Abstract
Sequential sampling models have been used to account for latency distributions of saccades directed to a single target (1). In these models saccade generation is a process of accumulating activity or evidence from a baseline level to a response criterion.
We present an extension of this basic scheme. Saccade direction is coded by an one-dimensional motor map. The units within this map accumulate evidence in favour of a target being in their “response field”. A saccade is generated whenever the activity of one unit reaches criterion. The saccade direction is determined by the direction coded by the winning unit.
The model contains a number of critical features: 1) population coding: the onset of a target activates a large number of units in a graded fashion so that the drift rate is highest for the central unit, but drops-off gradually with increasing distance away from the target; 2) passive decay: the accumulation of activity is leaky; 3) two forms of noise: variability in the input into the motor map triggered by the onset of a target, and variability in the units' activity during accumulation.
The model was tested against empirical data from a double-step study (2). It effectively accounts for: 1) variability in saccade latency and direction under single target conditions; 2) the shape of direction transition functions under double-step conditions; 3) global effect saccades when the angular separation between two target steps is small; 4) the basic gap effect in saccade latency and its lack of influence on the saccadic dead time; 5) the variation in saccadic dead time with increasing target separation.
1. R. H. S. Carpenter, M. L. L. Williams. (1995) Nature. 377, 59-62.
2. C. J. H. Ludwig, J. W. Mildinhall, I. D. Gilchrist. (2007). Journal of Neurophysiology.