ARIA roles for graphics

When reviewing the accessibility API mappings, I was once again struck by
how useless the repeated "group" role is.  I would really like to push
forward with a basic set of graphics roles.

Yes, our long-term goal is to come up with a nuanced taxonomy for
describing the semantic meaning of charts, maps, and so on.  But in the
short term, can we decide on some roles for simple graphics, that we can
present to the ARIA group for inclusion in ARIA 1.1?  And equally
importantly, that we can use as the default ARIA roles in the SVG API
mappings document.

Now, I know that adding new ARIA roles does not necessarily mean new API
mappings -- but it gives a nudge in the right direction to the people
behind the APIs.  And, since the complete role string is available to
accessibility tools even when it isn't mapped, it does make a little extra
information available if anyone wants to use it.

To recap the roles currently available for graphics, and their limitations:

   - img
   - Should be used only when this element and all child content forms a
      single image.
      - Currently used for everything from simple icons to photographs to
      complex charts and maps (although hopefully not SVG charts and maps).
      - Child content (if not used for the name/description) is assumed
      presentational and ignored.
      - Therefore, img is not useful for structured graphics with
      meaningful parts.

      - complementary, region, article
      - Possible choices for labelling a figure (as a whole) if you want
      the component parts to still be accessible.
      - Do not convey the idea that the content is graphical.

      - text (new in ARIA 1.1)
   - Complex content that should be treated as plain text with a value
      equal to the accessible name.
      - Good if the graphic directly represents text, or can be replaced
      with a few words of text without loss of meaning, but not useful
for other
      cases.

      - group
      - Conveys a structure, without specifying what it is or giving it any
      great importance.
      - Doesn't have any semantic meaning beyond that.
      - Is nonetheless what we've been using as the default mapping because
      it is the only role that doesn't convey *incorrect* semantics.

The following roles I think would cover most basic graphics.  In
combination with proper use of other ARIA attributes (such as labelled by
and flow to) they could greatly enhance the ability to describe
infographics and labelled diagrams.

   - graphic
   - Complex visual content with a meaningful structure to it.  The
      component parts may be interactive.
      - A large graphic can contain component graphics so long as they
      still meet the criteria.
      - This would be the default role for <svg> and <g> elements that meet
      the criteria for inclusion in the accessibility tree.
      - The fallback role would either be group or region depending on how
      significant the graphic is.

      - figure
      - A graphic that is also a complementary section of the document, and
      should be included in the table of contents.
      - The fallback role would be complementary.

      - icon
      - A graphical element which conveys a simple concept or category
      using a symbolic image.
      - Differs from an image in that a short name is all that is expected;
      a detailed description of the visual representation is not required to
      convey the meaning of the icon.
      - Children are presentational -- an icon is an atomic element.  It
      should never have component parts with interactivity of their own
      descriptions -- use "graphic" instead.
      - The fallback role would be img.
      - This could be the default role for <use>, so that authors would
      have to explicitly over-ride the role if they wanted the browser
to include
      the cloned content in the accessibility tree.

      - shape
      - A basic geometric shape.
      - The main purpose of this role, as distinct from img or icon, is so
      that accessibility tools can communicate what type of shape it is.  A new
      ARIA property would allow the author to convey the shape type separately
      from the label and description.  For example, in a flow chart,
the shape of
      nodes often conveys meaning, but you don't want to repeat that in the
      label, which should focus on the substantive information.
      - Children are presentational.
      - The fallback role would be img.
      - This would be the default for all the basic SVG shapes if they met
      the criteria for inclusion in the accessibility tree.
      - All the standard SVG shapes except path would have default
      shape-type descriptions that the browser should localize by default.
      - You could also use the shape role on a group if that group
      represented a single graphical element (e.g., if it contained a
basic shape
      plus its visible label, or a shape that has been duplicated and
layered for
      graphical effect).

If we can agree on a simple schema such as this, hopefully we can get it
into ARIA 1.1 -- and into implementations -- while we are still working on
the more complex charts taxonomy.

Best,
Amelia BR