The Significance of Linguistic Generalizations

Abstract

This paper addresses the problem of defining the notion ‘linguistically significant generalization’; it separates the notion of ‘generalization’ from that of ‘significance’, and proposes a rigorous objective method for determining the significance of a generalization, based on probability theory. The proposal is illustrated by applying it to examples central to theorizing in generative grammar. Precisely formulated versions of claims made by Chomsky on the use of certain abbreviatory notations (e.g. parentheses) and by Halle on certain ‘natural classes’ are investigated and shown to be not significant; i.e., they tell us nothing interesting about the nature of language. Other generalizations—on active/passive selectional restrictions, word-order universal, and noun phrase accessibility—are shown to be highly significant.