- From: John Foliot <jfoliot@stanford.edu>
- Date: Wed, 28 Apr 2010 22:15:47 -0700 (PDT)
- To: "'Laura Carlson'" <laura.lee.carlson@gmail.com>, "'Sean Hayes'" <Sean.Hayes@microsoft.com>
- Cc: "'Janina Sajka'" <janina@rednote.net>, "'Matt Morgan-May'" <mattmay@adobe.com>, "'HTML Accessibility Task Force'" <public-html-a11y@w3.org>, "'Gregory J. Rosmaita'" <oedipus@hicom.net>
Laura Carlson wrote: > > I tweaked this a bit more [1]. It now reads: > > <draft text> > > Outcomes of Creating a Missing Attribute > > A missing attribute provides a practical method of detection, > handling, and repair of missing text alternatives, after a conscious > decision has been made by the author to deliberately publish images > without text alternatives. It would: > > * Allow an image without alt text be honestly labeled for it is: > missing, incomplete, lacking substance. > * Affirm that the author did not (and does not intend to) provide a > text alternative. > * Provide a machine checkable mechanism to locate missing alt > text/enable tools to quickly discern where "missing" has been used. > * Afford a practical means to mitigate damages after all else has > failed, allowing for crowdsourcing or metadata repair. AT would be at > liberty apply a crowdsourced definition, to scour image metadata or or > both, since the AT knows that the author didn't apply a text > alternative, it can inform the user as to the potential deficiency in > the located text(s). > * Support ethical accountability by developing and promoting > responsible tools and by advocating an effective enabling environment. > > </draft text> > > Can anyone not live with that? Ideas for improvement? All input > appreciated. Hi Laura, I'm having a very hard time with this myself. How, fundamentally, is crowd-sourcing and mining of obtuse metadata going to accurately supply appropriate text alternatives to an image. What differentiates crowd-sourcing from OCR guessing - they are guesses that have an equal chance of being wrong as even partially right. I am not opposed to @missing (or @alt-not-asserted, or something of that nature) as this is accurate and precise - having others supply a guess as to what the image is (via crowd-sourcing, extrapolating from metadata, etc.) misses (for me) the whole point - the image was inserted by the author for a reason, and *only* the author can convey that reason. Crowd-sourcing neglects the fact that the "image" has a contextual component to it. For example, my daughter recently posted a picture of herself from Paris, where she was visiting as an exchange student. The photo is of her, with the Eiffel tower in the background - the tower being contextual to family and friends of her location, but the *point* of the photo is/was that she (my daughter, Victoria) was in Paris. In a crowd-sourcing scenario, the number of people who would recognize and know my daughter's name would be extremely small, yet the Eiffel tower is iconic - thus the crowd-sourced ALT text would be "Eiffel Tower" or "girl in front of the Eiffel Tower". Does that add any value at all? Does it perhaps instead detract from the value, or at the very least miss the entire point of the photo? In a private conversation earlier today, I was reminded of this story - one I've heard myself before: "...anthropologists showed a group of remote Papua-New Guineans 15 minutes of film shot on a busy Melbourne street -- after the film played, the anthropologists asked the sample group what images they recognized in the film, and after some hushed conversation, an answer emerged: "the chicken"; this greatly confused the anthropologists because they weren't aware that in one brief span a person is seen holding a freshly killed rooster by its bound legs; it wasn't until the anthropologists sat down with the film frame-by-frame that they caught the fleeting image of the chicken, which to the Papua-New Guineans was the only recognizable component of the 15 minute film..." As it stands now, I can support the current change proposal at: http://www.w3.org/html/wg/wiki/ChangeProposals/ImgElement20090126to Further, I can support the addition of: "A missing attribute [that]... * Allow[s] an image without alt text be honestly labeled for it is: missing, incomplete, lacking substance. * Affirm[s] that the author did not (and does not intend to) provide a text alternative. * Provide[s] a machine checkable mechanism to locate missing alt text/enable tools to quickly discern where "missing" has been used." I *Strongly Oppose* however references to crowd-sourcing, metadata mining or OCR processing, as all are guessing strategies that I find unacceptable. JF
Received on Thursday, 29 April 2010 05:16:20 UTC