Abstract
Psychophysical performance improves with practice but the improvement is often specific to the trained stimulus configuration. The improvement on easy tasks often transfers more than that on difficult tasks (Ahissar & Hochstein, 1997; Liu, 1999). These results were first established using experimental designs that confounded the difficulty during training and subsequent generalization tests. When these two factors were manipulated separately between subjects in an orientation discrimination task (Jeter et al., in press), the observed thresholds challenged the widespread notion that the training regimen determines the locus of learning. We replicate this surprising finding with d‘ measures of motion discrimination within subjects.
Method: The stimuli were moving filtered-noise textures (speed 10 deg/sec, duration 400 msec, bandpass filter centered on 3 cycles/deg). Each participant was pre-tested on two easy motion-direction discriminations (-59 vs -51; 31 vs 39) and two difficult ones (-57 vs -53; 33 vs 37 deg). The 20 participants then practiced one particular task for four sessions, counterbalancing between subjects. All four tasks were post-tested on Day 6.
Results: The group-averaged d' improved from 0.97 to 1.58 for the difficult task and from 1.78 to 3.12 for the easy task. The improvement was partially specific to the practiced direction: Post-test d' at the orthogonal direction was 1.22 for the difficult and 2.15 for the easy task. Importantly, the post-test performance in the easy-training group was identical to that in the difficult-training group.
Conclusion: Easy generalization tests yielded higher specificity indices (SI=70%) than did difficult tests (SI=40%), contradicting the Reverse Hierarchy Theory (Ahissar & Hochstein, 1997). Moreover, in agreement with Jeter et al. and contrary to the traditional interpretation, these indices did not depend on the training condition. Within a range of difficulty levels, practicing a given task seems to induce similar internal changes that produce different outcomes on different tests.