Remembering words in context as predicted by an associative read-out model

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Psycholinguistics

Background:

• Interactive activation models (IAMs) simulate word processing but lack associative and explicit memory components.
• Existing models do not fully capture the associative nature of word memory.

Purpose of the Study:

• Introduce the associative read-out model (AROM) to extend IAMs.
• Incorporate long-term word associations and explicit memory performance.
• Explain recognition memory phenomena using computational modeling.

Main Methods:

• Extended IAM with an associative layer based on Hebbian learning and corpus co-occurrence.
• Simulated a study-test task with varying levels of word association.
• Modeled test-phase performance using initialized activation for learned vs. non-learned items.

Main Results:

• Increased association rates enhanced "yes" responses for both learned and non-learned words.
• Model predicts lower slopes in z-transformed receiver-operating characteristics (z-ROCs) for recognition memory.
• AROM accurately predicted item-level recognition probabilities.

Conclusions:

• AROM successfully integrates form-based processing with semantic word relations.
• The model provides a unified account for implicit and explicit memory phenomena.
• AROM offers a more comprehensive simulation of word memory and recognition.