Virtual ICCM I

Over the last century, a large variety of cognitive models for syllogistic reasoning have been developed, thereby advancing our understanding about the way humans process reasoning tasks. Most of the research was performed on a restricted set of quantifiers from first-order logic, which simplified model evaluations and comparison due to a well-defined set of tasks and the availability of complete and extensive datasets. However, as everyday reasoning and communication relies on a large variety of quantifiers, the scope and potentially also the generalizability of the models was severely limited. The present work aims at extending the domain of syllogistic reasoning to a wider set of quantifiers by (I) presenting a benchmarking dataset that includes the quantifiers ``Most'' and ``Most not'', (II) evaluating two state-of-the-art models (the Probability Heuristics Model and mReasoner) with respect to their ability to account for individual reasoners and (III) set the predictive performance of the cognitive models into perspective by comparing them to upper bounds and providing in-depth insights about their strengths and weaknesses.

We compare the qualitative predictions of an existing quantum model and a novel multinomial processing tree (MPT) model of the interference effect using parameter space partitioning (PSP). An interference effect occurs when categorizing a stimulus changes the marginal probability of a subsequent decision, leading to a violation of the law of total probability. The PSP analysis revealed that our MPT model can produce the same qualitative patterns as the quantum model. Further analysis, however, revealed that the models differ in several important ways. First, a larger volume of the MPT model's parameter space produces a smaller number of interference effects compared to the quantum model. Second, the distribution of volume across patterns is more diffuse for the MPT model, indicating it is more flexible than the quantum model. We discuss limitations and future directions.

Mind-wandering occurs as emotional arousal decreases, which is related to the level of mastery of the current task. As a worker becomes more proficient in a task, the cognitive resources required to perform the task decrease. Then, surplus resources emerge and are naturally directed to “default-mode thinking,” which people usually engage in outside the task. As mind-wandering continues, this default-mode thinking becomes more active and affects the task performance. In this study, we describe this process by combining the basic functions of the cognitive architecture Adaptive Control of Thought-Rational (ACT-R). The chunk activation mechanism represents the on- and off-task thinking loops. Furthermore, we introduce stochastic fluctuation in the chunk activation to change the transition probability between these loops. This fluctuation is assumed to be driven by parasympathetic activity, which increases over time and is suppressed by novel stimuli. To develop this physiological change, this study uses the ACT-R temporal module. Simulations using these modules demonstrate the inverse U-shaped relations between task performance and task continuation. Such a process is consistent with theories of optimal levels of arousal.

Motivation is the driving force that influences people’s behaviors and interacts with many cognitive functions. Computationally, motivation is represented as a cost-benefit analysis that weighs efforts and rewards in order to choose the optimal actions. Shenhav and colleagues (2013) proposed an elegant theory, the Expected Value of Control, which describes the relationship between cognitive efforts, costs, and rewards. In this paper, we propose a more fine-grained and detailed motivation framework that incorporates the principles of EVC into the ACT-R cognitive architecture. Specifically, motivation is represented as a specific slot in Goal buffer with a corresponding scalar value, M, that is translated into the reward value Rt that is delivered when the goal is reached. This implementation is tested in two models. The first model is a high-level model that reproduces the EVC predictions with abstract actions. The second model is an augmented version of an existing ACT-R model of the Simon task, in which the motivation mechanism is shown to permit optimal effort allocation and reproduce known phenomena. Finally, the broader implications of our mechanism are discussed.