DONALD J. SALMON, JOSEPH J. PEAR, AND BEvERLEY A. KUHN. ST. AMANT ... of the St. Amant Centre, Dr. Carl Stephens, Director of. Psychology of theSt.
JOURNAL OF APPLIED BEHAVIOR ANALYSIS
19861191,53-58
NUMBER
1
(SPRING 1986)
GENERALIZATION OF OBJECT NAMING AFTER TRAINING WITH PICTURE CARDS AND WITH OBJECTS DONALD J. SALMON, JOSEPH J. PEAR, AND BEvERLEY A. KUHN ST. AMANT CENTRE AND UNIVERSITY OF MANITOBA
Generalization of four retarded children's object naming responses to stimuli in the natural environment was assessed after training with either objects or pictures of the objects. Generalization was typically greater after training with objects. In a second experiment, half of the stimuli that showed little generalization were retrained by alternating the original training object with an object that belonged to the same stimulus class as the training stimulus. The other half were simply retrained using the object. The alternating procedure resulted in substantial increases in generalization to untrained objects. DESCRIPTORS: generalization, retarded children, picture and object naming, verbal training
Children are more likely to generalize an object name to objects encountered in the natural environment if the training stimulus used in object name training is the object than if it is a photograph of the object or a picture card depicting the object (Welch & Pear, 1980). Experiment 1 is an extension of this finding, by comparing the amount of generalization that would occur to objects of the same stimulus class in the child's environment if the training stimulus was an object or a picture card depicting the object.
ing at a residential center. They were selected on the basis of being good vocal imitators, but having a small naming repertoire. Each child was individually taught in a room at the center. Assessments of the children's picture and object naming repertoire were conducted in the same room as the training sessions. Assessments of the children's object naming repertoire were also conducted in the children's natural environments (i.e., their wards or cottages). Generalization tests were conducted in the children's natural environments.
EXPERIMENT 1
Training Stimuli All picture cards used in the research were from a Peabody Articulation Kit. The training objects that were used matched the picture cards as dosely as possible. For each picture card and its object, four objects belonging to the same class were selected by the experimenter on the basis of convenience, with each differing from the object on some dimension (e.g., size, color, form, or any combination of these three).
MEOHOD Subjects and Setting The participants in this study were four developmentally delayed children (CA = 5-8 year) livThis study was conducted as part of a research program under the supervision of J. J. Pear on developing verbal behavior in retarded children and was supported in part by Grant No. MA-5647 from the Medical Research Council of Canada. Data for three of the four children are based on a thesis submitted by the first author in partial fulfillment of requirements for the MA degree in Psychology. We are grateful to Sr. Bertha Baumann, Administrator of the St. Amant Centre, Dr. Carl Stephens, Director of Psychology of the St. Arnant Centre, and to the staff of Riverside and the Developmental Day Program for their cooperation and help. Reprints may be obtained by writing to D. J. Salmon or J. J. Pear, Department of Psychology, University of Manitoba, Winnipeg, Manitoba R3T 2N2, Canada.
53
Experimental Design The experiment consisted of four phases. In each phase, seven stimuli were randomly selected, six for training purposes and one as a control against extraneous learning variables. During Phase 1, Chris and Bruce were trained using objects, and Laurie and Robin were trained using picture cards. Dur-
54
DONALD J. SALMON et al.
ing Phase 2, Chris and Bruce were trained using picture cards, and Laurie and Robin were trained with objects. Phase 3 was a replication of Phase 1, and Phase 4 was a replication of Phase 2. Bruce was exduded from Phase 4 because of his fairly extensive naming repertoire. In the following, the term "training object" refers to the object that was used as the training stimulus, or whose corresponding picture card was used as the training stimulus. The term "testing object" refers to the nontraining objects used in the generalization test. Overview of the General Procedures Prior to the start of the study we determined which of the objects and pictures each child could name and the picture and object names each child could imitate. The procedure used for this was similar to that used by Welch and Pear (1980). Each phase of the study proceeded through five different procedures: (a) a pretraining test, (b) a test in the natural environment, (c) training, (d) a posttraining test, and (e) a generalization test in the natural environment. In the pretraining test the stimuli to be trained were again assessed in the training room. Each training stimulus was presented three times in a random order, and motor instructions were interspersed with the training stimuli and intermittently reinforced with an edible. Responses to training stimuli were never reinforced. In the test in the natural environment, each of the seven training objects and the respective four test objects belonging to the same class were randomly presented three times, with motor instructions again interspersed and intermittently reinforced with an edible. Following the test in the natural environment one of the seven stimuli was randomly chosen as a control stimulus and received no training. During training, the picture or object names were trained using a procedure similar to that used by Olenick and Pear (1980), whereby two stimuli were trained concurrently to a specified criterion. A child proceeded through a series of prompt trials in which the experimenter named the stimulus (e.g.,
"What's this? Cup.") and probe trials on which the experimenter simply asked the name of the stimulus (e.g., "What's this?") until the naming behavior reached criterion. On Step 1 the first unknown stimulus was presented on a probe trial. This trial served as a further control against extraneous sources of learning. If the child responded correctly to the training stimulus, it was discarded and a new stimulus was selected. If the child made an error (i.e., an incorrect response or a response omission within an 8-second time limit) a prompt trial was presented. An error on this trial resulted in a further prompt trial. A correct response resulted in the probe trial being again presented. A correct response to the probe trial in Step 1 resulted in Step 2, which involved following the same procedure for a second unknown stimulus. A correct response on the probe trial of Step 2 resulted in two alternations with probe trials for the first and second unknown stimuli. Errors resulted in a return to the prompt trial with the sequence being repeated from that point. When the stimuli had gone through the above sequence three times without error, they were considered known and two more stimuli were randomly selected for training. Correct responses on probe trials were intermittently reinforced with an edible. All correct responses on prompt and probe trials were reinforced with praise (cf. Olenick & Pear, 1980). The posttraining test consisted of presenting the trained stimuli 10 times each in a random order. If a child failed to respond correctly to 7 of the 10 presentations of a stimulus, the stimulus was retrained and the posttraining test was repeated. Correct responses were intermittently reinforced with an edible. The generalization test in the natural environment was identical to the pretraining test conducted in the natural environment. Interobserver Reliability
Interobserver reliability was assessed during l/6 of the training sessions and V3 of the posttraining and generalization tests. Reliability was calculated as the ratio of agreements to agreements plus dis-
GENERALIZATION OF OBJECT NAMING
agreements on those responses that the experimenter scored as correct and the ratio of agreements to agreements plus disagreements on those responses that the experimenter scored as incorrect. Reliability ranged from 0.93 to 0.97. RESULTS AND DISCUSSION Figure 1 shows the percentage of correct responding during the posttraining test (black dots) and during the generalization test to the training objects and the testing objects in the natural environment over the four phases for all children. Note that all naming responses were well known on the posttraining test. During the generalization tests all children, with the exception of Laurie dur-
ing Phase 2, failed to respond correctly to any of the control stimuli indicating that learning outside the training situation was not a confounding factor. The figure reveals that more generalization to the training objects in the natural environment occurred when objects were used as training stimuli, with the exception of one phase with Robin. Note also that more generalization was found to the testing objects after training was conducted with objects, with the exception of one phase with Laurie and one phase with Robin. The results of this experiment replicate the findings of Welch and Pear (1980). Greater generalization was found to the training object in the natural environment after training was conducted with objects than with picture cards. In addition, the results extend the findings of Welch and Pear in that training with objects was more likely to increase generalization to other objects belonging to the same stimulus dass. EXPERIMENT 2 In a series of experiments with pigeons, Kalish and Guttman (1957, 1959) found that a generalization gradient can be broadened by training at two or more stimulus values. Similarly, generalization of naming responses to objects may be enhanced by training the same naming response to more than one object of a given stimulus dass. Essentially, this procedure seems to be what Stokes
55
i.-
0 LU
I-. z
LU
'U
IU 'U
I--
o8J. P.C. OSJ. P.C. - PICTURE CARDS OBJ. - OBJECTS M- %
P.C. OBJ. P.C. OBJ. *-S KNOWN ON
ESTT
CORRECT TRAINING OBJECTS
U- % CORRECT TESTIN OBJECTS
Figure 1. Percentage of correct responses emitted by all children to the objects and the testing objects during tests for generalization in the natural environment.
and Baer (1977) call "training sufficient stimulus exemplars." The expected effect of this procedure can be conceptualized in two quite different ways. First, it can be conceptualized as an increased tendency to respond to different stimuli represented by the different members of a dass of objects due to "summation" of separate generalization gradients established around the training stimuli (cf.
56
DONALD J. SALMON et al.
Kalish & Guttman, 1957, 1959). Second, it can be conceptualized as an increased degree of stimulus control brought about by varying irrelevant dimensions of the training stimuli (e.g., color and size) so that more circumscribed stimuli (e.g., form) come to control the response. In either case, the procedure should increase the amount of generalization to objects belonging to the same stimulus dass. The purpose of Experiment 2 was to test this possibility. METHOD Subjects and Setting
the object as the training stimulus. The remaining two pools were retrained, alternating the object with the least similar object. This procedure is referred to as the alternating procedure. Chris and Robin received the alternating procedure prior to retraining, and Laurie received retraining prior to the alternating procedure. Regardless of whether training was originally conducted using picture cards or the objects, during the alternating and retraining conditions objects were used as the sole training modality.
Overview of General Procedures For all stimuli that were simply being retrained, The participants were the same as those who had served in Experiment 1, with the exception of the training procedure was identical to that in ExBruce who was unavailable. The settings were the periment 1, with the exception that a correct response on the first trial of Steps 1 and 2 did not same as those used in Experiment 1. result in the stimuli being discarded. Training in Training and Testing Stimuli the alternating procedure was also similar to that Training stimuli in this experiment were those of Experiment 1, with the exception that two obon which the child had scored 60% or lower dur- jects belonging to one dass were alternated with two other objects belonging to another dass. Howing the generalization tests. ever, the number of training steps was the same Experimental Design as in Experiment 1. The posttraining test was idenTo increase the probability that the procedure tical to that used in the first experiment for those would enhance generalization, the additional object stimuli that were simply retrained. For those stimthat was trained in each stimulus dass was the one uli that were retrained using the alternating proleast similar to the training object or picture card, cedure, during the posttraining test the object was as determined where possible from the child's re- presented five times and the object least similar to sponding in Experiment 1. If this could not be the object was also presented five times. The gendetermined from the child's responding, the object eralization test was identical to that used in the least similar to the training object or picture card first experiment. was determined on the basis of three judges' rating of the stimuli. To control for the possibility that Interobserver Reliability Reliability was assessed in a manner similar to simply retraining a particular naming response may promote generalization, half of the responses were that used in Experiment 1 and ranged from 0.95 tested for generalization after simply being re- to 0.97. trained. RESULTS AND DISCUSSION Responses that had been trained with picture cards were randomly divided into two pools. ReFigure 2 shows the percentage of correct responses that had been trained with objects were sponses for those naming responses that were origalso randomly divided into two pools. From these inally trained using objects or pictures cards before four pools, the naming responses from one of the alternating or retraining and after alternating and pools that was trained with picture cards and the retraining. Because the object and the object least naming responses from one of the pools that was similar to it has both been trained in the alternattrained with objects were simply retrained using ing procedure, these objects were exduded from
GENERALIZATION OF OBJECT NAMING
CHRIS
LAURIE
57
ROBIN
C)
w
a:
0 0 z
w
w
a.
OBJECTS
OBJECTS PICTURE CARDS
PICTURE CARDS
OBJECTS
PICTURE CARDS
r AFTER BEFORE AFTER m BEFORE RETRAINING- EI RETRAINING Wd ALTERNATING U; S ALTERNATING Figure 2. Percentage of correct responses emitted by Children 1, 2, and 4 exduding the object and the object least similar to the object. Data shown were obtained prior to and after either alternating or retraining responses that were initially trained using either objects or picture cards.
the calculations. Comparable calculations were made for those responses that had been retrained (i.e., the object and the object least similar to it were excluded from these calculations). Thus, the only objects represented in the graph are the ones that were tested in the natural environment but never received any training. For all three children, those responses that were originally trained using objects and then alternated with another object showed substantial increases in generalization compared to the amount of generalization found prior to alternation. For those naming responses that were simply retrained there were small increases in
generalization. For those naming responses that were originally trained with picture cards, the alternating procedure using objects resulted in a large increase in generalization to the testing objects. This increase
in generalization was similar to that found with the alternating procedure when objects were the original training stimulus. Retraining with the object resulted in increases in generalization to those levels found to have occurred prior to the alternating procedure when objects were the original training stimulus for Chris and Laurie. For Robin, retraining with objects after training with picture cards increased generalization almost as much as the alternating procedure did. The results of Experiment 2 thus support the suggestion that generalization can be increased by "training sufficient stimulus exemplars." Retraining the stimuli proved to be rather ineffective in promoting generalization with the exception of the one phase with Robin for those stimuli that had initially been trained using picture cards. It would be of interest to determine whether optimal gen-
DONALD . SALMON et al.
58
summation hypothesis. Journal of Experimental Psyeralization to objects requires that all examples be chology, 57, 268-272. objects. The results with Robin suggest that this Olenick, D. L., & Pear, J. J. (1980). Differential reinforcemay not be the case because training picture cards ment of correct responses to prompts and probes in picture naming training with severely retarded children. followed by training with objects resulted in subJournal of Applied Behavior Analysis, 13, 77-90. stantial increases in generalization. Stokes, T. F., & Baer, D. M. (1977). An implicit tech-
REFERENCES Kalish, H. I., & Guttman, N. (1957). Stimulus generalization after equal training on two stimuli. Journal of Experimental Psychology, 53, 139-144. Kalish, H. I., & Guttman, N. (1959). Stimulus generalization after training on three stimuli: A test of the
nology of generalization. Journal of Applied Behavior Analysis, 10, 349-367. Welch, S. J., & Pear, J. J. (1980). Generalization of naming responses to objects in the natural environment as a function of training stimulus modality with retarded children. Journal of Applied Behavior Analysis, 13, 629-644. Received December 12, 1984 Final acceptance October 19, 1985