Artificial neural networks as models of stimulus control*

Distance from S+ (arbitrary units)

Figure 3: Intensity generalization profiles for three different kinds of stimuli: lights, bells
and whistles. On the horizontal axis the intensity of the test stimuli is plotted in arbitrary
units as a distance from the conditioning stimulus S⁺ , while on the vertical one is shown
the proportion of responses emitted for each stimulus in comparison with S⁺ . Data from
Razran (1949), table II, summarising data from Pavlov’s laboratory (Pavlov, 1927).

as sound frequency (Jenkins & Harrison, 1960, 1962), light wavelength (Hanson,
1959; Kalish & Guttman, 1959) and line orientation (Hearst, 1968). A point of
interest in distinguishing between these two classes of generalization tests is that
they give very different results in real experiments, and are therefore a good first
test for any model of stimulus control.

4 Results

4.1 The Intensity Test

The general appearance of intensity generalization gradients is clearly illustrated
in figure 3, where data coming from tests involving three different kinds of stimuli
(lights, bells and whistles) are reported (Razran, 1949; Mackintosh, 1974). All
three experimental gradients are monotonic, with increased number of responses
as the intensity of stimulation increases. All stimuli more intense than S⁺ are thus
supernormal, eliciting more responses than S⁺ itself.

More data are provided in Weiss & Schindler (1981), which we show in fig-
ure 4. In this experiment two groups of rats were conditioned for lever pressing
in the presence of clicks coming from a loudspeaker, the variable changing in the
generalization test being the number of clicks per second (directly related to the
amount of stimulation per second). A third group, serving as control, yielded a
substantially flat generalization profile (line 1 in figure 4), showing that the in-
tensity of stimulation has no control per se when the considered stimulus is not
correlated with any meaningful outcomes. Rats in the first experimental group

<<   4   5   6   7   8   9   10   >>

More intriguing information