See also: IRC log
<fesch> https://www.w3.org/wiki/SVG_Accessibility/Chart_Taxonomy#Known_Bug_in_Navigation
trackbot, start telcon
<trackbot> Date: 17 July 2015
<fesch> https://www.w3.org/wiki/SVG_Accessibility/Chart_Taxonomy#Known_Bug_in_Navigation
Fred: I put together a sample chart taxonomy, extending the basic graphics-document and graphics-object for the ARIA roles for graphics.
Doug: Is this just based on your taxonomy from previously, or did you try to integrate mine?
Fred: No, I haven't integrated
that yet.
... In addition to roles, I'm using the aria-type attribute to
clarify between different features when there are multiple
objects of the same role, such as x or y axes.
AmeliaBR: For axes, we could use the specific ARIA role for defining horizontal vs vertical direction.
Fred: Maybe, but it isn't just
for axes. E.g., on the box and whisker plot, there are multiple
types of data for each object: median, quartiles, ranges,
etc.
... I have a definition there. Basically, it depends on the
role which type of information is used in aria-type.
Doug: Is there any precedent in ARIA for an open-ended string to be used to distinguish between different objects? Basically, you have no restrictions on the values for aria-type. Are there any other ARIA attributes that take string instead of enumeration?
Fred: Well, aria-label, I guess.
Doug: I understand why you did it this way, but you've now got an attribute that is supposed to be context-dependent and flexible but also machine readable for distinguishing functionality.
Fred: Well, I have labels as way, but I expect the browser would also have a way to localize the role and type and add that to the name.
Doug: That's not what I'm getting at.
AmeliaBR: I agree that it is a problem to use one attribute for many things. We want to keep things simple, but we would need to enumerate all the possible values at some point.
Fred: I'm not sure that's possible.
AmeliaBR: If you want flexible distinctions, you could use aria-roledescription, which accepts freeform strings. But don't expect AT to do anything with that except repeat it to the user.
Doug: This is similar to my point. aria-type could mean different things in different contexts, which make it different to implement. It varies depending on the role. So aria-type only has meaning when it is paired with something else.
Fred: Paired with the role. So maybe call it aria-roletype.
Doug: It introduces a little ambiguity with regards to what values it can take. I understand you want to be flexible for any data facets, many different axes or legends for different types of charts.
Fred: And you can have multiple chart panels that are themselves nested in other axes, e.g. a SPLOM a giant matrix of charts.
Doug: There's any number of type
of axes you can have, so I'm sensitive to that. But there are
two things to consider: the conventions around what type of
axes are used for a given chart and what they are called, but
also the machine has to distinguish between them.
... Honestly, I was always uncomfortable with this part. There
need to be defined semantics, but it also needs to be
extensible.
Fred: Where there are conventions, I have tried to give things distinct roles, such as graphics-charttitle.
Doug: I'm mostly talking about
aria-type. I'm uncomfortable with it being able to be anything.
That creates incredibly loose semantics.
... It's not machine-readable.
Fred: If you expect the machine
to do anything other than announce it.
... but it is really just for announcing and navigation. You
can't make the machine do everything, you can't capture all the
possibilities.
AmeliaBR: We need to distinguish which parts of information need to be machine-readable and which need to be open-ended.
<fesch> https://www.w3.org/wiki/SVG_Accessibility/Chart_Taxonomy#Navigation_Between_Features
Doug: I'm looking at this as someone who has implemented a screen reader, albeit a very simple one, for charts. Axis type means something to my screen reader as a grouping mechanism. I also called the axes x and y, but you could also say, e.g., dependent and independent.
Fred: But the screen reader shouldn't have to figure that out. The aria-type just get appended on the end.
Doug: Would this be translated?
Fred: No, the role axis would be,
but the type would not. It would be given in the language of
the document.
... The author should know what's meaningful in that domain,
more than an AT developer.
... It guides the chart navigation to split the feature when
the type is different. If elements have the same parent and
same role & type, they are treated the same. If different,
they are treated differently.
Doug: For one thing, I'm not
comfortable with a generic name like aria-type, which could
conflict with other taxonomies, and that it can be used for so
many different things. It is being used for data-related things
but also for other things like annotations.
... I would like to separate out the data facets (axes,
legends, etc.) from other things in the chart. I don't what to
use the same property for everything.
... My screenreader would deal with those completely
differently. For data facets, it doesn't just announce them
when you get to the axis, but it uses them as a set of values
as the basis for summary statistics for the whole chart,
regardless of where in the graphic they are displayed.
Fred: I think the problem there would be if you've got mutliple-panel charts, where axes can be different for each nested chart.
Doug: I kind of treat that as a
different feature of the chart from axes.
... I would call each one a separate chart. But I understand
why you're not, since they may share common features.
<jasonjgw>
<jasonjgw>
Fred: For panels, they share the same data set, they are just different columns of the graphic. But in other examples, you can have paired charts which are separate data sets but share axes, e.g., the box and whisker charts on the wiki page.
<fesch> =__AUtBR_65T5wBRYE2Fp_hJmGbDUU%3D
<fesch> http://www.masswerk.at/minard/
Doug: I think it is usually easier to just create two separate charts, and just not show the axis on one of the charts than to try to explain to the machine that the axis should be used from a separate chart.
Amelia: That is asking the author to insert extra markup that they wouldn't otherwise do. And yes, if you're using software it wouldn't necessarily be difficult, but it is still asking authors to do something extra, keep two things coordinated, just to make things easier for a computer.
Fred: I've pasted the link to the Minard chart of Napoleon's retreat. This is one of the most famous data visualizations. The axes are re-used and you can't divide it into distinct sub-charts.
Doug: This is a particularly
complex example. How you recreate it, and how you make all the
data facets accessible, is going to be much more difficult than
simply defining extra axes and then making them
invisible.
... I'm not sure why you're bringing it up.
Fred: Simply because I think you're trying to over simplify data visualization, and say that everyone has to make charts a certain way.
Doug: That's basically what we're trying to do in this working group. Create simple rules for how people can make simple charts.
Fred: I don't think you can ask people to put extra elements in the chart that they don't want to draw.
Doug: But we're asking them to add in values, which might not ever be printed on the chart.
Amelia: Doug, why don't you try to come up with some alternative proposals for things you have a concern with? And I'll do the same.
trackbot, end telcon
This is scribe.perl Revision: 1.140 of Date: 2014-11-06 18:16:30 Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: RRSAgent_Text_Format (score 1.00) Succeeded: s/I do not see a conference matching that name scheduled within the next hour, trackbot/Scribenick: AmeliaBR/ Found ScribeNick: AmeliaBR Inferring Scribes: AmeliaBR Present: AmeliaBR shepazu fesch jasonjgw WARNING: No meeting chair found! You should specify the meeting chair like this: <dbooth> Chair: dbooth Found Date: 17 Jul 2015 Guessing minutes URL: http://www.w3.org/2015/07/17-svg-a11y-minutes.html People with action items:[End of scribe.perl diagnostic output]