We can model scientific discovery as automated reasoning and learning, e.g., using a logic-based representation of knowledge, which we will here call an ontology. Much of what Kuhn calls 'normal science' may be modelled as problem solving within the shared ontology of a scientific community. However, to model what Kuhn calls a 'paradigm shift', we need mechanisms for changing this ontology. This is what w3c call ontology evolution. As we will see, ontology evolution can also happen during 'normal science'. Moreover, ontology evolution requires not just belief revision, but changes to the underlying signature of the ontology. For instance, functions might be split or merged; new arguments might be added to or removed from them.