MEG Evidence That Modality-Independent Conceptual Representations Contain Semantic and Visual Features

Area of Science:

• Cognitive Neuroscience
• Neuroimaging

Background:

• Semantic knowledge can be accessed via various sensory inputs (e.g., images, words).
• Previous research suggests modality-independent brain representations exist, but their content is not fully understood.
• These representations may encompass semantic, perceptual, or orthographic features.

Purpose of the Study:

• To investigate the content of modality-independent conceptual representations in the human brain.
• To determine if these representations are purely semantic or also include perceptual features.

Main Methods:

• Utilized magnetoencephalography (MEG) on 25 human participants.
• Developed a novel cross-condition decoding approach using neural network classifiers.
• Compared learned latent representations with semantic, sensory, and orthographic feature models.

Main Results:

• Modality-independent representations showed significant correlations with both semantic and visual features.
• No evidence was found for correlations due to picture-specific visual features.
• Orthographic features did not drive the observed correlations.

Conclusions:

• Modality-independent conceptual representations in the brain integrate both semantic and perceptual information.
• These findings advance our understanding of how the brain stores and accesses knowledge across different sensory modalities.