Abstract
We used a visual exclusive-or (XOR) task to probe the processes that support the abilities of monkeys to learn to recognize large numbers of individuals within an object class. This performance is similar to people recognizing faces — a process that is generally assumed to rely on feature conjunctions and configural information, and therefore it has been implicitly assumed that the same processess support the monkeys' abilities. This has not been explicitly tested. Three macaque monkeys were tested on a variety of visual classification task using an XOR paradigm. The images were pictures of common objects (e.g. butterflies) or shapes in which image parts were reuseable and thus different images might require 1, 2, or 3 features for correct assignment. We manipulated object scale, geometry, and the use of cues to determine if the monkeys were sensitive to the configural properties of the objects. All three monkeys were able to eventually master a visual XOR object classification task, demonstrating that macaque monkeys can solve visual classification tasks requiring the use of conjunctions. However, in every circumstance, the XOR tasks were “hard”, i.e. they took longer to learn, were classified less accurately, and were classified more slowly than objects with diagnostic dimensions, even after 1000's of trials. There was a monotonic relationship between the number of conjunctions required for classification and RT. Manipulating the orientation of the images and their internal features implied the monkeys were using different strategies and also that they were sensitive to the global geometric properties. The discrepancy between performance on typical tasks and the XOR tasks challenges the assumption that feature conjunctions are a typical component of a monkey's approach to the usual laboratory visual object classification task.
Burroughs Wellcome Fund, James S. McDonnell Fund