Compound stress assignment by analogy: the constituent family bias

Abstract

Abstract This paper tests the hypothesis that stress assignment to English compounds works on the basis of analogy. In particular, the role of the constituent family, i.e. the set of compounds that share the same right or left constituent with a given compound, is investigated. On the basis of large amounts of data from three different corpora it is shown that the tendency towards a certain kind of stress pattern within the constituent families of a given compound is a strong predictor for stress assignment. This challenges rule-based approaches to compound stress assignment and lends independent evidence to exemplar-based approaches to language structure.