Analogy as an Organizational Principle in the Construction of Large Knowledge-Bases

A capacity for analogy is an excellent acid test for the quality of a knowledge-base. A good knowledge-base should be balanced and coherent, so that its high-level generalities are systematically reflected in a variety of lower-level specializations. As such, we can expect a rich, well-structured knowledge-base to support a greater diversity of analogies than one that is imbalanced, disjoint or impoverished. We argue here that the converse is also true: when choosing from a large pool of candidate propositions, in which many propositions are invalid because they are extracted automatically from corpora or volunteered by untrained web-users, we should prefer those that are most likely to enhance the analogical productivity of the knowledge-base. We present a simple and efficient means of finding potential analogies within a large knowledge-base, using a corpus-constrained notion of pragmatic comparability rather than the typically less-constrained notion of semantic similarity. This allows us to empirically demonstrate, in the context of a substantial knowledge-base of simple generalizations automatically extracted from the Google n-grams, that knowledge acquisition proceeds at a significantly faster pace when candidate additions are prioritized according to their analogical potential.