sity of Maryland, College Park, Maryland 20742. The authors would ... periments. 'Now at St. Olaf College. .... which was mounted a Lehigh Valley pigeon feeder.
JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR
VOLUME
9,
NUMBER
6
NOVEMBER, 1966
SOME EFFECTS ON GENERALIZATION GRADIENTS OF TANDEM SCHEDULES' MATrTHEW YARCZOWER, JAMES F. DICKSON,2 AND LEWIS R. GOLLUB UNIVERSITY OF MARYLAND
The relationship between training conditions and stimulus generalization gradients was examined using tandem schedules of reinforcement. Schedules were selected so that frequency of reinforcement and rate of responding were varied somewhat independently of each other. A peak-shift in the generalization gradient was obtained when extinction had been associated with one of the stimuli. No comparable peak shift was obtained when there were equal response rates in the training stimuli even with dissimilar frequencies of reinforcement. The data imply that response rates at the end of training, rather than reinforcement frequency per se, determine the characteristics of the generalization gradient.
multiple schedules of reinforcement. The response rate during the presentation of one stimulus may change when the schedule associated with a different stimulus is altered. "A positive contrast . . . would be an increase in the rate of responding in one component in a direction away from the rate prevailing in the other component" (Reynolds, 1961a, p. 115). When positive contrast occurs, the stimulus generalization gradient shows a "peak shift" (Hanson, 1959; Terrace, 1964; Friedman and Guttman, 1965). The peak shift has been defined as a difference in the distribution of responses to the several stimuli presented during the generalization test of subjects trained under a discrimination procedure, compared to those trained only with the positive stimulus (Hanson, 1959; Terrace, 1964). For example, consider stimulus generalization with wavelength stimuli. One group of subjects is given discrimination training involving a negative stimulus with a longer wavelength than that of the positive stimulus. A second group is given single stimulus training, involving only training to the same positive stimulus. The generalization gradient of the discrimination group will show a greater number of responses to stimuli of shorter 'This research was supported in part by USPHS wavelength than the positive stimulus comGrants No. MH 08819-01 and MH-01604-09 from the pared to the gradient of the group given only National Institutes of Health. Reprints may be obtained single stimulus training. Conversely, when from Matthew Yarczower, Dept. of Psychology, Univer- positive contrast is not observed during dissity of Maryland, College Park, Maryland 20742. The authors would like to thank Dr. Stanley S. Pliskoff for crimination training, a peak-shift does not his advice and aid throughout the course of these ex- occur (Terrace, 1964). Peak-shifts have been periments. obtained when extinction was one component of the training schedule (Hanson, 1959; Ter'Now at St. Olaf College. 631
Studies of the effects of multiple schedules (two or more schedules of reinforcement each consistently associated with a different exteroceptive stimulus) of reinforcement on generalization gradients can be placed into two classes. In one class, usually called "discrimination" experiments, different rates of responding are generated in the training stimuli by different schedules of reinforcement (Guttman, 1959; Hanson, 1959). In the other class, usually called summation experiments, identical rates of responding are established in the training stimuli by using the same schedule of reinforcement in each stimulus (Kalish and Guttman, 1957, 1959). A review of the first class suggests that when certain types of rate changes occur during discrimination training, the form of the stimulus generalization gradient also changes. The most important feature of the discrimination training is whether the response rate in the presence of the stimulus correlated with positive reinforcement (i.e., the positive stimulus) is increased over some control value. This increase, when it occurs, is called "positive contrast." Reynolds (1961a) has defined positive contrast by reference to rate changes during
632
MATTHEW YARCZOWER, et al.
race, 1964; Friedman and Guttman, 1965) as well as when both components involved positive reinforcement (Guttman, 1959). The present experiment investigated the effects of pairs of training schedules involving extinction as well as pairs of different positive reinforcement schedules. There are fewer experiments in the second class but, as one example, consider Kalish and Guttman (1957). They studied the case in which identical reinforcement schedules were paired with each of the different training stimuli to produce what were probably equal rates of responding in each stimulus. The stimulus generalization gradients following this training were quite different from gradients following training with a single stimulus. However, the differences cannot be ascribed to reinforcement frequency alone, since response rates and reinforcement frequency co-varied. It is possible to conceive of an experiment intermediate between these two classes. It could provide different frequencies of reinforcement, a characteristic of the first class of studies, but generate equal rates of responding in the presence of the two discriminative stimuli, a characteristic of the second class. Perhaps in this way we may begin to investigate the effects on generalization gradients of differential frequency of reinforcement as well as of differential response rates. The schedule chosen in an attempt to vary rate of responding and frequency of reinforcement somewhat independently of each other was a tandem (tand) variable interval (VI) differential reinforcement of low rates (DRL). In this schedule, reinforcement is made available at varying periods after the previous reinforcement, according to the VI schedule. When a reinforcement is available, only a response that occurs with a specified minimal interresponse time (IRT) can be reinforced. This specified minimal interresponse time is the value of the DRL schedule. In other words, the VI schedule determines the frequency of food reinforcement and the DRL schedule determines the response rate, by placing an upper limit on the rate that can be reinforced (cf. Ferster and Skinner, 1957; Blough, 1963).
throughout the experiment. The experimental chamber, made of Plexiglas except for the aluminum response panel, measured 13 in. long, 151/2 in. wide, and 181/2 in. high. An aluminum disc with 1½V2-in. circular openings into which Bausch and Lomb interference filters (44-78 series) and Kodak neutral density filters (No. 96 series) were placed, was used to program the visual stimuli. The illumination source was a GE 18a T10/2P-6v microscope illuminator bulb with a ribbon filament. The collimated monochromatic light transilluminated the translucent response key. All wavelength stimuli were equated for brightness with neutral density filters according to data on the pigeon's photopic sensitivity curve (Blough, 1957). The response panel consisted of a standard Gerbrands pigeon key below which was mounted a Lehigh Valley pigeon feeder.
Procedure The 20 naive pigeons were divided into five groups of 3, 4, 4, 4, and 5 birds. All subjects were magazine trained on the first day and then given 50 continuous reinforcement trials on the second day with the key illuminated by a 550 nm light. The unit "nm" refers to "nanometer" which is mathematically identical to millimicron. The subsequent treatment is described in detail below, and is summarized in Table 1. The first four groups were exposed during training to a multiple schedule (Ferster and Skinner, 1957) with two stimuli, 550 nm and 570 nm, and the fifth group to only a single stimulus, 550 nm. Group 1, composed of three subjects, received between 13 and 17 days of training to the schedule, mult VI 30 sec EXT, in which the schedules were correlated with the 550 nm and 570 nm stimulus, respectively. In this multiple schedule, when the key was illuminated with the 550 nm light, key pecks were reinforced after varying periods of time after the previous reinforcement, with an average time of 30 sec (VI 30 sec), and when the key was illuminated with the 570 nm light, no key pecks were reinforced, i.e., extinction (EXT) was scheduled. This group was used to insure that the peak shift effect could be replicated Subjects and Apparatus under the conditions of our apparatus and Twenty Silver King pigeons, three to five procedures, using schedules similar to those years old, were maintained at approximately reported by other investigators. Group 2 (four subjects) was exposed to a 75% of their free-feeding body weights
GENERALIZATION GRADIENTS AND TANDEM SCHEDULES
similar multiple schedule; 570 nm was paired with extinction, but 550 nm was paired with a tand VI 30 sec DRL 4 sec. As described above, this schedule provides for reinforcement of low rates (DRL) by allowing reinforcement only for interresponse times greater than 4 sec; reinforcements became available on the average of 30 sec after a preceding reinforcement (VI 30 sec). This group enabled us to see whether peak shift would occur after reinforcement on tand VI DRL schedules. The DRL feature of this schedule was expected to limit or prevent any rate increase (i.e., positive contrast) in the presence of 550 nm. The lack of positive contrast would be most interesting if a peak shift occurred in later testing. Group 3 (four subjects) was exposed to a multiple schedule made up of two positive reinforcement schedules, mult (tand VI 30 sec DRL 4 sec) (tand VI 4 DRL 8 sec). The parentheses set off the two schedules, the first paired with 550 nm and the second, with 570 nm. Comparing the second schedule to the first, it can be seen that reinforcement is scheduled only one-eighth as frequently (VI 4 versus VI 30 sec) and that a longer interresponse time and hence a lower response rate is required (DRL 8 sec versus DRL 4 sec). The purpose of establishing this training schedule was similar to that for Group 2, namely to see whether a peak shift would occur when positive contrast was not allowed in 550 nm by the tand VI 30 sec DRL 4 sec schedule. Another schedule, which might produce a peak shift, is, rather than extinction, a schedule with occasional but less frequent reinforcement than during the 550 nm stimulus. The selection of such a schedule was based on the study by Guttman (1959) in which it was found that a peak shift occurred when two VI training schedules differed in frequency of reinforcement.
633
The schedules for Group 4 (four subjects) selected to produce equal response rates in each of the stimuli despite greatly different frequencies of food reinforcement. Using tandem VI DRL schedules in each stimulus, this effect can be produced by combining the VI schedule that provides reinforcement more frequently with the longer DRL, and the less frequently reinforcing VI schedule with the shorter DRL. The multiple schedule arbitrarily selected to produce equal rates and unequal reinforcement frequencies was mult (tand VI 30 sec DRL 4 sec) (tand VI 3 DRL 2 sec), paired with the 550 nm key light and the 570 nm key light, respectively. Group 5 (five subjects) was given training only in the presence of a 550 nm key light, and was the "control" group to which all the other groups were compared. The reinforcement schedule for this single stimulus training was tand VI 30 sec DRL 4 sec. Table contains a summary of the design. All groups given training to two wavelength stimuli had either extinction (Groups 1 and 2) or less frequent reinforcement (Groups 3 and 4) in the presence of the 570 nm light. The column labeled "rft. ratio 570 nm/550 nm" shows the ratio of reinforcement frequencies in the presence of the two stimuli. The response rate in the presence of 570 nm relative to the rate in the presence of 550 nm is shown for each group in the right column. The extinction schedule for Groups 1 and 2 was expected to reduce responding to zero in the presence of 570 nm, whereas the selection of tand VI DRL schedules for Groups 3 and 4 was expected to produce lower and equal response rates, respectively. The terms in parentheses roughly summarize the results of training, and indicate the extent to which the design was met. We are in a position to evaluate, independently, the effects on stimulus generalization of response rates and reinwere
Table I Summary of schedules of reihforcement and expected reinforcement and response ratios.
Group mult VI 30"-EXT 1 2 mult tand VI 30" DRL 4"-EXT 3 mult tand VI 30"' DRL 4"1-tand VI 4 DRL 8"' 4 mult tand VI 30" DRL 4"-tand VI 3 DRL 2" tand VI 30" DRL 4" 5 *obtained data are in the parentheses.
Rft. ratio 570 nm/550 nm
Resp. rate 570 nm/550 nm
0 0
0
.20 (.20) .20 (.33)
