The reliability of ratings by linguistically untrained subjects in response to stress in speech

Abstract

To determine the reliability with which untrained raters could identify stress in the speech of a single person, two forms of the same material, (1) speech broken into short utterances and (2) speech in its conversational context, were presented to 40 linguistically naive psychology students who were asked to underline those syllables that they perceived as stressed. High reliabilities were obtained from both interrater measures (r=0.96 for each treatment) and a test-retest estimate (r=0.88). However, significantly larger total stress scores were recorded under the short utterance presentation than under the context condition. It was suggested that this result occurred because each of the few syllables in short utterances received greater attention than did the stream of syllables in context. Subsequent regression analysis led to the prediction that, for a short passage to attain a mean score equal to that which it would receive if rated in context, it should contain approximately 40 syllables.