Abstract
Purpose: When perceptual learning (PL) occurs in simple tasks, the learner is commonly unaware of the specific stimulus variables determining a classification. In contrast, the discovery of abstract relations in PL is sometimes thought to necessarily involve conscious awareness. We sought evidence for unconscious PL in tasks requiring the learner to discover either concrete attributes or an abstract relational basis for a classification. Method: Subjects viewed grid displays each consisting of 144 small squares in a 12 × 12 array. Each square in a grid had one of 4 gray levels. Participants judged whether each grid contained an embedded pattern, which was not described; learning could occur based on accuracy feedback. “Concrete” patterns were defined by 10 pixel locations having specific grayscale values. “Abstract” patterns were defined by pixel relations, not specific values. Experiment 1 tested concrete pattern discovery in three presentation and response conditions (presence/absence judgments for single instances, left/right judgments for paired comparisons, or “odd-one-out” triplets, in which one grid contained a target and two others did not). Experiment 2 tested discovery of abstract relations based on spatial or color relations. For patterns defined by spatial relations, pixel color was uniform and constant, but the pattern could appear anywhere in a grid. For patterns defined by color relations, position was constant and target color was uniform for each display, but color varied across displays. Speed of acquisition and conscious report of patterns were assessed. Results: Paired comparisons resulted in fastest learning. Learning occurred for both concrete and abstract embedded patterns even when participants could not describe the pattern. A pattern-drawing task designed to measure explicit knowledge found that most successful learners typically had little or no knowledge of target shape. Conclusion: Neither concrete nor abstract perceptual learning appear to require conscious awareness.