Oxford university press how the br ain evolved language

different from 
those of the adult learner’s native language. For example, German does not 
have an equivalent of the English [w], and when w is written in German, it is 
pronounced as [v]. Most German speakers have difficulty pronouncing w as 
[w]; they pronounce it as [v] in English as they do in German. The third type 
of difficulty, first-language interference, occurs when pronouncing L2 sounds 
which are similar to L1 sounds (Flege 1988). As an example of this difficulty, 

192  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
we might cite the /
I
/-/i/ contrast discussed in chapter 7. As noted in that 
chapter, these difficulties were paradoxical under behaviorist assumptions but 
can be substantially reconciled in terms of cerebral phonological motor maps 
constructed over on-center off-surround anatomies: in such anatomies similar 
sounds can mask each other. 
The fourth problem is difficulty pronouncing the same problem phones 
L1 children have difficulty pronouncing. Examples of these problem phones 
include English /r/ and /I/. The fifth problem is maintaining accurate pro­
nunciation in rapid speech. L1 problem phones tend to require “nonquantal” 
articulations which, like rapid speech, require exceptionally fine motor con­
trol. Adaptive grammar suggests that rapid and problem phones are difficult 
for adult learners because the phones require fine cerebellar learning. As we 
saw in chapters 4 and 5, the cerebellum lacks the on-center off-surround ar­
chitecture of the cerebrum and is therefore incapable of rapid, rebound-
mediated learning. Consequently, both the L1 child and the L2 adult learn these 
pronunciations slowly. 
Adults’ capacity to master these last two categories of pronunciation also 
appears to decrease with age more rapidly than their capacity to master the 
first two categories. Insofar as this is true, it appears to be akin to the adult’s 
decreasing capacity to learn fine motor skills like playing the violin, so it seems 
misleading to attribute the child’s advantage to some innate, language-specific 
capacity. Moreover, the decrease in these capacities appears to be relatively 
linear, so it also seems misleading to characterize it as the result of some “criti­
cal period.” Adaptive grammar attributes the child prodigy’s advantage to the 
fact that the cerebellum is a tabula rasa at birth and a slow, nonplastic learner 
thereafter. 
Because basic syntax does not directly involve the cerebellum, it is normally 
learned much better than pronunciation, but insofar as L2 inflectional mor­
phology involves sound and rhythm, it depends more heavily upon cerebellar 
processes and presents an intermediate degree of difficulty. Failure to com­
pletely control morphology is often called fossilization (Selinker 1972), an apt 
term within adaptive grammar, implying as it does the long-term memory cal­
cification of postsynaptic membranes. 
Writing 
After reading, there is writing. At first, this is simply “penmanship,” a fine motor 
skill requiring endless drill in order to train the punctilious cerebellum. But 
writing—and the more inclusive term, literacy—have consequences far beyond 
penmanship. Tannen’s book Spoken and Written Language (1982) is a seminal 
analysis of the systematic stylistic differences between spoken and written com­
munication; other authors (e.g., Ong 1982; McLuhan 1965; Olson 1991) have 
also called attention to the broad, cultural consequences of literacy. Here I 
would note only that in the family and in the neighborhood school, the pupil’s 
language, whether spoken or written, tends at first to be parochial. The words 

WHAT  IF  LANGUAGE  IS  LEARNED  BY  BRAIN  CELLS
?  •  193 
she uses are the words of the caregiver and teacher. But once a word is com­
mitted to writing, it lives beyond the moment, beyond the neighborhood, 
beyond the time when its network can be modified, modulated, or even re­
bounded by immediate feedback from an interlocutor. The written word acts 
on a universal and eternal stage. Once the pupil learns this fact, writing be­
comes more than a motor act, it becomes a public behavior in the world and a 
public behavior in history. 
In the limit, the activated network which wills an act of writing must be 
planned in the context of all the competing networks of all possible readers 
that the author can imagine, in the context of all consequences of conceivably 
practical bearing. If it is in reading that students first really sample the diver­
sity of other minds, other times, and other cultures, then it is in writing that 
this learning becomes behavior. In writing, the child achieves a kind of socio­
synaptic maturity and becomes a publicly communicating node in the society 
of Homo loquens. In the task of writing to a pluralistic and changing social envi­
ronment, the student learns to say No! to the parochial, to rebound her ex­
pectancies, and to imagine other cultures and contexts where meaning might 
be found. 

This page intentionally left blank 

NOTES
CHAPTER  1 
1.  Aristotle, De partibus animalium 2.7. 
2.  See DeFelipe and Jones 1988 for a recent appreciation of Ramón y Cajal’s work. 
3.  The force of this argument still holds over a wide range of assumptions. Esti­
mates of the number of human neurons vary between 10
10
 and 10
12
, so Jackendoff’s 
10
10
 is reasonably conservative. While 1.8 overestimates 
ν by implying that two subnet­
works of neurons differing by only one connection can encode distinct representa­
tions (a most unlikely case), it also underestimates 
ν by disregarding orderings among 
the synapses. If order were important, and were computed as permutations, it would 
be larger by a factor of k! Order is important, but it is not that large a factor. In gen­
eral, k can vary by ±4 orders of magnitude and still 
ν >> 10
1111
. Throughout the text, I 
estimate 
ν by 10
7,111,111
 to emphasize the fact that 
ν is big. 
4.  As just one example of this presumption, consider Simon and Feigenbaum 
1979: “[we assume that] the central processing mechanism operates serially and is 
capable of doing only one thing at a time.” 
5.  The story bears repeating. Einstein could not abide Heisenberg’s conclusion 
that wave-particles could not be definitively measured but only statistically estimated. 
“Nonsense, Niels!” Einstein exclaimed. “God doesn’t play at dice!” “Albert!” Bohr 
replied. “Stop telling God what to do!” 
CHAPTER  2 
1.  On the evidence of electroencephalograph (EEG) data, Poizner et al. (1987) 
concluded that ASL is organized in the same cortex as spoken language, but recent 
functional magnetic resonance imaging (fMRI) evidence points to a much greater 
involvement of visual and right-hemisphere cortex in sign language than in spoken 
language (Neville and Bavelier 1996, see also Kimura 1988). 
2.  I found this account of Miller’s experiment in Isaac Asimov’s Beginnings (1987). 
3.  As has been often remarked, the cockroach has been around a good while 
longer than Homo loquens and shows every sign of being around a good while longer. 
195 

196  • 
NOTES  TO  PAGES 
34–66 
One should not mean disrespect when using terms like “lower phyla”; “higher phyla” 
may not get the last laugh. 
4.  These are analogous to proprioceptive stretch receptors in the vertebrate 
locomotor system. 
5.  The first vertebrates in the fossil record, the fishy ostracoderms, appeared in 
the late Cambrian period. 
CHAPTER  3 
1.  A few classes of neurons have short axons, but these are very much the exception. 
2.  In a sense, every neuron is an interneuron, since every neuron communicates 
between two points. Nevertheless, the term is loosely used to locally distinguish neu­
rons that are not major input or output neurons. 
3.  As of this writing, nearly 1,000 human G-protein receptors have been reported 
in World Wide Web databases. Fortunately for neuroscience, which is already compli­
cated enough, most of these are not CNS receptors; unfortunately, CNS receptors still 
number in the hundreds. 
4.  The principal non-NMDA receptors are quisqualate and kainate receptors. 
5.  There are also autoreceptors for serotonin and other messengers. These respond 
to the cell’s own released neurotransmitters. 
CHAPTER  4 
1.  In recent years the term “neocortex,” or simply “cortex” (from the Latin for 
“rind”) has come to be preferred to “cerebrum.” Unfortunately, the cerebellum also 
has a “rind,” but its cortex is of an entirely different neural design from that of cere­
bral cortex, and for a clear understanding of language, it is necessary to understand 
the differences between the two architectures. To keep these differences in the fore­
front, I will persist in using the older term, “cerebrum.” 
2.  This pyramidal tract was originally named, not for the pyramid-shaped pyra­
midal cells which were later found to give rise to it, but rather for the pyramid-shaped 
gyrus of the medulla from which the spinal nerves descend. 
3.  Given three squares and an ordered stack of playing cards on Square 1 (ace 
on top, 2, 3, . . . J, Q, K), the Tower of Hanoi task is to re-create the ordered stack on 
Square 3 by moving one card at a time to another square, never placing a higher 
card on a lower card. The solution is recursive, and since recursion is a powerful 
computational tool, a generation of researchers took this otherwise trivial puzzle to 
be cognitively significant. 
Damasio et al.’s 1996 finding that proper names localize to the left temporal pole, 
proximate to the hippocampus, also suggests that HM’s particularly dramatic inabil­
ity to remember even his doctor’s name may have been due not only to resection of 
the hippocampus but also to resection of adjacent brain regions. 
4.  As we saw in chapter 3, matters are more complicated still.  Different postsyn­
aptic receptors can respond differently to a single neurotransmitter.  Thus, for example, 
the (
α
1
-adrenergic noradrenaline receptor is excitatory, admitting Ca
2+
 into the cell 
interior, but the ß-receptor is inhibitory, suppressing the cAMP second-messenger 
system (Shepherd 1997). 
5.  Most aphasias present a complex of deficits, and “pure” aphasias of one sort 
or the other are rare. 

NOTES  TO  PAGES 
71–82 
•  197 
6.  “Stellate” has been used to describe any vaguely star-shaped cell, whether in­
hibitory or excitatory; here I use it to describe only excitatory cells.  Chandelier cells 
did not stain well in early preparations and were not positively identified until 
Szentágothai himself did so in 1974. 
7.  A “bell-shaped curve” is an approximate example of a Gaussian probability 
distribution.  The “waves and troughs” of these synaptic distributions are exemplified 
in figure 5.4b. 
CHAPTER 5 
1.  This and related chromatic “optical illusions” are known in the literature as 
the McCollough effect (McCollough 1965). 
2.  Note that besides cessation of input and NSA, a rebound could also theoreti­
cally be effected by momentarily increasing or decreasing the inhibitory effects of i
gr 
and i
rg 
. 
3.  Technically, chloride (Cl
–
) channels can still be open and further hyperpolar­
ize a membrane after all Na
+
 channels are closed, but this process is limited, too. 
4.  As we shall see, there are two especially important exceptions to the principle 
of reciprocal, resonant connectivity: inhibitory cells cannot resonantly excite each 
other, and the cerebellum (being principally inhibitory in its Purkinje cell output) is 
not resonant. 
5.  It should be noted that the ART computer simulation presented in this chap­
ter is derived from Grossberg’s earlier, more psychological work. Subsequently Car­
penter and Grossberg developed and popularized a number of computer algorithms 
called ART-I, ART-II, etc. (Carpenter and Grossberg 1987; Carpenter et al. 1991). To 
achieve efficiency, these algorithms abstract away the biological details of Grossberg’s 
original ART. Our interest here is more in biological fidelity than computational effi­
ciency, so the reader should not expect to find a detailed correspondence between 
our models and the later ART models. 
6.  In vivo, ocular dominance columns become convoluted through interactions 
with orientation maps of the selective response of striate cortex barrels to lines in the 
vertical-horizontal plane. See Grossberg and Olson 1994 for a detailed ART model of 
this interaction. 
7.  Technically, Minsky and Papert only claimed to have proved that one class of 
simple perceptrons was incapable of computing XOR. However, they used this result 
to argue more broadly against all parallel computational models. 
CHAPTER  6 
1.  “Vocal fold” is a more descriptive term for these liplike organs, but idealizing 
them as “vocal cords” helps illustrate the acoustic principles behind voiced speech. 
2.  Cycles per second, abbreviated cps is now usually expressed in hertz (Hz) in 
honor of Heinrich Hertz, the German physicist who pioneered classical wave theory. 
3.  Texts in acoustic phonetics normally denote formants using F plus superscript, 
but the neural-modeling literature uses F with superscripts, like F 
1
, to identify neuron 
fields. We therefore use F plus subscript to denote formants (e.g., F
1
). 
4.  Vowels are also described with the tongue-place features of high, middle, and 
low and of front, central, and back. 
5.  In English and many other languages, /g/ and /k/ have a variant articulation, 
a palatal pronunciation, in which the tongue is positioned further forward. This 

198  • 
NOTES  TO  PAGES 
117–139 
allophone typically occurs when /g/ or /k/ precedes a front vowel like /i/. Its formant 
transitions have a different spectrographic signature from velar /g/ and /k/. 
CHAPTER  7 
1.  The effects discussed in this section have also been discussed as a “perceptual 
magnet effect” (Kuhl 1991) or an “anchor effect” (Sawusch and Nusbaum 1979; 
Sawusch et al. 1980). See also Flege 1988 and Rochet 1991. 
2.  A number of these issues surrounding NSA and bilingualism are addressed in 
Rose 1993. 
3.  As we will see in chapter 12’s discussion of Kim et al. 1997, bilinguals who learn 
their second language later in life can apparently also recruit wholly distinct A
3
–level 
polypoles for the second language. 
CHAPTER  8 
1.  In this chapter we use z
ji
 instead of z
ij
 because we are mainly concerned with 
feedback, top-down signals from some higher node xj to some lower node x
i
. 
2.  If a learner is prepped to “memorize the following (long) list,” he will normally 
focus on the beginning of the list, and a primacy effect can result, but very long lists, 
or lists for which the learner is not primed, will exhibit pronounced recency effects. 
3.  Recall from chapter 5 that the 0.5 mm inhibitory radius of an individual in­
hibitory cell also agrees well with the average width of ocular dominance columns in 
visual cortex reported by Hubel and Wiesel (1977). 
4.  See Bullock et al. 1994 and Fiala et al. 1996 for detailed ART models of cer­
ebellar learning. 
5.  Ataxia is the inability to coordinate voluntary movement. Dysarthria is the in­
ability to articulate words, as distinct from the ability to plan or comprehend words, 
which is termed aphasia. 
6.  The Kent et al. (1979) model differs from my account in positing short, in­
hibitory, cerebellar loops which, if handicapped by disease, force the system “to rely 
on longer loops to control movement. Consequently segment durations in speech are 
increased to allow time for the longer loops to operate.” 
CHAPTER  9 
1.  Dolphins and whales, while not exactly bipedal, do also move with a two-beat 
rhythm of their flukes. Similarly, parrots and other birds move with graceful two-beat 
wing strokes, if not with graceful two-beat footsteps. 
2.  Some spectrographic studies (e.g., Delattre 1965) have disputed this claim for 
French, and there is some evidence of stress-timed meter in French poetry (e.g., the 
classical alexandrine). Chinese has been widely taken to be an “isolating” language of 
essentially monosyllabic structure. However, a close study of the Chinese lexicon shows 
that most Chinese words are in fact multisyllabic. In particular, Chinese cheng-yu 
(loosely, “proverbs”) assume a canonical, four-beat structure divided into two feet of 
two beats each. The analogy of this structure to “beats” (as in music) or “stress” (as in 
English) is strengthened when one considers also the Chinese tone alternations which 
occur on the offbeats. Where English words have “stress patterns,” Chinese words have 
“tone patterns,” and just as English offbeats undergo vowel reduction, as in reciprocal-
reciprocity, Chinese offbeats undergo reduction to “neutral tone” (Wu 1992). 

NOTES  TO  PAGES 
139–164 
•  199 
3.  There are instances of stressed inflections and grammatical morphemes, for 
example, in Russian and Turkish. In many cases, these are only stressed where the 
inflection carries propositional content as well as grammatical content. 
4.  Generative grammar called such linguistic regularities as 9.12 “rules.” Instead 
of rule, which implies serial processing, as by a Turing machine, adaptive grammar 
prefers to say that regularities like 9.12 actually reflect patterns of neural resonance. 
5.  However, there are occasions on which children will overgeneralize, produc­
ing plural forms like mens or childrens. 
CHAPTER  10 
1.  We will see in chapter 11 that sentences 10.1 and 10.2 normally do not mean 
the same thing. 
2.  In LISP, dynamic scoping means that a variable is “bound” and has a specific value 
only below a specific tree node. Dynamic scoping is roughly comparable to notions 
like subjacency or c-command in generative systems. 
3.  John McCarthy is generally credited with designing the LISP computer language, 
but McCarthy himself credits Chomsky with convincing him of the value of a rigorous 
implementation of Church’s lambda calculus on a pushdown-store automaton. 
4.  A pidgin language is a simple language that is typically used for trade and 
commerce and that is invented when speakers of two mutually unintelligible languages 
come into contact. A pidgin language evolves into a creole when mothers begin to 
teach pidgin to their children as a first language. In this process, creoles begin to evolve 
the offbeat morphology that is the eventual hallmark of “fully developed” languages. 
5.  See Comrie 1989 for a discussion of accusative and unaccusative languages. 
6.  Linguists refer to unusual or contrasting-in-context linguistic constructions as 
being marked. 
7.  Less centrally, a proposition may also have attached to it a number of “outer” 
case arguments expressing such adverbial or prepositional relations as location, time, 
and manner. These are often expressed or implied “outside” the sentence. 
8.  There are other constraints on particle movement and dative movement. For 
example, we can say He threw back the ball, but not *He threw back the ball to Ted. See 
Fraser 1976 for a detailed analysis. 
9.  Generative linguistics, which recognized the existence of null pronouns in “pro-
drop” languages like Spanish, was reluctant to admit them to the analysis of English. 
In the last several years, generative linguistics appears to have moved toward the posi­
tion developed here. 
CHAPTER  11 
1.  Some scholars have found it ironic that the epigraph to this chapter appeared 
in Peirce’s essay “How to Make Our Ideas Clear” (1878). Perhaps the epigraph has 
suffered in translation from Peirce’s original French. 
2.  Copernicus would no doubt have gotten into trouble, too, if he hadn’t had 
the good sense to die on the day he published De revolutionibus. 
3.  There were, to be sure, earlier translations of the Bible into the vernacular. 
Wycliffe’s translation of the Bible into English (1380–93) was hardly approved by the 
Church, but because of the Great Schism (which began in 1378 and at one point re­
sulted in three contending popes), the Church did not have its act together, and 
Wycliffe died a natural death. However, Wycliffe’s follower, Jan Huss (1369?–1415) was 

200  • 
NOTES  TO  PAGES 
164–178 
burned at the stake for the very idea of a vernacular Bible, and in the same year the 
Church exhumed Wycliffe’s body and burned it, just for good measure. Even Luther’s 
contemporary William Tyndale was executed for his English translation (1534), which 
became the basis of the English King James Version. Luther’s version contains some 
beautiful German, but he’s mostly famous for getting away with it. 
4.  In countries where English is foreign, the term “English as a foreign language,” 
or EFL, is used, but where English is not foreign, one obviously should not call it English 
as a foreign language. In contexts where English is the norm, one speaks of English as 
a second language, or ESL. 
5.  The teaching of ESL remains something of an exception. As in the days of 
behaviorism, the field continues to be an exceptional laboratory for the study of the 
human mind. Indeed, it has raised many of the critical issues addressed by adaptive 
grammar in this book. But the United States no longer needs to import cheap labor. 
As a result, English learners in the United States are stereotyped as “illegal aliens.” 
Illegal or not, they are increasingly unwelcome and disenfranchised. Thus, in this 
political climate, the entire enterprise of ESL has become stigmatized. This is less true 
in non-English-speaking countries of the world. There, EFL has obvious practical bear­
ing and is regarded as meaningful. 
CHAPTER  12 
1.  Kim et al. (1997) did not find separate L1 and L2 regions in Wernicke’s area. 
This could be an artifact of their design, but it might also be evidence in support of 
the single-polypole, contextual dipole model of figure 7.8. 
2.  Derived from the Greek