Editorial for the special issue entitled: Mechanisms of control in selective attention tasks

As we view a scene or hear a cacophony of sounds, we often process just a small sample of the incoming sensory information. Sometimes the information we focus on is relevant to a particular goal we want to achieve; other times the information is irrelevant to our intended goals but captures our attention nonetheless. The term selective attention refers to those mechanisms which prioritise certain information over others and lead our experience to be dominated by one thing rather than another (Driver, 2001; Johnston & Dark, 1986; Treisman, 1969). The word ‘selective’ implies the operation of a mechanism whose purpose is to enable the processing of certain aspects of the incoming signal. Moreover, it also implies a need to do so; a need to filter out information that is less useful, perhaps to protect a limited-capacity system and/or to support goal-oriented behaviour, freeing us from the constraints of automaticity and stimulus-driven information processing and behaviour.

Historically, the field of selective attention focused on the type of information that selection is based on (e.g. discriminating useful vs. useless information on the basis of a particular physical characteristic such as colour or pitch). This led to a debate about whether selection was based on physical (early) or semantic (late) characteristics of the stimuli. It was later argued that selection could be determined by exogenous properties such as the number of items in a display whereby the greater the ‘perceptual load’, the less the irrelevant information was processed (Lavie, 1995; Lavie & Tsal, 1994; Lavie et al., 2014; see also Murphy et al., 2016). However, factors such as perceptual dilution (Kahneman & Chajczyk, 1983; Tan et al., 2018; Tsal & Benoni, 2010), spatial proximity of the irrelevant and relevant stimuli (Gatti & Egeth, 1978; Schneider, 2019), object-based attention (Chen, 2003), discriminability of dimensions (Melara & Algom, 2003), the exogenously determined span of visuo-spatial attention (Augustinova & Ferrand, 2014), and alerting cues (Weinbach & Henik, 2012) have been shown to modify selection success independent of perceptual load (Yeshurun & Marciano, 2013). In addition, issues related to task contingencies, such as correlations between the relevant and the irrelevant dimensions and variability in the irrelevant dimension (Melara & Algom, 2003) have been shown to modify the influence of irrelevant stimuli. All such findings highlight the influence of bottom-up factors in determining where attention is focused.

On the other hand, endogenous properties such as cognitive load also appear to affect the selectivity of attention; the higher the cognitive load, the more irrelevant information is processed (de Fockert et al., 2001; Kalanthroff & Henik, 2014; Lavie et al., 2014), indicating that a top-down mechanism – determining what is selected for processing – is in operation when cognitive load is low (but see Murphy et al., 2016, for a discussion of the modulating effect of simple vs. complex working memory loads). Moreover, findings such as apparent conflict adaptation effects (i.e. the reduction of the influence of the irrelevant information observed on trial N+1 because control has been activated by the experience of this conflict on trial N; Gratton et al., 1992) and proportion congruency effects (smaller interference from the irrelevant dimension when most trials are incongruent, i.e. involve conflict between relevant and irrelevant information) provide potential evidence for different mechanisms that operate over and control the top-down allocation of attention to facilitate selection (Botvinick et al., 2001; Braem et al., 2019). In an influential model, conflict (defined as the activation of multiple units in the same layer of a neural network) has been argued to be the trigger for top-down control (Botvinick et al., 2001). However, such effects also have counterpart bottom-up accounts (Schmidt, 2013, 2019) and there is a debate in the selective attention literature about just how much cognitive control influences the focus of attention (Algom & Chajut, 2019; Algom et al., 2022; Braem et al., 2019; Schmidt, 2013, 2019). Under these accounts, it is the bottom-up factors and statistical regularities of the task context that determine what is processed.

To investigate these different mechanisms in the lab, researchers employ tasks that involve an irrelevant dimension or stimulus that must be ignored or filtered whilst attempting to process (classify/name) a relevant dimension or stimulus. To illustrate, in one of the most commonly used selective attention tasks, the Stroop task (Stroop, 1935), participants are asked to identify the font colour a word is printed in whilst ignoring the meaning of the word. In this paradigm, control is thought to be evidenced by a smaller interfering effect from the irrelevant dimension. To measure performance on this task, one can average performance on particular trial types over an entire experiment, conceptualising control as static and time-invariant. For example, in the Stroop task performance is often measured by subtracting the averaged RTs of various trial types (e.g. mean RTs for colour-incongruent – colour-neutral trials, known as the Stroop interference effect), and control is specifically evidenced by a reduced magnitude of this difference. Indeed, a large interference effect is interpreted as a sign of impaired function (Kane & Engle, 2003). This measure of ‘static’ performance is often used to measure the what and the where of selective attention and gives us indices such as response, semantic, phonological and task conflict (see e.g. Parris et al., 2022).

In contrast, cognitive control can be conceptualised and measured as the influence of one trial on the next, control that happens in a time-varying manner, over the course of an entire experiment. Conflict adaptation and proportion congruency effects are taken as prime examples of this type of top-down control and indicate that participants are sensitive to the context in which each stimulus is presented. This type of control, referred to as dynamic control (Botvinick et al., 2001; Braem et al., 2019; Bugg, 2014; Bugg & Crump, 2012; Egner, 2023; Tzelgov et al., 1992), has been described as being either proactive or reactive (Braver, 2012) and captures the notion of learning and adapting to a changing environment.

Both static and dynamic approaches to the measure of control, as well as bottom-up factors affecting selective attention, are represented in this Special Issue. Studies in this issue address the question of whether there is evidence for an endogenous, top-down control mechanism in selective attention tasks (and at what level of processing that control is needed) or whether changes in indices of performance can be accounted for by exogenous, bottom-up, factors. In a cogent opinion paper, Schmidt argues that conflict adaptation, oft-heralded as a key marker of dynamic, proactive control, may not actually be particularly adaptive for performance, arguing that assumptions (often hidden or non-obvious) of conflict monitoring theory are non-trivial and, in many cases, imply relatively non-adaptive processes. Using pupillometry, Hasshim and colleagues address the question of whether the mechanism that produces the proportion congruency effect, thought to evidence dynamic, proactive control, is effortful as might be predicted by control accounts of the effect. Spinelli and Lupker address the issue of whether an imbalance in the relative frequencies of distractor-target combinations (e.g. where the word RED in red in the Stroop task is more common than other combinations) can account for the magnitude of the Stroop effects as has been predicted by input-driven accounts of Stroop effects. Gallego et al. assess the contributions of cognitive control, associative learning and episodic effects to the item-specific proportion congruency effect.

Several contributions to this Special Issue consider task conflict and its control in selective attention tasks. Task conflict refers to conflict between the task sets associated with the relevant and irrelevant dimensions of the stimuli, a notion that was borne of studies investigating mechanisms in task switching (Monsell et al., 2001). For example, in the colour-word Stroop task, the task set for colour naming is endogenously activated in line with task instructions. However, this competes with the task set for word reading, which is unintentionally activated on the presentation of a real-word stimulus (Goldfarb & Henik, 2007; Keha & Kalanthroff, 2023). Whilst historically competition was thought to exist between response, semantic and phonological representations, this relatively newly identified form of conflict is thought to occur and be controlled earlier in processing (see Goldfarb & Henik, 2007; Kalanthroff et al., 2018; Littman et al., 2019 for a review; see Parris et al., 2023, for an alternative viewpoint).

Keha and Kalanthroff contrast proactive control adaptation and contingency learning accounts of Stroop effects using a correlation-free design with inducer and diagnostic items to measure task conflict (RT for neutral-word > RT for neutral symbols) in mostly incongruent, mostly congruent, mostly neutral-word and mostly non-lexical (shape strings) contexts. Hershman et al. explored the contribution of the difficulty of processing the irrelevant dimension, and thus the extent of task conflict, on Stroop task performance. Morretti et al. aimed to investigate the mechanisms involved in resolving task conflict and determine if cognitive control could be dynamically upregulated when dealing with stimulus-based task conflict, akin to the adaptation of control commonly observed in response conflict, and ask whether different control mechanisms may be employed for different types of conflict.

Two papers in the current Special Issue consider where control might be needed in the Stroop task. Following a recent paper critically evaluating the notion of task conflict in selective attention tasks (Parris et al., 2023), Parris et al. consider a phonological account of a key marker of task conflict known as negative facilitation (congruent RTs > non-word neutral RTs). Finally, Qiu and van Heuven use Chinese colour-word homophones that are completely orthographically distinct from colour words to ask whether conflict happens at the level of phonological processing in the Stroop task and how this conflict changes as a function of manual vs. vocal responding.

In sum, this Special Issue joins similar endeavours dedicated to various issues surrounding cognitive control (Henik et al., 2018; Parris et al., 2019; Schmidt et al., 2015) and aims to stimulate future research in this direction. The studies presented herein attempt to provide answers to the questions of if, how and where cognitive control happens and how much it contributes to attentional selection.