Distributed Code for Semantic Relations Predicts Neural Similarity during Analogical Reasoning.

Area of Science:

• Cognitive Neuroscience
• Computational Linguistics
• Neuroimaging

Background:

• Semantic relation processing is crucial for human cognition.
• Debate exists on whether semantic relations are coarsely coded (network links) or finely coded (distributed patterns).
• Empirical evidence for conceptual and neural representations of semantic relations is lacking.

Purpose of the Study:

• To empirically investigate the neural coding of semantic relations.
• To compare computational models of relational dissimilarity against brain activity.
• To adjudicate between network-based and distributed representation accounts.

Main Methods:

• Used sequential presentation of verbal analogies to elicit analogy judgments.
• Compared measured neural activities with predictions from computational models.
• Focused on neural activity within a frontoparietal network.

Main Results:

• A frontoparietal network was identified as encoding three types of semantic relations.
• A computational model with distributed representations over abstract relations predicted neural activity.
• This model specifically predicted activity in the left superior parietal cortex for individual relations and a broader left-lateralized network for second-order comparisons.

Conclusions:

• Findings support a model where semantic relations are finely coded as distributed patterns over abstract relations.
• Neural evidence implicates the left superior parietal cortex and broader left-lateralized networks in semantic relation processing and comparison.
• This research advances understanding of the neural basis of analogical reasoning and semantic representation.