Intelligibility of Connected Discourse

Abstract

An approach to measuring the intelligibility of free-running connected discourse that contrasts with the quasi-statistical methods generally used with nonsense syllables, words, or sentences is described. With this approach, intelligibility is viewed as a decision by the listener that specifies how well the message was understood. The goal of testing, therefore, is to quantify the listener’s response with meaningful indices of performance. The various speech-Bekesy procedures are sensitive to a listener decision, but they fail to specify how the criterion of intelligibility was interpreted by the listener and have the additional disadvantage of specifying intelligibility as a threshold level in decibels rather than as a magnitude of intelligibility in percent. Experiments are reported in which systematic modifications were made to the speech-Bekesy procedures. The two major methods are “tracking to constant percentage criteria of intelligibility” and “estimation of intelligibility.” The results specify intelligibility of connected discourse by a performance-intensity function. The data are stable across alternative testing methods and compare favorably with sentence-repetition scores.