Abstract
We examined the idea that perceptual learning is driven by the learning of diagnostic features, which are generalized from the training stimuli. A visual search task was used, which required participants to detect the presence or absence of a "v"-shaped motion trajectory (dot moving down to the right and then up to the right), among distractor trajectories (dots moving up to the right and then down to the right). Over the course of ten training sessions, participants improved dramatically on this task, indicative of learning. After the end of training, we examined which feature(s) of the complex trajectory stimulus were learned during training. The most obvious assumption would be that participants learned the precise features and feature relationships underlying the motion trajectories that were trained. However, both stimuli contain identical horizontal motion cues and differ only in the temporal sequence of the vertical motion (down then up vs. up then down). Learning would be more efficient if it prioritized the diagnostic feature itself (i.e., the vertical motion sequence) in order to distinguish target and distractors. To explore this aspect, trained participants performed a control experiment for which the horizontal motion in the trajectories was entirely removed: the target moved vertically down and then up while the distractors moved up and then down. Results show that participants transferred their learning benefits to this condition even though it was missing part of the original trained motion pattern (the common horizontal motion). However, when the vertical component of the training motion was reduced by making the angle of the "v" and inverted "v"s wider, transfer to the purely vertical tests deteriorated. Overall, our results suggest that perceptual learning involves the independent learning of diagnostic features. As long as these diagnostic features remain intact, learning will generalize to other stimuli that contain them.
Meeting abstract presented at VSS 2017