Mining the Semantic Similarity of Spatial Relations from Text

Spatial relations are one of the most important components in a location description, conveying information about proximity, direction, adjacency and topology among other things. However, despite being studied for many years, the semantics of spatial relations are still not well understood,
particularly given that the use of spatial relations can vary with context. In this paper we investigate whether it is possible to mine the semantics of spatial relations from text, particularly focusing on semantic similarity, but also exploring the extraction of richer semantic information
about the relationships between spatial relations, with the long term goal of moving towards the automation of the interpretation and generation of locative expressions. We test three similarity methods, including a bag of words technique, with both general and geospatial corpora, and
using word embeddings. We compare the results to ground truth data from human subjects experiments.