Beyond Letters: Optimal Transport as a Model for Sub-Letter Orthographic Processing

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Psychology

Background:

• Letter processing is crucial for visual word recognition.
• Current models often simplify or ignore early perceptual letter recognition processes.

Purpose of the Study:

• To propose optimal transport theory as a computational framework for letter shape processing.
• To investigate the alignment of optimal transport cost with neural activity in early visual processing stages.

Main Methods:

• Representational similarity analysis was employed.
• Optimal transport cost (Wasserstein distance) was calculated between letter pairs.
• Neural activity elicited by visual letter stimuli was recorded and analyzed.

Main Results:

• Optimal transport cost aligns with neural activity within 225 ms of stimulus onset.
• This approach outperforms models based on simple shape overlap.
• Optimal transport successfully captures emergent geometric invariances in letter perception.
• Wasserstein distance predicts neural activity comparably to features from trained artificial neural networks.

Conclusions:

• Optimal transport theory provides a computationally explicit framework for modeling early orthographic processes.
• This theory offers a more constrained and interpretable model of letter perception compared to unconstrained artificial network representations.