Integrating Language Model Embeddings into the ACT-R Cognitive Modeling Framework

We use the ACT-R cognitive modeling framework to simulate the associative priming effect in lexical decision tasks, where prior exposure to a prime word influences response times to a target word, depending on the degree of semantic association between them. Currently, ACT-R relies on manually coded similarity scores, which are labor-intensive and prone to inaccuracies, especially for large vocabularies. In this study, we replace these hand-coded relationships with language model embedding vectors, including Word2Vec, along with state-of-the-art large language models such as BERT and Gemma, to enhance both the accuracy and scalability of the model. Our results demonstrate a strong correlation between ACT-R predictions and observed human response times from a dataset. We see this as a positive proof of concept for how embedding vectors can be integrated within algorithmic-level models of language processing.