Hierarchical conceptual spaces for concept combination

In natural language, unlike formal logic, concepts are inherently flexible and graded. They give different degrees of importance to defining features, allow for different degrees of typicality, and permit a rich array of combination rules. Applications such as natural language processing and semantic search can greatly benefit from a richer representation of natural concepts to capture this flexibility. In this paper we introduce a new hierarchical representation of concepts inspired by Gardenfors’ work on conceptual spaces, which incorporates a probability model of prototypes to account for vague concept boundaries, semantic uncertainty and feature weights within concept ontologies.