Abstract
In a series of studies (reported at VSS '03–'05), we have shown that perceptual learning for contrast detection may be more complex and/or less dramatic than is assumed. Specifically, we have shown that reductions in detection threshold are regularly accompanied by liberal shifts in response bias. These shifts are due to false alarm rates that either do not change or that increase as a function of practice. To date, all of this evidence has been obtained in tasks that require explicit presence/absence responses, responses that are typically not required in most studies of visual perceptual learning. This leaves open the possibility that the effects in bias and false alarm rates might be due to the act of repeatedly giving presence/absence responses. We tested this hypothesis in experiments in which observers' performance was assessed in two ways across 12 days of practice. On days 1 and 12, thresholds and false alarm rates were obtained using the method of constant stimuli. On days 2–11, thresholds were estimated using a variety of two-interval two-alternative forced-choice staircase procedures. Thus, on days 2–11, no presence/absence decisions were made. Results showed reliable decreases in threshold, reliable increases in false alarm rates, and an absence of bias for choice of interval. This suggests that even in the absence of making presence/absence decisions, observers experience a shift in decisional criteria that is specific to sensory detection. Implications for what is being learned are considered, with emphasis on the need for a distinction between sensory detection and perceptual labeling.