Interactive computing environments provide facilities to support and assist the range from novice to expert users, but casual and novice users tend to rely on a small starter set of commands. This proposal for thesis work addresses this problem through the implementation of GENIE (GENerated Informative Explanations), a system that answers users' questions about how to accomplish tasks in the domain of Berkeley Unix Mail. This work unifies three new perspectives on consulting. First, the decision on what to tell a user, including the "best" plan for the user's goal, is based on an evaluation of the user's current computational goal, and the goals the user has attempted in the past. Secondly, the decision on how to phrase the answer relies on a careful mixture of tutoring strategies. Finally, both an expert and user model are represented as declarative structures of goals with alternative plans that include explicit semantic relationships between plans.
展开▼
