Abstract
Perceptual learning, a form of adult brain plasticity that produces improved discrimination, has been studied in various tasks and senses. However, it is unknown whether and how this improved discrimination alters stimulus appearance. Here, in addition to a discrimination task, we used an estimation task to investigate how training affects stimulus appearance in human adults. Before and after training, observers were shown stimuli composed of dots moving slightly clockwise or counterclockwise horizontal, whose appearance has been shown to be biased away from horizontal. Observers were subdivided into three groups: Those who (1) trained in a discrimination task; (2) trained in an estimation task; (3) did not train. Training improved discrimination accuracy and decreased coherence thresholds. Counterintuitively, training also distorted appearance, substantially exacerbating estimation biases. These changes occurred in both training groups (but not in the notraining control group), suggesting a common learning mechanism. We developed a computational observer model that simulates performance on both discrimination and estimation tasks. The model incorporates three components: (1) the internal representation favors cardinal motion directions, which are most common in the natural environment; (2) in the estimation task, observers implicitly categorize motion, conditioning their estimates on this; and (3) both types of training induce an increase in the precision of representation of trained motions. We find that the simulations of the model, fit to individual observer data, can account for their improved discrimination and increased estimation bias. We conclude that perceptual learning improves discrimination while simultaneously distorting appearance.
Significance Perceptual learning refers to the increase in sensitivity to small stimulus differences (for example, distinguishing two similar perfumes or shades of blue) that arises from training. Another important perceptual dimension is appearance, the subjective sense of stimulus magnitude. It seems intuitive that training-induced improvements in discrimination would be accompanied by more accurate appearance. We used both discrimination and estimation tasks to test this hypothesis, and find that that while training improves discrimination ability, it leads to increases in appearance distortion. To explain this counterintuitive finding, we propose a model of how distortions of appearance can arise from increased precision of neural representations and serve to enhance distinctions between perceptual categories, a potentially important property of real-world perceptual learning.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
↵* Joint first authorship
