A neural network model of inhibitory processes in subliminal priming

Masked priming experiments have revealed a precise set of facilitatory and inhibitory visuomotor control processes. Most notably, inhibitory effects have been identified in which prime-target compatibility induces performance costs and prime-target incompatibility induces performance benefits. We argue that this profile of data is commensurate with an “emergency braking mechanism”, whereby responses can be retracted as a result of changing sensory evidence. The main contribution of this paper is to provide a neural network-based explanation of this phenomenon. This is obtained through the use of feedforward inhibition to implement backward masking, lateral inhibition to implement response competition, and opponent processing mechanisms to implement response retraction. Although the model remains simple, it does a very good job of reproducing the available masked priming data. For example, it reproduces a large spectrum of reaction time data across a number of different experimental conditions. Perhaps most notably, however, it also reproduces lateralized readiness potentials that have been recorded while subjects perform different conditions. In addition, it provides a concrete set of testable predictions.