A diverse range of behaviors including language, event segmentation, and learning to play an instrument involve parsing complex temporal sequences. Item co-occurrence in a sequence can exist between pairs of items shown consecutively, as well as triplets, quadruplets and other higher order sets. For example, when learning a language both humans and natural language processing algorithms make use of first order statistics about which words tend to follow others, and higher-order statistics about the grammar of sentences and paragraphs. The statistics of this co-occurrence between sets of different sizes can influence how well people learn the sequence. We can demonstrate this learning by presenting stimuli in a sequence determined by a walk on a highly clustered graph where stimuli within a large cluster tend to co-occur more frequently than stimuli in different clusters. When subjects are shown motor cues generated from a walk on such a graph, they exhibit shorter reaction times and more similar BOLD activity for successive cues drawn from the same cluster than when they are drawn from different clusters. Generalizable theories of hippocampal function suggest that its role in learning and predicting abstract relationships could subserve this behavior. In line with this theory, neuroimaging data implicates the hippocampus and downstream cortical regions in this behavior. However, direct electrophysiological measures of neural activity during the performance of this task have not been examined, leaving the description of the neural underpinnings of the behavior incomplete. We propose to use electrocorticography (ECoG) to test the hypothesis that activity in the hippocampus reflects behavioral change across clusters, and is sensitive to higher-order associations between stimuli. Our investigation is an important next step in testing theories of how the hippocampus aids relational learning, and it will add to knowledge of how information in the world is represented in the brain and how it aids or hinders learning.
PROJECT NARATIVE Identifying complex relationships between items in a sequence is a crucial component of learning in humans, and is thought to rely on activity in the hippocampus. However, the neural substrates that allow humans to learn higher-order statistics among stimuli in a sequence have not been fully explored, greatly limiting our understanding of sequential information encoding in the brain. We propose to use electrocorticography (ECoG) to test the hypothesis that hippocampal activity reflects these complex relationships between items, and tracks behavioral changes that demonstrate learning.
