Oxford university press how the br ain evolved language

Figure 12.3. 
The Grossberg model of babbling. Babbling creates “self-generated 
auditory feedback,” a Piagetian “circular reaction” which is mapped from sensory to 
motor cortex at F 
2
 and learned at F 
3
 and above (Grossberg 1986. Reprinted by 
permission of Academic Press.) 

184  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
Children walk before they talk 
As we noted in chapter 9, Jakobson (1968) suggested that something like 
mamamamama is universally children’s first word.
13
 The next universal of lan­
guage learning is that children walk before they talk. Human children finally 
begin to walk at about one year of age. Until that time, there are a few “words” 
like  mamamamama and dadadadada. But just after children start to walk, 
mamamamama becomes Mama, and dadadadada becomes Dada. Finally, at or 
just before age two, there is an exponential explosion of language. What trig­
gers this sudden burst of language, if not walking? 
The rhythm of walking entails a rhythmic dipole, which entrains and seg­
ments perseverative babblings into metrical feet: mamamamamamama becomes 
Mama, in the manner suggested by figure 9.2. Babbling gets rhythm and be­
comes speech. Of course, even paraplegic children can talk: phylogenetically 
the bilateral dipole extends back to the fishes and the bilateral brain. Even in 
the crib, infants exhibit vestigial tendencies to bilaterally organized motion, 
and crawling is also a bilateral movement.
14
 But walking involves a massive di­
pole. When we walk, the left foot swings forward, right arm swings back; right 
foot swings forward, left arm swings back. A large bulk of brain becomes en­
trained in the rhythm of moving our bulk, even when we are babies. The neu­
rons controlling speech articulators are not islands. They are swept up in this 
global undulation. And as every parent knows, when children start to walk, they 
start to plan. Walking rapidly becomes purposive. Toddlers start to plan how 
to get in trouble. 
Just as the tip-of-the-tongue (TOT) phenomenon (chapter 9) revealed 
rhythmicity to lie near the center of word representations, so we should ex­
pect vocabulary growth to develop with walking. The child’s words come slowly 
at first in sucking-rhythm syllables gathered into perseverative breath-group 
babbles like mamamamamamama. Then another rhythm is added, and the 
syllables organize themselves in paired feet like Mama and bye-bye. 
Now the child is introduced to literature in the form of rhythmic nursery 
rhymes, Kinder- und Hausmäerchen. Parents begin to teach children to say not 
just bye-bye but also bow-wow and moo-cow, and soon the child’s syllables are no 
longer just being reduplicated; they are being combined. As Branigan 1979 
showed, two word utterances like Car go are now being uttered in the same 
rhythmic time frame as the child’s previous one-word utterances. As the child 
begins to run and her speech becomes fluent, there is an explosive growth in 
vocabulary. By the third year, the child may be learning new words at the rate 
of ten per day (Miller and Gildea 1987). This marks the beginning of morphol­
ogy and syntax as on the offbeats the child begins to balance nouns and verbs 
with the -ings and -es’s of fluent, grammatical speech.
15 
Imitation 
Forty years ago, it was easy for Skinner to explain language learning. The child, 
who simply “imitated” adults, was positively reinforced for a good imitation and 

WHAT  IF  LANGUAGE  IS  LEARNED  BY  BRAIN  CELLS
?  •  185 
negatively reinforced for a poor imitation. In this manner, good language habits 
were established, and bad language habits were extinguished. As the genera­
tive network recruited linguists to the attack on behaviorism, it was widely re­
ported that children do not imitate adults’ speech. One famous dialogue was 
transcribed by Braine (1971) and has been widely quoted ever since as evidence 
against imitation (e.g., Fromkin and Rodman 1974; Pinker 1989, 1994). 
Child  Want other one spoon, Daddy.
Father  You mean, you want “the other spoon.”
Child  Yes, I want other one spoon, please, Daddy.
Father  Can you say “the other spoon”?
Child  Other . . . one . . . spoon.
Father  Say . . . “other.”
Child  Other.
Father  Spoon.
Child  Spoon.
Father  Other . . . spoon.
Child  Other . . . spoon. Now give me other one spoon?
And then there was the dialogue from McNeill 1966 that we quoted in chapter 1: 
Child  Nobody don’t like me.
Mother  No, say “Nobody likes me.”
[seven more times!]
Mother  Now listen carefully. Say “Nobody likes me.”
Child  Oh, nobody don’t likes me. 
Such anecdotes illustrate many points about the language learning pro­
cess, but the irrelevance of imitation is not one of them. As Fromkin and Rod­
man properly noted, parents do not often correct children as in the preceding 
dialogues. Absent the watchful eye of a psycholinguistics researcher’s camcorder 
hovering over the parent-child dyad like a censorious Big Mother, parents are 
usually concerned more with the meaning than the form of their children’s 
speech. They usually do not try to teach something the child is not ready to 
learn. To make this point more plainly, consider the following, hypothetical 
dialogue: 
Child  Want other one spoon, Daddy.
Father  Where are the Himalaya mountains?
Child  Yes, I want other one spoon, please, Daddy.
Father  Tuesday is National Kumquat Day.
Child  Other . . . one . . . spoon.
Father  Pass the Chateaubriand, please.
Child  Other.
Father  You have a leak in your radiator.
Child  Spoon.

186  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
Father  I do solemnly swear to tell the whole truth.
Child  Other . . . spoon. Now give me other one spoon?
Now this dialogue is patently absurd, and what is most absurd about it is that 
the father’s speech at no point resonates with the child’s speech. There is no 
topical continuity. 
Fortunately, there are in the annals of child language only a few cases of 
such pathological parenting. In the notorious case of “Genie” (Curtiss 1977), 
a young girl was kept sequestered in a closet until the age of thirteen. Such 
“wolf children,” who are not exposed to natural language until late childhood, 
seem unable to learn language normally in later life. The generative explana­
tion was that, because language is innate, children do not need to be taught it, 
they need only to be “exposed” to language during a critical period of child­
hood. Wolf children, the generative model explained, simply failed to gain 
exposure to language during their critical period. 
But as we have seen, “critical periods” seem mostly to occur before birth, 
and as the preceding absurd dialogue illustrates, more is needed than simple 
“exposure”: Genie was exposed to language through the closet door, but she 
did not learn language. Neither do hearing children of deaf parents do not 
learn spoken language by watching television (Sachs et al. 1981). Because lan­
guage is learned in resonance with behavioral plans, exposure alone is not 
enough. Language must have meaningful consequences in the social and physi­
cal environment. Consider instead a different hypothetical dialogue: 
Mother  What’s this? 
Child  Koo. 
Mother  That’s right! It’s a cow. And what does a cow make? (pointing) 
Child  Mik. 
Mother  Yes! It makes milk. 
From this much more typical dialogue, we see that both the behaviorists 
and their generative critics had the imitation game backward. Of course, chil­
dren cannot imitate their parents. You might as well ask me, a pathetic neo­
phyte pianist, to imitate a Horowitz recording. Rather, it is parents who imitate 
children! The mother’s expansions or recasts of the child’s utterance in the pre­
ceding dialogue are characteristic of caregiver/teacher speech and stand in 
stark contrast to the atopical preceding dialogue. 
Child language researchers found that parents almost never gave children 
explicit grammatical corrections or judgments, and a branch of generative 
philosophy known as learnability theory (Gold 1965, 1967; Wexler and Culicover 
1980; Berwick and Weinberg 1984) argued that language could not be learned 
without such “negative evidence.” This formed yet another generative argu­
ment for the innateness of language. However an increasing body of research 
began to find expansions and recasts to be an effective “teaching method,” one 
that is used universally by first-language caregivers as well as second-language 

WHAT  IF  LANGUAGE  IS  LEARNED  BY  BRAIN  CELLS
?  •  187 
teachers (Cross 1978; Barnes et al. 1983; Bohannon and Stanowicz 1988; Bohan­
non et al. 1990). These researchers have taken the general position that ex­
pansions and recasts constitute speech acts of implicit negative evidence or 
“unconscious” learning (Schmidt 1993, 1994) which vitiate the argument of 
learnability theory. Adaptive grammar concurs with this analysis, but it finds 
that the negative feedback is more deeply implicit still. 
When the child says [ku] and the mother says [kau], what the mother says 
will resonate with the child’s motor plan across the child’s arcuate fasciculus, in 
much the same preconscious fashion as a circular reaction (figure 12.3). But only 
what both the mother and the child say alike will resonate: only what is grammati-
cal will resonate. In our example, only the [k] and the [u] will resonate. What 
the child says incorrectly will not resonate: in our example, the child’s [k-u] 
formant transitions will not resonate with the mother’s [k-a] formant transitions. 
It is not necessary for the mother to say No, that’s wrong! It’s a [kau]. The 
inhibitory surrounds of cerebral cytoarchitecture provide “negative evidence” 
for free. In parental expansions or recasts like That’s right, it’s a cow, correction 
need not be overt; it is automatic. If, as generative philosophy would have it, 
the function of a teacher could be reduced to only saying No! whenever a 
mistake is made, then as surely as birds learn to fly without a teacher, it would 
be true that children learn language without a teacher. 
No! 
But No! is a powerful word, even if parents and teachers don’t have to use it. In 
chapter 10, I suggested that No! unleashes nonspecific arousal which can re­
bound active dipoles. This nonspecific arousal may be nowhere more evident 
than in the screamed No! that heralds the advent of “the terrible twos.” If the 
first word the child learns is mama, then the second word is No! Subsequently, 
the child begins to use no in combination with other words to rebound more 
specific conceptual networks, as in “pivot grammar”
16
 sentences 12.2–12.4 
(Bloom 1970): 
No ’chine. 
(12.2) 
No more. 
(12.3) 
No more noise. 
(12.4) 
These forms are then often succeeded by forms like 12.5–12.9. 
No Fraser drink all tea. 
(12.5) 
No put in there. 
(12.6) 
Don’t want baby. 
(12.7) 

188  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
Allgone milk. 
(12.8) 
Nobody don’t like me. 
(12.9) 
The multiplicity of negative forms (no, don’t, allgone, nobody, etc.) in En­
glish obscures general patterns, but there does seem to be a tendency for an 
emphatic and nonspecific NEG to assume a primacy position in the child’s early 
syntax (Bellugi 1967). Only later does the NEG become reordered, albeit of­
ten still imperfectly realized, into the fluent rhythm of standard English syn­
tax, as in 12.10–12.12: 
Fraser no drink all tea. 
(12.10) 
Milk allgone. 
(12.11) 
I’m not a little girl; I’m a movie star. 
(12.12) 
In chapter 10 I followed Ross in analyzing NEG as being applied to the 
rightmost element of the sentence. I noted that, if this were true, it would imply 
that, contra a central tenet of adaptive grammar, there is a pushdown-store 
automaton somewhere in the brain. In the preceding examples one can see 
what was wrong with that analysis: what is negated is not the rightmost element 
of the sentence containing NEG. What is negated is the rightmost element of 
the preceding sentence, of the preceding conversational turn. Eventually, negative 
morphemes are encoded in the offbeat relation gradient (sentences 12.10–12), 
but as sentences 12.2–9 show, negation is fundamentally not a morphosyntactic 
phenomenon. Negation is primarily—and ontogenetically—a discourse phenom­
enon. It is the rejection of the topicalization of new information. 
Syntax 
Syntactic evidence has long been the foundation of the generative claim that 
language is not learned. We saw in chapter 10 that sentences of considerable 
complexity can be readily accounted for by the general mechanisms of adap­
tive resonance theory, without special appeal to innate homunculi. Neverthe­
less, it might be well to consider one final example, a classic line of argument 
that Pinker (1989) calls “Baker’s paradox” (Baker 1979): 
Irv loaded eggs into the basket. 
(12.13a) 
Irv loaded the basket with eggs. 
(12.13b) 
Irv poured water into the glass. 
(12.14a) 
*Irv poured the glass with water. 
(12.14b) 

WHAT  IF  LANGUAGE  IS  LEARNED  BY  BRAIN  CELLS
?  •  189 
Sentence 12.13a supposedly allows a “locative movement transformation” 
and admits 12.13b, but 12.14a does not admit *12.14b. The questions asked 
are (1) how do children come to produce sentences like *12.14b, which they 
do, when they never hear adults speak such sentences;(2) how do children 
come to stop using such overgeneralized solecisms if adults never correct them; 
and (3) since adults never correct them, how do children distinguish such valid 
and invalid constructions, whose verbs seem otherwise synonymous? Baker’s 
paradox leads to some amusing syntactic puzzles, but like the puzzles created 
by the generative deduction, they are based on unwarranted premises; namely, 
(1) that something “moves” and (2) that this “movement” is governed by logi-
cal, computer-like rules. 
In general, instead of insisting that language is a rule-governed, computer-
program-like system, I assume that language is learned by brain cells. Then 
children can overgeneralize and produce patterns like *12.14b because cere­
bral competition has not yet contrast-enhanced and partitioned their linguis­
tic concepts. Just as a child can call a cow a doggie, a child can say I poured the 
glass with water. Eventually, children learn that there are many different verbs 
which admit many different case frames. Children can make these many and 
fine distinctions because the massively parallel architecture of their cerebrum 
“computes” these patterns with a granularity approaching 1 part in 10
7,111,111
. 
Discrete rules like “locative movement” fail to account for 12.15–12.17 because 
they deny that language can be complex to this degree. Sentence *12.15 illus­
trates this complexity with the verb to fill, which does not admit a locative case 
role in the first place. 
*Irv filled water into the glass. 
(12.15) 
Irv poured the glass with ice water. 
(12.16) 
The waiter poured the glasses with Chateau Petrus 1961. 
(12.17) 
Sentences 12.16 and 12.17 raise paradoxes within Baker’s paradox. If we 
assume language is rule-governed behavior, not only must we explain how 
children stop saying sentences like *12.14b, but now we must also explain why 
adults, once having “acquired” the rule, then go on to violate it with sentences 
like 12.16 and 12.17. The problem with rule-governed explanations of language 
has always been that the rules are more observed in the breach than in the 
observance. 
The brain is complex, more complex by far than I have made it seem in 
these pages. Even so, it seems simpler to allow that language is adaptive: we 
grow, we change, we learn, and our language grows and changes with us. So, 
for example, children stop overgeneralizing to forms like *12.14b when they 
learn that water becomes a kind of default patient of the verb pour: water is what 
one usually pours. In *12.14b, water is hardly the sort of thing one would ex­
plicitly present as new information. In 12.16, however, ice water could be new 
information. More clearly still, in 12.17 or in any context, Chateau Petrus 1961 

190  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
would definitely be rare and new information. As we adapt to new informa­
tion, our grammar must adapt with us. 
Reading 
By the time the child goes to school, a great deal of language has been learned. 
The basic motor plans of language have been laid down in cerebral cortex and 
their rhythms coordinated through cerebellar learning. Many verb patterns 
have also been learned and associated with many appropriate nouns and case 
roles. A basic inflectional grammar has also been learned, a neural network 
which inserts copula and -en into English passive sentences and in a thousand 
other particulars maintains an offbeat grammatical commentary on the seman­
tic substance of the sentence. For the most part, the syntactic order of nouns, 
verbs, and their modifiers need not be learned. These parts of speech follow a 
universal topic primacy gradient which is innate, but which is neither language-
specific nor species-specific. 
In school, reading is the pupil’s first task, and in English it is notoriously 
difficult. Until Chomsky and Halle’s Sound Pattern of English (1968), a consid­
erable amount of research was devoted to the sound-spelling correspondences 
of English, many of which are quite irregular. This research often led to some 
rather fanciful theories. The most infamously entertaining illustration of these 
was probably G. B. Shaw’s spelling of fish as ghoti, using the gh of enough, the o 
of women, and the ti of nation. Linguists were quick to point out that, although 
the o of women is idiosyncratic, Shaw had ignored the fact that the other pho-
neme-grapheme correspondences were context-sensitive: gh only assumes the 
sound of f  in syllable-final position and ti only assumes the sound of /
Ê/ be­
fore a following vowel. These observations point out the fact that reading is 
context-sensitive, but they largely ignored the fact that it is also rhythm-sensi-
tive, as in the reciprocate/reciprocity contrast cited in chapter 8. 
Reading is a double serial process. In the first place, the serial array of the 
written word must be visually processed. As we saw above, one type of dyslexia 
may affect the cerebellar control of eye saccades, which subserves this process. 
Then, the visually scanned information must be associated with one or several 
words—phonological motor maps.
17
 A second type of dyslexia could impede 
this association, and this type of dyslexia may in fact be induced by instruction. 
In English, there is a significant tendency for many poor and “dyslexic” 
readers to “plateau” around the age of ten. One clue to a possible cause of 
reading problems at this level in English comes from Holmes and Singer (1966), 
who found that at this age, knowledge of Greek and Latin roots was a signifi­
cant predictor of subsequent reading achievement. It is mostly only English 
polysyllabic words which have Greek or Latin roots, and it is mostly only these 
same words which exhibit stress alternations in English. Given the implication 
of stress patterns in lexical retrieval by the TOT phenomenon, one suspects 
stress alternations might be implicated in this reading problem. Indeed, many 
poor and/or dyslexic readers can be heard to attempt to read a word like offi-
cial as “off . . . awfick . . . often . . . awful.” 

WHAT  IF  LANGUAGE  IS  LEARNED  BY  BRAIN  CELLS
?  •  191 
Part of the student’s problem here is that he is trying to do exactly what 
his teachers have told him to do: he is trying to sound out the word from left 
to right. But the English stress patterns are predicted by the end of the word 
(Chomsky and Halle 1968; Dickerson 1975; Dickerson and Finney 1978). In 
words like official, failure to first find the right stress pattern can entail failure 
to find the right initial segment. Without the stress or the initial segment, the 
TOT phenomenon predicts the student will be unable to find the word he is 
looking for. Instead, the student should be decoding polysyllabic words from 
right to left! Other factors in reading (e.g., socioeconomic status of the student, 
the intrinsic interest of the reading material) may be more significant to suc­
cessfully learning to read, but the rhythmic integration of syllables into feet 
and words should be no less important to reading than it is to speech, and the 
failure of reading theory to integrate syllables and whole words in this fashion 
may be one reason for the inconclusiveness of the perennial debate between 
“phonics” and “whole-word” approaches to English reading instruction. 
Once the pupil can use printed words to access words in his mind—by 
sound, by meaning, and by rhythm—whole other worlds of vocabulary open 
up. Indeed, whole other languages open up. There is the language of geogra­
phy, the language of history, the language of biology—all the literatures of 
letters and jargons of science, all the languages of the world that use the Roman 
alphabet. (The languages written in other scripts, Korean in hangul, Hindi in 
devanagari—not to mention mathematics—are quite another matter, as are the 
nonalphabetic scripts of Chinese and Japanese.) Reading is fundamental, but 
it is not parochial, and it is a learning activity which extends well into adulthood. 
Adult Learners 
Pronunciation is the aspect of language in which adult second-language learn­
ers most frequently fail to achieve the proficiency of young language learners. 
In learning the pronunciation of a second language, adults encounter five 
difficulties which must be explained. 
The first is in some ways the easiest. Adults often simply don’t hear the 
difference between two sounds. This disorder appears to begin in adolescence, 
when children begin not to hear when their parents ask them to do something, 
and it continues into adulthood. In chapter 7, I discussed a Spanish bilingual 
who distinguished /
Ê
I
t/ and /
Êit/ by tone of voice instead of by vowel formants. 
Adults have already learned a thing or two, and they can figure out what words 
mean without figuring out exactly how they sound. 
The second difficulty is in pronouncing sounds which are