Abstract

A connectionist model is used to verify a possible dissociation between two quantification processes: "subitizing", the rapid enumeration of small patterns, and "counting", the serial quantification process based on some formal principles. A three-layer neural network, with the hidden layer selectively connected with output units for subitizing, is better able to specify numerosity of new sets of elements than either a perception or a three-layer neural network with hidden units connected only with outputs units for counting.