Abstract
Perceptual learning is thought to be specific for the features relevant to a task, given that several studies have shown a failure of any learning benefit to transfer from one task to another (Shiu & Pashler, 1992; Ahissar & Hochstein, 1993). Thus it has been suggested that attention driven by task demands successfully selects an early population of units which code for a relevant feature. Here we show a counter-example. In training, the subjects (n=6) were instructed to report a white letter presented in a sequence of otherwise black letters at the center of the display (RSVP task, e.g., Raymond, 1992; Joseph & Nakayama, 1998). The background of the display consisted of 5% signal dots which moved coherently in the same direction and 95% noisy dots which moved in random directions. The direction of the coherent motion was constant throughout training. In the test stages preceding and following training, we tested both the detectability of the coherent motion with the same display at the same S/N ratio as in training and the discriminability of 7 motion directions in the display that consisted only of coherently moving signal dots. The performance in the detectability task was at chance level both before and after the training, indicating that in training coherent motion was not perceived in the background. Nevertheless, the performance of the discrimination task for the same direction as presented during training was significantly higher than for the other directions which were not presented during training. These results indicate that mere repetitive exposure to a motion direction can modify the mechanism specific for that direction even if it is not perceivable. This suggests that perceptual learning occurs passively, without any conscious control.
Supported by NSF (BCS-9905194) to TW.