General Laws of Learning

Some of the general laws of learning seem to apply in the spatial domain. The next section reviews evidence for overshadowing, spatial generalization, and peak shift.

Overshadowing in use of landmarks

Pigeons and humans were trained with two situations, with nonovershadowed and overshadowed landmarks, respectively (Spetch, 1995). The square indicating the goal was absent after preliminary training. The overshadowed and nonovershadowed landmarks were the same distance from the target (goal). The nonovershadowed landmark was nearest to the target, but the overshadowed landmark was not the nearest to the target. The identities and locations of overshadowed and nonovershadowed landmarks were counterbalanced across subjects. The data show the proportion of responses in the goal area. Both pigeons and humans learned to use the nonovershadowed landmark more effectively for locating the target.

Spatial generalization

Pigeons were trained to peck at a touch screen monitor with one stimulus shown on it (Cheng, Spetch, & Johnston, 1997) as shown in the upper figure. During training, the stimulus appeared at one of two locations shown in red. When the stimulus was at the S+ location (the red square on the left in this figure), pecks were rewarded. When the stimulus was at the S- location, pecks were not rewarded. The birds were then given occasional unrewarded tests at all the locations shown. On generalization tests, the location of the stimulus varied in the vertical dimension, up and down from the training location on the monitor. Results in the next figure show a generalization gradient much like that found for other stimulus dimensions, with the training location (red point) receiving the most pecks, and progressively falling off to each side (blue points).

Spatial Peak Shift

In the same experiment, the birds were also given tests in which the stimulus location varied along the horizontal dimension. In this case, because they were trained with an unrewarded location horizontally displaced from the rewarded location (S+), the peak of responding was not at S+, but displaced from S+ in the direction away from S-. This phenomenon of peak shift is commonly found in other stimulus dimensions, for example, in wavelength of light (Hanson, 1959).

We think that other general laws of learning in the spatial domain are also likely (see next section). Blocking is an example. In fact, in some ways, spatial learning is much like learning in other domains. Undeniably, however, spatial learning also relies on special, dedicated mechanisms such as a sun compass, an odometer, or a geometric module. Spatial cognition is a mix of special mechanisms and general laws. We do not believe that it is either exclusively general all-purpose learning or exclusively a special domain unto its own.