Unimodal statistical learning produces multimodal object-like representations

Area of Science:

• Cognitive Science
• Neuroscience
• Perception

Background:

• Object concept formation is key to cognition.
• Traditional views emphasize visual or haptic boundaries for object segmentation.
• The emergence of abstract object concepts remains poorly understood.

Purpose of the Study:

• To test if statistical properties, rather than boundaries, are fundamental to object concept formation.
• To investigate cross-modal generalization of object learning.
• To explore the role of statistical learning in scene segmentation.

Main Methods:

• Developed a novel visuo-haptic statistical learning paradigm.
• Familiarized participants with objects defined by across-scene statistical properties.
• Assessed learning via visual familiarity and haptic pulling tasks.

Main Results:

• Demonstrated strong within-modality learning (visual and haptic).
• Showcased significant 'zero-shot' cross-modality generalization.
• Found a high correlation between within-modality learning and cross-modality generalization.

Conclusions:

• Boundaries are not essential for object segmentation; statistical properties are more fundamental.
• Humans can segment scenes into objects using only statistical information.
• Statistical learning underpins cross-modal object concept generalization.