RE: Issue 556 and 669

On Mon, 17 May 2004, Gregg Vanderheiden wrote:

>
>
> Regarding
> 1. "Structures and relationships of the content can be derived
> programmatically (for example, through a markup or data model)
>
> One question I have always wondered.
>
> If structure or relationships can be determined through automated
> heuristics, would that count too?    If so, then having the headers all
> "look" like headers (and be consistent within levels) would satisfy if
> common tools or converters were available that could 'determine'  the
> headers, lists, etc by just evaluating a page and its visual formatting.
> After all, that is all that the sighted person has.

Are you supposing that the heuristics would be 100% reliable across a wide
variety of content, and amenable to implementation in server-side or
client-side software? If so, but only under these conditions, I would
agree that 1.3 is met.

The problem in practice is that styles are not completely uniform across
documents, with the result that any heuristics are prone to error, whereas
markup supplied by the author (given an appropriate authoring environment)
is highly reliable.

Suppose there were metadata supplied by the content author or a third
party that established a completely correct mapping of style properties to
element types, but the content itself was encoded presentationally. Or
suppose there were a structure tree with pointers back to the
presentational format in which the content was stored (essentially what
tagged PDF is). Both of these in my opinion would count as satisfying 1.3,
even if the structure resided on another server, whether maintained by the
author or a third party, as long as server-side or client-side software
implemented the necessary mechanism for matching the structure with the
content and constructing a presentation to satisfy the user's
requirements.

Work on RDF mechanisms in this area, notably by Lisa, should also be
mentioned here.

Received on Monday, 17 May 2004 02:07:01 UTC