Sept. 25, 2003
professor and author Steven
Pinker reflects on his research on language, cognition and
creativity and on his career as one of MIT's most admired teachers.
Pinker was appointed Johnstone Family Professor of Psychology
at Harvard University this fall, after twenty years in MIT's
Department of Brain and Cognitive
Sciences. One of America's most famous public intellectuals,
he is the author of many essays and books, including The
Language Instinct (1994) and The Blank Slate: The Modern
Denial of Human Nature (2002).
Jay Keyser is the Peter de Florez Professor Emeritus
in Linguistics and Special Assistant to the Chancellor at MIT.
of Pinker's Farewell is now available.
PINKER began with recollections of his early career at MIT
as a postdoctoral fellow at the Center for Cognitive Science.
He recalled his involvement in the center's Lexicon Project,
which included theoretical work on how the meaning of a word
governs its usage in a sentence. The project helped him solve
a paradox in language development that had troubled him for
cases, the verb dictates the structure of a sentence. One of
the things a child has to learn is what verbs go with what sentence
structures. For example, consider the pair of sentences:
loaded hay onto the wagon.
John loaded the wagon with hay.
In the first
sentence, the verb has the direct object "hay," followed
by the prepositional object "wagon." In the second
construction, the direct object becomes the prepositional object,
and vice versa. In English, there are sixty to seventy verbs
that allow you to make this alternation, another example being:
John splashed the wall with water; John splashed water
onto the wall. With this pattern, it would behoove a child
to extract this regularity and use it to generalize to new verbs.
Children are not restricted to parroting the sentences they
hear. Generalizations are essential to language acquisition,
because language is an infinite system with no limit to the
number of thoughts you can express.
there are exceptions to the generalization of making these flips.
For instance, you could say: Amy poured water into the glass;
but saying: Amy poured the glass with water sounds incorrect,
even though its meaning is clear. How could this sound odd given
the generalization? Although children make these errors, evidence
shows that most parents do not correct them on sentence form.
How children grow up to find these sentences ungrammatical was
to this paradox came from ideas that emerged from the Lexicon
project. Two rules were identified. The first rule is concerned
with how a situation is conceptualized. For example, when you
load hay onto a wagon, the object you are affecting depends
on how you construe the situation. You can be affecting the
hay by moving it, or holistically affecting the wagon by making
it full. Language captures those alternative construes of the
same event in the sentences above.
rule maps the affected entity onto the position of direct object.
In the case of the sentence: John loaded hay onto the wagon,
"hay" is the direct object because it is primarily
these rules, you begin to see why some verbs go into the alternation,
while others do not. Languages dictate what kind of situations
can undergo a conceptual shift by dividing the world of actions
according to geometry and force.
verbs of simultaneous contact and motion (such as brush, smear,
rub, streak), as a class, all undergo the flip: Brush butter
onto the turkey, brush the turkey with butter. Verbs
of imparting force causing ballistic motion (such as splash
or splatter) also undergo the alternation.
But of the
verbs enabling gravity to cause motion (such as pour, drip,
dump, spill), none permit the alternation. This is the same
with other verbs of mediated attachment, when there is another
force between the object and the destination (for instance pin,
fasten, attach). Therefore, you can "brush the turkey with
butter," but you cannot "pour the turkey with butter."
cognitively decompose an event into the sequence of physical
events that underlie the meaning of each word. When you learn
a language, you must subdivide your verb classes very finely
according to this intuitive physics. For Pinker, this uncovered
a world of conceptual structure that could be revealed through
S. J. Keyser (left) moderating, Pinker paused to address
questions from the audience.
Given that children at some point do make these errors, how
do they get to the point of not making them?
Children make these errors sporadically; even adults make them.
The errors aren't prohibited, but just sound a little odd. If
you look at the number of such errors that children make, as
a proportion of the number of opportunities to make them, it
is very small. I think historically, a language changes when
one of these classes is stretched, so that an event is easier
to construe. When enough people hear it from one another, the
language embraces the change.
The lengths of the descriptions of the verb lists are as long
as the lists themselves. What makes you think that children
aren't just memorizing the lists?
First of all, there is a generalization tendency in language
that shows it is not about memorizing lists. As for the lengthy
descriptions, the features that language cares about tend to
be universal, and are reused over many classes.
spoke of three great visionaries who were among his colleagues
at MIT: David Marr, Marvin Minsky, and Noam Chomsky.
was a scientist and professor in the Department of Psychology
from 1977 to 1980. People were in awe of him, and some of his
proposals were looked upon as dogma. Pinker's research on vision
was influenced by Marr's work in shape recognition.
mechanism for recognizing a shape would be to have a template:
a stored representation that mimics the shape of the object.
However, a template matching mechanism will make errors if the
object is shifted, rotated, or presented in other variations.
Yet these are things a human can recognize easily.
that shapes were defined on a coordinate system centered on
the object itself. As a shape moves, the coordinate system moves
with it, and the description of the shape remains invariant.
This is in contrast to a viewer-centered reference frame, where
the object is described with respect to the viewer's coordinate
Marr's theory in a number of ways. In one experiment, people
learned shapes that were too complex to be recognized by a simple
feature. The shapes were learned in one orientation, but were
presented in another orientation in a test to see how long it
took people to name them.
was correct in that humans had a representation of shape that
was invariant across all orientations, a person should be equally
fast at recognizing objects in all orientations. Alternatively,
if humans were good at shape recognition because they memorized
shapes at various orientations they have seen, then a person
could faster identify an orientation he or she has seen before.
that Marr's prediction of orientation independence was only
partly right. Subjects took a longer time in recognizing orientations
they did not learn. However, when the shape was symmetrical,
they recognized it equally as quickly in all orientations. It
seemed that humans have an object-centered coordinate system
that could be mapped directly on an object regardless of orientation,
but only for one dimension at a time.
the shapes of a square and a diamond with 90-degree angles.
In terms of Euclidean geometry, they are identical. Nonetheless,
they have different names. This shows that the perception of
shape is not just the process of matching templates, but also
depends on the larger context that defines a coordinate system.
spoke of Marvin Minsky's influence. Minsky was a scientist,
philosopher, and seer who helped to demystify intelligence in
the 1950s. Minsky argued that people could make sense of thinking
as a kind of computation.
there was the influence of Noam Chomsky. Two of Chomsky's ideas
were especially inspiring. First was the notion that the mind
was a complex system composed of a number of faculties, including
language. The other was the idea that the organization of the
mind was innate, and that children were preprogrammed to learn
language. Pinker's work tries to flesh out what exactly a child
is born with in terms of a learning algorithm.
his work from Chomsky's in a several ways. While Chomsky's analysis
of language is more abstract, Pinker stays close to actual surface
he differs from Chomsky on the larger question of where language
fits in human nature. For Pinker, natural selection is the only
process that can result in evolution of complex innate structures.
A complete story of language asks what role it played in evolution,
and what were the selective advantages in having complex grammar.
does not see it this way. In fact, over the years he has become
rather hostile to the idea that language is a system designed
for communication. He believes that language evolved for beauty,
not for use. Chomsky's skepticism about evolution extends far
enough to say that there is nothing about language that is particularly
well adapted for communication.
theory of human nature links his theories of language and politics,
Pinker said. He believes that people have a spontaneous tendency
to cooperate and create for the sheer sake of it without regard
for reward or consequences. That is the deepest root of Chomsky's
belief system. This leads his radical politics. It also leads
to a conception of language that emphasizes creativity, but
devalues the utility of language as a system of communication.
It cannot be explained in terms of its beneficial consequences,
which is the essence of natural selection. For Pinker, Chomsky's
view is fascinating and beautiful, but very different from his
Why isn't there some middle ground on the development of language?
Perhaps language was born as a method of expressing creativity,
and by accident became important for communication, and then
was passed on by natural selection.
Well, evolution has to work like that in the sense that the
initial variation cannot be for anything. The reason why I wouldn't
put it that way is that the system for expressing thought has
enough complexity that it couldn't have arisen because of random
mutation; it is too organized. There exists a reproductive benefit
to striking bargains and exchanging know-how. It seems more
plausible that the initial function was to communicate, and
that sheer expression was a spandrel of communication, as opposed
to the other way around.
consideration is that language comes at a cost. Humans are the
only mammals that cannot breathe and drink at the same time;
we have a vocal tract that is adapted for language, and are
at risk of choking whenever food passes over our trachea with
some chance of being lodged in it.
KEYSER: So could it be possible that the first language
was sign language?
It is not a crazy idea. In fact, it is an old idea that has
been revisited by psychologist Michael Corballis in his book
From Hand to Mouth, where he argues that gesture was
an intermediate stage. This is something we cannot really know,
but there is some circumstantial evidence in the fact that chimpanzees
have hand gestures, or that sign language is so easy for children
to acquire. Much language use involves a hybrid of speech and
gesture; sometimes it is impossible not to use gesture. For
example, how do you define the word "spiral"?
You've mostly talked about the way children learn language.
Does an adult learn language in a fundamentally different way?
That's a very good question. We know that adults are not as
good at it as children. Adults are more likely to be saddled
with an accent, as well as minor syntax and inflection mistakes
that a native speaker would not make. This is probably because
of a change in the plasticity of the brain that occurs with
learning a second language do better than one would think, because
they have a first language to fall back on. Much of learning
a second language in adulthood is mentally translating from
the first language. There are two reasons why I think that.
One is that adults make errors in their second language that
would be grammatical in their first language. The second reason
is based on the study of deaf people acquiring sign language.
There was a study that looked at two groups: people deafened
in adulthood who subsequently learned sign language were compared
to people who were deaf from birth, but were kept away from
signing until adulthood. Of the two groups, the adults who learn
sign language as a second language did better.
I have a question about articulation. Most people don't have
the vocabulary to express all the continuous and analog thoughts
in their heads. How much is language binding, considering the
fact that you have to express these continuous signals in your
head with a finite basis of words in your vocabulary?
I wouldn't think of language as binding in the sense that it
prevents you from thinking in certain ways, even if you had
a smaller vocabulary. Although thought is so rich and analog,
language does multiply the value of thinking, because you can
acquire new concepts through communicating with other people.
The way language changes also suggests that it doesn't confine
us. I think that when there's something we want to express,
we change the language, rather than being unable to think it.
Is there an evolutionary explanation for why Chomsky is so romantic?
Perhaps not an evolutionary one. I think there are great themes
in our intellectual tradition that differentiate thinkers, as
well as people on the political left and right, who have different
conceptions of human nature. The conventional alignment is that
people on the right believe in a stronger conception of human
nature, while people on the left embrace the idea of a blank
slate. Chomsky was so radical because he upended that equation;
he was firmly on the left, but was an innatist.
What I gathered from your writings is that you seem to think
most of the complexities of our language are unique to humans.
Can you talk about why you think so?
Language abilities emerge spontaneously in any normal child
in a natural human community, whereas the abilities of chimps
are the results of human training pushing them to their limits.
It is not a level playing field to begin with. Added to that
is the fact that the ability of trained chimps to use sign language
is extremely rudimentary. I don't think it is meaningful to
ask whether other animals have language, because that is a semantic
question of how broadly you want to define language. A better
question would be whether other animal communication systems
work according to the same principles as human language. The
differences are rather extreme - the two main ones being our
dependence on combinations as opposed to individual signs, which
is the central importance of grammar; and our ability to refer
to things independent of our immediate emotional states.
One thing that frustrates me about cognitive science is how
great thinkers can have such radically divergent ideas. Do you
think that with the advent of technologies, there can be some
strong non-trivial predictions that can be made in the future,
that might distinguish some of these contradictory hypotheses?
Yes, definitely, I am an optimist in terms of trying to make
ideas more testable and empirically responsive. In terms of
how new techniques might further resolve this, functional neuroimaging
might be one way, but I think genomic analysis might be more
informative. There was a study published in Nature a
year ago that was relevant to the issue of whether language
was a target of natural selection. It was an analysis of a gene
that was responsible for a speech disorder, and the results
of the study support the idea of language as a target of natural
--photos by Walter Holland