Re: Alt text and descriptions

to follow up on what Daniel Dardailler said:
> 
> ALT="" is really telling something about the image: that it is
> pure decoration, but that the user-agent is still free to try to guess
> what it is (using some HTTP metadata, the filename, etc).
> 

Yes, the user agent has latitude.  I hope it also has latitude to
elide all evidence of the image (in the initial page
presentation) when this is the ALT, so long as there is a
follow-up method available (just two examples: in Netscape, the
Document Information page; in Lynx the IMAGE_LINKS mode) that
would list all used and suppressed references.

One of the points that may bear repeating is the idea that text-to-speech
is a narrowband medium and hence the speech-using visitor may have
a bias toward an executive summary of the story on the first go, as
opposed to the quantity of information that can be packed on one
GUI Web Page.  So I am hoping that we can couch the message to browser
developers in the following way:

	- REQUIREMENT: (on information accessibility):
	applies_to_information:  LongDesc[Accept=text/*](image_object). 
	methods_required: GET

The point about the "information definition" is that the browser
must support some method for satisfying a query in which the user
only has the image [or its name or handle] and wants to get [any,
the designated] textual description of that image.  Text here
refers to the Internet Media primary type text/* so as to include
HTML, which is the preferred medium for LongDesc.

The method can be multi-step and interactive.  It is not required to
be an atomic action.

Example, suggested, or recommended methods may be defined by the
guidelines but in this case I think less is more and moral suasion
should be focused on the more functional requirement.

This leaves up to browser experimentation guided by user feedback the
layering of whether certain classes of information (discriminated by
ALT="" in this case) are represented on the first pass at what level
of detail.
	
My current thinking revolves around the idea that this is a special
case of an About method, and that the Web should be driving toward
providing a very short list of pervasive methods.  About is on my
current list of candidate pervasive methods.  Do, About, ...(I
am prepared to hide Help under About but I might lose that one...).

The point is that the About method would be implemented in different
ways in different implementation contexts but that it would be
enough of a workalike at all levels so that at least what shows
through at the UI level could be consistent.

	An example low-level implementation is to 
	introduce an IREF attribute
	of HTML tags, a variant on HREF.  
	HREF is used on Do and IREF is used on About.
	
	Example of a high-level implementation is that this About
	method could be implemented in HTTP via a more vigorously	
	populated HEAD response or new META method.

--
Al Gilman

Received on Friday, 11 July 1997 10:28:04 UTC