W3C home > Mailing lists > Public > w3c-wai-gl@w3.org > October to December 2000

Re: General Exception for Essential Purpose

From: Al Gilman <asgilman@iamdigex.net>
Date: Sun, 29 Oct 2000 10:12:19 -0500
Message-Id: <200010291445.JAA18533@smtp2.mail.iamworld.net>
To: <w3c-wai-gl@w3.org>
At 07:53 PM 2000-10-28 -0700, Kynn Bartlett wrote:
>In general, servers should never be sending raw data unless they
>believe that the user agent _will_ be able to process the information.


Sharing the data model means you are not just sending 'raw' data.  You are
sending documented data.  And it doesn't matter whether the data so
document is
sent raw or formatted.  The key is in sharing the key.

>If they believe this to be the case, then yes, it's a good idea to
>send XML + Schema, but absent that information, it's irresponsible
>to consider a structured data model as "accessible" in any sense
>of the word.
>I hope I understand what you're saying here -- I admit that your
>pseudo-mathematics have a tendency to lose me.  I feel as if we
>are not using the clearest language possible in this discussion.


Let's take this down to specific examples, then.

In one of the Bristol presentations, 

The Web Production Cycle

.. there is a good example to illustrate the difference.

In the presentation, Miles Sabin develops some multi-targeted content and then
chases through it to correct some data errors.

The way he develops the multi-targeted content is with a very shallow model,
however.  The only model used is the list of final form flavors.  It is all
done with smil:switch clone structures selecting final form fragments each of
which is identified by target medium, not any deeper model of what it is good
for, like precision or conciseness.  This is like using style names for CLASS
tokens.  The indirection is there, but it has been wasted, not used to gain
beneficial abstraction.

Miles put two errors in the content, to demonstrate that in the above approach
the errors are not localized, but still spread around through the source even
in the multi-targeted source.  The two errors were in were the date of an
upcoming event where Judy Brewer and Bill Clinton were both to speak, and the
acronym for the organization of which Judy is an officer.  A point that Miles
failed to make or exploit is that both of these errors could have been
automatically corrected by synchronizing the press release with what would be
in Judy's Palm device already.  What is needed is the two model-based

the _date_ of an _event_ where [the WAI officer] _Judy Brewer_ and [the U.S.
President] _Bill Clinton_ will both speak.

the _standard acronym_ for _an organization_ where _Judy Brewer_ is an

This is another case of "If 'WAI' is the answer, what is the question?"  

Had these questions (role descriptions, pattern assertions) been associated
with the relevant data appearing in the various targeted forms of the
hypertext, then correcting the distributed places where the same information
was referenced would have been an easy matter.

This kind of model knowledge would be widely understood.  It takes care
consolidating the dictionary, schema, or ontology of supertypes which are both
effective in supporting adaptive transformations and widely understood.  But
these common patterns are there, in the genres.

This is not a black art.  There are commercial tools that have been available
to work this way for a long time.  Compare Phil Darringer's presentation (if
you can read it):

  Adapting Content for Wireless



>Kynn Bartlett  <kynn@idyllmtn.com>                   
>Director of Accessibility, Edapta              
>Chief Technologist, Idyll Mountain Internet  
>AWARE Center Director                     
>What's on my bookshelf?                        
Received on Sunday, 29 October 2000 09:45:08 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:47:08 GMT