To study commonsense reasoning and to write programs that reason effectively about the physical world, we must understand the nature of perspective-taking in qualitative reasoning — which perspectives to take, how to represent them, and when to shift from one perspective to another.

This dissertation defines a task-driven approach to perspective-taking for automated qualitative reasoning. Central to our approach is the notion that model formulation and selection is an integral pan of reasoning about complex physical systems. Given a task, models can be created that reflect different configurations of the system topology, at various structural granularities, and in distinct language forms.

This dissertation describes a computational framework that integrates three perspective-taking dimensions in reasoning about electronic circuits: topological configuration, structural aggregation, and ontological choice. The research shows that by using a task-driven, perspective-taking approach, we can extend the range of automated qualitative causal reasoning about complex physical systems.