In this paper, we will consider the alignment of heterogeneous ontologies in multi agent systems. We will start from the idea that each individual agent is specialized in solving a particular task and therefore requires its own specialized ontology that is, in principle, not understandable for other agents. This heterogeneity of ontologies, of course, poses problems for the communication between agents. In our framework, we assume that the agents share some minimal common ground which can be used to learn new concepts. We will discuss which concepts of the different ontologies the agents should learn from each other in order to establish a communication vocabulary that enables optimal communication.