Similarity, Comparability and Analogy in WordNet: Squaring the Analogical Circle with Mondrian

Similarity and comparability are complementary notions that are easy to confuse and difficult to tease apart. Semantic similarity makes comparison meaningful, while comparison is often the pragmatic means through which similarity is perceived and constructed. To date, WordNet has found wide application as a basis for modeling and measuring semantic similarity, but is lacking as a basis for wellformed comparability judgments. In this paper we describe a corpus-based approach to learning a model of sensible comparability, and show how this pragmatic model can be integrated with a WordNet-based notion of semantic similarity. We go on to show how this model of comparability, called Mondrian, provides a convenient and efficient means of supporting simple relational analogies between WordNet terms, and how analogies of this kind can be used for the automatic acquisition of enriching relational knowledge in WordNet.