Predicting stress in English verbs – analogy, morphology, and the level of abstraction of phonological patterns

Abstract

Stress position in English words is well-known to be influenced by phonological, morphological, and (morpho)syntactic factors (esp. verbs vs. nouns). All these influences, however, are far from categorical, and very often stress varies from one word to another in a seemingly unpredictable manner. What is more, recent corpus-based work points to the relevance of opaque, i.e. etymological prefixes. It is, however, unclear why effects of opaque morphology should be productive synchronically. Moreover, there is an increasing body of research indicating that stress assignment works on the basis of lower-level information, i.e. analogy. In this paper we analyze the results of a pseudo-word experiment testing the production of trisyllabic verbs. Structural predictors tested in the analysis comprise syllable weight and prefixation (measuring the transparency of prefixes). To test analogy, we had a computational analogical model predict stress in the pseudo-words based on all verbs in the Cambridge Pronouncing Dictionary . Lexical support measures were then used as predictors alongside weight and prefixation in a mixed-effects logistic regression model. The results show that all three variables are significant predictors of main stress in the experimental data. Lexical support is strongly correlated with actual productions, but AML underpredicts some of participants’ preferences. These are, however, explained by syllable weight and transparency of the prefix in our regression model. We conclude that properties on different levels of representation are relevant for the prediction of stress position in English pseudo-verbs.