Decoding Neural Representations of Rules and Rule Order

Abstract

Event Abstract Back to Event Decoding Neural Representations of Rules and Rule Order Kai Görgen1, Carlo Reverberi1* and John-Dylan Haynes1, 2 1 Charité-Universitätsmedizin, Bernstein Center for Computational Neuroscience, Germany 2 Max Planck Institute for Cognitive and Brain Sciences, MPI, Germany Conditional rules of the form "if x then y" are of vital importance for our everyday life. Recent research has started to investigate the neural substrate underlying conditional rule processing. A critical aspect of rule processing that has so far been neglected is that many situations require the rules to be executed in a specific order. For example, it might be wise to apply the rule “If you don't have any money, go to the ATM machine” BEFORE the rule “If you want ice-cream, go to the ice-cream shop”, instead of applying them vice versa. Here, we present the results from a functional magnetic resonance imaging (fMRI) study in which we successfully decoded the identity of two simultaneously active rules as well as their application order from local patterns of fMRI data. A modified cued task-switching paradigm was administered during fMRI scanning. Participants were required to retrieve, maintain, and apply two conditional rules (e.g. “First, if there is a musical instrument, press left. Second, if there is a food item, press right”) to a set of target stimuli. We identified regions containing representations of rule identity and application order by applying multivariate pattern classifiers on local patterns of activation during the delay period between cue presentation and task execution. Using a redundant coding scheme for the rules (each rule was alternatively represented by one of two visual cues in different trials), we were able to disentangle representations of rules and representations of visual features of the cue. Representations of rule identity were found in bilateral supero-lateral parietal cortices, right dorso-lateral prefrontal cortex (dlPFC), and right anterior cingulate cortex (ACC). Representations of application order were found in bilateral supero-lateral parietal cortices and right ACC. In ACC, we found an overlap between regions that contained representations of rules and rule order. In contrast, a dissociation between both regions was found in supero-lateral parietal cortices, as well as in lateral PFC (where only representations of rules but not of rule order were found). The locations of rule identity representations fit well with recent findings by Reverberi & Haynes [1]. The finding that regions containing representations of rule and rule order were dissociated in some regions suggests that both features of rule processing are handled in different ways by the brain. In conclusion, we were able to successfully localize representations of two defining features of complex rule sets, namely rule identity and rule order. Additionally, we found evidence that the brain treats these two types of information in different ways.