Neural Components of Reading Revealed by Distributed and Symbolic Computational Models

Area of Science:

• Cognitive neuroscience
• Computational psychology
• Neuroimaging

Background:

• Understanding the neural basis of reading involves mapping cognitive components like orthographic, phonological, and semantic processing to brain activity.
• Two prominent computational models, Artificial Neural Network (ANN) and Dual-Route Cascaded (DRC), offer differing theoretical frameworks for reading processes and their neural instantiation.
• Previous research has not directly compared the neural plausibility of these distinct computational models.

Purpose of the Study:

• To compare the neural plausibility of ANN and DRC models of reading.
• To determine which computational model better accounts for brain activity patterns during reading aloud.
• To investigate the neural coding of reading through representational similarity analysis.

Main Methods:

• Representational Similarity Analysis (RSA) was employed to compare model-generated representations with neural data.
• Participants engaged in reading aloud tasks while their brain activity was recorded.
• Statistical analyses controlled for the contributions of each model to assess unique variance explained.

Main Results:

• Both ANN and DRC models showed some correspondence with neural activity during reading.
• ANN model representations correlated more strongly with relevant cortical areas associated with reading.
• After controlling for DRC model contributions, ANN models significantly explained neural data variance, whereas DRC models did not.

Conclusions:

• Artificial neural networks utilizing distributed representations offer a more accurate account of neural coding for reading compared to the DRC model.
• This study validates a framework for comparing computational cognitive models against neural data to understand brain function.
• Findings support the utility of non-symbolic, distributed representations in explaining the neural basis of reading.