The presented thesis provides a new approach for the integration of heterogeneous data in contexts, which (1) require data to remain in their original structural and semantic form in order to allow a dynamic ad-hoc integration based on individual use-cases and (2) need a particular focus on lexical and syntactical patterns, which are only bound to the instance-level of data and thus not covered by schema level integration. Based on preliminary work, the concept of a data transformation framework is developed on the contextual base of the formal interpretation of semi-structured schemata as regular tree grammars. As the content of the terminal nodes has been identified to be often not in an atomic form, the rule framework is applied on contained data in order to describe implicitly existing patterns and translate the data into semantically enriched forms.