Constructing Hierarchical Concepts via Analogical Generalization

Chen LiangNorthwestern University
Ken ForbusNorthwestern University

Abstract

Learning hierarchical concepts is a central problem in cognitive science. This paper explores the Nearest-Merge algorithm for creating hierarchical clusters that can handle both feature-based and relational information, building on the SAGE model of analogical generalization. We describe its results on three data sets, showing that it provides reasonable fits with human data and comparable results to Bayesian models.

Files

Constructing Hierarchical Concepts via Analogical Generalization (352 KB)



Back to Table of Contents