Resolution of linear entity and path geometries expressed via partially-geospatial natural language

Abstract

When conveying geospatial information via natural language, people typically combine implicit, commonsense knowledge with explicitly-stated information. Usually, much of this is contextual and relies on establishing locations by relating them to other locations mentioned earlier in the conversation. Because people and objects move through the world, a common and useful kind of geospatial phrase is the path expression, which is formed by designating multiple locations as landmarks on the path and relating those landmarks to one another in sequence. These phrases often include nongeospatial information, and the paths often include linear entities. This thesis builds upon the work done for the GeoCoder spatial reasoning system, by addressing several of its limitations and extending its functionality.