Noam Chomsky, On the Nature of Language, 1976

7/27/2019 Noam Chomsky, On the Nature of Language, 1976

1/12

Part 11. Formulating the Target (11)ON T H E N A T UR E O F L A N G U A G E

Noam C homskyDepartment of LinguisticsMassachusetts institute of Technolog-vCam bridge. M assachusetts 02139

Imagine a creature so magnificently endowed as to be in a position to regardhumans rather in the way that we regard fruit flies. Faced with the problem ofdetermining the nature of language, this creature m ight exp loit a variety of means. Itmight undertake the study of genetic m utations or intrusive experim entation ofvarious sorts. Or, it might investigate the variation in know ledge of language attainedas experience available is experimentally controlled. U nder the latter app roa ch , thegenetically determined human language faculty, which specifies the nature oflanguage, is considered as a function ass ociating kno wled ge attai ned with experie nceavailable. Taking the grammar of a language to be a representation of theknowledge attained, or at least a fundamental component of such knowledge, thelanguage faculty may be regarded as a fixed function, characteristic of the species,one component of the human mind, a function which maps experience into gra mm ar.The method of concomitant variation is a natural way to study such a functiondirectly.We may construe a general linguistic theory or a theory of universal gram maras a hypothesis concerning this acquisition function. A prior i , on e might imagine arange of possibilities. Th us it m ight be th at generalized learning strategies involvingprocedures of induction, analogy, and generalization in some dimensional systemexhaust the con tents of the language faculty. O r, the course of language acquisitionmight reflect cognitive growth in other domains, say, constructions of sensory-m otor intelligence. It has been proposed in the past tha t tax ono m ic proced ures ofiterated segmentation an d classification suffice t o dete rmine the gra m m ar of particu-lar languages and thus express the nature of language.Alternatively, one might argue that languages have a partially determinatestructure as a ma tter of biological necessity, much a s the gene ral cha racte r of bodilyorgans is fixed for the species. Such structure must therefore be spelled o u t in thetheory of the language faculty. T he theory of distinctive features is perhaps the mostfamilia r case. It has been propo sed th at a certain set of features is available inprinciple for phonetic representation; each language must make its selection fromam on g these, observing certain imp licational principles governing su ch systems. S eeReference I and much subsequent w ork. Th e theory of distinctive features ca n beregarded as a hypothesis concerning the acquisition function for language. In myview, work of the past years has provided con siderable sup po rt fo r a conceptio n ofthe language faculty along these lines, as a schematism t ha t narrowly constrains theform of the grammar attained, rather than a system of generalized inductive andtaxonomic procedures for acquiring knowledge. Thu s it seens to me not unreason-able to approac h the study of language as we would the study of som e organ of thebody.Suppose that a hypothesis is advanced asserting that grammars must have theproperty P, that is, having the property P esults fr om th e structure of the languagefaculty itself rather than from experience, though relevant experience may berequired to trigger the prope r growth a nd functioning of a system with the prop ertyP. n principle, the hyp othesis m ight be tested in vari ous ways, say, by designing a n

46

Annals of the New York Academy of Science, Volume 280, Origins and

Evolution of Language and Speech, pages 4657, October 1976


2/12

C h o m s k y : On t h e N a t u r e of L a n g u a g e 47environment neutral with regard to P, and d etermining whether gram ma rs invariablyhave the property P rather than not-P, no less com patible with presented experience,in such an environment. Or, one might design an environment with experienceinconsistent with P to verify the prediction that the language violating P is notlearned under normal conditions of time and access to data, or used with normalfacility, and so on.Of course, no such course of inquiry is available to humans invest igat ing hu ma nlanguage. Par t of the intellectual fascination of the study of hu ma n language derivesfrom the fact that it is often necessary to devise intricate and complex argu me nts tosuppor t , or reject, the hypothesis that gram ma rs must meet some condit ion P a s ama tter of biological necessity. T hi s unavo idable continge ncy of inq uiry in no waydeprives the study of language of its empirical character, a lthough it do es bear o n theforce and persuasiveness of part icular empirical t h e o ~ e s .

Given the actua l conditions of inquiry, a natural line of arg um en t is the following.Suppose we find that a particular language has the proprty P, that is, speakersjudgments and o ther behavior conform to P where clear an d reliable, construc tionsviolating P are rejected, and so on. Suppose, furthermore, that P is sufficientlyabstract and evidence bearing on it sufficiently sparse and contrived so that it isimplausible to suppose that all speakers, or perhaps any speakers, might have beentrained or taught t o observe P or might have constructed gra mm ars sat isfying P byinduction from experience. The n it is plausible to postulate that P i s a property of thelanguage faculty, that languages conform to P a s a m att er of biological necessity. W ear e assum ing, then, that the environment-which we ca nn ot control-is neutral withrespect to P, nd we seek properties for which this assumption is plausible.A familiar and very simple example of this kind of reasoning concerns theproperty of structuredependence of linguistic rules. In this case, the argument,which I will no t review (see References 2, 3) seems to me qu ite compelling, an d thereis a theory of linguistic rules with o th er virtues as well that a cco unts for the facts.Th us the argument is that the s tructu rede pen den t property of rules is not somethingthat has to be learned, but is a precondition of learning. The language facultyrequires that rules be formulated in these terms, th ou gh fo r a n organism differentlydesigned, a system violating this principle could serve the fu nctions of language noless well.Note tha t in su ch cases as these we may plausibly postulate th at P is a property ofuniversal grammar on the basis of investigation of a single language. There is n oparadox here. The argument rests on the al leged fact that something is knownwithout relevant experience so that knowledge must be at t r ibuted to the languagefaculty itself, a faculty com mo n to the species. D ee p analysis of a single language m ayprovide the most effective means for discovering nontrivial properties of universalgrammar .Of course, the argum ent is nondemonstrat ive an d is therefore o pen to refutationby broader inquiry into the same language or other languages. Consider thefol lowing case, pat terned o n the famil iar argu me nt concerning struc tured epe nde nceof rules.

We find in English two kinds of relative classes, restrictive and nonrestrictive. In( I ) , the italicized relative clause is restrictive; in (2), nonrestrictive:( I ) People who g o to M IT like math(2) J o h n , who goes to MI T, likes math

These tw o types of relatives have quit e different syntactic an d sema ntic properties. Inparticular, restrictives c an stack, giving suc h sentences as (3), whereas nonrestric-tives cannot, so that (4 ) is ruled out:


3/12

48 A n n a l s N e w York A c a d e m y of Sciences(3 ) People who go to M IT w ho like m ath will get job s(4) Jo hn , who goes to M IT, wh o likes ma th, will get a jo b

I t seems implausible that all people capable of these distinctions have sufficientconfirmatory evidence to establish the principle governing stacking of relatives.Following the line of argument just outlined, it is therefore reasonable to postulatethat as a property of universal grammar, biologically necessary, nonrestrictiverelatives ca nn ot stac k.Reasonable, but apparently incorrect. It seems that in Korean and Japanese, forexample, such sentences as (4) are quite acceptable. Therefore the hypo thesis, thoughreasonable, is untenable.Bearing in mind that experience to ground the distinction between (3) and (4 )does not seem generally available, we might propose a more abstract hypothesis.Perhaps some other linguistic property, which must itself be learned, determineswhether no nrestrictive s can sta ck by virtue of a general principle of universalgramm ar. We observe, for example, tha t Korean and J apa nese d o not distinguishtwo types of relatives, restrictive and nonrestrictive, and that speakers of theselanguages often find difficulty in d is ce rn in g t h e difference of s e n ~ e . ~ . ~e mighttherefore postulate tha t in languages that distinguish two types of relatives, restrictiveand nonrestrictive. only restrictives can stack. We are postulating. then, that thelanguage learner must dete rmin e from experience whether the language to which heis exposed distinguishes restrictives and nonrestrictives. Given this information, hecan appeal to a general principle of universal grammar to establish which relativescan stack, without a ny experience with stacked relatives. He need no t be taught tha t(4) is excluded and (3) grammatical. Rather, given that English regularly distin-guishes ( I ) from (2). it follows by a general principle that it excludes (4). T heconclusion is reasonable, since otherw ise we would be hard put to exp lain why p eopled o not generalize fr om the stacked relative (3) to the stacked nonrestrictive (4). I notein passing tha t o th er learned prop ertie s of the language might be suggested as the keyto bringing a general principle to bear, in this case.Our new, more abstract hypothesis is no less empirical than the earlier, simplerone. We know just wh at evidence would refute o r confirm it. Tho ug h the direct test ofcontrolling the environment is excluded, we may nevertheless search for relevantevidence a m on g the existing languages or in observation of language acquisition, an dperhaps in other ways.Pursuing a similar line of argument, we may seek universal conditions in anydomain of grammar. Suppose, say, that some principle of rule ordering serves toexplain som e phone tic prope rties of English. The n-particu larly if the evidence is"exotic"-we may reason ably prop ose tha t the principle belong s to universalgram mar, a nd then ask whether this conclusion is suppo rted by furt he r inquiry intoEnglish and other languages. We will not be surprised to discover, again, that alearned property of language determines when and how the universal principle isapplied.Or, consider the dom ain of interaction of syntax an d seman tics. It seems to be aproperty of certain types of co m m on nouns t ha t they can be used with either abstractor concrete reference. Consider the sentences ( 5 ) and (6):

( 5 ) Joh n w rote a book(6) The book weighs five pounds

In ( 5 ) . the reference of book is abstract. T hus, Jo hn may have written the book in hishead, comm itting nothing to paper, but ( 5 ) would still be app ropr iate. Or, if ther e ar etwo copies of the boo k before me, 1 may say of each, "Jo hn wrote this book." but I d o


4/12

C h o m s k y : On the N a t u r e of Language 49not conclude that John wrote two books. In contrast, the reference of book in (6) isconcrete. A specific material object weighs five pounds. I f a hardcover and paperbackcopy of the book t hat J oh n w rote lie before me, I may say that this book weighs fivepounds and that book, three pounds.We have here what appears to be a systematic ambiguity over a certain ca tegoryof words. as distinct from the idiosyncratic amb iguity of such words as English trun k,which may refer to an oversized suitcase or an app endag e of a n elephant. Typically,in such cases of systematic ambiguity, we can co mb ine the two senses in a relativeconstruction, as in (7) o r (8):

( 7 ) ( i ) John wrote a book that weighs five pounds( i i ) The book that John wrote weighs five pounds(8 ) ( i ) John wrote a book, which weighs five pounds( i i ) This book, which John wrote, weighs five pounds

Such com binations ar e excluded in oth er cases of ambiguity. For example, insentence (9). the phrase.flyingplanes may be construed a s referring to objects tha t fly,which are dangerous, or to the ac t of piloting, which is dangerous:(9 ) Flying planes can be dangerou s

In this case, the am biguity of,fl.ying pl an es is syntactically dete rm ine d, no t a gen eralproperty of phrases with a certain semantic function or character. In this case, ofcourse, we can no t com bine the tw o senses in a relative co nstruction . Sen tence (10) sexcluded, and in ( I I ) , we must construe the head of the relative t o be pla ne s rathertha n ,flving planes:

(10) Flying planes, which is a nuisance, are dangerous( I I ) Flying planes, which are a nuisance, is dangerousIn general. relativization seems to be free in the case of systematic ambiguity, a s in(7). (8). Consider, for example, such words as "temperature" which designatefunctions of some sor t . Th us we can say (12), meaning tha t the function is increasingover a certain interval including the present:

(12) The temperature is rising rapidlyBut such terms can typically be used also to designate a value of the funct ion a t agiven time and place, as in ( I 3):

( I 3) The temperature was 70' here a t sunriseTh ere are problems in determining the m eaning of suc h expression s (see References 6,note 12, and 7). But again we find that the two senses can be combined, as in (14):

(14) The temperature, which was 70' here at sunrise, is rising rapidlyPresumably the systematic ambiguity of words such a s book, temperature, an d

others (Reference 8 . p. 19 ff for som e examples) is determ ined in th e lexicon, perhap sby means explored in Reference 9 o r 10.Th us , we have a single for m al element with afixed range of mea ning, and relativizatio n is possible, d espite a shift of sense. But inthe case of,flying pla ne s (syntactically determined amb iguity) o r tru nk (idiosyncraticamb iguity) we have two fo rmal elements, in each case, with the s am e phonetic fo rm .Relativization is impo ssib le in su ch cases.One might imagine that the constrain t has some thing to d o with "semanticincoherence" or the like, but this is far from clear. Notice that nonrestrictives have a


5/12

50 Annals New York Academy of Sciencessense rather like conjunction. Thus the pairs (i) and ( i i ) of (15)-(17) are close inmeaning:

( 1 5 ) (i) Joh n, who goes to M IT, l ikes math(16) (i) The temperature, which was 70. is rising(17) (i ) This book, which J oh n wrote, weighs five pound s

(ii ) Jo hn l ikes math an d Jo hn goes to M IT( i i ) T h e temperature is rising and the temperature was 70( i i ) This book weighs five pounds and John wrote this book

Why, then, does not (l8)(i) have approximately the meaning of (18)(ii), and (l9 )(i)approximately the meaning of (19)(ii):(18) (i ) Flying planes, which is a nu isance, ar e dan ger ous(19) (i) Jumbo waved his trunk, which was full of clothes( i i ) Flying planes are dangerous and flying planes is a nuisance

( i i ) Ju m bo waved his trunk (of a n elephant) and the trunk (container) was fullTh ere is nothing inco herent ab ou t (18)(ii) an d (19)(ii). We might as k, then , why we d onot generalize from (15)-( 17) to the absurd (18)-( 19). I t seems that a general principleof syntax-semantics interaction is involved, and again it seems plausible to attributeit to universal gra mm ar, although much remains obscu re in this case.Let me turn now to some more complex cases. Consider the operat ion ofquestion-formation. A rule of wh-movement forms the sentence (20) by preposingthe wh-word of (21):

of clothes

(20) Wh o did John see(21) Joh n saw whoWe ignore here a number of irrelevant details. Th e wh-word may be deeplyembedded, as in (22) deriving from (23):

(22) Who did Mary believe that Bill wanted her to see(23) Mary believed that Bill wanted her to see whoBut if the wh-word is in a sentence emb edded within a noun phrase, i t cannot heextracted. Thu s we can no t derive (24) from (25):

(24) W ho did Mary believe the claim tha t Jo h n saw(25) Mary believed the claim that John saw whoThe latter fact follows fro m a general condit ion tha t Ross cal ls the complex no unphrase constraint (CN PC ; cf. Reference I I ) , which implies, in particular. th at a wordcannot be extracted from a sentence embe dded within a noun phrase, to a positionoutside that noun phrase.Again. it is difficult to imagine that everyone has received relevant instruction orbeen exposed to relevant experience. No r is there a ny semantic incoherence o r

othe r sim ilar defect in (24); were it a well-formed sentence, we would know exactlywhat i t means. Th us i t seems plausible to at t ribute C N P C to the language facultyitself. Otherwise, it is hard to see why some, or all. speakers would not simplygeneralize the rule of wh-movement to the case of a sentence within a no un ph rase.In this case, the constraint CNPC seems so special and artificial that we mightsearch for some deeper and more natural principle from which it follows. I haveargued elsewhere 12.3 that we can explain this an d othe r constraints on the assumption


6/12

C h o m s k y : On the Nature of Language 51that all transformational rules are bounded and cyclic; tha t is. they app ly within asentence or noun phrase or across adjacent phrases of the type sentence, noun phrase,but not across more widely separated domains. I t can be shown that many crucialinstances of C N P C follow from this general principle of rule applicatio n if we ad d a nempirical observation: that sentences, but n ot no un phrases, have a complementizerposition that can be filled by the wh-word or by othe r words such as that ("that Johnleft surprised me") or for ("for John to leave would be a mistake"); 1 3 ~ ~The explanation is controversial. If it is on the right track, as I believe, itillustrates again the expected interaction between general principles that belong tothe la n g u a g faculty a nd that determine the natur e of language, an d specific facts thatmust be learned. In this case, the general principle is the principle of boundedcyclicity, and th e fact th at m ust be learned is tha t English sentences hav e acomplementizer position that can be filled by a wh-word. as in (20), (22). or (26)( f rom (27)) . although we cannot, of course, derive (28) f rom (29), noun phraseshaving no complementizers:

(26) 1 wonder [who John saw](27) I wonder [COM P Jo hn saw who](28) I saw [who John's picture of](29) I saw [John's picture of who]Let us explore the dom ain of syntax-semantics interaction a bit further. Co nsiderEnglish reciprocal constructions. We have, basically, two types, as in (30). (31):

(30) Each of the men saw the others( 3 I The men saw each other

These seem close to synon ym ous a t first glance. In general, we can replace the pair(each of the men. the o the rs ) by the pair ( t h e m en. each o the r ) with little if an y ch an geof sense. Such substitution gives (31) from (30) in the context (32):(32) . . . s a w . . .

But such substitution without chang e of sense is not always possible. Con sider thetwo discourses (33) and (34):(33) Each of the women saw some of the films. Each of the men saw the others.(34) Each of the women saw some of the films. The men saw each other.

In (33), the phrase th e others may refer to the films not seen by th e women. In (34).the phrase each other must refer to the men. Th us substitution of the pair ( t h e men,each other) for (each of the men, the o the rs ) in (33) radically changes the sense. Thepoint is that the phrase each other must have its anteced ent within the sa me sentence,while the phrase th e others need not. The sentences (30) an d (31) may have close tothe same meaning, but (30) has a broader range of meaning. The reference of eachother is fixed by a princip le of (sentence-) gr am m ar , but the reference of the others isnot. though the principle governing it may apply within a sentence.This distinction h as an interesting range of consequences. Con sider, for exam ple.the sentence (35):(35) Each of the candidates wanted [John to vote for the others]

Supp ose that we replace the pair (each o f t he candidates, th e others) by the pair ( t hecandidates. each othe r). Then we derive (36):(36) The candidates wanted [Joh n to vote for each other]


7/12

52 A n n a l s N e w York A c a d e m y of SciencesBut (36) is no sentence of English. Again, the problem is not semantic. I f compelledto assign it a sense, we w ould, pre sum ably, select the perfectly sensible (35). We m ightsuppose that the problem is that each oth er must find its antecedent within the samesentential structure, that is, within the brackets in (36). But this condition is toostrong, as we can see from (37):

(37) The candidates wanted [each other to win]In the sentences (35). (36), (37). the bracketed phrases are ~e nte nt i a1 . l~ut only (35)and (37) are well formed. (36) is not, It is fair to ask why not. W hy does th e speakernot generalize, notin g that the su bsti tutio n of phrases is permissible in (30J, (31), or in(38). (37):

(38) Each of the candid ates wanted [the others to win]Why is the substitution impermissible in ( 3 3 , contrary t o the natural generalization?It is difficult to believe, again, that this is the result of specific tra inin g or experience,in each (o r any) case. Language learners are not corrected for such mistakes as (36);nor d o they actually make s uch mistakes. In fact, relevant exam ples are rare, and aperson might well live most or all of his life witho ut com ing across any, although hewould mak e the prop er distinctions on first presen tation with a relevant exam ple, soit appears.The answer to the problem, I believe, lies in still another general condition ongrammar that I have called elsewhere the specified subjec t con ditio n (SS C) .l?J T hecondition entails that a phrase X within a phrase P cannot be related to a positionoutside of P if P contains a subject distinct from X (or, in fact, distinct from theoccupant of the outside position. in a well-defined sense). T h us if the bracke ted.expression in (39) is a P, then Y ca nn ot be related by rule t o X if Z is the sub ject of P:

(39) . . . Y . . [ Z . . . x . .]In (36). the candidates is Y , John is Z, and each o ther is X. Therefore the rule ofreciprocal interpretation cannot apply to the pair (the candidates. each other), bySSC . But the principle SS C is a principle of (sentence-) g ram m ar. Therefore it doesnot apply to the case (35). Similarly, it is inapplicable in (37) because each other isitself the embedded subject.The principle S S C applies much mo re broadly. Con sider again wh-movement.We have (40) but not (41):

(40) W ho did you hear s tor ies abo ut(41) Who did you hear [Johns stories about]((40) may also be blocked in a style tha t imposes more stringent conditio ns onstrand ed prepositions, bu t this is irrelevant here). Th e reason, 1 think, is that t hebracketed noun phrase of (41) has a subject, Jo hn , while the phrase stories ab ou twho, which underlies (40), has no subject. Therefore SSC blocks wh-movement in(41), but not (40). We a re relying here o n a generalization of the notion subject oftraditional grammar, a generalization which 1 believe is independently well moti-vated.8 Both this extended n otion of subject an d the general principle S S C ar e faircandidates for universal gramm ar. T ha t is , i t seems to m e quite reasonable to proposethat they are determined by the 1anguage.capacity itself.Consider on e final case, on e step more com plex. C onsider the sentences (42), (43):

(42) Jo hn seemed to each of the men to like the o thers(43) Jo hn seemed to the m en to l ike each other


8/12

Chomsky: On t h e N a t u r e of Language 53Here again, substitution of (the men, each oth er) fo r (each o ft h e men, the others) in(42) is impossible. Why is this so?Notice that the verb like in (42). (43) does not have a subject before it as it do es in(44):

(44) The men wanted [Bill to like each other]Thus (44) has the form of (39); the rule of reciprocal interp retatio n is blocked becauseof the presence of the subject Bill (= Z of (39)). But in (43), there is no sub-ject corresponding to Bill of (44). Nevertheless, (43) is no more grammatical than(44).In traditional terms, the verb like in (43) does indeed have a subject, namely, anunderstood subject. John is the un derstood subject of like in (43). Apparently,this mentally present but physically absent subject behaves just as the physicallypresent subject Billdoes in (44). ringing SSC into operat ion. Thu s we may think of(43) as having the abstract structu re (45):

(45) John seemed to the men [Z to like each other]Here Z is the null subject, interpreted a s referring to Joh n. But (45) has the structure(39). so that SSC blocks the reciprocal rule.Again, there is no logical necessity for a language to be designed in such a wayth at phonetically null mentally prese nt subjects behave ju st a s physically presentsubjects do . If languag e were de signed differently, (43) would have approximately th emeaning of (42). just as (31) has approximately the meaning of (30) and (37)approximately the meaning of (38). Th e language wou ld be n o less well designed fo rcommunicat ion o r expression of thought (for a creature designed to handle thislanguage). Again. a person might live most or all of his life without ever beingexposed to relevant evidence (let alon e training) th at would indicate to him t ha t thenatural generalization is (for som e reason) blocked in (43). Thus, it seems absurd toclaim that experience provides the basis for these judgments. N or d o they seemto be explicable in functional terms on grounds of communicative efficiency orthe like.Rather, i t seems that on ce again, the facts reflect a biologically given preco nditio nfor learning. The child learning English simply imposes the requirement thatmentally present su bjects function a s though they were physically present. H e do esthis even in the absence of relevant evidence, so i t appears. A theory of universalgram mar-tha t is, a theory of the languag e faculty-must seek a n exp lana tion as t owhy this is so . Th e answer, 1 think, lies in the trace theory of mov emen t rules,which requires that when a phrase is moved by transformation it leaves behind aphonetically null but morphologically real element trace (a so-called zero mor-pheme) that functions semantically as a kind of bound variable. Other rules ofsyntax an d m orpholog y have n o way of knowing tha t this element will (ultimately) bephonetically null. Hence it operates as a specified subject, an d in o th er ways in th esystem of rules, while playing an essential role in semantic interpretation.Space does not permit a discussion of syntactic an d sem antic conseq uence s of thetrace theory.)J2.Is.l6 1 will de scrib e briefly a curious phonetic effect of trace, whichgives indep enden t evidence th at it is morphologically nonnull. Con sider the sentences(46). (471, (48):

(46) Wh o do you want to h it(47) W ho d o you w ant to see Bill(48) Wh o do you want to choose


9/12

54 A n n a l s N e w York A c a d e m y of SciencesThese derive, respectively, from (49). ( SO ) , (51):

(49) You want to hit who(50) You want who to see Bill(51) (i) You want who to choose(ii) You want to choose whoTh u s (48) is ambiguous a s between the two interpretations (5 I)(i). (51)(ii).It has been observed by a number of people that in (46) and sense (51)(ii) of (48),the sequence want to can be elided to wanno; but this is impossible in (47) or in sense(51)(i) of (48). Th us consider ( 5 2 ) . (53). (54):

(52) Who d o you wanna h it(53) W ho d o you wanna see Bill(54) Who d o you wanna choose

Sentence (54) can mean only (51)(ii) and (53) s impossible. Why is this so?corresponding to (49). (SO), (5 ) , respectively:Notice tha t unde r the trace theory, we have the abstract s tructures ( 5 5 ) , (56). (57).

(55) W ho do you want to h i t I(56) Who d o you wan t I to see Bill(57) (i) Wh o do you wan t 1 to choose( ii ) W ho do you w ant to choose IIn each case, I is the trace left by wh-movement. Th ere is a rule attac hin g to to apreceding occurrence of want in certain styles. T he resulting phonological word th enreduces to wanna. Obviously, the rule will not apply if there is intervening morpho-logical material, as in (58):

(58) John will want Bill to goFor exactly the same reason, the attachm ent rule will not apply in (56) and (57)(i). I tfollows tha t wanna can appe ar in (46) but not (47). and that w hen i t appears in (48),we know that (48) was derived from (57)(ii) rather than (57)(i) and therefore means(51)(ii) rather than (51)(i).Although the trace is phonetically null, it is syntactically and morphologicallyreal and enters into the computations that determine the form and meaning ofsentences no less than the zero plural of sheep. See References 17 and 18 an dreferences cited there. Note that 1 have not explained why th e understood subject ofhi t does not block elision in (46). or the understood subject of choose in (51)(ii)).Here lies ano the r tale.The trace theory too is controversial and not without its problems. But it doesserve to explain quite a range of appare ntly unrelated facts and it lays the basis fo rother promising developments in linguistic theory, 1 believe. I f it proves to be correct,or perhaps to be a special case of a still mo re ade qu ate theory, the n again we will haveevidence concerning the intrinsic structure of the language faculty, a species propertytha t fixes the essential na tur e of language.Exam ples such a s these only scratch the surface of th e problem , needless to say. Ioffer them to illu strate a certain pattern of argumen t that has proved quite productivean d tha t offers a great deal of promise for the futu re as well, in tha t its potentialitieshave barely been tapped . W e might wish t ha t m ore direct tests could be devised fo rhypotheses concerning universal grammar. But even in their absence, strictlylinguistic investigation can lead us to some plausible general principles of somesubtlety, principles that we may hope to relate to the results of other lines ofinvestigation into the nature of language.


10/12

Chomsky: On the Nature of Language 55

I .2.3.4.5.6.7.8.9.

10.I I .12.13.14.15 .16.17.18.

R E F E R E N C E SJAKOHSON.., G. FANT M. HALE. 1963. Preliminaries to Speech Analysis. M IT Press.C I IO M S K Y ,. 1965. Aspects of the Theory of Syntax. MIT Press. Cambridge. Mass.CHOMSKY,. 1976. Reflections on Language. P anth eon . New York, N.Y.KI JNO , S. 1973. The Structure of the Japanese Language. MIT Press. Cambridge, Mass.ANDREWS.. 1975. Studies in the Syntax of Relative and Comparative Clauses. Ph.D.

Cambridge, Mass.

dissertation. MIT. Cambridge. Mass.C H O M S K Y , . 1975. Questions of form an d interpretation. Linguistic Analysis I ( I ) :75-109.H A C K I N G .. 1975. All kinds of possibility. Philos. Rev. 84 3): 321-37.C H O M S K Y ,. 1972. Studies on S eman tics in Generative Grammar. Mou ton. The Hague,KATZ , J. J . 1972. Semant ic Theory. Harper and Row. New York, N .Y .J A C K E N W F F ,. S. 1972. Semantic Interpretation in Generative Grammar. MIT Press.Ross, J. R. 1967. Constraints in Variables in Syntax. Ph.D. dissertation. MIT. Cam-C H O M S K Y ,. 1973. Canditions o n transformations. In A Festschrift for M orris Halle. S.BRESNAN.. 1970. On complementizers: towards a syntactic theory. Foundations ofB R E S N A N .. 1972. The Theory of Complementizers in English S ynta x. Ph.D. dissertation.WASOW, . Anaphoric Relations in English. To be published.FIENG O, . W. 1974. Seman tic Cond itions on Su rface Structure. Ph.D. dissertation. MI T.SELKIRK.. 1972. Th e Phrase Phonology of English and French. Ph .D. dissertation. M IT .LIGHTFOOT.. 1975. Traces and Doubly Moved NPs. Mimeograph. McGill University.

The Netherlands.

Cambridge, Mass.bridge, Mass.Anderson & P. Kiparsky, Eds. Holt, Rinehart and Winston. New York. N.Y.Language 6: 297-32 I .MIT. Cam bridge, Mass.

Cambridge, Mass.Cambridge, Mass.Montreal, Quebec.

DISCLJSSIONF T H E P A P E RD R . VO N GLASERSFELD:may have misunderstood w hat Professor Chom sky said,but the last exam ple, "Wh o d o you w ant t o see Bill," surely is a perfectly go odEnglish sentence?D R . C H O M S K Y :t's a perfectly good English sentence, but you can't run togetherthe w ords "want" an d "to" to give the elided "wanna." T ha t is, you can't say. "Whod o you wa nna see Bill?" That's th e point. It's a phonetic point. The sentence is fine,but it can't have that phonetic realization.DR. V O N GLASERSFELD:wouldn't be able to judge that.M A R YSMITHU n iv . of Connecticut, Storrs, Conn.): Do you have any evidence

for o ther languages that this sort of subject constraint is really built in?DR. CHO MSK Y: ell, I person ally don't, becaus e I don't know any oth erlanguages well enoug h t o work on them. But there is some work o n oth er languagestha t has brou ght ou t some very interesting evidence, I think. Fo r example, there is aBrazilian linguist named Carlos Quicoli who has been working recently on Portu-guese clitic movement, movement of pronominal elements from their originalposition to attach to verbs; and I think he's got som e very striking evidence tha t itobserves the specified subject conditio n. Jean-R oger Verneaud, a French linguist, ha spresented some evidence from French, again bearing on the same things. There is


11/12

56 A n n a l s New York A c a d e m y of S c ie n c e sfurther evidence in an im portant new book by Richard Kayne, an d there are oth erthings of that sort.

1 should say, incidentally, th at it would be im port ant. of cou rse, to try to find ou twhe ther there is negative evidence. I don't know of any clear cases but tha t may bethrough lack of looking. If there is contrary evidence. of course, that means that thecondition can't be right as formulated, because it's intended to be a universalcondition.But I thin k o ne would ex pect th at relevant evidence will not be easily obtained inother languages any more than it is easily obtained in English. That is. any bit ofevidence that is to be relevant to a principle of this degree of abstractness has to bepresented within a pretty rich framework of analytic assumptions; so tha t asuperficial investigation of a lan gua ge won't tell you very mu ch.For example, you may superficially find sentences that seem to support or toreject the theory. But it won't tell you much unless you really show what rules areinvolved t ha t generated those sentences. And t ha t requires independe nt evidence an dstudy of some depth. T hus, a condition on rules can only be confirmed or refuted byrules, not by observed phenomena, as a point of logic.DR. OUIS EVIC:Univ .of Wisconsin, Madison, W is): f these things are built in,why does someone like Professor von Glasersfeld have difficulty figuring on yourexample?

DR. H O M S K Y :he difficulty he had was becau se he didn't know , and couldn'tknow as a nonnative sp eaker, that there is a n elision rule in English t ha t form s"wanna" from "want to." T he fact that you ca n elide "want to" to "wanna" is no tbuilt in. Okay'? In oth er w ords, there could very well be a dia lect of English in whichthat's just false. Sim ilarly, it's n ot bu ilt in tha t the wo rd "want" begins with "wa"instead of "pa", let's say.DR. EV IC : see now tha t you wanted to use that as an example in yourargument, but-DR. H O M S K Y :he a rgum ent is th at given this rule, which has to be learned, wesee that in fact this rule obse rves some very stra nge con ditio ns. It applies in som ecases and it doesn't apply in other cases.Now, the question we have to ask is how d o we know, h ow does anybody w hoknows the language know, th at you can app ly the rule in som e cases an d no t in othe rcases? Well, if you don't even know that the rule applies, of course the questiondoesn't arise. But for a speaker of English w ho know s that there is such a rule, whoknows you can say "I wa nna go", let's say, the ques tion arises, how do es he also kno wtha t you can't say "who d o you w ann a see Bill?" And th e answe r, I thin k, is in thiscondition.DR. EV IC : kay, that's just fo r the elision rule, but for the oth er rule you wouldexpect that anybody who's learning English should be able to pick that up.

DR. H O M S K Y :hich o ther rule d o you mean?DR. E V IC : he reciprocal rule, with the senses you gave.DR. H O M S K Y :ell, there, too, is a learned element. Don't forget, all of thesethings are going to involve inte raction of learned elem ents and fixed principles. So in

the case of the reciprocals there is a learned element; namely, that "each other" and"the others" differ, in t ha t "each o ther" is govern ed by rules of sentence gr am m ar an d"the others" is not govern ed by rules of sentence gr am m ar . I don't think there is an yway to know that, by one's biological constitution.But given that you know that, man y consequences follow by virtue of principlestha t I think are part of one's biological constitution. So from this point of view, onewould be studying language exactly in the way you'd stud y a n organ , say, the hea rt.Some aspects of each human heart depend on the embryological environment, let 's


12/12

C h o m s k y : On t h e N a t u r e of Language 57say the nutritional level or something. I don't know what. But other aspects of itdon't. And we want to try and sort those out.Well, in general it's very tricky to so rt them ou t in the case of a developed organ .What I was trying to illustrate is a pattern of reasoning which I think enables us ininteresting cases to sort out these factors.S. H A R N A D :Let me just ask a question which everyone else who has beenfaithfully atte ndin g these sessions is surely bu rnin g t o ask. If som e rules you h avedescribed constitute universal con straints on all languages, yet they a re no t learned,nor are they somehow logically necessary a priori, how did language get that way?

DR.CHOMSKY: Well, it seems to me tha t would be like asking the question h owdoe s the heart get th at way'? 1 mea n, we don't learn t o have a heart , we don't learn tohave arm s rather tha n wings. W hat is interesting t o me is tha t the question shou ld beasked . It seems to be a natural question ; everyone asks it. And I think we should askwhy people ask it.M y feeling is th at if, say, the M artia n scientist tha t I was imagining were to lookat earthlings and investigate them, he would have n o reason to do ub t tha t language isas much an organ of the bod y a s the eye or t he heart o r the liver. It's strictlycharac teristic of the species, has a highly intricate s tructure, develop ed mo re o r lessindependently of experience in very specific ways, and so on. It has all the generalproperties of an organ of the body.Why does i t seem so strange to us to think in those terms? In othe r wor ds, whydoes i t seem so strange to us to consider hu m an beings in the m ental aspe cts of theirlife as we would consider any object of the natural world? Why is it so natural toinsist on a dichotomy in accordance with which we treat the physical development ofhuman beings as if humans belong to the physical world, but not their mental

M y own feeling is that what we have here is a kind of inversion of a historicaltradition that really ultimately has religious and other roots. Traditionally it wasregarded as impossible, for all sorts of reasons, to study the human soul by themethods of science.Now, what's interesting to me is th at precisely within tho se tendencies in sciencethat tried to be naturalistic, that tried to defend science against religious barbarismand so on, precisely in those bran che s of science the same curio us refusa l to de al withthe facts persisted, so that the question you raise seems to pose an overwhelmingparado x to an investigator. The p roblem seems overwhelming of explaining thegrowth of this particular mental org an, hum an language, through t he interaction of agenetically determined system with experience, althoug h an alo go us question s ab o utoth er organs d o not seem to arise with the same force.

I think the question arises in th e case of language exactly as it d oe s in the case ofthe eye. or the heart , or sexual organs (to pick something that matures long afterbirth), and so on.There is every reason to suppose that this mental organ, human language,develops in accordance with its genetically determined characteristics, with someminor modifications that give one language o r ano ther , depending o n experience. Butthen, one would say the same abo ut any bodily o rgan a s far as I can see.

development?

Noam Chomsky, On the Nature of Language, 1976

Documents

Transcript of Noam Chomsky, On the Nature of Language, 1976