RE: Add a 'Date reviewed' field to techniques and failures

From: Gregg Vanderheiden [mailto:gregg@raisingthefloor.org]
Sent: Thursday, May 19, 2016 5:35 PM

I was thinking more of Failures than techniques.

And what is a guaranteed failure today - may not be failure in the future with better AT - or may only be a failure funder specified conditions.

So the review date is just to make it easier to periodically review all failures to see if the failure needs to be modified, qualified or removed over time.

I think review dates make more sense on failures than on techniques. They wouldn't even have to be visibly included in the published document as they would be provided for internal use by the working group.

If a failure ceases to be a failure due to a technological advance, the appropriate response would be to document this clearly in a revised publication. The same holds if it's only a failure under limited circumstances.

I expect advances in artificial neural networks to start raising interesting issues for WCAG in the coming years. Machine learning algorithms can apparently recognize images much more accurately than was the case even severl years ago, especially when trained on large databases. If the user has access to a Web-based image recognition tool that automatically seeks to identify images (perhaps as a browser extension), then failing to provide a text alternative becomes only a probable failure of SC 1.1.1, not a definite failure as it now is for all practical purposes.

Of course, the image recognition algorithms could be even more effective in the hands of the author as a means of pre-populating alt attributes, and this is how I think they should be used; but the difficulties pointed out above nevertheless remain for WCAG.

The same is probably true of sound recognition algorithms, although (at least in my admittedly very limited experience) much of the sound on the Web occurs in music, podcasts, radio broadcasts and video/multimedia, not as discrete sounds that a recognizer could identify.

Recognizing structure from presentational cues is another possible application, raising questions about failures of SC 1.3.1 in the event that someone is successful in training machine learning algorithms to identify common page and document structures. (I expect that the failure cases would remain, however, since "programmatically determined" demands a deterministic rather than a probabilistic solution, so the machine learning techniques wouldn't qualify). There may be pressure to revise WCAG if assistive technologies founded on machine learning bcome available and effective in many of the common cases.


________________________________

This e-mail and any files transmitted with it may contain privileged or confidential information. It is solely for use by the individual for whom it is intended, even if addressed incorrectly. If you received this e-mail in error, please notify the sender; do not disclose, copy, distribute, or take any action in reliance on the contents of this information; and delete it from your system. Any other use of this e-mail is prohibited.


Thank you for your compliance.

________________________________