ile i. -;a i;;i iiil r following this line of thinking, s ... file-;a :e*;$it $ii...

8
-;a :E*;$it $ii l;$aF;a'f;t*Eei;;i E=ra,E E;. J.E: s* Z=Et:?;€ji €TE*iEi; r*r;*i'r p i E ;:; I 3 ?:+ uE rr^'F o.- ZfEHE:l'.- 6-i = "n !# ; :?€E i. +z=Ee;T3;i 3;+i*Ei1=>i}:g :;it } E;i; n 5€ Tia i a= :iz*=Z;?4. leZ; F";i-*;r u.E i-: ; ; = a :jE i E-i v;3*:i;E*r :+€EiF€:c!E€tr:Ees €a=r t =Z EFri;e: i!13 ?,8?: i: ;Ei€E;Et;ii €::f S ZZ ;=isoE r:: *, i =E=2ii3 i=; iEc =+; ii lg F ej i ; E tn;i :: $i;€i;; rjn5;+iEa, as #T ; #* # *f si+i i;if 1F55EFi tgt+*i€;E i*tig+ = ? rri*r ;;;; r; rE-*t ; :.5t F;;?; *Ag.igt ag i 1,21r; ? al:ii ltzi3e lz;i=' ;i E : a$ E ?e i i z#E i r e 7:!!;Et;;?;PE* *g r g ,: H.Ex $gf E:;; =: = I ;:,;p!;i g;.' I:==Erf iei* q'5[f$**p ^ ijES iltlE ;F:s ;,1= ;il:; ; i3;E€ i i iiil i*Eiz=:?igt:s=Eii; ati+; *+grg; i; z i;;i €+€=:i iiiyi===;EEiEsi;l+;j:€;.:: I ; E|+i i,:.;i r -3€;iFi;;j lfiE eie;F:i;3i ; ilEg EE a; €'iE; i: FEF;=i'i€ 1t ritji c: Ed - 3 x ile I ryi E {d€ 6;i;;;€tt re*;*ilE:"*= :E;;E I = f; i,i;E€;j ;g-,j'= +':i3r;6: :i€ ;;g:s2*E; d *?t r??;;?:F s F:iitt;*?;fFg;til?t;;izz* - ?s 5 i€; *;" ;E :?;f ar: += ifFE?j?g!;g;:;;l $ Ei:t;:gi at;jgiiie!€?i?j fl.iFi;EEij Ari; REFLECTIONS ON LANGUAGE formed (13d). Again SSC explains the difference, if We take the abstract phrase structure to be as indicated bv "_ " in (13), a conclusion for which there is independent evi- dence. Thus once again, analogies are blocked by the gen- eral condition SSC. It is worth emphasizing that each of these examples adds further support to the conclusion that approaches of the character of E, which assigns a fundamental role to seg- mentation, classification, analogy, and generalization, are mistaken in prinCiple. Furthermore, in these examples at leas t, "semantic explanations" seem beside the pOint. Thus there is no semantic consideration that blocks the inter- pretation (13d) for (13b), or tha t prevents (8) from having the perfectly sensible interpretation "it appears to each of John's friends that Mary hates the other(s)," just as (2) has the interpretation "it appeared to Mary that each of John's friends hates the other(s)." In such cases, we might trivialize the notion "analogy" by building into it some condition equivalent to SSC, but this is clearly beside the point. (Cf. note 5.) Notice that sse itself might well have a functional or semantic explanation, but this is a different matter. (Cf. Chomsky [1973a] for some discussion.) Such examples might lead our scientist S to postulate a grammar containing such rules as NP-preposing, reciprocal interpretat ion, UP, a rule giving the meaning of not . .. many, and so on. The rules would be governed by such principles as SSC, and by the general principles of interpre- tation of initial phrase markers and surface structure men- tioned earlier (d. note 6, and chapter 3, pp. 94- 117). If S is inclined to believe that the theory of language learning is of the character of E, he will attempt to design procedures of association, habit formation, induction, or analysis by segmentation and classification that will give this grammar as "output" on the basis of a record of data. A system of procedures \",ill be tested for adequacy in terms of its suc- cess in this task. As just noted, an analysis of properties of Problems and Mysteries 153 the grammar suggests that no such approach is feasible and that a theory of the character of R is much more likely to be succ es sful. Following this line of thinking, S might ask what ele- ments of th e grammar might be candidates for universal grammar, th e schematism that constitutes an element of the initial state (the property P). Th e empirical conditions of his problem are plain enough. The variety of languages provides an upper bound on the and speCifiCity of the properties that he may assign to ul1lversal grammar. The necessity of accounting for acquisition of particular gram- mars provides a lower bound. Between these bounds lies the theory of language learning, what we have called "LT(H,L)" in chapter 1- S might propose, for example, that SSC, some of the prin- ciples of interpretation of initial phrase markers and surface structures (d. note 6), the conditions on permissible gram- matical ru les, and so on are elements of universal grammar, . whereas such rul es as NP-preposing are specific to English or have properties speCific to English. Thus he might con- clude that a child must learn the rule of NP-preposing or some of its properties, but need not learn SSC, or the gen- eral properties of grammar. Rather, he would have this formation available as an element of P, a property of hIS initial state (though, as noted earlier, this genetically de- termined property may, as in other familiar cases, function only at a certain maturational stage or under appropriate triggering conditions). The child will thus select a grammar meeting these conditions and containing the rule of NP- preposing with its particular properties. Evidence for the latter rule is considerable, as noted in chapter 3 and refer- ences cited there. Since there has been much confusion about these matters, perhaps it may be useful to recall a still simpler case that illustrates the general point. Consider again the discussion of the principle of structure-dependence of rules discussed

Upload: hahanh

Post on 08-Nov-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

-;a :E*;$it $ii l;$aF

;a'f;t*Eei;;i

E=

ra,EE

;. J.E

:s* Z

=E

t:?;€ji €TE

*iEi; r*r;*i'r

p i E

;:;I 3 ?:+

uE rr^'F

o.- ZfE

HE

:l'.- 6-i = "n !# ;

:?€Ei. +

z=E

e;T3;i 3;+

i*Ei1=

>i}:g :;it } E

;i;n 5€ T

iai

a= :iz*=

Z;?4. leZ

; F";i-*;r u.E

i-: ; ; = a :jE

i E-i v;3*:i;E

*r :+€E

iF€:c!E

€tr:Ees €a=

rt =

Z E

Fri;e: i!13 ?,8?: i: ;E

i€E;E

t;ii €::fS

ZZ

;=isoE

r:: *, i =E

=2ii3 i=

; iEc =

+; ii lg

F ej i ; E

tn;i :: $i;€i;; rjn5;+iE

a, as #T ;

#* # *f si+i i;if 1F

55EF

i tgt+*i€;E

i*tig+=

? rri*r ;;;; r; rE-*t ; :.5t F

;;?; *Ag.igt ag i

1,21r; ? al:ii ltzi3e lz;i=' ;i

E : a$ E

?e i i z#E

i r e 7:!!;E

t;;?;PE

* *g r g ,: H.E

x $gf E:;;

=: =

I ;:,;p!;i g;.' I:==

Erf iei* q'5[f$**p

^ ijES

iltlE ;F

:s ;,1= ;il:; ; i3;E

€ ii iiil i*E

iz=:?igt:s=

Eii; ati+

; *+grg; i;

z i;;i €+

€=:i iiiyi=

==

;EE

iEsi;l+

;j:€;.:: I; E

|+i i,:.;i r -3€;iF

i;;j lfiE eie;F

:i;3i; ilE

g EE

a; €'iE; i: F

EF

;=i'i€ 1t ritji c: E

d -

3 x ile I ryi E

{d€ 6;i;;;€tt re*;*ilE:"*=

:E;;E

I =

f; i,i;E€;j ;g-,j'=

+':i3r;6: :i€ ;;g:s2*E

;d *?t r??;;?:F

s F:iitt;*?;fF

g;til?t;;izz*- ?s 5 i€; *;" ;E

:?;f ar: +=

ifFE

?j?g!;g;:;;l

$ Ei:t;:gi at;jgiiie!€?i?j fl.iF

i;EE

ij Ari;

REFLECTIONS ON LANGUAGE

formed (13d). Again SSC explains the difference, if We take the abstract phrase structure to be as indicated bv "_ " in (13), a conclusion for which there is independent evi­dence. Thus once again, analogies are blocked by the gen­eral condition SSC.

It is worth emphasizing that each of these examples adds further support to the conclusion that approaches of the character of E, which assigns a fundamental role to seg­mentation, classification, analogy, and generalization, are mistaken in prinCiple. Furthermore, in these examples at least, "semantic explanations" seem beside the pOint. Thus there is no semantic consideration that blocks the inter­pretation (13d) for (13b), or that prevents (8) from having the perfectly sensible interpretation "it appears to each of John's friends that Mary hates the other(s)," just as (2) has the interpretation "it appeared to Mary that each of John's friends hates the other(s)." In such cases, we might trivialize the notion "analogy" by building into it some condition equivalent to SSC, but this is clearly beside the point. (Cf. note 5.) Notice that sse itself might well have a functional or semantic explanation, but this is a different matter. (Cf. Chomsky [1973a] for some discussion.)

Such examples might lead our scientist S to postulate a grammar containing such rules as NP-preposing, reciprocal interpretation, UP, a rule giving the meaning of not . .. many, and so on. The rules would be governed by such principles as SSC, and by the general principles of interpre­tation of initial phrase markers and surface structure men­tioned earlier (d. note 6, and chapter 3, pp. 94- 117). If S is inclined to believe that the theory of language learning is of the character of E, he will attempt to design procedures of association, habit formation, induction, or analysis by segmentation and classification that will give this grammar as "output" on the basis of a record of data. A system of procedures \",ill be tested for adequacy in terms of its suc­cess in this task. As just noted, an analysis of properties of

Problems and Mysteries 153

the grammar suggests that no such approach is feasible and that a theory of the character of R is much more likely to be

successful. Following this line of thinking, S might ask what ele-

ments of the grammar might be candidates for universal grammar, the schematism that constitutes an element of the initial state (the property P). The empirical conditions of his problem are plain enough. The variety of languages provides an upper bound on the richn~ss and speCifiCity of the properties that he may assign to ul1lversal grammar. The necessity of accounting for acquisition of particular gram­mars provides a lower bound. Between these bounds lies the theory of language learning, what we have called "LT(H,L)" in chapter 1-

S might propose, for example, that SSC, some of the prin-ciples of interpretation of initial phrase markers and surface structures (d. note 6), the conditions on permissible gram­matical rules, and so on are elements of universal grammar,

. whereas such rules as NP-preposing are specific to English or have properties speCific to English. Thus he might con­clude that a child must learn the rule of NP-preposing or some of its properties, but need not learn SSC, or the gen­eral properties of grammar. Rather, he would have this i~­formation available as an element of P, a property of hIS initial state (though, as noted earlier, this genetically de­termined property may, as in other familiar cases, function only at a certain maturational stage or under appropriate triggering conditions). The child will thus select a grammar meeting these conditions and containing the rule of NP­preposing with its particular properties. Evidence for the latter rule is considerable, as noted in chapter 3 and refer-

ences cited there. Since there has been much confusion about these matters,

perhaps it may be useful to recall a still simpler case that illustrates the general point. Consider again the discussion of the principle of structure-dependence of rules discussed

$ aizzt!-.Eiii1=

r: tE. *i;E

;: +

fE€,=

.6EE

;ij:!if :E

:";:ill|*iih?r= Iirs H

; $* z "in iE;E

i ; ! E;.t;E

:=s?iE

;g€ ==

2i lii g€: i; +

'-; *g;=

E 3ss=

=t *st ?'"iE

;;fJgg;; E:Z

-if,&,r-:;E

! lzz; i€ eizEE

i i+g i r; .E

EE

'; EE

:: Higi €r;

g EetgE

_E - .+

+* r_c **r nr t, I+

Fi

E i :!;ag}g

E'E

i Eri+

zz;: i;iil=,;i i t$; irr t E

; ii€E ?*:?i

i g:: liri*i;Ei:=

iiEili;ii;i

;= ;{;jarg+

tE

=E

=rF

3: ?;= rE

i e ; iE=

E rE

rz| = : =

iE s.E

: " +s -

* i rE =

,Ez': ! E

ii= i g lE

i;: f ;=e tte *iF

i;Ei€

: j ; " e7: g F"-i E

E i E

;E t I F

: i ; ; e E !. : s *,* : s I I s

;iaa?zl;-#-a F

r; ?1'::*3-i€ i*;,+g,ir=

E: p E

i E

:€ it i:;:: 'ziE E

j t"s t gii fr *s"= j gf €f i F

f 5 fl

:....|......:.:.|:,.':...'...:..i,..:...::''i':.:

i : T;f

E F

T r : g I

F

ijf;€!j::gf:g::i

= E

T=

- a.:T g *J

€i6;iEg€u ii;a

3 - *5 r 9€^3 E g: q ;.

= € =

la = t J : 27 3 e

E E

f fs r iy;: rf:.s*t;;;?$'i;;€E:€:iE

EE

r€Elji

#a).-rrF--A

E iE

s;c;f E; E

I Fli::€!E

;ii:;:E

;t5:l{y;-EzP

r5: i i€ g,* f: sT

.: ;

;EiiE

FjF

i'; ii ; E

z:i i gjE=

z ?g r* 5j i:j: T

7=": ; : I;?

-:: ? -=

^-

E

i" i'a

C -

* i

boo- A s

j =

- j

. 2=

i 5 s. l-= * i

=..i e F

'a xE

1 ;=

i I

i i:Ei ;:ii: i; +

r :i5g : #ti :;z:; i z ie; las f i:: fs' 'Z

ii:EF

; .i+E

P*r"iiilT

ff jE

3 :f ii ;i; F* ; : fli;f i;j i=

E* i

; ;ii :r : !i ij : tf {5t gi;j F;j

iE * tj i'rf ; iif"j E

TifE

iEigi

$ : ? :3:F: E

;: jFg; j i:jl F

i F€

154 REFLECTIONS ON LANGUAGE

in chapter I, pp. 30-3. For reasons explained there, S ~v~L:ld naturally conclude that some feature of the child's ll1Ibai st~te leads him to reject the structure-independent hypothesIs 1 to account for question formation select· . , Iller lllstead tl~e more abstract and complex structure-dependen~ hypothesIs 2, on the basis of evidence compatible w 'tl bh'IS'1 · 11 ot,' mIg lt conclude that universal grammar provides I~otation for rules that simply does not permit the formula~ bon of structure-independent rules, despite the advantaO'es that they would have for some different organism or devi~' This conclusio~ would constitute one part of S's theory ~f language leanung. The example again supports the more general conclusion that a theory of language learning is of the character of Ii. rather than E.

If we sufficiently enrich t~e system of universal grammar, postulat.ed as an element of the innate property P, it may be possIble to account for acquisition of grammar on the basIs of the limited evidence available. At least, this seems a feasible prospect (but see note 4).

Supp~se that S succeeds in developing a tentative theory ?f l:armng for common sense and for language. If my guess IS nght~ these would be of the character of R; that is, they would ll1volve fixed and highly restrictive schemata which come into operation under limited conditions of exposure to data, determine the interpretation of these data as ex­

perience, and lead to the selection of systems of rules (gram­mar, common sense) which are put to use in human action and interaction. It is not excluded that the schematisms for grammar and for common sense have nontrivial common ele:?ents, al:d S will naturally search for these. 10 There may be generalIzed learning strategies" that form part of both of these empirical theories. On the other hand it seems likely .that these schematisms will have unique elements as well: J~ISt a.s common sense may well have special devices ~o dIstll1gmsh faces from other geometrical objects. There IS no reason to expect to find in the property P significant

P1'Oblerns and M ystel'ies 155

(lI1alogues to the analyzing mechanisms for identification of faces or determination of line, angle, and motionY There is no reason to expect that the principle of structure-depen­dence or SSC will appear in the theory of common sense. At some sufficiently abstract level, one may find analogies; S might ask whether the system for identifying faces in­volves an abstract representation or model and a "transfor­mational system" of projection. But as a scientist, S would have no dogmatic beliefs as to the character of the various systems of learning and their interrelation. Ii.ather, this · would be an empirical problem that he would hope to solve.

To the extent that S succeeds in characterizing the innate properties of mind that make possible the learning of gram­mar and common sense, he would be able to explain why these systems are qualitatively so djfferent from the third cognitive structure mentioned earlier, knowledge of physics. That is, he would now regard the properties of mind that underlie the acquisition of language and common sense as biological properties of the organism, on a par in this re­spect with those that enable a bird to build a nest or repro­duce a characteristic song; or, for that matter, comparable to the properties that account for the development of par­ticular organs of the body. (Cf. chapter 1.) Humans are not specially adapted, in the same way, to the learning of physics.

Parenthetically, S might conclude that something similar is true of physics as well. The human mind is a biologically given system with certain powers and limits. As Charles Sanders Peirce argued, "Man's mind has a natural adap­tation to imagining correct theories of some kinds .... If man had not the gift of a mind adapted to his requirements, he could not have acquired any knowledge" (ed. Tomas, 1957). The fact that "admissible hypotheses" are available to this specific biological system accounts for its ability to Construct rich and complex explanatory theories. But the same properties of mind that provide admissible hypotheses

r i2eEe;E

; E7E

EE

.?Z 7E

:=e*z?u:i=

Ei

E';E

iEuaE

4a*i *EE

{IT€,"*zr a; a+

ta :lu1| EE

;r:E

li?iii ; re;#,! l; E

:i€: FE

gEE

:e: r :iE r

* E;+

z;+=

= 'lzrgiii g:trF

I++

,f,i;cjEi y;i

i ?ii|i=z:i E

;=i;z;r i;iiili|iir! zT

= sp

: \zriEr=

* z;E,s*4 7 E

Eiiiur; €:E

ii=gt;i5

: a4EiE

tiF E

iZ E

ZE

T', E

Et z z ? i i'- e i i€ =

! ai-i j : : ; e *T

E i?.=

=i*ci * af€ i I *? ?21e,?r:i;{

€ is=F

e E ia ;i+

;;; +isf i fii j€ fr';iE

*:: ; ; ryX- ezig* j : H

E iii i+

7tn: =r* ii;i

: E iiz=

_:Z=

' a=

+ii| qE

E -=

' s E

ET

iElS

r? ;; q i i B

E E

: et € re 8 e s rf iE

; :+=

; ry;* $q^:r E i.: : gE

E t; E

:?t;.#;€*;;1E

; ,,=,;E

= |iiE

ElE

!€ i;1:ii*iE=

+;

; E#E

;?i!

ts+ i E

e r- E2z=

zz'=! ;; i=

.* E=

&tr z d E

tEzE

' +E

FntriE

*;S=

= E

*gr3 *= * s -T

a*:, a a lt,; rEg1T

; f

=E

it:1!*i* !?!niiiig; 1z;{iFi! :i€ ;i;i3

; !:iia;fii Ei!4ig! izl!'i:5g E

if €;;jE;-

3 g?;Ti5E

;j * ;; sryE

+g5 h;€e et*;H

fE€ ;€j* I

z j+"7lt]=

iE iiii s€*E

Err€i ;;t;P

! aff;j3lz i?E

;i{€:iiilA: l+

ii11ifi lt1l* E$f }x=

Ee

i ; EZ

i:"t'= y3 ;-=

i*?7: r??= i4C

if : $ii 5"+*?23

I i;igs+;i; \i+

Et€j zzzi:.

EE

E s F

iiiE::i'

' **j;isiErE

+i€iE

lE "E

iE:i iffE

1+ligE

$aE Il

E ?E

E;;=

iAag:=

*EE

€; 'ri==

i+E

i i;:€+ai gisi; f :

REFLECTIONS ON LANGUAGE

may well exclude other successful theories as unintelligible to humans. Some theories might simply not be among the admissible hypotheses determined by the specific properties of mind that adapt us "to imagining correct theories of some kinds," though these theories might be accessible to a differently organized intelligence. Or these theories might be so remote in an accessibility ordering of admissible hy­potheses that they cannot be constructed under actual em­pirical conditions, though for a differently structured mind they might be easily accessible.

If S regards humans as part of the natural world, such speculations will seem by no means strange or incompre­hensible to him, and he might in fact attempt to investigate them and establish specific conclusions about these matters through scientific inquiry. In pursuing this effort, he would note that while man's mind is no doubt adapted to his re­quirements, there is no reason to suppose that discovery of scientific theories in particular domains is among the re­quirements met through natural selection. S might go on to develop a theory of problems and mysteries for the human organism. There would be problems in domains where ad­missible (or readily accessible) hypotheses are close to cor­rect, and mysteries elsewhere-for this organism. To take a specific case, consider the question of causation of be­havior discussed briefly in chapter 1. It might be that our inability to deal with this question reflects a temporary con­dition of ignorance, a defect that can be overcome, in prin­ciple, as science progresses. But S might discover that this optimistic view is incorrect and that the human mind is in­herently incapable of developing scientific understanding of the processes by which it itself functions in certain domains. Kant suggested that the "schematism of our understanding, in its application to appearances and their mere form, is an art concealed in the depths of the human soul, whose real modes of activity nature is hardly likely ever to allow uS to discover, and to have open to our gaze" (trans. Kemp,

Problems and Mysteries 157

1958). Perhaps this is true, in some. respects. at le~st. !here is nothing contradictory in the belIef that lllvesbgatlOn of the inherent intellectual capacities of a speCific biological

r'(1anism humans might lead S-even if S is human him-o b' , . .

self-to a scientific demonstration that some possIble SCI-ences lie beyond human grasp, perhaps the science of cau­sation of behavior among them. I am not urging this conclusion, but merely noting that it is not ruled out a priori.

Returning to the more comfortable matter of problems that do not seem to be mysteries, let us suppose that S has now established the basis for the fundamental distinction between grammar and common sense, on the one hand, and knowledge of phYSiCS, on the other. Though the latter, too, is derived on the basis of specific properties of mind, it does not reflect these properties in the same way as language and common sense do. Hence the vast qualitative difference in relative accessibility. .

S could not, I believe, determine that common sense and grammar are qualitatively different in this respect from physics merely by inspecting the three cognitive structures that he has been led to attribute to humans within his com­prehensive theory of their attained intellectual organization. If our present beliefs are near correct, the grammar of lan­guage is a highly intricate system of rules and principles. There is no absolute sense of the notion "simplicity" in terms of which grammar is "simpler" than, say, atomic phYSiCS, and although common sense has not been investigated in a comparable way, much the same may well be true in this case. The qualitative differences that S would discover in his investigation of humans doubtless reflect the structure of the mind as a contingent biological system. This is the

- only rational conclusion, I believe. This Peircean view of acquisition of cognitive structures

should not be strange to physiological psychologists, at least. (Cf. chapter 1, pp. 8-9.) It might be suggested that

$ =i=

,+=

: s- ' =

=::3+

:i?,8"7:EE

Ei it::5i E

iEE

i'^72; ? * i;;; 842 ia; E

F; lE

n q; : g; ; H;u *s LE

>-? 3 j s:; *g; : I iE

;EE

:*= u; =

.: t= i! fr"::::; E

i;; zi€," : i1 E

+ i*Z

ii i?jiliilT"zi

E g il i i i-;Z

+ :{ i i

Fi f e=

-EiZ

E;?E

= e gZ

="rrrg:rE

=*: A

s iE!;,i=

A;

f izi:; ai 3* F; iE

\tz\; i r* iEE

E ;fF

i tii?*E j

s = :=

i'z!,8E ii"leii|i;:a; iu5 +

;*$: *E€ d.E

i.Ep # F

ft ie F

i ii'rE

:EE

ir ifi i?? ;'c1= : i:;€?l -zE

a,i: ;€ ;ii=

r i - =E

E=

t,=i"?:.8?8,--=

r *;z:: rE

?l;?.Ei=

;t,zE

=_a ;i i tE

r : c Yu,+

i iiZZ

=iE

Z* : : tE

;?Ez: T

:; i I v:'; aa=

ia:E E

E it7**:E

=:-r E

= ?,t ; ;'-;3 :- >

i711?=

zLs-iEA

t € =z:E

zi E i;

AE

{=, E

E E

EE

; E i

E 7=

:z

i/€iZ

z:,.i=;i=

=E

=:i q=

* =E

iEA

i =:=

;::=i-

€i tz i" ta; ErE

t,iE";!

= ats iz7; i ;g* ;i2;-

:Etr? s ';; F

: ,i=E

'iiiTav,aE

=>

? ; j!iZit:t

; !.2:,+z 'i!; ; l: lii+

€1i:il::g€ Eiiiitl*

i ZZ

Z+

; zi=i *iie +

zi;iiiisi;i; :+;7$iiz

, | =

:==

=+

ii i:=:

r :::*ZzzE

zi ;i =

Ei::;=

ii +

??i: i:++

EF

:";lE=

iLE=

':L!=ti =

-.z1 iE=

!: i i:c:" iis l;;E

ii:: S

t:rZE

; i i fi ::; :bil-=i ?: :Z

: ?€;r:s ;i7;=:;r

=" *E

:i fg . - lEif'E

E

=: ;

*=-=

7= E

=z!:;i;T

i:=r!1t1t-tE

i-: ;;* 2'.E:r=

ii=i

,p =e i :. i i t l: ; ; Z

Z4E

:*i:lj E;; iZ

EE

:Zi i;:

; =

:Eii ii::Z

=i*i tZ

,a=z f ; +

€ =E

= +

!E: i;

= -

jEi i:,

- : r?.e .,sf; jg E=

A ;i":E

:EE

* ;=E

l; iis1=

t="Z

e .?F.i_-p; 3f E

yi lg=i

€ 4;p*"ii I ==

A E

i? =

,= :€*E

= i =

,t fr Ei e+

+?€ :ig'j F

€ .i; g 5"= E

i.TA

:g€

'trl. .:.:t: :,1.:

REFLECTIONS ON LANGUAGE

much of learning theory has been investigatin a an artifa .t " " b c_ unnatural learning under experimental conditions dC'"ised

so as to be outside the organism's inherent calJ(lcities c( , )11-

~itions which thus provide smooth learning curves and the lIke, but perhaps tell us very little about the organism' stl.ldied.l~ s

. :;Yhe~l a learning theory is formulated with sufficient pre- " clSlOn, It becomes possible to ask whether it is capable in principle of accounting for the attainment of a state of in­tellectual organization or a cognitive structure that we have reason to postulate for the mature organism. Suppose, for example, that a specific learning theory can be dernoll­strated to have the following property: an otherwise un­struc:tured system that can be modified in accordance with the mechanisms of this theory can approach, in the limit, any finite-state device that produces strings left to right as it shifts from state to state, but nothing other than such a device. Since it is well known that even the syntax of ex­tremely simple systems (e.g., propositional calculus) cannot be represented by such a device and that the syntax of language surely cannot, we can conclude at once that the learning theory is inadequate as a theory of lanauage learn­ing.

13 Thus a theory which predicts c~nverge~ce towards

demonstrably inadequate systems must obviously be re­jected as a theory of the actual attainment of systems that are far richer. In this case, then, S could go beyond the conclusion that general learning theories of known sorts are completely implausible.

In fact, the scientist S, if indeed unencumbered by our intellectual tradition, would, I think, be unlikely ever to have considered the notion that there is a "theory of learn­ing" in any interesting sense. Thus, suppose that there is a general theory of learning that applies to rats as well as humans, and that humans differ from rats simply in that they make use of its mechanisms more rapidly, fully, and effectively, and can thus attain more complex states by

Problems a11d M ysteries 159

jl1eans of the devices postulated in this theory. \Ve would then conclude that humans must be as far superior to rats jn maze-running ability as they are in ability to acquire language. But this is grossly false . Similar observations would lead a rational scientist to conclude at once that the human ability to learn language involves a special faculty or cognitive system of some sort, distinct from the cognitive system that underlies the ability to learn how to run a maze, and unavailable to rats or, as far as we know, any other organism.H For if language were simply acquired by a "gen­eralized learning strategy" that applies as well in other domains (say, maze running), we would expect other organ­isms that ar~ comparable to humans in these other domains (and thus, by assumption, use strategies similar to those employed by humans in these domains) to have comparable language-acquisition ability as well. (Cf. pp. 18, 19.)

The properties of common sense that are involved in the special human ability (if such exists) to deal with faces as compared with other geometrical figures, the properties of universal grammar, the properties that distinguish grammar and common sense from knowledge of physics, would also lead S to reject the hypothesis that there is a general learning theory, common to all organisms, undif­ierentiated in a single organism with respect to cognitive domain. These observations would lead S to the natural conclusion that the intellectual organization of a mature human is a complex integrated system that includes cog­nitive structures that are acquired on the basis of rather speCific initial adaptations. The nontrivial content of the "theory of learning" would be given by specifying these initial adaptations and the ways in which they change through maturation and experience. The steady states that are attained uniformly and rapidly should give particular insight into the nature of the organism. These steady states might well reflect quite different physiological properties and structures. There is little reason to suppose that there

160 REFLECTIONS ON LANGUAGE

are "laws of learning" of any substance and generalit \ that account for the acquisition of these complex and sp:' cHic steady states, or for the integration of cognitive struc_ tures that constitute the mature mind.

Problen.1s of acquisition of knowledge and belief have generally been investigated in a way that to a natural sci­entist might seem rather perverse. There has been little attention to the problem of characterizing "what is ' learned." Rather, certain a pTiori assumptions have been presented as to how learning takes place: principles of as­sociation, habit formation, and the like. These a priori as­sumptions have been pursued with speculative and experi­mental studies of the systems that might be acquired by these methods, with virtually no effort to establish that the systems that can be acquired are those that are acquired. A more natural approach, it seems to me, is the one sketched above: analysis of the states attained, followed by attempts to determine the nature of systems capable of attaining these states under given conditions of time and access to data, and investigation of the physical basis for these achievements, whatever it may be.

Psychologists sometimes go so far as to define their dis­cipline so as to exclude consideration of the states attained. Thus it is common to distinguish "linguistics," taken as the study of grammar, from "psychology," which is concerned with behavior and learning. l

;; To a scientist following the course outlined for S, this would seem a senseless distinc­tion. Linguistics is simply that part of psychology that is concerned with one speCific class of steady states, the cog­nitive structures that are employed in speaking and under­standing. The study of language learning is concerned with the acquisition of such cognitive structures, and the study of behavior is concerned with the ways in which they are put to use. It is self-defeating to construct a discipline that is concerned with use and attainment of some cognitive

Problems and Mysteries

Cture but that excludes consideration of the structure

stru ' itself. . .

Equally misleading, I think, is the tendel:cy 11: phllo-hical discussion to speculate on the ways 111 whIch lan­

s~~lCfe and its use might be taught. Language is not really fau~lt, for the most part. Rather, it is learned, by m~re. ex-

osure to the data. No one has been taught the prmclple ~f structure-dependence of rules, or sse, or language-spe­cific properties of such rules as NP-preposing. Nor is th~re any reason to suppose that people are taught the meanmg of words. It may be true that "teaching someone how to use an expression is the native soil from which talk a~out meaning has grown," lG but this historical comment gIVes little reason to suppose that explanations of meaning are exhausted, or even advanced, by an account of teaching. The study of how a system is learned cannot be identified with the study of how it is taught; nor can we assume that what is learned has been taught.

To consider an analogy that is perhaps not too remote, consider what happens when I tum on the ignition in my automobile. A change of state takes place. We might in­vestigate the characteristics of the new state by exa~ining fumes from the exhaust, the vibration level, the motIon of the car when I press the accelerator, and so on. A careful study of the interaction between me and the car that l~d to the attainment of this new state would not be very Il­luminating. Similarly, certain interactions between me and my child result in his learning (hence knowing) English. We can learn something about this new state in ways out­lined earlierY But a careful study of the interactions be­tween me and my child that result in his attaining this new state might give little insight into what it is that he has learned or what kind of an organism he is.

No doubt John Austin (1940) is correct in saying that when we are asked, "What is the meaning of (the word)

;i{;it?igf! €F

? iE gsl F

3E iit?; ; ! $ss; ?

I {+tg; +

; fF; giiait ii;E

ig;lj si ggil i{g

li; j E*+

itr; ieiig €;;iti; gitr :a;ii;al;iE

tqiisi ;gtisf $ i*i; j=rrE

,iif:EiiE

' lr;gTirri 5i E

tZ*i;+

E gflr :$;r:;i€; r5=

a:

q t *: ** F E

E+

385;EE

iE,:;i;f; E

€ieiiiaHi+

.'1.. -

.,, ,,,

REFLECTIONS ON LANGUAGE

rat?" we can reply in words or get the questioner to im. agine an experience or situation in which the word would or would not be used. But from this observation we cannot go on to conclude, as he does, that this description tells us all that we might reasonably want to know about the mean. ing of the word. It seems that he is relying on the implicit assumption that when we have described how we might teach, we need no longer ask what is learned. Austin is thus limiting himself to a description of turning on the ignition, whereas the model of language that he is condemning in these remarks is concerned rather to give an account of the state of the system that is activated by these manipulations.

Let us ask finally how S might go about describing the resuJts of his inquiries. Specifically, consider the much-de­bated question whether the cognitive structures S is at­tributing to the organism constitute some kind of belief or knowledge.

Consider first the case of common sense. S is attributing to his subject a system of rules and principles concerning the organization and behavior of objects. S claims that his subject differs from a rock or a bird in that this cognitive structure is an element of his final state. Since the subject is a physical organism, the system attributed to him must have a finite representation. Evidently, there are many conceivable finite representations of the system, however much we know about it, but S will not therefore conclude that there is no way to choose among these on empirical grounds. The empirical evidence is never exhausted. Fur­thermore, discoveries now unanticipated may reveal the relevance of certain types of evidence that seem now to . have no bearing on the issue. S might arrive at a general principle of organization that excludes some finite repre­sentations but not others, and he might show that acquisi­tion of cognitive structures in this and other cases can be explained on the assumption that this general principle is

Problems and Mysteries

erative as an innate schematism. Such a demonstration op ld IJrovide evidence in support of a finite representa-wOll . . . h . fi .

tllat observes the pnnclple and agall1st anot el mte tion ( .' ' . . . bI

'esentation that VIOlates It. In thIS and ll1nUmela e ~pl .

1 . "'ays S might tr)' to determine which of the vanous ot leI V> • • •

. gl'llable finite representatIOns of the cogmtlve system JIll a f

. )l 'ausible candidates for a theor)' of the final state 0 ,u e 1 ( . f his subject. On the basis of a postulat~d represen~atIOn 0

a cognitive system, a~ld other assum~trons about ll1forma­tion processing, S Will try to explam many phenom~na, such as why it is that his subject takes two presentatIOns

t be the same face but not two others. S will thus try to o 1 .

. account for the fact that his subject believes 1e is seemg the sanlP face twice. One might imagine various direct or indirect ways in which such specific beliefs and expecta­

tions could be tested. S might refer to the postulated cogniti~e structure as a

"system of beliefs." The finite representatlon postu~ated as the subject's characterization of his system of beh~fs and many beliefs that are implied by it will be uncon~cIOus, no doubt, and inaccessible to introspection. In many mstances, the subject does express beliefs. S will explain this fact ~y showing how these expressed beliefs follow from the fi~Ite representation. Suppose that the subject expresse~ belIefs that do not follow, or rejects beliefs that do, or neIther ac­cepts nor rejects beliefs that are assigned a definite status under this characterization, or acts in such a way that S must attribute to him beliefs that are inconsistent with the characterization. Then S will try to explain this result in terms of the inter~ction of other systems with the cognitive system of belief; failing in this effort, he will have to re-

. h fi . t t' 18 vise te mte represen a IOn. . ' It may be expected that conscious belIefs WIll form a

scattered and probably uninteresting subpart of the full cognitive structure. At least, if this is so, there should be

r i#::? ; s i* eFsE

tX;^ F

t**:IIafl iFtE

F €€*

?;:.; iT !:i

= T

i l{!regi;s; eiFi nE

?FsT

i iaE,4E

',;ivEzE

z $ltdi"l E: €l +

tE =

=F

;# E&

s ir *!=

z;+eii?;i: €i?ii :ii ii:iiE

aiiiEgt it-

F s H

n; i;5 E

i iE lt? fr,E

EE

i; i|,=ieta K

ii.# e sftE =

r o a I d

: : ; ; =+

*8il;ei ; ; i ;=8123* r !*re o l:f,?s E

;E; =

E:€jj " : i cE

='flt s yj!F

f;Fi;i:f ;€; U

Ei:^

€ ?;f"eif,;+,ii ;rE

.!;s iE ':*;:*q 877; E

B:, i'eE

i *3 r 1* T

g{fE E

ii-I E

EE

=5 :F

Ed";=

EE

T# ;e€ s"f 3:

-ae il* E 3*: ;.i€3E

i Ei;;*i ! i*€ E

€ ;Z=

;ipg: i:i i:* ' si:i

E ; € E

..1EE

EF

i?il I!: iil +

* if ;i;P

EJ

d,!

B:: i g: rriE

S: r E

,iE I ry; E

:-; t = z=

? i:tg H

*Hilt a a a"t+

*€:' t= E

i g $€F! **"x"- vE

z- E?

?E frE

EziA

E 2 F

^i+=

E

I s 6 HE

:€E,q ;+

E ij E

";ge€

.iE {{ i;itli;i i* IlF

; iE iiaE

tiel iil :FI lE

it:r; *l;3=€; ;t; ningi+

#+ga; E

i $Et

i ;i "iiEi;7:i=

Er!r ili: i:g;j E

ffi: ; r;E;

, -E,+

7?,;:,2;8,;iiA+

2iEE

, ;€.- s"*p;*5 i:E a3

, :: =*?; E

ityq r€E =

: j FE

;€ :=H

$gi u;s .; ;i e:i i*; i:: i+

;f I; ;t: tr;i€rilei E;:;{ if ;ei

fr: i"Y

*'t =i =

:ii:;€q; tA;lr;; +

: = tE

gi+F

if ir;

7+Z

Eu a lfr=

UJ te i s.;rc E

zitai=:g: ; ;? i;t ["

= : I't_lii:^: ; i i saiif+

+ ;.t :+

Ef €; iE

€f i : ii;€?t|i;:zi;z i=

EE

;=if !i ! E

2 iz; -: Fi eE

ir1l;.q ; iE

- =E

:+; E

=*E

! **E-.E

+ i;!i+

fg;; #€j;t?: i

REFLECTIONS ON LANGUAGE

no reason for surprise. Nor do I see any objection to S's practice in referring to the cognitive structure attributed to the subject as a system of beliefs.

Consider now the case of grammar. If S speaks English, he will say that some of his subjects have "learned French" and now "know French." Furthermore, in many specific instances, they can articulate their knowledge, as knowl­edge that so-and-so. Again, the set of such cases is unlikely to be of much interest in itself. S will attempt to explain these instances by showing how they follow from the grammar of French, interacting with other cognitive struc­hlres. In this way, he will try to account for facts analogous (for French) to those mentioned earlier (cf. (1)-(13)). The problems of confll-mation and choice of theories are anal­ogous to those that arise in the case of investigation of common sense.

Obviously, the grammar is not in itself a theory of per­formance (behavior). It is, however, proper for S to propose that the grammar is a component of such a theory and to proceed at this point to construct a theory of interaction of structures that would serve as a theory of performance for his subjects. 1D S might refer to the grammar attributed to the speaker as a representation (or model) of his knowl­edge of his language. S might also want to say that the subject who knows the language knows the grammar, and that in his initial state he knew universal grammar. Thus S's subject differs from an English speaker, a rock, or a bird in that he knows the grammar of French (to use the suggested terminology). He is like an English speaker, and different from a rock or a bird, in that in his initial state he knew universal grammar.

Since some might object to this terminology, S might prefer to invent some technical terms. Let us say that if a speaker knows the language L then he cognizes L. Further­more, he cognizes the linguistic facts that he knows (in any uncontroversial sense of "know") and he cognizes the

Problems and M ystel'ies

)rinciples and rules of his internalized grammar, both ~hose that might be brought to awareness and those that are forever hidden from consciousness. Furthermore, he cognizes the principles that underlie the acquisition of language in the first place, the principles of universal grammar (assuming that the approach o~tlined ~arlier is correct). Thus a person who knows EnglIsh cogmzes cer­tain facts, for example; that necessarily bachelors are un­married and that "is" goes with singular subjects. He also cognizes that specific rules are ordered in a certain way relative to one another. Furthermore, the person cognizes that transformations apply in a cyclic ordering and obey SSC, that initial phrase markers and surface structures con­tribute to semantic interpretation in ways described earlier, and that transformations are structure-dependent. The lat­ter examples constitute part of "innate cognization" (as­suming that the theory suggested earlier is correct).

If we decide to use the word "know" in a narrow sense, restricting it to conscious "knowledge of" or to "knowing how" ("why, who ... ," and so on) as this notion is often construed, then "knowledge of" as in "knowledge of lan­guage" will have to be explicated in terms of the new tech­nical vocabulary, so it appears. 20 In this usage, what is "known" will be a rather ill-defined and, perhaps, a scat­tered and chaotic subpart of the coherent and important systems and structures that are cognized. For psychology, the important notion will be "cognize," not "know."

Or, we might make the decision to sharpen and perhaps extend the term "know" so that it has just the properties of "cognize," thus eliminating the new terminology. Then We will be able to explain explicit knowledge of certain facts by showing how these cases are related to the system of "tacit knowledge." 21

I doubt that this question can be settled by considera­tion of "ordinary usage," which seems to me vague and in­explicit at just the crucial points. The philosophical tradi-

S .gE

I?f;i:E'i sf: i

ii 6v*E;+

Pj * F

^"i?5:-8;. 3, g5;:rg€iri H

g,i;€€j$fEi:

? l€ 'H: l t ;t

5zl; t;!i;i ri€ if :iii3,: ; s;b ::; '; E

? g"Er;i:{€ A

; EZ

', i E

;* E?E

::?a s€z rE =

:=€+

s:5.,E

aEE

; i : i;: EE

|E p :E

iE€:E

id er i . r3 sH

! =

g ; E gti;fiE

;iEgi E

Ei*E

b i'! t:iii ai:€F:

E - "a E

?E r;sE

g ei ;; ij$E*+

;s ;?* t Bsv?-e

i =

L€ *4: ; = 7*y ? +

;= ilE

6;; t g; i i it "f€r EE

|E",

i iiiigZV

fzlilii nz{:; 6?i }i*i ;i ii !i€:=i

+ i: ;1: I 14rs tli

+ E

{i zt7+7 z V

z 5 = e-E

T: ; ;E

=. n7 f'tE

:*Eg; ju,i-a';

= 1#E

TiT

rErltF

,- rf t IZZ

SzE

-::?:,i g i* € = i:+

*+zr; ! ; i?=

'! a ; i+; E

?1 E E

E g?€ i ?E

i3 : E E

-.EE

E iE

ZA

ii}EE

id 4,E f E

i 5E ?E

€g:;Ei sE

E*:E

E€*e:

i

.{iif F

i 3+

b:.ts,'!5-o" *$'l

= j;ig,,:; 5F

t€g!+€i,fli *j;

; ;F;:;i ;;i:;!j;::i ;t;

i rt7;ji tg:jryi=

q=oF

eF:

: l;;lE: !j'*:'=

=.9ijt5{ f+

5i : s 3ieb ;:ggj ;;tijs €;iF

'=

-

" u

.: -:

=

=

" c, *

=

i".; ^.'-

, =

|

I E;i i;i ;+

:;iijg;;; : a;; ;;?t;: ji;E

; g eF

! r i ;:;:t:!i,1 r;f; iE

iT r;;, =

;*i ;3+

;€: ,! 1i;; Ei;;: *;; E

:;

ji=€€€i!-ti,=

*ig;: si g !'-;v€; g g;F

fgf;IF3F

IfJs:5; E

tr;::l fE;

166 REFLECTIONS ON LANGUAGE

tion is varied. Leibniz, for kId one, spoke of l "now e ge, though he seems to ha . mconsciollS

edge as accessible to co ' . eve regarded all know1_ . nsclOusness, Hume d "b d

stmcts as those parts of .' '" eSCll e in-"d . an ammal s knowl'da" 1

enve[s] from the original hand f t, ~ ~e t lat it to the "parts of knowledge" tl t ,

0 1 na 1Il e, 111 contrast

tion. la It earns from observa_

It seems to me that the " . systems of knowledge and b 11).llfn~lples that determine OUr . e Ie mteract so co 1 t 1 mseparably with "our knowled e" . , mp e e y and term, that it would be diffi Igt' Idn anyone s sense of this

cu to eveIop aI' count that \vould l'soIat "t, k . co leI ent ac-

e e lUe nowled . " H ' . unclear that more than te' . 1 . ge. owever, It IS S might choose to aba dllmnol ogy IS at stake here. Thus

" ' n on t le terms "kno 1 d " even knowledge of langua e" (if ' w e f?e and while noting that the1'e 1', I~ttl s.ome find that offensive),

s 1 e walTa t· ,d' for these decisions If so 1 '11 . n m 01 mary usage

. . ", le WI speak of acn . " l11zmg, and competence, instead of 1 .' k'l Ulll~lg, cog­knowledge. eammg, nowmg, and

As long as We are clear about what we a' d . . approach seems to me quite all right "p, l~d dOIng, eIther about the thina it i - d . IOVI e we agree (Hume). b' s nee less to dispute about the terms"

that have been [ capnot survey ases that I think .e question that tamely, whether these accounts. at we can spec-

1 formal system y that [he] has I." 22 This obser­Dry is ever "im­(artz apparently m. He suggests

Problems and Mysteries

the following example to illustrate the "messy issue" that he thinks arises. Suppose a certain device, D, labels spheres "+" if their density is greater than 1 and "-" if their density is less. Suppose that a system of equations E is proposed, involving relations between volume, density, weight, and so on, which describes the output of D. But, Schwartz notes, D "might not employ a set of principles anything like [E]." Thus it might never consider weight, volume, and so on, but might contain a liquid of density 1 and label" +" any sphere that sinks, and" -" any sphere that floats. "\Vould it be reasonable to claim that our equa­tions [E] are internally represented in this machine?" His answer is that it would not, even though "in some sense the liquid in the machine could be held to 'stand for' the equa­tions [E]."

Suppose that S, observing D, concluded that its "cogni­tive state" involves calculations of the sort expressed in E. Further investigation might convince S that this conclusion is wrong and that a very different principle is employed, namely, the one Schwartz suggests. It would not, of course, be "reasonable" to persist in maintaining that D "employ[s] a set of principles anything like [E]," since this conclusion has been refuted. The fact that E continues to describe input-output relations correctly is uninteresting to S, once he has discovered the actual principles employed. There is no "messy issue" here. ,

Of course, if S were simply satisfied to say that E de­scribes input-output relations, he would not have pro­ceeded with further investigation to determine whether D actually employs principles like E. But as a scientist, S would have been interested to discover what it is about D that makes E an accurate description. To determine this, he might put forth the working hypothesis that D actually calculates in accordance with E, or that E enters in some important way into the actual performance of D. And he would then seek evidence bearing on this hypothesis. He