Current knowledge systems lack an effective architecture to bridge the well-acknowledged gap between flexibility and rigidity - users’ preference of flexible expression and machine’s need for rigorous representation. This paper focuses on finding an adequate set of relationship types (or link types) and map flexible expressions from users to link types in propositional knowledge systems, in which knowledge elements are stored as “concept-relationship-concept” triplets. We call our approach “hybrid link architecture”. This architecture has two levels of presentations, an open link name layer and a closed link type layer. Between these two layers is a matching mechanism based on a psycholinguistic thesaurus to map names to types. The link types are derived by synthesizing literature of knowledge organization, semantic network, and educational taxonomies. A two-stage evaluation is conducted based on 39,706 triplets. The evaluation shows that the matching mechanism is accurate and the proposed link types are mutually exclusive and cover most link names in our dataset.