We present a transformation scheme that mediates between description logics (DL) or RDF-encoded ontologies and type hierarchies in feature logics (FL). The DL-to-FL direction is illustrated by an implemented offline procedure that maps ontologies with large, dynamically maintained instance data to named entity (NE) and information extraction (IE) resources encoded in typed feature structures. The FL-to-DL translation is exemplified by a (currently manual) translation of so-called MRS (Minimal Recursion Semantics) representations into OWL instances that are based on OWL classes, generated from the the type hierarchy of a deep linguistic grammar. The paper will identify parts of knowledge which can be translated from one formalism into the other without loosing information and parts which can only be approximated. The work described here is important for the Semantic Web to become a reality, since semantic annotations of natural language documents (DL) can be automatically generated by ...