Sensitivity to Temporal Community Structure in the Language Domain

The interrelatedness of lexical items, typically defined in terms of semantic or phonological overlap, has been shown to influence language learning. Given that language also contains sequential structure, we investigate here whether temporal overlap among words, formalized in graph theoretical terms as displaying the property of community structure, might also have consequences for learning. We create a graph organized into clusters of densely interconnected nodes with relatively sparse external connections. After assigning a novel pseudoword to each node in the graph, we generate a continuous sequence of visually-presented items by walking along its edges. Word-by-word reading times suggest that learners are indeed sensitive to temporal overlap. Compellingly, we also demonstrate that prior exposure to sequences organized into temporal communities influences performance on a subsequent word recognition task.