Fig. 8.
Predicted results for the cue-weighting model based on the distributional statistics of the two cues. These responses were obtained by setting the parameters of the Gaussians in the input-level MOGs to the means and standard deviations of each cue and training the model (i.e. keeping the parameters at the cue-level MOGs constant while allowing the model to learn the parameters at the combined MOG). The model shows a much smaller trading relation than the one observed in Simulation 2 and in the data from listeners.