Hi Bruce,
The ability to disambiguate - and correctly pronounce - homographs is
really a function of the Text To Speech (TTS) engine rather than the
screen reader (which just feeds text to the TTS engine).
Taking a look at the page you referenced:
VoiceOver, tested with Vicki, Agnes and Fred TTS engines shows about
a 66% success rate in terms of pronouncing homographs correctly. This
puts the Mac OS TTS engines right about in the middle in terms of
their ability to distinguish between homographs. I've seen success
rates as low as 54% and as high as 84%.
A closer look at the page you linked provides some clues as to where
we can expect the TTS engines to fail to disambiguate homographs.
What the TTS engine does well:
1. Distinguish between parts of speech.
2. Distinguish between nouns and Proper Nouns
What the TTS engine does *not* do well:
Distinguishing between semantic differences like: lead, bass and row.
Notice the TTS engines get numbers 5, 8, 11, 12, 14 and 15 wrong.
This is pretty much what you can expect with modern TTS engines.
There is plenty of research happening in this area. I don't have any
of commercial TTS engines available here for testing. It might be
worth trying some of the for-purchase TTS engines and see what kind
of results you get. Commercial TTS engines are available from from
AT&T, Cepstral, etc.
Hope the information helps!
Joe
On Feb 27, 2006, at 9:30 PM, Bruce Bailey wrote:
I remember trying an experiment some time ago with VoiceOver and being
disappointed. Was I careless in my testing? Or has VoiceOver gotten
better with words that can be pronounced different ways depending on
context? Here is one page to test with:
http://www.ahajokes.com/ethnic07.html
