A computer program that learns to decode sounds from
different languages in the same way that a baby does helps to shed new light on
how people learn to talk, researchers said on Tuesday.
They said the finding casts doubt on theories that babies
are born knowing all the possible sounds in all of the world's languages.
"The debate in language acquisition is around the
question of how much specific information about language is hard-wired into the
brain of the infant and how much of the knowledge that infants acquire about
language is something that can be explained by relatively general purpose
learning systems," said James McClelland, a psychology professor at
McClelland says his computer program supports the theory
that babies systematically sort through sounds until they understand the
structure of a language.
"The problem the child confronts is how many categories
are there and how should I think about it. We're trying to propose a method
that solves that problem," said McClelland, whose work appears in the
Proceedings of the National Academy of Sciences.
Expanding on some existing ideas, he and a team of
international researchers developed a computer model that resembles the brain
processes a baby uses when learning about speech.
He and colleagues tested their model by exposing it to
"training sessions" that consisted of analyzing recorded speech in
both English and Japanese between mothers and babies in a lab.
What they found is the computer was able to learn basic
vowel sounds right along with baby.
"It learns how many sounds there are. It figures that
out," he said in a telephone interview.
And if the computer can do it, he said, a baby can, too.
"In the past, people have tried to argue it wasn't possible for any machine to learn these things, and so it had to be hard-wired (in humans)," he said. "Those arguments, in my view, were not particularly well grounded."