How the Brain Evolved
Language
Donald Loritz
OXFORD UNIVERSITY PRESS
HOW THE BR AIN
EVOLVED LANGUAGE
This page intentionally left blank
·
·
·
·
·
HOW THE BR AIN
EVOLVED LANGUAGE
·
·
·
·
·
Donald Loritz
1
1
Oxford New York
Athens Auckland Bangkok Bogotá Buenos Aires Cape Town
Chennai Dar es Salaam Delhi Florence Hong Kong Istanbul Karachi
Kolkata Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi
Paris São Paulo Shanghai Singapore Taipei Tokyo Toronto Warsaw
and associated companies in
Berlin Ibadan
Copyright © 1999 by Donald Loritz
Published in 1999 by Oxford University Press, Inc.
198 Madison Avenue, New York, New York 10016
First issued as an Oxford University Press paperback, 2002
Oxford is a registered trademark of Oxford University Press
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
electronic, mechanical, photocopying, recording or otherwise,
without the prior permission of Oxford University Press.
Library of Congress Cataloging-in-Publication Data
Loritz, Donald, 1947–
How the brain evolved language / Donald Loritz.
p. cm.
Includes bibliographical references and index.
ISBN 0-19-511874-X; 0-19-515124-0 (pbk.)
1. Language and languages—Origin. 2. Biolinguistics. 3. Grammar,
Comparative and general. 4. Human evolution. I. Title.
P116.L67 1999
401—dc21
98-29414
1 3 5 7 9 8 6 4 2
Printed in the United States of America
on acid-free paper
This book is dedicated to my family
and
to the memory of
Walter A. Cook, S.J.
This page intentionally left blank
ACKNOWLEDGMENTS
This book has been so long in the writing that I can hardly begin to recount
and thank the many people who have helped it on its way. Foremost among
them are certainly Paula Menyuk and Bruce Fraser, who introduced me to the
many, many subtleties of generative psycholinguistics. Then I must thank Steve
Grossberg for teaching me to avoid “homunculi” as cheap solutions to hard
psychological problems.
In the late 1970s and 1980s, there was little appreciation of neural networks.
During these years I must thank in particular Jan Robbins, Walter Cook, and
Sam Sara for supporting me on little more than simple faith. By the late 1980s,
however, neural networks had become more widely respected, and with the help
of Bernard Comrie, Charles Ferguson, Winfred Lehmann, and Betty Wallace
Robinett, and the support of my colleagues, I succeeded in securing tenure
and setting aside some time to think heretically about how language could
possibly work without a computational homunculus.
For such sense as the following pages might make, I am indebted also to
Allen Alderman, Kathy Broussard, Craig Chaudron, Dick Chen, Brad Cupp,
Lisa Harper, Beatriz Juacaba, Jee Eun Kim, Elena Koutsomitopoulou, Bernard
Kripkee, Mark Lewellen, Lise Menn, Afsar Parhizgar, Lisette Ramirez, Bill Rose,
and Elaine Shea for their insights and support. I fear I have not yet achieved
the rigorous standards my teachers have set, but the time has come to collect
my attempts at explaining language so that better minds might bring them to
fruition.
December 1998
D. L.
This page intentionally left blank
ONE
·
TWO
·
THREE
·
FOUR
·
FIVE
·
SIX
·
SEVEN
·
EIGHT
·
NINE
·
TEN
·
ELEVEN
·
TWELVE
·
CONTENTS
Lought and Thanguage 3
Jones’s Theory of Evolution 21
The Communicating Cell 36
The Society of Brain
Adaptive Resonance
Speech and Hearing
52
74
90
Speech Perception 109
One, Two, Three 123
Romiet and Juleo 133
Null Movement 143
Truth and Consequences 161
What If Language Is Learned by Brain Cells? 171
Notes 195
References 203
Index 219
This page intentionally left blank
HOW THE BRAIN
EVOLVED LANGUAGE
This page intentionally left blank
•
O
N
E
•
Lought and Thanguage
A baby wildebeest, born on the Serengeti, learns to walk and run in a matter
of minutes. The precocious wildebeest is marvelously adapted. But once a wilde
beest can outrun a lion, its adaptation is largely done. Humans must adapt,
too, but they must continue to adapt, first to mother, then to brother, then to
teacher, then to work, then to him, then to her, then to babies, and so it goes.
The world to which humans must adapt is a world of words: his word, her word,
my word, your word, word of mouth, word of honor, word of scripture, word of
law, the adaptive word.
A baby wildebeest, born on the Serengeti, learns to walk and run in a matter
of minutes, but it takes two full years before a baby human makes a word, learns
to talk. And it is another ten years before the human child learns to talk back.
Even then, it is an immature criterion that would claim the teenager has mas
tered language. Human beings’ fascination with language persists as long as
their fascination with life. How does a baby learn language? How does one talk
to a teenager—or, for that matter, to a parent? What do words mean? Or, since
I know what I mean, what do your words mean? And what do you think my words
mean? Does she think I’m a nerd? Does he think I’m a bimbo? Will my boss
think I’m disloyal if I don’t say yes? How can I adapt?
Where Is Language?
One doesn’t have to be a philosopher to ask such questions, and in fact, it may
be better if one is not. The celebrated learned men of history never concerned
themselves much about the learning of language. (After all, teaching children
language is women’s work!) Thus absolved of responsibility for explaining lan
guage in its most human terms, the wise men of history were freed to conflate
and confuse thought and language as they pleased.
3
4
•
HOW THE BRAIN EVOLVED LANGUAGE
This is not to deny that down through the ages many amusing books have
been written on the topic of language. For example, in 1962 the philosopher
J. L. Austin published a book exploring the question Is the King of France bald?
(Since there is no longer a King of France, Austin concluded the question is
neither true nor false, but simply void.) By and large however, previous phi
losophers of thought and language have simply been ignorant of one impor
tant modern fact: thought happens in the brain!
Aristotle, to pick one particularly influential example, thought that the
main function of the brain was to cool the blood.
1
In hindsight, the ancients’
ignorance of the brain and its function was quite understandable. Locked up
in its bony carapace, the brain, which resisted exposure to the warrior’s sword,
resisted as well the anatomist’s scalpel. And even when the ancients noticed
there was a brain beneath all that bone, they couldn’t see it do anything. It didn’t
beat or breathe or bend. What ancient could have imagined that the brain
created ideas with the same electrical forces as Zeus’s thunderbolts? Real knowl
edge, Aristotle thought, was lodged in the heart, and even today, when we know
something well, we say we “know it by heart.” So we can understand as well the
ancients’ belief that knowledge was mysteriously dissolved in the blood.
Finally, 2,000 years after Aristotle, Harvey showed that the heart pumped
blood through the body, circulating nourishment to the organs of the body.
Knowledge had to be somewhere else. But the microscope had not yet been
invented, and when the seventeenth-century eye looked at the brain, the first
feature it noted was that the brain, like the heart, had several connected, fluid-
filled chambers, called ventricles. (In Figure 1.1, a horizontal section of the brain
exposes the main, lateral ventricles.) To seventeenth-century philosophers, the
meaning was obvious: the brain was just another pump. Following Galen and
Harvey, Descartes thought it pumped an animating fluid (animus) through the
nerves, thereby causing muscles to move. He specifically thought that the
pineal gland at the base of the ventricles was a kind of master valve, which con
trolled hydraulic pressure in the system. To Descartes, this brain-pump was just
so much more plumbing. Hydraulically moving the muscles was important, but
it was just machinery; it could have nothing very much to do with thought. For
Descartes, thought happened somewhere else. Thought happened in the mind,
not in the brain.
But where was the mind? For Descartes, language, mind, and thought were
all essentially the same thing. Descartes would have asserted that it makes no
more sense to ask Where is the mind? than it does to ask Where is language? or
Where is algebra? Such questions, to use Austin’s term, were simply void. Lan
guage, thought, and mind were abstract sets of formal relations. They could
relate things in places to other things in other places, but they were not them
selves in some place. For Descartes, thought and language, mind and mean
ing, algebra and geometry, were all essentially the same sort of thing, which is
to say they weren’t really things at all.
In the seventeenth century this dualism of mechanics and mind, of things-
in-the-world and things-not-in-the-world, had a confirming parallel in the
Church’s natural-supernatural dualism of life and afterlife. In a sense, Descartes
LOUGHT AND THANGUAGE
•
5
Figure 1.1.
Horizontal section of the cerebrum. The (lateral) ventricles are
exposed. Occipital lobe (O), parietal lobe (P), frontal lobe (F), lateral ventricle
(L), corpus callosum (C), fissure of Rolando (R). (Kirkwood 1995, 15. Reprinted by
permission of Churchill Livingstone.)
extended the conception of the supernatural to include not only angels but
also algebra, algorithms, and language. These otherworldly entities had a truth
that, to Descartes, was obviously true, a priori, before and independent of any
empirical experience. One could only find this truth by doubting empirical,
in-the-world experience and by believing in a priori, not-in-the-world truths,
truths like the existence of God.
But how could you or I, mere mortals both, know in the mind that even
we ourselves exist, let alone so sublime a being as God? Was there a less pre
sumptuous a priori truth from which we could deduce these larger Truths?
Perhaps the most famous “rationalist” deduction of this sort was Descartes’s
proof of his own existence:
cogito ergo sum
(1.1)
think-I therefore exist-I
“I think therefore I am.”
Unfortunately, as Nietzsche later observed, Descartes’s “proof” turns out
to be uselessly circular: in Latin, the -o on cogito and the form sum itself both
indicate first-person I. Consequently, as the literal gloss of 1.1 emphasizes, the
premise I think presupposes the conclusion I am. To illustrate this point, con
6
•
HOW THE BRAIN EVOLVED LANGUAGE
sider 1.2 and 1.3 (here and elsewhere a * means “something seems wrong with
this sentence”):
*You think therefore I am.
(1.2)
or
*Thought exists therefore I am.
(1.3)
Without its presuppositions, Descartes’s proof fails utterly. In Descartes’s
defense, we should perhaps consider the context of his times. The Reforma
tion had put reason at odds with God, and Descartes had a larger agenda than
to vainly prove his own existence. But the proof is still false. Even a genius can
not deduce truth from faulty premises.
Tabula Rasa
Well before Nietzsche, many philosophers objected to Descartes’s dualistic
method. Descartes’s contemporary Francis Bacon strenuously objected to
Descartes’s introspective method. Francis Bacon (and, coincidentally, 400 years
earlier, Roger Bacon) espoused a rather distinctively English empiricism. Un
like Descartes’s dualism, this empiricism was a triadism that divided the uni
verse into Soul, Mind, and Matter. Leaving the supernatural aspects of Soul to
God, empiricism proceeded to focus on the material aspects of Matter. But
neither Bacon was a rigorous scientist by modern standards. (In what was ap
parently his first and only scientific experiment, Francis Bacon stuffed a chicken
with snow to see if the snow would inhibit decay. The only reported result was
that Bacon caught cold and died.) The relationship of Mind to Soul and Mat
ter was little advanced by their methods. It wasn’t until a hundred years after
Descartes that empiricism found a clear voice in the philosophy of John Locke.
For Locke, Mind was just a blank slate, an erased tablet of Matter, a tabula rasa.
Experience wrote upon the tablet, thus creating Mind. Of course rationalists
objected that this explained no more than cogito ergo sum. If, as the empiricists
would have it, there was such a tablet, then where was it? Where was Mind?
And if this tablet were writ upon in language, then where was language? Void
questions all! So rationalism survived until 1849, when Claude Minié invented
the conical bullet.
Before 1849, bullets were musket balls. Musket balls had a frustrating habit
of curving unpredictably in flight, so prior to 1849, opposing armies would line
themselves up, shoulder to shoulder, in order to give the opposing team a rea
sonable chance. Even then, when a musket ball did happen to score, it tended
to shatter the skull, causing massive damage to the brain beneath. Minié’s
conical bullet, on the other hand, flew true. Even better, it was frequently able
to create a surgically clean hole in the skull and a nice, focused wound (a focal
lesion) in the underlying brain tissue.
LOUGHT AND THANGUAGE
•
7
As a result of this technological advance, a young doctor in France, Pierre
Paul Broca, obtained a sizable cohort of war casualties whose brain lesions dis
turbed their ability to speak but otherwise left the patients’ minds and be
haviors remarkably intact. In 1861, Broca presented the discovery that such
aphasia occurred almost exclusively when injury was sustained to a relatively
limited area of the left half of the brain. Several years later a Viennese doctor,
Karl Wernicke, discovered that injuries to another region on the left side of
the brain caused a second kind of aphasia. Whereas “Broca’s aphasics” had
difficulty speaking but relatively little difficulty comprehending language,
“Wernicke’s aphasics” had no difficulty speaking but great difficulty compre
hending. Where is language? had seemed a void question, but suddenly—and
quite unexpectedly—it had an answer.
Where Language Is
Language was in the brain! This finding, utterly implausible to the ancients,
was supported by copious and irrefutable evidence: spoken output was gener
ated in Broca’s area, and heard input was processed in Wernicke’s area. The
scientific community instantly and earnestly undertook the study of the brain.
It was no longer the seventeenth century. Leeuwenhoek had long since
invented the microscope, and within a generation of Broca, scientists had
trained it on the brain. In 1873, Camillo Golgi discovered that chromium-
silver salts would selectively stain brain cells, thus making them clearly visible
under the microscope. Using Golgi’s staining method, Santiago Ramón y Cajal
charted the microstructure of the brain in encyclopedic detail, and by the dawn
of the twentieth century, it had become an established scientific fact that mind
was brain. And since brain was made up of white matter and gray matter, mind
was matter. Rationalism was dead.
For their discovery of the brain’s previously invisible structure, Golgi and
Ramón y Cajal were awarded the 1906 Nobel Prize.
2
Their work also engaged
them in a famous debate. Ramón y Cajal believed each cell was a separate cell,
wholly bounded by its cell membrane and unconnected to its neighbors, but
his microscopes weren’t powerful enough to prove it. On the other hand,
Galvani had long before shown that electricity made a dissected frog’s leg twitch.
It could therefore be readily inferred that there was electrical communication
among nerve cells. But how could electrical impulses be transmitted if the wires
weren’t connected? Golgi maintained that the myriad cells of the nervous sys
tem must form a continuous network.
In the early 1900s many more researchers joined in this debate. Using ever-
more-powerful microscopes, they took ever-closer looks at nerve cells. In the
end, Sherrington, Adrian, Dale, Loewi, and others proved that Ramón y Cajal
was right, earning in the process Nobel Prizes for their efforts. Neurons were
discrete cells separated by a synaptic gap. This gap was small, but it was big
enough to electrically insulate each cell from the next. So how did neurons
pass their messages across the synapse? They passed their electric messages
8
•
HOW THE BRAIN EVOLVED LANGUAGE
using chemicals, called neurotransmitters. Doubt of the world and belief in truth
were now clearly behaviors of the brain:
Thus, both doubt and belief have positive effects upon us, though very differ
ent ones. Belief does not make us act at once, but puts us into such a condition
that we shall behave in a certain way, when the occasion arises. Doubt has not
the least effect of this sort, but stimulates us to action until it is destroyed. This
reminds us of the irritation of a nerve and the reflex action produced thereby;
while for the analogue of belief, in the nervous system, we must look to what are
called nervous associations—for example, to that habit of the nerves in conse
quence of which the smell of a peach will make the mouth water. (Peirce 1877:9)
Never Mind the Mind
But Peirce was ahead of his time. Twenty years later, in America, Peirce’s “prag
matic” perspective developed into behaviorism. Behaviorism came in many
flavors, but one lineage descended from Peirce to Dewey to Thorndike to
Watson to Lashley. In the formulation of John B. Watson, behavior could be
observed and scientifically reduced to a series of stimulus-response events,
“habits of the nerves,” occurring along a chain of neurons. Mind was just an
unobservable and useless abstraction. All of creation, from the lowliest animal
to the highest form of social organization (then widely believed to be either
the assembly line or the Prussian army), could be pragmatically analyzed solely
in terms of stimulus-response chains of command. Behaviorism, in the social
form of totalitarianism, promised a well-regulated society in which every ani
mal want could be provided by eliciting strict, learned, obedient responses to
the stimuli of an all-powerful, all-loving dictator.
Predictably, this utopian vision was especially popular among the ruling
and managerial classes, who had never worked on an assembly line or directly
experienced the new, improved, conical bullet. Many, following Herbert Spen
cer (1862) and later “social Darwinists,” envisioned themselves to be “super
men,” a new species which had evolved through natural selection to a point
“beyond good and evil” (Nietzsche 1883). However, after World War II and the
likes of Hitler and Stalin, this utopian vision began to lose some of its appeal,
even among the controlling classes. In his 1948 utopian novel Walden Two, the
celebrated Harvard behaviorist B. F. Skinner attempted to dissociate behavior
ism from these infamous European practitioners. As Skinner spun the story,
everyone—more or less regardless of race, creed, color, or, for that matter,
genetics—could be educated to perfection through the application of “pro
grammed learning.” In programmed learning, students were methodically
rewarded for correct answers and punished for incorrect answers. In this way,
it was believed that good habits would be efficiently “learned” and bad habits
would be efficiently “extinguished.”
In the United States, however, there was a new class of university students:
World War II veterans whose college tuition was paid as a war benefit. These
LOUGHT AND THANGUAGE
•
9
students and vocal, war-hero labor union leaders let it be known that they did
not consider any chain of command to be utopian. Whether on the front line,
the assembly line, or the school registration line, they did not want to be pro
grammed! By the mid-1950s, opposition to Skinner had become widespread,
but it was inchoate. Behaviorism had been politically refuted by the European
experiment with totalitarianism, but Skinner’s scientific authority as a Harvard
professor was still unassailable, and there were no viable alternatives to his
psychological theories.
In 1957, amid mounting popular disdain for behaviorism, Skinner pub
lished a scholarly book, Verbal Behavior. In it, he sought to show that behavior
ism had developed far enough beyond the study of lab rat behavior to undertake
the explanation of human language. In 1959, two years after the publication
of Verbal Behavior, Noam Chomsky, a young linguist at the Massachusetts Insti
tute of Technology, published a disdainful review of it in the journal Language.
Not only did Chomsky find Skinner’s analyses of language naïve, but he found
them to be proof of the vacuity of behaviorism in general.
Skinner didn’t reply directly to Chomsky’s review, but he did write another
book, Beyond Freedom and Dignity, to which Chomsky also gave a bad review. These
reviews of Skinner and behaviorism made Chomsky an instant, popular cham
pion of freedom and dignity, opening a new chapter in the confusion of thought
and language.
Finite Mind, Infinite Language
Reaching back to rationalism for support, the thrust of Chomsky’s argument was
that language was not a “thing” like a stimulus or a response, a punishment or a
reward. Language was a unique—and uniquely human—module of mind. Thus,
twentieth-century generative grammar became grafted onto a Cartesian dualism.
The resulting generative philosophy has depended heavily on what I call the
“generative deduction,” the basic form of which may be given as follows:
(1a) The human brain is finite, but
(1b) an infinity of sentences exists,
(1c) which can be generated by rule,
proving language is infinite. Nevertheless,
(2a) normal human children acquire language quickly and effortlessly,
(2b) even though no one teaches language to young children,
(2c) and only human children so acquire language.
Therefore,
(3) language is innate. It is not so much learned as it is “acquired.”
10 •
HOW THE BRAIN EVOLVED LANGUAGE
The premises of the generative deduction have come under attack from
many quarters, but it has not yet been refuted. Consider, for example, Jacken-
doff’s 1994 witty defense of premise 1. First, Jackendoff opens the dictionary
at random and generates a large number of sentences by a simple rule:
A numeral is not a numbskull.
A numbskull is not a nun.
A nun is not a nunnery.
. . .
These are all completely absurd, but they are sentences of English nevertheless.
There will be something like 10
4
× 10
4
of them = 10
8
. Now let’s put pairs of these
sentences together with since, like this:
. . .
Since a numeral is not a numbskull, a numbskull is not a nun.
. . .
Since a numeral is not a numbskull, a numbskull is not a nunnery.
. . .
And so on it goes, giving us 10
8
× 10
8
= 10
16
absolutely ridiculous sentences. Given
that there are on the order of ten billion (10
10
) neurons in the entire human
brain, this divides out to 10
6
sentences per neuron. Thus it would be impossible
for us to store them all in our brains. (Jackendoff, 1994:21)
Although 10
16
does not quite qualify as mathematical infinity, it certainly
seems infinite for human purposes. This infinity of language was at the nub
of Chomsky’s arguments against Skinner in 1959, and premise 1 of the gen
erative deduction has stood unrefuted and irrefutable until the present day.
For the past forty years, a variety of biologists, psychologists, teachers, and
child-language researchers have contested premise 2, arguing that children are
taught language and do in fact learn in the process. But premise 1 forms the
basis for a strong logical defense of premise 2. Chomsky has introduced that
defense with a different quotation from Peirce:
You cannot seriously think that every little chicken that is hatched has to rum
mage through all possible theories until it lights upon the good idea of picking
up something and eating it. On the contrary, you think that the chicken has an
innate idea of doing this; that is to say, that it can think of this, but has no fac
ulty of thinking anything else. . . . But if you are going to think every poor chicken
endowed with an innate tendency towards a positive truth, why should you think
to man alone this gift is denied? (Peirce, quoted in Chomsky 1972, 92)
Peirce called the ability to come up with new theories abduction, a logico
cognitive process which he believed was more important than either of the
logical processes of induction or deduction. Chomsky asked essentially the same
question of children and language: one cannot seriously think every little child
that is born has to rummage through all possible grammatical theories until it
lights upon the one right way of making words into sentences. Language could
LOUGHT AND THANGUAGE
•
11
not be learned unless every child was endowed with an innate tendency toward
a correct, universal grammar.
Following Chomsky’s suggestions, researchers undertook a series of math
ematical analyses, collectively referred to as “learnability theory,” to investigate
the conditions under which language could be learnable (Gold 1965, 1967;
Hamburger and Wexler 1975; Wexler and Culicover 1980; see Pinker 1984,
1989, for approachable reviews). The gist of their argument was the following.
If you say potayto and I say potahto, how is a child to learn which one to say? This
argument becomes more convincing as one considers, not just the 5,000 or
10,000 words that a child might memorize, but also the fact that the child knows
how to transform these words à la Jackendoff into an infinite number of sen
tences (premise 1 again). Chomsky’s seminal example was the “passive trans
formation,” as of 1.4 into 1.5:
John saw her.
(1.4)
She was seen by John.
(1.5)
Instead of 1.5, why doesn’t a child ever say 1.6*,
*Saw by John was she.
(1.6)
or any of the other 118+ possible permutations of 1.5? “Because the child
never hears those other 118+ permutations,” you may say. But the child has
likely never heard the exact permutation which is 1.5, either. Nevertheless,
every child has learned to produce passive sentences like 1.5 by the age of
six or so (premise 2a).
“Well, the child doesn’t memorize rote sentences,” you reply. “He remem
bers patterns.” But exactly how does he remember patterns? No one in his right
mind sits down and teaches a child of four that “to transform an active sen
tence pattern into a passive sentence pattern, one positionally exchanges the
subject and direct object, prefaces the subject with the word by, appropriately
changes the grammatical case of the moved subject and direct object, precedes
the main verb with the tensed auxiliary of be, agreeing in number and person
with the new subject, and replaces the main verb by its past participle.”
You might instead argue that the child learns language patterns by imitat-
ing adult speech, and this was in fact the explanation proposed by behavior
ists. Unfortunately, child-language researchers quickly found that children don’t
imitate adult speech. Consider the following, oft-quoted transcript from McNeill
1966:
Child Nobody don’t like me.
Mother No, say “Nobody likes me.”
Child Nobody don’t like me.
Mother No, say “Nobody likes me.”
12 •
HOW THE BRAIN EVOLVED LANGUAGE
Child Nobody don’t like me.
Mother No, say “Nobody likes me.”
Child Nobody don’t like me.
Mother No, say “Nobody likes me.”
Child Nobody don’t like me.
Mother No, say “Nobody likes me.”
Child Nobody don’t like me.
Mother No, say “Nobody likes me.”
Child Nobody don’t like me.
Mother No, say “Nobody likes me.”
Child Nobody don’t like me.
Mother Now listen carefully. Say “Nobody likes me.”
Child Oh, nobody don’t likes me.
To maintain that language is “learned,” it appears one needs a better theory
of learning than imitation.
Although generative philosophy has demonstrated the failure of behav
iorism to most observers, it has not been without its critics. For example, the
claim that language is rule-based (premise 1c) extends back to the foundations
of modern linguistics in the eighteenth century, but for forty years, nonlinguists
have objected that language cannot be rule-governed, because semantics, the
meaning system of language, is not rule-governed. After all, what rule could
definitively tell you what I mean when I say I love you? But semantics has little
to do with the generative deduction. Chomsky has argued that “such under
standing as we have of [language] does not seem to be enhanced by invoking
thoughts that we grasp, public pronunciations or meanings, common languages
that we partially know, or a relation of reference between words and things”
(1993, 25), and as Jackendoff’s A nun is not a nunnery illustrates, sentences can
be grammatical even if they are meaningless. That is, leaving meaning aside,
how is one even to explain syntax, if not as acquired through the agency of an
innate, rule-governed system?
Recently, many cognitive psychologists have attacked premise 1c by dem
onstrating that pattern-based neural networks can exhibit linguistic behaviors
similar to that of rule-based systems (Rumelhart and McClelland 1986a). But
to date these demonstrations have been more semantic than syntactic. Also,
the fact that rulelike behavior can be elicited from an artificial neural network
does not preclude the possibility that the brain functions at some other, more
interesting level like a rule-based digital computer.
My discomfort with the generative deduction originated with premise 2a,
that children learn language “effortlessly.” To be sure, childhood in middle-
class America in the latter half of the twentieth century has been mostly child’s
play, but even privileged children display the temper tantrums of the “terrible
twos,” and these are nothing so much as results of the child’s frequently frus
trated efforts at communication. Nor do mommies and au pairs find the ter
rible twos “effortless.” Nevertheless, the claim that toddlers learn language
effortlessly seems never to have been challenged directly, and I am unaware
LOUGHT AND THANGUAGE
•
13
that generative philosophers have ever independently proposed an objective
measure of child effort. The problem, no doubt, is that effort is an intrinsically
subjective, “nonscientific” concept. Society devalues child labor because no one
pays children a salary, and no one hears children complain—no one except
mommies and au pairs, that is, but “scientific” society doesn’t pay them salaries,
either.
“Hard science” often tries to distance itself from such social issues, but when
the object of scientific inquiry is language, it is hard to maintain distance. As a
kind of compromise, sociolinguists (Ferguson and Slobin 1973) and “function
alists” (Bates and MacWhinney 1982; MacWhinney 1987a) have attacked premise
2b by redefining learning away from the narrow terms of behaviorism into more
general terms of interaction in the social environment. We learn that the sky
is blue, that birds fly, and that ice is slippery from the physical environment
without a teacher, but no one claims this knowledge is innate. Sociolinguistic
functionalism argues that we learn language from the social environment in
much the same way. But how do we learn that birds fly and ice is slippery? Gen
erative philosophers have justifiably objected that this sort of learning ( a) is
not itself well understood and so ( b) barely begins to address deeper problems
like how we understand the sentence I don’t think penguins can fly.
Finally, biologists have often attacked premise 2c, the human uniqueness
of language, citing dancing bees and signing apes as evidence of the evolution
and learning of language in other species. Nevertheless, not even the proud
est trainer invites his animals to cocktail parties. Whatever their language,
animals’ language is still a far cry from human language.
Although locally convincing, none of these attacks has proved generally
fatal to the generative deduction, much less added up to a viable alternative
theory of thought or language. Taken together, though, they indicate that
something is amiss with the generative deduction. Forty years after first postu
lating that children have an innate “language acquisition device,” generative
philosophers have as yet been unable to find its place in human biology, and
generative theory has found itself increasingly at odds with the rest of science
and society. Chomsky himself has become defensive, asserting that “no one
knows anything about the brain” (Chomsky 1988, 755), and asking,
how can a system such as human language arise in the mind/brain, or for that
matter, in the organic world, in which one seems not to find systems with any
thing like the basic properties of human language? That problem has sometimes
been posed as a crisis for the cognitive sciences. The concerns are appropriate,
but their locus is misplaced; they are a problem for biology and the brain sci
ences, which, as currently understood, do not provide any basis for what appear
to be fairly well-established conclusions about language. (Chomsky 1994, 1)
The preceding is neither a crisis for biology nor a crisis for linguistics; it is
a crisis for Science. The assertion that no one knows anything about the brain
may have been defensible in 1936, when Turing initiated “the study of cogni
tive activity from an abstract point of view, divorced in principle from both
biological and phenomenological foundations” (Pylyshyn 1979). It may also
14 •
HOW THE BRAIN EVOLVED LANGUAGE
have been defensible in the late 1950s, when the foundations of generative
philosophy were being laid. But since then, some two dozen Nobel Prizes have
been awarded for discoveries in brain science. To date, at the end of the twen
tieth century, some thirty-three Nobel Prizes have been awarded for discover
ies about the human brain and nervous system (Ramón y Cajal in 1906, Golgi
in 1906, Sherrington in 1932, Adrian in 1932, Dale in 1936, Loewi in 1936,
Erlanger in 1944, Gasser in 1944, Hess in 1949, Békésy in 1961, Hodgkin in
1963, Huxley in 1963, Eccles in 1963, Hartline in 1967, Wald in 1967, Granit
in 1967, Axelrod in 1970, von Euler in 1970, Katz in 1970, Guillemin in 1977,
Schally in 1977, Yalow in 1977, Sperry in 1981, Hubel in 1981, Wiesel in 1981,
Levi-Montalcini in 1986, Cohen in 1986, Sakmann in 1991, Neher in 1991,
Fischer in 1992, Krebs in 1992, Gilman in 1994, Rodbell in 1994). The prob
lem today is not that “no one knows anything about the brain.” The problem
is that we know so much about the brain and its abnormalities in so much detail
that it becomes difficult to step back and see how the brain might do some
thing so normal and so large as language.
The Von Neumann Limit (
ν)
The great mathematician John Von Neumann is credited with having invented
the modern serial computer’s organization into “procedural memory” (pro
gram) and “declarative memory” (data). Although Ramón y Cajal had won the
1906 Nobel Prize for showing that the brain is a massively parallel processor,
Von Neumann declared that “the nervous system [is] a computing machine in
the proper sense, and that a discussion of the brain in terms of the concepts
familiar in computing machinery is in order” (Von Neumann 1958, 75).
Von Neumann went on to claim that since serial computers could do
everything parallel computers could do and then some, they were in principle
of design superior to parallel computers. It follows logically from this premise
that the digital computer is, in principle of cognitive design, superior to the
human brain, and that the computer scientist could be in this respect supe
rior to God Almighty. Of course, most computer scientists have been too modest
to make these deductions in public, but in the privacy of classified documents,
these obvious implications sold a lot of computers to a world military/security
establishment bent on being almighty (Roszak 1986). Flush with money and
power from these contracts, Von Neumann and his followers overlooked one
small factor in their calculations, however, death.
It turns out that the fatal flaw in the generative deduction is in its least-
examined premise, premise 1a. While a serial computer might well be able to
do everything a parallel computer can do, it can’t always do those things in
the 10
9
-odd seconds of a human’s allotted lifetime. The reason every little child
doesn’t have to rummage through all of the words and sentences his little head
can hold is not so much that every little child is born with a language acquisi
tion device endowing him with an innate tendency toward universal grammar.
LOUGHT AND THANGUAGE
•
15
It is that the human mind is infinite. To see how this is so, let us recall Jackendoff’s
conclusion:
And so on it goes, giving us 10
8
× 10
8
=10
16
absolutely ridiculous sentences. Given
that there are on the order of ten billion (10
10
) neurons in the entire human
brain, this divides out to 10
6
sentences per neuron. Thus it would be impossible
for us to store them all in our brains.
Recall how Jackendoff got to these numbers. He combinatorially paired 10
4
words
into 10
8
simple sentences and then combinatorially paired those 10
8
simple sen
tences to create 10
16
compound sentences. But if words and sentences can com
bine, why can’t neurons combine? In fact, Ramón y Cajal showed that neurons do
combine: each brain cell makes synaptic connections to thousands and thousands
of other brain cells. Brain cells function in subnetwork combinations with other
brain cells. And just how many subnetwork combinations can brain cells make,
you ask? Well, it so happens there is a formula for combinations:
n!
————
(1.7)
(n – k)!k!
If we assume that each of Jackendoff’s 10
10
neurons is used in language
and that each makes some 10
3
synaptic connections, and if we further assume
that every word is represented by 10
6
connections combining in simultaneous
activation, then with n = 10
13
synapses taken in combinations of k = 10
6
, 1.7
becomes 1.8:
10
13
!
ν = ——————— ≈ 10
7,111,111
(1.8)
(10
13
– 10
6
)!10
6
!
10
There is no computer large enough to compute the value of
ν exactly, but
7,111,111
is a reasonable approximation.
3
And how big is 10
7,111,111
? Well, 10
300
is a generous estimate of the number of atomic particles in the known universe,
so not only can your brain store Jackendoff’s 10
16
+ sentences, it can also store
a name for every particle in the universe! And it will still have room for nam
ing all the particles in 10
7,110,811
more universes!! Compared to Jackendoff’s 10
16
sentences—or the brief candle of a human life—your mind’s capacity is, for
all human purposes, infinite.
10
But are we to seriously think that every little child that is born has to rum
mage through all possible 10
7,111,111
combinations until it lights upon the one
good grammatical idea for a sentence? Only if we process them serially. Be
cause letter follows letter, word follows word, and sentence follows sentence,
the wise men of history from Aristotle to Von Neumann supposed that serial
language must be the product of a serial process: because language is serial,
they supposed thought must also be serial.
4
But unlike Von Neumann’s serial
computer, it takes no more time for a parallel-processing brain to abduce
7,111,111
theories than it does to abduce 1 theory! Think about it this way. When
you go to the zoo and see a zebra, do you start going through all the names in
16 •
HOW THE BRAIN EVOLVED LANGUAGE
your head— aardvark, ant, anteater, antelope . . . —until you come to zebra? Of
course not. Zebra comes to mind as quickly when you see a zebra as aardvark
does when you see an aardvark. Neither is this trick of fast, “content-address-
able” memory unique to humans. When a lion sees an aardvark or a zebra, I’m
sure it knows immediately what’s for dinner, even though it doesn’t have a
printed menu. When a pigeon sees a hawk or a fox, it immediately decides
whether to duck or to fly. In the meantime, the little serial computer, which
had to rummage through all its possible plans of action, would have been lunch.
If we are prepared to think that every poor pigeon is innately endowed with a
fast, content-addressable, parallel-processing brain, why should we think to
Homo loquens alone this gift is denied?
Adaptive Grammar and the Plan of the Book
It is important not to confuse thought and language. Just because language
is manifestly serial, it does not necessarily follow that language must be com
puted by a serial processor. Although language is serial, thought is parallel.
Although Turing machines are serial processors, the human brain is a paral
lel processor.
While generative philosophy’s fundamental confusion of thought and lan
guage cannot be ignored, the primary objective of this book is not an attack
upon generative linguistics or artificial intelligence. Within their just premises,
generative linguistics has identified numerous previously unnoticed cogni
tive phenomena which demand explanation, and serial computers have often
proved themselves to be valuable, if not very adaptive, devices. On the other
hand, children are not Turing machines, and so long as children are needed,
they need a theory of their own. What is needed is a theory of how human
language, which functions to serve our minute-by-minute social adaptation,
arose as part of the same adaptive, evolutionary process which led to Homo
loquens. This book attempts to develop such a theory. Chapters 2–5 outline the
theory’s foundations, from elementary evolutionary and biological principles
of life through Stephen Grossberg’s adaptive resonance theory (ART). Chap
ters 6–12 apply ART to language, creating in the process that specific applica
tion of ART which, to give my critics a convenient target, I will call adaptive
grammar.
Adaptation occurs on many timescales. Over the ages, each phylum evolved
as an adaptation to a changing Earth. Over the millennia each species evolved
as an adaptation to changing habitats. These are the timescales on which ad
aptation is usually discussed. But each individual human also must adapt to
society over the course of a lifetime, and the neurons which encode our daily
thought and language must adapt even faster, on a scale of seconds. English
speakers, for example, must adapt every a or the to their listeners’ knowledge.
Thus, we say a dog until our listener knows which dog we’re talking about, and
thereafter, we say the dog. On these latter timescales, it is common usage to call
adaptation learning.
LOUGHT AND THANGUAGE
•
17
Because adaptation is such a universal property of life, adaptive grammar
is rooted in many disciplines, and this book must range from Cambrian to
Cenozoic, from molecules to minds to language, across evolutionary biology,
neurobiology, psychology, linguistics, mathematics, and computer science. Even
if I could master all of these fields and their subfields, I would need to sim
plify, and even still, every reader would find some chapter difficult and another
chapter oversimplified. For this, and for the many lacunae in my own knowl
edge, I beg the reader’s indulgence. In the end, however, a viable adaptive
grammar will be the product of many minds, and to attract a broad resonance,
I have sacrificed much detail to readability.
Chapter 2 begins by taking a long, long view of adaptation and adaptive
communication, starting with the evolution of the brain cell out of the “pri
mordial soup.” That soup has left few fossils, so this is an admittedly specula
tive view, but it is justified on three accounts. First, the story of evolution
establishes the biological principle of self-similarity, which is needed to extend
adaptive grammar beyond the simplest biological and linguistic examples.
Second, imagining how the simplest, two- and four-celled brains evolved is a
preliminary thought experiment in Grossberg’s method of minimal anatomies,
the reasonable method of explaining complex neural processes in terms of
simpler neural processes. This preliminary thought experiment shows how basic
neural mechanisms that would eventually be needed for language had already
evolved as early as 600 million years ago. Finally, the story of evolution coinci
dentally gives the uninitiated reader a narrative thread that can tie together
the numerous biological facts upon which subsequent chapters will build.
Chapter 3 studies the single-celled organisms called neurons. But whereas
chapter 2 was necessarily speculative, there is relatively little that is speculative
about chapter 3. Twentieth-century science has laid neurons out plainly before
us, photographed by electron microscopes and dissected with biochemical
scalpels. With the rapid progress of biological science, every speculation in this
domain quickly becomes a testable—and tested—hypothesis.
The neuron is ultimately a social creature, so chapter 4 brings the neuron’s
billion-year evolution to its culmination in the complex society of the human
brain. Chapter 4 first looks at the large-scale organization of the brain as it
appeared to Broca and Wernicke: hindbrain, midbrain, and forebrain, right
hemisphere and left hemisphere, front and back. These views are now fairly
familiar to the educated reader, but chapter 4 at its end takes some more un
usual, less traditional views: it also looks at the brain top to bottom, inside out,
and splayed flat.
Once upon a time, there were four myopic neuroscientists. Peering through
the microscope, they happened upon a brain cell. The first, observing a cell
body shaped like a pyramid, said, “This is a pyramidal cell.” The second, ob
serving the spines on the cell’s dendrites, said, “This is a spiny cell.” The third,
observing its spherical neurotransmitter vesicles, said, “This is a spherical cell.”
The fourth, noting the concentration of glutamate in those vesicles, said, “This
is an excitatory cell.” Of course, they all were right, but by using ever-thicker
lenses to study ever-smaller objects, our myopic neuroscientists eventually
18 •
HOW THE BRAIN EVOLVED LANGUAGE
“could no longer see the forest for the trees.” Only a handful of neural net
work researchers persisted in studying the neural forest that Ramón y Cajal
discovered, explored, and charted.
Following these researchers, chapter 5 turns off the electron microscope
and uses Grossberg’s adaptive resonance theory to describe the midscale orga
nization of the brain. Earlier theories, like behaviorism, were built on the de
tails of how individual neurons behaved (as described in chapter 3), while later
theories, like generative philosophy, were built on the organization of gross
modules of the brain like Broca’s area and Wernicke’s area (described in chap
ter 4). But these theories could not integrate their levels of description to ex
plain how an entire brain full of neurons is capable of writing War and Peace—
or, for that matter, guiding me to my office every morning—even though no
individual neuron supervises the process. ART, by contrast, describes such a
brain in terms of minimal anatomies that are organized on a scale of two to a
few thousand neurons. Only when we understand how and why these minimal
anatomies work do we begin to understand how and why larger systems like
thought and language must work as they do.
Adaptive resonance theory has been constructed principally by the analy
sis of mathematical models. Insofar as adaptive grammar is more concerned
with its linguistic validity than its mathematical validity, chapter 5 will only touch
upon the mathematical foundations of ART. To help the general reader visu
alize many of the most important features of ART—contrast enhancement,
noise suppression, resonance, self-similarity, and neural rebounds—without
close mathematical study, the main points of chapter 5 are presented by means
of a graphical computer simulation. This will be a relief to many readers, but
it would be wrong to suppress the mathematics entirely. Unlike first-year
calculus’s thin gruel of missile trajectories, the mathematics of ART is humane
and exciting, so I have tried to keep just enough of it to encourage the intrepid
reader to venture into ART’s primary literature.
Linguistics often views language as atomic sounds ( phones) built up into
the progressively larger structures of phonemes, morphemes, words, phrases,
sentences, and discourse. It happens that this approach also suits the method
of minimal anatomies, so chapters 6–12 follow the same general plan, apply
ing ART to these successive levels of language, in the process constructing that
corpus of explanations I call adaptive grammar.
Chapter 6 is transitional between neuroanatomy and linguistics. It describes
how the mouth produces phones and how they are received by the ear and
passed along auditory nerve pathways to the brain.
After the physical, physiological, and phonetic description of speech and
speech sounds in chapter 6, the problem of phonemes is addressed in chapter 7.
This is the basic problem of how you can say potayto and I can say potahto, yet
we can both still mean the same tuber. Similar questions have been investigated
by ART, but these have mostly been about vision—for example, how we can
stably identify grass as “green” even though daylight itself changes in color from
dawn to noon to dusk. At this phonemic level, where neurons in the minimal
anatomies between the ear and Wernicke’s area process the sound spectrum
LOUGHT AND THANGUAGE
•
19
of speech, previous ART analyses map quite directly onto issues of linguistic
perception.
But speech spectra change more quickly than Apollo drives his team across
the sky: language happens quickly in time, and serial order becomes a funda
mental problem. For example, when children first learn to count, they tend to
count “one, two, three, eight, nine, ten.” What happens to the middle of this
series? Behaviorist accounts in terms of stimulus-response chains (one stimu
lates two stimulates three, etc.) obviously had some missing links, but the issue
is also critical for parallel models of cognition: how can a parallel brain encode
serial order? Among major “connectionist” theories, only ART offers a suffi
ciently detailed analysis of neural architecture to solve this fundamental lin
guistic problem. Chapter 8 outlines the general solution.
Intending to honor the queen, Rev. William Archibald Spooner, Fellow and
Warden of New College, Oxford, is claimed to have offered a toast to “our queer,
old dean.” In his memory, such quaintly twisted phrases now bear the slightly
derisive eponym of spoonerisms. Psychologists, however, are not at all derisive.
In 1951, it was the lowly spoonerism that spelled the beginning of the end for
behaviorism. In that year, Karl Lashley (a student of J. B. Watson, the “Father
of Behaviorism” himself) first noted the impossible problem spoonerisms posed
for behaviorism: in spoonerisms, stimulus-response chains not only lose some
links but they must also split apart and rejoin in totally nonhabitual recombi
nations. Worse, Lashley noted that these recombinative reversals occurred fre
quently and ubiquitously not only in disorders like dyslexia but also in normal
speech and common behaviors like dance and tpying [sic]. Chapter 9 explains
the spoonerism as a natural interaction of ART rebounds (chapter 5) and ART
serial-learning anatomies (chapter 8). Rather unexpectedly, this leads to a
deeply rhythmic analysis of word structure (morphology), one uncannily remi
niscent of recent linguistic theories of “metrical phonology.”
Lashley died in 1958, so it was left to Chomsky to administer the coup de
grace to behaviorism. Chomsky saw that spoonerisms were not only isolated
error phenomena but actually instances of a more general linguistic process
called metathesis. In particular, Chomsky saw that sentences like 1.9 and 1.10
could also be related by metathesis:
Spoonerisms slew behaviorism.
(1.9)
Behaviorism was slain by spoonerisms.
(1.10)
Chomsky’s theory of linguistic metathesis was built on the mathematical
principles of Alonzo Church’s lambda calculus (1941). The lambda calculus is a
recursive grammar of algebra and the design specification for a kind of com
puter called a pushdown-store automaton. Pushdown-store automata are especially
suited to recursive operations upon data that are structured in binary trees, so
Chomsky explained that sentence 1.10 was derived from 1.9 by a “passive trans
formation” on syntactic trees, which moved behaviorism to the tree position of
spoonerism and vice versa. Chomsky’s theories generated widespread and well
20 •
HOW THE BRAIN EVOLVED LANGUAGE
funded enthusiasm: if recursion and pushdown-store automata could explain
language, then certainly they could explain human intelligence in general!
Although Chomsky cautiously distanced himself from such glib enthusiasm,
his work became very much a cornerstone of a generation’s research in “artifi
cial intelligence.”
But if language is produced by brain cells, what could it mean to “move
behaviorism to the tree position of spoonerism”? Could it mean that in transforming
1.9 into 1.10, some behaviorism-cell and some spoonerism-cell actually exchange
places in the brain? Of course not. So chapter 10, “Null Movement,” states the
obvious and rejects Chomsky’s basic explanation of metathesis: nothing moves.
It then extends the analyses of chapters 7–9 to develop an alternative explana
tion of how words can “move” if neurons can’t. In place of movement, chapter
10 borrows the serial organization principles of chapter 8 and organizes syntax
around the neural representation of topicality.
Chapter 11 is a kind of cadenza. As we progress from phoneme to mor
pheme to phrase to sentence, we require larger and larger minimal anatomies
to describe phenomena. When we finally reach the stage of social discourse
and meaning, our anatomies are no longer minimal, and as we approach the
Von Neumann limit, they take on the behavior of free will. Just as Einstein
rejected Heisenberg’s uncertainty principle,
5
many “hard” scientists today re
ject the notion of free will, but when, as sometimes happens, the activity of
1 subnetwork in 10
7,111,111
causes an entire brain to change its “mind-set,” sci
entific prediction and explanation can no longer be 100% accurate. In chap
ter 11 the multiple themes of the earlier chapters become intertwined in ways
that science is reluctant to entertain. Truth and meaning are explored as logi
cal and social constructs, respectively, and self-similarity is reinvoked with the
epigram “A human being is a neuron’s way of making another neuron.”
Chapter 12, “What If Language Is Learned by Brain Cells?” is like ontog
eny recapitulating phylogeny. The themes of the preceding chapters are reca
pitulated, but this time from the perspective of the individual language learner.
At the outset, the unborn fetus is not only affected by its genetic inheritance
but also exposed to a raft of environmental hazards, so phenomena of disor
dered language learning are treated first. When I began this book after many
years away from the language disorders literature, I confess to having been
skeptical of the many and varied complaints of “learning disability.” But as it
evolved, adaptive grammar itself began to convince me not only that learning
disabilities were real but also that nearly everyone is learning disabled. But just
as the chapter appears ready to end with a prescription of learning pills for
everyone, it returns to a reconsideration of learning—normal language learn
ing, nature’s way of enabling a less-than-perfect assemblage of neurons in the
human brain to adapt and survive. Our computers are wonderful and our
medicines are wonderful, but it is most wonderful of all that nearly every
human child, despite genetic defects and a hostile environment, learns a
human language and survives.
EVOLUTION
•
21
•
T
W
O
•
Jones’s Theory of Evolution
The Sanskrit language, whatever be its antiquity, is of a
wonderful structure; more perfect than the Greek, more
copious than the Latin, and more exquisitely refined than
either, yet bearing to both of them a strong affinity, both in the
roots of verbs and in the forms of grammar, than could possibly
have been produced by accident; so strong, indeed, that no
philologer could examine them all three, without believing
them to have sprung from some common source, which,
perhaps, no longer exists.
Sir William Jones (1786)
We identify transformational grammar with Chomskyan grammar, but ideas
about language “transformations” were in the air well before Chomsky’s name
was attached to them. We call the heliocentric solar system the Copernican
system, but Aristarchus first proposed it nearly 2,000 years before Copernicus.
We call the theory of evolution “Darwin’s theory of evolution,” but the central
notion of evolution was proposed nearly 100 years before Darwin by the emi
nent English philologist Sir William Jones. Following Jones, European philolo
gists embarked upon the reconstruction of what they called the “Aryan language,”
the ancestor of all modern Indo-European languages. (Because Hitler mis
appropriated the term “Aryan,” this ancestral language is now called “Proto-
Indo-European.”) Within a generation, it had been conclusively proved that
languages as diverse as English and Sanskrit had, in fact, descended from this
now-extinct language. Since these philological reconstructions did not extend
back more than a few thousand years, they did not challenge the biblical
account of creation, and Jones’s theory of evolution quickly became established
scientific fact. Within a generation, the philologists’ ideas and methods were
adopted by Lamarck, Chambers, Wells, and many other predecessors of Darwin
who proposed the evolution of nonhuman species within biblical time. So
Darwin did not really so much invent the theory of evolution as apply it cor
rectly to biology.
But even after the philologists explained the descent of modern languages
and Darwin explained the descent of man, there remained something about
21
22 •
HOW THE BRAIN EVOLVED LANGUAGE
humankind that evolutionary theories seemed still unable to explain. That
something was descent of language—not just English or French or Chinese,
not even Proto-Indo-European, but language itself. This great web of mean
ing that we humans call language—can it not be called our soul, or at least our
mind? And however might evolution account for that? For centuries, fabulous
fakirs and sober scientists had attempted to produce horses that count and dogs
that talk. In no case, however, did any succeed in producing or finding animal
language that seemed anything like human language. Absent Darwinian evi
dence that language has evolved gradually, generative philosophy set forth the
bold assertion that language is a specific and uniquely human development,
undescended from Darwinian nature. Generative philosophy allowed this de
velopment to be physiologically associated with brain regions like Broca’s area
and Wernicke’s area, but it implied that, in Chomsky’s words, any attempt to
study language as having evolved from general animal intelligence was “adapta
tionist hogwash” (Chomsky 1988). This further implied not only that behav
iorism had been defeated by language but that its essential procedure, the
extrapolation of research findings about animal behavior to human behavior,
was sterile.
Beginning in the 1960s, generative psycholinguists therefore began to
explore language anew, as an autonomous module of mind. But by the mid
1970s, there were some disturbing reports from these frontiers. Researchers
into sign languages of the congenitally deaf (especially ASL, American Sign
Language; Klima and Bellugi 1979) had established that these visual languages
were capable of expressing human thought as completely as spoken language.
We will revisit this topic in chapter 9, but since sign language does not involve
primary speech cortex or hearing cortex, language began to look like an ad
aptation of a more general intelligence.
1
Then Philip Lieberman showed that apes and other quadrupeds had been
incapable of speech, not because of some profound cognitive deficit, but sim
ply because of the shape of their vocal tracts. As we shall see in chapter 6, speech
sounds depend upon the shape of the vocal tract. When the human species
assumed an erect posture, the vocal tract bent 90°, its acoustic properties
changed dramatically, and speech became possible. But apes’ vocal tracts re
mained unbent and physically incapable of producing most speech sounds—
despite the best efforts of fabulous fakirs and sober scientists.
Armed with these two pieces of evidence, ethologists stopped trying to
teach apes to speak and started trying to teach them ASL. A series of chimps
and gorillas (Gardner and Gardner 1969; Patterson 1978; see Premack 1985
for a summary) rapidly learned vocabularies of hundreds of signs, and what
is more, they learned to combine them into new signs and sentences. For
example, upon seeing a swan, a chimp called Washoe combined the signs for
water and bird.
Thus, an evolutionary link was established between animal language and
human language—at least in the minds of many biologists and ethologists. But
generative philosophers remained unimpressed. To them, Washoe had signed
nothing more remarkable than “there’s some water; there’s a bird.” Apes still
JONES
’
S THEORY OF EVOLUTION
•
23
had not demonstrated the ability to generate novel and infinite sentences.
Syntax remained a summit which only humans had scaled.
Nevertheless, by the turn of the twenty-first century, even some of his MIT
colleagues began to abandon Chomsky’s hard line (Pinker and Bloom 1990,
Pinker 1994). It became acceptable, even in some linguistic circles, to say that
language had evolved. The naked ape stood upright, bending his vocal tract.
This gave him a significantly enriched inventory of “calls,” and the rest is his
tory. Unfortunately, this account still does not explain some of the more re
markable aspects of language, let alone of mind. The generative objection is
that this account does not explain why, for example, we can say sentences 2.1
and 2.2 but not 2.3:
The man who is
1
dancing is
2
singing a song.
(2.1)
Is
2
the man who is
1
dancing singing a song?
(2.2)
*Is
1
the man who dancing is
2
singing a song?
(2.3)
For a deeper explanation of language, we must look deeper into evolution.
We must go back in time, long before the hominids, long before the prehomi
nids. In order to understand how human language and communication make
survival possible, we must understand how intercellular communication among
the first one-celled life forms made multicelled life forms possible. And to
understand this, we must go back long before even the dinosaurs, back to when
the only living things were rocks.
Consider that a rock lives and dies. Take a common crystal of baking soda
(sodium bicarbonate, NaHCO
3
), and drop it into a supersaturated solution of
baking soda. Behold, the crystal grows. And while it is true that there is more
to life than growth, note how this growth is like life. The growth of a crystal is
self-similar: a small crystal grows into a large crystal, and just as a small boy grows
into a big boy, its essential structure remains unchanged.
Moreover, crystals reproduce. If we split our crystal in two, each half grows
into a crystal which is also self-similar to its parent. And while it is true that
there is more to life than growth and reproduction, consider also that our poor
crystal can die.
Baking soda is hardly a diamond among crystals, but it is spectacular in its
death. Complete this elementary-school thought experiment by dropping a
teaspoon of vinegar (flavored acetic acid) on our crystal. Immediately, the crys
tal undergoes a fiery death, leaving behind only a pile of soda ash, a puddle of
water, and a cloud of carbon dioxide. Thus, even the lowliest rocks have the
essentials for a tale of life, death, and, sometimes, transfiguration. What is
wanting is a little more personality, a salt of the earth with a little more spice.
In 1951, looking for that spice, Stanley Miller concocted a primordial soup
of water, methane, ammonia, and hydrogen. In his laboratory, he subjected
this mixture to artificial lightning, seeking to simulate early conditions on the
planet Earth. After a week, Miller began to find amino acids in his soup. At the
24 •
HOW THE BRAIN EVOLVED LANGUAGE
time, only a handful of scientists appreciated what Miller had found, but in
1953, Watson and Crick suggested “a structure for the salt of deoxyribonucleic
acid” (DNA). This salt of DNA organized the assembly of amino acids into
proteins, the stuff of life.
2
One molecule of baking soda is just like any other molecule of baking soda.
What gives proteins and DNA personality is their complexity. DNA chains four
nucleic acid bases—guanine (G), cytosine (C), adenine (A), and thymine (T)—
along a backbone of phosphate and deoxyribose-sugar molecules. Each sub-
chain of three bases defines a codon, which in turn defines one of twenty amino
acids. Chains of codons form genes, and chains of amino acids form proteins.
DNA is also sexy. One chain of nucleotides does not make a complete DNA
molecule; it must be embraced by a second chain, its complementary image.
When DNA reproduces, the two chains unwrap, and each builds a new mirror-
image partner (figure 2.1).
It is still a long way from an amino acid to a sexy, self-replicating strand of
DNA, and although Miller’s experiment has been superseded by much more
sophisticated studies (e.g., Orgel 1979; Cairns-Smith 1985; see Dawkins 1986
and Dennett 1995 for highly readable introductions to evolutionary biology),
his experiment is still superb allegory, for Stanley Miller’s test tube did not only
bring forth amino acids; it also brought forth that old villain, acetic acid. Life,
death, and simple salts. The life of the first nucleic acids cannot have been much
less treacherous than the life of a baking-soda crystal in a third-grade classroom.
Even the earliest molecules of life had to be constantly on the lookout for killer
chemicals.
Despairing of the odds of bringing forth life in a hostile primordial soup
(even given two billion years), I did not know how to continue this chapter.
Then I gazed out my window at the rain, and I saw a test tube in every drop
and a laboratory in every puddle, and when I multiplied billions of drops
by billions of puddles by billions of years, the emergence of complex self-
replicating molecules no longer seemed so miraculous. Like a Polynesian ex
plorer happening upon an island without snakes and without disease, all that
life needed was one puddle of Eden, protected against predators by an imper
meable barrier of rock.
Nearly impermeable, that is, because at some point, on some day, some one
self-replicating, complex molecule had to bravely go where no such molecule
had gone before, and it had to survive. To survive outside paradise, it had to
clothe itself; it had to bring a barrier with it. In billions of puddles over billions
of years, probably billions of colonies of replicating molecules tried on many
different clothes, but in a watery world, it was the cell that survived.
The Cell
In the cell, a bilayer of phospholipid hydrocarbon molecules created a puddle
within the primordial puddle. A phospholipid hydrocarbon is a fatty molecule,
one end of which is electrochemically repelled by water while the other is at
JONES
’
S THEORY OF EVOLUTION
•
25
Dostları ilə paylaş: |