The ability to correctly interpret geospatial information is undoubtedly a difficult exercise for artificial intelligence. While humans easily deal with this problem, it is an enormous challenge for computers. The importance of this problem is even more acute when considering that the Internet is crowded with geospatial references. More importantly, with the advent of mobile devices at a very large scale, the ability to find and classify locations becomes a very demanding feature.

This necessity has not been overlooked by the most prominent Internet providers who already offer a wide range of services for accessing geospatial information. However, the quality of the results is still far from our ability to deal with the same pieces of information.

As usually happens with automated systems that have to reason with incomplete yet imperfect information, a first step consists of creating an ontology that recognizes the most important concepts and relates them in a consistent manner. This paper introduces OnLocus, an ontology that recognizes the following concepts: place, territorial division, landmark, place descriptor, address, place name, and positioning expressions. In contrast with other similar ontologies, it is domain-dependent since it exploits concepts that are found in most urban communities. As a matter of fact, the ontology has been proven on a number of Brazilian Web pages. It can also be seen as an extraction ontology since it provides services for extracting geospatial information from the pages visited.

All in all, the paper is very well organized and makes an important contribution to this type of knowledge representation schema. It is very well written and is easy to read and follow.