Human versus ideal performance on a three-interval forced-choice task with stimulus uncertainty along one and two dimensions

Abstract

Discrimination of sinusoids was measured using a three-interval forced-choice ‘‘sample-discrimination’’ paradigm in which mean values of target and nontarget stimuli remain fixed but individual stimuli are selected from Gaussian distributions. The task was designed as a psychoacoustic analog of an existing three-interval forced-choice test of phonetic contrast perception. Performance of five human subjects was measured as a function of d′ and compared with that of an ideal observer. In three experiments, stimuli differed in terms of frequency alone, intensity alone, and both. Performance on the unidimensional frequency and intensity discrimination tasks was very similar. It approximated the ideal for very high and very low values of d′ but fell below the ideal for intermediate values (6≳d′≳3). This finding suggests a shift from a post- to a preclassification decision strategy as stimulus uncertainty increases. When uncorrelated uncertainties of frequency and intensity were introduced in the two-dimensional task, performance of the ideal observer improved, as expected, but that of the human observers did not. This finding suggests a tendency to attend to only one dimension in this unfamiliar nonspeech task. [Work supported by NIH Grant No. 2PO1DC00178.]