Comparison of the Reception of Heterogeneous and Relatively Homogeneous Messages under Repetition in Noise

Abstract

Receivers listened to sets of 60 messages, words from the Thorndike-Lorge list. The words were of different syllabic length, of several mean frequencies of occurrence (20 each at frequencies of occurrence of 64, 128, and 256 per 4.5 million words). Messages were sent at three speech-to-noise ratios. For each word presented the receiver recorded a message and, immediately following, gave a confidence rating of his accuracy of reception. Each message was repeated until confirmed or until six presentations had occurred. The main purpose of the experiments was to test a simple stochastic model relating the probabilities of correct reception, correct and incorrect categorization of confidence ratings of repeated messages in noise. The data are in excellent agreement with predictions of the total number of message events required before the entire set of N messages is confirmed, and of the total numbers of correct and incorrect identifications confirmed. An earlier finding was that the model gives a good description of the process for homogeneous messages. The description is shown here to be nearly as good for the heterogeneous messages used in these experiments. In addition, a stimulus-sampling model is tentatively proposed to account for confusions among messages. [Supported by Research Committee of University of California and Office of Naval Research Contract Nonr 233 (58).]