Old Brains Can Learn New Language Tricks
By SANDRA BLAKESLEE -- April 20, 1999
Pity the Japanese tourist asking for directions
in New York City: "Which way to Times Square?"
The answer might be: "Turn left after the next light."
The next what? Does that mean traffic signal or the
next street turning off to the right? To the typical
native speaker of Japanese, right and light are
hopelessly confused because the English sounds "L" and
"R" are indistinguishable. But they won't be confused
for long. In a fascinating set of experiments,
researchers at the Center for the Neural Basis of
Cognition in Pittsburgh have found a way to teach
native speakers of Japanese to hear the difference
between L and R reliably after just one hour of
training.
The new findings are "extremely
compelling," said Dr. Edward Jones,
a neuroscientist at the University of
California at Davis and president of
the Cognitive Neuroscience Society,
who is familiar with the research.
They shed light on how the adult
brain changes, a phenomenon called
plasticity, and on mechanisms that
make it resistant to change.
Dr. Helen Neville, a leading expert
on brain plasticity at the University
of Oregon in Eugene, called the
experiments "cool." They show that
the adult brain is capable of
substantial change, even late in life,
she said.
The research is being conducted by
Dr. Jay McClelland and his colleagues at the Pittsburgh
center, a joint program of Carnegie Mellon University
and the University of Pittsburgh. McClelland, the
center's co-director, has long been interested in how
the brain learns and, sometimes more importantly, how
it fails to learn.
He presented his findings last week at the annual
meeting of the Cognitive Neuroscience Society, which
was held in Washington.
The brain presents a puzzle, McClelland said, in that
many kinds of learning continue or even improve
throughout adulthood, but others, like the speech
sounds of a language, appear to slow almost to a stop.
Few people can learn a second language without an
accent after the age of 10.
Scientists call this 10-year window a critical period
for acquiring the sounds of a language. But what is the
neural basis of this critical period?
Clues are found in the way brain cells connect and
influence one another. In test tube experiments,
scientists take two nerve cells that are connected by a
kind of cable called an axon. When the first cell is
induced to fire an electric pulse down its axon, the
second cell also fires. Soon, physical changes develop
in both cells so that the first one almost always makes
the second one fire. This is how information is passed
throughout the brain in a process known as Hebbian
learning.
Early on, brain cells are not well connected,
McClelland said. But when experiences from the outside
world begin to flow into the brain, cells begin to
fire, and Hebbian patterns get stamped in. As
neuroscientists are fond of saying, cells that fire
together, wire together, forming circuits.
Later, a cell may come into direct contact with cells
from another circuit, but if it is committed to what it
has learned, it will not respond even to very strong
stimulation from those other cells. In other words, it
fails to learn.
This model of Hebbian learning can be applied to the
sounds of human language, McClelland said. Newborn
babies can discriminate all the sounds of every
language in the world. It is as if there were a space
inside their brains that is a blank slate, waiting for
sounds to enter.
When the sounds of the native language come pouring in,
each sound induces some cells to wire up and become
dedicated to its peculiar frequency.
Thus, there are clumps of cells that are tuned to the
Finnish "O," the Spanish "D" or the English "Th" -- all
of which are difficult for non-native speakers to hear
or pronounce. Within a short time, the baby's ability
to distinguish all sounds fades away.
Babies in English-speaking
families have cells dedicated to
hearing both L and R, whereas
Japanese babies have only one
phonetic category for a similar
sound, McClelland said. To an
American, the single Japanese R
sound, as in the word riokan,
meaning guest room, sounds rather
like a D, he said. In any case,
as the children in both cultures
grow up,
their sound categories become more sharply defined.
The brain has a relatively small amount of neural
tissue dedicated to speech sounds, McClelland said, and
so carves up that space with strong boundaries.
Sounds are stamped in early because the baby needs them
to build the foundation for language comprehension.
Thus, Hebbian processes come in early and lock in
speech sound circuits. In other parts of the brain,
Hebbian processes continue, but those circuits can be
more flexible, he said.
An adult learns to speak a second language by making
new connections in many circuits but cannot supplant
the locked-in native sound system with the new sounds
of the second language.
Moreover, the unfamiliar sounds of a foreign language
actually reinforce the sound system of one's native
language, McClelland said. When a Japanese speaker
comes to America, every time he hears an English L or
R, his single Japanese R phoneme is activated. Instead
of becoming more flexible, his ability to hear L and R
diminishes with increasing exposure to English.
The challenge, McClelland said, is to carve out two
spaces in the Japanese speaker's brain when he has only
one space for the L and R sounds. If the critical
period is over forever, he said, this should be
impossible. But if some plasticity or malleability
remains, there should be a way to override the embedded
circuits using Hebbian learning.
Thirty-four native Japanese speakers came to the
Pittsburgh laboratory, where they were given headphones
through which they heard pairs of words -- road and
load, light and right, and so on -- under one of two
conditions.
In one condition, subjects heard regular speech. They
had to say or guess when they heard an L-word or R-word
by tapping a response into a laptop computer.
In the second condition, subjects heard the same words
exaggerated by a computer, so that each sound's
peculiar frequency or formant was accentuated. As their
ability to distinguish L and R words improved, the
words were presented in regular speech. Finally, they
heard the words in sloppy or degraded speech so that
even native speakers would have to listen hard to hear
the difference.
The subjects, all of whom had great trouble with L and
R before training, used the computer for three
20-minutes sessions, involving hundreds of word pairs,
McClelland said. No feedback was offered, to make
conditions resemble the real world.
Those who heard natural speech barely improved and some
actually got worse, McClelland said. But those who
heard the exaggerated speech with gradual training
toward more natural speech all improved greatly. After
one hour, they could clearly distinguish light and
right.
At this point, the successful subjects do not
generalize what they have learned to all L and R
sounds, he said. But the experiment is just beginning.
If they go on to train on numerous pairs of words, they
may be able to retrain their entire sound system. It
appears that they have begun to carve out new,
independent circuits for the L and R sounds.
This approach may be effective for retraining other
embedded circuits, like those that underlie racial
prejudices or stereotypes, McClelland said. For
example, some people react with fear when they see a
stranger, based simply on his dress or skin color. That
response may be stamped in. "We could think about ways
of structuring situations to present a stimulus that
would originally elicit the fear response and then
teach the brain to have a different reaction," he said.
And it can definitely be used for second language
training.
During the World War II, American soldiers fighting the
Japanese infantry adopted the password "lollapalooza,"
figuring no Japanese speaker could pronounce it. So
much for that idea.
Copyright 1999 The New York Times Company |
