An axiomatic approach to the notion of semantic antagonism

Abstract

The concept of semantic antagonism refers to the human capability of characterizing those objects of an universe of discourse being dissimilar, significantly different from or opposite to a given concept, predicate or previous knowledge. This capability is essential in the formation of linguistic polarities, such as false/true or good/bad, that enable us to analyze, organize (classify) and give meaning to reality in terms of opposite poles of semantic reference. Though they are related, the notion of semantic antagonism is somehow more general than that of antonymy, since the former allows to characterize opposition even in the absence of antonym words and is not constrained by the assumption of symmetry that underlies the last. Therefore, the notion of semantic antagonism seems to be well suited for giving base to those knowledge representation frameworks which introduce some kind of bipolarity or distinction between positive and negative information. In this paper, an axiomatic approach is taken in order to describe the reasonable assumptions a dissimilarity operator acting on a set of predicates should obey. This enables to derive a basic differentiation of these operators, and particularly to show that antonyms are special cases of antagonistic predicates. Furthermore, through the proposed axioms it is possible to introduce the idea of dissimilarity structure over a set of predicates, and the application of this last notion in the context of supervised learning for classification tasks is briefly described.