A Two-Tiered Approach for Organizing Slots in Large, Frame-Structured Knowledge Bases

Abstract

Slots represent semantic relations and play a major role in frame-based representation systems; they not only act as an instruction set for knowledge entry but also support most forms of nterferencing. Thus, the organization of slots merits a systematic study in its own righL A taxonomic approach formalizes the organization of the slots and provides a principled intelpretation for their semantics. We discuss the organization of slots from three different perspectives---relation-element, slot-use, and slot-argument-and propose that all three are useful in providing interpretations for slots. We argue that each individual view, by itself, does not offer sufficient semantics for slot organization. On the other hand, forcing all perspectives together destroys the clarity of a principled taxonomy. Therefore, we propose using two taxonomic views: A first taxonomy for slots is based on the relation-element and slot-use views; this taxonomy is domain-independent and promotes knowledge reuse. A second taxonomy, based on a slot-argument view, is domain- dependent and parallels the nonslot taxonomy.