I started with RDF dumps of the concepts and tasks, which are available from the SPARQL endpoint page. Assertions about relations in the database are not currently available from the endpoint, so I got a custom dump of those as well. I then wrote a python script to finesse all of this into an OWL ontology. I have to say that this is a great way to learn about the structure of OWL! I used Protégé to help validate my OWL and examine the contents, which was really useful.
Part of the work in setting up this ontology was deciding how to align it with other ontologies, particularly with CogPO. I had a long meeting at OHBM2011 with Jessica Turner and Angie Laird, during which we hashed out a set of classes and relations that will serve to structure the task representation in both of our ontologies. For our representation of tasks, we are inheriting the BehavioralExperimentalParadigm class from CogPO.
I have also registered the Cognitive Atlas with the NCBO BioPortal site, which I hope will give it increased visibility. In the long run, I plan to create a pipeline by which the ontology will the automatically generated and updated directly from the database, but for now will do regular manual updates.