Semantic representations in working memory: A computational model.

Area of Science:

• Cognitive Psychology
• Computational Neuroscience
• Psycholinguistics

Background:

• Verbal working memory relies on semantic knowledge.
• Semantic similarity effects in immediate serial recall are complex and poorly understood.
• Existing models struggle to explain these unique semantic effects.

Purpose of the Study:

• To propose a mechanistic explanation for semantic similarity effects in immediate serial recall.
• To extend connectionist models of working memory with semantic representations.
• To account for the distinct nature of semantic similarity effects compared to other similarity types.

Main Methods:

• Developed an extended connectionist architecture incorporating semantic representations.
• Introduced a tagging mechanism for selective, retrospective encoding of shared categorical features.
• Modeled immediate serial recall tasks to observe semantic similarity effects.

Main Results:

• The proposed model successfully accounts for the majority of observed semantic similarity effects.
• Demonstrated the role of selective encoding of shared semantic features.
• Highlighted differences in how working memory represents semantic versus phonological information.

Conclusions:

• Working memory represents semantic information more restrictively than phonological information.
• The proposed model provides a viable mechanistic explanation for semantic similarity effects.
• This work advances computational models of verbal working memory and semantic processing.