ICCM: Processes

The parameters governing our behaviour are in constant flux. Accurately capturing these dynamics in cognitive models poses a challenge to modellers. Here, we demonstrate a mapping of ACT-R's declarative memory onto the linear ballistic accumulator, a mathematical model describing a competition between evidence accumulation processes. We show that this mapping provides a method for inferring individual ACT-R parameters without requiring the modeller to build and fit an entire ACT-R model. We conduct a parameter recovery study to confirm that the LBA can recover ACT-R parameters from simulated data. Then, as a proof of concept, we use the LBA to estimate ACT-R parameters from an empirical data set. The resulting parameter estimates provide a cognitively meaningful explanation for observed differences in behaviour over time and between individuals.

Self-report as a tool to understand different cognitive processing strategies has been criticised for decades, but to date there have not been many alternatives. To remedy this hiatus, we propose to apply a recently developed method for processing stage analysis (Hidden semi-Markov Model Multivariate Pattern Analysis, HsMM-MVPA) to a cognitive strategy prediction task. HsMM-MVPA uses specific patterns in EEG data to determine the most likely number of sequential processing stages. Under the assumption that cognitive processing strategies differ in the number of stages, we constructed a classifier using fitted HsMM-MVPA to try and differentiate between two cognitive strategies in unseen data. The method is applied to data from a multiplication verification task, in which participants are asked to verify the truth of a solution to a multiplication problem (3 x 9). We asked participants to indicate via self-report whether they knew the answer by heart (Strategy 1, Retrieval) or needed to compute the answer (Strategy 2, Procedural). The classifier could predict the self report labels above chance, suggesting that the number of processing stages identified using EEG can be used to track the cognitive processing strategy that are in use throughout a task.

Most cognitive models for human syllogistic reasoning aim to explain an average reasoner, i.e., the responses given by aggregating the response of the majority of reasoners. Studies show that individuals can deviate a lot from this average reasoner. So far, there have been very few models to explain and predict the responses of individual reasoner. In empirical studies, it can be observed that participants often rely on heuristic strategies (System 1 processes) to solve syllogistic problems but participants switch to analytical strategies (System 2 processes) occasionally. The study by Tse et al. (2014) demonstrated that inhibition of the matching heuristic is necessary to switch to the analytical processes in conflict problems that the output from the heuristic does not agree with that from analytical processes. This paper presents four mechanisms to incorporate individual differences in reasoning strategies and effect induced by problem type of the syllogism in predictive computational models built according to the mental model theory, mReasoner, and verbal models theory. Among these models, the composite model, which takes the highest accuracy model for individual reasoner, can reach a median accuracy of 86% in predicting the conclusions given by individual reasoner in the study

We describe a new approach for developing and validating cognitive process models. In our methodology, graphical models (specifically, hidden Markov models) are developed both from human empirical data on a task, as well as from synthetic data traces generated by a cognitive process model of human behavior on the task. Differences between the two graphical models can then be used to drive model refinement. We show that iteratively using this methodology can unveil substantive and nuanced imperfections of cognitive process models that can then be addressed to increase their fidelity to empirical data.