A model for landmark learning in the honey-bee

Summary

• 1. A single bee is rewarded with sugar for visits to the centre of a circle of eight landmarks.

• 2. After training, tests are given in which the sugar is removed and the flight path of the bee recorded as it searches for it. The bee persistently searches a small area in the centre of the circle, demonstrating that it has learnt the position of the sugar relative to the landmarks (Fig. 1A, C).

• 3. In a second test, three landmarks are removed leaving a semi-circle. The bee now searches a small area “inside” the semi-circle (Fig. 1B, D). In its chosen position none of the landmarks match in size, bearing, or angular separation with the values of the landmarks present during training (Fig. 2), showing that a “retinal matching” model of landmark learning does not predict these results.

• 4. It is suggested that the bee is not measuring the position of individual landmarks but the overall landmark configuration, a “Gestalt” hypothesis first put forward by van Beusekom (1948). It is hypothesised that the bee moves “inside” the semi-circle in an attempt to have landmarks “surrounding” it as it had when collecting sugar from the centre of the circle.

• 5. The term “sourroundedness”, a concept similar to the Gestalt idea of “degree of closure” is given a rigid definition. A term describing landmark average distance and a comparison process are also defined, these terms are incorporated into a computer program.

• 6. The computer program successfully predicts the results of a series of experiments with various training and testing situations (Figs. 1, 5, 6, 7) and some results found by van Beusekom (1948) with the digger wasp (Fig. 8).

• 7. The model is compared with a “retinal” model and its behaviour in real world situations discussed. It is concluded that the model, although containing some arbitrary features, represents a possible, rigidly defined and testable model of the landmark learning of the bee.