Figure 2.

A schematic representation of the cWCST and the parallel RL model. A negative feedback on the cWCST (top) indicates that the previously executed response (not depicted) was incorrect, implying that a category switch is required. The subsequent stimulus card is sorted by the color category, as indicated by observing response 1. The parallel RL model (bottom) assumes independent trial-by-trial cognitive and sensorimotor learning (upper and lower grey bar, respectively). Core to cognitive and sensorimotor learning are feedback predictions for the application of categories ($Q_C$) and the execution of responses ($Q_S$), respectively. Feedback predictions are updated in response to received feedback. Individual cognitive (${\alpha }_C$) and sensorimotor learning rates (${\alpha }_S$), which are further separated for positive and negative feedback, quantify the strengths of updating. For the subsequent target, cognitive and sensorimotor feedback predictions are integrated. Response probabilities are rendered from these integrated feedback predictions, with an individual inverse temperature parameter ($\mathsf{\tau }$) quantifies how well response probabilities accord to integrated feedback predictions. Retention mechanisms ensure that feedback predictions transfer to the next trial. Here, individual cognitive (${\gamma }_C$) and sensorimotor retention rates (${\gamma }_S$) quantify the strengths of retention of feedback predictions.