Artificial intelligence planning systems are being put to use to determine the activities of a wide range of intelligent interactive systems. The ability for these kinds of systems to explain their plans to human users is essential for the systems' successful adoption and use. This project is investigating the generation of natural language descriptions of computer plans.
This work is developing a cognitive and computational model of task context and its role in the generation of action descriptions, specifically, the means by which negative constraints and justifications are used to create more effective task descriptions. The project methodology includes both experimental and theoretical aspects; naturally occurring text corpora collected by the project is used to form a computational model for the production of plan descriptions that accounts for the discourse features described above. This model is then empirically evaluated to determine the model's efficacy.
The work demonstrates the effective use of automatically generated plan structures as underlying knowledge representations for task-based discourse. The results have a direct impact on the use of intelligent information technologies used in training and education. This enables applications that provide context-sensitive help to computer users that are themselves not experts in information technology, for example in the automatic generation of instructions in situations where pre-designed instructional materials or other resources are not available.
