The semantic Web envisions an Internet where data can be used by applications just as easily as it can be by humans. Ontologies are a key building block of the semantic Web, and can enable applications to have a shared understanding of data, and to be semantically interoperable. However, ontologies from heterogeneous sources may use different terms to represent the same concept. This can present problems for applications which are required to work interoperably with independently produced ontologies. Manually determining semantically equivalent terms between ontologies is a laborious and error-prone process. In this paper, we present an architecture which uses machine-learning techniques to produce mappings between semantically equivalent terms in heterogeneous ontologies. The architecture combines the results of several machine-learning algorithms, in order to be effective with a wider range of data than if a single algorithm were used.