Learning to Perform Role-Filler Binding with Schematic Knowledge

Through specific experiences, humans learn structural relationships underlying events in the world. Generalizing knowledge of structural relationships to new situations requires dynamic role-filler binding, the ability to associate specific "fillers" with abstract "roles". Previous work found that artificial neural networks can learn this ability when explicitly told what the roles and fillers are. We show that networks can learn these relationships even without explicitly labeled roles and fillers, and show that analyses inspired by neural decoding can provide a means of understanding what the networks have learned.