tests using aria-describedby on an image

Skip to end of metadata
Go to start of metadata

From: Geoff Freed [mailto:geoff_freed@wgbh.org]
Sent: Wednesday, June 19, 2013 5:23 AM
To: Gerardo capiel; George Kerscher; Markus Gylling; Daniel Weck; Marisa DeMeglio; Romain Deltour; Betsy Beaumon; Matt Garrish; Anh Bui
Subject: Re: Standards call on June 20, inviting my DAISY team about linking

Hi, everyone:

As Gerardo requested, below are the results of some tests I did to see how screen readers and browsers reacted to the use of aria-describedby to point to long image descriptions presented within a hidden iframe.  As you'll see, consistency is lacking in how the screen readers announce the presence of the descriptions (no surprise) but in the end the descriptions can, in fact, always be located by the AT.  

Here are a couple of points to ponder:

1.  Since we're hiding the image description in an iframe that is positioned off-screen or behind an image, is the iframe itself actually necessary?  Couldn't authors just stick the description in a <div> and hide that instead?  That would eliminate the need to store the description(s) in a separate file(s).  This might simplify the delivery process for some authors since they won't need to maintain separate files for each description or groups of descriptions.  Instead, everything would be held in one source file. 

2.  If EPUB decides to recommend the use of aria-describedby on hidden iframes or divs, then it might be necessary to file bugs with the screen-reader and browser vendors (and also vendors of EPUB reading systems) that don't make clear associations between aria-describedby and the @id.  As you'll see below, when reading down a page there is no describedby association read aloud-- users only get that when browsing from one image to the next.  Maybe that's another bug to file so users could hear the association when reading the page via either method.

George, you can forward this to the standards working group if you think it's useful.

Thanks, and let me know if there are questions.

Geoff/NCAM

=========

INTRODUCTION
Below are the results of tests using aria-describedby on an image, and an associated id on a hidden iframe, like this:

<img src="./data.png" alt="I am alt for the first image" aria-describedby="long1">
<iframe title="I am the title for the first iframe" src="./desc1.htm" id="long1">
</iframe>

The test file is available at http://ncamftp.wgbh.org/diagram/gf_iframe_test/iframe2.html .  Thanks to James Craig for providing the original examples from which this file was derived.

NOTE: "browsing by images" means pressing a shortcut key to jump from one image to the next.  "Reading down the page" refers to using the arrow keys or other shortcuts to read everything on the page in a linear order.  When reading down the test page, for example, users will always hear the image description contained within the iframe after hearing the image alt, even if the screen reader does not make the association provided by aria-describedby.

The following screen readers were used in the tests:
-- JAWS 14.0.3005 (Windows)
-- NVDA 2013.1 (Windows)
-- VoiceOver 10.8.4 (Mac)
-- ChromeVox 27 (Mac and Windows)

And the the following browsers:
-- Firefox 21 (Windows)
-- IE 10 (Windows)
-- Safari 6.0.5 (Safari)
-- Chrome 27 (Mac and Windows)

RESULTS
1. NVDA+Firefox (Windows)
Browsing by images:
-- reads alt, then reads the associated describedby (by reading the title of the iframe)
-- user can navigate forward; NVDA will read the contents of the iframe

Reading down the page:
-- reads alt; does not read associated describedby
-- user can navigate forward; NVDA will read the contents of the iframe

2.  NVDA+IE (Windows)
Browsing by images:
-- reads alt, then reads the associated describedby (by reading the title of the iframe), then immediately reads iframe contents

Reading down the page:
-- reads alt; does not read associated describedby
-- user can navigate forward; NVDA will read the contents of the iframe

3. JAWS+Firefox (Windows)
Browsing by images:
-- reads alt, then says "use jaws key + alt+r to read descriptive text"
-- user can navigate forward; JAWS will read the contents of the iframe

Reading down the page:
-- reads alt, then says "use jaws key + alt+r to read descriptive text"
-- user can navigate forward; JAWS will read the contents of the iframe

4.  JAWS+IE (Windows
Browsing by images:
-- reads alt but does not read associated describedby
-- user can navigate forward; JAWS will read the contents of the iframe

Reading down the page:
-- reads alt but does not read associated describedby
-- user can navigate forward; JAWS will read the contents of the iframe

5. CHROMEVOX+CHROME (Mac and Windows)
Browsing by images:
-- reads alt; does not read associated describedby
-- user can navigate forward; ChromeVox will read the contents of the iframe

Reading down the page:
-- reads alt; will read the associated describedby if user sets navigation to the *object* level and also uses the detailed-browsing-level commands (right/left arrow keys, as opposed to up/down arrow keys)
-- user can navigate forward; ChromeVox will read the contents of the iframe

6. VoiceOver+SAFARI (Mac)
Browsing by images:
-- reads alt but does not read associated describedby
-- user can navigate forward and enter the iframe; VoiceOver will read the contents of the iframe

Reading down the page:
-- reads alt but does not read associated describedby
-- user can navigate forward and enter the iframe; VoiceOver will read the contents of the iframe


On Jun 17, 2013, at 11:52 AM, Gerardo Capiel <gerardoc@benetech.org> wrote:

Geoff,

In preparation for this call, I think it would be helpful to get an analysis of what browser / screenreader combinations support the hidden iframe in describedBy perhaps using James Craig's example.

Though a reading system may not be fronted by a traditional screenreaders, we need to make it easy for content producers by not requiring them to have to remember to implement both longdesc and describedBy on top off potentially alt for images.

Gerardo

Gerardo Capiel
VP of Engineering
Benetech

On Jun 17, 2013, at 8:30 AM, "George Kerscher" <kerscher@montana.com> wrote:

Dear DIAGRAM Standards folks Plus Markus, Daniel, Marisa, and Romain,

Dial in and times below.

First, from my perspective  a lot of the accessible image work will involve
annotations, which will happen between  IDPF and the W3C Open Annotation
work. This will happen post EPUB 3.0.1, because how one can link into an
EPUB 3 will be defined.

Second, and here is why I have asked Markus, Marisa, Daniel, Matt, and
Romain to join us if they can:

We are down to the wire about linking  in a EPUB 3 document. The proposals
involve longdesc, describedBy and describedAt.

We have describedBy in ARIA right now and is available on most elements. It
is also supported in some of the browsers and AT.

DescribedBy can point to the ID of a hidden iFrame, an aside element, or
into another hidden element, such as a p or a div.

If it were to point out to an external item from the current page, this
would involve a level of indirection, i.e. it would first need to point to
the id of an element and then on to the URI of the external item.

describedAt would need to be in a draft recommendation in ARIA to be
considered for EPUB 3.0.1, which has some significant barriers.

Longdesc is in an extension specification and is only on img, which is a
problem.

Gulp, I am proposing that we go with describedBy.

If we get consensus on this, we would need to produce samples fairly quickly
and look at the behaviors as well.

I suggest we attack this first and then cover the other items in the
original agenda below. We have an IDPF call following this one, so we cannot
run over.

Date: Thursday June 20, 2013
Time: 1pm Pacific (4pm Eastern, 3pm Central, 2pm Mountain, 8pm in Finland)
Number to call: 1-866-210-1669 (International: 1-404-835-7069)
Participant code: 8319194#
Host code: 28262051#

Tentative agenda (subject to amendment & additions - based on last meeting's
notes here):

a.    IDPF discussion (George)
b.    Mapping from APIP to Content Model. (Madeleine)
c.    Epub 3.01 (George)

Best
George

Enter labels to add to this page:
Please wait 
Looking for a label? Just start typing.