6
M. J. WELLS
predict with a very fair degree of accuracy the proportion of errors that
octopuses will make (Sutherland, 1957; but see also Deutsch and
Sutherland, 1960 ; Dodwell and Sutherland, 1961).
There is evidence that horizontal extents are more accurately
assessed than vertical extents (Mackintosh et al., 1963). They also seem
to be more important in determining behaviour. Thus the shapes
shown in Fig. 3 are arranged in order of their similarity to octopuses.
The animals were trained to distinguish between the cross and the
Cross
I
2
3
4
5
6
+-+/Ase 7 8 9 1 0 II 1 2 1 3
1 4
15
1 6
1 7
1 8
19
20
2 1
22
23
24
25
26
27 Square
FIG. 3. Further shapes that octopuses c m distinguish by sight. The order of the
figures is determined by the number of attacks octopuses made on each in transfer
tests following training to distinguish between the cross and the square. (From
Sutherland, 1962.)
square, and subsequently tested with the other figures to determine,
from the number of attacks made upon them, how like the original
two figures they appeared to the trained experimental animals (Sutherland, 1962). The order of the figures appears to us to be determined
along a dimension perhaps best described as " open " (large perimeter
to area ratio) to " closed " (small perimeterlarea). Some of the details
of the order seem illogical unless one supposes that the animal pays
more attention to horizontal than to vertical extents. It is otherwise
not at all obvious why, for example, the horizontal rectangle 8 should
be treated as so much more like a cross than the vertical 19, and shape 3
so differently assessed from 20, again the same shape in a different
orientation.