Many theories of visual recognition hold that complex objects are represented as compositions of features and the relations between them.

This research is concerned with how people learn the statistical regularities among features of complex patterns.

In the example below, the pattern in the lower right corner is made up of horiozontal and vertical pairs, as well as 'intermediate' singletons.

In our experiments, such pairs were used to generate many differnt patterns and test whether people learn the regularities between them.
Note that the pairs are spatially separated so that other features appear intermediate between them. See here for more details.