Acquiring a domain-specific task model is an essential and notoriously challenging aspect of building knowledge-based systems. This paper presents machine learning techniques which are built into an interface that eases this knowledge acquisition task. These techniques infer hierarchical models, including parameters for non-primitive actions, from partially-annotated demonstrations. Such task models can be used for plan recognition, intelligent tutoring, and other collaborative activities. Among the contributions of this work are a sound and complete learning algorithm and empirical results that measure the utility of possible annotations. |
