An experiment on shape discrimination and signal detection in octopus

Ten octopuses were trained to perform a successive discrimination between the two shapes shown in Figure I (a). After 7 days of training, when performance was significantly above chance, transfer tests were given with other shapes that were either rotations or parts of the original training shapes. At least six theories have been put forward to explain shape discrimination in the octopus, but none of these are capable of explaining the present results. The transfer tests suggest that the discrimination was performed in terms of component parts of the shapes (vertical bars projecting upwards or downwards), and their relationship to the shape as a whole (terminal or central).

During successive discrimination training the general level of attack varies between animals, and fluctuates from day to day. As a result there are often more attacks on both the positive and negative shapes on some occasions than others, making it difficult to compare the levels of discrimination achieved. It is suggested that the concepts of signal detection theory can help overcome this difficulty. Attacks on the positive shape (“hits”) plotted against attacks on the negative shape (“false positives”) constitute an ROC curve from which a value of d', independant of the general level of attack, can be obtained.