W3C home > Mailing lists > Public > public-html@w3.org > July 2010

Re: Change proposals for ISSUE-31 and ISSUE-80

From: Aryeh Gregor <Simetrical+w3c@gmail.com>
Date: Thu, 22 Jul 2010 16:19:55 -0400
Message-ID: <AANLkTikSn-TxqonCf7LeP2-i1Yo7i2h5rCiTBmPjdAx5@mail.gmail.com>
To: Matt May <mattmay@adobe.com>, John Foliot <jfoliot@stanford.edu>, Steven Faulkner <faulkner.steve@gmail.com>
Cc: Jonas Sicking <jonas@sicking.cc>, Ian Hickson <ian@hixie.ch>, "public-html@w3.org" <public-html@w3.org>
On Wed, Jul 21, 2010 at 4:55 PM, Matt May <mattmay@adobe.com> wrote:
> I think it's pretty easy to be led astray using that strategy. Especially if you're picking pages at random, without knowing, for example, how frequently each page is accessed. Since, as I've said, I care about the outcome for the user, a million random pages may add up to, say, the Google homepage, in terms of how often a user accesses the alt text in those documents.

You could weight by popularity too.  If you have the right dataset
(e.g., you're Google), this should be no problem.

> Data's not the problem. All of us have data. The weakness of that approach is that we're all equally capable of getting it to say whatever we want. When you're dealing with subjective criteria, data is the wrong tool for the job.

There are limits to what data can do, but looking at the available
data is a necessary starting point.  Not all of us have data.  Most
HTMLWG members are not accessibility experts, and will not be familiar
with facts that might be obvious to accessibility experts.

On Wed, Jul 21, 2010 at 4:59 PM, John Foliot <jfoliot@stanford.edu> wrote:
> Which of course begs the question, what is your experience?

I have none.  Indeed, if I were an accessibility expert, I'd probably
know what data is out there.  But as it happens, the contents of the
HTML5 spec are not decided by accessibility experts.  I don't think
any of the chairs specializes in accessibility, although I'm sure that
at least one or two have some experience with it.  With our
decision-making structure, the experts need to present their data and
reasoning fully enough to convince non-experts.

I briefly looked over the change proposals for this issue, and I
didn't spot any actual evidence about how blind people use alt text in
any of them.  Steven Faulkner's simple presentation of how a typical
Flickr page sounds in a popular screen reader with and without
low-quality auto-generated alt text did far more to convince me than
all of the rationales presented in the various Change Proposals here.

> I don't ask to be malicious or antagonistic, but if you are like most
> sighted users, you likely don't think about alt text as part of your daily
> surfing experience, as you are seeing the image instead. Equally important
> is that it is *your* experience, which we all know can never be assumed to
> be *the* experience or even the predominant experience: it is the aggregated
> experience that we must glean facts and trends from.
> Since alt text primarily (but not exclusively) benefits the users of screen
> reading technology, it would be a logical next step to also suggest that
> users from that user-group would likely have a more realistic assessment of
> 'experience' than those who do not use AT on a daily basis. Is that a fair
> statement? If yes, and you are not yourself a member of that community, then
> your experience would likely be less informed than those who do use AT
> daily. This is not to judge your experience, but to put it in perspective.

This is why I asked for data on the subject, and phrased my post
largely as a question.

> I don't think hard, analytic data actually exists (I am not aware of any),
> any more than hard analytic data that suggests that "bad" alt text is
> "harmful" to end users exists. It's something we grok based not on empirical
> data, but rather via inference: accessibility specialists can gauge that
> overall the use of @alt has continued to improve both at the author end, and
> as a result to the end user. We make that assessment because those of us who
> focus on accessibility are exposed to that feedback on a more regular basis
> than those who do not work in the field.

A summary of feedback by screen-reader users constitutes data.  So
does a demonstration of how particular pages sound using a screen
reader.  I was just asking for something beyond appeals to authority.
Possibly I phrased myself poorly, and if so, I apologize.

On Thu, Jul 22, 2010 at 6:49 AM, Steven Faulkner
<faulkner.steve@gmail.com> wrote:
> here is a small study i did a while back (2007), which may be of interest
> Investigating the proposed alt attribute recommendations in HTML 5
> http://www.paciellogroup.com/resources/articles/altinhtml5.html

Thanks!  This is exactly the sort of thing that I was looking for.  It
makes it very clear that in at least some cases, poor-quality
auto-generated alt text is much better than no alt text.  The
filenames were numeric, so they were complete gibberish -- and of
course, this is very common for web applications, because they might
store huge numbers of files and need to use arbitrary identifiers to
avoid name conflicts.  So at a minimum, a typical web app could help
out by even providing the original name of the uploaded file as alt
text, so the screen reader doesn't have to use the gibberish name used
for storage.

On Thu, Jul 22, 2010 at 7:33 AM, Steven Faulkner
<faulkner.steve@gmail.com> wrote:
> Hi Aryeh,
> there is also some info provided in the following surveys:
> http://www.webaim.org/projects/screenreadersurvey/
> http://www.webaim.org/projects/screenreadersurvey2/
> A paper [1] from 2005 by Chieko Asakawa (a researcher and developer of
> assistive technology and accessibility testing tools at IBM research labs
> Tokyo, who also happens to be blind) provides some useful data and
> discussion about alt text usage.
> She states:
>> "If there is no alternative text for these images, blind users will lose a
>> lot of the information that is presented by using images. Alternative texts
>> are extremely important for them to obtain this information from the Web and
>> to navigate through Web pages."
> 3.1 Insertion of alternative texts
>> The use of images is rapidly increasing. Figure 1 shows the annual average
>> number of images used in one page for 11 tested sites over the past 9 years.
>> This shows that the number has increased more than four times during this
>> period. If there is no alternative text for these images, blind users will
>> lose a lot of the information that is presented by using images. Alternative
>> texts are extremely important for them to obtain this information from the
>> Web and to navigate through Web pages. All the guidelines and regulations
>> mention providing alternative texts when images are used in a page. Figure 1
>> also shows the number of missing alt texts in a page. It increased from 1997
>> to 2000. However, it decreased suddenly in 2001 from 13.2 images (38%) in
>> 2000 to 9.1 images (19%) in 2001. Investigating the data, we found that all
>> of the images on 5 tested Federal Agency sites had alt texts. US Section 508
>> became effective in June, 2001. This regulation apparently affected these
>> numbers. In 2004, it appears that private companies also started inserting
>> alt texts. The number of images without ALT text decreased to 3.1 images
>> (7%) in 2005. This was affected by one of the news sites that had only
>> inserted alt texts for 27% of the images (missing alt ratio 73%) in 2004,
>> but that site had improved to 94% (missing alt ratio 6%) in 2005. The use of
>> images has increased steadily, but the ratio for missing alt texts has been
>> decreasing quite consistently among these test sites since 2000.This
>> indicates that the Web accessibility is improving drastically.
> The paper also discusses issues arising from inappropriate text
> alternatives.
>  [1] What's the web like if you can't see it? (PDF File,
> 633kb) [http://citeseerx.ist.psu.edu/viewdoc/download?doi=]

This is also very informative.  Thanks a lot for the info -- I'll be
almost certain to refer to it in the future.
Received on Thursday, 22 July 2010 20:20:27 UTC

This archive was generated by hypermail 2.4.0 : Saturday, 9 October 2021 18:45:21 UTC