The Automaticity of Visual Statistical Learning.

Nicholas B. Turk‐Browne; Junge, J. A.; Scholl, Brian J. · 2005 · OpenAlex-citations

DOI: 10.1037/0096-3445.134.4.552

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates the fundamental units of visual statistical learning (VSL), specifically determining whether VSL operates over individual visual features (e.g., shape or color independently) or over multidimensional objects where features are bound together. While previous research established that VSL occurs automatically and without awareness, it remained unclear how the visual system processes sequences involving multiple feature dimensions. The authors hypothesized that the nature of VSL depends on the statistical covariance between features during learning. The researchers conducted five experiments using a standard VSL paradigm where observers passively viewed sequences of colored shapes moving behind an occluder. The sequences contained hidden statistical regularities in the form of triplets (sequences of three items appearing in a fixed order). In Experiment 1, each shape was paired with a unique color, creating perfect covariance. Observers showed robust learning of these colored-shape triplets. Experiment 2 tested whether this learning was object-based by presenting test items as either monochromatic shapes or color patches alone. Performance dropped significantly, suggesting that both features were necessary for expression, implying object-based learning. Experiment 3 ruled out the possibility that Experiment 1’s success was merely due to having more statistical information at test by re-pairing shapes and colors differently; performance remained attenuated, confirming that VSL relies on the specific feature bindings formed during learning. Experiments 4a and 4b explored how varying the covariance between features affects the units of learning. During familiarization, some shape triplets had unique colors (matched), while others had randomly assigned colors (random). When tested with monochromatic shapes, observers showed robust VSL for both matched and random triplets, indicating a shift to feature-based learning when color was not diagnostic of shape identity. Similarly, Experiment 4b demonstrated robust VSL for color triplets when shape information was removed, provided the color-shape covariance was disrupted during learning. These results indicate that VSL is flexible: it operates over bound objects when features covary perfectly, but switches to feature-based processing when feature correlations are weak or absent. The findings suggest that VSL is not strictly limited to low-level features or high-level objects but adapts based on the statistical structure of the input. The authors conclude that sensitivity to feature correlations may help define what constitutes an object for the learning system. This implies that VSL plays a role in perceptual organization, potentially helping to segment the visual world into coherent objects based on statistical regularities. The study provides critical insight into the mechanisms of implicit learning, demonstrating that the "currency" of VSL is determined by the diagnosticity of feature dimensions in the environment.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-25; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• automaticity skill acquisition