Do Sentence Transformers Learn Quasi-Geospatial Concepts from General Text?