Extracting Spatial Information From Place Descriptions

A computational model of understanding place descriptions is a cardinal issue in multiple disciplines and provides critical applications especially in dialog-driven geolocation services. This research targets the automated extraction of spatial triplets to represent qualitative spatial relations between recognized places from natural language place descriptions via a simple class of locative expressions. We attempt to produce triplets, informative and convenient enough as a medium to convert verbal descriptions to graph representations of places and their relationships. We present a reasoning approach devoid of any external resources (such as maps, path geometries or robotic vision) for understanding place descriptions. We then apply our methodologies to situated place descriptions and study the results, its errors and implied future research.