[Fwd: Re: Accessibility implications of Chickenfoot] (Relates to label guessing)

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

[Fwd: Re: Accessibility implications of Chickenfoot] (Relates to label guessing)

Aaron Leventhal-3


Hi all,

There are two answers to your question:

(1) We're planning to move to an MPL-like license for the entire
chickenfoot codebase.  Current target for doing that is the end of the
semester (late May).  So if you can wait until then, you can look at it
with no risk of pollution.

(2) If you don't need to look at the code, but just want to know what
the heuristics are, you can look at our papers.  Our original paper
describes the technique used by Chickenfoot to find the best matching
textbox for a set of keywords:

Automation and Customization of Rendered Web Pages, UIST 2005
http://groups.csail.mit.edu/uid/projects/chickenfoot/uist05.pdf

But you're probably more interested in an algorithm that goes the other
direction, from a textbox to label.  This technique was originally
implemented in Koala, but not described in the Koala paper.  It *is*,
however, described in our latest paper:

http://groups.csail.mit.edu/uid/projects/bookmarker/bookmarks.pdf

The relevant bits are in the "Recording the Action History" section,
pages 6-7.

Hope this helps,
Rob

Alex Faaborg wrote:

> Rob and Greg,
>
> During the CSUN accessibility conference several weeks ago a
> conversation started between some Mozilla contributers about heuristics
> for associating labels with HTML form controls.  This could improve the
> accessibility of Web browsing since it would enable screen readers to
> accurately describe form fields in the tab order, and voice recognition
> systems to identify form fields the user is referring to, even if the
> form fields were not correctly labeled by the Web page's creator.
>
> Many of the screen reader and speech recognition vendors that make
> software compatible with Firefox have been asking for this feature, so
> adding the functionality to Firefox 3 could have a major impact for our
> disabled user base.
>
> Several people who have grants from the Mozilla Foundation to improve
> the accessibility of Firefox would like to look at the heuristics used
> by Chickenfoot for associating labels with controls.  However, if they
> base the implementation on work done by your group, they could get into
> some licensing problems since Chickenfoot is currently only licensed
> under GPL, and looking at the code will contaminate them.
>
> I believe Chickenfoot attempts to locate form controls retroactively
> based on input from the user, as opposed to proactively after the page
> loads.  However, your heuristics based on position and distance may
> still be very useful.  Would it be possible to release just this subset
> of code under a MPL/GPL/LGPL tri-license? (http://www.mozilla.org/MPL/)  
> I believe most of the relevant methods can be found in Pattern.js
>
> Thanks,
> -Alex Faaborg
>
>
>
>

_______________________________________________
dev-accessibility mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-accessibility
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Re: Accessibility implications of Chickenfoot] (Relates to label guessing)

Aaron Leventhal-3
The label guessing that Chickenfoot uses doesn't
handle the table as layout case, AFAICT:

 From the paper:
"For checkboxes and radio buttons, the label
consists of the text following the element. For
textboxes and listboxes, the label is found by an
  pproach similar to Koala’s [9]: first checking
the element’s previous siblings for visible
nonempty  ext, and if that fails, then searching
the  previous siblings of the element’s ancestors
until text is found. This algorithm generally
finds a caption correctly when it appears on the
left or above the component."

That may be good enough for Firefox 3's label
guessing, however.

- Aaron
_______________________________________________
dev-accessibility mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-accessibility
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Re: Accessibility implications of Chickenfoot] (Relates to label guessing)

Jason White
On Tue, Apr 10, 2007 at 10:45:00PM -0400, Aaron Leventhal wrote:
 

>  From the paper:
> "For checkboxes and radio buttons, the label
> consists of the text following the element. For
> textboxes and listboxes, the label is found by an
>   approach similar to Koala?s [9]: first checking
> the element?s previous siblings for visible
> nonempty  ext, and if that fails, then searching
> the  previous siblings of the element?s ancestors
> until text is found. This algorithm generally
> finds a caption correctly when it appears on the
> left or above the component."
>
> That may be good enough for Firefox 3's label
> guessing, however.

I agree, except that it doesn't handle the case of a list or text field where
the label follows the input field. When we are at the first such field in a
form and the search for preceding text fails, should we reverse the direction
of the search (namely, by traversing subsequent siblings and, if necessary,
their ancestors?) Clearly this reversal would need to be in force for the
remainder of the form or option group, assuming consistency on the part of the
content author.

In general, my reaction to all of this is that it is time for Web content
authoring practices to catch up with where HTML 4.0 was a decade ago, and to
move on to embrace XForms. As for missing ALT attributes of IMG elements,
these have been mandatory in every HTML and XHTML DTD from version 4.0
onwards. I sometimes wonder whether the practice of user agents (allowed, if I
remember rightly, by successive HTML specifications) of rendering invalid
HTML, was a bad design decision, and whether instead user agents should be
required to treat invalid markup as a "fatal" error. However, I also realize
that even if HTML specifications had required user agents to reject invalid
markup, some corporate user-agent developers would almost certainly have
ignored this, proclaiming the interpretation of invalid documents as a
feature.

_______________________________________________
dev-accessibility mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-accessibility
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Re: Accessibility implications of Chickenfoot] (Relates to label guessing)

Aaron Leventhal-3
In reply to this post by Aaron Leventhal-3
Jason White wrote:
> I agree, except that it doesn't handle the case of a list or text field where
> the label follows the input field.
How common is that? Do JAWS and Window-Eyes handle
those cases really? I'd love to see an example in
the wild.

- Aaron
_______________________________________________
dev-accessibility mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-accessibility
Reply | Threaded
Open this post in threaded view
|

Re: [Fwd: Re: Accessibility implications of Chickenfoot] (Relates to label guessing)

Jason White
On Wed, Apr 11, 2007 at 05:02:27PM -0400, Aaron Leventhal wrote:
> How common is that? Do JAWS and Window-Eyes handle
> those cases really? I'd love to see an example in
> the wild.

I've encountered them but I can't remember the URI of any. I'm not a Windows
user, so I can't answer the other questions. However, if the heuristic can
handle these cases without too many inaccuracies then it should be included.
If the algorithm keeps track of where the last text label, if any, was (for
the previous field, if there is one), and the search doesn't find any text
between that and the current field, that's the condition under which I would
reverse the direction of the search.

Obviously, the concept of "between" in the above has to take into account tree
traversal.

_______________________________________________
dev-accessibility mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-accessibility