Decision Processes in Speech Discrimination as Revealed by Confidence Ratings

Abstract

Standard speech discrimination tests require the listener to make a decision about a given sequence of stimuli. For example, with the ABX test the listener is required to determine whether the third stimulus is most like the first or most like the second. This test along with several other forced-choice procedures has the excellent property that the E need not specify the dimension along which the stimuli differ, e.g., select the stimulus with the higher pitch. The typical measure of the listener's performance is the percentage of stimuli correctly discriminated. However, in recent attempts to examine the processes underlying the identification and discrimination of speech sounds, it has become apparent that the listener's task, i.e., providing just a single response, may not accurately reflect all the information that the listener may have available to him about the stimuli. For example, sometimes a listener may be quite certain that his response was correct whereas other times the listener may be very uncertain about his response. In this study we examined how listeners assign confidence ratings to discrimination judgments for a set of synthetic stop consonants and a set of steady-state vowels. The confidence ratings obtained with both ABX and 4IAX discrimination procedures carry additional information about the stimulus properties of consonants and vowels and provide some insight into the decision processes employed in the discrimination of these two classes of speech sounds.