The ability to perform automated conversions between different data formats is key to achieve interoperability between heterogeneous systems. Conversions require the definition of mappings between concepts of separate data specifications, which is typically a difficult and time-consuming task.In this article, we present a technique that exploits, in part, semantic web technologies, to automatically suggest mappings to users, based on both linguistic and structural similarities between terms of different data specifications. In addition, we show how a machine-learned linguistic model created by gathering data from domain-specific sources can help increase the accuracy of the suggested mappings. The approach has been implemented in our prototype tool, SMART (SPRINT Mapping & Annotation Recommendation Tool), and it has been validated through tests carried out using specifications from the transportation domain.