IMITATION AND LANGUAGE DEVELOPMENT'
James A . Sherman UNIVERSITY OF KANSAS
I. INTRODUCTION
.
.
.
.
.
.
.
.
.
.
.
11. SOME IMPORTANT CHARACTERISTICS OF IMITATION FOR
.
240
111. THE USE OF IMITATION IN LANGUAGE DEVELOPMENT .
246
THE ANALYSIS OF LANGUAGE
IV.
SUMMARY AND DISCUSSION REFERENCES.
.
.
.
. .
.
. .
.
. .
.
. .
.
. .
.
. .
.
239
. .
. .
263 270
I. Introduction During the first four years of life most children reared in a speech community increasingly exhibit speech forms common or typical of that community. At the same time, they exhibit an increasing ability to imitate the speech of people in their environments. The purpose of this paper is to review some of the recent research on the development of imitative behavior in children and the use of imitation procedures to develop various language forms in children, as a basis for suggesting the ways in which imitation might facilitate the development of speech, from more primitive forms toward closer approximations of the speech of the adult language community. Preparation of this chapter was supported by Program Project Grant HD 00870 to the Bureau of Child Research, University of Kansas. I thank Dr. Donald M. Baer for his helpful comments and criticism throughout the preparation of the chapter. 239
240
lames A . Sherman
11. Some Important Characteristics of Imitation for the Analysis of Language Most research on imitative behavior of children has been concerned with an analysis of the variables which are related to whether children with existing imitative repertoires exhibit these repertoires. (Reviews of such research have been published by Aronfreed, 1969; Bandura, 1962; Bandura & Walters, 1963; Flanders, 1968; among others.) Several learning theory accounts of the development of imitation have been offered (N. E. Miller & Dollard, 1941; Mowrer, 1960; Risley, 1966), but until recently relatively little research has been concerned with actually establishing imitative behavior in children. There appears to be two reasons for this. The first is that most children already display imitative behavior at a very early age. The second is that it is impossible to state definitely that a child lacks imitative behavior, because it is impossible to examine all of the possible conditions and situations in which a child might exhibit imitative behavior. Recently, however, several studies have appeared in which procedures were described for the development of imitative behavior in at least apparently nonimitative children. In each of these cases, the children used as subjects were from institutionalized, deviant populations. Although it is impossible to state that these children did not ever display imitative behavior, they clearly did not display such behavior in a variety of situations in which they were observed, particularly in experimental situations in which the development of imitation was to be analyzed. After the apparent lack of imitative behavior was assessed, training programs based upon reinforcement procedures were employed to establish imitative repertoires. Baer, Peterson, and Sherman (1967) have provided one example of such procedures. The subjects were retarded children, whose existing imitative repertoires were evaluated by observing the children on their wards and by bringing the children into an experimental situation and demonstrating a series of simple responses to them (such as raising the left arm, or tapping the table with the left hand) accompanied by the instruction, “DOthis.” None of the children initially matched the demonstrations of the experimenter. Furthermore, none appeared to imitate behaviors shown by the other children or personnel on the ward. On this basis, the children were classified as nonimitative. The initial training in this study involved the use of reinforcement procedures to establish simple matching responses. The children were brought to an experimental room at mealtimes and bites of the children’s meals were used to reinforce their behavior. Two procedures were used in con-
Imitation and Language Development
24 1
junction with reinforcement to establish the first matching responses: “shaping” and “putting through.” Shaping was the use of bites of food contingent upon behavior to reinforce successively closer and closer approximations to a desired target behavior. Putting through was a procedure in which the experimenter manually guided or assisted the children through the desired behavior. As training on a particular response progressed, the experimenter’s assistance was gradually removed or “faded out.” An example of such training is as follows: The first response of the program for Subject 1 was to raise an arm after the experimenter had raised his. The subject was presented with a series of arm-raising demonstrations by the experimenter, each accompanied by ‘Do this,’ to which she made no response. The experimenter then repeated the demonstration, reached out, took the subject’s hand and raised it for her, and then immediately reinforced her response. After several trials of this sort, the experimenter began gradually to fade out his assistance by raising the subject’s arm only part way and shaping the completion of the response. Gradually, the experimenter’s assistance was faded until the subject made an unassisted arm-raising response whenever the experimenter raised his arm (Baer et al., 1967, p. 408).
Although a child who had been trained in this manner might reliably raise his arm after the experimenter had demonstrated this response, it would be questionable to assert that imitative behavior had been established. For example, the experimenter might now say, “Do this” and tap the table. Given that only one matching behavior had been established, the child probably would raise his arm rather than tap the table. Thus, what had been taught was a specific response to a particular stimulus or cue (“DO this” plus probably any demonstration by the experimenter). In this study, however, training did not stop after one response had been established. A series of different matching behaviors was trained, each member of which was topographically similar to a prior demonstration by the experimenter. A result of this training is shown in Fig. 1. This figure shows the percentage of new demonstrations (within blocks of 10 new demonstrations) matched by this child the first time each demonstration was shown to the child. Over a training series encompassing 130 different responses, the child’s probability of imitating a new response increased from a very low level to 100% for the last block of 10 new demonstrations. Similar results were obtained for a second subject. (The third child required only a short list of imitative tasks and it was impossible to calculate the results in the same manner.) To the extent that subjects would match virtually any new response demonstrated by the experimenter, it could be said that they displayed an imitative repertoire. Correlated with this increase in the probability of imitating new demonstrations was a decrease in the amount of training required to establish new matching responses, even though
242
James A . Sherman
BLOCKS OF 10 NEW RESPONSES
Fig. 1. The development of imitation over successive blocks of new demonstrations (from Baer et al., 1967).
these responses might not have been matched by the child the first time they were demonstrated. Thus, the children learned new imitations faster and faster. (The increased probability of matching new responses the first time demonstrated, and more rapid learning of new responses as training progressed, are not unique to imitative training: very similar phenomena have been noted in other learning situations. This has been labeled “learning set” and refers to an increase in the rate of learning over a series of related problems.) As imitative training progressed, the experimenter performed certain demonstrations to the children, which, if imitated on the first presentation, were never reinforced. These demonstrations were probes to determine whether the child would continue to imitate them in spite of nonreinforcement. It should be emphasized that the never-reinforced imitations were different in topography from the reinforced imitations. The results for one subject are shown in Fig. 2. The solid line in the figure shows the percentage of those demonstrations imitated which resulted in reinforcement, and the dotted line shows the percentage of those demonstrations imitated which did not result in reinforcement. During Sessions 15 to 26 when certain imitations produced reinforcement, all imitations (reinforced and unreinforced) were maintained. Also shown in the figure are the data from a condition in which no imitations produced reinforcement. In this condition, labeled DRO in the figure (Sessions 27 to 31), reinforcement was delivered 20 seconds after an imitation had occurred or, if no imitation occurred, 20 seconds after the demonstration. When no imitations produced reinforcement, all imitative behavior declined in strength; but imitative behavior recovered when reinforcement again was made contingent upon only some of the imitations (Sessions 32 to 38). Similar results were ob-
Imitation and Language Developnierit
243
"\ t
L...
SESSlONS
Fig. 2 . The maintenance and extinction of imitative responses ( f r o m Burr et al., 1967). See text for further explanation of abbreviations.
tained for the other two subjects. Peterson (1968b), with one of the subjects used in the Baer et al. (1967) study, additionally showed that unreinforced imitations were maintained when demonstrations for these responses were interspersed among demonstrations for reinforced imitations, but were not maintained as well when demonstrations for unreinforced imitations were presented alone over a series of trials. Metz (1965) also has reported the use Qf similar reinforcement procedures to establish nonvocal imitative behavior in two autistic children. In this study, initial matching responses were taught by shaping and puttingthrough procedures. After this training the children displayed a higher probability of imitating new demonstrations on the first presentation than they had prior to imitative training. Lovaas, Berberich, Perloff, and Schaeffer (1966) have used similar procedures to establish imitative speech in two mute schizophrenic children (also see Hewett, 1965; Hingtgen, Coulter & Churchill, 1967; Lovaas, 1966). Each child was reinforced with bites of food first for simply emitting sounds, then only after he had emitted a sound shortly following a vocalization by the experimenter, and then only if his vocalization increasingly matched the experimenter's vocalization. A series of different sounds and words was then demonstrated by the experimenter. In each case, the experimenter delivered reinforcement contingent upon a child's producing closer and closer matches to the vocal demonstration of the experimenter. The result of these training procedures was that the children's rate of learning to match new sounds presented by the experimenter increased as training progressed. This result is similar to that obtained by Baer et al.
244
James A . Sherman
(1967) during training of nonvocal imitative responses with retarded children. In addition, Lovaas et al. reported that when reinforcement was no longer delivered contingent upon imitative behavior (but instead was delivered on a time basis rather than following correct imitations) imitative behavior decreased in strength. In a further manipulation, Lovaas et al. demonstrated Norwegian words to the children interspersed with a series of English-word demonstrations. The children’s accurate imitation of English words was reinforced, but no reinforcement was provided for imitation or partial imitation of the Norwegian words. The result was that the children’s imitation of the Norwegian words improved in accuracy over time, even though there were no direct consequences for this improvement. This result suggests that the improvement in imitation of Norwegian words was related to the ongoing reinforcement for accurate English imitations, since, in the previous manipulation, all imitation decreased once contingent reinforcement for imitations was removed. However, no experimental procedures were employed to determine directly the effect upon the Norwegian words of removing contingent reinforcement for English imitations. The finding of Lovaas et al. that imitations of certain words improved in accuracy, even though these imitations were never reinforced, was replicated with three normal preschool children by Brigham and Sherman (1968). In this study, an experimenter demonstrated both English and Russian words to subjects. Accurate imitation of the English words was reinforced (with praise and candy or tokens which could be traded for a toy), while imitation of Russian words produced no consequences from the experimenter. In a later condition, reinforcement was no longer presented contingent upon accurate imitation of English words, but instead was presented after a minimum time period following an English-word imitation. The results of these conditions for one subject are shown in Fig. 3. The two scales on the ordinate represent the percentage of words correctly imitated for the English words and a score representing the accuracy with which Russian words were imitated. (See Brigham and Sherman for a detailed description of the accuracy scores for Russian words and the reliability of the scoring system.) The DRO period marked on the figure refers to the condition in which reinforcement was no longer delivered immediately following accurate English imitations. The figure shows that when reinforcement was delivered immediately following correct English imitations, the percentage of correct English imitations and the accuracy of imitation of Russian words were high. When reinforcement was no longer delivered contingent upon correct English imitations, the percentage of English words correctly imitated and the accuracy of imitation of Russian words decreased. In addition, new
Imitation and Language Development
245
*--*---
4 i
q.
4
'
-+----t--t---t-
t --t --l 11 12 13
O R 0
Rcinlorcrmcnl
7
8
9
1
0
Fig. 3. Accuracy of imitation for English and Russian words (from Brigham & Sherman, 1968). See text f o r further explanation of abbreviations.
Russian words added during the DRO condition (Russian? words-filled triangles) were not imitated accurately. When reinforcement again was made contingent upon accurate English imitation, the percentage of English words correctly imitated and the accuracy of imitation of all Russian words, new and old, increased. Similar results were obtained for two additional subjects under these experimental conditions. In addition, the other two subjects were exposed to a condition in which reinforcement was delivered prior to their imitation of English words, rather than following correct English imitations. Percentage of English words correctly imitated and accuracy of Russian words imitated were markedly lower under these conditions than when reinforcement was contingent upon accurate English imitations. These studies were chosen to exemplify some of the characteristics of experimentally developed imitative repertoires. Other studies have used normal children who clearly had already existing imitative repertoires (Baer & Sherman, 1964; Brigham & Sherman, 1968; Burgess, Burgess, & Esveldt, 1970; Parton, 1970; Steinman, 1970; Waxler & Yarrow, 1970).
246
James A . Sherman
These studies have explored some of the experimental conditions which are related to the occurrence and nonoccurrence of both reinforced and unreinforced imitative behavior. In general, normal children also display unreinforced imitations when other imitative responses are produced in an experimental setting by reinforcement and/or instructions. On the basis of the studies cited, several summary statements can be made: ( a ) children who do not imitate behavioral demonstrations exhibited by a model can be taught to do so through the use of reinforcement procedures; ( b ) once the children have been taught to imitate a number of behavioral demonstrations, they exhibit an increased probability of imitating demonstrations which were not specifically taught; (c) some imitative responses can be maintained, even though unreinforced, as long as other, topographically different imitations are reinforced; and ( d ) the accuracy with which unreinforced demonstrations are imitated can be controlled by reinforcement operations performed on other imitative responses.
111. The Use of Imitation in Language Development The theoretical importance of imitation for the behavioral development of children has often been emphasized (e.g., Bandura, 1962; Bandura & Walters, 1963; Gewirtz & Stingle, 1968). One major reason for such an emphasis is to provide a learning mechanism for the rapid acquisition of novel behaviors by children. It is clear that novel behaviors can be produced in a child’s repertoire merely by providing an appropriate model for those behaviors, as well as by lengthy training methods such as those involved in shaping behavior (reinforcing successive approximations to some terminal state or topography). How and why imitative repertoires develop in children is not well understood. Several authors have presented theoretical accounts of the conditions under which imitative repertoires (both verbal and nonverbal) may be learned (N. E. Miller & Dollard, 1941; Mowrer, 1960; Risley, 1966). In addition, the studies cited previously (Baer et al., 1967; Lovaas et al., 1966; Metz, 1965) demonstrate learning methods which can produce imitative repertoires. The adequacy of learning accounts of the development of imitation in normal children depends upon a number of issues: ( a ) the amount of laboratory evidence supporting the assumptions of the theoretical accounts of imitative development; ( b ) the number and quality of studies demonstrating the development of imitative behavior in children lacking such repertoires; and (c) the degree to which variables or conditions demonstrated to be functional in experimentally producing imitative behavior are present in the “normal” environments of children. A number of these issues
Imiration and Language Development
247
with respect to the development of imitative vocal behavior have been discussed elsewhere (Siegel, 1969). Here, it will simply be assumed that most children do develop imitative repertoires (for whatever reason) in that they can and do match a wide variety of behaviors exhibited by other people in their environments. The question to be asked is how such an imitative repertoire may facilitate the development of language. Several accounts have emphasized the possible role of imitation in the development of phonetic systems, articulation, and vocabulary elaboration or labeling (e.g., Fry, 1966; Mowrer, 1960; Osgood, 1953; Peterson, 1968a). The basic reason for the emphasis on imitation as a process in the initial development of speech seems clear; it provides a mechanism or basis for the relatively rapid acquisition of new vocal responses. Other learning accounts of initial vocal development-such as those which simply cite differential reinforcement (or shaping) of speech by parents and caretakers of the child-appear to be unrealistic. First, the successive training of different sounds and words by the parents would simply take too long to account for their rather rapid development in young children. Second, parents do not appear to provide possible reinforcing consequences (e.g., social approval, attention) with sufficient consistency and regularity to establish the wide variety of vocal behaviors exhibited by, say, a twoyear-old child. In contrast, if a child does imitate some of the speech to which he is exposed, the problem of vocal acquisition seems considerable simplified. However, very young children typically do not exhibit perfect imitation of speech which is modeled to them. Frequently, young children apparently are able to imitate only portions of words or intonation patterns. What accounts for the increasing accuracy with which children can imitate? Fry ( 1966), among others, has suggested that various environmental events serve to increase the accuracy with which a child imitates vocal models. His account is as follows: The mother speaks to the child, using a word or expression appropriate to the situation; the child responds by imitating the word, and his own utterance thus begins to be associated with the situation. The sounds he makes will not be a very close copy at first, but the mother in her turn responds by repeating the word and thus repeatedly gives the pattern and usually shows her approval as the baby in the course of time gets closer to the pattern. In this way articulations that are not very close approximations to begin with are modified progressively until they at last become normal articulations for a given language [Fry, 1966, pp. 191-1921.
This account is similar to a shaping account in that it assumes that parents do provide differential consequences for speech that more closely approximates “correct” production. In addition, however, these differential conse-
248
James A . Sherman
quences are provided for more accurate imitative responses, which may have rather far-reaching effects. For example, in the studies cited in the previous section similar procedures were used to establish motor imitative responses in initially nonimitative children. After some amount of specific training, these children then began to imitate new motor responses the first time they were demonstrated and did not require specific training on these new responses. Further, the studies by Lovaas et al. (1966) and Brigham and Sherman (1968) showed that some verbal imitations, even though initially not perfect, improved in accuracy when other, already correct, imitations were reinforced. The way in which words or morphemes are ordered within an utterance by young children may also be analyzed, in part, as involving imitation. For example, Brown and Fraser (1964) recorded the utterances of young children ranging in age from 22 to 36 months old and attempted to develop provisional grammars which described these utterances. On the basis of their observations they suggested three characteristics of the children’s utterances. First, the order of words or morphemes in simple utterances of the children roughly corresponded to the order in which these words or morphemes might occur in an adult utterance. Second, the children were consistent in that certain words or morphemes were not exhibited in their spontaneous utterances, at a given level of verbal development. Third, when the children were asked to repeat sentences said by an adult, they consistently included some morphemes and excluded others, and these consistencies in imitated speech were similar to those observed in “spontaneous” speech. Thus, Brown and Fraser (1964) suggested that the utterances of young children can be characterized as grammatical adult sentences “from which certain morphemes have been omitted (p. 71 ) .” Brown and Bellugi ( 1964) have additionally suggested that parents often systematically provide adultlike models of speech to children by “expanding” a child’s utterance. In this case a child may emit an utterance, which although not grammatical by adult standards, nevertheless is understandable. The adult may then expand this utterance to correspond to a grammatical statement. An example reported by Brown and Bellugi is a child’s utterance of “baby highchair,” which was expanded by the mother into, “Baby is in the highchair.” This example, and many others observed in the normal interaction between mother and child, indicate that adults often systematically provide models for speech which may be important in the progressive development of the child’s speech toward more adultlike forms. However, while parents may systematically provide such models, there is no empirical evidence available to suggest that these models are important to the progressive development of children’s speech. In one of the few studies which has attempted to evaluate the possible function of
Imitation and Language Development
249
adult expansions, Cazden ( 1965 ) found that when immediate expansions of children’s speech were provided by adults, this did not result in more rapid development of adultlike speech forms, as compared with when grammatical models were provided noncontingently. There are two lines of descriptive evidence which have been put forward to suggest that imitation per se cannot account for progressive development of children’s speech. Ervin (1964) examined the spontaneous speech and the imitated speech of five children. A provisional grammar was then written which described or provided a possible set of rules categorizing the spontaneous utterances emitted by each child. Imitated utterances (immediate repetition of a model’s utterance) were then examined to see if they were grammatically consistent with the rules written for spontaneous utterances. Ervin reported that for four of the five children, the descriptive rules written for spontaneous utterances characterized the imitated utterances equally as well. Since imitations did not appear to be grammatically different from spontaneous utterances, Ervin suggested that immediate overt imitations do not appear to serve as a basis for progress in the development of grammatical skills. However, it is important to consider the criteria that might be used to judge the adequacy of any descriptive grammar. Brown and Fraser (1964), for example, attempted to specify grammatical rules describing the utterances of young children and pointed out the following: In our discussion of the techniques for the discovery of a grammar we have repeatedly pointed to the existence of equally reasonable alternative decisions. The arbitrariness of choice could be somewhat reduced by taking a larger corpus. However, we are sure that the best-founded grammar will not be uniquely determined but will only be a good provisional try [p. 681.
Given that a number of equally reasonable alternative decisions are available when constructing a grammar, Brown and Fraser (1964) suggest that one way of testing the adequacy or precision of the grammar is by looking at its predictive powers: the ability of the grammar to predict “sentences that are possible while not predicting sentences that are impossible [p. 691.” The ability to predict sentences that are possible simply refers to the power of the grammar to anticipate those sentences or utterances which have not yet occurred (and thus have not entered into the construction of the provisional grammar) but which will appear in the sample of utterances of the child if a second sample of utterances is taken in the near future. The restriction that the grammar not predict utterances that do not occur in subsequent samples seems equally important. As Brown and Fraser point out, if the grammar is written in the most general form, it will predict all (or most) of the utterances that occur in subse-
250
James A . Sherman
quent samples, However, since it will also predict many utterances that will not be obtained in subsequent samples of speech, its usefulness as a precise specification of a grammar is limited. Obviously, these restrictions are only rough criteria against which the adequacy of a provisional grammar can be evaluated. However, the findings of Ervin need to be evaluated within this context. If, in fact, the grammar derived from spontaneous utterances was written in too general a form, then it will of course encompass both spontaneous and imitated utterances, but it will also have inadequate sensitivity to reflect any possible differences in structure or form between these two classes of utterances or to detect any elaboration over time and/or experience. On this basis it would seem inappropriate to accept Ervin’s conclusion that the spontaneous utterances and imitated utterances of the sample of children she observed were not grammatically different. Alternatively, if the grammar derived from the spontaneous utterances was both precisely descriptive and predictive, then Ervin’s conclusion that imitated utterances do not appear to be grammatically different from spontaneous utterances would be justified. Unfortunately, the information available at this point does not appear to allow a clear choice between the alternatives. A second line of evidence which has been presented to suggest that simple imitation does not seem to account for or provide a clear basis for grammatical development, is simply that children produce language that apparently has not been modeled to them, but which nevertheless appears to be systematically related to the child’s past language exposure (Brown & Bellugi, 1964; Ervin, 1964). For example, children may exhibit “plural” forms such as ccfoots”and “mans.” Children may also exhibit “past tense” forms such as “buyed” and “corned.” In each of these cases it is unlikely that adults have modeled these forms to the child. Ervin (1964) stated: All of the children, over the period we studied them, regularized the plural for foot and man. They said man-mans, and foot-foots or feet-feets. Most preferred foot-foots. Very few of the children fluctuated between foot and feet, so although the word feet must have been heard by the children, we can clearly see a regularizing influence. If imitation alone were at work, we would have expected fluctuation between foot and feet [p. 17.51.
In addition, Ervin discussed the observation that plural forms regularly emitted by the child, such as “foots,” may be temporarily changed into forms such as “footsiz.” The fact that “even highly practiced, familiar plurals may be temporarily changed in form by overgeneralization of new patterns [p. 1771” suggests, to Ervin, that other processes outweigh simple imitative mechanisms. Additional evidence was cited by Brown and Bellugi (1964). They pointed to a series of utterances by two children which did not appear to be imitations. First, these utterances were ungrammatical,
Imitation and Language Development
25 1
and thus not likely to be modeled by adults. Second, it would be difficult to assume that these utterances represented reductions of adult utterances since it would be difficult to make simple grammatical sentences of the child‘s utterances by adding words or morphemes into the existing sequence of words in the utterances. The observations cited by Ervin (1964) and by Brown and Fraser ( 1964), as well as normative data collected by Berko ( 195 8), clearly indicate that children exposed to a language community develop language repertoires which extend beyond that to which they have been exposed. If this is correct, an account of language learning which rests solely upon imitative mechanisms and does not account for the development of a child’s “generative” ability appears to have serious difficulties. As stated by Brown and Bellugi (1964) : All children are able to understand and construct sentences they have never heard but which are nevertheless well formed, i.e., well formed in terms of general rules that are implicit in the sentences the child has heard. Somehow, then, every child processes the speech to which he is exposed so as to induce from it a latent structure. This latent rule structure is so general that a child can spin out its implications all his life long. It is both semantic and syntactic. The discovery of latent structure is the greatest of the processes involved in language acquisition and the most difficult to understand [p. 1511.
A number of authors have attempted to infer the “latent structure” of children’s language at various levels of complexity by observing and attempting to categorize the spontaneous utterances of young children (e.g., Braine, 1963; Brown & Fraser, 1964; Ervin, 1964; McNeill, 1966; W. Miller & Ervin, 1964). These observational studies have been concerned with descriptively specifying the development of initial grammars in children, primarily development of the “rules” which characterize the ways in which children combine words to form simple utterances. In the last few years, however, several investigators have approached the problem of the development of language systems in an experimental manner. That is, they specified a simple “rule governed” language system and, using subjects who did not initially display these types of language in their speech, attempted to train specific examples of the language form. In the process of this training, the investigators examined the possibility that subjects will begin to use new and untrained examples of the language form being trained. One of the first examples of the experimental development of language classes was provided by a study by Guess, Sailor, Rutherford, and Baer ( 1968). The purpose of this study was to teach the generative or productive use of plurality in labeling objects. A 10-year-old retarded girl was the
252
James A . Sherman
subject. The girl had been nonverbal, but prior to the beginning of the study she had been exposed to a verbal training program which had developed a vocal imitative repertoire and a small vocabulary of single words and simple phrases. At the beginning of the study, an experimenter showed the child a series of objects, singly and in pairs, and asked, “What do you see?” The child responded with singular labels to both single and pairs of objects. Following this, training was initiated. During training the child was taught to label single objects with a singular label and a pair of those objects with a plural label. This training was accomplished by using both imitation and differential reinforcement procedures in the following way. First, a single object was shown to the child and the experimenter asked, “What do you see?” If the child correctly labeled the object, reinforcement was delivered (praise from the experimenter and a bite of food). If the child labeled it incorrectly, or did not respond within a short period of time, the experimenter named the object correctly, withdrew it for a brief period of time, and then presented it again. This was repeated until the child labeled the object correctly. In the next stage of training, the procedure was repeated with a pair of the same objects. In this phase, however, a plural response from the child was required for reinforcement. As before, the experimenter provided a correct model (plural label) if the child did not respond or responded incorrectly. Finally, in the third phase, a random sequence of single objects and pairs of the objects was presented to the child until the child correctly labeled a single object with a singular label and pairs of the object with a plural label several trials in succession. Again, reinforcement was contingent upon correct labeling responses, and the experimenter modeled the correct response on trials when the child did not respond or responded incorrectly. Following this, identical training procedures were used to train correct singular and plural labels for a second object, and so forth. Each experimental session was devoted to the training of singular and plural labels for one type of object or item. Thus, in each session new labels were trained for new items. The purpose of the study was to attempt to train a simple generative language repertoire of plurality. The development of such a repertoire would be indicated if the child began to label correctly new pairs of objects without having been directly trained to emit that label in the plural and without having been exposed to models of that plural label by the experimenter. Thus, the child, having been taught the correct singular label for an object, might correctly label a pair of those objects the first time this pair was shown to her. Such a result is shown in Fig. 4. In this figure each session on the abscissa represents the training of one item for both singular and plural labels. The number of plural shifts listed on the ordinate refers to whether the child supplied the plural label correctly when shown the current training item as a pair the first time. Since only one
Imitation and Language Development
I
5
II
15
PO
25
X,
35
253
40
44
SESSlO NS Fig. 4 . Cumulative number of plural shifts ( f r o m Guess et al., 1968).
item, singly and in pairs, was trained in a session, a plural shift could occur or not occur only once each session. Plural shifts are plotted cumulatively: each increase in the graph shows the occurrence of a correct response the first time a particular pair of objects was presented. During the first condition of the study (I, reinforcement for normal, or conventional, plural usage) the figure shows that the child produced a correct plural label in session three when a new pair of objects was displayed to her. Further, she continued to produce correct plural labels during the next seven sessions when new pairs were shown to her. In order to evaluate the role that training played in developing the child’s generative plural labeling, the modeling conditions and the conditions for reinforcement were reversed. Now, the child was reinforced for stating a plural label when a single object was displayed to her and was reinforced for stating a singular label when a pair of objects was displayed to her. In addition, the experimenter modeled these responses to the child on trials when the child did not respond or responded “incorrectly.” With these exceptions, the training conditions were the same as those in the first condition and each session was devoted to the training of a new item. Figure 4 (Condition 11, reinforcement for reversed plural usage) shows that the subject did not label the pairs of objects “correctly” the first time they were presented for the first six training items, but then began to do so. All pairs after the ninth one were labeled correctly the first time presented in this condition. Following this, the original experimental conditions were reinstated and normal singular and plural labels were reinforced. As in the first application of this condition, the child emitted correct plural labels when shown a new pair of objects the first time. During later portions of the study, several types of new items were
254
lames A . Sherman
presented to the child to determine the extent of her generative plural labeling repertoire. In one condition, the child was shown three objects rather than the usual pair. In each case she labeled the three objects with an appropriate plural label. In another condition, words normally requiring irregular plural labels were presented to the child. When shown a picture of two men, the child responded “mans”; to a picture of two children she responded “childs” and to two leaves she responded “leafs.” A third condition was designed to investigate whether current training would affect previously trained responses. In this condition, normal or conventional singular and plural labels to single objects and pairs of objects respectively were being reinforced. Now, the child was shown single objects and pairs of objects to which she previously had been taught a reversed plural usage (single objects labeled with a plural label and pairs of objects labeled with a singular label). When these items were shown to the child, she exhibited singular and plural usage consistent with her current training history (single objects were labeled with a singular label and pairs of objects were labeled with a plural label) rather than the reversed usage which had been specifically taught to these objects earlier. Two additional studies (Guess, 1969; Sailor, 1969) have extended the generality of these findings. Guess (1969) examined the relationship between receptive speech or auditory comprehension and productive speech. In the process of this analysis, he demonstrated the establishment of a generative repertoire of plurality in two retarded boys using imitation and differential reinforcement procedures. Sailor ( 1969) extended the results of the Guess et al. (1968) study by showing the training of differential inflectional patterns within the response class of plurality with two retarded girls who previously did not exhibit plural usage. In conventional English speech, the unvoiced inflection I-sI is required to form the pluraI of most singular nouns ending in unvoiced sounds or phonemes and the voiced inflection 1-zI is required to form the plural of most nouns with voiced endings. In the Sailor study, one subject was first exposed to models and was reinforced for labeling plural items requiring the 1-s( inflection and then was exposed to models and was reinforced for labeling plural items requiring the 1-21 inflection. The second subject was exposed to the same conditions but in a reversed order. After each subject had been trained to a criterion performance within a condition, she was exposed to new plural items which, in conventional English speech, required the inflection not currently being trained. The results obtained showed a clear generalization of training, in that the subjects labeled plural objects with the inflectional form currently being trained. That is, if a subject was currently being trained with nouns requiring the I-sI inflection, she pluralized other nouns by adding an I-sI inflection even though conventional English
Imitation arid Language Developnieirt
255
speech requires a I-zI inflection. Conversely, if a subject was currently being trained with nouns requiring a 1-21 inflection to form a plural, she pluralized other items by adding a 1-21 inflection even though conventional English speech requires the I-sI inflection. Furthermore, the effects of current training were sufficient to overcome previous training. For example, a subject who had been trained to pluralize nouns requiring an I-s/ inflection would pluralize these same nouns by adding a 1-21 inflection if she currently was being exposed to training which involved nouns requiring the I-zI inflection to form plurality. Thus, with five different subjects who initially did not display a generative speech repertoire of plurality, the training conditions employed in these three studies were sufficient to produce such a repertoire. In each study, the basic procedures used were demonstrations or models provided by the experimenter and differential reinforcement of the child’s “correct” verbal responses. In each case, the training of some subset of items produced a generalization effect such that the child responded to new items in a way consistent with the “rules” characterizing the current training conditions. Furthermore, in two of the studies (Guess et al., 1968; Sailor, 1969), the way in which an item was pluralized by children could be changed by placing this item in a context where a reversed or alternative form was being trained to other items. Thus, the form of previously learned responses could be modified by training which did not directly involve that response. It is perhaps not surprising that a great deal of recent work in the experimental establishment of generative language repertoires has been confined to the development of plurality. Plurality represents one of the simplest yet appropriate langauge tasks for investigations of this type. However, six additional studies, three involving sentence usage, two involving word combinations, and one involving verb inflections, have extended the generality of the results obtained with plurality. These studies show, for several language forms, that a child’s prior training and/or exposure to language can produce more than that which has been directly trained and/or modeled to the child. Wheeler and Sulzer (1970), with a child who initially spoke “telegraphic” English, used imitation and reinforcement procedures to train the child’s usage of complete sentences in describing pictures. The subject in this study initially spoke in “sentences” which did not contain most articles and auxiliary verbs. In the experimental training procedure, the child was shown pictures and asked, “What do you see?” Appropriate sentences, or portions of the sentences which contained articles and auxiliary verbs were modeled to the child, and reinforcement (tokens which could be traded for various activities) was contingent upon the child’s
256
James A . Sherman
imitation of the model’s statement or correct production of a complete sentence without an imitative prompt. Under these conditions, Wheeler and Sulzer found that the child began to produce more nearly complete sentences (containing articles and auxiliary verbs) without imitative prompts to pictures on which the child had been trained as well as to pictures for which the child had received no direct training. When the child was then reinforced for “telegraphic” sentence forms and these were modeled to him, he again began to use more sentences of this type without imitative prompts, when describing pictures on which he had received training as well as pictures for which he had received no direct training. A return to the original training conditions produced more nearly complete unprompted sentences to trained as well as to untrained pictures. Fygetakis and Gray (1970) observed the speech of aphasic children during periods in which the children showed their toys to a teacher and talked about the toys ( a “show and tell” period). During these times, it was observed that the children used few descriptive or interrogative sentences containing the verb “is.” The children were then exposed to a training procedure in which the children first imitated portions of sentences and then imitated entire sentences which contained the verb “is.” Then the children were required to produce sentences containing “is” without an immediately preceding demonstration by the teacher. The children’s behavior was reinforced with stars which later could be exchanged for toys (also see Gray & Fygetakis, 1968a, 1968b). After the training program, the children exhibited an increased amount of “is” sentences in subsequent “show and tell” periods. Similarly, when the children had been exposed to the training program in which they first imitated and then produced interrogative sentences containing ‘‘is,’’ they showed an increased number of sentences of this form during subsequent “show and tell” periods. Bandura and Harris ( 1966) investigated the role of imitation, reinforcement, and instructions in increasing the number of passive constructions and prepositional phrases in sentences produced by normal second-grade children. The children in this study were initially shown cards with a word on each and were asked to make up a sentence containing the word (baserate measure), Following this, different groups of children were exposed, for short periods, to various experimental conditions involving three variables: ( 1 ) modeling, in which an adult modeled sentences (containing passive constructions or prepositional phrases) preceding the child’s opportunity to respond to cards; (2) reinforcement, which consisted of social praise (e.g., “very good”) and presentation of a star contingent upon a child’s production of a sentence containing a passive construction or a prepositional phrase; stars could be traded after sessions for a “present’’; (3) instructions, which involved telling a child to pay close
Imitation and Language Development
251
attention to the sentences which did and did not produce a star and asking the child to repeat sentences which did produce a star. The children in the study initially produced very few sentences containing passive constructions. Subsequently, however, the group of children exposed to the experimental condition containing all three variables (modeling, reinforcement, and instructions) produced significantly more passive constructions than did children exposed to none of the experimental conditions (control group) as well as children exposed to conditions containing modeling alone, reinforcement and instructions, and modeling and reinforcement. Nevertheless, even the children exposed to the experimental conditions containing modeling, reinforcement, and instructions, produced very few sentences containing passive constructions (a mean of 2.80 per child out of 20 opportunities). Sentences containing prepositional phrases were exhibited with a much higher frequency by the children. The experimental condition containing reinforcement and instructions and the one containing modeling, reinforcement, and instructions were more effective in producing these sentences than were the conditions containing none of the conditions, modeling and reinforcement, or modeling alone. Bandura and Harris additionally noted that the passive sentences generated by the children were varied and only rarely duplicated the production of the model. Thus, they suggested that the procedures were effective in generating novel combinations of responses. Hart and Risley (1968) recorded the speech of disadvantaged preschool children during periods of free play activity and found that these children exhibited a low frequency of color adjective-noun combinations in their speech. Play materials during free play were then made contingent upon the children’s production of a color adjective-noun combination. In addition, for the first few days of the procedure, the teachers prompted such combinations by asking appropriate questions or by modeling responses when a child did not respond or responded incorrectly. These conditions produced a large increase in the number of color adjective-noun combinations including a number of combinations which had never been recorded in the speech of the children previously. When play materials were no longer contingent upon color adjective-noun combinations, the children showed a decreased number of such combinations; but this number, including new combinations, was considerably above that seen initially. Similar findings have been reported by Hart (1969) in an extension of this research. Schumaker and Sherman (1970) have shown the training of generative speech repertoires involving past and present (present progressive) tense verb inflections. In this study retarded children were used as subjects. (For the purpose of this discussion, the procedures and results for only two of
258
lames A . Shermari
the subjects will be described. A third subject was also employed, although the procedure for this subject was slightly different.) Prior to the study, the children appeared to exhibit little or no appropriate usage of regular past tense verb forms in their speech. In spoken English, three types of regular endings or inflections can be added to verb stems to form the past tense: 141, I-dl, or I-adl. Which one is appropriately added depends upon the sound or phoneme terminating the verb stem. Verb stems ending in most voiceless phonemes require the I-tl inflection (e.g., baked, stopped, touched). Verb stems ending in most voiced phonemes require a I-dl inflection (e.g., climbed, played, smiled). Verb stems ending in either It1 or [dl require an I-adl inflection (e.g., painted, graded). For experimental purposes four classes of past tense usage were designated, verbs requiring a I-tl inflection; verbs requiring a I-dl inflection; verbs ending in It1 and requiring an [-ad[ inflection, and verbs ending in (dl and requiring an (-ad( inflection. (The verbs requiring an /-ad[ inflection were divided into two separate classes because it was found during initial training that considerable interaction occurred between verbs ending in It1 which required an [-ad[ inflection and those ending in Id1 which also required an /-ad/ inflection.) The present progressive tense involves the addition of “-ing” to all verb stems, but four present-tense classes were designated to correspond to the four past-tense classes. After an initial pretest to determine that the subjects did not exhibit regular past-tense verb forms in their speech, the experiment began. There were two types of experimental sessions: training sessions and probe sessions. During training sessions a child was taught to use a particular verb in both its present- and past-tense form. In training, the experimenter modeled correct responses to the child if the child did not respond correctly, and reinforced correct productions of the child with praise and poker chips (poker chips could be exchanged after sessions for a small toy or candy). When a child correctly produced both the present- and past-tense forms of the verb several times on successive trials without requiring the experimenter’s model, a probe session was scheduled. The purpose of the probe session was to determine whether training would generalize to the production of present and past-tense forms of verbs that had never been specifically trained. During probe sessions, verbs that previously had been trained were presented to a child as in the training sessions, and correct responses were reinforced (with praise and poker chips). In addition, other verbs on which the child had received no direct training were presented interspersed among the trained verbs. No responses to the untrained or probe verbs were reinforced whether the responses were correct or not, and no models were provided for these verbs.
Imitation and Language DeveIopment
259
The overall design of the study involved multiple training conditions or multiple baselines (Baer, Wolf, & Risley, 1968). In the first training condition, verbs requiring a particular inflection for the formation of the past tense were trained, one at a time, in both the past and present tense. Following the training of each verb, a probe session was scheduled. Within the probe session, all verbs which had been previously trained within that class were interspersed among novel untrained verbs which required each of the four types of inflections to form the past tense. Basically, then, what was done was to train a series of verbs within one of the classes and schedule probe sessions to determine the generalized effect of this training on novel probe verbs from all four of the classes. Next, a series of verbs within a second class was trained and probe sessions were scheduled to determine the generalized effect of this training. Then a discrimination between verbs in the two previously trained classes was taught, and the effects were evaluated during probe sessions. The results of the procedure are displayed in Fig. 5 , which shows the performance of one of the children on the untrained probe words over the various conditions of the study. Each set of axes shows the percentage of correct past-tense responses for untrained verbs within one class. The top set of axes shows correct past tense production of untrained verbs ending in It1 which require an [-ad\ inflection (labeled “TED”). The second set of axes shows correct past tense production of untrained verbs ending in Id1 which require an /-ad1 inflection (labeled “DED”). The third and fourth set of axes show correct past tense productions of untrained verbs requiring I-tl and I-dl inflections respectively (labeled “T” and “D”). The various training conditions of the study are marked at the top of the figure. Where training of verbs within a particular class occurred (and thus these words were interspersed among untrained verbs during probe sessions) the data points are connected by a dashed line. Otherwise, the data points are connected by a solid line. Thus, Ruth was taught the correct past-tense production to a series of “TED’verbs in Condition I, “DED7 verbs in Condition 11, and “T” verbs in Condition IV. In Conditions I11 and V, training sessions were devoted to discrimination training between verbs in two or more classes. In Condition 111, verbs within the “TED” and “DED” classes were trained. In Condition V, verbs within the “TED” “DED’ and “T” classes were trained. The results of the training seemed clear: as verbs within a single class were trained, Ruth increasingly showed correct past-tense production to untrained verbs within that class. When verbs from two or more classes were trained, she showed correct past-tense production to untrained verbs within these classes. Figure 6 shows similar data for another child in this study. For this
260
lames A . Sherman
PROBE SESSIONS Fig. 5 . Percent correct past-tense responses to untrained verbs (from Schumaker & Sherman, 1970). See text for further explanation of abbreviations.
child the sequence of training was as follows: Condition I, training on verbs within the “TEDyclass; Condition 11, training on verbs within the “T” class; Condition 111 discrimination training of verbs within the “TED” and “T” classes; Condition IV, training on verbs within the “D” class. The results with this child were similar to the results for the first, in that
Imitation and Language Development
PATTY
W
cn
z W
I-
& 8 I
cn W cn Z
2
cn w LT I-
8 LT
26 1
CONDITIONS
L
100
40
u) 0
LT
0 0 I-
z W
0 LT
W
a DE D
30
35
4(
-?45
PROBE SESSIONS
Fig. 6 . Percent correct past-tense responses to untrained verbs ( f r o m Schumaker & Sherman, 1970). See text f o r further explanation of abbreviations.
262
James A . Sherman
increases in the percentage of correct past-tense responses to untrained verbs were closely correlated with the type of training the child was currently receiving. The data of both subjects showed an additional noteworthy characteristic: a decline in the percentage of correct past-tense usage on untrained verbs when training began on a new class of verbs. For example, with Patty, after a number of words from the “TED” class had been trained and generalized effects of this training had been observed, training was started on verbs from the “T” class. After several verbs from the “T” class had been trained, Patty exhibited correct past-tense production to untrained verbs in the “T” class. However, her percentage of correct past tense responses to the untrained verbs from the “TED’ class dropped to zero, even though she had exhibited a large number of correct responses to verbs from this class in the previous condition. What apparently had happened was that the past-tense inflectional form currently being trained with some verbs was used to form the past tense, not only of verbs within that class but also for verbs in the other classes. Thus, a verb such as “paint” might be correctly produced in the past tense form during Condition I as “painted.” However, during Condition 11, when verbs within the “T” class were trained, the past tense form of “paint” would be incorrectly produced as “paint.” Similar effects were observed for the untrained probe verbs within the classes for which no examples had yet been trained. Thus, during training of verbs from the “T” class, a past tense form of “play” (a verb from the “D” class) might be “playt,” a form which would be scored as an incorrect production. These types of productions were scored as “overgeneralization” of the trained form, an effect which’ was consistently obtained with all three of the subjects. It should be noted, however, that during the discrimination training conditions, correct production to untrained verbs within two or more classes could be produced. For example, during Condition I11 for Patty, examples of verbs within the “TED” and “T” classes were concurrently trained. This produced correct past tense production to untrained verbs within both the “TED’ and the “T” classes. Thus, the “overgeneralization” effect of training within one class could be overcome by concurrent training of examples within two (or more) classes. The study by Schumaker and Sherman provides two lines of evidence which support the idea that the items trained had an effect upon pasttense production of untrained verbs. The first is that the percentage of correct past-tense productions of untrained verbs gradually increased only when correct past-tense productions were trained to other verbs from the same class. The second was the occurrence of “overgeneralization” of the past tense inflectional form currently being trained to verbs which normally
Imitation and Language Development
263
require other types of past-tense inflections. Thus, the results showed that the children’s language extended beyond that to which they had been trained or exposed (at least experimentally), but that these extensions were closely related to the past experimental histories of each child. For the purposes of this discussion, the generalization of trained pasttense inflectional forms to novel or untrained verbs has been emphasized. It should be noted that measures were obtained of both the correct pastand present-tense productions of the verbs in the probe sequence which had been directly trained and of correct.production of the present tense forms of untrained verbs from each class. In general, the following results were obtained: percentage of correct past- and present-tense productions to the trained verbs which were included in the probe sessions was high, even during the initial stages of a condition when a child exhibited little generalization of an inflectional form to untrained verbs; correct production of the present-tense forms of untrained verbs was closely correlated with the type of verbs being taught during training sessions and, thus, formation of the present tense also appeared to show the development of a generative or productive speech repertoire.
IV.
Summary and Discussion
The experimental studies described in the previous two sections represent a selected sample of recent attempts to teach or train imitative and language repertoires in children. These studies showed that teaching a series of specific stimulus-response relationships may result in a behavioral repertoire which goes beyond those stimulus-response relationships which were specifically taught. This was seen in the studies of imitation when children imitated new behavior demonstrated by the model, even though these responses had never specifically been trained. Further, these nevertrained imitations continued to be exhibited even though they were never reinforced, as long as other, topographically different imitations were reinforced. Several studies were described which showed the establishment of generative repertoires of plurality, simple sentences, and verbtense usage. In these studies, children were taught, by imitation and differential reinforcement procedures, to respond verbally to a series of specific stimulus cues. On the basis of this training, the children were able to produce new and appropriate responses to new stimulus cues. Thus, after training, children appropriately pluralized new items which had been labeled for them only in the singular; similarly, after training, children produced appropriate present- and past-tense forms of new verbs presented to them.
264
James A . Sherman
These results are, perhaps, not surprising. There is a large body of experimental literature available showing that humans, as well as infrahuman subjects, show successive improvements in performance when exposed to a series of related training tasks. This effect has been referred to as learning set (e.g., Harlow, 1959). The studies of imitation cited in this paper can be regarded as examples of a particular type of learning set, one in which a series of tasks exemplifying the “rule,” “Do as the model does,” are trained. Over a series of these tasks the children begin to “DO as the model does” for behaviors which have not been specifically included in the previous training. Similarly, the generative production of novel plural forms, novel word combinations and sentences, and novel presentand past-tense forms exemplifies “rule” governed behavior which is related to the type of training to which the children have been previously exposed. However, in almost all learning set studies the responses emitted by subjects had limited topographical variation, in that a simple response such as pressing a button, lifting a food cup, or pointing was required. These learning set studies showed that the number of trials required for a specified criterion performance for these topographically simple responses could be reduced by exposing the subjects to a series of related tasks. In the studies of the development of imitation, plurality, word combinations and sentences, and use of verb inflections, the responses were not so topographically simple. Instead, the topographies of the responses varied as did the stimulus tasks presented to the subjects. Thus, the recent research on imitation and generative language behavior may be viewed as an extension of previous research on simple learning sets. One of the objections to learning theory accounts of complex behavior, and in particular of the development of language, is that these accounts cannot deal with the occurrence of novel behavior, especially novel behavior which can be described as “rule governed.” However, the studies cited concerning the development of imitation seem to exemplify precisely this type of phenomenon: the occurrence of the novel “rule-governed” behavior. Furthermore, the development of this behavior could be directly related to rather simple training operations: the systematic use of differential reinforcement and fading procedures to teach the children to match a series of different behaviors of a model. This then may provide the simplest account of the occurrence of novel behaviors: the past history of the subjects with respect to related behaviors (Baer, Guess, & Sherman, 1969; Peterson, 1968a). Given that a child has acquired a widely generalized imitative repertoire, it seems possible that his language behavior could be markedly influenced by the language models to which he is exposed. However, there are several objections to proposing an imitative basis for the development of new
Imitation and Language Development
265
language forms. The primary one, as discussed earlier, is that children may construct or emit a variety of verbal responses that they have never heard. If one assumes that imitation plays an important role in language development, then the child‘s emission of verbal responses to which he has not been exposed presents a serious problem. However, experimental analogues of this type of phenomenon are provided by the studies of the development of plurality, simple sentences, and verb inflections. In these studies, children were taught, by imitation and differential reinforcement procedures, to respond verbally to a related series of stimulus situations. In the process of this training, the children learned to produce novel responses which had never been directly taught or modeled for them. Again, an analysis of the development would involve pointing to the past history of the subjects on related responses. A second characteristic of several of the studies described in this paper should be noted again here: the modification of previously learned responses by ongoing training conditions. In the speech of normal children, Ervin (1964) noted that previously learned responses were changed by “overgeneralization of new patterns [p. 1771.” This type of phenomenon was also noted in the studies of the development of plurality and verb inflections. Three examples can be cited. In the study by Guess et al. (1968), during one condition the child was exposed to training procedures which produced a generative repertoire of reversed plural labeling (plural labels to single objects and singular labels to pairs of objects). In a later condition, the child was trained, with different objects, to label a single object with a singular label and pairs of objects with a plural label. Now, when the child was shown objects to which she had been previously taught reversed plural labeling, she did not display the labeling behavior previously trained but rather displayed labeling behavior which was consistent with the current training conditions. That is, to these previously trained objects, she labeled single objects with singular labels and pairs of objects with plural labels. Sailor (1969) found that the way in which nouns were pluralized could be altered by current training conditions. In this study, subjects were trained to label pairs of objects with a plural noun which included either an I-sI or a I-zI inflection. After subjects showed a generative repertoire of plural labeling for one of these inflections, the training conditions were changed so that the plural label required the other inflection. Now, when the earlier objects were shown to the children, the plural labels emitted by them included an inflection which was consistent with the current training conditions rather than an inflection which previously had been exhibited. Schumaker and Sherman (1970) showed that correct past-tense responses to untrained verbs was lost when the training conditions involved teaching past-tense responses to verbs from another
266
James A . Sherman
class. In this case, the reduction in correct past-tense responses to untrained verbs within a class was correlated with “overgeneralization” of the inflectional form characteristic of the class currently being trained. Thus, these studies show that specific training can create correspondingly specific generative speech repertoires in language-deficient children, and that responses within these repertoires can be modified by subsequent training procedures which do not directly involve these responses. In most of the studies of the development of generative speech repertoires, the training conditions involved the joint use of differential reinforcement and imitation. In none of the studies with language-deficient children was there an attempt to separate or evaluate the relative contribution of these two procedural components. Thus, the development of the generative repertoires might have been a function of differential reinforcement alone, modeling alone, or a combination of the two. It should be noted, however, that the training procedures would have been made considerably more difficult if modeling techniques had not been used. If, for example, an experimenter had to rely exclusively upon differential reinforcement, it would have meant that each response exhibited by the child would have had to be trained by shaping procedures, involving reinforcement for behavior closer and closer to a target response. Given that the children had imitative repertoires, this would seem to be highly inefficient, since a correct response could be produced merely by providing an appropriate model or demonstration of the response. Thus, while it is impossible to determine the separate contribution of differential reinforcement and imitation to the development of generative language repertoires in these studies, it would seem, at a minimum, that the use of imitation procedures facilitated the training process. Bandura and Harris (1966), studying normal children, did attempt to investigate the effects of modeling and reinforcement separately, but their results varied, depending upon the response form manipulated. Furthermore, since their subjects were exposed to very short training conditions, it is difficult to extrapolate their results to other studies. While several of the studies cited have shown experimentally the development of generative language repertoires, it is important to add several qualifications. First, the speech repertoires trained represent only very simple kinds of generative language (plurality: Guess, 1969; Guess et al., 1968; Sailor, 1969; word combinations: Hart, 1969; Hart & Risley, 1968; verb inflections: Schumaker and Sherman, 1970; more complete usage of simple sentences: Bandura & Harris, 1966; Fygetakis & Gray, 1970; Wheeler & Sulzer, 1970). Second, in the majority of the studies, the use of these generative repertoires was restricted to carefully controlled experimental tasks and no attempt was made to evaluate the subjects’ usage of the experimentally trained language in nonexperimental situations or in
Imitation and Language Development
267
situations which might approximate “normal” conversation. [Exceptions to this statement are found in the studies by Fygetakis and Gray (1970), Hart and Risley (1968), and Hart (1969). In the latter two studies, language training took place in the “normal” environment in which it was to occur.] In large part, these restrictions seem to represent simply the preliminary nature of most of the recent studies. At this time, there is no logical reason to expect that the development of more complex forms of generative language repertoires would not yield to the same type of experimental analysis. Nor are there logical reasons to suppose that experimentally developed generative language repertoires would not be used by subjects in nonlaboratory situations and in “normal” conversation, given the use of appropriate training conditions. Nevertheless, a number of demonstrations of the development of more complex forms of language and its use in situations approximating “normal” speech remains to be accomplished and requires careful specification of what is it that is to be trained as well as specification of the situations and contexts in which it is to be exhibited. These demonstrations, if provided, should begin to specify the extent (and the limits) to which imitation and reinforcement procedures may be used to modify the language of speech-deficient children. A third limitation on the generality of the conclusions drawn from the studies cited, is based upon the type of subjects employed. Except in the study by Bandura and Harris (1966), the subjects used exhibited impoverished speech prior to the start of the study. Thus, they were children who, for unknown reasons, failed to develop or use language in the “usual” manner. It was for exactly this reason that they were selected as subjects in the studies. However, this raises a question: Would a variety of “normal” children who lacked such speech repertoires show similar results if exposed to the types of training procedures used in these studies? To answer this question for simple generative speech repertoires, it seems likely that it will be necessary to use very young children, for only young children may show the necessary deficits in speech to be appropriate subjects. The use of normal children, however, may involve problems of demonstrating experimental control. Eventually, these children can be expected normally to develop generative language repertoires without any special training procedures. Thus, within any experimental training program it will be necessary to use methods to separate out the contribution of the training program from the normal development of these speech responses which would be expected to occur independent of the experimental training program. However, the results of the study by Bandura and Harris (1966) suggest that even relatively short-term procedures with normal children may produce an effect upon the speech repertoires of the children. The fact that procedures involving differential reinforcement and imi-
268
James A . Sherman
tation have been shown to be functional in establishing generative speech repertoires in deviant children does not establish that these variables are responsible for the development of such language behavior in normal children. While it seems likely that models for appropriate speech are present in the environments of most children and that events which may serve as reinforcement occur following appropriate speech, it has not been demonstrated that these variables affect the normal development of language. Within this context it should be noted that some of the subjects in the experimental studies described were exposed to systematic and lengthy training conditions before they exhibited generative speech. It does not seem likely that children in normal environments are exposed to such systematic conditions. However, the children in most of these experimental studies were selected because of their failure to develop speech normally. Perhaps it is the case that these children simply require more extensive and systematic training conditions than would normal children. Nevertheless, it remains to be seen how important differential reinforcement and/or imitation are for the normal development of speech repertoires. Such an evaluation will require information from several sources. First, it will be necessary to obtain precise and reliable observations of the development of various language classes in normal children, the language models to which these children are exposed, and the environmental consequences of children’s language usage. Some of this information ,is currently available. However, an increasing emphasis needs to be placed upon observational reliability. For example, in almost all of the experimental studies described, an attempt was made to evaluate the degree to which independent observers agreed or disagreed on the occurrence of specific responses. Typically, the amount of agreement was high, probably reflecting, in part, the fact that the training procedures produced clearly articulated, identifiable responses. The problem of observational reliability seems more difficult to handle in “real life” situations where recording is more difficult and the articulation of words (particularly by very young children) is highly variable. However, if we are to obtain precise descriptive information, it seems necessary to develop methods to estimate the reliability of the recording methods; otherwise, it will not be clear whether reported observations of children’s language describe the behavior of children or are simply the autisms of observers of children’s language. Second, to investigate the possible causal nature of reinforcement and/or imitation in the development of normal speech, it seems necessary to vary systematically the kinds of language models presented to children and the consequences of their speech in the “normal” environment. In the experimental studies with speech-deficient children, the functional effectiveness of the procedures in developing generative speech was evaluated in
Imitation arid Language Development
269
several ways. One strategy was to attempt to show that the children initially lacked a particular speech repertoire but, following some amount of training, did exhibit the repertoire. However, this is not a sufficient demonstration, since it was possible that the children might have developed the particular language repertoire independently of the experimental procedure (in spite of the fact that they apparently had not done so earlier in life). Or, it is possible that these speech repertoires had been previously developcd although, for various unknown reasons, not exhibited during the early testing conditions of the study. Two control procedures for demonstrating the relationship between the experimental procedures and the development of speech repertoires were used. One was to change the current training task. For example, in the Guess et al. (1968) study, initially appropriate singular and plural labeling was trained. Following this, reversed singular and plural label was trained. The fact that the subject displayed behavior appropriate to each of these training conditions strongly indicated that the training procedures, and not other variables, controlled the generative speech repertoire. The study by Sailor (1969) and Wheeler and Sulzer ( 1970) used similar procedures to demonstrate experimental control over the language repertoires exhibited by the children in their studies. A second method of showing the relationship between the experimental procedures and the development of language repertoires was the use of multiple baselines. This approach was exemplified in the study by Schumaker and Sherman ( 1970). In this study, several different classes of past-tense usage were designated. Then, examples from one class were trained, next examples from a second class were trained, then examples from the two classes were trained, then examples from a third class were trained, and so forth. The fact that the children initially displayed generative performance to untrained verbs within a class only when other verbs from that class were trained indicated that the training procedures, rather than other variables, were responsible for the development of the particular speech repertoires. (It might be noted that when verbs from a second class were trained, previously correct performance to untrained verbs from the first class declined. This result, as in the other studies, provided additional evidence for the causal relationship between language behavior and the experimental procedures.) The control procedures discussed above are appropriate for demonstrating a functional relationship between experimental procedures and the development of behavior for an individual child. An alternative, and more conventional strategy, as exemplified in the study by Bandura and Harris (1966), is designed to show the functional effect of a variable or set of variables on the average performance of a group of children. In the study by Bandura and Harris, one group of children (the control group) per-
270
James A . Sherman
formed the experimental task but was not exposed to the variables of modeling, reinforcement, and instructions, combinations of which were used in the various experimental groups. Th,e use of experimental designs of this type demands the availability of a large number of subjects. When this large number is not available, or when an investigator wishes to establish the functional effectiveness of procedures in producing a language repertoire for an individual child, it will be necessary to employ experimental designs appropriate to individual subjects. In addition, where the training procedures require lengthy time periods, it may be more efficient to use experimental designs appropriate to individual subjects. Nevertheless, for the investigation of variables responsible for normal language development of children, it seems necessary to use one or more of these control conditions or control groups to separate the functions of the experimental manipulations from the possible functions of other variables in the child’s current environment. Such designs may show us, with confidence, that generative language can be produced by simple, well-known procedures, mirroring at least some of the “natural” conditions present in the environments of children. On the basis of the experimental evidence to date, it would seem reasonable to suggest that imitation and reinforcement offer a potentially important source of variables related to the development of language repertoires. The use of imitation and reinforcement has already been fruitful in the training of language-deficient children. Perhaps similar variables may be useful in the understanding of the “normal” development of language. At this moment, it does not seem possible to subject such a suggestion to any single crucial test which would distinguish between it and other suggestions concerning the normal development of language. Rather, what seems to be needed is an accumulation of evidence from various sources which will support, modify, or eliminate its feasibility. REFERENCES Aronfreed, A. The problem of imitation. In L. P. Lipsitt & H. W. Reese (Eds.), Advances in child developmerit and behavior. Vol. 4. New York: Academic Press, 1969. Pp. 209-319. Baer, D. M., Guess, D., & Sherman, J. A. Adventures in simplistic grammar. Paper presented at seminar of Research in Language of the Retarded, Lawrence, Kansas, February 1969. Baer, D. M., Peterson, R. F., & Sherman, J. A. Development of imitation by reinforcing behavioral similarity to a model. Journal of the Experimental Analysis of Behavior, 1967, 10, 405-416. Baer, D. M., & Sherman, J. A. Reinforcement control of generalized imitation in young children. Journal of Experimental Child Psychology, 1964, 1, 3 7 4 9 .
Imitation and Language Development
27 1
Baer, D. M., Wolf, M. M., & Risley, T. R. Some current dimensions of applied behavior analysis. Journal of Applied Behavior Analysis, 1968, 1, 91-98. Bandura, A. Social learning through imitation. In M. R. Jones (Ed.), Nebraska symposium on motivation. Lincoln, Neb.: University of Nebraska Press, 1962. Pp. 211-269. Bandura, A., & Harris, M. B. Modification of syntatic style. Journal of Experiniental Child Psychology, 1966, 4, 341-352. Bandura, A., & Walters, R. H. Social learning and personality development. New York: Holt, Rinehart & Winston, 1963. Berko, J. The child’s learning of Englirh morphology. Word, 1958, 14, 150-177. Braine, M. D. S. The ontogeny of English phrase structure: the first phase. Language, 1963, 39, 1-13. Brigham, T. A., & Sherman, J. A. An experimental analysis of verbal imitation in preschool children. Journal of Applied Behavior Analysis, 1968, 1, 151-158. Brown, R., & Bellugi, U. Three processes in the child’s acquisition of syntax. In E. Lenneberg (Ed.), N e w directions in the s/udy of language. Cambridge, Mass.: M.I.T. Press, 1964. Pp. 131-161. Brown, R., & Fraser, C. The acquisition of syntax. In U. Bellugi & R. Brown (Eds.), The acquisition of language. Monographs of the Society for Research in Child Development, 1964, 29, 43-79. Burgess, R. L., Burgess, J. M., & Esveldt, K. C. An analysis of generalized imitation. Journal of Applied Behavior Analysis, 1970,3, 39-46. Cazden, C. B. Environmental assistance to the child’s acquisition of grammar. Unpublished doctoral dissertation, Harvard University, 1965. Ervin, S. M. Imitation and structural change in children’s language. In E. Lenneberg (Ed.), N e w directions in the study of language. Cambridge, Mass.: M.I.T. Press, 1964. Pp. 163-189. Flanders, J. P. A review of research on imitative behavior. Psychological Bulletin, 1968, 69, 316-337. Fry, D. B. The development of the phonological system in the normal and the deaf child. In F. Smith & G. A. Miller (Eds.), The genesis of language: A psycholinguistic approach. Cambridge, Mass.: M.I.T. Press, 1966. Pp. 187-206. Fygetakis, L., & Gray, B. B. Programmed conditioning of linguistic competence. Behaviour Research and Therapy, 1970, 8, 153-163. Gewirtz, J. L., & Stingle, K. G. Learning of generalized imitation as the basis for identification. Psychological Review, 1968, 75, 374-397. Gray, B. B., & Fygetakis, L. Mediated language acquisition for dysphasic children. Behaviour Research and Therapy, 1968,6, 263-280. (a ) Gray, B. B., & Fygetakis, L. The development of language as a function of programmed conditioning. Behaviour Research and Therapy, 1968, 6, 455-460. (b) Guess, D. A functional analysis of receptive language and productive speech: Acquisition of the plural morpheme. Journal of Applied Behavior Analysis, 1969, 2, 55-64. Guess, D., Sailor, W., Rutherford, G., & Baer, D. M. An experimental analysis of linguistic development: The productive use of the plural morpheme. Journal of Applied Behavior Analysis, 1968, 1, 297-306. Harlow, H. F. Learning set and error factor theory. In S. Koch (Ed.), Psychology: A study of a science. Vol. 2. New York: McGraw-Hill, 1959. Pp. 492-537. Hart, B. M. Investigations of the language of disadvantaged preschool children. Unpublished doctoral dissertation, University of Kansas, 1969.
272
James A . Sherman
Hart, B. M., & Risley, T. R. Establishing use of descriptive adjectives in the spontaneous speech of disadvantaged preschool children. Journal of Applied Behavior Analysis, 1968, 1, 109-120. Hewett, F. M. Teaching speech to an autistic child through operant conditioning. American Journal of Orthopsychiatry, 1965, 35, 927-936. Hingtgen, J. N., Coulter, S. K., & Churchill, D. W. Intensive reinforcement of imitative behavior in mute autistic children. Archives of General Psychiatry, 1967, 17, 36-43. Lovaas, 0. I. A program for the establishment of speech in psychotic children. In J. K. Wing (Ed.), Childhood autism. New York: Macmillan (Pergamon), 1966. Lovaas, 0. I., Berberich, J. P., Perloff, B. F., & Schaeffer, B. Acquisition of imitative speech by schizophrenic children. Science, 1966, 151, 705-707. McNeill, D. Developmental psycholinguistics. In F. Smith & G. A. Miller (Eds.), The genesis of language: A psycholinguistic approach. Cambridge, Mass.: M.I.T. Press, 1966. Pp. 15-84. Metz, J. R. Conditioning generalized imitation in autistic children. Journal of Experimental Child Psychology, 1965, 2, 389-399. Miller, N. E., & Dollard, J. Social learning and imitation. New Haven: Yale University Press, 1941. Miller, W., & Ervin, S. M. The development of grammar in child language. In U. Bellugi & R. Brown ( U s . ) , The acquisition of language. Monographs of the Society for Research in Child Development, 1964, 29, 9-34. Mowrer, 0. H. Learning theory and the symbolic process. New York: Wiley, 1960. Osgood, C. E. Method and theory in experimental psychology. New York: Oxford University Press, 1953. Pp. 688-690. Parton, D. A. Imitation of an animated puppet as a function of modeling, praise, and directions. Journal of Experimental Child Psychology, 1970, 9, 320-329. Peterson, R. F. Imitation: A basic behavioral mechanism. In H. N. Sloane & B. D. Macaulay (EMS.), Operant procedures in remedial speech and language training. New York: Houghton-Mifflin, 1968. Pp. 61-74. (a ) Peterson, R. F. Some experiments on the organization of a class of imitative behaviors. Journal of Applied Behavior Analysis, 1968, 1, 225-235. (b) Risley, T. R. The establishment of verbal behavior in deviant children. Unpublished doctoral dissertation, University of Washington, 1966. Sailor, W. An experimental analysis of linguistic development. Unpublished doctoral dissertation, University of Kansas, 1969. Schumaker, J., & Sherman, J. A. Training generative verb usage by imitation and reinforcement procedures. Journal of Applied Behavior Analysis, 1971, 3, 273287. Siege], G. M. Vocal conditioning in infants. Journal of Speech and Hearing Disorders, 1969, 34, 3-19. Steinman, W. M. Generalized imitation and the discrimination hypothesis. Journal of Experimental Child Psychology, 1970, 10, 79-99. Waxler, C. Z., & Yarrow, M. R. Factors influencing imitative learning in preschool children. Journal of Experimental Child Psychology, 1970, 9, 115-1 30. Wheeler, A. J., & Sulzer, B. Operant training and generalization of a verbal response form in a speech-deficient child. Journal of Applied Behavior Analysis, 1970, 3, 139-147.