fig. 1. Stimulus generalization gradient for subjects that were trained to identify the target length which is of 75 as the dimension of length and were tested in the presence of the other dimensions of length. fig. 2. Effect of intradimensional discrimination training on stimulus control. Subjects received discrimination training in which the S- was 85 as the dimension of length, keeping the S+ 75 as the dimension of length. 2 . In Figure 1, it shows the comparison of the stimulus generalization gradients of two different subject groups, between the class average and me as an individual.
To commence with, lets talk about the curve representing the class mean. The shape of the graph is basically symmetrical between 55 and 80, which these numbers are the indication of the length of the line presented. Though 75 is the target length, the highest rate of response occurred in response to the length of 65 and 70, both encountering 100% of responses. The subjects also made substantial numbers of responses when length of 60 and 75 were tested. However, when length of 55 and 80 were tested, response rates decreased quite dramatically, only 20% of the total responses were recorded, making the graph a bell-shaped one.
As for length of 85 or above, no responses were detected. For the curve representing my own result, it is more or less a symmetrical graph as well. The highest rate of response occurred in response to the original length of 75. Once again, substantial numbers of responses were also detected when length of 70 and 80 were tested. Yet there is a huge decrement of response for much shorter or longer lengths (i. e. , 55-65 and 85-95), no responses were detected. These two curves both serve to indicate the phenomenon of stimulus generalization, though the individual result better explains it.
The peak of response rate lies on or around (for the class mean) the target length,75. But quite a similar percentage of total responses were also made to the lengths around the target one, this indicates that responding generalized to the 70 or even 65 and 80 stimuli, therefore a gentle gradient could be plotted. On the other hand, as the length of the test stimuli became increasingly different from the target length, progressively fewer responses occurred. It is because the difference between the tested length and the target length was significant enough to be recognized and differentiated, thus quite a ramatic decrement of responses occurred when a comparatively very short or very long line were tested. The results shown a gradient of responding as a function of how similar each test stimulus was to the original training stimulus (target). One reason for not having the highest percentage of responses at 75 for the class statistics would be because it was reflecting the class average responses and there might be out-lyers whose results affected the norm. Stimulus generalization gradients provided precise information about how much a stimulus has to be changed to produce a change in behavior.
A gentle slope shows the variation in the stimulus is not significant enough to produce a respond to the variation while a steep slope shows the variation in the stimulus is large enough for the subjects to respond to it. In Figure 2, it shows the comparison of the intradimensional discrimination gradients of the two subject groups (me as an individual and the class average). To commence with, I will first talk about the curve representing the class mean. The shape of the curve is asymmetrical, with the highest percentage of response again occurred in response to the length of 65 and 70, achieving 100% responses.
But this time, once the length increased gradually from 70, the percentage of response decreased steadily until it reached 0% when the length of 85, which is the S- (discriminative stimulus), is tested. Although the target length was again, 75, there is a counterintuitive phenomenon known as the peak-shift effect to explain the peak of response lying on 65 and 70 instead of 75. Quite high percentages of response were occurred when 75, the target length was presented. Yet the percentage of responses was higher to 65 and 70 than to 75.
This shift of the peak responding away from the original S+ is recognizable after discrimination training with the length of 85 as S-. This shift of the peak has an explanation other than the generalization. During the earlier phase of discrimination training, responding was never reinforced in the presence of the 65 and 70 stimuli. However, because the target stimulus and the discriminative stimulus are similar in intradimensional discriminative tasks, the generalization gradients of excitation and inhibition will overlap.
This is due to the inhibitory response learnt when S- is presented in the discriminatory training. Moreover, the degree of overlap will depend on the degree of similarity between S+ and S-. Since then, generalized inhibition from S- will suppress responding to S+ resulting this peak-shift effect. As for the curve representing my own result, the graph is more or less like symmetrical, with the peak occurring at the length of 75. The percentage of response increased significantly from the length of line varies from 65 to 70, creating a steep slope.
For lines which differed from the target length comparatively much, like 55-65 and those above 90, no response were made, therefore 0% of the total response were recorded. Though I have also gone through the discrimination training, the absence of peak-shift effect may be due to individual difference, or insufficient training, therefore I still responded most to the S+ stimulus (target length,75) and responded progressively less as the length of the test stimuli deviated from the S+ stimulus.