Knowledge of

. Language:
Its Nature, Origin, and Use
Noam Chomsky

CONVERGENCE
A Series Founded, Planned, and Edited
by Ruth Nanda Anshen

Westport, Connecticut
IPMIEGIER london
I
)

I
t
I

1 Kowledge of Language

) as a Focus of Inquiry

The study of language has a long and rich history, extend

ing over thousands of years. This study has frequently been
understood as an inquiry into the nature of mind and thought
on the assumption that "languages are the best mirror of the
human mind" (Leibniz). A common conception was that "with
respect to its substance grammar is one and the same in all
languages, though it does vary acciden tally" (Roger Bacon).
Th e invariant "substance" was often taken to be the mind and
its acts; particular languages use various mechanisms-some
rooted in human reason, others arbitrary and adventitious-for
the expression of thought, which is a constant across languages.
One leading eighteenth century rational grammarian defined
"general grammar" as a dedw;:tive science concerned with " the
immu tabJe and general principles of spoken or written lan
guage" and their consequences; it is "prior to all languages, "
because its principles "are the same as those that direct human
re son in its intellectual operations" (Beauzee). Thus, "the
science of language does not differ at all from the science of
thought. " " Particular grammar" is not a true "science" i n the
sense of this rationalist tradition because it is not based solely
on universal necessary laws; it is an "art" or technique that
shows how given languages realize the general principles of
human reason. As John Stuart Mill later expressed the same
leading idea, "The principles and rules of grammar are the
means by which the forms of language are made to correspond
with the universal forms of thought. . . . The structure of every
sentence is a lesson in lo_g-ic." Others. oarticularlv chuinP" thP
2 1 KNOWLEDGE OF LANGUAGE KNOWLEDGE OF LANGUAGE AS A FOCUS OF INQUJRY I 3

Roma ntic period, argued that the nature and conten t of thoug!lt The generative grammar of a particular language (where
are determ ined in part by the devices made available for Its "generative" means nothing more than "explicit") is a theory
expression in partic ular languages. These devices may i nclude that is concerned with the form and meaning of expressions of
contrib utions of individ ual genius that affect the "chara cter" this language. One can imagine many different kinds of
of a langua ge, enrich ing its means of expression and the approach to such questions, many points of view that might be
though ts expres sed withou t affecti ng its "form ," its sound adopted i n dealing with them. Generative grammar limits
system and rules of word and senten ce forma tion \ Humb ldt). itself to certain elemen ts of this larger picture. Its standpoint is
With regard to the acquis ition of knowl edge, tt was wtdely that of individual psychology. It is concerned with those aspects
held that the mind is not "so m uch to be fil led therew ith from of form and meaning that are determined by the "language
withou t, like a vessel, as to be kindled and awaked " ( Ralph faculty, " which is understood to be a particular component of
Cudwo rth); "The growth of knowledge . . . [rather resemb lesJ. .. the human mind. The nature of this faculty is the subject
the grow th of Fruit; however externa l causes may in some mtter of a general theory of lin g uistic structure that aims to
degree cooper ate, it is the interna l vigour , and virtue of the tree, discover the framework of principles and elements common to
that must ripen the j uices to their j ust maturi ty" (james Harris) .1 attainable human languages; this theory is now often called
Applie d to language, this essenti ally Platon istic conception "universal gram mar" (UG), adapting a traditional term to a
wou ld suggest that know ledge of a particu lar l anguag e grows new context of inquiry. UG may be regarded as a characteriza
and matures along a course that is in part intrins ically deter tion of the genetically determined language faculty. One may
mined, with modific ations reflecti ng observed usage, rather in think of this facufty as a "language acq uisition device, " an
the manner of the visual system or other bodi ly "organ s" that innate component of the human mind that yields a particular
develop along a course determ ined by genetic instructions under language through interaction with presented experience, a
the trigger ing and shaping effects of environ mental factors: device that converts experience into a system of knowledge
Wi th the excepti on of the relativi sm of the Roman tics, attained: know ledge of one or another language.
such ideas were genera lly regarded with m uch disapproval in The study of generative grammar represented a significant
the mainst ream of linguis tic research by the late ninetee nth shift of focus in the approach to problems of language. Put in
century and on through the 1 950s. In part, this attitude developed the simplest terms, to be elaborated below, the shift of focus was
u nder the impact of a rather narrow ly constru ed empiri cism from behavior or the products of behavior to states of the
and later behavi orist a nd operat ionalis t doctrin e. In part, it mind/brain that enter into behavior. If one chooses to focus
resulted from the quite real and impressive successes of historical attention on this latter topic, the central concern becomes
and descrip tive studies conduc ted within a narrower compass, knowledge of language: its nature, origins, and use.
specific ally, the discovery of "sound laws" that pro ided m .uch The three basic questions that arise, then, are these:
understand ing of the history of languages and theu relation (i) What constitutes knowledge of language? (1)
ships. I n part, it was a natura l conseq uence of the investig ation (ii) How is knowledge of language acquired?
of a much richer variety of languages than were known to (iii) How is knowledge of language put to use?
earlier scholar s, languages that appeared to violate many of the
al legedly a priori concep tions of the earlier rati nalist traditi n. 2 The answer to the first question is given by a particular
After a century of genera l neglect or obloqu y, tdeas resemb hng generative grammar, a theory concerned with the state of the
those of the earlier tradition re-emerged (initial ly, with virtual ly rnind/brain of the person who knows a particular language.
no awaren ess of historical antecedents) in the mid- 1 950s, with The answer to the second is given by a specification of UG
the develop ment of what came to be called "generative along with an account of the ways in which its principles
grammar" -again , revivin g a long-lapsed and largely forgotten interact with experience to yield a particular language; UG is a
tradition. 3 theory of the "initial state" of the language faculty, prior to any
4 I KNOWLEDGE OF LANGUAGE
KNOWLEDGE OF LANGUAGE AS A FOCUS OF INQUIRY 1 5

linguistic experience. The answer to the third question would

matters of fact. Similarly, one may argue that the topic of
be theory of how the know ledge of language attained enters generative grammar does not exis t, al though it is hard to see
into the expression of thought and the u nder, s tanding of pre how to make this position minimally plausible. Wi thin the
sented specimens of language, an derivatively, into commun study of generative grammar there have been many changes
ication and other special uses of language. and differences of opinion, often reversion to ideas that had
So far, this is nothing more than the outline of a research been abandoned and were later reconstructed in a di fferent
program that takes up classical questions that had been put light. Evidently, this is a healthy phenomenon indicating that
.
aside for many years. As j ust described, it should not be particu the discipline is alive, although it is sometimes, oddly, regarded
larly controversial, since it merely expresses an interest in certain as a serious deficiency, a sign that something is wrong with the
problems and offers a preliminary analysis of how they might basic approach. I will review some of these changes as we
be confronted, al though as is often the case, the initial formula proceed.
tion of a problem may prove to be far-reaching in its implica In the mid-1950s, certain proposals were advanced as to the
tions, and ultimately controversial as it is developed. form that answers to the questions of ( I ) might take, and a
Some elements of this picture may appear to be more research program was inaugurated to investigate the adequacy
controversial than they real ly are. Consider, for example, the of these proposals and to sharpen and apply them. This program
idea that there is a language faculty, a component of the mind/ was one of the strands that led to the development of the
brain that yields knowledge of language given presented cognitive sciences in the contemporary sense, sharing with
experience. It is not at issue that humans attain knowledge of other approaches the belief that certain aspects of the mind/brain
English, Japanese, and so forth, while rocks, birds, or apes do can be usefully construed on the model of computational systems
not u nder the same (or i ndeed any) conditions. There is, then, of rules that form and modify represen tations, and that are put
some property of the mind/brain that differentiates humans to use in in terpretation and action. From its origins (or with a
from rocks, birds, or apes. Is this a distinct "language facul ty" longer perspective, one might say "its reincarnation" ) abou t 30
with specific structure and properties, or, as some believe, i s it years ago, the study of generative grammar was undertaken
.
the case that humans acquire language merely by applymg with an eye to gaining some insight into the na ture and origins
generalized learning mechanisms of some sort, perhaps with of systems of knowledge, belief, and understanding more
greater efficiency or scope than other organisms? These ae ot broadly, in the hope that these general ques tions cou ld be
topics for speculation or a priori reasoning but for empmcal illuminated by a detailed investigation of the special case of
inquiry, and it is clear enough how to proceed: namey, by human language.
.
facing the questions of (1). We try to determme what 1s the This research program has since been running its course,
.
system of know ledge that has been attained and what properues along a number of different paths. I will be concerned here with
must be attributed to the i nitial state of the m ind/brain to only one of these, with the problems i t faced and the steps that
account for its attainment. Insofar as these properties are were taken in an effort to deal with them. During the past 5-6
language-specific, either i ndividualy or i n the way they are years, these efforts have converged in a somewhat unexpected
organized and composed, there is a distinct language faculty. way, yielding a rather different conception of the nature of
Generative gram mar is sometimes referred to as a theory, language and its mental represen tation, one that offers interest
advocated by this or that person. In fact, it is not a theory any ing answers to a range of empirical questions and opens a
more than chemistry is a theory. Generative gram mar is atopic, variety of new ones to inquiry while suggesting a rethinking of
which one may or may not choose to study. Of course, one can the character of others. This is what accounts for an unmistak
adopt a poin t of view from which chemistry disappears as a able sense of energy and an ticipation-and also uncertain ty
discipline (perhaps it is all done by angels with m irrors). In this which is reminiscent of the period when the study of generative
sense, a decision to study chemistry does stake out a position on grammar in the modern sense was initiated about 30 years ago.
6 I KNOWLEDGE OF LANGUAGE KNOWLEDGE OF LANGUAGE AS A FOCUS OF INQUIRY 1 7

Some of the work now being done is quite different in character with some general commentary and often insightful observa
from what had previously been possible as well as considerably tions. Generative grammar, in contrast, is concerned primarily
broader in empirical scope, and it may be that results of a rather with the intelligence of the reader, the principies and procedures
new kind are within reach, or at least within sight. I would like. brought to bear to attain ful l knowledge of a language. Structur
to try to explain why this may be so, beginning with some alist theories, both in the European and American traditions,
remarks about goals, achievements, and failures of the past did concern themselves with analytic procedures for deriving
years. aspects of grammar from data, as in the procedural theories of
To avoid misunderstanding, I am not speaking here about Nikolay Trubetzkoy, Zellig Harris, Bernard Bloch, and others,
all of the study of language but rather of generative grammar, but primarily in the areas of phonology and morphology. The
and even here I will not a ttempt anything like a real history of procedures suggested were seriously inadequate and in any
the course of research but rather will give a somewhat ideal ized event could not possibly be understood (and were not intended)
picture that is in part clearer in retrospect than it was at the to provide an answer to question (Iii), even in the narrower
time. Furthermore, what I am describing has represented a domains where most work was concentrated. Nor was there an
minority position throughout, and probably still does, although effort to determine what was i nvolved in offering a compre
in my view it is the correct one. A number of differen t current hensive account of the knowledge of the speaker/hearer.
approaches share properties of the sort discussed here and may As soon as- these questions were squarely faced, a wide
be intertranslatable to a considerable extent. I will not consider range of new phenomena were discovered, i ncluding quite
this important topic here and will also make no effort to survey simple ones that had passed unnoticed, and severe problems
the range of ideas, often conflicting, that fall within the partic arose that had previously been ignored or seriously misunder
ular tendency that I will discuss-what is now sometimes called stood. A standard belief 30 years ago was that language acquisi
"government-binding (GB) theory." tion is a case of "overlearning." L;mguage was regarded as a
I want to consider, then, two major conceptual shifts, one habit system, one that was assumed to be much overdetermined
that inaugurated the contemporary study of generative gram by available evidence. Production and i nterpretation of new
mar, and a second, more theory-internal, that is now in process forms was taken to be a straightforward matter of analogy,
and that offers some new perspectives on traditional problems. 4 posing no problems of principle.5 Attention to the q uestions of
Traditional and structuralist grammar did not deal with ( 1) quickly reveals that exactly the opposite is the case: l an guage
the questions of ( I ), the former because of its implicit reliance poses in a sharp and clear form what has sometimes been cal led
on the unanalyzed intelligence of the reader, the latter because "Plato's problem," the problem of "poverty of stimulus," of
of i ts narrowness of scope. The concerns of traditional and accounting for the richness, complexity, and specificity of shared
generative grammar are, in a certai n sense, complementary: a knowledge, given the l imitations of the data available. This
good traditional or pedagogical grammar provides a full list of difference of perception concerning where the problem lies
exceptions (irregular verbs, etc.), paradigms and examples of regu overlearning or poverty of evidence-reflects very clearly the
lar constructions, and observations a t various levels of detail effect of the shift of focus that inaugurated the study of generative
and generality about the form and meaning of expressions. But grammar.
it does not examine the question of how the reader of the A great many examples have been given over the years to
grammar uses such information to a ttai n the knowledge that is illustrate what clearly is the fundamental problem: the problem
used to form and interpret new expressions, or the question of of poverty of vidence. A familiar example is the structure
the nature and elements of this knowledge: essentially the dependence of rules, the fact that without instruction or direct
questions of (I), above. Without too much exaggeration, one evidence, children unerringly use computationally complex
could describe such a grammar as a structured and organized structure-dependent rules rather than computationally simple
version of the data presented to a child learning a language, rules that involve only the predicate " leftmost" in a linear
8 I KNOWLEDGE OF LANGUAGE KNOWLEDGE OF LANGUAGE AS A FOCUS OF INQUIRY I 9

sequence of words. 6 To take some other examples, to which we intransitive forms differ from normal intransitives in other
will return, consider sentences (2)-(7): respects; for example, we can form "the dancing bear" (corres
'.ponding to " the bear that dances" ), bu t not " the eating man"
I wonder who (the men expected to see them] ( 2) (corresponding to ''the man who eats'' ) .8 Such facts pose further
. [the men expected to see them] (3) problems of poverty of stimulus.
John ate an apple (4) Children do not make errors about the interpretation of such
John ate (5) sentences as (6)-( 7) past a certain stage of development, and if
John is too stubborn to talk to Bill (6) they did, the errors would largely be uncorrectable. It is doubtful
John is too stubborn to talk to ( 7) that even the most compendious traditional or teaching grammar
notes such simple facts as those i l lustrated in ( 2)-( 7), and such
Both (2) and (3) include the clause bounded by brackets, but observations lie far beyond the domain of structural grammars.
only in (2) may the pronoun th e m be referentially dependent on A wide variety of examples of this sort immediately come to
the antecedent the men; in (3) the pronoun is understood as attention when one faces the questions formulated in ( l ).
referring in some manner i ndicated in the situational or dis Know ledge of language is often characterized as a practical
course con text, but not to the men. Numerous facts of this sort, a bility to speak and understand, so that questions ( l i) and ( Iiii)
falling under what is now generally called "binding theory," are closely related, perhaps identified. Ordinary usage makes a
are known without relevant experience to differentiate the cases. much sharper distinction between the two questions, and is
Such facts pose a serious problem that was not recognized in right to do so. Two people may share exactly the same knowledge
earlier work: How does every child know, unerringly, to interpret of language but differ markedly i n their ability to put this
the clause differently in the two cases? And why does no peda knowledge to use. Ability to use language may improve or
gogic grammar have to draw the learner's attention to such decline without any change in k nowledge. This a bility may
facts (which were, in fact, noticed only quite recently, in the also be impaired, selectively or in general, with no loss of
course of the study of explicit rule systems in generative knowledge, a fact that would become clear if inj ury leading to
grammar)? impairment recedes and lost ability is recovered. Many such
Turning to examples (4)-( 7), sentence (5) means that John considerations support the commonsense assumption that
ate something or other, a fact that one might explain on the knowledge cannot be properly described as a practical ability.
basis of a simple inductive procedure: ate takes an object, as in Furthermore, even if this view could somehow be maintained,
( 4), and if the object is missing, it is understood as arbitrary. it would leave open all of the serious questions. Thus, what is
Applying the same inductive procedure to (6) and (7), it shou d the nature of the "practical ability" manifested in our interpre
be that (7) means that John is so stu bborn that he (John) will tation of the sentences (2)-( 7), how is i t properly described, and
not talk to some arbitrary person, on the analogy of (6). But the how is it acquired?
meaning is, in fact, quite different: namely, that John is so Often it is not immediately obvious what our knowledge
stubborn that some arbitrary person won' t talk to him (John). of language entails in particular cases, a fact illustrated even
Again, this is known without training or relevant evidence. 7 with short and simple sentences such as (8)- (10):
The situation is, in fact, more complex. Although plausible,
the inductive procedure suggested for the relatively straight h is wife l oves her husband (8)
forward examples (4)-(5) does not seem correct. As noted by John is too clever to expect us to catch Bill (9)
Howard Lasnik, the word eat has a somewhat different meaning John is too clever to expect us to catch (10)
in its intransitive usage, something like dine. One can say
"John ate his shoe, " but "John ate" cannot be understood to In the case of (8), it takes some thought to determine whether
include this case. The observation is general for such cases. The his can be referential ly dependent on h er hu sband if her is
lO I KNOWLEDGE OF LANGUAGE KNOWLEDGE OF LANGUAGE AS A FOCUS OF INQUIRY 1 II

dependent on his wife-that is, if the reference of either he or problems remain, exactly as before, now embedded in termino
she is not somehow contextually indicated.9 Examples (9) and logical confusion. The task of determining the nature of our
( 10) are, in fact, analogous to (6) and ( 7), respectively, but again, knowledge (= abi lityz), and accounting for its origins and use,
it takes some though t to discover that ( 10) means that John is so remains exactly as challenging as before, despite the termino
clever that an arbi trary person can not expect us to catch him logical i nnovations.
(John), al though it is clear at once that it does not mean that O ther examples similar to ( 8) -( 10) raise further questions.
John is so clever that he (John ) cannot catch some arbitrary

Consider the following sentences:

person, on the analogy of (9) (and (4), (5)). Our abilities seem
limited somehow in such cases (and there are far more complex John is too stubborn to expect anyone to talk ( ll)
ones), but it would make little sense to speak of our knowledge to Bill
of language as "limited" in any comparable way.
Suppose we i nsist on speaking of knowledge of language John is too stubborn to visit anyone who talked (12)
as a practical ability to speak and understand. Then normal to Bill
usage m ust be revised in numerous cases such as those jusl
discussed. Suppose that Jones takes a public speaking course Suppose we delete Bill from ( I I ) and (12), yielding ( 13) and ( 14),
and improves his ability to speak and understand without any respectively:
change in his knowledge of English, as we would describe the
situation in normal usage. We must now revise this common John is too stubborn to expect anyone to tal k to (13)
sense usage and say, rather, that Jones has improved his ability1
to use his abilityz to speak and understand; similar translations John is too stubborn to visit anyone who talked to (14)
are required in the other cases. But the two occurrences of
"ability" in this description are hardly more than homonyms. Sentence ( 13) is,structurally analogous to ( I 0), and is understood
Ability1 is ability i n the normal sense of the word: it can in the same manner: it means that Joh n is so stubborn that an
improve or decline, can be inadequate to determine consequen arbitrary person would not expect anyone to talk to him (John).
ces of knowledge, and so on. Ability2, however, remains stable "By analogy," then, we would expect sentence (14) to mean that
while our ability to use it changes, and we have this kind of John is so stubborn that an arbitrary person would not visit
"ability" even when we are unable to detect what it entails in a nyone who talked to him (John). But it does not have that
concrete cases. I n short, the neologism "ability2" is i nvested meaning; in fact, it is gibberish. Here we have a double failure
with all the properties of knowledge. Note that there are cases of analogy. Sentence ( 14) is not understood "on the analogy" of
when we do speak of abilities that we cannot put to use: for (4), (5), (6), (9), and ( 12) (hence meaning that John is so stubborn
example, the case of swim mers who cannot swim because their that he (John) would not visit anyone who tal ked to some
hands are tied, al though they retain the a bility to swim. The arbitrary person) , nor is it understood "on the analogy" of (7),
cases in question are not of this sort, however. ( 10), and ( 13); rather, it has no interpretation at all. And while
The purpose of the attempt to reduce knowledge to abi lity the status of ( I I ), ( 12), and ( 14) is iptmediately obvious, it takes
is, presumably , to avoid problematic features that seem to some thought or preparation to see that ( 13) has the interpreta
inhere in the concept of know ledge, to show that these can be tion it does have, and thus to determine the consequences of our
explained in dispositional or other terms more closely related know ledge in this case.
to actual behavior (whether this is possible even in the case of Again, these are facts that we know, however difficul t it
abili ty1, the normal sense, is another question). But nothing of may be to determine that our system of k nowledge has these
the sort is achieved by this departure from ordinary usage; the consequences. We know these facts without instruction or even
12 I KNOWLEDGE OF LANGUAGE
KNOWLEDGE OF LANGUAGE AS A FOCUS OF INQUIRY I 13

direct evidence, surely wi thout correction of error by the speech facuJty of the mind-the language facu lty-with its specific
com munity. I t would be a bsurd to try to teach such facts as properties, structure, and organization, one "modu le" of the
these to people learning English as a second language, j ust as mind.10
no one taught them to us or even presented us with evidence
that cou ld yield this knowledge by any generally reliable
procedure. This is knowledge without grounds, without good
reasons or support by reliable procedures in any general or NOTES
otherwise useful sense of these notions. Were we to insist that
knowledge is a kind of a bility, we would have to claim that we I. On these and many other discussions, primari ly i n the seven
lack the ability to unders tand "John is too stubborn to talk to" teenth-nineteenth centuries, see Chomsky ( 1966). For di scussion of
s ome misin terpretation of this work, see Bracken (1984).
as meaning "John is too stu bborn to talk to someone or other"
2. The a l leged a priorism of work in this tradition has of ten
(on the analogy of "John ate an apple"- "John ate"), and that
been exaggera ted. See Chomsky ( 1 966) and more recent work for
we lack the ability to understand ( 1 4) on the analogy of "John discussion of this point.
ate an apple"- "John a te" (so that it means that John is too 3 . The tradition, i n this case, is a different one, represented in
stubborn to visit anyone who talked to someone or other) or on its most advnced for m in the early work of the I ndian gram maria ns
the analogy of "John is too stu bborn to talk to, " with the 2,500 years ci go. See Kiparsky ( 1 982). A modern coun terpar t is Bloom
" inversion strategy " tha t we somehow use in this case (so that field ( 1 939), whic h was radically different in c haracter from the work
(14) means that John is too stubborn for someone or other to of the per iod and i nconsistent with his own theor ies of language, and
visit anyone who talked to him, John). But these would be odd remai ned virtually without influence or even awareness despite
claims, to say the least. These are not failures of ability. lt is not Bloomf ield's great prestige.
that we are too weak, or lack some special skill that could be 4. See New meyer ( 1 980) for one view of the history of this
acquired. We are perfectly capable of associating the sentence per iod pr ior to the second major conc eptual shift; and for some more
personal comments, the introduction to Chomsky ( 1 975a), a somewhat
( 1 4), for example, with ei ther of the two meanings that would
a bbreviated version of a 1 956 revision of a 1 955 man uscri pt, both
be provided "by analogy " (or others), but we know that these
unpublished. See Lightfoot ( 1 982) and Hornstei n and Lightfoot ( 1 98 1 )
are not the associations that our knowledge of the language for discussion of the general backgrounds for m uch current wor k, and
provides; ability is one thing, knowledge something quite dif Radford ( 1 98 1 ) for an i n troduction to the work tha t led to the second
ferent. The system of knowledge that has somehow developed conceptual shift. See Chomsky ( 1 98 1 ) for a more technical presentation
in our minds has certain consequences, not others; it relates of some of the ideas that entered i n to this conceptual shift and van
sound and meaning and assigns structural properties to physical R iemsdijk and Williams ( 1 985) for an i n troductor y study of this
events in certain ways, not others. c urrent wor k .
It seems that there is little hope in accounting for our 5. Although basica l l y adopting this point of view, W . V. Quine,
knowledge in terms of such ideas as analogy, induction, associ however, argued that there is a very severe, in fact, insuperable problem
ation, reliable procedures, good reasons, and justification in of underdetermination affecting a l l aspects of language and grammar,
and much of psychology more general ly ( Q ui ne, 1 960, 1 972). I do not
any generally useful sense, or in terms of "generalized learning
think that he succeeded in showing tha t some novel form of i ndeter
mechanisms" (if such exist). And i t seems that we should follow
m i nacy affects the study of language beyond the normal underdeter
normal usage in distinguishing clearly between knowledge m i na tion of theory by evidence; his own for m u la t i ons of the thesis
and ability to use that knowledge. We should, so it appears, fur thermore i n volve i n ternal inconsistency (see C homsky, 1 9 75b ,

think of knowledge of language as a certain state of the mind/ 1980b). T here seems no reason on these grounds , then, to disti nguis h
brain, a relatively stable element in transitory mental states linguistics or psychology in pr inciple from the natural sciences in
once it is attained; furthermore, as a state of some distinguishable accordance with what Hockney (1975) cal ls Quine's "bifurcation
14 . I KNOWLEDGE OF LANGUAGE

thesis." A s i m i lar conclusion is rea ched by P u tnam ( 1 98 1 ) i n his

abandon men t of metaphys ical rea l ism on Q u i nea n grounds. His s tep
also a bandons the bifurca tion thesis, a l though in the opposite
direction. .,
"

6. See C homsky ( l 975a). See Cra i n and Na kayama ( 1984) for

e mpirical study of this q ues tion with 3-5-year-old children.
7. The reaction to such phenomena, also unnoticed until
2 Concepts of Language
recen tly, aga i n i l l ustrates the difference of outlook of structuralist
descriptive and genera tive grammar. For some practitioners of the
former, the sta tement of the facts, which is straightforward enough
once they are observed, is the answer-nothing else is necessary; for
the latter, the s ta tement of the facts poses the problem to be solved. Cf.
Ney (1983), pa rticularly, h is puzzle men t about the " peculiar view of 2.1 THE COMMONSENSE CONCEPT AND
grammar [ that) u n n ecessarily complica tes the w hole ma tter" by seek DEPARTURES FROM IT
ing an expla na tion for the facts. Note tha t there is no q ues tion of right
or wrong here, but ra t her of topic of inquiry.
Let us turn now to the questions of ( I ) of Chapter I. To
8. In early work, such facts were used to motivate an analys is of
begin wi th, let us distinguish the intui tive, pretheoretic com
intransitives such as eat as derived from corresponding transi tives by
a system of ordered rules that excluded the unwan ted cases; see
monsense notion of language from various technical concepts
'
C homsky (1962). that have been proposed with the inten t of developing an
9. On s tructures of this type, and problems of binding theory, eventual science of language. Let us call the latter "scientific
more genera l ly, see H iggin botha m ( 1938a), among much other work. approaches " to language, with an eye directed more toward a
10. See Fodor (1983). But it is too narrow to regard the ''language possible future than a present reality, some might argue. The
module" as an input system in Fodor's sense, if only because it is used scientific approaches, I believe without exception, depart from
in s peaking and thought. We might consider supplementing this the commonsense notion in several ways; these departures also
picture by adding an "outp u t system," but pla i n ly this must be linked affect the concepts of know ledge or understanding of language,
to the input system; we do not expect a person to speak only English use of language, rule of language, rule-guided linguistic
and u n derstand only Japanese. Tha t is, the input and output systems behavior, and others.
must each access a f ixed system of k nowledge. The latter, however, is a
In the first place, the commonsense notion of language has
cen tral system which has essential problems of modularity, a fact that
a crucial sociopolitical dimension. We speak of Chinese as "a
brings the en tire picture into q uestion. Furthermore, even regarded as
an input system, the la ngua ge module does not appear to have the
language, " al though the various "Chinese dialects" are as
property of rapidity of access that Fodor discusses, as indicated by diverse as the several Romance languages . We speak of Dutch
(8)-{ 14). Note also tha t even if Fodor is right in believi ng that there is and German as two separate languages, al though some dialects
a sharp distinction between modules in his sense and "the res t," of German are very close to dialects tha t we call "Dutch" and
which is holis tic in several respects, it does not follow tha t the res idue are not mutually intelligible with others that we call "German . "
is unstructured. In fact, this seems highly u n l i kely, if only because of A standard remark in i ntroductory linguistics courses i s that a
the "epistemic boundedness" that he notes. Many other questions language is a dialect with an army and a navy (attribu ted to
arise concerning Fodor's very i n triguing discussion of these issues, Max Weinreich). That any coherent account can be given of
which I will not pursue here. "language" in this sense is doubtful; surely, none has been
offered or even seriously attempted. Rather, all scien tific
approaches have simply abandoned these elements of what is
called "language" in common usage.1
 16 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 17

The commonsense notion also has a normative-teleological tions (I) of Chapter I , considering only the case of a person
element that is eliminated from scientific approaches. I do not presented with uniform experience in an ideal Bloomfieldian
refer here to prescriptive grammar but to something else. Con speech com munity with no dialect diversity and no variation
sider the way we describe a child or a foreigner learning English. among speakers.
We have no way of referring directly to what that person knows: We should also make note of a more subtle theory-internal
It is not English, nor is it some other language that resembles assumption: The language of the hypothesized speech com
English. We do not, for example, say that the person has a munity, apart from being uniform, is taken to be a "pure"
perfect knowledge of some language L, similar to E nglish but instance of U G i n a sene that must be made precise, and to
still different from it. What we say is that the child or foreigner which we will return. We exclude, for example, a speech com
has a ''partial knowledge of E nglish," or is "on his or her way" munity of uniform speakers, each of whom speaks a mixture of
toward acquiring knowledge of E nglish, and if they reach the Russian and French (say, an idealized version of the nine
goal , they will then know E nglish. Whether or not a coherent teenth-century Russian aristocracy ). The language of such a
accoun t can be given of this aspect of the commonsense termi speech community would not be "pure" in the relevan t sense,
nology, i t does not seem to be one that has any role in an because i t would not represent a single set of choices among the
eventual science of language. options permitted by UG but rather would include "contradic
I will follow standard practice in disregarding these aspects tory" choices for certain of these options.
of the commonsense notions of language and the associated Questions ( 1) of Chapter I, then, arise initially under these
notions of rule-following and so forth, although the departure idealizations, and the same is true, in effect, of other approaches
should be noted, and one may ask whether it is entirely innocent. to language, although the fact is often not explici tly recognized
Modern linguistics commonly avoided these questions by and may even sometimes be denied.
considering an idealized "speech com munity" that is internally The legitimacy of these idealizations has sometimes been
consisten t in its linguistic practice.2 For Leonard Bloomfield, questioned, but on dubious grounds. 4 Indeed, they seem indis
for example, a language is "the totality of utterances that can be pensable. Surely there is some property of mind P that would
made in a speech com munity, " regarded as homogeneous enable a person to acquire a language under conditions of pure
(Bloomfield, 1928/1957). In other scientific approaches, the and uniform experience, and surel y P (characterized by UG) is
same assumption enters in one or another form, explicitly or put to use under the real conditions of language acquisition.
tacitly, in identification of the object of inquiry. No attempt is To deny these assumptions would be bizarre indeed: It would
made to capture or formulate any concept with the sociopolitical be to claim either that language can be learned only under
or normative-teleogical aspects of informal usage of the term conditions of diversity and conflicting evidence, which is absurd,

"language. " The same is true of approaches that understand or that the property P exists-there exists a capacity to learn

language to be a social product in accordance with the Saussur language in the pure and uniform case-but the actual learning
ean concept of "langue." of language does not involve this capaci ty. In the latter case, we
Of course, it is u nders tood that speech communities in the would ask why P exists; is it a "vestigial organ" of some sort?
Bloomfieldian sense-that is, collections of individuals with The natural approach, and one tha t I think is tacitly adopted
the same speech behavior3-do not exist in the real world. Each even by those who deny the fact, is to attempt to determine the
individual has acquired a language in the course of complex real property of mind P, and then ask how P functions under
social interactions with people who vary in the ways in which the more complex conditions of actual linguistic diversity. It
they speak and interpret what they hear and in the in ternal seems clear that any reasonable study of the nature, acquisition,
representations that underlie their use of language. S tructural and use of language in real l ife circumstances must accept these
linguistics a bstracted from these facts in its attempts at theory assumptions and then proceed on the basis of some tentative
construction; we also a bstract from these facts in posing ques- characterization of the property of mind P. In short, the ideali-
18 I KNOWLEDGE OF LANGUAGE
CONCEPTS OF LANGUAGE I 19

zations made explicit in more careful work are hardly contro in scientific approaches. Weaker assumptions than strict identity
versial; they isola te for examination a property of the language would suffice for the discussion below, but this stronger
facul ty the existence of which is hardly in do4bt, and which is assumption seems a reasonable one, to a very good approxima
surely a crucial element i n actual language acquisition. tion, and I wilJ keep to it here.
By making these idealizations explicit and pursuing our
inquiry in accordance with them, we do not in any way prejudice
the s tudy of language as a social product. On the contrary, it is 2.2 EXTERNALIZED LANGUAGE
difficult to i magine how such studies might fruitfully progress Scientific approaches to language, i n the sense of the term
without taking i nto account the real properties of mind that used earlier, have developed various technical notions of lan
enter into the acquisition of language, specifically, the properties guage to replace the commonsense notion. The term "grammar''
of the i n i tial s tate of the language faculty characterized by U G. has also been used i n a variety of ways. In conventional usage, a
Note a lso that the study of language and UG, conducted grammar is a description or theory of a language, an object
within the framework of individual psychology, allows for the constructed by a linguist. Let us keep to this usage. Then
possibility that the state of knowledge attained may itself include associated with the various technical notions of language there
some kind of reference to the social nature of language. Consider, are corresponding notions of grammar and of universal gram
for example, what Putnam (1975) has called "the division of mar (UG).
linguistic labor. " In the language of a given individual, many Structural and descriptive linguistics, behavioral psychol
words are semantically i ndeterminate in a special sense: The ogy, and other contemporary approaches tended to view a
person will defer to "experts" to sharpen or fix their reference. language as a collection of actions, or utterances, or linguistic
Suppose, for example, that someone knows that yawls and forms (words, sentences) paired with meanings, or as a system
ketches are sa iling vessels but is unsure Qf the exact reference of of linguistic forms or events. In Saussurean s tructura lism, a
the words "yawl" and "ketch," leaving it to specialists to fix language (langue) was taken to be a system of sounds and an
this reference. In the lexicon of this person's language, the associated system of concepts; the notion of sentence was left in
entries for "yawl" and "ketch" will be specified to the extent of a kind of limbo, perhaps to be accommodated w i thin the s tudy
his or her k nowledge, w i th an indication that details are to be of langu(lge use. For B loomfield, as noted earlier, a language is
filled in by others, an idea that can be made precise in various "the totality of utterances that can be made in a speech com
ways but without going beyond the study of the system of munity." The American variety of s tructural-descriptive lingu
knowledge of language of a particular i ndividual. O ther social istics that was heavily i nfluenced by Bloomfield's ideas further
aspects of language can be regarded in a like manner-although more concentrated primarily on sound and word s tructure,
this is not to deny the possibility or value of other kinds of study apart from various proposals, notably those of Zel lig Harris, as
of language that incorporate social s tructure and interaction. to how larger units (phrases) could be constructed by analytic
Contrary to what is sometimes thought, no conflicts of principle principles modelled on those i ntroduced for phonology and
or practice arise in this connection. morphology.!i Many researchers today adopt a posi tion of the
We are also assuming another idealization: That the prop sort l ucidly developed by David Lewis, who defines a language
erty of m ind described by U G is a species characteristic, common as a pairing of sentences <!nd meanings ( the latter taken to be
to all humans. We thus a bstract from possible variation among set-theoretic constructions in terms of possible worlds) over an
humans in the language facu l ty. It is plausible to suppose that infinite range, where the language is " used by a population"
apart from pathology (potentially an i mportant area of inquiry), when certain regularities "in action or belief" hold among the
such variation as there may be is marginal and can be safely population with reference to the l anguage, sustained by a n
ignored across a broad range of linguistic investigation. Again, interest in communication. 6
the assumption is c onventional , though generally explicit, in scientific
20 1 KNOWLEDGE OF LANGUAGE
CONCEPTS OF LANGUAGE / 21

Let us refer to such technical concepts as instances of ian" view that "languages could differ from each other without
"externalized language" (-language), in the sense that the lim t and i n unpredictable ways," echoi ng William Dwight
construct is understood independently of the properties of the .
Whitney's reference to "the infinite diversity of human speech "
mind/brain. U nder the same rubric we may include the notion and Edward Sapir's notion that "language is a human activity
of language as a collection (or system) of actions or behavior of that varies without assignable limit. "7 Such statemen ts reflect a
some sort. From a point of view such as this, a grammar IS a fairly broad consensus of the time. Although they could hardly
collection of descriptive statements concerning the -language, have been intended li terally, they did express a relativistic
the actual or potential speech events (perhaps along with some impulse that denigrated the study of UG. More precisely, it
account of their context of use or semantic content). In technical cannot be that human language varies without assignable lim it,
terms, the grammar may be regarded as a function that enu although it might be true that it is "infinitely diverse"; it is an
merates the elements of the -language. Sometimes, grammar empirical question of some in terest whether UG permits an
has been regarded as a property of -language, as in Bloomfield's infinite variety of possible languages (or a variety that is infinite
remark tha t a grammar is "the meaningful arrangement of in more than structurally trivial respects, say, with no bound
forms in a language" (Bloomfield, 1933). Despite appearances, on vocabulary), or only a finite diversity. a
the problem of accounting for the unbounded character of the Nevertheless, significant con tribu tions were made to UG
-language and the person's knowledge of language inluding in our sense within these traditions. For example, the theory of
this fundamental property is not squarely addressed In such distinctive features in phonology, which greatly infl uenced
approaches , a matter to which we will return. s tructuralist studies in other fields, postulated a fixed inventory
The -language is now understood to be the real object of of "atomic elements" from which phonological systems could
study. Gram mar is a derivative notion; the linguist is free to be drawn, with certain general laws and implicational relations
select the gram mar one way or another as long as it correctly .
governing the choice. And it was generally assumed that such
identifies the -language. Apart from this consideration, ques notions as topic and comment, or subject and predicate, were
tions of tru th and falsity do not arise. Quine, for example, has universal features of language, reflecting the fact that a declara
argued that it is senseless to take one grammar rathr than tive sentence is about something and says something about it.
another to be "correct" if they are extensionally equivalent, Later, important work on linguistic universals was conducted
characterizing the same -language, for him a set of expressions by 1 oseph Greenberg and others, yielding many generalizations
(Quine, 1972). And Lewis doubts that there is any way "to make that require explanation, for example, the fact that if a language
objective sense of the assertion that a grammar G is used by a has subj ect-object-verb order, it will tend to have post positions
population P whereas another gram mar G', which generates rather than prepositions, and so on.
the same language as G, is not." Along these lines, then, we rna y develop a certain technical
The notion of -language is familiar from the study of concept of language (-language), and an associated concept of
formal systems, as is the concl usion j us t cited: In the case of the grammar and UG, as a basis for a scientific study of language.
"language of arithmetic," for example, there is no objective Many different specific ideas fall roughly within this general
sense to the idea that one set of rules that generates the well framework.
formed formulas is correct and another wrong.
As for UG, to the extent that such a study was recognized as
legitimate, this theory would consist of statements that are true 2.3 INTERNALIZED LANGUAGE
of many or all human languages, perhaps a set of conditions
sa tisfied by the -languages that count as human languages. A rather different approach was taken, for example, by
Some appeared to deny the possibility of the enterprise, for Otto Jespersen, who held that there is some "notion of structure"
example, Martin Joos, who put forth what he called the "Boas- in the mind of the speaker "which is definite enough to guide
22 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 23

him in framing sentences of his own," i n particular, "free i t is a bout H's brain ( i n particulqr, i ts language faculty) that
expressions" that may be new to the speaker and to others.9 Let corresponds to H's knowing L, that is, by virme of which R(H, L)
us refer to this "notion of structure" as an "internalized lan holds and the s ta tement that R(H, L) is true.
guage" (llanguage). The 1-language, then, is some element of I t is natural to take L to be !-l anguage, Jespersen ' s " notion
the mind of the person who knows the language, acquired by ,
of structure, ,. regarding this as an entity a bstracted from a s tate
the learner, and used by the speaker-hearer. of the language facul ty, the latter being one component of the
Taking language to be llanguage, the grammar would mind. Then, for H to know L is for H to have a certain
then be a theory o f the 1-language, which is the object u nder 1-language. The s tatements of a grammar are statements of the
investigation. And if, indeed, such a "notion of structure" theory of mind about the !-language, hence s tatements about
exists, as Jespersen held, then ques tions of truth and falsity structures of the brain formulated at a certain level of abstraction
arise for gram mar as they do for any sci enti fic theory. This way from mechanisms. These s tructures are specific things in the
of approaching the questions of language is radically different world, with their specific properties. The statemen ts of a
from the one sketched above and leads to a very different con grammar or the s tatement that R(H , L) are similar to statements
ception of the nature of the inquiry. of a physical theory that characterizes certain entities and their
Let us return now to the poi nt of view outlined in Chapter properties in abstraction from whatever may turn out to be the
I. Knowing the language L is a property of a person H; one task mechanisms that account for these properties: say, a n ine
of the brain sciences is to determine what it is about H's brain teenth-century theory about valence or properties expressed in
by virtue of which this property holds. We suggested that for H the periodic table. Statements about 1-language or the statement
to know the language L is for H's mind/brain to be in a certain that R(H, L) ( for various choices of H and L) are true or false,
state; more narrowly, for the language faculty, one module of much in the,way that statements a bout the chemical structure
this system, to be in a certain s tate SL.10 One task of the brain of benzene, or about the valence of oxygen, or about chlorine
sciences, then, is to discover the mechanisms that are the physical and fluorine being in the same column of the periodic table are
real ization of the s tate SL. true or false. The I -language L may be the one used by a speaker
Suppose we analyze the notion "H knows language L" in but not the !-language L', even if the two generate the same class
relational terms, that is, as involving a relation R (knowing, of expressions (or other formal objects) in whatever precise sense
having, or whatever) holding between H and an abstract entity we give to this derivative notion; L' may not even be a possible
L. One might question this move; we speak of a person as human !-language, one attainable by the language facul ty.
knowing U.S. history withou t assumi ng that there is an entity, UG now is construed as the theory of human !-languages,
U.S. history, that the person knows, or knows in part. Let us, a system of conditions deriving from the human biological
however, assume the move to be legitimate in this case. The endowment that identifies the !-languages that are humanly
assumption wil l be j ustified to the extent that this move con accessible under normal conditions. These are the !-languages
tri butes to providing i nsight into the questions that primarily L such that R( H, L) may be true ( for normal H , under normal
concern us, those o f ( l ) of Chapter l ; this would be the case, for conditions).ll
example, if there are significant principles governing the set of Of course, there is no guarantee that this way of approach
postulated entities L. Suppose that we proceed further to regard ing the problems of (I) i n Chapter I is the correct one. This
talk of mind as talk about the brain undertaken at a certain level approach may turn out to be thoroughly m isguided, even if i t
of abstraction at which we believe, rightly or wrongly, that achieves substan tial success-j ust as a theory of valence, etc.
significant properties and explanatory principles can be discov might have turned out to be completely off the track, despite its
ered. Then statements a bout R and L belong to the theory of subs tantial success in nineteenth-century chemistry. It is always
mind, and one task of the brain sciences will be to explain what reasonable to consider alternative approaches, if they can be
24 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 25

devised, and this will remain true no matter what successes are then, a matter of adding to one's stpre of rules, or modi fying
achieved. The situation does not seem different in principle this system, as new data are processed. Question ( I iii) breaks
from what we find in other areas of empirical inq uiry. I will down into two parts: a "perception problem" and a "production
suggest directly that i n certain fundamental respects early ideas problem." The perception problem would be dealt with by
about !-language were misguided and should be replaced by a construction of a parser that incorporates the rules of the 1-
rather differen t conception, although one formu lated in the language along with other elements: a certain organization of
same general framework. The reasons, however, do not derive memory and access (perhaps a determinis tic pushdown structure
from any incoherence or flaw in the general approach but with buffer of a certain size; see Marcus, 1980), certain heuristics,
rather from empirical considerations of description and expla and so forth. A parser should not map expressions into their
nation. structures in the way that these are associated by the !-language.
For example, a parser should fail to do so in the case of so-called
"garden-path sentences" 1 2 or sentences that overload memory
2.4 THE SHIFT OF FOCUS FROM -LANGUAGE for left-to-right pass, it should mir:ror the difficul ties experienced
TO I-LANGUAGE wi th sen tences such as (8)-(14) of Chapter 1 and so forth. The
production problem is considerably more obscure; we will
On the Reasons for the Shift of Focus
return to that.
2.4. 1
The E-language that was the obj ect of s tudy in most of
In Chapter I , we saw that the study of generative grammar traditional or structuralist grammar or behavioral psychology
shifted the focus of attention from actual or potential behavior is now regarded as an epiphenomenon at best. Its s ta tus is
and the products of behavior to the system of knowledge that similar to that of other derivative objects, say, the set of rhyming
u nderlies the use and understanding of language, and more pairs, which is also determined by the !-language that constitutes
deeply, to the innate endowment that makes it possi ble for the system of knowledge attained. One might argue that the
humans to attain such knowledge. The shift in focus was from status of the E-languagc;:.is considerably more obscure than that
the study of E-language to the study of !-language, from the of the set of rhyming pairs, since the latter is determined in a
study of language regarded as an externalized object to the fairly definite way by the !-language whereas the bounds of
study of the system of knowledge of language attained and E-language can be set one way or a nother, depending on some
internal ly represented in the mind/brain. A generative grammar rather arbitrary decisions as to what it should include.
is not a set of statements about externalized objects constructed Summarizing, then, we have the following general picture.
in some manner. Rather, it purports to depict exactly what one The language facu'lty is a distinct system of the mind/brain,
knows when one k nows a language: that is, what has been with an initial state So com mon to the species ( to a very close
learned, as supplemented by innate principles. UG is a charac first approximation, apart from pathology, etc.) and apparently
terization of these innate, biologically determined principles, unique to it in essential respects.13 Given appropriate experience,
which constitute one component of the human mind-the this facul ty passes from the state So to some relatively stable
language faculty. steady state Ss, which then undergoes only peripheral modifica
W ith this shift of focus, we at once face the questions ( 1) of tion (say, acquiring new vocabulary items). The attained state
Chapter I . In the earliest work, the answer to ( I i) was taken to incorporates an !-language (it is the state of havi ng or knowing
be that knowledge of language is knowledge of a certain rule a particular 1-language). UG is the theory of 50; particular
system; the answer to ( I ii), that this knowledge arises from an grammars are theories of various !-languages. The !-languages
initial state So that converts experience to a "steady state" Ss, that can be attained with So fixed and experience varying are
which incorporates an 1-language. Acquisition of language is, the attainable human languages, where by " language" we now
26 . I KNOWLEDGE OF LANGUAGE
CONCEPTS OF LANGUAGE I 27

mean !-language. The steady state has two components that brain and their components (under the idealizations already
can be distinguished analytical ly, however, they may be merged discussed). UG and theories of !-languages, universal and par
and intertwined: a component that is specific to the language ticular grammars, are on a par with scientific theories in other
in question and the contribu tion of the initial state. The former domains; theories of -languages, if sensible at all, have some
constitutes what is "learned" -if this is the appropriate concept different and more obscure status because there is no corre
to employ in accounting for the transition from the initial to sponding real-world object. Linguistics, conceived as the study
the mature state of the language facul ty; it may well not be. 14 of !-language and So, becomes part of psychology, ultimately
The system of knowledge attained-the !-language biology. Linguistics will be incorporated within the natural
assigns a status to every relevant physical event, say, every sciences insofar as mechanisms are discovered that have the
sound wave. Some are sentences with a definite meaning (literal, properties revealed in these more abstract studies; indeed, one
figurative, or whatever). Some are intelligible with, perhaps, a would expect that these studies will be a necessary step toward
definite meaning, but are ill-formed in one way or another serious investigation of mechanisms.15 To put it differently,
("the child seems sleeping"; "to whom did you wonder what to E-language, however construed, is further removed from
give? " in some dialects; "who do you wonder to whom gave the mechanisms than !-language; at a higher order of abstraction.
book?" in all dialects). Some are well formed but unintel ligible . Correspondingly, the concept raises a host of new problems,
Some are assigned a phonetic represen tation but no more; they and it is not at all clear whether they are worth addressing or
are identified as possible sentences of some language, but not trying to solve, given the artificial nature of the construct and
mine. Some are mere noise. There are many possibilities. Dif its apparent uselessness for the theory of language.
ferent !-languages will assign status differently in each of these The shift of focus is also, arguably, a shift toward the
and other categories. The notion of E-language has no place in commonsense notion of language. This matter is less important
this picture. There is no issue of correctness with regard to than the move toward realism and also much less clear, because,
-languages, however characterized, because -languages are as noted, all of these approaches deviate from the commonsense
mere artifacts. We can define "-language" in one way or concept in certain respects. But it seems that when we speak of a
another or not at all, since the concept appears to play no role person as knowing a language, we do not mean that he or she
in the theory of language. knows an infinite set of sentences, or sound-meaning pairs
The shift of focus from E- to !-language, reviving and taken in extension, or a set of acts or behaviors; rather, what we
modi fying much older traditions, was very much in order. The mean is that the person knows what makes sound and meaning
technical concept of E-language is a dubious one in at least two relate to one another in a specific way, what makes them "hang
respects. In the first place, as j ust observed, languages in this together," a particular characterization of a function, perhaps.
sense are not real-world objects but are artificial , somewhat The person has "a notion of structure" and knows an !-language
arbi trary, and perhaps not very in teresting constructs. In con as characterized by the linguist's gram mar. When we say that it
trast, the steady state of knowledge attained and the initial state is a rule of English that objects follow verbs, as distinct from the
So are real elements of particular mind/brains, aspects of the r ule of Japanese that verbs fol low objects, we are not saying
physical world, where we understand mental states and repre that this is a rule of some set of sentences or behaviors, but
sentations to be physically encoded in some manner. The ! rather that it is a rule of a system of rules, English, an !-language.
language is abstracted directly as a componen t of the state The rules of the language are not rules of some infinite set of
attained. S tatements about !-language, about the steady state, formal objects or potential actions but are rules that form or
and about the initial state So are true or false statemen ts about constitute the language, like Articles of the Constitution or
something real and definite, abou t actual states of the mind/ rules of chess (not a set of moves, but a game, a particular rule
28 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 29

system). Of the various technical notions that have been devel derstanding. In the literature of generative gram mar, the term
oped in the study of language, the concept of !-language seems "language" has regularly been used for E-language in the sense .
closer to the commonsense notion than others. of a set of well-formed sentences, more or less along the l ines of
The shift of perspective from the technical concept E Bloomfield's definition of "language" as a " totality of utteran
language to the technical concept ! -language taken as the object ces . " The term "grammar" was then used with systematic
of inqu iry is therefore a shift toward realism in two respects: ambiguity, to refer to what we have here called "!-language"
toward the study of a real object rather than an artificial con and also to the linguist's theory of the I -language; the same was
struct, and toward the s tudy of what we really mean by "a true of the term UG, introduced later with the same systematic
language" or "knowledge of language" in informal usage ambiguity, referring to So and the theory of S0. Because the
(again, abstracting from sociopolitical and normative-teleo focus df'. a ttention was on !-language, E-language being a
logical factors). derivative and largely artificial construct, we find the paradox
Of these two considerations, the first is the clearer and i cal situation that in work devoted to language, the term "lan
more important. It is not to be expected that the concepts that guage" barely appers. In my 1965 book Aspects of the Theory
are appropriate for the description and understanding of some of Sy ntax, for examp le, there is no entry for "languag" in the
sys tem of the physical world (say, !-language and So) will index, but many entries under "grammar ," general ly referring
include the sometimes similar concepts of normal discourse, to !-language.
just as the physicist's concepts of energy or mass are not those of It would have been preferable to use the term "language"
ordinary usage. Furthermore, many questions arise about the in something closer to the i ntuitive sense of informal usage;
usage of the intuitive concepts that have no obvious relevance that is, to use the term "language" as a technical term in place
to the inquiry into the nature of the real objects, !-language and of "(generative) grammar" (in the sense of !-language) while
S0. Suppose, for exam ple, that a Martian with a quite different adopting some technical term (perhaps "E-language") for what
kind of m ind/brain were to produce and to understand sentences was called "language." The term "(generative) grammar" would
of English as we do, but as investigation would show, using then have naturally been used for the linguist's theory of the (I-)
quite different elements and rules-say, without words, the language, along the lines of the preceding discussion. Much
smal lest units being memorized phrases, and with a totally confusion might have been spared in this way. I suspect that the
different rule system and UG. Would we then say that the debate in past years over the alleged problems concerning the
Martian is speaking the same language? Within what limits concepts grammar and knowledge of gram mar may in part be
would we say this? Similar q uestions arise as to whether a n traced to these unfortunate terminological choices, which rein
artificial system i s exhibiting some form o f intelligence or forced inappropriate analogies to the formal sciences and gave
understanding. These may be reasonable questions concerning rise to the erroneous idea that the study of grammar poses new,
the intuitive concepts of language and the like in colloquial complex, and perhaps i ntractable philosophical issues com
usage, but it is not clear that they have much bearing on the pared with the study of E-language . 18
inquiry into the real-world objects, !-language and the initial The misleading choice of terms was, i n part, a historical
state S0 16 accident. The study of generative gram mar developed from the
The conceptual shift from E-language to !-language, from confluence of two intellectual traditions: traditional and struc
behavior and its products to the system of knowledge that turalist grammar, and the s tudy of formal systems. Although
enters into behavior, was in part obscured by accidents of there are importan t precursors, it was not until the mid-1950s
publishing history, and expository passages taken out of context that these intellectual currents truly merged, as ideas adapted
have given rise to occasional misunderstanding.n Some ques from the s tudy of formal systems came to be applied to the far
tionable terminological decisions also contributed to misun- more complex systems of natural language in something
30 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 31

approaching their actual richness, and in subsequent years, tion) a s w e please. But the F.-language is n o t "given. " What is
their actual variety, thus making i t possible, really for the first "given" to the child is some finite array of data, on the basis of
time, to give some substance to Humboldt's aphorism that which the child's mind (incorporating So) constructs an !
language involves " the infinite use of fini te means, " the "finite language that assigns a status to every expression, and that we
means" being those that constitute the 1-language. may think of as generating some -language under one or
But the study of formal languages was misleading in this another stipulated convention (or we may dispense with this
regard. When we study, say, the language of arithmetic, we may apparently superfluous step). What is given to the linguist are
take it to be a "given" abstract object: an infinite class of finite arrays of data from various speech communities, including
sentences in some given notation. Certain expressions in this much data not available to the language learner, on the basis of
notation are well-formed sentences, others are not. And of the which the lin guist will a ttempt to discover the nature of So and
wel l-formed sen tences, some express arithmetical truths, some of the particular !-languages attained. The account presented
do not. A "grammar" for such a system is simply some set of by Quine, Lewis, and others has the story backwards:
rules that specifies exactly the well-formed sentences. In this languages are not given, but are derivative, more remote from
case, there is no further question of the correct choice of gram data and from mechanisms than !-languages and the grammars
mar, and there is no truth or falsity to the matter of choosing tha t are theories of !-languages; the choice of -language there
among such grammars. Much the same is true of alternative fore raises a host of new and additional problems beyond those
axiomatizations, al though in this case we know that none of connected with grammar and I-language. Whether it is worth
them will capture exactly the truths. It is easy to see how one while addressing or attempting to solve these problems is not at
might take over from the study of formal languages the idea all clear, because the concept of -language, however construed,
that the "language" is somehow given as a set of sentences or appears to have no significance. The belief that -language is a
sentence-meaning pairs, while the grammar is some characteri fairly clear notion whereas !-language or grammar raises serious,
zation of this infinite set of objects, hence, it might be thought, perhaps intractable philosophical problems, is quite mistaken.
a construct that may be selected one way or another depending J ust the opposite is true. There are numerous problems con
on convenience or other extraneous concerns. The move is cerning the notions !-language and grammar, but not the ones
understandable, but misguided, and it has engendered much raised in these discussions.
pointless discussion and con troversy. . It should be noted that famOiar characterizations of "lan
Recall Qui ne's conclusion, cited above (p. 20), that i t is guage" as a code or a game poin t correctly toward !-language,
senseless to take one grammar rather than another to be "correct" not the artificial construct -language. A code is not a set of
if they are extensionally equivalent, and Lewis's doubts that representations but rather a specific system of rules that assigns
there is any way " to make objective sense of the assertion that a coded representations to message-representations. Two codes
grammar G is used by a population P whereas another grammar may be differen t, although extensionally identical in the
G', which generates the same language as G, is not. " It is quite message-code pairings that they provide. Similarly, a game is
true that for every -language, however we choose to define this not a set of moves but rather the rule system that underlies
notion, there are many grammars (i.e., many grammars, each of them. The .Saussurean concept of langue, although far too
which is a theory of a particular !-language that, under some narrow in conception, m ight be interpreted as appropriate in
conven tion tha t one has adopted, determines this -language). this respect. The same is true of Quine's definition of a language
But this is a matter of no consequence. In the case of some as a "complex of present dispositions to verbal behavior" insofar
formal system, say ari thmetic (presumably the model i n mind), as i t focuses on some internal state rather than -language,
we assume the class of well-formed formulas in some notation although i t is unacceptable for other reasons: Thus, two indi
to be "given," and we select the "grammar" ( the rules of forma- viduals who speak the same language may differ radically i n
32 I KNOWLEDGE OF LANGUAGE CONCEPTS O F LANGUAGE I 33

their dispositions to verbal behavior, and if dispositions are introduce technical terms for our purposes, giving them the
characterized in terms of probability of response under given meaning required for this discussion.
conditions, then it is impossible to identify languages in these Sometimes it has been suggested that knowledge of lan
terms; and again, the fundamental question of the use and guage should be understood on the analogy of knowledge of
understanding of new sentences is left without any explanation. arithmetic, arithmetic being taken to be a n abstract "Platonic"
Perhaps the clearest account is Jespersen's in terms of the entity that exists apart from any mental structures.19 It is not in
" notion of structure" that guides the speaker "in framing sen question here that there does exist what we have cal led an
tences of his own . . . , " these being "free expressions. " internalized language (described by what Thomas Bever calls
As we have seen, these ideas became the focus of a ttention "a psychogrammar") and that it is a problem of the natural
in the study of generative gram mar, although not without sciences to discover it. What is claimed is that apart from
controversy. Saussurean structuralism had placed Jespersen's particular !-languages, there is something else additional , what
observation about "free expressions" outside of the scope of the we might' cal l "P-languages" ( P-English, P-Japanese, etc.),
study of language structure, of Saussure's langue. Bloomfield existing in a Platonic heaven alongside of arithmetic and (per
( 1 933) held that when a speaker produces speech forms that he haps) set theory, and that a person who we say knows English
has not heard, "we say that he utters them on the analogy of may not, in fact, have complete knowledge of P-English, or,
similar forms which he has heard," a position later adopted by indeed, may not know it at all. Similarly, the best theory of the
Quine, C. F. Hockett, and the few others who even attempted to !-language, of what this person actually knows, might not be
deal with the problem. This idea is not wrong bu t rather is the best theory of what is selected on some grounds to be
vacuous until the concept of a nalogy is spelled out in a way that P-E nglish.2
explains why certain "analogies" are somehow valid whereas The a nalogy to arithmetic is, however, quite unpersuasive.
others are not, a task that requires a radically different approach In the case of arithmetic, there is at least a certain initial
to the whole question. Why, for example, are sentences (6) and p lausibility to a Platonistic view insofar as the truths of arith
(7) of Chapter 1 (p. 8) not u nderstood "on the analogy" of (4) metic are what they are, independent of any facts of individual
and (5 )? Why is sentence ( 1 4) not understood "on the analogy" psychology, and we seem to discover these truths somewhat in
of any of the earlier examples, in fact given no interpretation at the way that we discover facts about the physical world. In the
all? We can give substance to the proposal by explaining case of language, however, the corresponding position is wholly
"analogy" i n terms oU-language, a system of rules and princi without merit. There is no initial plausibility to the idea that
ples that assigns representations of form and meaning to lin apart from the truths of grammar concerning the !-language
guistic expressions, but no other way to do so has been proposed; . and the truths of UG concerning S0 there is an additional
and with this necessary revision i n the proposal, it becomes domain of fact about P-language, independent of any psycho
clear that "analogy" is simply an inappropriate concept in the logical states of individuals. Knowing everything about the
first place. mind/brain , a Platonist would argue, we still have no basis for
I have been freely using various commonsense notions determining the truths of arithmetic or set theory, but there is
such as "knowledge, " "rule-fo11owing, " and so forth in this not the slightest reason to suppose that there are truths of
account. Various questions have been raised about the legitimacy language that would still escape our grasp. Of course, one can
of this usage. I will put these questions off for now, returning to construct abstract entities at will, and we can decide to call
them in Chapter 4, but meanwhile continuing to use the terms. some of them "English" or "Japanese" and to define "linguis
I think the usage here is reasonably in accord with common tics" as the study of these abstract-objects, and thus not part of
usage, but nothing of great moment is at stake, and one could the natural sciences, which ;:tre concerned with such entities as
34 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 35

!-language and S0, with grammar and universal grammar in To clarify what is a t stake, suppose that two proposed
the sense of the earlier discussion. But there seems little point to gram mars G 1 and G2 differ in the choice of phonological
such moves. features postulated: G1 postulates the system F I> and G2, the
A somewhat similar conception is advanced by Soames system F 2 Suppose that G 1 and G2 are not distinguishable with
( 1 984). He dis ti nguishes between two disciplines, psychology respect to a data base consisting of what Soames stipulates to be
and linguistics, each defined by certain "Leading Questions," the "linguistically relevant" facts. Suppose that perceptual
which are different for the two disciplines. The study of ! experiments of the sort Sapir conducted in his classic work, or
language and S0, as described above, is part of psychology. other more sophisticated ones, yield results that can be explained
However, "If one's goal is to answer the Leading Ques tions of in terms of the features of F 1 but not F 2 Imagine further that
linguistics, one will a bs tract away from psycholinguistic data s tudies of aphasia and child language show that language
that are not constitutive of languages" (and similarly, from breakdown and growth can be explained along J akobsonian
neurophysiological data, etc.). The "Leading Questions" of lines in terms of F 1 but not F2, and that the choice of F 1 but not
'
linguistics include, for example, the questions, "In what ways F 2 provides a n accoun t for speech production and recogni tion,
are English and Italian alike?," "In what ways has English again along Jakobsonian lines. Soames agrees that there is a
changed" in the course of its history?, and so forth. The concepts field of inquiry, call i t "C(ogni tive)-linguistics, " which would
English and Italian are taken to be clear enough pretheoretically use this evidence to select G 1 over G2 as the theory of l anguage
to give these Leading Questions content, a highly dubious that is represen ted in the mind/brains of the members of this
assumption for reasons already discussed, and surely not one speech communi ty. B u t he proposes that there is another disci
made in actual linguistic research . Again, no q uestion is raised pline, call it "A(bstract)-linguistics," which dismisses this evi
here about the legi timacy of the investigation of !-language dence and regards G1 and G2 as equally well supported by
and S0; rather, the question is whether this s tudy fal ls under "relevant" empirical evidence; in fact, a practitioner of A
what we will decide to call "l i nguistics" and whether there is, linguis tics would choose G2 over G 1 if i t were "simpler" on
as Soames urges, "a theoretically sound, empirically significant some general grounds. There is no doubt that Sapir and J akob
conception of linguistics" that restricts i tself to a certain stipu son, among many others, would have followed the path of
lated domain of evidence, to facts that are "constitutive of C-linguistics in such a case, selecting G1 as the grammar and
language. " applying this conclusion to the study of " Leading Questions"
One might point out that the termi nological proposals concerning the historical evolu tion of l anguages, and so on.2 '
that Soames advances are a bit eccentric. It seems odd, to say the The burden ofproof clearly falls on those who believe that
least, to define "linguistics" so as to exclude many of its major alongside C-linguistics, the status of which is not here in ques
practitioners-for example, Roman Jakobson and Edward tion, there is sol'I\e poin t in developing the new discipline of
Sapir, who would surely not have agreed that what Soames A-linguistics, which not only differs from linguistics as it has
regards as extralinguistic data are irrelevan t to the q uestions of actually been practiced by major figures in the field but also is
linguistics as they u nderstood them, including the "Leading radically different from anything known in the sciences: It
Questions, " and who, in support of their analyses, adduced would be regarded as strange i ndeed to restrict biology or
evidence of a sort that Soames places outside of that "constitutive chemistry in some a priori fashion to ques tions and concepts
of language. " B u t pu tting aside terminology, the real question defined so as to delimit in advance the category of relevan t
that arises is whether there is any reason to establish a discipline evidence. In the sciences, at least, disciplines are regarded as
of "linguistics " that restricts i tself on a priori grounds to some conveniences, not as ways of cut ting nature a t i ts joints or as the
particular data and constructs a concept of "language" that can elaboration of certain fixed concepts; and their boundaries shift
be studied within this choice of relevant data. or disappear as knowledge and understanding advance.22 In
36 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 37

this respect, the study of language as understood in the discus sources apart from j udgments concerning the form and meaning
sion above is like chemistry, biology, solar physics, or the of expressions: perceptual experiments, the study of acquisition
theory of human vision. Whether the burden of proof faced by and deficit or of partially invented languages such as creoles, 2 s
advocates of A-linguistics can be borne, I will not speculate, or of literary usage or language change, neurology, biochemis
except to observe that even if it can, the fact would have no try, and so on. It was one of the many contributions of the late
consequences with regard to the legitimacy or character of the Roman Jakobson to have emphasized this fact, in principle,
en terprise we are discussing, as Soames makes clear. and in his own work in practice. As in the case of any inquiry
Note that the issue is not the legitimacy of abstraction. It is into some aspect of the physical world, there is no way of
perfect! y proper to develop the subject of rational mechanics, a del imiting the kinds of evidence that might, in principle, prove
branch of mathematics abstracted from physics that treats r.elevant. The study of language structure as currently practiced
planets as mass points obeying certain laws, or to develop should eventually disappear as a discipline as new types of
theories that consider aspects of 1-language in abstraction from evidence become available, remaining disti nct only insofar as
their physical realization or other properties; indeed, that is the its concern is a particular faculty of the mind, ultimately the
s tandard practice, as outlined earlier. But one is not misled brain: its initiat"s.tate and its various attainable mature states.
thereby into believing that the subject matter of rational To be sure, the j udgments of native speakers will always
mechanics is an entity in a Platonic heaven, and there is no provide relevant evidence for the study of language, j ust as
more reason to suppose that that is true in the study of lan perceptual judgments will always provide relevant evidence for
guage .2 3 the study of human vision , although one would hope that such
evidence wiJl eventual Jy lose its uniquely privileged s tatus. If a
theory of language failed to account for these j udgments, it
2.4.2 The Empirical Basis for tbe. Study of 1-language
would plainly be a failure; we might, in fact, conclude that i t is
In actual practice, linguistics as a discipline is characterized not a theory of language, but rather of something else. But we
by attention to certain kinds of evidence that are, for the moment, cannot know in advance j ust how informative various kinds of
readily accessible and informative: l argely, the j udgments of evidence will prove to be with regard to the language faculty
native speakers. Each s uch j udgment is, in fact, the result of an and its manifestations, and we should anticipate that a broader
experiment, one that is poorly designed but rich in the evidence range of evidence and deeper understanding will enable us to
it provides. In practice, we tend to operate on the assumption, identify in j ust what respects informan t judgments are useful
or pretense, that these informant judgments give us "direct or unreliable and why, and to compensate for the errors intro
evidence" as to the structure of the !-language, but, of course, duced under the tentative working assumption , which is indis
this is only a tentative and i nexact working hypothesis, and any pensable, for today, and does provide us with rich and significant
skilled practioner has at his or her disposal an armory of information.
techniques to help compensate for the errors introduced. In I t is important to bear in mind that the study of one
general, informant j udgments do not reflect the structure of the language may provide crucial evidence concerning the structure
l an guage directly; judgments of acceptability, for example, of some other language, if we continue to accept the plausible
may fail to provide direct evidence as to grammatical status assumption that the capacity to acquire language, the subject
because of the intrusion of numerous other factors. The same is matter of UG, is common across the species. This conclusion is
true of other judgments concerning form and meaning. These implicit in the research program outlined earlier. A study of
are, or should be, truisms.24 English is a study of the realization of the initial state So under
In principle, evidence concerning the character of the ! particular condi tions . Therefore, it embodies assumptions,
language and ini tial state could come from many different which should be made explicit, concerning S 0 But So is a
38 I KNO WLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 39

constant; therefore, Japanese must be an instantiation of the inquiry into mechanisms in the brain sciences. In principle,
same initial state under different conditions. Investigation of discoveries about the brain should influence the theory of mind,
Japanese might show that the assumptions concerning S0 and at the same time the abstract study of states of the language
derived from the study of English were incorrect; these assump faculty should formulate properties to be explained by the
tions might provide the wrong answers for Japanese, and after theory of the brain and is l ikely to be indispensable in the
correcting them on this basis we might be led to modify the search for mechanisms. To the extent that such connections
postulated grammar of English. Because evidence from Japanese can be established, the study of the mind-in particular, of
can evidently bear on the corrrectness of a theory of S0, it can 1-language-will be assimilated to the mainstream of the natural
have indirect-but very powerful-bearing on the choice of the sciences.
grammar that attempts to characterize the !-language attained So little is now known about the relevant aspects of the
by a speaker of English. This is standard practice in the study of brain that we can barely even speculate about what the connec
generative grammar. For this reason alone it is quite wrong to tions might be. We can, however, i magine how they might be
suppose that there are no grounds to choose among "exten established in principle, however remote the goal. Suppose
sionally equivalent grammars" for a "given language" (see pp. that the study of !-language establishes certain general principles
20, 30- 1 ) : One of these m ight, for example, require a theory of of binding theory that explain facts of the sort discussed in
S0 that is demonstrably inadequate for some other language. Chapter 1 . Then a -task of the brain sciences is to determine
On the highly relativistic assumptions of certain varieties what mechanisms are responsible for the fact that these princi
of descriptive linguistics that held that each language must be ples hold. Suppose hat we have. two grammars-two theories
studied in its own terms, this research program may seem to be of the state of knowledge attained by a particular person-and
senseless or illegitimate, although one should note that this suppose further that these theories are "extensionally equiva
point of view was, in part, an ideology that was not observed in lent" in the sense that they determine the same -language in
practice. If we are interested in discovering the real properties whatever sense we give to this derivative notion . It could in
of the i nitial state of the language faculty and of its particular principle turn out that one of these grammars incorporates
realizations as potential or actual !-languages, the ideology properties and principles that are readily explained in terms of
must be abandoned, and we must regard a theory of one language brain mechanisms whereas the other does not. Similarly, two
as subj ect to change on the basis of evidence concerning other theories of UG that are equivalent in that they specify exactly
languages ( mediated through a theory of UG), or evidence of the same set of attainable !-languages might be distinguishable
other sorts. in terms of properties of the brain. For example, one might
We observed that it is a task for the brain sciences to contain certain principles and possibilities of variation that
explain the properties and principles discovered in the study of can be readily explained in terms of brain mechanisms, and the
mind. More accurately, the interdependency of the brain sciences other not.
and the study of mind is reciprocal. The theory of mind aims to It is easy enough to imagine cases of this sort. Suppose that
determine the properties of the initial state S0and each attainable theory I contains the principles P 1, ,P n and theory II contains
.

state SL of the language faculty, and the brain sciences seek to the principles Qb . . . , Qm, and that the two theories are logically
discover the mechanisms of the brain that are the physical equivalent: The principles of each -can be deduced from the
realizations of these states. There is a common enterprise: to principles of the other so that any description of behavior or
discover the correct characterization of the language faculty in potential behavior in terms of one of these theories can be
its initial and attained states, to discover the tru th about the reformulated in terms of the other. It could be that the brain
language faculty. This enterprise is conducted at several levels: sciences would show that each Pi corresponds to some determi
an abstract characterization in the theory of mind, and an nate complex of neural mechanisms, whereas there is no such
40 I KNO WLEDGE OF LANGUAGE
CONCEPTS O F LANGUAGE I 41

account of the Q/s; some brain injury, for example, might the study of rules and representations. This shift also led at
selectively modify the P/s but not the Q/s. In such a case, facts once to a recasting of many of the traditional questions of
about the brain would select among theories of the mind that language study. Many new a nd challenging problems arose,
might be empirically indistinguishable in other terms. Although while a number of familiar problems dissolved when viewed
results of this sort are remote in the current state of u nderstand from this perspective.
ing, they are possible. The relation of brain and mind, so Consider the study of sound structure, the primary focus of
conceived, is a problem of the natural sciences. attention in structural and descriptive linguistics. Taking E
language as the topic of inquiry, the problem is to discover the
2.4.3 Some Consequences of the Shift of Focus elements i nto which the stream of speech is subdivided and
their properties and structural arrangements: phonemes and
To summarize, we may think of a person's knowledge of a features, regarded as segments of an acoustic wave form or of a
particular language as a state of the mind, realized in some series of articulatory motions. Much of phonological theory
arrangement of physical mechanisms. We abstract the ! consisted of analytic procedures for accomplishing this task.
language as "what is known" by a person in this state of Focusing on the I-language, however, the problem is a rather
knowledge. This finite system, the I-language, is what the different one: to find the mental representations that underlie
linguist's generative grammar a ttempts to characterize. If I say the production an<Lperceptioh of speech and the rules that
that this system has such-and-such properties, what I say is true relate these representations to the physical events of speech.
or false. I am, in short, proposing a theoretical account of the The problem is to find the best theory to account for a wide
properties of certain mechan isms, an account presented at a variety of facts, and we do not expect that analytic procedures
level of abstraction at which we believe that significant proper exist to accomplish this task, j ut as there are no such procedures
ties of these mechanisms can be expressed and principles in other fields.
governing these mechanisms and their functions elucidated. Consider, for example, the words listed below, where
The study is in some ways similar to what Gunther Stem has column I is the conventional orthography, column II appears
cal led "cerebral hermeneu tics, " referring to the abstract i nves
tigation of the ways in which the visual system constructs and I II III
interprets visual experience (Stent, 1 98 1 ). Similarly, UG is the
bet bet bet
study of one aspect of biological endowment, analogous to the
bent bent bet
study of the innate principles that determine that we will have a
bend bend bend
human rather than an insect visual system. The technical con
knot nat nat
cept "knowl edge of !-language" is a reasonably close approxi
mation to what is i nformally called "knowledge of language, " nod nad nAd
abstracting from several aspects of the commonsense notion as write rayt rayt
discussed earlier, al though this consideration is a secondary ride rayd rAyd
one for reasons already mentioned. writer rayt+r ray Dr
The shift of point of view to a mentalist interpretation of rider rayd+r rAy Dr
the study of language was, as noted earlier, one factor in the
development of the contemporary cogni tive sciences, and con to be the correct phonological representation, and column III,
stituted a step toward the incorporation of the study of language the approximate phonetic representations in one dialect of
within the natural sciences, because it helps pave the way to an English, taking [a] to be a short vowel and [A] a corresponding
inquiry into the mechanisms with the properties exhibited in long vowel (their exact phonetic character is irrelevant here),
 42 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 43

[eJ a nasalized counterpart to [e ], and D a tongue flap rather like tions (II). The latter representations are not derived from the
a tril led [ r]. speech sounds by analytic procedures of segmen tation, classifi
We may assume that the phonetic representations of column cation, extraction of physical features, and so forth, but are
III correspond to actual speech events by universal principles of established and justified as part of the best theory for accounting
interpretation that essentially preserve linearity; that is, the ultimately for the general relation between sound and meaning
sequence of phonetic symbols corresponds to the sequence of of the !-language. Further syntactic and semantic rules apply to
sounds (the matter is not this simple, as is well known). The the representations of (II) in the expressions in which these
phonological representations of the second column, not the words appear. The !-language, i ncorporating the rules that
phonetic representations of the third, correspond to the way form the representations (II) and the rules that relate them to
that we intuitively "hear" these words. Although phonetic (Ill), is acquired by the child by applying the principles i ncor
analysis reveals that bet and bent differ only in nasalization of porated in the i nitial state So to the presented facts; the pro blem
the medial vowel, and that each has three phonetic segments as for the grammarian is to discover these principles and show
distinct from the four-segment word bend, this does not corre how they lead to the choice of the representations (II) ( assuming
spond to the intuitive perception; we hear knot and nod as these to be correct). The failure of taxonomic procedures is of
differing only i n one feature, voicing of the final consonant, no significance, because there is no reason to believe that such
not in both the vowel a nd the consonant (as, e.g., knot versus procedures play any role in language acquisition or have any
Ned). The representations of writer and rider that we intuitively standing as part of U(;.
perceive and that clearly relate to lexical and syntactic structure As these very simple examples illustrate, even at the level of
are as indicated in the second column (with + standing for the sound structure, mental representations may be relatively
break between the lexical item and the agentive affix), not the abstract-i.e., not related in a simple way to actual specimens
third, although the latter expresses the phonetic fact that the of linguistic behavior (in fact, this is even true of the phonetic
words differ only in vowel quality. Examples such as these representations, as a closer analysis would show). As we move
posed difficult problems for an approach to phonology that to other levels of inquiry into the 1-language, we find increasing
sought to determine phonological u nits by analytic procedures evidence that men tal representations are abstract in this sense.
applying to actual speech events. The question is the status of The systems of rules and principles that form and modify them
the representations of column II, which were always recognized are fairly simple and natural, although they interact to yield
to be "correct" in some sense although their elements do not structures of considerable complexity and to determine their
correspond point-by-point to the actual sounds of speech, the properties in quite a precise fashion. In short, the language
subparts of the actual specimens of E-language. faculty appears to be, at i ts core, a computational system that is
Shifting the focus of attention to 1-language, the problems rich and narrowly constrained in structure and rigid in its
quickly dissolve. The representations of column II are essentially essential operations, nothing at all like a complex of dispositions
the mental representations o f the lexicon, which enter into the or a system of habits and analogies. This conclusion seems
syntax and semantics. The phonetic representations of column reasonably well established and has been given 'considerable
III derive from these by s traightforward rules, most of them substance; there is no known a lternative that even begins to
quite general: Vowels assume a particular quality before voiced deal with the actual facts of language, and empirically mean
and unvoiced consonants and become nasalized before nasal ingful debate takes place largely within the framework of these
consonants, the n asal consonant drops before an unvoiced assumptions.
dental, and (in this dialect) the dental stops merge as [D] medially Nevertheless, it should be observed that the conclusion is
under this stress contour. Applying these rules, we derive the i n many ways a rather surprising one. One might not have
phonetic forms (III) from the lexical-phonological representa- expected that a complex biological system such as the language
44 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 45

faculty would have evolved in this fashion, and if indeed it has, first sentence refers. Or, consider such widely discussed examples
that discovery is of no small significance.26 as "everyone who owns a donkey beats it, " problematic because
The scope of the shift to a mentalist or conceptualist the pronoun it does not appear to be formally within the scope
interpretation, to internalized rather than externalized language, of the quantified noun phrase a don key that binds it. One
is broader than has been sometimes appreciated. Quite explicitly, might try to apprqach the analysis of such sentences by con
it included the study of syntax, phonology, and morphology. I structing a representation with the property that for every pair
think it also includes much of what is m isleadingly cal led "the (man, donkey), if own holds of the pair, then so does beat. Then
seman tics of natural language" -I say "misleadingly" because we should say the same abou t "everyone who has a chance
I think tha t m uch of this work is not seman tics at all, if by wastes it," without, however, committing ourselves to the belief
"semantics" we mean the study of the relation between language that among the things in the world there are chances. Even if we
and the world-in particular, the study of truth and reference. restrict ourselves to the context "there are . . . ," we can hardly
Rather, this work deals with certain postulated levels of mental assume that there are entities in the world, or in the world as we
representation, including representations of syntactic and lexical believe it to be, that correspond to the terms that appear ( "there
form and others called "models" or "pictures" or "discourse are looks that inj ure and others that charm, ' ' ' 'there are chances
representations" or "situations, " or the like. But the relation of that are too risky to take, " "there are opportunities that should
these latter systems to the world of objects with properties and not be passed up, " etc. ).
relations, or to the world as it is believed to be, is often intricate One can think of many still more extreme examples.
and remote, far more so than one m ight be led to believe on the Although there has ben much concern over the status of fic
basis of simple examples. The relation cannot, for example, be tional and abstract objects, the problem, in fact, cuts far deeper.
described as "incorporation" or element-by-element association. One can speak of "reference" or "coreference" with some intel
Consider, for example, the principles of pronominal refer ligibility if one postulates a domain of mental objects associated
ence, which have been central to these quasisemantic investiga with formal enti ties of language by a relation with many of the
tions. If I say "John thinks that he is intelligent," he may refer properties of reference, but all of this is internal to the theory of
to John, but not if I say "he thinks that John is intelligent. "27 mental representations; it is a form of syntax. There seems no
We can account for such facts by a theory of the structural obvious sense in populating the extra-mental world with cor
configurations in which a pronoun can acquire its "reference" responding entities, nor any empirical consequence or gain in
from an associated name that binds it. The same principles, explanatory force in doing so. Insofar as this is true, the study of
however, apply to such sentences as "the average man thinks the relation of syntactic structures to models, "pictures," and
that he is intelligent, " "he thinks that the average man is the l ike, should be regarded as pure syntax, the study of various
intelligent" (or "John Doe thinks that he is intelligent, " where men tal representations, to be supplemented by a theory of the
"John Doe" is introduced as a designation for the average relation these mental objects bear to the world or to the world as
man). But no one assumes that there is an entity, the average it is conceived or believed to be. Postulation of such mental
man (or John Doe), to which the pronoun is permitted to refer representations is not innocuous but must be justified by
in one bu t not the other case. If I say "John took a look at him, empirical argument, just as in the case of phonological or other
but it was too brief to permit a positive identification, " it can syn tactic representations. Thus, the shift toward a computa
refer to the look that John took; but the near synonym "John tional theory of mind encompasses a substantial part of what
looked at him" cannot be extended in this way with the same has been called "semantics" as well, a conclusion that is only
interpretation, al though no one believes that there are looks fortified if we consic;ler more avowedly "conceptualist" approaches
that a person can take, to one of which the pronoun it in the to these topics.
 46 I KNOWLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 47

To proceed, we are now concerned with !-language and NOTES

the initial state of the l anguage facul ty, with the l inguist's
grammars and UG. As a tentative empirical hypothesis, we 1 . These o bservations, genera l l y considered truisms, are rejected
might take the !-language to be a rule system of some sort, a by Katz ( 1 98 1 , pp. 79-80) o n the gro u n ds that to recognize the fact that
specific real ization of the options permitted by U G, fixed by the concepts language and dialect of co l lo q u ia l usage i nvo lve a
presented experience. The rule system assigns to each expression sociopol it ical dimension wo u l d be " like cla i m i ng that the co ncept of
a structure, which we may take to be a set of representations, n u m ber is no t a concept o f mathematics but a socio po l itical o ne."
There is no reaso n to accept this curious conclusion.
one on each linguistic level, where a l inguistic level is a partic
2. However, there were exceptions, for example, the theory of
ular system of mental representation. Th is structure must pro
"overal l pa tterns , " o f which each English dialect was held to be a
vide whatever information about an expression is available to
s u bsystem. See Trager and Smith ( 1 95 1 ). Note that the q ues tion o f
the person who knows the language, insofar as this i nformation "variable rules," as discussed by some socio linguists, i s no t relevan t
derives from the language faculty; its representations must here.
specify j ust what the language faculty contributes to determining 3. We put aside here j ust wha t this term wo u l d mean in B loo m
how the expression is produced, used, and understood. fieldian or any other variety of "behavioris t" l inguis tics. Purs uing
A linguistic level is a system consisting of a set of minimal s uch an approach, one wo u l d have to explain j us t w ha t it means to
elements ( primes), an operation of concatenation that forms say that people speak the very sam e language al though they do no t
strings of pri mes, as much mathematical apparatus as is neces tend to say the same things in given circumsta n ces. The same q uestion
sary to construct appropriate formal objects from these elements, arises if la nguage is defined as a "complex of present dispositions to
the relevant relations that hold of these elements, and a class of verbal behavior' ' (Qui ne, 1 960), as do o ther pro blems that seem insol
u ble if the technical co n-structed co ncept "language" is to be a useful
designated formal objects (markers) that are assigned to expres
term for the investiga tio n of language, or to have any relation to what
sions as their representations on this level. The rule system
we cal l " language." On this matter, see Choms k y ( l 975b, pp. 1 92- 1 95).
expresses the relations among the various levels in the language
4. One might also note some unintentiona l l y comica l o bjections,
i n question and determines the elements and properties of each such as the charge by O x ford professor of linguis tics Roy Harris
level. At the level of phrase structure, for example, the primes ( 1 983) that the s ta n dard ideal iza tion ( which he ascribes to Saussure
are the minimal elements that enter i nto syntactic description Bloomfield-Chomsky) reflects "a fascist concept o f language if ever
(jo h n, run, past-tense, N, V, S, etc.) , the basic relation is is-a there was o ne, " because i t takes the "ideal" speech co m m u n i ty to be
(foh n is an N , john ran is a n S, etc.) , and the phrase-markers " to ta l l y homogeneo us."
will be certain formal objects constructed out of primes that 5. For so m e discussio n, see Chosky ( 1 964) and Postal ( 1 964).
express completely the relation is-a. The phrase-markeT for the For co mpariso n of transformational generative gra mmar with Harris's
string joh n ran will indicate that the full string is an S (sentence), early theory of transfo rmations, regarded as an a na l y tic procedure
that]ohn is an N ( noun) and an N P (noun phrase), and that ran applying beyon d the sentence level of "structural grammar," see the

is a V (verb) and a VP (verb phrase); examples appear below. i ntroductio n to Cho ms k y ( 1 975a).
6. Lewis ( 1 975). Lewis provides one of the clearest presentations
The theory of linguistic structure ( UG ) will have the task
of an "extensio na l " a pproach to lang uage and also a critiq ue o f
of specifying these concepts precisely.28 The theory must provide
studies o f "internalized language" in t h e sense described belo w. For
grammars for the !-languages that can, in principle, be attained critical discussion, see Cho msky ( 1 980b).
by a human mind/brain, given appropriate experience,29 and it 7. Editorial co mments i n Joos ( 1 957); Whi tney ( 1 872); Sapir
must furthermore. be so constrained that just the right !-language ( 1 92 1 ). Whi tney, who exerted a major i nfl uence on Sauss ure a n d
is determined, given the k ind of evidence that suffices for lan American linguistics, was criticizing Steinthal's Humboldtian approach,
guage acquisition. We turn next to these questions. which I believe, fal ls na tura l l y i n to the earlier tradi tion referred to
48 I KNO WLEDGE OF LANGUAGE CONCEPTS OF LANGUAGE I 49

above. Humboldt, who is widely regarded (e.g., by Bloomfield) as an 1 23- 1 28. As for the publ ishing history, the earliest publications on
extreme relativist, in fact held that "all languages with regard to their generative grammar were presented in a framework suggested by
gram mar are very similar, if they are investigated not superficially, certain topics in automata theory (e.g., my Syntactic Structures,
b u t deeply i n their i nner nature." See Chomsky ( 1 966), p. 90, and 1 957 -actually course notes for an u ndergraduate course at MIT and
references cited, for further discussion. hence presented from a poi n t of view related to in terests of these
8. This question, however, was surely n o t w h a t Whi tney h a d i n st udents). Specifically linguistic work, such as Chomsky ( l 975a), was
mind. not publishable at the ti me. In the latter, considerations of weak
9. Jespersen ( 1 924). On Jespersen's notions as compared to generative capacity (i.e., characterizabi lity of E-languages), fini te
those of contem porary genera tive grammar, see Reynolds ( 1 97 1 ); automata and the like were compktely absent, and emphasis was on
Chomsky ( 1 977), Chapter I . 1-Ianguage, a l though the term was not used.
I 0. One might argue that the systems we are considering consti 1 8. For further discussion of this matter, see Chomsky ( 1 980b).
t u te only one element of the faculty of language, understood more 19. See Katz ( 1 98 1 ) a n d Bever ( 1 983).
broadly to encompass other capacities involved in the use and under 20. This would follow if the evi dence stipu lated to be releva n t
standing of language, for example, what is sometimes called "com t o i dent ifying a cer tain P l a tonic l anguage a s P-English is distinct
municative com petence, " or parts of the human conceptual system from the evidence that bears on the theory of the !-language actually
that are speci fical ly related to language. See Chomsky ( 1 980b). I will represented in the mi nd/brains of speakers of English, or if some
p u t such questions aside here, con ti nui ng to use the term "langu age novel canons are adopted for i nterpreting evidence. By a simi lar

faculty" i n the narrower sense of the previous discussion. procedure, we could establish " P la tonistic biology , " concerned, for
I I. For a rela ted but somewhat di fferent way of viewing t hese example, with what Katz ca1ls the "essential property" of a heart ( that
questions, see Higginbotham ( 1 983b). it is a pump) and t h us abstracting from the physical laws that make i t
1 2. Those that tend to yield a false parse, such as Thomas beat ( a nonessential property). We might then find that the best
Bever's example " the horse raced past the barn fel l," where the first biological theory iS distinct from the best t heory of Platonistic biology
six words are general l y taken to cons t i t u te a full clause, leavi n g no j us t as the best ( u l timately, biological) theory of !-language might be
i nterpretation for the final word, a lthough on reflection i t is clear that distinct from the best theory of Platonistic language ( however i t i s
the expression is a well-formed sentence stating that a certain horse specified; for Katz, b y analysis of "our concept of t h e abstract object
fel l , namely, the one that was raced past the barn. natural language") .
1 3. Obviously, the questions of innateness and species-specificity 21. For some recent discussion of the matter i n connection w i t h
are distinct. It has been al leged that I arid others have taken "in nate' ' historical linguistics, see Lightfoot ( 1 979).
and "species-specific" to be"synonyms" (Cartmill, 1 984). I am unaware 22. Katz i nsists t ha t disciplines such as chemistry, biology, and
of any examples of such confusion, a lthough there are a number of so forth have inherent, conceptually determined boundaries. Indeed,
articles refu ting it. he regards the claim as uncontroversial, the alternative being a form
1 4. See Chomsky ( 1 980b), pp. 1 34- 1 39. of "ni hilism" that "would turn the spectrum of well-focused academic
15. On this matter, see Marr ( 1 982). Note that the q uestion of disciplines i n to chaos" (op. cit . ) .
the legi timacy or sense of a realist interpretation of science in general 23. Arguments that have been o ffered t o the con trary seem t o
is not at issue here; rather, nothing new in principle seems to arise i n m e question-begging o r otherwise flawed. Thus, Katz argues agai nst
t h e case of t h e study o f !-language and i ts origins. If one wants to Hilary P u tnam that if what we call "cats" were discovered to be robots
consider the question of realism, psychology and li nguistics seem controlled from outer space, then they would not be cats, because the
poor choices; the question should be raised with regard to the more meaning of "cat" i n the Platonic entity P-English is "feline animal";
advanced sciences, where there is much better hope of gai ning i nsight this would remain true even if it were determined that in the !-language
i n to the matter. See Chomsky ( I 980b) for further discussion. of each speaker of English, "cat" is understood in accordance w i t h
1 6. For some commentary on the general issue, see Enc ( 1 983). Putnam's analysis, which takes cats t o be of t h e same natural kind ( a
I 7. On some misunderstandings, whiCh are repeated i n su bse concept of science) a s particular exemplars. The argument goes
quent work that I will not discuss here, see Chomsky ( l 980b), pp. through, trivially, with regard to P-English as Katz stipulates i ts
50 I KNOWLEDGE OF LANGUAGE

properties. But Pu tnam was proposing a theory concerning human

languages and conceptual systems, concerning English, not P-English
as Katz defines it, and Katz offers no reason to believe that his Platonic
object meri ts the name "English" any more than an equally legitimate
abstract object that would i ncorporate Pu tnam's assumptions.
Throughou t, the arguments are of this sort. Katz also presents an
account of the history of generative grammar and of documents he
cites that is seriously inaccurate, as is often evident even on in ternal
3 Facing Plato's Problem
grounds. See also Chomsky ( 1 98 1 ), pp. 3 1 4- 3 1 5 .
24. For discussion of some common misu nders tandings about
these and related matters, see Newmeyer ( 1 983).
25. O n the relevance of this material, see Bickerton ( I 984) and
references ci ted, and discussion in the same issue of the journal.
3. 1 . A MODEL O F EXPLANATION
26. For some discussion, see Chomsky ( 1 980b, 198 1 ); and
Chomsky, H u y bregts, and van Riemsdij k ( I 982).
27. The matter is more complex. See Evans ( I 980) and Higgin
With the shift of focus from -language to !-language,
botham ( 1 983a). B u t we can put aside the required sharpening of from the study of behavior and its products to the study of
these notions here. systems of mental representation and computation, a number
28. For an early effort, see Chomsky ( 1 975a), dating from of questions arise. Some relate to the legitimacy or the proper
1 955-56. bounds of this move; these I will put aside until Chapter 4.
29. A stronger req uirement would be that UG specify exactly Others arise internally to the research program that develops
the !-languages attainable under normal conditions. It is not obvious, naturally from this shift of focus. These are substantive questions
however, that UG meets this conditions. The attainable languages as to how the gener:l ideas should be fleshed out. Let us now
are those that fa l l in the i n tersection of those determined by UG and
turn to these questions.
the humanly learnable systems, and conditions on Iearnabi lity might
The central task is to find the basic elements of !-lan
excl ude certain grammars perm i t ted by UG. S i m i lar remarks hold
guage-henceforth, language. One must, in the first place,
w i th regard to parsi ng. For background on these matters, see Wexler
and Cul icover ( 1 980) and Berwick and Weinberg ( I 984).
show that the devices provided by the theory of UG are adequate
to the descriptive task at hand-i.e., that they are rich enough to
account for the attested variety of languages and, indeed, for
their possible variety. A second task is to show that these devices
are meager enough so that very few languages are made available
to the language-learner, given data that, in fact, suffice for
language acquisition. Unless this condition is satisfied by the
theory of UG, it will be impossible to account for the fact that
languages are learned. The transition from the initial state to
the steady state takes place in a determinate fashion, with no
conscious attention or choice. The transition is essentially
uniform for individuals in a given speech community despite
diverse experience. The state attained is highly articulated and
very rich, providing a specific interpretation for a vast array of
sentences lacking close models in our experience. These condi
tions of our variant of Plato's problem set the second task that