Limits of generalisation in categorical judgments

Abstract
Categorisation is central to perception and cognition. Category-selective neurons in parietal and prefrontal cortex appear to respond in many different contexts (a single neuron might be selective for dogs vs. cats, but also for sports cars vs. sedans). Category-selective neural representations that generalise across contexts may increase efficiency and facilitate learning. However, they can potentially lead to interference – for instance when multiple potential category judgments can be applied to the same set of stimuli. Computational modeling suggests that in these situations, context-generalised neural representations are suboptimal. I will present evidence from behaviour and EEG in humans that is consistent with this framework.