The problem of data semantics is establishing and maintaining the correspondence between a data source and its intended subject matter. We review the long history of the problem in Databases, and contrast it with recent research on the Semantic Web. We then propose two new directions for research on the problem and sketch some open research questions.