Categorisation judgments based on multiple sources of evidence: neural and computational mechanisms
Much research has been devoted to understanding how humans and other primates choose an appropriate course of action on the basis of noisy sensory information. Many current models assume that evidence integration is a linear, accumulation-to-bound mechanism. But cognitive psychologists instead emphasize that central processing is capacity-limited, and that sensory information must pass through a performance-limiting 'bottleneck' before influencing behaviour. I will describe two EEG experiments supporting the idea that the passage of sensory evidence into an accumulation process is gated according to the phase of slow (2Hz) oscillatory changes in excitability over the parietal cortex. Information that successfully passes the bottleneck is encoded in an additive fashion in lateralized beta-band EEG activity over motor cortex. When attention is divided, information in the irrelevant stream fails to reach this accumulator phase and instead leaks away, as proposed by 'late selection' models of attention. In a third study, we propose a neural network model of this integration and decision process and assess its predictions in human fMRI signals. This study confirms the distinction between central (multiplicative, decision-weighting) and motor (additive, accumulation) stages in the decision. Together, these data suggest that the integration of evidence during human perceptual categorisation is a fluctuating, capacity-limited process.