A natural language interface for crime-related spatial queries

Web-based mapping applications such as Google Maps or Virtual Earth have become increasingly popular. However, current map search is still keyword-based and supports a limited number of spatial predicates. In this paper, we build towards a natural language query interface to spatial databases to answer crime-related spatial queries. The system has two main advantages compared with interfaces such as Google Maps: (1) It allows query conditions to be expressed in natural language, and (2) It supports a larger number of spatial predicates, such as ldquowithin 3 milesrdquo and ldquoclose tordquo. The system is evaluated using a set of crime-related queries run against a dataset that contains many spatial layers in the Denton, Texas area. The results show that our approach significantly outperforms Google Maps when processing complicated spatial queries.