3. THE EXPERIMENTAL ANALYSIS OF VISUAL BEHAVIOR
65
Gestalt school) still guide behavioral research. Since we ourselves recognize many shapes despite transformations of size, position, rotation, or
brightness contrast, we are tempted to ask whether our animal subjects
also have ways of categorizing shapes that transcend these transformations. For example, it is commonly assumed that shapes are easily recognized by animals, despite changes in size (Sutherland, 1968). However,
the usual method of demonstrating “size invariance” is inadequate since
an animal approaching the stimulus (for food reward) experiences considerable variation in size of the associated retinal images of each shape.
To overcome this difficulty, Ingle ( 1971) trained goldfish to discriminate
14” wide circles from equal-sized squares that were presented at a fixed
distance lateral to one eye during a conditioned avoidance paradigm.
When well trained, five fish learned the opposing habit using smaller
(7”) shapes. For example, fish avoiding the large circle but not the square
now learned to avoid the small square but not the circle. When subjects
were retested (without reinforcement) using the original large shapes,
they retained the initial habit despite the intervening training. We cannot
say that these fish found no resemblance between large and small squares,
but rather that they used two different sets of rules in distinguishing the
two pairs of shapes. These results do not settle the important problem
of size invariance, but they do indicate that size is sometimes an important determinant of the way a fish can classify particular shapes.
Studies by Bowman and Sutherland (reported in Sutherland, 1968)
indicate that goldfish do not generalize the properties of a square through
a 45” rotation to a diamond. Following training on a circle vs. square
discrimination, their subjects failed to distinguish a circle from a diamond.
However, other discriminations may be transposed through rotations.
Both Schulte (1957) and Saxena (1966) found good transposition of a
discrimination of stripe number (and width) from vertical to horizontal
settings (Fig. 3 ) . This task requires attention to number of line elements
rather than to their spatial relationship. Saxena noted that a square vs. X
discrimination would generalize to a diamond vs. cross pair. Here the
discrimination is not one of orientation but a topological distinction between open shapes vs. intersecting lines. These experiments show that
fish can judge two stimuli (e.g., squares and diamonds) either as equivalent or as distinctive, depending upon which basis of comparison our
training procedure forces upon them. It is the aim of the generalization
test to indicate the kinds of “similarity” that may exist. As Kliiver (1933)
has documented in elegant detail, it is difficult to disentangle the actual
dimensions of visual analysis from even an assortment of such tests. The
assumption that fish possess inherently fewer and more rigid classifications
than the higher mammals is plausible but is by no means demonstrated.
Ingle ( 1971 ) has speculated that large-scale “gestalt” features such
65
Gestalt school) still guide behavioral research. Since we ourselves recognize many shapes despite transformations of size, position, rotation, or
brightness contrast, we are tempted to ask whether our animal subjects
also have ways of categorizing shapes that transcend these transformations. For example, it is commonly assumed that shapes are easily recognized by animals, despite changes in size (Sutherland, 1968). However,
the usual method of demonstrating “size invariance” is inadequate since
an animal approaching the stimulus (for food reward) experiences considerable variation in size of the associated retinal images of each shape.
To overcome this difficulty, Ingle ( 1971) trained goldfish to discriminate
14” wide circles from equal-sized squares that were presented at a fixed
distance lateral to one eye during a conditioned avoidance paradigm.
When well trained, five fish learned the opposing habit using smaller
(7”) shapes. For example, fish avoiding the large circle but not the square
now learned to avoid the small square but not the circle. When subjects
were retested (without reinforcement) using the original large shapes,
they retained the initial habit despite the intervening training. We cannot
say that these fish found no resemblance between large and small squares,
but rather that they used two different sets of rules in distinguishing the
two pairs of shapes. These results do not settle the important problem
of size invariance, but they do indicate that size is sometimes an important determinant of the way a fish can classify particular shapes.
Studies by Bowman and Sutherland (reported in Sutherland, 1968)
indicate that goldfish do not generalize the properties of a square through
a 45” rotation to a diamond. Following training on a circle vs. square
discrimination, their subjects failed to distinguish a circle from a diamond.
However, other discriminations may be transposed through rotations.
Both Schulte (1957) and Saxena (1966) found good transposition of a
discrimination of stripe number (and width) from vertical to horizontal
settings (Fig. 3 ) . This task requires attention to number of line elements
rather than to their spatial relationship. Saxena noted that a square vs. X
discrimination would generalize to a diamond vs. cross pair. Here the
discrimination is not one of orientation but a topological distinction between open shapes vs. intersecting lines. These experiments show that
fish can judge two stimuli (e.g., squares and diamonds) either as equivalent or as distinctive, depending upon which basis of comparison our
training procedure forces upon them. It is the aim of the generalization
test to indicate the kinds of “similarity” that may exist. As Kliiver (1933)
has documented in elegant detail, it is difficult to disentangle the actual
dimensions of visual analysis from even an assortment of such tests. The
assumption that fish possess inherently fewer and more rigid classifications
than the higher mammals is plausible but is by no means demonstrated.
Ingle ( 1971 ) has speculated that large-scale “gestalt” features such