Figure 5. Accounting for the dual effects of training on audiovisual integration using a Bayesian causal inference model.

(A) Improvement in the precision of SOA estimates explains narrowing of the temporal binding window, but predicts a concomitant increase in the magnitude of the ventriloquist effect with near-synchronous stimuli. Model simulations are shown for three levels of σ_Δt on audiovisual simultaneity discrimination (left panel), the temporal tuning of the ventriloquist effect (middle panel) and the spatial tuning of the ventriloquist effect (right panel). (B) Changes in prior expectations explain the reduction in the amplitude of the ventriloquist effect. Model simulations are shown for the same three tasks, with different functions representing different levels of prior expectation about whether audiovisual stimuli relate to a common cause (p_common). (C) Coupling reductions in both σ_Δtand p_common captures training-induced changes in performance across all three tasks. Unless otherwise indicated, simulations were carried out using the following parameter set: σ_Δt = 80 ms, σ_υ = 1deg., σ_a = 2deg., p_common = 0.5.