description only read when element gains focus; has no effect on content of virtual buffer (both jaws and nvda)
button text in virtual buffer changed to reflect only description text; initial button content missing (jaws only)
has effect on input, link, and button (jaws and nvda)
has no effect for paragraphs and headings - generally no effect on non-intrinsically focusable items which are made focusable via tabindex (jaws and nvda)
nvda responds to aria-describedby on heading tags (nvda only)
nvda responds to it on list elements, although rather inconsistently (nvda only)
I suggest the following behavior for aria-describedby:
Only read text of element pointed to by aria-describedby when element gains focus.
Arrowing to the element should not cause it to be read unless arrowing also causes it to gain focus.
Only speak message about using alt+jawsKey+r to read associated text once, and only when element gains focus.
Consider only speaking it when verbosity set to beginner!
For form elements, read all explicit and implicit labels, followed by the text contained within the element, followed by any descriptions.
Consider adding a delay between the last generated speech and the start of the description.
In reference to the last point above, if the user is already familiar with the element (say a text input field to which a description is attached), then she can simply begin filling it out without a pause, thus causing Jaws to silence the description. If she were to wait to specified time interval, say 7 seconds, Jaws would read the attached description.
This is description text assigned via aria-describedby.