Abstract
Purpose. Perceptual learning (PL) in real-world domains leads to fluent extraction of abstract relational structures but often requires years of exposure. Recent research (e.g. Kellman, Massey, and Son, 2009), however, suggests that PL in real-world learning domains, such as mathematics, can be accelerated using appropriately designed computer-based learning technology. Such efforts differ from most PL research in two ways: They involve more complex task domains, such as mathematical structures, and they typically employ realistic, domain-focused assessments, as in high-stakes standardized math tests. That PL effects drive learning gains in these situations may be inferred but has seldom been tested directly. Here, we studied PL in a complex domain and examined transfer to a basic information extraction task. Method. We trained participants for PL of abstract relational patterns in Chinese characters. Different groups were trained to classify based on either (1) overall configurations (structures), (2) common feature relations (components), or (3) non-relational information (stroke count). All groups used a common set of stimuli. After PL, we tested for changes in information extraction using a visual search task (which had been pretested before the PL phase). Search displays contained novel exemplars, involved manipulations of target-distractor similarity using structures and components, and included both heterogeneous and homogeneous distractor displays. Results and Conclusions. We found robust changes in visual search specific to the type of perceptual classification training. Structure-based classification training markedly improved search efficiency when target and distractors shared the same structure, but only for heterogeneous distractors. For homogeneous distractors, component-based classification training produced most improvement in search efficiency. We also found improvement in search efficiency with exemplars of untrained relational categories. Results suggested that high-level PL produces changes in basic information extraction tasks, and that sensitivity induced by PL for both relational structure and specific components transfers to novel structural categories.
Supported by the U.S. Department of Education (IES) & National Institutes of Health.