 |
|
January 20, 2003 Vol.73, no. 1S
 |
| THE AWARE PAIR: EECS professors Mankoff
and Dey are trying to improve on the speed and efficiency of current
communication programs for the speech and physically disabled.Photo
by Angela Privin. |
EECS professors
design “aware chair” communication system for physically and
speech-impaired
Anind Dey is helping disabled
people communicate better by watching sitcoms. The Intel researcher
and adjunct EECS professor has worked with graduate student Jeff Heer
to transcribe many hours of television programming to better study how
people talk. In collaboration with his wife, EECS professor Jennifer
Mankoff, Dey is trying to improve on current word prediction keyboard
technology.
Some speech-impaired people communicate by typing into a machine that
generates speech. Word prediction technology attempts to speed up the
process by guessing the intended word from the letters typed. Current
technology is error-prone and cumbersome, producing conversation speeds
of less than 10 words per minute, while average conversations hover
between 100 to 150 words per minute.
The pair hopes to revolutionize word prediction by adding context such
as location, schedule, time of day, and previous speech patterns to
specify vocabulary choices. “If you go the deli and type in ‘roast
b...’ the software will predict beef as your next word instead
of more common words like ‘because’ or ‘before’,”
says Dey.
After proving itself in the lab, the software will be installed on a
mobile unit that can be attached to wheelchairs for field tests. “The
more the system is used, the more data it has on what people say in
different settings, and the better its predictions should become,”
says Mankoff.
Mankoff hopes this project will shed more light on managing ambiguity
in language and give hints on the kind of data needed to make the technology
faster and more efficient. “Right now we are just working on intuition,”
she says.
The pair gleaned the vocabulary for their program from English as a
second language Web sites.
While the work is still in its nascent stages, the couple hopes to develop
technology proficient in predicting a vocabulary set based on the task
being accomplished. “Once you are able to do those kinds of predictions,
the applications are endless,” says Anind.
|
|
