behind my last post.

-- 
Nick Kew

Site Valet - the essential service for anyone with a website.
<URL:http://valet.webthing.com/>

++++

<nick_kew> Have some thoughts; wondering whether WAI/ER is the best forum
<wendy> best forum for your thoughts?
<nick_kew> Developing a parser based on WCAG
<nick_kew> - and reports
<wendy> ooooh. sounds interesting. what are you thinking? (or is that what you are composing?)
<wendy> :)
<nick_kew> well, thinking as I write ...
<nick_kew> WCAG isn't going to be as easy as a formal spec to work from, so I'll need to discuss it at length (I think)
<wendy> cool. i look forward to it.
<wendy> what would the parser parse?
<wendy> e.g., if it is based on WCAG is it parsing HTMl for alt attributes?
<wendy> e.g. language X for Y attribute on Z element?
<nick_kew> I'm thinking it through.  It'll have to extract tags.  One option would be DOM (preprocess HTML->XML), followed by set recipe of per-tag checks
<wendy> have you seen the schematron?
<wendy> http://www.ascc.net/xml/resource/schematron/WAI-example.html
<wendy> you could use HTML Tidy to help get the HTML->XML
<nick_kew> Tidy: possible, but I had in mind SX (see Code Valet)
<wendy> ok. i'll have to check it out. i'm not familiar with SX.  It does the same type of thing as tidy?
<nick_kew> schematron: been on my TBD list for far too long
<nick_kew> No, it doesn't correct anything; just transforms SGML->XML
<wendy> ah. cool.
<nick_kew> It's from SP (James Clark's toolkit)
<wendy> re: schematron: it would be cool if schematron's report produced earl.
<wendy> i've wanted to play with, but no time.
<nick_kew> metoo
<nick_kew> Someone suggested schematron when I talked about Validating for Accessibility several months ago
<wendy> it's a cool thing. with a bit of work it could be really, super wonderfully cool. :)
<wendy> it tries to encode WCAG, but messes up on a few things.
<nick_kew> (and still haven't found time to follow that through :-( )
<nick_kew> hmmm ... must revisit
<wendy> it also is fairly limited, but w/some tinkering, I think it could do more.
<nick_kew> hmmmm .. needs thinktime.
<wendy> yes. so, i look forward to your message. i ought to correct some of the errors that i noticed.
<wendy> (dan c. published a version on the w3c site and asked people to keep working on it)
<wendy> but, i doubt i'll have time to do that b4 i hit the road on thurs.
<nick_kew> OK, stopped thinking and hit send now
<wendy> :)
<nick_kew> (talking to you here and to #validator simultaneously :-( )
<nick_kew> EARL:  *what* should it report?
<nick_kew> (opens can of worms; ducks rapidly)
<wendy> :)
<wendy> i'm not sure what you're asking.  earl should report how well something meets a spec, whether it be WCAG, UAAG, HTML, SVG, etc.
<wendy> but, I sense you are getting at something else. :)
<nick_kew> WCAG is more interesting, because you can't YESNO much of it
<wendy> if you ask specific enough questions you can y/n, can't you?
<nick_kew> No, I mean Site Valet's WCAG tool can point to 'alt="bullet"' and say 'are you sure you don't mean alt="*"', but no automated tool can say for certain that it's wrong
<wendy> right. so, when you ask, "EARL:  *what* should it report?" are you asking what it should report for the example you just gave?
<wendy> i.e. 'are you sure you don't mean  alt="*"'
<wendy> if so, that's an interesting question.
<nick_kew> My proposal: my tool reports "I attach a probability of 50% to the hypothesis this is wrong"
<wendy> hmmm.
<nick_kew> A human evaluator can then override with "I say with 100% certainty it's right"
<wendy> in other words: automatic check result could be yes, no, or "please check it for me"
<wendy> right.
<nick_kew> - or another automated tool can say "probably wrong" and we aggregate weights
<wendy> right.
<wendy> but, how would you determine weights?
<wendy> would you allow complete spectrum of 0 to 100%?
<nick_kew> But there's more: we can then accept human inputs from the public, and assign them weights
<wendy> weighting gets sticky. if human's are rating their own, i think we can only say 50% and 100%. Since I think humans interpret percentages differently. :)
<nick_kew> - on the premise that a comment from the public ranges from accurate to stupid
<nick_kew> No, not what I mean:
<nick_kew> * have a feedback mechanism
<nick_kew> * visitor says "this doesn't work"
<nick_kew> * inference engine weights comment
<nick_kew> * weight of advers comment reaches threshold; server notifies maintainer "you have a real problem here"
<nick_kew> (but visitor is never asked a percentage - too subjective)
<wendy> right. We could give people a scale, such as, "choose the phrase that best fits your experience:
<wendy> I can not use the web site at all.
<wendy> i can use it with some difficulty
<wendy> i can use it with ease.
<wendy> "
<nick_kew> Possibly.  More discussion and real-life testing wanted.
<wendy> sure.
<nick_kew> But I'd like to build a testbed
<wendy> what would you do w/the earl? make it an annotation?
<nick_kew> Tell you that when I've reviewed sbp's latest
<wendy> (assuming EARL could express something like the above)
<wendy> ok. although i asked him this morning on the call if he thought 1.0 would effect where/how earl is stored and he said "no."
<nick_kew> OK, I'm asking for it to be able to express uncertainty, possibly with numeric values
<wendy> yes, understand and I don't think that is in 1.0 yet.  regardless, deciding how/where to store and access earl is something i'm curious about.
<wendy> it sounds like an annotation is the best fit for the scenario you just outlined.
<wendy> (user rating a site for accessibility)
<wendy> (sending advise/feedback to author)
<nick_kew> Could be.
<nick_kew> Valet runs with an old-fashioned SQL backend, but the interesting question is what the next layer looks like
<nick_kew> Re: testbed - could this in principle become a WAI project?
<wendy> re: next layer. not sure what you mean.
<wendy> re: testbed - do you mean hosted by w3c or using WAI resources to develop or both?
<nick_kew> Data representation
<nick_kew> Either or both
<wendy> re: testbed - sure. we have something that is really basic already (and i think completely unused).
<wendy> refer to:
<wendy> http://www.w3.org/WAI/report/
<wendy> basic report to send to authors about site.
<wendy> I wrote an implementation plan to dress it up a bit:
<wendy> http://www.w3.org/WAI/ER/WART/plan.html
<wendy> This plan keeps expanding. primarily will be used like the ATR
<wendy> (http://www.aprompt.ca/ATR/ATR.html))
<nick_kew> Ah - I saw that a while ago.  The idea is it doesn't take comments as one-off, but aggregates evidence
<wendy> right - that's the real beauty of EARL.
<nick_kew> Starting point could be authoring tool's own report
<wendy> It would be useful if ATR aggregated info, but it only produces it.
<nick_kew> + other automated tools
<nick_kew> ...
<wendy> yes!
<nick_kew> An altogether more ambitious project
<nick_kew> - but an interesting one
<wendy> yes, but one we've wanted to see come to life since the beginning.
<wendy> right.
<wendy> it's one of the reasons for EARL: taking output from Bobby, A-prompt, siteValet, etc. and letting the user
<wendy> compare results
<nick_kew> [if only I could get a nice job in academia ... :-(((( ]
<wendy> :(
<nick_kew> Would it be worth posting this discussion to wai-er-ig for review?
<nick_kew> (some of it adds to my post)

Received on Tuesday, 30 October 2001 15:28:40 UTC