Confidence ratings in discrimination of prosodic features of unintelligible sentences

Abstract

Three types of sentence materials were presented to test perception of prosodic features. The sentences differed in grammatical phrase structure, major constituent boundary (MCB), or contrastive stress. Subjects made forced-choice responses to the prosodic features and assigned a confidence rating estimating of the probability of being correct. Sentences were presented under three experimental conditions: audition alone (AUD), vision alone (VIS), and audiovisual (AV). The acoustic signal was low-pass noise, amplitude modulated by the sentence materials. Irrespective of correctness, highest ratings were assigned to responses under the AV condition, next highest to VIS, and lowest to AUD conditions. The relationship of accuracy to rating assignment varied with sentence material and experimental condition. For the test of grammatical phrase structure (5 AFC) low ratings were assigned to correct responses under AUD conditions, suggesting that these were correct guesses. In locating the MCB (16 AFC), rating assigned to correct responses were not greatly different from ratings assigned to error responses. If subjects use ratings as instructed, the probability of obtaining a correct response would correspond to the ranges of certainty for each rating category, independent of experimental condition. Under AUD and VIS conditions, subjects over- or underestimated the probability of being correct. Under AV conditions, subjects were most successful in estimating accuracy with confidence ratings. [Work done while author was at University of Virginia.]