Modeling Response Time Variation in Addition Tasks: An ACT-R Extension with EVC-Guided Pause States

Reproducing variability in human response times is a central challenge in cognitive modeling. While conventional ACT-R models can replicate the general shape of response time (RT) distributions, they often fail to capture trial-by-trial fluctuations in cognitive processing time. In this study, we propose an extended ACT-R model that introduces a novel 'pause state,' a transient suspension of cognitive processing hypothesized to facilitate fatigue recovery. The initiation of this pause state is governed by the Expected Value of Control (EVC) theory, which selects the action with the highest expected utility based on trade-offs between cognitive rewards and costs. To validate the model, we designed a new simple addition task that enables fine-grained measurement of RT variability. The task allows for the decomposition of response times into two phases using gaze-tracking data: one corresponding to cognitive processing and the other to retrieval and motor response. Model parameters were optimized using a genetic algorithm, and model performance was evaluated by comparing simulated and empirical RT distributions. Results demonstrate that the proposed model accounts for trial-level RT variability, particularly in cognitive processing time, more accurately than a traditional ACT-R model. These findings highlight the importance of EVC-guided pause states in modeling subtle fluctuations in cognitive activity and contribute to a deeper understanding of the mechanisms underlying RT variability.