Assessing how much two geographic databases reflect the same point of view is a key issue for data integration. We argue that this task requires developing ontologies revealing the point of view of each piece of information, and neglecting the technical choices behind the organization of information in data schemas. These ontologies need then to be aligned and globally compared. In this paper, we describe techniques allowing doing that in the fields of ontology alignment and natural language processing. We illustrate those points through results from experiments made on actual data with a semi-automated analysis of their specifications.