Using commonsensical cardinal directions to describe bordering objects.

Geographic information systems (GIS) customarily encode spatial information using geometric objects (points, polylines and polygons) and their locations. But people frequently use qualitative relations, such as topological relations (e.g., connection or overlap) or cardinal direction relations (e.g. North or Southeast), to describe spatial scenes. While topological relations have been integrated into modern GIS, direction relations have remained isolated from GIS and are not available for user interaction. Instead, a user must visually infer them from map depictions. This work uses the problem of generating and interpreting cardinal direction labels that describe the direction between a region and its surrounding neighbors (e.g., all neighbors of a US state) to identify principles for computing more qualitative descriptions of directions that are intuitive to people and to correctly interpret descriptions commonly used by people. This is a step towards bridging the qualitative-quantitative divide between spatial information systems and human conceptualizations of space.