Generating descriptions of complex activities

Intelligent systems are often called upon to form plans that direct their own or other agents' activities. For these systems, the ability to describe plans to people in natural ways is an essential aspect of their interface. To generate task-related discourse, computational linguists have put the data structures that make up plans to effective use as an underlying knowledge representation. There is a mismatch, however, in the level of detail present in the simplest of plans and the detail found in people's descriptions of them. As a result, computer systems responsible for automatically generating descriptions of plans must pay careful attention to the interaction between the quality of a description and the quantity of information it contains. In this thesis, I present the Cooperative Plan Identification (CPI) architecture, a computational model that generates concise textual descriptions of plans. The CPI architecture selects appropriate content to include in a plan description based on Grice's conversational maxim of quantity. In this model, speakers and hearers are implicit collaborators in their communication about a plan. A hearer interprets a concise plan description by filling in the missing detail using plan reasoning. A cooperative speaker selects the content of a plan description based on his expectation that the hearer is able to complete the description in much the same way that a planning system completes a partial plan. The CPI architecture generates plan descriptions by heuristic search, using a model of the hearer's interpretation process--her plan reasoning, her plan-related preferences and her reasoning resource bounds--to select candidate plans that are both concise and effective. The architecture has been empirically evaluated in an experiment which I also describe here. In this experiment, subjects were given written instructions and asked to carry out the instructions within a virtual environment. The results of this experiment clearly showed that subjects following instructions produced by the CPI architecture performed their tasks with fewer execution errors and achieved a higher percentage of their tasks' goals than did subjects following instructions produced by alternative methods.