The interrelatedness of lexical items, typically defined in termsof semantic or phonological overlap, has been shown toinfluence language learning. Given that language also containssequential structure, we investigate here whether temporaloverlap among words, formalized in graph theoretical terms asdisplaying the property of community structure, might alsohave consequences for learning. We create a graph organizedinto clusters of densely interconnected nodes with relativelysparse external connections. After assigning a novelpseudoword to each node in the graph, we generate acontinuous sequence of visually-presented items by walkingalong its edges. Word-by-word reading times suggest thatlearners are indeed sensitive to temporal overlap.Compellingly, we also demonstrate that prior exposure tosequences organized into temporal communities influencesperformance on a subsequent word recognition task.