Re: operational concept for table browsing

Having reviewed the section on tables in the September draft, it appears
that most of the requirements which Al and T.V. Raman have identified are
adequately supported in HTML. Of course, it would be possible for a user
agent to employ more sophisticated techniques for identifying which header
cells correspond to each data cell, then those stated in the algorithm
which is outlined in the specification. For example, if column groups are
defined, the first heading in each group might be taken by default as a
branch in the tree structure, as though it had been labeled with the AXIS
attribute.

I would appreciate clarification of what a "row header" is, and of how
such a header is to be marked up in the HTML. Row headers are mentioned in
the algorithm for determining which headers correspond to a data cell, but
nowhere else in the discussion of tables generally can I find any further
reference to them.

Another issue which deserves to be raised at this point is the default
reading order of a table, in cases where interactive navigation is not
desired. Obviously it would be best if a speech-based user agent could
read the entire table in a comprehensible fashion, without requiring the
user to issue navigational commands to control the rendering. This need is
at the centre of T.V. Raman's requirement that readings such as
"Population growth in [cell 1] was [cell 2]". We need to ascertain to what
extent such readings can be automatically generated from the HTML source,
and whether there is a need for further semantic information to be
supplied. In the latter case, what kind of additional content would a
speech-based user agent need?

It appears to me that the relationships between data and header cells can
be adequately expressed by way of the AXES attribute, and that hierarchies
can be built by means of a default algorithm (for example, by recognising
the first column in a column group as a first-level branch in the tree
structure), together with the AXIS attribute which may, in some cases, be
supplied by the author. What additional semantic content can a table
contain which would not be captured in such a model? Perhaps Al could
clarify the types of relationships between cells which he wishes to take
into account, but which he thinks are not adequately representable in the
existing markup repertoire.

Received on Saturday, 4 October 1997 21:26:08 UTC