Statistical Discrimination of the Synonymy/Antonymy Relationship Between Words

Abstract

A basic hypothesis is stated about the contextual and co-occurrence properties of synonymous words. On the basis of this hypothesis, several statistics are derived for use in discriminating between pairs of words which are synonymous and pairs of words which are nonsynonymous. The discriminating power of these statistics is tested on a corpus consisting of titles of physics theses. The tests indicate that two of the derived statistics have relatively high discriminating power. The results are interpreted and the possibility of obtaining better discriminating power is discussed.