Most of our daily tasks, from making coffee to crossing the road, require us to keep information in mind, to use it for guiding future actions and to update it whenever newer information arrives. Working memory (WM) is the cognitive system that enables these abilities (Baddeley and Hitch, 1974; Miyake and Shah, 1999), and WM updating is the ability to modify the stored information upon need (Morris and Jones, 1990). Updating is an ability rather than a single process, which is carried out by removing outdated items (Oberauer, 2001; Ecker et al., 2014), by adding new items, and/or by substituting the existing information with new one (Kessler and Meiran, 2006, 2008; Ecker et al., 2010; Kessler and Oberauer, 2014, 2015).

Theoretical models of WM emphasize the conflict between maintenance and updating (Frank et al., 2001; Miller and Cohen, 2001; O'Reilly, 2006; Badre, 2012; see Fallon et al., 2017, for a recent demonstration). Specifically, WM enables to maintain relevant information in a highly accessible state, shielded from being interfered by irrelevant information. However, this shielding should be removed rapidly when needed, in order to enable updating the representations held in WM. Therefore, rapid updating must counteract maintenance. The tension between these two forces is assumed to be regulated by cognitive control, in the form of a decision process that determines which items will be updated and when. The above-mentioned theoretical models account for this control by assuming a gating system, which separates WM from the flow of information that arises from internal (thoughts) or external (perception) sources. The gate serves as a selective filter. When closed, the gate enables robust maintenance within WM, while blocking irrelevant information. In contrast, opening the gate allows WM to be updated with goal-relevant input. Hence, the gate has two “states,” open and closed. Controlling these states is critical to the optimal function of WM. It is worth noting at this point that the gate metaphor typically applies to situations in which relevant and irrelevant information is presented sequentially, rather than simultaneously. In these situations, it is possible to block irrelevant information and preserve relevant one by changing the state of the gate over time, while alternating between receiving relevant and irrelevant information. Tasks that presumably involve temporally-based gating include the AX-CPT task (Braver and Cohen, 2000; D'Ardenne et al., 2012; Kessler et al., 2017), attentional blink (Raymond et al., 1992), and complex span (Daneman and Carpenter, 1980). As will be elaborated below, the goal of the present work is to examine the role of gating in task switching, and specifically the relationship between the process of gate opening to that of switching a task-set.

As mentioned above, updating is not a unitary process. Rather, based on the gating model, I suggest that WM updating is carried out by a complex cascade of sub-processes. These include: (1) detecting a change in the environment (Rensink, 2002; Hyun et al., 2009); (2) identifying the changed information as one that requires robust maintenance. In other words, identifying it as goal-relevant; (3) opening the gate to WM; (4) removing the outdated item from WM (Oberauer, 2001; Ecker et al., 2014); (5) encoding the new information into WM; and (6) closing the gate in order to shield the updated information for future distractors. Due to the large number of processes involved, it is a challenge to identify these processes, tease them apart, understand the conditions under which they take place, and specify the relationship among them. For example, it is conceivable that some processes are not required in some cases, while others are optional and depend on the task structure or individual differences. Also, the order in which these processes take place it is still not well-understood, including whether they operate serially or in parallel.

Kessler and Oberauer (2014) demonstrated that the time required to update WM is composed of two components: switching the state of the gate (namely, opening or closing), and modifying the associations between items in WM and their context (e.g., position). Participants were presented with series of trials, each trial comprised a set of 4 letter. After a varied number of trials, they were prompted to recall the most recent set of letters. Accordingly, the participants had to update their WM with the new letter set in each trial, and were not required to remember the letters presented in earlier trials. In each trial the participants had to press a key to in order to proceed to the following trial. This keypress enabled measuring the duration of WM updating. Critically, in each trial, some of the items were repeated and some were updated, compared to the previous trial. The number and serial positions of updated and repeated letters was manipulated. Updating times were explained by a scanning and gate-switching model. According to this model, in each trial participants scan the letter set according to the reading direction (i.e., left-to-right for English letters, right-to-left for Hebrew letters; see Kessler and Oberauer, 2015). While scanning, they are encountered with new items (that need to be updated into WM), and items that are repeated from the previous trial (that do not require updating). Accordingly, moving from a repeated item to an updated one requires to open the gate to WM, while moving in the other direction requires closing the gate. Updating costs increased linearly with the total number of gate switch operations. In addition, within a sequence of updated items, updating times were proportional to the number of updated items, reflecting the added cost of constituting an association between the new item and its context.

A more direct evidence for gate switching costs was provided using the reference-back task (Rac-Lubashevsky and Kessler, 2016a), that was developed with the aim of disentangling the sub-processes of WM updating. The reference-back is based on the n-back task, arguably one of the most commonly used measures of WM updating (e.g., Jonides et al., 1997; Chatham et al., 2011). In the “standard” n-back task, participants are presented with a stimulus in each trial, and are required to decide whether or not it is identical to the stimulus presented n trials before. Since each presented stimulus will later serve as a reference for comparison, WM updating is required in each trial of this task (with the possible exception of 1-back, see Rac-Lubashevsky and Kessler, 2016a). This fact, along with the overall computational complexity of the task, does not permit one to extract the updating process in isolation due to the lack of baseline. Also, it is not clear which trials of the n-back task trigger gate opening and closing.

The reference-back paradigm was developed in order to overcome these shortcomings. This task is composed of two types of trials, reference and comparison, which are indicated by different colors, a red or blue frame surrounding the stimulus, respectively. In each trial, participants are required to indicate whether the presented item is the same as, or different from, the most recent item that appeared within a red frame. Accordingly, each trial in this task requires a comparison to the reference followed by a same/different decision. In addition, reference trials (indicated by a red frame) require one to update WM with the presented stimulus, because it should serve as a reference to which the following trials should be compared. Thus, reference trials require opening the gate to WM, in order to enable updating. On the other hand, comparison trials (indicated by a blue frame) do not require WM updating. Instead, these trials require one to continue maintaining the last reference stimulus in WM. Because each comparison trial is also compared to the last reference trial, the reference needs to be protected from being overwritten by changes in comparison trials. Hence, the gate to WM should be closed in these trials. Previous results using this paradigm (Rac-Lubashevsky and Kessler, 2016a,b; Rac-Lubashevsky et al., 2017) demonstrated that (a) performance in reference trials is slower than in comparison trials, supporting the additional updating process required in the former, and (b) switching between the two trial types is associated with an additional cost, reflecting the time taken to open or close the gate to WM. This cost implies that the state of gate tends to remain constant, either open or closed, until a change is required (see also Kessler et al., 2017, for a similar finding using the AX-CPT task).

The behavioral evidence for gate-switching costs that were presented above arrive from paradigms that involve declarative materials (such as letters, digits) as memoranda. Oberauer et al. (Oberauer, 2009; Oberauer et al., 2013) suggested that WM is composed of analogous structures and processes that apply to declarative and procedural information. In their framework, procedural WM is the ability to maintain, update and manipulate task-sets and task-rules. As with declarative information, procedural WM is limited in capacity, is composed of associations between items (responses) and contextual information (stimuli), and depends on a selection mechanism. While the two WM systems are analogous, they process information in parallel, suggesting that they are independent (Souza et al., 2012). Based on the distinction between procedural and declarative WM, the present study sought out to examine whether gate switching also takes place in procedural WM, and how it is related to other phenomenon observed in this sub-system.

More specifically, the present study examined whether switching between task-sets (Grange and Houghton, 2014) involves gate opening. Task switching is a core executive function (Miyake et al., 2000). In a typical task switching experiment, participants are required to alternate between different sets of S-R rules (“task-sets”) that can be applied to the same stimuli. Therefore, switching between task-sets reflects cognitive flexibility, since it requires from the participant to react differently to the same perceived input. Typically, a task-cue is presented either before or together with the stimulus. The cue indicates which of the possible task-sets is relevant in the present trial (Meiran, 1996). The most important and robust finding with this paradigm is a task-switching cost. Specifically, performance (in reaction times and accuracy) is worse for switch trials, in which the relevant task is different from the one that was relevant in the preceding trial, compared to repetition trials, in which the relevant task did not change.

The ability to apply the cued task-set to a stimulus is a prime example of goal-directed behavior, in which the required action depends on the internal representation of the desired goal. In contrast to automatic performance, in which the stimulus is directly associated with its response (i.e., in long-term memory), in goal-directed behavior the response is mediated by the on-line representation of the relevant goal, which is often selected among several options. WM plays a crucial role in maintaining the relevant task set and biasing performance according to it (Miller and Cohen, 2001; Kane and Engle, 2003). Therefore, switching the relevant task is a special case of (procedural) WM updating (see Oberauer et al., 2013).

In the present work, the possible role of gating in updating WM with task-sets is examined using a procedural version of the reference-back paradigm. This is essentially a reference-back task, in which the memoranda are tasks rather than letters. Participants were required to switch between tasks upon a cue, as commonly used in the cued task-switching paradigm. Importantly, the relevance of the task cue was manipulated between trials. In attend trials, the participants were required to perform the task indicated by the cue, as in standard task switching experiments. In ignore trials, however, they had to ignore the cue and continue to perform the task that was relevant in the previous trial (c.f. rule violation; Pfister et al., 2016). In other words, ignore trials required to filter out the cue information, while maintaining the previously-relevant task-set in WM. Accordingly, the attend and ignore conditions correspond to updating and maintenance modes of WM operation, respectively (c.f. Kessler and Oberauer, 2014). Critically, switching between these conditions corresponds to opening and closing the gate to WM. Specifically, switching from ignore to attend involves gate opening, and switching from attend to ignore involves gate closing (see Figure 1).

The procedural reference-back task enables to examine the interaction between gate opening and task switching. Specifically, ignore trials always involved a task-set repetition, since the cue identity was irrelevant. Attend trials, in contrast, could either involve a task repetition or a task switch. In addition, an attend trial could either follow an ignore trial, implying a gate opening, or follow another attend trial.

Back to the question of the relationship between task-switching and gate opening, three possible results are conceivable. One possibility is that task-switching cost is additive with gate opening cost. That is, the same task switching cost will be observed when moving from an ignore trial to an attend trial, and within a sequence of attend trials. Such a result would imply that gate opening is independent of task switching, suggesting that task switching cost does not include the duration of gate opening. A second possibility is an under-additive interaction, namely a smaller gate-opening cost in task-switch trials. This result implies a that the two processes take place in parallel. A third possibility is an over-additive interaction, reflecting a cross-talk or shared resources between the two operations.