Oxford university press how the br ain evolved language

Figure 10.6. 
A nominal topic gradient and a verbal relation gradient combine 
rhythmically to generate a sentence. 
in outputting the articles a and the or Prof. Plum or Mrs. White. All such detail is 
suppressed in figure 10.6 in order to more clearly illustrate the essential orga­
nizational principle of the topic/relation dipole. 
The fundamental order in figure 10.6 is established by the topic gradient 
P > W > H > K: Prof. Plum is output first. (This topic gradient is contextual and 
transient. It is defined in STM and diagrammed in figure 10.8 with ordered 
STM arrows.) After cerebellar deperseveration and bottom-up rebounds have 
deactivated P and T, R rebounds into activity. V is then activated. Loc (location) 
and Inst (instrument) are “primed” (subliminally activated), because kill has 
a learned association with these case roles in long-term memory (LTM). Killed 
is output, and V and R are deactivated. The T/R dipole rebounds again. Mrs. 
White, the next most active nominal element in the topic gradient, is activated 
and output. The T/R dipole then rebounds again. Loc and Inst are both equally 
activated by R, since in the learned relation gradient, either could be output 
next, but under the current topic gradient, H > K, so Loc is more primed than 
Inst, and in is output next. 
The T/R dipole rebounds again, and the hall is activated and output. The 
T/R dipole rebounds to R, activating Inst and outputting with. Finally, the T/R 
dipole rebounds one last time, outputting a knife. 

NULL  MOVEMENT 
•  157 
Pronouns 
In chapters 8 and 9, we saw how the performed motor nodes of serial lists are 
deperseverated by inhibitory feedback and rebounds. This process seems also 
to explain fundamental universal features of pronouns, clitics, and other pro­
nounlike words. Pronouns and related pro-forms are found in all natural lan­
guages, and figure 10.7 explains why sentences of the type 
Sid hit himself. 
(10.37) 
are universally preferable to sentences of the form 
*Sid hit Sid. 
(10.38) 
Figure 10.7 models pronominalization subnetworks where (a) Sid saw Bill, and 
then either (b) Sid hit Bill or (c) Sid hit Sid. 
For simplicity, figure 10.7a collapses the topic and relation gradients of 
figure 10.6 into a single topic-relation gradient. After (a) Sid saw Bill, /bIl/ is 
deperseverated, so that in (b) the semantic relation hit (Sid, Bill) is expressed 
Figure 10.7. 
Simple pronominalization. 

158  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
as Sid hit him. (By the same principles, He hit him is also predicted; for clarity, 
figure 10.7 only diagrams one pronoun.) 
Figure 10.7c considers the semantic relation hit (Sid, Sid). After /sId/ is 
initially pronounced and deperseverated, only the motor plans for /hIm/ and 
/hIms
°lf/ can be activated without contrastive stress. Although /hIm/ is the 
more frequent (and so has the larger LTM trace), /hIms
°lf/ is also activated 
by T1, the primary topic, Sid. By contrast, in (b), because Bill is not the pri­
mary topic, /hIm/ is output. There is a certain similarity between this expla­
nation and the generative notion of traces. 
When a linguistic element was moved, generative linguists believed that it 
left behind a residual trace. Thus, in 10.39 a trace t
i
 of Neil
i
 was believed to re­
main in the embedded clause of the surface structure: 
Neil
i
 was believed [t
i
 = Neil
i
] to have destroyed the evidence. 
(10.39) 
Traces explained why, after hearing 10.39, one can reply Neil without any hesi­
tation to the question Who might have destroyed the evidence? But since nothing 
moves, adaptive grammar analyzes this trace as simply a “null pronoun,” the 
completely inhibited motor plan of its antecedent.
9 
The Scope of Negation 
Adaptive grammar also offers an explanation of the “scoping of negation.” 
Consider 10.40, for which four interpretations (10.41–10.44) are possible: 
John didn’t eat the pizza quickly. 
(10.40) 
John didn’t (NEG eat the pizza quickly). 
(10.41) 
John didn’t (NEG eat) the pizza quickly. 
(10.42) 
John didn’t eat the (NEG pizza) quickly. 
(10.43) 
John didn’t eat the pizza (NEG quickly). 
(10.44) 
Example 10.41 interprets NEG as negating the entire scope of the verb 
phrase eat the pizza quickly, but it is more likely that John did eat the pizza—he just 
didn’t eat the pizza quickly. Examples 10.42 and 10.43 are possible readings, 
but normally would be spoken with contrastive stress on the italicized words. 
The normally preferred specific reading is that quickly is being negated (10.44), 
and this pattern is common enough that Ross (1978) proposed a “rightmost 
principle of negation,” which assigns negation to the final constituent of a sen­
tence. Adaptive grammar makes a similar analysis. In 10.40, quickly, eat, and pizza 
are all activated in STM and so are potential “attachment points” for NEG. At 
the end of the sentence, NEG would be applied globally, presumably as a burst 

NULL  MOVEMENT 
•  159 
of nonspecific arousal, and the least-activated conceptual subnetwork, that 
which encodes the newest information, is rebounded. 
But once NEG is encountered in the sentence, how is NSA suppressed until 
the end of the sentence? Is there a pushdown-store automaton in the human 
brain after all? And when NSA is finally released, how is it constrained so as to 
rebound only the rightmost element? Adaptive grammar has answers to these 
questions, but they are not syntactic. They must wait until chapter 12. 
Questions: Extraction and Barriers 
Finally, we return to the questions raised by sentence 2.2/10.18. 
Is
2
 the man who is
1
 dancing p
2
 singing a song? 
(10.18) 
Generative linguists thought the generation of 10.18 involved (a) the extrac­
tion of an element (is
2
) from one place (p
2
), (b) its “movement” to another 
place, and (c) an elaborate set of principled “barriers” which would, for ex­
ample, prevent is
2
 from moving to the front of the sentence. Figure 10.8 ac­
counts for 10.18 without recourse to metaphors of movement. 
English yes/no questions like 10.18 are initiated by an auxiliary verb. En­
glish Aux and related modal verbs carry the epistemological status of a propo­
sition (Givón 1993). In English, this association between epistemological status 
(? in figure 10.8) and Aux is learned as part of the grammar, so in figure 10.8, 
LTM traces order Aux before the rest of the sentence, S. After Is is output, the 
Aux-S dipole rebounds, and S initiates activation of the T/R dipole at T. (The 
dashed LTM trace from S to R suggests that in VSO languages, if in fact there 
are such, S can learn to initiate activation of the T/R dipole at R.) Thereafter, 
the T/R dipole oscillates in phase with the foot dipole of chapter 9. As was men­
tioned in the discussion of figure 10.6, T and R need not rebound on every 
foot. 
The first nominal concept to be activated, N
1
, is the topic, man. All sub­
stantives can be phonologically realized as either a phonological form 
Φ or Pro. 
For simplicity, figure 10.8 only diagrams 
Φ and cerebellar deperseveration for 
the instance of man (i.e., /mæn/). At t
2
, /mæn/ is output and deperseverated. 
Now the relative clause S
rel
 is activated. This activation is displayed with an STM 
arrow because relative clauses are not always attached to nominals. (The or­
dering of relative clauses, however, is language-dependent and must be learned 
at LTM traces, which, for simplicity, are not diagrammed in figure 10.8.) 
The relative clause, S
rel
, (re)activates the sentential rhythm dipole at T. In 
this case, the topic of the embedded relative clause is also the nominal con­
cept man. Since 
Φ has been deperseverated, Pro now becomes active, and who 
is output at t
3
. At t
4
, the dipole switches back to R. Aux and V are activated and 
is dancing is output. 
Bottom-up deperseveration and rebounds now deactivate Vp, S
rel 
, and N
1
. 
The top-level dipole rebounds to R, and the top-level Vp is activated. Aux, how­

160  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
Figure 10.8. 
Generation of questions and relative clauses. 
ever, has already been performed and deactivated, so V = singing is output at 
t
5
. Finally, the top-level T/R dipole rebounds back to T; N
2
 is activated and a 
song is output at t
6
. 
Having dispensed with the need for generative linguistics in this chapter, I should 
close by crediting generative theory with anticipating many of the key elements 
of adaptive grammar. Figure 10.8, for example, builds on generative trees, which 
were generally correct in their structure, if not in their operation. Generative 
linguistics also correctly predicted the existence of an “abstract, autonomous” 
grammar, a relational system which functions quite independently of “real-world, 
substantive” cognition. However, the generative assumption that sentences are 
generated by movement proved a bad choice of metaphor. Nothing moves. Lan­
guage needs relevance, and syntax is ordered by topicality. To be useful for sur­
vival, grammar must relate to a topic; otherwise, it has no meaning. 

TRUTH AND CONSEQUENCES
•
161
• 
E 
L 
E 
V 
E 
N
• 
Truth and Consequences
Consider what effects, that might conceivably have practical 
bearings, we conceive the object of our conception to have. 
Then, our conception of these effects is the whole of our 
conception of the object. 
C. S. Peirce, the Pragmatic Maxim
from “How To Make Our Ideas Clear (1878) 
In the last chapter we saw that the topic—that which we are talking about— 
plays a privileged role in ordering our unfolding motor and language plans, 
our sentences. But some topics never seem to arise. For example: 
The King of France is bald. 
(11.1) 
You think therefore I am. 
(11.2) 
The human race has never existed. 
(11.3) 
Every bachelor is an unmarried man. 
(11.4) 
One would be very surprised to stray into a discussion on one of these topics at 
a cocktail party. As we first noted in connection with 11.1, the problem seems 
to be not so much that such sentences are false as that they are simply void. 
They are meaningless. Even 11.4, which is very, very true, is very, very trite. 
While it is easy to say that sentences like 11.1–11.4 are meaningless and 
that topics must be meaningful, it is quite a bit more difficult to clarify just what 
makes an idea meaningful, as Peirce’s above attempt illustrates.
1
 So let us first 
try to clarify Peirce. Consider the following sentences: 
Hands up or I’ll shoot! 
(11.5) 
Global thermonuclear war will begin any minute. 
(11.6) 
161 

162  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
Unlike sentences 11.1–11.4, these sentences bear on matters of life and death. 
Presumably, they have a great deal of what Peirce would call “practical bear­
ing.” Being in the “future tense,” neither one would be strictly True before the 
fact, but either would, in sincere context, be Very Meaningful. Truth and 
Meaning are not necessarily the same thing. 
Sentences 11.5–11.6 are over-the-top, “Hollywood” examples of Meaning, 
and as a philosopher of science, Peirce would no doubt have found them crass. 
Only in a footnote to a later (1893) edition of his essay did Peirce deign to 
give popular expression to his “Pragmatic Maxim”: 
Before we undertake to apply this rule, let us reflect a little upon what it im­
plies. It has been said to be a skeptical and materialistic principle. But it is only 
an application of the sole principle of logic recommended by Jesus: “Ye may know 
them by their fruits,” and it is very intimately related with the ideas of the Gos­
pel. We must certainly guard ourselves against understanding this rule in too 
individualistic a sense. (Peirce quoted in Wiener 1958, 181n) 
The too-individualistic sense against which Peirce warns us was William 
James’s sense of pragmatism. Born the first son of a wannabe Harvard profes­
sor (Henry James, the elder), William James succeeded where his father had 
not. In that previous heyday of American capitalism at the turn of the last cen­
tury, James popularized Peirce’s notion of pragmatism with movie marquee 
rhetoric: “the cash-value of true theories,” “truth is what works.” Blessed with 
this clear (some would say pandering) style, James succeeded in becoming a 
Harvard professor and celebrated as the “Father of American Pragmatism.” 
By contrast, Peirce was the precocious son of a respected Harvard math­
ematics professor. He no longer aspired to status. In his 1859 Harvard class book 
he inscribed the following: 
1855  Graduated at Dixwell’s and entered College. 
Read Schiller’s Aesthetic Letters & began the study of Kant. 
1856  Sophomore: Gave up the idea of being a fast man and 
undertook the pursuit of pleasure. 
1857  Junior: Gave up the pursuit of pleasure and undertook to 
enjoy life. 
1858  Senior: Gave up enjoying life and exclaimed “Vanity of 
vanities!” 
Disdainful of vanity, Peirce was an intensely original thinker whose writing 
seems always contorted to avoid the popular clichés of his day. No member of 
the Get-along-Gang, Peirce was dismissed as arrogant and was little appreciated 
in his own time. For many years, history regarded Peirce’s students and col­
leagues (including John Dewey, E. L. Thorndike, and his sometimes-antagonist 
Josiah Royce) more highly than Peirce himself. Had it not been for the patron­
age of the powerful and influential James, it is possible that Peirce’s work would 
have been totally lost. But as it happened, James’s patronage was also patron­
izing, and his popularization of Peirce’s pragmatism with overly simplistic for­

TRUTH  AND  CONSEQUENCES 
•  163 
mulae like “the true is the expedient” and “faith in a fact helps create the fact” 
would have been plagiarism had it been more astute. 
In Peirce’s view, James confused Truth and Meaning. Meaning resides in 
the practical consequences of the objects of our conception, but what we find 
meaningful may not be True. We are fallible. This insistence on “fallibility” led 
Peirce to rename his philosophy “pragmaticism, which [is a term] ugly enough 
to be safe from kidnappers” (Peirce 1905). As it happened, the times found 
James’s “truth pays” more appealing than Peirce’s Jesus. “Truth pays” had more 
“cash value.” Despite James’s patronage, Peirce died a failure by Hollywood 
standards, impoverished and forgotten. 
To be fair, we should note that from a psychologist’s perspective James’ 
jingles were perhaps defensible definitions of workaday truth, of the rational­
izations and convenient fictions of everyday psychopathy. The difference be­
tween Truth and Meaning may be less of quality than it is of quantity. I suspect 
Peirce would not have objected so strongly if James had said, “What works for 
a long time is true.” James was a psychologist of his day, but Peirce was a scien­
tist, and in the scientific ideal, eternal truths work eternally. The problem is 
that even in science revolutions occur. An Einstein detects a small wrinkle in 
space-time, and suddenly the entire edifice of Newtonian mechanics is reduced 
to a convenient fiction of workaday physics. Science’s quest for long-term 
replicability is certainly noble, but for the individual (and sometimes for the 
species), survival often comes down to short-term, lower case, Jamesian truths. 
If we can’t have truth, we must settle for meaning. 
Truth and Survival in Science 
In his classic study of scientific revolutions, Kuhn’s central example was the 
Copernican Revolution (Kuhn 1957, 1962). He paints a picture of licensed Ptole­
maic astronomers doodling with epicycles, while outside the ivied halls of the 
scientific establishment, Copernicus was meticulously noting small discrepancies 
in measurements and creating the future science of the cosmos. Kuhn exam­
ines the historical and sociological dynamics of these paradigm shifts in engag­
ing detail, but for my money, he doesn’t sufficiently credit economics. The “cash 
value” of Copernicus’s theory wasn’t in its Truth but in its Meaning. 
In the fifteenth century, the expansion of maritime trade led intrepid sail­
ors to challenge the popular notion of a flat Earth. Fifty years before Copernicus’s 
text was published in 1543, Columbus had already reached the East by sailing 
West, and twenty years before Copernicus, Magellan had already circumnavigated 
the globe (1522). To be sure, Ptolemy thought the Earth was spherical, and the 
heliocentric system did not directly improve navigation, but it was still the pros­
pect of riches from world trade and the accompanying need for improved navi­
gation by the stars that paid the salaries of Ptolemaic and Copernican astronomers 
alike. Columbus and Magellan were the ones who conducted the empirical ex­
periments with practical consequences. To paraphrase James, the meaningful 
theory was what people would buy. By 1543, no one was buying Ptolemy, so 

164  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
Copernicus could publish De revolutionibus orbium coelestium, claiming what expe­
rience had found meaningful to also be True. This is what got Galileo into trouble 
with the Church.
2
 The Earth could be round and go around all it wanted, and 
the Church didn’t really care how much money merchants made thereby; it only 
cared that the heliocentric universe not be declared an Eternal Truth. 
Cash value and Truth have been confused in linguistics, too. For Plato and 
Aristotle, linguistics may have been basic research into eternal truths, but for 
the Holy Roman Empire, linguistics had practical consequences. It meant lan­
guage teaching and language learning: teaching and learning the Greek of 
Scripture and the Latin of the Church. Grammar was a core course of the 
medieval trivium, and linguists were primarily language teachers . . . at least 
until the Reformation. 
The Reformation was as much a linguistic revolution as it was a social, 
political, and religious revolution. Luther’s original Ninety-Five Theses (1517) 
are now largely forgotten, but his translation of the New Testament from Latin 
to German (1534) remains a cultural bible.
3
 Coupled with Gutenberg’s inven­
tion of the printing press (ca. 1456), the mass-produced Lutheran Bible soon 
had God speaking directly to the people—in German. Job prospects became 
bleak for Latin and Greek teachers in Germany. 
Although German had a Bible, it still lacked the cultural history and pres­
tige the Romance languages had inherited from Latin. But after Jones’s theory 
of evolution (chapter 2), a new generation of linguists set to work reconstruct­
ing an earlier Germanic language, a sister to Latin, Greek, and Sanskrit. After 
Napoleon’s demise, this newly discovered classical pedigree became German 
nobility’s title to empire, and while demand may have dwindled for Latin and 
Romance-language teachers, the aspiring young German philologist could hope 
for a court appointment to study Germanic and “Aryan.” One such aspiring 
young philologist was Jakob Grimm. In 1808, Grimm was appointed personal 
librarian to the king of Westphalia. Germanic, unlike Latin and Greek, had 
left no written literature from which it could be reconstructed, so Grimm and 
his younger brother, Wilhelm, studied Germanic oral literature. In 1812, they 
published their first collection of fairy tales, Kinder- und Hausmärchen (Children’s 
and Home Tales). In 1830, Jakob and Wilhelm Grimm were given royal appoint­
ments to the University of Göttingen. Germany was no longer a third-world 
country, and the Brothers Grimm were no longer publishing fairy tales. By 1835, 
they had published Die deutsche Heldensage and Deutsche Mythologie (German Hero 
Sagas and German Mythology). 
At the same time that philology was being celebrated in Germany, linguists 
were still being employed as language teachers in the United States. Needing 
a steady influx of immigrants to settle the frontier and expand labor-intensive 
industry, the young nation founded “grammar schools” which employed lin­
guists to teach English as a second language (ESL)
4
 in a New World trivium of 
readin’, writin’, and ’rithmetic. In the United States, bilingualism had practi­
cal bearings, and language teaching was meaningful. It remained meaningful 
until World War I limited immigration and the rise of communism discredited 
bilingualism. To please their patrons and prove their patriotism, Americans 

TRUTH  AND  CONSEQUENCES 
•  165 
became monolingual, and soon language teacher–linguists were no longer 
needed in the New World either. 
After World War II and Hitler’s appropriation of the term “Aryan,” the job 
market for philologists collapsed. But as the world’s only surviving economy, 
the United States suddenly found itself an international power. United States 
soldiers returning home from the war reported with surprise, “No one in Eu­
rope speaks English!” Within a decade, study of modern foreign languages 
became required in every U.S. college and high school. At the same time, the 
“baby boom” produced a 40% increase in the U.S. birthrate. Eventually, the 
baby boom became a student boom, and the demand for linguists to teach 
foreign languages redoubled. Suddenly, linguists could get jobs again. 
Leadership in this new, foreign-language teaching movement came from 
linguists trained in the incompatible methods of philology (the comparative 
method and the contrastive analysis hypothesis) and psychology (habit forma­
tion and interference). As crude as those methods seem today, I still remem­
ber my first pattern practice drill in German: 
Willi 
Was gibt es denn zum Mittagessen. 
Hans  Wahrscheinlich Bratwurst. 
Willi 
Ich habe Bratwurst nicht gern. 
But when the first cohort of multilingual U.S. students and I went abroad, eager 
to strike up conversations about bratwurst, we found that everybody else in the 
world had already learned English! 
Almost simultaneously, oral contraceptives were invented and the baby 
boom became a baby bust. Within a generation, English became the lingua 
franca of the “new world order.” In the United States, there was suddenly no 
longer a pressing national need for foreign languages. Before long, colleges 
and universities had removed their foreign-language requirements. Soon there 
were few foreign-language students, and there were fewer jobs for foreign-
language teachers.
5
 Fortunately, there were other job opportunities for Ameri­
can linguists, but they were top secret. 
At the heart of German war communications in World War II was the Enigma 
Machine. The Enigma Machine was a kind of cryptographic cash register which 
took in a message, letter by letter, and then, by a complex system of gears, put 
out an elaborately transformed and encrypted code. For example, if today were 
Tuesday and e were input as the 1037th letter of the message, then x might be 
the output code. To defeat Germany, the Allies needed to defeat the Enigma 
Machine, and they needed to do it fast. As it happened, in 1936 Alan Turing 
published a paper which mathematically described a universal cryptographic 
cash register, one which could be configured to emulate any kind of real cryp­
tographic device. With the outbreak of hostilities, the cash value of Turing’s 
theory skyrocketed. The German’s Enigma Machine was a “black box”: from 
enemy actions, cryptographers could see what had gone in, and from inter­
cepted enemy radio messages they could see what had come out, but they 
couldn’t see how it did it. The black box had to be “reverse-engineered.” To 

166  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
that end, the Allies immediately began a major war program to build a “Tur­
ing machine” which could emulate the German’s Enigma Machine. At the end 
of the war, the Turing machine was upstaged by the atomic bomb, but the gen­
erals knew that the triumph of the Allies was in large measure the triumph of 
the Turing machine and of a new linguistics, a computational linguistics. 
In 1949, on behalf of the U.S. military and espionage establishments, 
Warren Weaver of the Rand Corporation circulated a memorandum entitled 
“Translation” proposing that the same military-academic complex which had 
broken the Enigma code redirect its efforts to breaking the code of the Evil 
Empire, the Russian language itself. Machine translation became a heavily 
funded research project of both the National Science Foundation and the 
military, with major dollar outlays going to the University of Pennsylvania and 
the Massachusetts Institute of Technology. In 1952, Weaver outlined a strategy 
before a conference of these new code-breakers. The strategy was to first ana­
lyze, or parse, Russian into a hypothetical, abstract, universal language, which 
Weaver called machinese, and then to generate English from this machinese. At 
MIT, the machine translation effort became organized under the leadership 
of Yehoshua Bar-Hillel, and in 1955, Bar-Hillel hired a University of Pennsylva­
nia graduate student who just happened to have written a dissertation outlin­
ing a theory for generating English from machinese. His name was Noam 
Chomsky, he called machinese “deep structure,” and his theory was “genera­
tive grammar.” 
By 1965, however, Bar-Hillel had despaired of achieving useful machine 
translation. The main problem was that the MIT Russian parsing team had “hit 
a semantic wall.” It never succeeded in producing deep structures from which 
Chomsky’s theory could generate English. In describing this semantic impasse, 
Bar-Hillel noted how hard it would be for a machine to translate even a simple 
sentence like 
Drop the pen in the box. 
(11.7) 
The problem was meaning. The problem with 11.7 was that drop, pen, and 
box all have three or more senses (meanings with a small m). Theoretically, some 
3
3
 different sentences could be generated from a deep structure containing 
just those three substantive terms. Consider for example *11.8 and 11.9: 
*Drop the pen in the
det
 box
verb
. 
(11.8) 
?Drop the pen
playpen
/pen
ballpoint
 in the box
trailer
/box
container
. 
(11.9) 
Sentence *11.8 is fairly simple to solve. It can be rejected as an ungram­
matical sentence by a simple generative grammar rule, something like a verb 
may not immediately follow a determiner. But 11.9 is more problematic. In 11.9, both 
pen and box are nouns. Each could be translated by two different Russian words, 
but how was a poor computer to know which one was the right one? Ostensi­
bly for this reason the U.S. government gave up on machine translation in 1966 

TRUTH  AND  CONSEQUENCES 
•  167 
(ALPAC 1966). In point of fact though, the reason was more economic. As evil 
as the Evil Empire might have been, the United States was at the time gearing 
up its war on Vietnam, and Russian-English machine translation was not going 
to be of immediate help. Some research had to be sacrificed for the war effort. 
Physics, computer science, and mathematics all took one step back, and lin­
guistics was volunteered. 
Two years later, in 1968, three discoveries obviated the ALPAC report’s 
criticism of machine translation: (1) Bobrow and Fraser’s description of the 
augmented transition network, (2) Fillmore’s case grammar, and (3) Quillian’s 
semantic networks. 
pen
In fact, Bar-Hillel was much too skeptical. Chomsky (1965) had already 
made considerable progress on sentences like 11.9 with his work on selectional 
restrictions. We can drop a pen
ballpoint
 into a box
container
, but we can’t drop a pen
prison 
into a box
container
. If our friendly neighborhood lexicographer were to define 
ballpoint 
to have the feature +object and pen
prison
 to have the “semantic feature” 
+institution, then a simple grammar rule restricting drop to the selection of a 
direct object which was either +object or –institution would reject *11.10: 
*Drop the pen
prison
 in the box. 
(11.10) 
This is a kind of agreement rule. We could say the semantic features of the 
verb must agree with the semantic features of the direct object, but this solution 
still posed several technical problems. The first problem was finding a way to 
compute agreement between separated phrases—so-called long-distance dependen-
cies. Within just two years of the ALPAC report, Bobrow and Fraser (1969) solved 
the general problem of long-distance dependencies with the augmented transi­
tion network (ATN). Whereas the lambda calculus and the pushdown-store au­
tomaton had two Turing machines working together, the ATN formalism had 
three: one for program, one for data, and one for agreement. 
A second problem arose when the verb and the direct object underwent a 
passive transformation, as in 11.11: 
The pen was dropped in the box. 
(11.11) 
In this case, semantic agreement needs to be enforced between subject and 
verb, not between verb and direct object. This problem was also solved in 1968 
by Fillmore’s case grammar, which as we saw in chapter 10 replaced terms like 
“subject” and “direct object” with terms more appropriate to computing se­
mantic agreement on selectional restrictions, terms like “actor” and “patient.” 
Finally in 1968, Quillian published his ideas on semantic networks. The gist 
of Quillian’s idea is illustrated by figure 11.1. Pen has (at least) three senses. 
Pen
1 
is a tool. This is represented in figure 11.1 with an ISA link from pen
1 
to tool. In figure 11.1, a tool also ISA instrument, which illustrates the easy 
linking of a semantic network to case grammar. Pen
2 
ISA enclosure, as is 
pen
3
. Pen
1
, however, is FOR writing, while pen
2
 is FOR animals and pen
3
 is FOR 
criminals. 

168  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE