Executive Functions & Cognitive Control

Humans can satisfy a large variety of abstract goals, such as driving to work or doing the weekly grocery shopping. This requires the maintenance of contextual information to guide neural information processing through top-down mechanisms (Miller & Cohen, 2001). Functions subserving this ability are collectively referred to as “cognitive control”. Experimental investigations of cognitive control often present multi-dimensional stimuli with potentially conflicting information (e.g. the Stroop task or the Flanker task) or require switching between multiple tasks with different response demands for the same stimuli (e.g. task-switching studies). Previous work has found that allocation of cognitive control in these tasks is dynamic, adjusting to recent history (e.g. Gratton et al, 1992), but also long-term expectations (e.g. Bugg, 2014; Siqi-Liu & Egner, 2020) and item-specific expectations (e.g. Bugg et al., 2011; Chiu & Egner, 2017). In three new experiments, we show how “temporal context” (Howard & Kahana, 2002) can be seen as a primary driver of cognitive control allocation and develop a novel computational model that can explain these and several previous effects (Gonthier et al., 2016; Dignath & Kiesel, 2021) that earlier models (Botvinick et al., 2001; Verguts & Notebaert, 2008; Lieder et al., 2018) cannot account for.

This is an in-person presentation on July 21, 2024 (15:20 ~ 15:40 CEST).

The ability to adapt verbal and behavioral output to current goals depends on various cognitive control mechanisms. The distinction among different control mechanisms often relies on their differential effects on response latency, which can be formalized using evidence accumulation models. Indeed, previous research suggests that cognitive control can modulate different components of evidence accumulation, including the rate of evidence accumulation and the amount of evidence required to reach a threshold. However, our subjective experience of exerting control often involves the experience of withholding the expression of a response that has already won this internal race. Such post-accumulation inhibition is especially relevant for behaviors that depend on memory retrieval, as in the case of free recall, semantic fluency, and the free association task. Thus, whereas the dynamics of memory retrieval can be formalized as a race among competing memories, the verbal expression of a retrieved memory could be inhibited if it turns out inconsistent with goals or task instructions (e.g., a word that has already been reported when such repetitions are prohibited). I will present a recently developed tractable model formalizing post-accumulation inhibition (semi-Markov process model). I will then demonstrate evidence for the involvement of post-accumulation inhibition in a free association task, highlighting its specificity to cases where the control criterion does not depend on evidence strength (i.e., when repeated associations are prohibited but not when weak associations are prohibited). I will conclude by discussing the implications of ignoring post-accumulation inhibition in other tasks involving cognitive control demands.

This is an in-person presentation on July 21, 2024 (15:40 ~ 16:00 CEST).

We examined whether selective attention, which is mainly theorized as the ability to focus on the category-relevant dimension, is a sole construct in understanding category learning. As the attention literature dissociates selective attention into focusing and filtering, we argue that filtering is another component that should be considered to fully understanding category learning. In the study, we provide an experimental paradigm that can dissociate filtering from focusing. By utilizing the paradigm along with collecting individual attention control measures, we show that filtering is related to the ability to inhibit irrelevant information. We also present that computational models (e.g., ALCOVE) that incorporate selective attention only as an ability to focus can not explain the results from the current study.

This is an in-person presentation on July 21, 2024 (16:00 ~ 16:20 CEST).

We report the development of evidence-accumulation models (EAMs) for a large data set from 128 participants performing tasks from the Dual Modes of Cognitive Control (DMCC) battery. Each participant performed two replications of 15 approximately half-hour sessions over two three-week cycles (Tang et al., 2023). DMCC is designed to probe individual differences in two modes of controlling choices, anticipatory (“proactive”) and choice-stimulus-driven (“reactive”). The battery consists of four tasks, two requiring binary manual responses, Task Switching (consonant/vowel or odd/even choices) and Sternberg (positive vs. negative for presence on an immediately preceding study list). Another, the AX-CPT task, adds a no-go choice to a binary manual response task, and the fourth, a Stroop task, requires an 8-choice vocal colour-naming response. Each task has a baseline form and two variants incorporating manipulations encouraging proactive and reactive processing. Although EAMs with a race architecture can handle the diverse response types in a single framework, estimation challenges arise in providing a comprehensive account of performance in all conditions at the individual level because some reactive and proactive manipulations have very few responses even in this large data set. We attempt to address these challenges using hierarchical Bayesian modeling and parameterizations that balance parsimony, theoretical informativeness, and descriptive adequacy. We apply this approach to examine the reliability with which reactive and proactive control are measured and compare the outcomes to descriptive analyses employing descriptive measures based on a subset of each participant’s trials reported by Snijder et al. (2023).

This is an in-person presentation on July 21, 2024 (16:20 ~ 16:40 CEST).