23 November 1979, Volume 206, Number 4421

Can an Ape Create a Sentence?H. S. Terrace, L. A. Petitto, R. J. Sanders, T. G. Bever

The innovative studies of the Gardners(1-3) and Premack (4-6) show that achimpanzee (Pan troglodytes) can learnsubstantial vocabularies of "words" ofvisual languages. The Gardners taughtWashoe, an infant female chimpanzee,signs of American Sign Language (ASL)(7, 8). Premack taught Sarah, a juvenilefemale, an artificial language of plastic

song when asserting territory. Such ri-gidity is typical of the communicative be-havior of other genera, for example, beescommunicating about the location andquality of food or sticklebacks engagingin courtship behavior (14).Human language is most distinctive

because of a second level of structurethat subsumes the word-the sentence

Summary. More than 19,000 multisign utterances of an infant chimpanzee (Nim)were analyzed for syntactic and semantic regularities. Lexical regularities were ob-served in the case of two-sign combinations: particular signs (for example, more)tended to occur in a particular position. These regularities could not be attributed tomemorization or to position habits, suggesting that they were structurally constrained.That conclusion, however, was invalidated by videotape analyses, which showed thatmost of Nim's utterances were prompted by his teacher's prior utterance, and thatNim interrupted his teachers to a much larger extent than a child interrupts an adult'sspeech. Signed utterances of other apes (as shown on films) revealed similar non-human patterns of discourse.

chips of different colors and shapes. In arelated study, Rumbaugh et al. (9) taughtLana, also a juvenile chimpanzee, to useYerkish, an artificial visual language.These and other studies (10), one ofwhich reports the acquisition of morethan 400 signs of ASL by a female gorillanamed Koko (11), show that the shiftfrom a vocal to a visual medium cancompensate effectively for an ape's in-ability to articulate many sounds (12).That limitation alone might account forearlier failures to teach chimpanzees tocommunicate with spoken words (13).Human language makes use of two

levels of structure: the word and the sen-tence. The meaning of a word is arbi-trary. This is in contrast to the fixedcharacter of various forms of animalcommunication. Many bird species, forexample, sing one song when in distress,one song when courting a mate, and one

SCIENCE, VOL. 206, 23 NOVEMBER 1979

SCIENCE

quences produced and understood bytheir pongid subjects were governed bygrammatical rules. The Gardners, for ex-ample, note that "The most significantresults of Project Washoe were thosebased on comparisons between Washoeand children, as . . . in the use of orderin early sentences" (3, p. 73).

If an ape can truly create a sentencethere would be a reason for asserting, asPatterson (11) has, that "language is nolonger the exclusive domain of man."The purpose of this article is to evaluatethat assertion. We do so by summarizingthe main features of a large body of datathat we have collected from a chim-panzee exposed to sign language duringits first 4 years. A major component ofthese data is the first corpus of the multi-sign utterances of an ape. Superficially,many of its utterances seem like sen-tences. However, objective analyses ofour data, as well as of those obtained byother studies, yielded no evidence o an

........... ....

ape's ability to use a grammar. Each in-stance of presumed grammatical compe-tence could be explained adequately bysimple nonlinguistic processes.

Project Nim

Our subject was a male chimpanzee,Neam Chimpsky (Nim for short) (16, 17).Since the age of 2 weeks, Nim was raisedin a home environment by human surro-gate parents and teachers who communi-cated with him and amongst themselvesin ASL (7,8). Nim was trained to sign bya method modeled after the techniquesthat the Gardners (2) and Fouts (18) havereferred to as molding and guidance. Ourmethods of data collection paralleledthose used in studies of the developmentof language in children (19-24). Duringtheir sessions with Nim, his teacherswhispered into a miniature cassette re-corder what he signed and whether his

H. S. Terrace is a professor of psychology at Co-lumbia University, 418 Schermerhorn Hall, Colum-bia University, New York 10027. L. A. Petitto is agraduate student in the Department of Human De-velopment at Harvard University, Cambridge, Mas-sachusetts 01238. R. J. Sanders is a graduate studentin the Department of Psychology at Columbia Uni-versity and a visiting instructor at the State Univer-sity ofNew York in Utica. T. G. Bever is a professorof psychology and linguistics at Columbia Universi-ty.

(15). A sentence characteristically ex-presses a complete semantic propositionthrough a set of words and phrases, eachbearing particular grammatical relationsto one another (such as actor, action, ob-ject). Unlike words, most sentences can-not be learned individually. Psycholo-gists, psycholinguists, and linguists arein general agreement that using a humanlanguage indicates knowledge of a gram-mar. How else can one account for achild's ultimate ability to create an 'in-determinate number of meaningful sen-tences from a finite number of words?Recent demonstrations that chim-

panzees and gorillas can communicatewith humans via arbitrary "words" posea controversial question: Is the ability tocreate and understand sentences unique-ly human? The Gardners (1, 3), Premack(6), Rumbaugh (9), and Patterson (11)have each proposed that the symbol se-

0036-8075/79/1123-0891$02.00/0 Copyright C 1979 AAAS 899

on

Nov

embe

r 15

, 201

3w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n N

ovem

ber

15, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Nov

embe

r 15

, 201

3w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n N

ovem

ber

15, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Nov

embe

r 15

, 201

3w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n N

ovem

ber

15, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Nov

embe

r 15

, 201

3w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n N

ovem

ber

15, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Nov

embe

r 15

, 201

3w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n N

ovem

ber

15, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

on

Nov

embe

r 15

, 201

3w

ww

.sci

ence

mag

.org

Dow

nloa

ded

from

o

n N

ovem

ber

15, 2

013

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

signs were spontaneous, prompted, Table 1. Number of tokens and types of com-molded, or approximations of the correct binations containing two, three, four, and fivesign (25). or more signs.Nim satisfied our criterion of acquiring Length of Tokens Types

a sign when (i) on different occasions, combinationthree independent observers reported its Two signs 11,845 1,138spontaneous occurrence and (ii) it oc- Three signs 4,294 1,660curred spontaneously on each of five Four signs 1,587 1,159successive days. By spontaneously we Five or more signs 1,487 1,278mean that Nim signed the sign in an ap-propriate context and without the aid ofmolding, prompting, or modeling on the stances of more tickle, the conventionalpart of the teacher. As of 25 September English juxtaposition of these signs. Ac-1977, Nim had acquired 125 signs (26). cordingly, there is no basis for deciding

Combinations of Signs

The Gardners' analyses of Washoe'ssign combinations prevents one fromstudying their grammatical structure.With but two minor exceptions, theGardners did not report the order ofsigns of Washoe's multisign combina-tions (27). For example, more tickle andtickle more were both reported as in-

whether Washoe's multisign combina-tions obeyed rules of sign order (28). Onecould conclude that Washoe had learnedthat both more and tickle were appropri-ate ways of requesting that tickling reoc-cur and that when Washoe signed bothsigns it was because of her prior trainingto sign each sign separately.We defined a combination of signs as

the occurrence of two or more differentsigns that were not interrupted by the oc-currence of other behavior or by the re-

turn of the hands to a relaxed position(29). Of Nim's combinations, 95 percentconsisted of sequences of distinct signsthat occurred successively. These are re-ferred to as "liiear sequences." Twoother kinds of combinations were not in-cluded in the corpus: contractions of twoor more signs and simultaneous combi-nations in which two distinct signs oc-curred at the same time. Even thoughsuch combinations can occur in ASL,they were excluded from our corpus be-cause it was impossible to specify thetemporal order of the signs they con-tained. Figure 1 shows a typical linearcombination, me hug cat, in which thereis no temporal overlap between any ofthe signs.

In no instance were specific se-quences, contractions, or simultaneouscombinations reihforced differentially.Indeed, Nim was never required to makea combination of signs as opposed to asingle sign. However, Nim's teachers of-ten signed to him in stereotyped ordersmodeled after English usage, and they

Fig. 1. Nim signing the linear combination, me hug cat to his teacher (Susan Quinby). (Photographed in classroom by H. S. Terrace.)

I

SCIENCE, VOL. 206892

may also have unwittingly given him spe-cial praise when he signed an interestingcombination. Such unintentional reac-tions do not, however, appear to differfrom the reactions parents exhibit whentheir child produces an interesting utter-ance or one that conforms to correctEnglish.Nim's linear combinations were sub-

jected to three analyses. First, we lookedfor distributional regularities in Nim'stwo-sign utterances: did Nim place par-ticular signs in the first or the second po-sition of two-sign combinations? Sec-ond, having established that lexicalregularities did exist in two-sign com-binations, we looked for semantic re-lationships in a smaller corpus of two-sign combinations for which we had ade-quate contextual information. The re-sults of these analyses were equivocal. Athird, "discourse," analysis of videotapetranscripts shows that Nim's signs wereoften prompted by his teacher's priorsigns.Corpus and distributional regularities.

From Nim's 18th to 35th month histeachers entered in their reports 5235types of 19,203 tokens of linear combina-tions of two to five or more signs. Dif-ferent sequences of the same signs wereregarded as different types (for example,banana eat or eat banana). The numberof types and tokens of each length ofcombination (Table 1) in each case grewlinearly (30, 31).The sheer variety of types of combina-

tions and the fact that Nim was not re-quired to combine signs suffices to showthat Nim's combinations were not vlearned by rote. The occurrence of morethan 2700 types of combinations of two-and three-sign combinations wouldstrain the capacity of any known esti-mate of a chimpanzee's memory. As wasmentioned earlier, however, a large vari-ety of combinations is not sufficient todemonstrate that such combinations aresentences; that is, that they express a se-mantic proposition in a rule-governed se-,-quence of signs. In the absence of addi-tional evidence, the simplest explanationof Nim's utterances is that they are un-structured combinations of signs, inwhich each sign is separately appropriateto the situation at hand.The regularity of Nim's combinations

suggest that they were generated byrules and was most pronounced in thecase of two-sign combinations. Asshown in Table 2, more + X is more fre-quent than X + more, give + X is morefrequent than X + give, and verb + meor Nim is more frequent than me orNim + verb. An example of the regular-ities in Nim's two-sign combinations,23 NOVEMBER 1979

consisting of all transitive verbs com-bined with all references to himself (meor Nim), is shown in Table 3 (32). Thenumber of tokens with the verb in thefirst position substantially exceeds thereverse order. Also, Nim combined tran-sitive verbs as readily with Nim as withme (33). Nim's preference for using meand Nim in the second position of two-sign combinations was also evident inrequests for various ingestible and non-ingestible objects (Table 2).

Different frequency patterns, such asthose shown in Tables 2 and 3, are notsufficient to demonstrate that Nim's se-quences are constrained structurally.Nim could have a set of independentfirst- and second-position habits thatgenerated the distributional regularities

we observed. A conservative inter-pretation of these regularities, one thatdoes not require the postulation of syn-tactic rules, would hold that Nim usedcertain categories as relatively initial orfinal irrespective of the context in whichthey occur. If this were true, it should bepossible to predict the observed frequen-cy of different constructions, such asverb + me or verb + Nim, from the rel-ative frequency of their constituents inthe initial and final positions.The accuracy of such predictions was

tested by allocating each sign of a two-sign sequence to a lexical category andthen using the relative frequencies ofthese lexical categories to predict theprobabilities of each two-sign lexicaltype. The predicted value of the proba-

Table 2. Frequency of particular signs in first and second positions of two-sign combinations.

Combination Types Tokens

Transitive verb

47275124

+ X+ more+ X+ giveme

+ orNim

+ Transitive verb

Noun (food/drink)

97412427177

78825

19

me+ orNim

+ noun (food/drink)

Noun (nonfood/drink)me

+ orNim

+ Noun (nonfood/drink)

158

34

26

35

26

775

261

181

99Nim

Table 3. Two-sign combinations containing me or Nim and transitive verbs [V(t)].

V(t) + me V(t) + Nim me + V(t) Nim + V(t)

Types Tokens Types Tokens Types Tokens Types Tokens

bite me 3 bite Nim 2 me bite 2break me 2brush me 35 brush Nim 13 me brush 9 Nim brush 4clean me 2 clean Nim 1 me clean 2

me cook Idraw Nim I

finish me I finish Nim 7 Nim finish Igive me 41 give Nim 23 me give 11 Nim give 4

Nim go 4groom me 21 groom Nim 6 Nim groom 1help me 6 help Nim 4 me help 2hug me 74 hug Nim 106 me hug 40 Nim hug 23kiss me 1 kiss Nim 6 me kiss 1 Nim kiss 2open me 13 open Nim 6 me open 10 Nim open 5

pull Nim 1tickle me 316 tickle Nim 107 me tickle 20 Nim tickle 16

515 283 98 60

Total types: 25 Total types: 19Total tokens: 788 Total tokens: 158

893

Table 4. Twenty-five most frequent two- and three-sign combinations.

Two-sign Fre- Three-sign Fre-combinations quency combinations quency

play me 375 play me Nim 81me Nim 328 eat me Nim 48tickle me 316 eat Nim eat 46eat Nim 302 tickle me Nim 44more eat 287 grape eat Nim 37me eat 237 banana Nim eat 33Nim eat 209 Nim me eat 27finish hug 187 banana eat Nim 26drink Nim 143 eat me eat 22more tickle 136 me Nim eat 21sorry hug 123 hug me Nim 20tickle Nim 107 yogurt Nim eat 20hug Nim 106 me more eat 19more drink 99 more eat Nim 19eat drink 98 finish hug Nim 18banana me 97 banana me eat 17Nim me 89 Nim eat Nim 17sweet Nim 85 tickle me tickle 17me play 81 apple me eat 15gun eat 79 eat Nim me 15tea drink 77 give me eat 15grape eat 74 nut Nim nut 15hug me 74 drink me Nim 14banana Nim 73 hug me hug 14in pants 70 sweet Nim sweet 14

bility of a particular sequence was calcu-lated by multiplying the probabilities ofthe relevant lexical types appearing inthe first and second positions, respec-tively. In predicting the probability ofmeeat, for example, the probability ofme inthe first position (.121) was multiplied bythe probability of eat in the second posi-tion (.149), yielding a predicted relativefrequency of .016.The correlation between 124 pairs of

predicted and observed probabilities was.0036. It seems reasonable to concludethat, overall, Nini>s two-sign sequencesare not form ent vositionhabits. Furthermore, it is not possible topredict the observed relative positionfrequencies of lexical types of three-signcombinations from the relative frequen-cies of their constituents. The correla-tion between the 66 pairs of predictedand observed probabilities was only .05.

Relation between Nim's two-, three-andfour-sign combinations. As childrenincrease the length of their utterances,they elaborate their initially short utter-ances to provide additional informationabout some topic (20, 22). For example,instead of saying, sit chair, the childmight say, sit daddy chair. In general, itis possible to characterize long utter-ances as a composite of shorter constitu-ents that were mastered separately.Longer utterances are not, however,simple combinations of short utterances.In making longer utterances, the childcombines words in short utterances injust one order; he deletes repeated ele-ments, and he treats shorter utterances

894

as units when they are used to expandwhat was expressed previously by asingle word.The apparent topic of Nim's three-sign

combinations overlapped considerablywith the apparent topic of his two-signcombinations (Table 4). Eighteen ofNim's 25 most frequent two-sign combi-nations can be seen in his 25 most fre-quent three-sign combinations, in virtu-ally the same order in which they appearin his two-sign combinations. Further-more, if one ignores sign order, all butfive signs that appear in Nim's 25 most

Table 5. Most frequent four-sign combina-tions.

Four-sign combinations Fre-quency

eat drink eat drink 15eat Nim eat Nim 7banana Nim banana Nim 5drink Nim drink Nim 5banana eat me Nim 4banana me eat banana 4banana me Nim me 4grape eat Nim eat 4Nim eat Nim eat 4play me Nim play 4drink eat drink eat 3drink eat me Nim 3eat grape eat Nim 3eat me Nim drink 3grape eat me Nim 3me eat drink more 3me eat me eat 3me gum me gum 3me Nim eat me 3Nim me Nim me 3tickle me Nim play 3

frequent two-sign combinations (gum,tea, sorry, in, andpants) appear in his 25most frequent three-sign combinations.We did not have enough contextual in-formation to perform a semantic analysisof Nim's two- and three-sign combina-tions. However, Nim's teachers' reportsindicate that the individual signs of hiscombinations were appropriate to theircontext and that equivalent two- andthree-sign combinations occurred in thesame context.Though lexically similar to two-sign

combinations, the three-sign combina-tions (Table 4) do not appear to be in-formative elaborations of two-sign com-binations. Consider, for example, Nim'smost frequent two- and three-sign com-binations: play me and play me Nim.Combining Nim with play me to producethe three-sign combination, play meNim, adds a redtndant proper noun to apersonal pronoun. Repetition is anothercharacteristic of Nim's three-sign combi-nations, for example, eat Nim eat, andnut Nim nut. In producing a three-signcombination, it appears as if Nim is add-ing to what he might sign in a two-signcombination, not so much to add new in-formation but instead to add emphasis.Nim's most frequent four-sign combina-tions (Table 5) reveal a similar picture. Inchildren's utterances, in contrast, therepetition of a word, or a sequence ofwords, is a rare event (34).

Differences Between Nim's and aChild's Utterances

The fact that Nim's long utteranceswere not semantic or syntactic elabora-tions of his short utterances defines amajor difference between Nim's initialmultiword utterances and those of a

/ child. These and other differences in-dicate that Nim's general use of combi-nations bears only a superficial similarityto a child's early utterances (35-38).The mean length of Nim's utterances.

As the mean length of a child's utter-ances (MLU) increases, their complexityalso progressively increases (20-22). InEnglish, for example, subject-verb andverb-object construction merge into sub-ject-verb-object constructions.

Figure 2 shows Nim's MLU (the meannumber of signs in each utterance) be-tween the ages of 26 and 45 months (39).The most striking aspect of these func-tions is the lack of growth of Nim's MLUduring a 19-month period. Figure 2 alsoshows comparable MLU functions ob-tained from hearing (speaking) and deaf(signing) children (40), including thesmallest normal growth of MLU of a

SCIENCE, VOL. 206

speaking child that we could locate. Allchildren start at an MLU similar toNim's at 26 months, but, unlike Nim, thechildren all show increases in MLU.

Another difference between Nim's andchildrens' MLU has to do with the valueof the MLU and its upper bound. Ac-cording to Brown, ". . . the upper boundof the (MLU) distribution is very reliablyrelated to the mean.... At MLU = 2.0the upper bound will be, most liberally,5 ± 2" (41). Nevertheless, with an MLUof 1.6 Nim made utterances containingas many as 16 signs (give orange me giveeat orange me eat orange give me eat or-ange give me you). In our discourse anal-yses of Nim's and Washoe's signing (seebelow), we suggest mechanisms that canlengthen an ape's utterance but that donot presuppose an increase in se anticor syntactic competence. \/)Semantic-reraie s hips expressed in

Nim's two-sign combinations. Semanticdistributions, unlike the lexical ones wediscussed above, cannot be constructeddirectly from a corpus. In order to derivea semantic distribution, observers haveto make judgments as to what each com-bination means. Procedures for making

4.4

4.0

3.6 -

3.2

2.8H

2.4

2.0 F

1.6

1.2

Children:Hearing Deaf

A Eve o--o Ruth* 'Sarah *--- Pola

4

4.1/.

A~

& .x

such judgments, introduced by Bloom(19, 20) and Schlesinger (42), are knownas the method of "rich interpretation"(21-23, 42). An observer relates the utter-ance's immediate context to its contents.Supporting evidence for semantic judg-ments includes the following observa-tions. The child's choice of word order isusually the same as it would be if the ideawere being expressed in the canonical

( adult form. As the child's MLU increas-es, semantic relationships identified by arich interpretation develop in an orderlyfashion (20, 22, 43). The relationships ex-pressed in two-word combinations arethe first ones to appear in the three- andfour-word combinations. Many longerutterances appear to be composites ofthe semantic relationships expressed inshorter utterances (20, 22).

Studies of an ape's ability to expresssemantic relationships in combinations'of signs have yet to advance beyond thestage of unvalidated interpretation. TheGardners (44) and Patterson (11) con-cluded that a substantial portion ofWashoe's and Koko's two-sign combina-tions were interpretable in categoriessimilar to those used to describe two-

Nim:

*-* Classroom sessionsx---x Home sessionsD--O Videotape samples

...----

/

~~~//z ~~~//.. , /

/ n/ O P

0

II

II

II

I

0

o~~~~~~E E s s t l~ ~ -6 \ It El l

I 0I I I I 40 4 4 51

20 24 28 32 36 40 44 48 52

0.14

0.10

0.06

word utterances of children (78 and 95percent, respectively). No data are avail-

,Jable as to the reliability of the inter-pretations that the Gardners and Patter-son have advanced.A widely cited example of Washoe's

ability to create new meanings throughnovel combinations of her signs is her ut-terance, water bird. Fouts (45) reportedthat Washoe signed water bird in thepresence of a swan when she was askedwhat that? Washoe's answer seemsmeaningful and creative in that it juxta-poses two appropriate signs in a mannerconsistent with English word order.Nevertheless, there is no basis for con-cluding that Washoe was characterizingthe swan as a "bird that inhabits water."Washoe had a long history of beingasked what that? in the presence of ob-jects such as birds and bodies of water.In this instance, Washoe may have sim-ply been answering the question, whatthat?, by identifying correctly a body ofwater and a bird, in that order. Beforeconcluding that Washoe was relating thesign water to the sign bird, one mustknow whether she regularly placed anadjective (water) before, or after, a noun

n 11 nth0.02

ID) UoC<0,oD Q+ 0+ Q Do'Q C

csLc w La.0 0 to

Semantic relationship

v. 0.18_

,0.14

0.10 F

0.06 F

0.02

Age (months) .; _

Fig. 2 (left). Mean length of signed utterances of Nim and three deaf children and *V @ z ;mean length of spoken utterances of two hearing children. The functions showing T O EC0Nim's MLU between January 1976 and February 1977 (age, 26 to 39 months) are 0f r

> + 3.1~0based on data obtained from teachers' reports; the function showing Nim's MLU + '- C +between February 1976 and August 1977 (age, 27 to 45 months) is based upon : +

videotranscript data. [See (39) regarding the calculation of MLU's for signed ut-terances.] Fig. 3 (right). Relative frequencies of different semantic relation-ships. The bars above I and II show to the relative frequencies of two-sign combi- Semantic relationship

nations expressing the relationship in the order specified under the bar, for example, an agent followed by an action. The bars above I show

the relative frequencies of two-sign combinations expressing the same relationship in the reverse order, for example, action followed by an agent.

0

Q

c

c

,=6- r" Cl=

23 NOVEMBER 1979 895

1.U - _ Children Nim our semantic analyses of an ape's two-Bloom, Rocissano& Hood (1976) 00'O" sign combinations nor those of any oth-

0.8 Adjacent responses---- 3 er studies have produced such evidence.

0.6 1' 8' , In some cases, utterances were inher-CL - o\ / ' ,, 'ently equivocal in our records. Accord-

0.6 j Xozo . ingly, somewhat arbitrary rules were2 * /i used to interpret these utterances. Con-

0.4 ~

* . sider, for example, combinations ofNime 0.4 /t/ r. 1 and me with an object name (for ex-/ Expansions ample, Nim banana). These occurred

A when the teacher held up an object that0.2 Imitations the teacher was about to give to Nimwho, in turn, would ingest it. We had no

0.o 2, , * 6---,-'.*. clear! basis for distinguishing between the21?6 36 26 31 36 41 following semantic interpretations ofAge (months) combinations containing Nim or me and

Fig. 4. Proportion of utterances emitted by aobject name: ngeNt-obje bn-children (left-hand function) and by Njm an object name: agent-object, ben-(right-hand functions) that are adjacent to, im- eficiary-object, and possessor-possesseditative of, or expansions of an adult's prior ut- object.terance. An equally serious problem is posed

by the very small number of lexical itemsused to express particular semantic

(bird). That cannot be decided on the roles. Only when a semantic role is rep-basis of a single anecdote, no matter how resented by a large variety of signs is itcompelling that anecdote may seem to an reasonable to attribute position prefer-English-speaking observer. ences to semantic rules rather than toWithout prejudging whether Nim ac- lexical position habits. For example, the

tually expressed semantic relationships role of recurrence was presented exclu-in his combinations, we analyzed, by the sively by more. In combinations pre-method of rich interpretation, 1262 of his sumed to relate an agent and an object ortwo-sign combinations, which occurred an object and a beneficiary, one wouldbetween the ages of 25 to 31 months (46). expect agents and beneficiaries to be ex-The results of our semantic analysis are pressed by a broad range of agents andshown in Fig. 3. Twenty categories of se- beneficiaries, for example: Nim, me,mantic relationships account for 895 (85 you, and names of other animate'beings.percent) of the 957 interpretable two-sign However, 99 percent (N = 297) of thecombinations. Brown (47) found that beneficiaries in utterances judged to bethere were 11 semantic relationships that object-beneficiary combinations wereaccount for about 75 percent of all com- Nim and me, and 76 percent (N = 35) ofbinations of the children he studied. Sim- the agents in'utterances judged to beilar categories of semantic relationships agent-object combinations were you.were used by the G'ardners and by Pat- From these and other examples, it is dif-terson (48). ficult to decide whether the positional

Figure 3 shows several instances of regularities favoring agent-object and ob-significant preferences for placing signs ject-beneficiary constructions (Fig. 3)expressing a particular semantic role in are expressions of semantic relationshipseither the first or the second positions. or idiosyncratic lexical position habits.Agent, attribute, and recurrence (more) Such isolated effects could also be ex-were expressed most frequently in the pected to appear from statistically ran-first position, while place and beneficiary dom variation.roles were expressed most frequently by Discourse analysis. An analysis ofsecond-position signs (49). video transcripts revealed yet anotherAt first glance, the results of our se- contribution to the semantic look of

mantic analysis appear to be consistent Nim's combinations; his utterances werewith the observations of the Gardners often initiated by his teacher's signingand Patterson. But even though our judg- and they were often full or partial imita-mnents were reliable, several features of tions of his teachers' preceding utter-our results suggest that our analysis, and ance. Since full or partial imitations werethat of others, may exaggerate Nim's se- included in the corpus, it is possible thatmantic competence. One problem is the the semantic relationships and positionsubjective nature of semantic inter- preferences we observed are, to somepretations. That problem can be reme- extent, reflections of teachers' signingdied only to the extent that evidence cor- habits. Those that were imitated cannotroborating the psychological reality of be regarded as comparable to a child'sour interpretations is available. Neither nonimitative constructions.

896

Table 6. Discourse between Washoe (W) andB. Gardner (B.G.). See Fig. 5. This is a tran-script of a tape shown on television.

Time Frame

00.00 7 B.G:Iwhat1.46 8 time1.96 9 W: Itime2.25 10 now?! eatl4.50 11 Itimel4.84 12 eat!

--------- splice ---------

There has been increasing interest inthe way parents speak to their children(50) and in the ways children adjust theirspeech to aspects of the prior verbal con-text (51). Fillmore (52) has likened adultconversations to a game in which twoparticipants take turns moving a topicalong. Children learn quite early thatconversation is such a turn-taking game(53). However, our discourse analysisrevealed a fundamentally different rela-tionship between Nim's and his teach-er's utterances.

fhe corpus we analyzed was derivedfrom transcripts of 31/2 hours of video-tapes from nine sessions recorded be-tween the ages of 26 to 44 months (54). Acomparison of Nim's discourse with histeachers and children's discourse withadults, characterized by Bloom et al.(51), is shown in Fig. 4. Adjacent utter-ances are those that follow an adult ut-terance without a definitive pause (51).At 21 months (MLU - 1.3), the mnost ap-propriate stage of development withwhich to compare Nim, the average pro-portion of a child's utterances that areadjacent is 69.2 percent (range, 53 to 78percent). A somewhat higher percentage(87 percent) of Nim's utterances wereclassified as adjacent (range: 58.7 to 90.9percent).

Adjacent utterances can be classifiedin four categories. (i) Imitations arethose utterances that contain all of thelexical items of the adult's utterances,and nothing else; (ii) reductions are thosethat contain some of the lexical items ofthe adult's utterance and nothing else;(iii) expansions are those that containsome of the lexical items of the adult'sutterance along with some new lexicalitems; and (iv) novel utterances are thosethat contain none of the lexical items ofthe aduft's utterance. Among the chil-dren studied by Bloom et al. (51), imita-tions and reductions accounted for 18percent (Fig. 4) of all of the children's ut-terances at stage 1 (MLU = 1.36). That18 percent decreas,ed with increasingMLU, accounting for only 2 percent ofthe children's utterances at stage 5

SCIENCE, VOL. 206

(MLU = 3.91). In contrast, 39.1 percentof Nim's adjacent utterances (N = 509)were imitations or reductions (range,19.5 to 57.1 percent).At stage 1, 21.2 percent of a child's ut-

terances are expansions of the adult'sprior utterance (range, 10 to 28 percent).On the average, only 7.3 percent ofNim's utterances were expansions of histeacher's prior utterance (range, I to 15percent). As the child gets older, the pro-portion of its utterances that are expan-sions increases. Bloom et al. (51) notedthat many of the child's utterances aresystematic expansions of verb relationscontained in the adult's prior utterance.No such pattern was discernible inNim's expansions. Indeed a preliminary',/analysis of Nim's expansions indicatesthat aside from the teacher's signs, hisutterances contain only a small numberof additional signs, such as me, Nim,you, hug, and eat. Since these signs arenot specific to particular contexts, theydo not add new information to the teach-er's utterance.By definition, adjacent utterances may

include interruptions of a teacher's or anadult's utterance. Such interruptions de-tract from true conversation since theyresult in discourse that is simultaneousrather than successive. In 71 percent ofthe utterances that have been examined(425 out of 585), Nim signed simulta-neously with his teacher. Of these over-lapping utterances, 70 percent occurredwhen Nim began an utterance while theteacher was signing. When the teacherinterrupted one of Nim's utterances, itwas generally the case that Nim had justinterrupted the teacher and the teacherwas, in effect, asserting his or her right tohold the floor. Nim's interruptions showno evidence that they are in response tothe teacher's attempts to take the floorfrom him. Few data are available con-cerning the relative frequency or dura-tion of simultaneous utterances that oc-cur in dialogues between children andadults in either spoken or sign language.In the most relevant study we could lo-cate, McIntyre reports that her video-tape transcripts of a 13-month deaf childsigning with her mother revealed virtual-ly no interruptions of the mother's utter-ances (54a). Bloom (55) and Bellugi (56)have observed that interruptions are vir-tually nonexistent in their videotapes ofchildren learning vocal and sign lan-guages (56a).None of Nim's teachers, nor the many

expert observers who were fluent in signlanguage, were aware of either the extent "

to which the initiation and contents ofNim's signing were dependent on the23 NOVEMBER 1979

Table 7. Discourse between\Washoe (W) andB. Gardner (B.G.).

Time Frame(sec- (seeond) Fig. 5)

00.00 1 B.G: leat00.42 2 mel02.38 3 Imore02.80 4 me(mine)l03.34 5 (W feeds B.G)07.09 6 /thank you/10.92 7 /what12.38 8 time12.88 9 W: /time13.17 10 now?l eatl15.42 11 Itime15.76 12 eatl

---------splice---------00.00 13 B.G:/what00.46 14 now?l00.29 15 /what W: lin04.79 16 now in,05.33 17 Ime05.67 18 eat06.17 19 time06.38 20 4 ?l eatl

teacher's signing or the degree to whichNim imitated or interrupted his teachers.That information can be obtained onlyfrom ifim or videota nspts. Thecontrast between the conclusions thatmight be drawn from our distributionalanalyses and those that follow from ourdiscourse analysis provides an importantmethodological lesson. In the absence ofa permanent record of an ape's signing,and the context in which that signing oc-curred, even an objectively assembledcorpus of an ape's utterances does notprovide a sufficient basis for drawingconclusions about the grammatical regu-larities of those utterances.

Unanticipated, but instructive, ex-

amples of the influence of the teacher'ssigning on Nim's signing were noted iftphotographs such as those shown in Fig.1 (a series of still photographs taken witha motor-driven camera). Examination ofFig. 1, prompted by the results of ourdiscourse analysis, reveals that Nim'steacher signed you while Nim was sign-ing me, then later signed who? while Nimwas signing cat. Because these were theonly four photographs taken of this dis-course, we cannot specify just when theteacher began her signs. It is not clear,for example, whether the teacher signedyou simultaneously or immediately priorto Nim's me. However, it is unlikely thatthe teacher signed who? after Nim signedcat.

Comparison of Nim's Discourse withThat of Other Signing Apes

Two valuable sources of informationthat suggest that Nim's discourse withhis teachers was not specific to the con-ditions of our project are a film producedby Nova for television, entitled, TheFirst Signs of Washoe (57), and a film,produced by the Gardners, TeachingSign Language to the Chimpanzee:Washoe (58).

Consider the scene from First Signs ofWashoe shown in Table 6 and in the left-hand portion of Fig. 5 (59). In this con-versation, Washoe's utterances eitherfollowed or interrupted B. Gardner's ut-terance. It is also the case that the signtime was uttered by B. Gardnerjust priorto Washoe's utterance time eat (60).Teaching Sign Language to the Chim-

panzee: Washoe presents a longer ver-

Table 8. Discourse between Washoe (W) and S. Nichols (S.N.).

Frame(see

Fig. 6)

12

S.N: Ithatl (points to cup)

(brings cup and doll closerto W; S.N. allows W to touchit; S.N. slowly pulls itaway)

3 S.N: Ithatl (points to cup)

S.N: (brings the cupand doll closer to W)

4

567

S.N: (brings cup closer to W)

S.N.: Ithatl (points to cup)

W:/babyl

W:/in/(looks away fromS.N.)

W: (looks back at cupand doll)

W: Ibabyl

W: linlW: /my

drinkl

897

sion of the same conversation. As can beseen in Table 7 and Fig. 5 both of Wash-oe's signs (time) and (eat) were signed byB. Gardner immediately prior to Wash-oe's having signed them. Time eat can-not be considered a spontaneous utter-ance for two reasons. It was a responseto a request to sign by B. Gardner and itcontained signs just signed by her. Thesignificance of a full record of discoursebetween a chimpanzee and its teacher isalso revealed by the segment that followsthe splice in the film. Consider Washoe'scombination me eat time eat in isolation.Without knowledge of the teacher's priorutterances it would be all too easy to in-terpret Washoe's utterance as one thatsignifies a description of future behaviorand a knowledge of time. Our transcrip-tion of the discourse between B. Gardnerand Washoe also shows that three out ofWashoe's four utterances interrupted B.Gardner's utterances.Another instructive example of the in-

,fluence of the teacher on the productionof Washoe's signs is provided by the ut-terance glossed as baby in my drink (Fig.

Table 9. Discourse between Washoe (W) andS. Nichols (S.N.).

Time(sec- Discourseonds)

00.00 S.N:/ who stupid?l00.42 W:/ Susan, Susanl05.30 S.N:/ who stupid?l05.58 W:/ stupidl06.42 S.N/: who?l06.72 W:IWashoel07.04 S.N:/ Washoel07.36 S.N:/ (tickles

Washoe)/

6 and Table 8), a combination of foursigns described in both films as a creativeuse of sign language by Washoe. In this(run-on) sequence, the order of Wash-oe's signs reflects the order in which theteacher (Susan Nichols) signed to Wash-oe to sign about a baby doll inserted in acup. The sequence of the teacher's signs(pointing to the doll and then pointing tothe cup) follows the order called for byan English prepositional phrase. Onlythe last two signs, my and drink occurred

without intervening prompting on thepart of the teacher. The sign glossed infilm as my is confiiurationally identicalto the sign me shown in Fig. 5, frame 17.Both signs conform to the specificationof my in the Gardners' description ofWashoe's sign (1, p. 264). For these rea-sons alone, Washoe's actual sequence ofsigns, baby in baby in my drink, cannotbe regarded as a spontaneously gener-ated utterance.

In the immediately preceding scene ofthe film, Susan was shown drilling Wash-oe extensively about a baby in shoe andan apple in hat while Washoe was tryingto grab the desired objects from theteacher. This suggests that Washoe'ssign my, in baby in baby in my drink, wassigned to convey to her teacher that shewanted the doll. Given this type of drill,and the teacher's pointing to the objectsto be named in the appropriate sequence,it is gratuitous to characterize the utter-ance shown in Fig. 6 as a creative juxta-position of signs that conveyed themeaning "a doll in Washoe's cup."As a final example of Washoe's dis-

I w

B.G.:/what

B.G.:/w hat

W.:/time

/what W.: /in

W.: /me

7? )

time

now D

in/

eat

eat/

Fig. 5. Tracings (made from a film) of Washoe signing with B. Gardner. See (S9).

now?

I

II

II

I

now/

eat/ /time eat/

898 SCIENCE, VOL. 206

course with her teachers, consider theconversation about Washoe's intelli-gence shown in Table 9. This sequencealso appears to be a drill. The importantquestion it raises, however, is whetherWashoe actually understood the mean-ings of stupid (and smart). Her usage ofstupid was clearly imitative of her teach-er. The exchange between Washoe andthe teacher (Susan Nichols) was also ter-minated at the point at which the teacherinduced Washoe to make the signs stupidand Washoe. The circumstances underwhich this sequence of signs occurredraises questions about the Gardners' se-mantic analysis of combinations such asNaiomi good (44). That combination waspresented as an example of attribution,an interpretation that would be appropri-ate only in the absence of the kinds ofprompting and reward shown in the filmsof Washoe signing.The longer of these films, Teaching

Sign Language to the Chimpanzee:Washoe, showed 155 of Washoe's utter-ances of which 120 were single-sign ut-terances. These occurred mainly in vo-cabulary testing sessions. Each of Wash-oe's multisign sequences (24 two-sign, 6three-sign, and 5 four-sign sequences)were preceded by a similar utterance or aprompt from her teacher. Thus, Wash-oe's utterances were adjacent and imita-tive of her teacher's utterances. TheNova film, which also shows Ally (Nim'sfull brother) and Koko, reveals a similartendency for the teacher to sign beforethe ape signs. Ninety-two percent of Al-ly's, and all of Koko's, signs were signedby the teacher immediately before Allyand Koko signed.The data provided by a few films are

admittedly much more limited in scopethan data of the type we obtained fromour nine videotapes. It seems reasonableto assume, however, that the segmentsshown in the films, the only ones avail-able of apes signing, present the best ex-amples of Washoe's, Ally's, and Koko'ssigning. Even more so than our tran-scripts, these films showed a consistenttendency for the teacher to initiate sign-ing and for the signing of the ape to mirV/ror the immediately prior signing of theteacher.

Other Evidence Bearing on an

Ape's Grammatical Competence

In evaluating the claim that apes canproduce and understand sentences it isimportant to keep in mind the lack of asingle decisive test to indicate whether aparticular sequence of words qualifies as23 NOVEMBER 1979

S.N.:/that/W.:/baby/ /in/

W:/baby in my drink/

Fig. 6. Tracings of Washoe (made from a film) signing with S. Nichols. See (59).

a sentence or whether a particular per-formance qualifies as an instance ofgrammatically guided sentence compre-hension. It has been observed widelythat the early sequences of words utteredby a child do not necessarily qualify assentences (19, 24). If, indeed, the onlyevidence that children could create andunderstand sentences were their initialutterances, and their responses to theirparents' utterances, there would be littlereason to conclude that a child's produc-tion and comprehension of words aregoverned by a grammar.A rich interpretation of a young child's

early utterances assumes that they areconstrained by structural rules (20, 22).It is difficult, however, to exclude sim-pler accounts of such utterances. Achild's isolated utterance of a sequenceof words could be a haphazard concate-nation of words that bear no structuralrelationship to one another (22) or rou-tines that the child learns by rote as imi-tations of its parent's speech (24). How-ever, as children get older, the varietyand complexity of their utterances grad-ually increase (21, 61). Especially tellingis the observation that children passthrough phases in which they producesystematically incorrect classes of utter-ance. During these phases, the child ap-parently "tries out" different sets ofrules before arriving at the correctgrammar. Children are also able to dis-criminate grammatically correct from in-correct sentences (62). Accordingly, ex-planations of their utterances that are notbased upon some kind of grammar be-come too unwieldy to defend.Production of sequences. Before re-

garding a sequence of words as sen-tences, it is necessary to demonstrate theinsufficiency of simpler interpretations.Consider some examples of sequence

production on the part of Sarah andLana. As a result of rote training, bothSarah and Lana learned to produce spe-cific sequences of words, for example,please machine give apple (9), or Marygive chocolate Sarah (6). Subsequently,both Sarah and Lana learned to sub-stitute certain new words in order to ob-tain other incentives from the same orfrom other agents (for example, Randygive Sarah apple, please machine givedrink, or please machine show slide).The last sequence was presented as evi-dence that Lana learned to use different"verbs" (give and show) in conjunctionwith a different category of incentives,slide, window, and music (9).

Sarah's and Lana's multisign utter-ances are interpretable as rotely learnedsequences of symbols arranged in partic-ular orders; for example, Mary give Sa-rah apple, or please machine give apple.There is virtually no evidence that Lanaand Sarah understood the meaning of allof the "words" in the sequences theyproduced. Except for the names of theobjects they requested, Sarah and Lanawere unable to substitute other symbolsin each of the remaining positions of thesequences they learned. Accordingly, itseems more prudent to regard the se-quences of symbols glossed as please,machine, Mary, Sarah, and give as se-quences of nonsense symbols (63).

Consider comparable performance bypigeons that were trained to peck arraysof four colors in a particular sequence,A--B--*C-->D, regardless of the physicalposition of the colors (64). In these ex-periments, all colors were presented si-multaneously and there was no step-by-step feedback after each response. Evi-dence that the subjects learned the over-all sequence, and not simply the specificresponses required by the training arrays

899

was provided by performance that wasconsiderably better than chance on novelarrays. It has yet to be shown that pi-geons can master ABCX problems(where X1 could refer to one type ofgrain, X2 to a different type of grain, X3to water, X4 to the opportunity to see orto attack another pigeon, and so on). If apigeon can learn such sequences (a notunlikely outcome) one wonders what isto be gained by assigning "names" toeach member of the sequence, for ex-ample, referring to the sequence,green--white--*red--blue, as machinegive R42 grain.

Sequences of symbols produced by anape may seem grammatically related toone another in the eyes of human observ-ers. It does not, however, follow that thechimpanzee had any knowledge of therelationships that a human observer mayinfer (65). As difficult as it may be totrain an ape, or any organism, to producea sequence of arbitrary responses thatmay look like a sentence, it is even moredifficult to show that those sequenceshave the structural properties of humansentences (63).Comprehension of multiword se-

quences. An inherent difficulty in usingapparent comprehension as an indicatorof a child's syntactic competence is thefrequent presence of nonsyntactic cuesto meaning (22, 23). This can be con-trolled if sentence comprehension exper-iments are designed to exclude semanticand extralinguistic cues. However, manypurported examples of sentence compre-hension by chimpanzees *can be ex-plained as nonsyntactic problem-solvingbehavior. Even complex problems,which seem to require an understandingof the syntactic structure of the instruc-tion (for example, conditional instruc-tions and instructions presented in hier-archical form), could be solved by apply-ing nonsyntactic rules (63).

Demonstrations by Premack, Rum-baugh, and the Gardners that their chim-panzees can answer wh-questions cor-rectly is evidence of the memory capac-ity of a chimpanzee. There is littlereason, however, to conclude that thesechimpanzees comprehended wh-ques-tions. In each case, the chimpanzeeswere drilled extensively on the correctanswers to questions such as color that?,what that?, and only a limited choice ofanswers (usually two) were available.The constant setting in which repeatedproblems of the same nature were ad-ministered provided ideal conditions forthe establishment of learning sets and theuse of nonsyntactic strategies in solvingthese problems. Without a greater varie-ty of problems and a greater range of

900

possible answers, the results of such<tudies cannot be interpreted as "linguis-tic" demonstrations of the interpretationof wh-questions (64).

In their effort to demonstrate compre-hension of wh-questions, the Gardnersaccepted as correct any response theydesignated as lexically appropriate. Forexample, if Washoe signed blue in an-swer to what color? when she was showna red ball, blue was considered "cor-rect" because it was a color. The signifi-cant correlation that the Gardners reportbetween question forms and responseforms shows that Washoe learned to re-spond to category questions with signsfrom the appropriate category: colors,trainers' names, actions, and so on.However, many of her specific answerswere clearly inappropriate. The Gard-ners nevertheless concluded that Wash-oe's performance is comparable to thatof a child at stage 3 in Brown's schemefor describing the development of lan-guage in children (22). At this stage, chil-dren are not only able to produce correctanswers to simple wh-questions, butthey are also able to produce a variety of

/ constructions whose mean length ex-ceeds 2.75 morphemes. The significanceof analyzing child language in terms ofstages derives largely from the structuralcomplexities that a child masters, in acumulative fashion, at each point of itsdevelopment. The Gardners' conclusiondoes not take into account these aspectsof a child's language at stage 3.

Conclusions

Projects devoted to teaching chim-panzees and gorillas to use languagehave shown that these apes can learn vo-cabularies of visual symbols. There is noevidence, however, that apes can com-bine such symbols in order to create newmeanings. The function of the symbolsof an ape's vocabulary appears to be notso much to identify things or to conveyinformation [as, for example, Skinner'sconcept of "tacts" (66)] as it is to satisfya demand that it use that symbol in orderto obtain some reward [Skinner's con-cept of "mands" (66)].

In our study more than 20,000 combi-nations of two or more signs, producedby Nim, an infant chim,panzee, were ex-amined for evidence of syntactic and se-mantic structure. Lexical regularities, inwhich particular signs tended to occur inparticular positions, were observed inthe case of two-sign combinations. It isimpossible to explain these regularitiesas ovet-all position habits or the memori-zation of many individual sequences. As

such, these regularities provide superfi-cial evidence that Nim's two-sign combi-nations followed rules of sign order.However, other aspects of Nim's use ofsign language suggest that it is unwar-ranted to conclude that his combinationswere primitive "sentences."The mean length of Nim's combina-

tions fluctuated between 1.1 and 1.6 dur-ing the last 19 months of the project.During that time, the size of his vocabu-lary more than doubled (from 42 to 125signs). Nim's three-sign combinationsshowed no evidence of lexical regular-ities, nor did they elaborate or qualifywhat he signed when he produced a two-sign combination.Our semantic analysis of Nim's two-

sign combinations showed that 85 per-cent of these combinations could be as-signed unambiguously to one of 20 se-mantic categories. Going beyond the re-sults of other studies, we demonstratedthe reliability of our semantic judgmentsand also observed that certain semanticroles were expressed in particular orders

\1of signs. However, our data also suggestthat it is premature to apply the methodof "rich interpretation" to the utterancesof an ape. Not only did the number oflexical examples of each semantic roleseem too few to justify the designation oforder regularities as semantic (ratherthan lexical), but there were also toomany idiosyncratic order regularities incombinations of particular signs. Thus,the evidence necessary to demonstrate aknowledge of categorical semantic rulesis insufficient.A discourse analysis of Nim's use of

sign language, which related Nim's utter-ances to his teacher's immediately prioruse of sign language, produced furtherevidence that Nim's use of language dif-fered fundamentally from that of a child.Nim imitated and interrupted his teach-ers' utterances to a much larger extentthan a child imitates and interrupts anadult's speech. This suggests that Nimwas less creative than a child in produc-ing utterances and that he had notlearned the give-and-take-aspect of con-versation that is evident in a child's earlyuse of language. Analyses of films of oth-er apes signing with their teachers re-vealed a similar lack of creativity in oth-er apes' utterances, and a similar depen-dence of these utterances on the priorsigning of their teachers.

In sum, evidence that apes create sen-tences can, in each case, be explained byreference to simpler nonlinguistic pro-cesses. Sequences of signs, produced byNim and by other apes, may resemblesuperficially the first multiword se-quences produced by children. But un-

SCIENCE, VOL. 206

less alternative explanations of an ape'scombinations of signs areieliminated, inparticular the habit of partially imitatingteachers' recent utterances, there is noreason to regard an ape's multisign utter-ance as a sentence.Our results make clear that any new

study of an ape's ability to use languagemust collect a large corpus of utterances,in contexts that can be readily docu-mented by reference to a permanent vi-sual record (67). With such data onewould be left with an incomplete basisfor comparing an ape's and a child's useof language.For the moment, our detailed investi-

gation suggests that an ape's languagelearning is severely restricted. Apes canlearn many isolated symbols- (as candogs, horses, and other nonhuman spe-cies), but they show no unequivocal evi-dence of mastering the conversational,semantic, or syntactic or tion oflanguage.-r

Rae im-is gaid NOWe1. B. T. Gardner aad R. A. Gardner, J. Exp. Psy-

chol. Gen. 1I (No. 3),.244 (97S).2. R. A. Gardner and B. T. Gardner, Ann. N. Y.

Acad. Sdt. A09.31(197).-3. __, ibid. 157,752(1WS).4. D. Pre k Sdese 172, 84 (1971).5. ,J. Fz. Anal. Beh'. 14, 107 (1970).6. ,Intellee iApt and fan (Lawrence

ErlbaumA , Hlladale, N.J,X, 1976).7. J. C. Woodward, Sign Lan. Stud. 3, 39

(1973); H. Borsein, in Sign L48age of theDeaf.: Psychological, Lingustic. Wa Sociologi-cal Perspectives,! . .Sh sad L. Nam-ir, Eds. (Academi9a, N7 Yo 1978), pp.338-361; W. C. StIkoe et a. . t pften as-sumed erroneousy thati alfual-vi-seal communication are ASL. w r, signlanguages vary along a c m. At oneextreme is ASL, ic p uiwque gram-mar, exprEsive devices;, a toQpoogy. Thenatural language i- N a i can deafpeople, ASL u learned as a e bymany deaf people, especially *e chidrenof deaf parent. At the oite extreme isSigned English, a c for e i English ina manual-vinsual mode. I Sgednlish-butnot in ASL-sps are ued is word or-der with signed equivaeat of MORes, suchas -ed and -ly. 1n SiWgi sh (PSE),ASL s toeher with s t expressivedevices in English word order without thegrammatical inorphemes of English, are used.Hearing persocs terely achieve "uativelike"ctrol over th copesttuan ad grammarof ASL. Ia this oh se oa signinglbaviorinE w¢n1Al. used. Inthis sense, its -tihing to t their signing"ASL" since it do exhibitlhe gamaticalucc it doe_.4'-- '' 'Dstrauctre 01f thatlumas

8. W. C. Stoloe, J.. C l,C. C. Cron-berg, Eds.,A Dkitioy n Sign Lan-iaenL bGggG ud College

9D. M. Rumbaugh, E4., Leqrning by aChdeqpanzee: T>e La F iet(AcademicPress. New po,97N.,515

(1972); M. K.T n Lut*G Up Hu-man (Scieue and ehaiori Books, Palo Alto,

FF. G. Ptten, BriA Lang. 12, 72 (1978).P. ea m. N.F. Acad. Sci. 280, 660

. 1975).< C. Hayes, The APe in Ovr Hoses (Harper &

Row, New York, Wl)s. kF. H yes and C.Hayes, I4c. 4A ?A ;. St. O, IOS (1951);

W. N.K Scie**ce 423 (1968);Wand. .ad the Child(>Graw- ,NeW ouk, l93¢N. N. Lady-:ua-Kots, Apg6nJItlusCeAI4 (Museum

14. Univ.

23 NOVEMB 1979

Press, London, 1961); K. von Frish, The Danc-ing Bees (Methuen, London, 1954); N. Tin-bergen, The Study of Instinct (Clarendon, Ox-ford, 1951).

15. N. Chomsky, Aspects of the Theory of Syntax(MIT Press, Cambridge, Mass., 1965); SyntacticStructure (Mouton, The Hague, 1957); M.Gross, M. Halle, M. P. Schuztenberger, Eds.,The Formal Analysis of Natural Languages(Mouton, The Hague, 1973); J. Katz and P. M.Postal, An Integrated Theory of Linguistic De-scription (MIT Press, Cambridge, Mass., 1964).

16. Nim was born on 21 November 1973 at the Insti-tute for Primate Studies in Norman, Okla. On 3December 1973, Nim was flown to New Yorkaccompanied by Mrs. Stephanie LaFarge, who,along with her family, raised Nim in their homeon New York's West Side for 21 months. Be-tween 15 August 1975 and 25 September 1977,Nim lived on a university-owned estate(Delafield) in Riverdale, N.Y., in a large housewith private grounds where he was cared for byfour to five resident trainers. Nim formed partic-ularly close attachments with certain membersof the project. H. S. Terrace was the only proj-ect member who maintained a strong and a con-tinuous bond with Nim throughout the project.During the first 18 months of the project, Mrs.LaFarge was the most central person in Nim'slife. After he was moved to Delafield, Nim be-came closely attached to L. A. Petitto, who su-pervised his care both at Delafield and in a spe-cial classroom built for Nim in the PsychologyDepartment of Columbia University. After Pet-itto left the project (when Nim was 34 monthsold), Nim became closely attached to two resi-dent teachers at Delafield, W. Tynan and J. But-ler. From September 1974 until August 1977,Nim was driven to his classroom at Columbiathree to five times a week where he was givenintensive instruction in both the expression andthe comprehension of signs and where he couldalso be observed, photographed, and videotapedby other teachers with a minimum of distraction(through a one-way window). Nim was alsotaught regularly at Delafield, although in a lessformal manner. Extensive analyses of his sign-ing at Delafield and in the Columbia classroomrevealed no systematic differences in any of theaspects of Nim's signing reported in this study.During the 46 months in which he lived in NewYork, Nim was taught by 60 nonpermanent vol-unteer teachers. As he grew older, Nim's emo-tional reactions to changes in personnel in-tensified, and it became increasingly difficult fornew teachers to command Nim's attention. BySeptember 1977, it was clear that we did nothave the resources necessary to hire a staff ofqualified permanent teachers who could ad-vance the scientific aspects of the project. Ourchoice was to provide "babysitters" who couldlook after Nim, but who were not uniformlyqualified to further Nim's understandin8 of signlanguage, or to terminate the project. With greatreluctance, we decided on the latter course ofaction. On 25 September 1977, Nim was flownback to his birthplace in Oklahoma. A fuli de-scription of Nim's history and socialization canbe found in (17).

17. H. S. Terrace, Nim (Knopf, New York, 1979).18. R. S. Fouts and L. Groodin, Bull. Psychon. Soc.

4 (No. 4a), 264 (1974).19. L. Bloom, Language Development: Form and

Function in Emerging Grammars (MIT Press,Cambridge, Mass., 1970).

20. __, One Word at a Time: The Use of SingleWord Utterances Before Syntax (Mouton, TheHague, 1973).

21. L. Bloom and M. Lahey, Language and Devel-opment and Language Disorders (Wiley, NewYork, 1978).

22. R. Brown, A First Language: The Early Stages(Harvard Univ. Press, Cambridge, Mass., 1973).

23. J. A. Fodor, T. G. Bever, M. F. Garrett, ThePsychology of Language (McGraw-Hill, NewYork, 1978).

24. M. Braine, Monogr. Soc. Res. Child Dev. 41(1), Serial No. 164 (1976).

25. An imitative sign is one which repeats the teach-er's immediately prior utterance. A spontaneoussign is one that has not occurred in the teacher'simmediately prior utterance. A prompted signfollows a teacher's prompt. A prompt uses onlypart of the proper sign's confiration, move-ment, or location. For example, the teachermight prompt Nim (first and second fingersdrawn down the temple) by extending those twofingers from a fist. Agreement between teachers'reports and transcripts of an independent ob-server (who either watched Nim and his teacherthrough the one-way window of his Columbiaclassroom or who transcribed videotapes)ranged between 77 and 94 percent. Virtually all

of the disagreements occurred when the teacherfailed to record a sign, when, for example, theteacher was busy preparing an activity, whenNim was signing too quickly, or when the teach-er was signing to Nim. There was no evidencethat the omissions of the teacher were system-atic. At worst, the teachers' reports under-estimated the extent to which Nim signed.

26. A complete listing of Nim's vocabulary, includ-ing the topography of each sign and context inwhich it occurred, can be found in (17, Ap-pendix C).

27. B. T. Gardner and R. A. Gardner, Bull Psy-chon. Soc. 4 (No. 4a), 264 (1974); Minn. Symp.Child Psychol. 8, 3 (1974).

28. Word order is but one of a number of ways inwhich a language can encode different mean-ings. When, however, regularities of sign ordercan be demonstrated, it does provide strong evi-dence for the existence of grammatical struc-ture. Given the difficulty of documenting otheraspects of an ape's signing, regularities of signorder may provide the simplest way of demon-strating that an ape's utterances are grammatical[U. Bellugi and E. S. Klima, The Signs ofLanguage (Harvard Univ. Press, Cambnidge,Mass., 1979); H. W. Hoemann, The AmericanSign Language: Lexical and GrammaticalNotes with Translation Exercises (National As-sociation of the Deaf, Silver Spring, Md., 1975);E. S. Klima and U. Bellugi, Ann. N.Y. Acad.Sci. 263, 255 (1975); H. Lane, P. Boyes-Braem,U. Bellugi, Cog. Psychol. 8, 262 (1976)].

29. In ASL, the segmentation of combinations intosigns has a function similar to that of the seg-mentation of phrases into words in spoken lan-guage: segmentation delineates sign sequencesthat are immediately related to one another.Other segmentation devices as well are used inASL. In addition to the relaxation of the hands,specific head and body shifts, and specific facialexpressions signal constituent boundaries[Stokoe et al. (8); E. Brown and S. Miron,-J.Verbal Learn. Verbal Behav. 10. 658 (1971);H. Lane and F. Grosjean, J. Exp. Psychol. 97,141 (1973)].

30. Two rules were used in combinations containingsigns which were repeatedsuccessively. Theserules ensured the shortest possible descriptionof a particular combination. For homogeneouscombinations, if all signs in a sequence were thesame (eat eat eat), the sequence was treated as asingle sign utterance (eat). Homogeneous se-quences of signs were not tabulated as combina-tions. For heterogeneous sequences, if a partic-ular sign was repeated, immediate repetitions ofthat sign were not counted. For the purpose oftabulation within the corpus, a sequence such asbanana me me me eat was reduced to bananame eat. Whereas the original sequence con-tained five signs, this combination was enteredas a three-sign sequence. In general, the sign Y,repeated in succession n times, was counted as asingle occurrence of Y, independently of the val-ue of n. The entries shown in Table I representthe number of types and tokens of Nim's linearcombinations observed to occur before the re-duction rules were used. A complete listing ofNim's combinations that resulted from the appli-cation of the reduction rules can be found in Ter-race et al. (31). In ASL, repetitions of a signconvey particular meanings. One type of con-trast between repeated and nonrepeated signs isexemplified by the contrast between the formsof certain nouns and verbs. Many verbs (sweep,fly, drive) are made with a single motion. Re-lated nouns (broom, airplane, car) are made byrepeating a sign twice (the so-called doublebounce form; see T. Suppala and E. Newport,personal communication). None of Nim's teach-ers could distinguish between the meanings ofutterances that did and did not contain signs thatwere repeated successively. Emphasis appearsto be their sole function. Overall, less than 15percent of the linear utterances we observedcontained successively repeated signs. As far aswe could tell, the outcomes of our analyseswould be unchanged if we included utterancescontaining signs that were repeated successive-ly.

31. H. S. Terrace, L. A. Petitto, R. J. Sanders,T. Bever, in Children's Language, K. Nel-son, Ed. (Gardner Press, New York, in press),vol. 2.

32. Other complete listings of the regularities ofNim's two-sign combinations can be found in(t7).

33. That there are more tokens of two-sign combina-tions containing me than containing Nim is per-haps best explained by the fact that Nim learnedthe sign me before he learned the sign Nim. Sub-sequently the frequencies with which Nim andme were combined with transitive verbs was es-

901

sentially the same. In the case of combinationscontaining more, it might be argued that Nimmodeled the construction more + X after histeachers' utterances. Often a teacher would signto Nim, more + X?, to see if Nim would signmore orX in reply. On this view, Nim learned tosign more + X by first imitating a few instancesof more + X and then generalizing this con-struction to new objects and actions. Such gen-eralization would be necessary because Nimproduced most of the first tokens of each type ofmore + X combination without any modeling bythe teacher. This type of explanation seems lesscogent in the case of other classes of combina-tions, for example, those containing transitiveverbs shown in Table 2, because Nim was notasked questions with these signs that couldserve as models.

34. N. Colburn, thesis, Teachers College, ColumbiaUniversity (1979).

35. Most of the comparisons we will make draw up-on data obtained from studies of the acquisitionof spoken language by hearing children of hear-ing parents. Nim was taught by hearing teacherswho were not uniformly fluent signers. Accord-ingly, studies describing the acquisition of signlanguage by deaf children of hearing parentswould provide the most relevant point of refer-ence for evaluating the data we obtained fromNim. However, to the extent that data are avail-able from deaf children (of either deaf or hearingparents) there is no evidence that any major dif-ferences exist between the general features oflanguage acquisition by deaf and hearing chil-dren. For further details see (36-38).

36. E. F. Ashbrook, "Development of semantic re-lations in the acquisition of American Sign Lan-guage" (working paper, Salk Institute, June1975).

37. R. S. Hoffmeister, Proceedings of the First Na-tional Symposium on Sign-Language Researchand Teaching, Chicago, 1977.

38. E. S. Klima and U. Beliugi, in Communicationand Affect: A Comparative Approach, T. Allo-way, L. Krames, P. Pliner, Eds. (AcademicPress, New York, 1972), p. 67.

39. In calculating a child's MLU, certain conven-tions are followed that cannot be applied direct-ly to sequences of signs. A spoken utterance, forexample, is often broken down into morphemesrather than words: running, and run there wouldeach be regarded as a two-morpheme utterance.In sign language, the utterance, run there, canbe expressed by a single sign [E. S. Klima andU. Beilugi (38)]. While the MLU of utterances insign language has yet to be calibrated as a mea-sure of linguistic development, it is neverthelessof interest to ask how the mean length of Nim'scombinations varied during the time in which werecorded his utterances. The rules used to calcu-late Nim's MLU resulted from a conservativeadaptation of those followed by Brown (22). SeeTerrace et al. (31) for further details.

40. As far as we can determine, the only MLU func-tions that have been described for deaf childrenacquiring sign language are those described byE. S. Klima and U. Bellugi (38); H. S. Schlesin-ger, in preparation; R. J. Hoffmeister (37). Oneexplanation for the apparent delay in the growthofMLU in deaf children has to do with the prop-erty of sign language that allows one to encode anumber of morphemes within a single sign [seeE. S. Klima and U. Bellugi (38), p. 85].

41. R. Brown, personal communication.42. I. M. Schlesinger, in Ontogenesis of Grammar,

D. I. Slobin, Ed. (Academic Press, New York,1971), p 63.

43. In some cases, word order is also used contrast-ively, and in at least one child, intonational dif-ferences were observed that were associatedwith differences in meaning [L. Bloom (20); M.Bowerman, Early Syntactic Development: ACross-Linguistic Study with Special Referenceto Finnish (Cambridge Univ. Press, London,1973)1.

44. B. T. Gardner and R. A. Gardner, in Behavior ofNonhuman Primates: Modern Research Trends,A. M. Schrier and F. Stollnitz, Eds. (AcademicPress, New York, 1971), vol. 4, pp. 117-184.

45. R. S. Fouts, in Society and Psychology of Pri-mates, R. H. Tuttle, Ed. (Mouton, The Hague,1975), pp. 371-390.

46. The semantic corpus covered data from roughly5.5 months for each teacher. Data were based on48 reports of J. Butler, 58 of R. Sanders, and 41of W. Tynan. After interpreting the utterancesof their own sessions, each teacher interpretedthe utterances of one of the other two teachers.They agreed in their interpretations of 717 utter-ances (57 percent of the original corpus). If a dis-agreement could not be resolved, the utterancein question was considered ambiguous and dis-regarded. Contextual notes in the teachers' rec-ords included sufficient information for theteachers to agree as to the interpretation of the967 two-sign combinations (76 percent of theoriginal corpus). In the remaining cases(N = 295), no interpretation could be made(N = 260), or two or more equally reasonableinterpretations were made that could not be dis-ambiguated (N = 35). These combinations werenot included in our analysis. None of our con-clusions would be altered if we used either of theinterpretations of the combinations that werenot distinguishable.

47. R. Brown, Psycholinguistics; Selected Papersby RogerBrown (Free Press, New York, 1970).

48. The number of categories used for interpreting achild's or an ape's early combinations is arbi-trary. Our 20 categories could be collapsed intoseven in Brown's with two left over; into eight inthe Gardners' system, with two left over; andinto 11 in Patterson's system, with one left over.

49. In addition to the relational combinations we ob-served (Fig. 3), 166 combinations were assignednonrelational interpretations, apparently ex-pressing the conjunction of elements from twopropositions, for example, tickle hug, dirty run,apple nut, and in play. Such expressions are rarein children's speech and have not been reportedin previous studies on training apes to learn signlanguage [L. Bloom, P. Lightbown, L. Hood,Monogr. Soc. Res. Child Dev. 40 (2), Serial No.160 (1975).

50. E. Newport, in Cognitive Theory, N. J. Casel-lan, D. B. Pisoni, G. R. Potts, Eds. (LawrenceErlbaum Associates, Hillsdale, N.J., 1976); C.Snow, Child Dev. 43, 549 (1972).

51. L. Bloom, L. Rocissano, L. Hood, Cog. Psy-chol. 8, 521 (1976).

52. C. Fillmore, On Deixis (Indiana LinguisticsClub, Bloomington, 1973), vol. 9, p. 269.

53. D. Stern, J. Jaffee, B. Beebe, S. Bennett, Ann.N.Y. Acad. Sci. 263, 89 (1975).

54. These analyses were performed by R. J. Sand-ers. Each tape was transcribed by the teacherwho worked with Nim (L. Petitto, four tapes; R.Sanders, four tapes; J. Butler, one tape; W. Ty-nan, one tape). Only single signs and linear com-binations were used in our discourse analysis.They accounted for 95 percent of the transcribedutterances. An informal check of the remaining5 percent (simultaneous combinations and con-tractions) indicated that the results would re-main the same if all of Nim's utterances wereincluded in the analysis. In order to check thereliability of our transcripts, short segments offive tapes were transcribed by two independenttranscribers, both of whom were teachers fromthe project. The average agreement between thetwo transcribers was 81.8 percent.

54a.M. McIntyre "Learning to take your turn inASL," unpublished manuscript.

55. L. Bloom, personal communication.56. U. Bellugi, personal communication.56a.C. Baker [in On the Other Hand: New Per-

spectives on American Sign Language, L. A.Friedman, Ed. (Academic Press, New York,1977)] notes that, in some instances the "speak-er's" sign overlaps with the "addressee's" sign,for example, the last sign of a question is heldmomentarily. However, Baker (personal com-munication) observes that this kind of overlap isnot regarded as an interruption and that inter-ruptions in dyadic signing pairs are as infrequentas interruptions in dyadic oral pairs.

57. The First Signs of Washoe, WGBH Time-Life,Nova (Time-Life Films, New York, 1976).

58. Teaching Sign Language to the Chimpanzee:Washoe, produced by R. A. Gardner and B. T.Gardner (Psychological Cinema Register, Uni-versity Park, Pa., 1973), No. 16802.

59. The transcripts of Washoe signing with herteachers (Tables 6 to 9) were prepared by L. A.Petitto and verified by H. S. Terrace. In pre-paring our transcripts we followed the practiceof psychologists and linguists studying sign lan-guage acquisition by deaf children, reviewingeach utterance in normal and in slow motion. Itshould be clean that static descriptions of isolat-ed signs (Tables 6 to 9 and Figs. S and 6) cannotconvey information about movement. Such in-formation would be crucial, for example, in dis-tinguishing between signs where repetition influ-ences their meaning ["double bounce" signs de-scribed in (30)]. Our reviews of both films ofWashoe yielded no instance in which the addi-tion of information about movement would alterthe interpretation of Washoe's or her teachers'signing. With the one exception noted in Table 7and Fig. 5, the segments we analyzed were un-edited. Our transcripts and tracings show all ofthe teacher's and Washoe's signs, at the earliesttime at which they were discriminable during thesegments we selected. Since the films we ana-lyzed consisted of edited segments, it was notpossible to establish a single temporal referencefor the occurrence of each sign. In transcripts,the time of occurrence of each sign was speci-fied with respect to the first sign of that portionof the transcript (arbitrarily designated 00:00).Time was measured by counting the number offrames between the beginning of each scene.The frame numbers in the transcript refer to theframes of the figure in this text that show a trac-ing of the sign. In Tables 6 to 9, the beginning andthe end of each utterance is marked by backslashes. Other behavior is described in paren-theses. Tracings were prepared by L. Roberts.

60. The Gardners have not presented evidencethat Washoe understood the meaning of the signtime. In this and in other examples of its usage,it seems as if Washoe learned that it was an ap-propriate response when requesting food orsome other incentive [the Gardners' descriptionof the usage of time (1, p. 266)].

61. P. S. Dale, Language Development, Structureand Function (Holt Rinehart & Winston, NewYork, 1976).

62. T. G. Bever, in Georgetown University RoundTable on Languages and Linguistics, D. P.Dato, Ed. (Georgetown Univ. Press, Washing-ton, D.C., 1975), pp. 63-75.

63. H. S. Terrace, J. Exp. Anal. Behav. 31, 161(1979).

64. , Bull. Psychon. Soc. 9, 269 (1977);R. 0. Straub, M. S. Seidenberg, T. G. Bever, H.S. Terrace, J. Exp. Anal. Behav. 32, 137 (1979);R. 0. Straub, thesis, Columbia University(1979).

65. J. Limber, Am. Psychol. 32 (No. 4), 280 (1977);G. Mounin, Curr. Anthropol. 17 (No. 1), 1(1976); M. S. Seidenberg and L. A. Petitto, Cog-nition 7, 177 (1979).

66. B. F. Skinner, Verbal Behavior (Appleton-Cen-tury-Crofts, New York, 1957).

67. Our experience also suggests that in attemptingto extend the mastery of sign language beyondthat which we observed in Nim, it is importantto guarantee that the subject of this type of studybe raised and taught by a small and stable groupof teachers [see (17)].

68. Supported in part by grants from the W. T.Grant Foundation, the Harry Frank Guggen-heim Foundation, and NIH (ROI MH29293).We thank L. Bloom and R. Brown for helpfulcomments of an earlier draft of this manuscript;S. LaFarge for the care with which she raisedNim during his first 18 months; B. Stark, S. Ler-man, and T. Blumenfeld for supervision ofNim's medical care, and G. A. Tate for assist-ance in producing photographs of Nim. We alsothank the more than 100 volunteers who assistedin working with Nim and on data analysis; andW. Benesch, I. Brody, J. Butler, B. Johnson, S.Quinby, A. Schachter, and W. Tynan for theirassistance.

SCIENCE, VOL. 206902