Using Event Semantics for Toponym Disambiguation

Abstract

This chapter discusses a method for improving the disambiguation of location names using limited event semantics. Location names are often ambiguous, as the same name may refer to locations in different states, countries, or continents. Ambiguous location names, also known as toponyms, need to be disambiguated (or grounded) when resolving many spatial relations expressed in textual documents. Previous methods for disambiguating toponyms have utilized simple heuristics, statistical ranking, and ontological methods in order to resolve a location reference. However, since toponyms are used in documents that refer to events, semantic knowledge characterizing events can be used to ground location names. We propose an ontology-based method with a technique that considers the participants in events such as people, organizations, and other locations. Event semantics are integrated into an ontology that is used to distinguish geographical names through a probabilistic approach based on logistic regression. Our experimental results on the SpatialML corpus (Mani et al., 2008) indicate that using event structures improves the quality of disambiguated toponyms.