Testing computational models of letter perception with item-level event-related potentials

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Visual Perception

Background:

• Understanding letter perception is crucial for cognitive science.
• Computational models aim to explain the mechanisms of visual recognition.

Purpose of the Study:

• To test computational models of letter perception.
• To investigate the neural processes underlying letter identification.

Main Methods:

• Recorded event-related potentials (ERPs) during letter and pseudoletter identification tasks.
• Computed item-level ERP measures to index the transition from feature analysis to letter identification.
• Fitted computational models, including interactive-activation models, with ERP data.

Main Results:

• Identified a key time window (100-200 ms) for the transition from feature analysis to letter identification.
• ERP measures supported models with specific connection architectures.
• Results favored a model incorporating feedforward excitation, lateral inhibition, and feedback excitation.

Conclusions:

• The findings support a detailed computational model of letter perception.
• The model includes interactions between feature and letter processing levels.
• This research advances our understanding of visual word recognition mechanisms.