Semantic integration by pattern priming: experiment and cortical network model

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Computational Linguistics

Background:

• Semantic priming effects are often modeled using neural networks with Hebbian learning, focusing on pairwise word associations.
• Classical Hebbian learning struggles to account for the frequent occurrence and processing of multi-word patterns (e.g., "by the way") found in large text databases.

Purpose of the Study:

• To investigate the impact of three-word patterns on semantic priming.
• To evaluate a novel inter-synaptic learning algorithm's ability to reproduce experimental findings on pattern learning.

Main Methods:

• An experiment was conducted manipulating the frequency of three-word patterns.
• A biologically inspired inter-synaptic learning algorithm was developed and tested.
• Simulations were performed to assess the algorithm's capacity for learning three-word patterns.

Main Results:

• Target words received significantly more priming when presented within a three-word pattern compared to pairwise associations.
• The inter-synaptic learning algorithm successfully potentiated synapses based on the activation of multiple neurons.
• Simulations demonstrated the network's ability to learn and replicate the observed effects of word patterns.

Conclusions:

• Multi-word patterns play a crucial role in semantic processing and priming, challenging traditional pairwise association models.
• Biologically inspired inter-synaptic learning offers a promising mechanism for neural networks to learn complex word patterns.
• This research provides a computational model for understanding how the brain processes and learns linguistic structures beyond simple word pairs.