At 15:25 21-02-2001, Ben Haskell wrote:
>Boumas don't exist because it is easier for the brain to recognize 26
>glyphs than to memorize thousands of boumas, and multiply this
>"thousands" by dozens of fonts. A bouma recognition mechanism implies
>HUGE "look-up tables" stored in the brain.
This isn't quite accurate. The human brain doesn't process things by
specific look-ups, but by more generic "shapes". I did some
computer-vision recognition work in college, and it turns out that the
Gaussian "shock" function of an outline[*] tends to uniquely identify an
object regardless of light conditions and orientation. This is why you can
recognize your friend's face even if he's turned at a slightly different
angle to you than he was when you first met him.
>Bouma's work was in the 1950s, as I recall from Hrant's previous posts.
>This was in the era of primitive serial computers and "look up table"
>algorithims. If Bouma's work was conducted today, would he not be
>influenced instead by massively parallel computers and "neural network
>object recognition" algorithims?
The word-shape hypothesis holds up quite well in a massively parallel
[*] Roughly, this is the skeleton that you get if you take an outline and
successively make an internal outline a fixed distance from the outline
towards the middle.
"Years of learning the hard way have convinced us it is better for the
staff to amuse the octopus than to allow the octopus to amuse itself." -
Helen Tozer, California Academy of Sciences
Freelance XML Consultant - <URL:http://crism.maden.org/>