WordNet Sits the S.A.T. - A Knowledge-Based Approach to Lexical Analogy

One can measure the extent to which a knowledge-base enables intelligent or creative behavior by determining how useful such a knowledge-base is to the solution of standard psychometric or scholastic tests. In this paper we consider the utility of WordNet, a comprehensive lexical knowledge-base of English word meanings, to the solution of S.A.T. analogies. We propose that these analogies test a student's ability to recognize and estimate a measure of pairwise analogical similarity, and describe an algorithmic formulation of this measure using the structure of WordNet. We report that the knowledge-based approach yields a precision at least equal to that of statistical machine-learning approaches.