Syntactic rules of spatial relations in natural language

Abstract

Spatial relations are the main part of geographical information in natural language. Their extraction and semantic interpretation play a significant role in bridging the gap between geographical information system and natural language. Normally spatial relations are described with certain spatial terms and syntactic rules in natural language. To overcome the disadvantage of manual induction of syntactic rules, this paper proposes a new machine learning approach based on a sequence alignment algorithm. Firstly, the description instances of spatial relations in a large-scale annotated corpus are extracted and analyzed, and the sequence alignment algorithm is used to calculate the pattern similarity between instances of spatial relations. Then, the instances with high similarity are generalized aspopularly used syntactic rules. Finally, these rules are used for extraction spatial relations in a test data to evaluate their validation. The experimental results indicate that the generalized rules can achieve better performance than those rules induced according to occurrence frequency in the corpus. © 2013 Springer Science+Business Media New York.