Re: Good Summary Values and Layout Table Detection (Was: Issues of @summary and use of data for "decisions")

Shelley Powers writes:

> Though CSSquirrel addressed this issue humorously, several of us have
> expressed concern about decisions being made on proprietary data,

There was data shared with the list which suggested very strongly that
nearly all uses of summary are so misguided as to be useless -- so
fortunately in this case[*1] we don't need to take on trust analysis of
private data.  (Those who have seen private data are not making any
claims from it which can't also be seen in the public data.)

> This is then used as some kind of empirical proof that the @summary
> attribute should be pulled.

As somebody with no previous view on the matter, and who is open to
being persuaded one way or t'other, I initially found that to be
compelling evidence that summary is so misused that it's useless,
because nobody would ever find the few good values among the cornucopia
of irrelevance.

However another data point was added: screen readers will ignore most of
those summaries, having heuristically determined their host tables to be
layout rather than data tables.  So those errant summary values are
mostly benign, and of summaries actually read to users a large enough
proportion may be helpful to make it worth the feature's existence -- it
completely changes the balance on summary's utility.

So what it would be really useful to see now is the algorithm screen
readers use for determining whether a table is for layout or data, for a
couple of reasons:

* It would enable determining which of the summary values Philip listed
  are heard by screen reader users and which are ignored.  This lets us
  see how good the technique is at skipping the irrelevancies, and
  enables examination of just the summary values which users hear.

* If summary is part of HTML 5 then it will depend on user agents
  applying that algorithm, so it needs to be specified.  In particular,
  a new talking browser entering the market would need to implement the
  algorithm to be interopable with the current web.

Are any developers of software which already includes such an algorithm
on this list, and if so are they willing to share it?

If not then it seems we'll have to reverse engineer it to get summary in
HTML 5.  Is anybody in a position to help with that?

Cheers.

Smylers

[*1]  Your general point is of course a good one.  There may be places
in HTML 5 where justification of a decision involves non-public data, in
which case we do need to work out what to do there.  But the debate
about summary -- which seems complicated enough as it is -- can be
separated from that.

Received on Wednesday, 24 June 2009 11:16:11 UTC