Abstract:

Most previous work on learning task models, a special case of the well-known knowledge acquisition bottleneck, has dealt with non-hierarchical models. We present and analyze techniques for inferring a hierarchical task model from partially-annotated examples of task-solving behavior. We show our algorithm has desirable formal properties and that both restrictive and preference biases are useful for generating effective models. Finally, we describe experiments that explore the appropriate division of labor between the learning algorithm and the person who provides the annotated examples.