SPATIAL LEARNING: ANIMALS

Resources that animals need are usually distributed patchily within their home range, and many animals learn where they are and how to reach them. Stuart Altman describes how one troop of baboons responded to ripe berries on an isolated bush in the center of their home range as a sign of their availability elsewhere and trekked to a remote patch of bushes bearing the same fruit.

Some animals cache food when it is abundant, remember the precise locations of the caches for long periods, and return to empty the caches when food is scarce. Clark's nutcracker provides a dramatic example. The birds collect tens of thousands of pine seeds in the autumn for recovery during the subsequent winter and spring. Scrub jays caching food in the laboratory remember not only where they have cached it but also which items they have stored in which locations. Proof came from an experiment by Nicola Clayton and Anthony Dickinson in which birds were given two different foods to cache. Before the birds were allowed to recover either food, they were prefed with one. Prefeeding caused the birds to focus their search on sites containing the other untasted and so more appealing food—a result that implies that birds know where they have hidden particular food items. By using food that rapidly rots, the same experimenters showed that scrub jays also know when they made the cache. Lastly, birds remember that they have emptied a cache, and they avoid revisiting empty sites. Other resources that animals remember include watering sites, nests, places where mates are to be encountered, shelters, and bolt holes. Spatial knowledge is closely integrated with other forms of knowledge that may influence what spatial information is stored, when it is retrieved, and how long it is remembered.

Some animals have evolved set procedures for acquiring needed navigational information. Indigo buntings learn the constellations of stars around the North Star, and they use that constellation for guidance in their migration. The birds are preprogrammed to learn the unique pattern of stationary stars that lies close to the Earth's axis of rotation and so do not move across the night sky. Wasps and bees perform highly structured learning or orientation flights when first leaving a new feeding site or their nest. The flight is designed to emphasise landmarks that are close to the goal and that can thus provide precise navigational information. In a single such flight, they learn enough about the arrangement of landmarks to be able to return there. Rats explore a new environment and reexplore a familiar one when changes occur. During exploration they learn the layout of the environment without the benefit of any immediate reward. In one experiment to demonstrate such learning, rats explored a T-maze with two visually distinct goal boxes and with all extra-maze cues screened off. After the rats had explored the empty maze several times, they were placed singly in one of the goal boxes and allowed to eat there. When re-placed at the entrance, most rats returned directly to the box where they had been fed. A control group fed in one goal box without prior exploration of that particular maze showed no tendency to return to the same box. Exploration allows the rats to learn the paths to different recognized locations that only later come to be associated with a valued resource.

It is remarkable that a wide array of animals, from insects to primates, acquire and store the same two distinct types of spatial information. One kind is derived from dead reckoning, also known as path integration. An animal leaving its home base continuously monitors the direction and distance of the path that it takes and uses this path-derived information to keep an updated record of its direction and distance from home. Consequently, it is always able to take a direct path home, even after a circuitous outward journey. On finding a good source of food, both insects and mammals store the path integration coordinates of the food site. Equipped with these stored coordinates, they can later return to that site by subtracting the coordinates of the site from their current coordinates as given by path integration.

The second kind of stored spatial information does not have positional coordinates attached to it. It comes from memories of visual landmarks that can indicate a site or a route or signal what kind of action the animal ought to perform there. It is still unclear whether there exists in any animal an intimate connection between memories of landmarks and their positional coordinates. In insects the available evidence suggests that memories of landmarks and of positional coordinates are kept separate.

Some insects, birds, and possibly mammals learn the position of a site in terms of the distance and direction of one or more visual landmarks from it. They then return to the site by moving until their current distances and directions from those landmarks match the stored ones. Precision in locating the site is enhanced by learning the distances and directions of several landmarks and by emphasizing information that comes from landmarks that are nearer to the site. It is likely that mammals also learn the spatial relationships between landmarks, but experimental corroboration has proved elusive.

The richness of animals' memories of the arrangement of landmarks is illustrated by an experiment by David Brodbeck on chickadees. It showed that these birds learn a caching site in terms of several potential retrieval cues. Birds were trained to find seeds in one wooden block in an array of four differently decorated blocks that were attached to a wall in a large aviary. The site is thus specified by room cues, position in the array, and the appearance of the baited block. On test trials the array was shifted as a group, so that one block was moved to the location occupied by the baited block, and the other blocks in the array were rearranged. Birds in these tests looked for their seeds following a consistent order. They first searched at the site that was specified by the landmarks in the room, then at the site defined by position within the array of blocks, and lastly in the block that was correctly decorated. Birds had thus learned all these properties of the caching site and were guided by them in a set hierarchy.

A similar example of a predisposition to learn multiple features of the surroundings of a significant site comes from insects. Honeybees learn both the local landmarks that pinpoint a site and the panoramic context within which the local landmarks are set. As a bee flies within the vicinity of the site, the appearance of the distant panorama changes less with the bee's movements than does the appearance of local landmarks. Consequently the distant panorama can be recognized more reliably than the local landmarks. Insects exploit this piece of ecological geometry and use their memory of the panorama to prime the recall of local landmarks, which can thus be identified when viewed from unusual viewpoints or under unusual lighting conditions.

To reach a desired goal animals either follow familiar paths or, more rarely, plan a new route from an arbitrary starting point to a goal that is not visible from the starting point. Evidence for such route planning comes from a study by Charles Menzel and colleagues on a young Bonobo chimpanzee, Kanzi. Kanzi could be told through lexigrams which out of several goals in a familiar wooded area he should head for. The chimpanzee was led to an arbitrary starting point and then given a lexigram signaling a familiar location that was out of sight and that harbored food or some other desirable item. On all occasions Kanzi led a human companion to the correct location. Sometimes Kanzi took direct trails, and sometimes he followed a more indirect path. There is as yet little understanding of the behavioral mechanisms underlying route planning.

Although ants and bees can reach a familiar feeding site solely by means of path integration, they notice and approach prominent objects on the route and rapidly come to learn the appearance of these landmarks. The landmarks are used to segment the route, and insects learn vectors of the appropriate direction and distance to take them from one landmark to the next. The division of a route into sections increases the insect's chances of reaching its goal. First, prominent beacons that attract the insect at the end of each segment make it easy to correct for inaccurate vectors or crosswinds that deflect the insect from its proper path. Second, the precision with which the insect is delivered to its goal depends on the accuracy of the vector from the final beacon rather than on a vector covering the whole path from start to finish. Again we see that animals are naturally biased to learn features of their environment that improve navigational accuracy. Spatial learning is strongly guided by built-in mechanisms and strategies that predispose animals to learn about navigationally useful features of their environment, often in anticipation of their need for later guidance.