A neural network model of semantic memory linking feature-based object representation and words.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Neuroscience

Background:

• Semantic memory theories propose distributed cortical processing and multimodal object representation.
• Previous models focused on object representation but lacked integration with semantic memory.

Purpose of the Study:

• To extend a computational model for object representation to create a semantic memory system.
• To link sensory-motor object features with lexical representations (words).

Main Methods:

• A model representing objects as feature collections in topological cortical areas was developed.
• Gamma-band synchronization segmented features; Hebbian learning connected feature and lexical areas.
• Simulations involved presenting objects with corresponding words for training.

Main Results:

• The network successfully associated objects with words using multimodal inputs (sensory-motor and acoustic).
• Object segmentation and association were achieved even with incomplete information.
• The model demonstrated semantic links between words representing objects with shared features.

Conclusions:

• Semantic memory is an integrated process retrieving content via co-activation of multimodal regions.
• The model supports theories of distributed semantic memory and multimodal integration.
• Future versions can test conceptual theories and analyze neural deficit data.