JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR
JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR
THE DEVELOPMENT OF IMITATION BY REINFORCING BEHAVIORAL SIMILARITY TO A MODEL1
DONALD M. BAER,2,3 ROBERT F. PETERSON, AND JAMES A. SHERMAN3
UNIVERSITY OF KANSAS, UNIVERSITY OF ILLINOIS, AND UNIVERSITY OF KANSAS
This research demonstrated some of the conditions under which retarded children can be taught to imitate the actions of adults. Before the experiment, the subjects were without spontaneous imitative behavior, either vocal or motor. Each subject was taught, with food as reinforcement, a series of responses identical to responses demonstrated by an experimenter; i.e., each response was reinforced only if it was identical to a prior demonstration by an experimenter. Initially, intensive shaping was required to establish matching responses by the subjects. In the course of acquiring a variety of such responses, the subjects’ probability of immediate imitation of each new demonstration, before direct training, greatly increased. Later in the study, certain new imitations, even though perfect, were never reinforced; yet as long as some imitative responses were reinforced, all remained at high strength. This imitativeness was then used to establish initial verbal repertoires in two subjects.
CONTENTS
Method Subjects First training procedures Further training procedures
Probes for imitation Non-reinforcement of all imitation Imitative chains Verbal imitations Generalization to other experimenters
Results Reliability of scoring imitative responses First training procedures DRO procedures Imitative chains Verbal behavior Generalization to other experimenters
Discussion The development of a class of behaviors
which may fairly be called “imitation” is an interesting task, partly because of its relevance to the process of socialization in general and language development in particular, and partly because of its potential value as a train- ing technique for children who require special methods of instruction. Imitation is not a spe- cific set of behaviors that can be exhaustively listed. Any behavior may be considered imita- tive if it temporally follows behavior demon- strated by someone else, called a model, and if
its topography is functionally controlled by the topography of the model’s behavior. Spe- cifically, this control is such that an observer will note a close similarity between the topog- raphy of the model’s behavior and that of the imitator. Furthermore, this similarity to the model’s behavior will be characteristic of the imitator in responding to a wide variety of the model’s behaviors. Such control could re- sult, for example, if topographical similarity to a model’s behavior were a reinforcing stim- ulus dimension for the imitator. There are, of course, other conditions which
can produce similar behaviors from two orga- nisms on the same occasion, or on similar occa- sions at different times. One possibility is that
1A portion of this research was presented at the biennial meeting of the Society for Research in Child Development, Minneapolis, Minnesota, March, 1965. This research was supported by PHS grant MH-02208, National Institute of Mental Health, entitled An Experimental Analysis of Social Motivation. Mr. Frank Junkin, Superintendent, Dr. Ralph Hayden, Medical Director, and other members of the staff of the Fir- crest School, Seattle, Washington, made space and subjects available. We wish to thank Mrs. Joan Beavers for her help as a “new” experimenter in the tests of generalization and for assistance in the preparation of this manuscript.
2Reprints may be obtained from Donald M. Baer, Department of Human Development, University of Kansas, Lawrence, Kansas 66044. 3Research Associates, Bureau of Child Research,
University of Kansas. 405
1967, 10, 405-416 NUMBER 5 (SEPTEMBER)
DONALD M. BAER, et al.
both organisms independently have been taught the same responses to the same cues; thus, all children recite the multiplication ta- bles in very similar ways. This similarity does not deserve the label imitation, and hardly ever receives it; one child’s recitation is not usually a cue to another’s, and the similarity of their behavior is not usually a reinforcer for the children. Nevertheless, the children of this example have similar behaviors. The fact that the world teaches many chil-
dren similar lessons can lead to an arrange- ment of their behaviors which comes closer to a useful meaning of imitation. Two children may both have learned similar responses; one child, however, may respond at appropriate times whereas the other does not. In that case, the undiscriminating child may learn to use this response when the discriminating one does. The term imitation still need not be ap- plied, since the similarity between the two children’s responses is not functional for either of them; in particular, the second child is not affected by the fact that his behavior is similar to that of the first. This arrangement ap- proaches one which Miller and Dollard (1941) call “matched-dependent” behavior. One or- ganism responds to the behavior of another merely as a discriminative stimulus with re- spect to the timing of his own behavior; many times, these behaviors will happen to be alike, because both organisms will typically use the most efficient response, given enough experi- ence.
It should be possible, however, to arrange the behavior of two organisms so that one of them will, in a variety of ways, produce precise topographical similarity to the other, but nothing else. A study by Baer and Sherman (1964) seemingly showed the result of such prior learning in several young children. In that study, reinforcements were arranged for children’s imitations of three activities of an animated, talking puppet, which served both as a model and a source of social reinforce- ment for imitating. As a result of this rein- forcement, a fourth response of the puppet was spontaneously imitated by the children, although that imitation had never before been reinforced. When reinforcement of the other three imitations was discontinued, the fourth, never-reinforced imitation also decreased in strength; when reinforcement of the original imitations was resumed, imitation of the
fourth response again rose in rate, although it still was never reinforced. In short, these chil- dren apparently generalized along a stimulus dimension of similarity between their behav- iors and the behaviors of a model: when simi- larity to the model in three different ways was reinforced, they thereupon displayed a fourth way of achieving similarity to the model. Thus, similarity between their behavior and the model’s was a functional stimulus in their behavior. Metz (1965) demonstrated the development
of some imitative behavior in two autistic chil- dren who initially showed little or no imita- tive response. In this study, responses similar in topography to demonstrations by the ex- perimenter were reinforced with “Good” and food. Metz found that, after intensive training, several imitative responses could be main- tained in strength even when not reinforced with food, and that the subjects had a higher probability of imitating new responses after training than before. However, in one of the conditions used to evaluate the subjects’ imi- tative repertoire before and after imitative training, “Good” was still said contingent upon correct new imitations. Thus, for one subject who initially showed a non-zero rate of imitation, it could be argued that the in- creased imitation in the test after training was due to an experimentally developed reinforc- ing property of “Good”, rather than to the imitation training as such. Further, in the Metz study, due to a lack of extinction or other manipulation of the behavior, it is dif- ficult to specify that the higher probability of imitating new responses, and the maintenance of unreinforced imitative responses, were in fact due to the reinforcement of the initial imitative responses during training.
Lovaas, Berberich, Perloff, and Schaeffer (1966) used shaping and fading procedures to establish imitative speech in two autistic chil- dren. They reported that as training progressed and more vocal behavior came under the con- trol of a model’s prior vocalization, it became progressively easier to obtain new imitative vocalizations. When reinforcement was shifted from an imitative-contingent schedule to a basically non-contingent schedule, imitative behavior deteriorated. In an additional ma- nipulation, the model presented Norwegian words interspersed with English words for the children to imitate. Initially, the children did
406
THE DEVELOPMENT OF IMITATION
not reproduce the Norwegian words perfectly. However, the authors judged that the subjects gradually improved their imitations of the Norwegian words even though these imitations were not reinforced. The studies by Baer and Sherman (1964),
Metz (1965), Lovaas et al (1966), and other re- ports (Bandura, 1962) suggest that for children with truly imitative repertoires, induction has occurred, such that (1) relatively novel behav- iors can be developed before direct shaping, merely by providing an appropriate demon- stration by a model, and (2) some imitative responses can be maintained, although unrein- forced, as long as other imitative responses are reinforced. The purpose of the present study was to ex-
tend the generality of the above findings and to demonstrate a method of producing a truly imitative repertoire in children initially lack- ing one.
METHOD
Subjects Three children, 9 to 12 years of age, were
selected from several groups of severely and profoundly retarded children in a state school. They were chosen not because they were re- tarded, but because they seemed to be the only children available of a practical age who ap- parently showed no imitation whatsoever. (The success of the method to be described suggests that it may have considerable practi- cal value for the training of such children.) The…
