Abstract
Perceptual learning is often studied using Same/Different and 2AFC tasks and the stimulus-specific improvement is attributed to perceptual factors. However, two-interval tasks involve visual memory and decision strategies that can change in stimulus-specific ways too. To test whether the improvement is due to perceptual or non-perceptual factors, we compared single-interval (Yes/No) and two-interval (Same/Different) tasks in two within-subject studies of visual motion discrimination. Method: Stimuli were random-dot cinematograms (RDC, 100% coherence, speed 10 deg/sec, duration 400 msec). Each block involved two motion directions 5° apart (e.g., -52.5° vs -47.5°). The implicit reference direction (e.g., -50°) and the task were fixed in each block. In single-interval trials, observers indicated which direction was presented; In two-interval trials, observers indicated whether two sequentially presented RDCs moved in same or different directions. Each session began with single-interval blocks and switched halfway to Same/Different blocks. Auditory feedback throughout. Experiment 1 pretested one reference direction, trained an orthogonal direction for four days, and post-tested the original direction. 13 naive observers. Experiment 2 was similar, but two reference directions (e.g., -50° and -37.5°) were trained in alternating, homogenous, 60-trial blocks. 15 more observers. Results: Both tasks improved with practice: d′ increased by approximately two thirds along identical learning curves. Small but reliable switch costs between tasks. The stimulus specificity index (SI) varied between 40% and 60% across conditions in the group average, with large individual differences. No evidence of stimulus-specific improvement of visual STM or of changes in Same/Different decision strategy. The bootstrap standard deviation of the SI estimate was two times higher in the two-interval than the single-interval data. No evidence that interleaved training improved transfer. Conclusion: Learning of motion direction discrimination is driven by perceptual factors. Specificity indices are hard to measure because of individual variability. The Yes/No task affords higher statistical power than Same/Different.