E-mail List Archives

Re: Guessing ALT text for product images that are generated automatically


From: John Foliot
Date: Feb 28, 2011 10:33PM

Jared Smith wrote:
> On Mon, Feb 28, 2011 at 6:32 PM, Kerryn Sues wrote:
> > The closest we can get is to *guess* the content of the image by the
> > product title (e.g., ALT="Bali Holiday - 5 nights" or ALT="Bali") --
> > we can't get as specific as ALT="Couple walking along a beach".
> If this product title is already present within the page (maybe as an
> image caption), then it probably not make sense to also add this as
> the image alt attribute value as it would then be redundant and read
> twice by screen readers.


> It sounds like "closest we can get" is merely a reflection of
> limitations of the current system. Certainly there must be a mechanism
> for somebody to provide alternative text for such an image somehow,
> no?

This topic has surfaced before in other fora, and specifically the HTML5
Working Group lists. It has been suggested that "heuristic analysis" could
be used to attempt to repair or remediate images with no alt text. While
there was certainly fans of this idea (referring to existing tools such as
Google Goggles) after a fair round of discussion most accessibility
specialists concurred that any "guess" could impart as much confusion and
"harm" as any potential help, primarily due to the fact that appropriate
alt text must be taken in context, and is based upon author intent (in
other words, the author has chosen the image for a reason, and that reason
will influence the value of the alt text).

This is also consistent with W3C's Authoring Tool Accessibility Guidelines
(ATAG) Guideline 3.4 "Do not automatically generate equivalent
alternatives. Do not reuse previously authored alternatives without author
confirmation, except when the function is known with certainty." [Priority
1] (http://www.w3.org/TR/WAI-AUTOOLS/#gl-prewritten-descs)

At this time, "heuristic analysis" has been ruled out as a viable means of
determining @alt values in HTML5.

However, if your system allows for human author intervention, then by all
means a 'crowd-sourcing' like mechanism could possibly work.

> > So is it preferable to make a best guess at the content of the image,
> > or to leave it blank with ALT="" or ALT=" " (seeing as other text
> > content within the page describes the image).
> Best guess doesn't sound very good. If the content of the image is
> conveyed in content, then give it alt="" (assuming it is not the only
> thing inside a link, in which case it *must* be given alternative
> text, even if it is redundant). Barring that, my first suggestion
> would be to fix the system to allow appropriate, descriptive
> alternative text.

Based on what we know from the posting, I think alt="" would be the most
appropriate, with the addition of aria-role="presentation"
(http://www.w3.org/TR/wai-aria/roles#presentation). While the image might
help convey a certain mood or feeling for sighted users, the actual image
(if we are to understand correctly) is being randomly generated, and as
such is as much toss-away visual fluff as useful content.

> Alt=" " (quote, space, quote) is never appropriate.