Re: Investigating the proposed alt attribute recommendations in HTML 5

On Sep 11, 2007, at 13:58, Steve Faulkner wrote:

>> That behavior is that alt='' suppresses
>> the image altogether but the omission of the attribute causes a
>> bearable placeholder to be presented so that the user knows that
>> there's an image.
> The problem with this is that there is no way for the software to  
> know whether the images presence needs to be announced to the user.

Hence, the need for the announcement to be bearable.

> for example in the cases of decorative images or layout images  
> without an alt attribute how are these to be filtered out from  
> images that contain "critical content" without an alt attribute.  
> What is clear (to me) is that using the omission of the alt  
> attribute is not and will not be a sufficent flag to assitve tech.

The problem with any "reliable "flag is that it can be auto-generated  
just to pass validation.

> Currently something like 30% (not sure of the exact stats) of  
> images on the web do not have alt attributes, of these how is the  
> assistive tech to discern which are worthy of announcing to a user  
> and which are to be safely ignored?

With existing content, explicit annotation won't be available  
regardless of what this WG does. I suggest considering various  
characteristics of image files (dimensions, format, spikiness of  
color distribution histogram, etc.) that can be extracted in  
software, taking a large sample of Web images, asking humans to  
classify the images into announcable and non-announcable and feeding  
the data into a Bayesian classifier. I'm pretty sure a probabilistic  
classifier could tell which images are non-decorative photos, which  
are thumbnails and which images have the traits of standard-size ad.  
Telling apart UI icons and small decoration might be harder.

> One suggestion has been to provide a noalt attribute I am as yet  
> unsure of the merits of this proposal. Another suggestion would be  
> to require the alt unless the image has meta data explicitly  
> associated with it in some other way. I also asked the question on  
> the list ( 
> 0150.html) whether alt=" " (quote space quote) could be used to  
> flag images that should have an alt text but none is supplied.

Considering JAWS, alt=' ' degrading into alt='' may be a better  
backwards compatibility story than noalt degrading into no alt. Is  
there any single ASCII punctuation character that could be used to  
denote announcable but altless image and that would have a reasonable  
degradation story in current screen readers? (To make the current  
screen readers say something and to have something that is visible to  
sighted authors looking at the source.)
>> it is clear that the current state of JAWS is so bad
> I don't know what you would expect a screen reader to do?

Examples in

Henri Sivonen

Received on Tuesday, 11 September 2007 13:34:28 UTC