A Computer That's Patient, Listens Well
Perched at a computer in his classroom at Fort Pitt Elementary School, 7-year-old Leon is getting a private reading lesson.
The screen displays the sentence "Jim got a spot on his bib." Leon, reading aloud into a microphone headset, substitutes the word "sport" for "spot."
"Hmm," the computer says and waits.
Leon corrects himself. "Jim got a spot on his bib," he reads.
"Dynamite," the computer answers.
Did that computer actually hear Leon's error?
In a sense, yes, according to Jack Mostow, a scientist at Carnegie Mellon University here. With a $3.65 million grant from the Interagency Education Research Initiative, Mostow is leading a project to test and perfect a computer tutor that deciphers spoken language.
Ten years in development, Mostow's project was once underwritten solely by the National Science Foundation. But adding the U.S. Department of Education and the National Institute of Child Health and Human Development to the funding mix has strengthened his project, he says.
"This is just the opposite of a turf battle," he says. "These guys really talk to one another."
While many computers "talk," understanding speech is trickier because accents and dialects differ, Mostow says. His software program relies on a speech-recognition system called Sphinx II that was also developed at Carnegie Mellon.
The tutor intervenes when the reader makes mistakes, clicks for help, or comes across a word that is likely to be difficult. For example, if Leon had not caught his error, the tutor might offer as a clue a rhyming word with similar spelling, such as "not," or another word starting with the same letters. The assistance is based on studies of what expert teachers do to guide beginning readers.
"Teachers don't have very much time to spend listening to children read," Mostow says. "This kind of infinitely patient individual assistance is something that's natural for computers to do."
Hardly perfect, the tutor misses about half the mistakes that readers make.
Even so, the technology has met with some success in pilot tests. Here at Fort Pitt Elementary, for example, some struggling 3rd graders who used the device with supervision from an aide made two years' reading progress in eight months on a school reading measure. More tests are under way this year.
But to Tish Rygalski, a 2nd grade teacher here who uses the tutor, the result is already obvious. "Any child who reads more becomes better at reading," she says. "Of course, they're learning more."
Vol. 20, Issue 6, Page 36