Our research concentrates on understanding and generating intelligent behavior with artificial neural networks. On one hand, the goal is to better understand human information processing, that is, how intelligent behavior in humans arises from neural network mechanisms. On the other, the research aims at building more intelligent artificial systems. Our approach is to develop algorithms and architectures that explicitly represent and make use of the structure in the task, such as schemas, subgoals, and modularity. This way it is possible to build neural network models of more complex behavior than is possible with traditional uniform network architectures. For example, high-level processes such as schema learning, sentence understanding, and game playing can be implemented with modular neural networks, and such systems can often be more efficient and cognitively valid than traditional models.