Re: strawman with legend fix

I just took a close look at Doug's Data Visualization Taxonomy, and I
certainly would not want to see that level of complexity in an ARIA chart
annotation system.

It's still a very useful document, as a sampling of all the diverse types
of charts that would need to be describable with the new annotation
system.  However, I don't see a purpose in declaring separate role values
for each chart type.  Not only is that an impossible task -- there will
always be charts that don't fit the taxonomy -- but it isn't terribly
useful.  What are accessibility technologies supposed to do with that
information?  Would they have separate rules for how to present each chart
type?

Instead, I think the focus should be on describing the *data* -- the
semantic meaning of a chart -- rather than describing its presentation.
That allows data visualization designers to develop complex custom designs
and still have a way of annotating them.

For example, most data charts would contain

   - A collection of datapoints each associated with certain categories or
   certain values on a quantitative scale.

   - Categorical scales or legends would contain the set of possible values
   and an indication of whether they have a strict ordering.  From each
   category value, it would ideally be possible to access the list of
   datapoints that are labelled by this category.

   - Quantitative scales that would have
      - max value, min value, units, and possibly step values;
      - a description of the associated visual dimension
      (horizontal/vertical position, element width/height/radius, start angle,
      angle extent, color hue, color saturation, brightness, etc.) to make it
      easier to understand prose descriptions of the chart;
      - the syntax could probably re-use some attribute names from the ARIA
      range inputs.

Ideally, an accessibility technology would be able to synthesize the
information from the datapoints so that a user could start from information
about a quantitative scale and then access information like max, min,
median, etc. on that scale dimension, and jump to the associated data
points in order, similar to using a sortable table column.

I'm thinking a screen reader would say something like:

   - "Scale 'Product Mass', displayed as vertical position, ranges from min
   0kg to max 30kg.  Associated with 33 data points, max value 27kg, min value
   15kg.  Option: more statistics. Option: list datapoints from max to min.
   Option: list datapoints from min to max..."

Other charts would have contours instead of discrete datapoints (not sure
how to describe those with words!).  And then of course there would be flow
charts and graphs, where you have nodes and connectors, but both nodes and
connectors might still be associated with categories or quantitative scales.

Fred's taxonomy looks like a good starting point.  I have a couple quibbles
-- e.g., I'm not sure why timelines need an "event" element separate from
an "item", and I don't find "item" a particularly useful term -- but I
expect to be on Friday's call so I look forward to a more detailed
discussion.

Best,
Amelia Bellamy-Royds

On 17 December 2014 at 09:12, Fred Esch <fesch@us.ibm.com> wrote:
>
> If you compare the taxonomy Doug and I did, you will see Doug dives into
> specifics which is a nest of worms to work through.  I think we should
> avoid diving down and rather spend our time identifying the concepts of
> graphics which need to be relayed to a blind user.
>