Augmented Reality Cues Facilitate Task Resumption after Interruptions in Computer-Based and Physical Tasks

Many work domains include numerous interruptions, which can contribute to errors. We investigated the potential of augmented reality (AR) cues to facilitate primary task resumption after interruptions of varying lengths. Experiment 1 (N = 83) involved a computer-based primary task with a red AR arrow at the to-be-resumed task step which was placed via a gesture by the participants or automatically. Compared to no cue, both cues significantly reduced the resumption lag (i.e., the time between the end of the interruption and the resumption of the primary task) following long but not short interruptions. Experiment 2 (N = 38) involved a tangible sorting task, utilizing only the automatic cue. The AR cue facilitated task resumption compared to not cue after both short and long interruptions. We demonstrated the potential of AR cues in mitigating the negative effects of interruptions and make suggestions for integrating AR technologies for task resumption.


INTRODUCTION
Interruptions are frequent in many work domains, such as healthcare [15,24,32,39], aviation [63], software development [1] or office work [19].Interruptions decrease efficiency [3,89], lead to higher error rates [23,47,83,94] and decrease satisfaction with one's own work performance [6].Although efforts have been made to avoid disruption altogether [16], this approach requires deep workflow restructuring and is not applicable to all tasks.For example, time-critical tasks that must be addressed immediately may occur and thus require the interruption of the primary task.A different approach to interruption management is to attempt to mitigate the negative effect of the interruptions.For this purpose, various types of visual cues have been explored, demonstrating varying degrees of efficacy.For example, carrying a syringe enhanced task resumption in nurses [31].Conversely, cues, such as the simple act of placing the mouse cursor on the correct task point for task resumption did not improve task resumption [90].
We investigated whether cues on a head-mounted augmented reality (AR) display can mitigate the negative effect of interruptions.The application of cues via AR holds the promise of being task-independent, possible for task were users are mobile and customizable, but, to the best of our knowledge, the efficacy of AR cues has not yet been investigated.Based on the so-called Memory for Goals theory [2], we used the resumption lag (i.e., the time between the end of the interruption and the resumption of the primary task) to measure the disruptive effect of an interruption on the performance of the primary task.In two experiments, we investigated whether AR cues at the point of task resumption could reduce the resumption lag.In Experiment 1, we used a computer-based primary task, which is frequently used in the context of the Memory for Goals theory.We chose such a task to test that AR cues can be studied using the Memory for Goals theory and to investigate what kind of AR cues provide good task resumption support.More specifically, we investigated whether a cue automatically placed by task-tracking software or manually placed through a hand gesture by the user provided better support for task resumption.In Experiment 2, we used a tangible primary task to investigate the effectiveness of the automatically set cues in a task that more closely resembled a tangible working task, such as medication sorting.Considering the relevance of spatial memory for task resumption [76], our goal was to evaluate how cues performed in a task with a more notable spatial dimension than the computer-based task in Experiment 1.We contribute to HCI-interruptions research by providing (1) empirical data that show that AR cues in computer-based and tangible tasks facilitate the resumption of primary tasks after an interruption; (2) empirical data that show that manually set cues provide no further advantage over automatically set cues; and (3) a new task paradigm for studying interruptions of tangible primary tasks that allows for easy adaptation of task complexity.

RELATED WORK 2.1 HCI Research on Interruptions
In the past two decades, interruptions have been extensively investigated in the field of HCI [64,65].Salvucci et al. [82] introduced a Multitasking Continuum, encompassing concurrent multitasking and sequential multitasking.Concurrent multitasking includes more task switches and shorter actions on the different tasks and frequently requires self-initiated switching (e.g., driving and talking).Sequential multitasking is characterized by fewer task switches and longer periods of attention devoted to the single tasks (e.g., cooking and reading a book).Interruptions are considered sequential multitasking, and numerous studies have highlighted the disruptive effects of interruptions in laboratory settings and various work domains, including, for example, office work [19,45], computer programming [102], healthcare [25,34], aviation [57,63] or assembly work in manufacturing [50].
Early work on interruption coordination by McFarlane and Latorella [65] differentiated between immediate, negotiated, mediated and scheduled.HCI work attempted to reduce the disruptive effects of interruptions through mediated or scheduled interruptions [44,87,102].For example, Iqbal and Bailey [44] developed and evaluated a notification management system for computer-based work.The system reduced frustration and confirmed the users' preference to schedule interruptions during phases of low workload.Whereas Iqbal and Bailey's system used task data to determine the next fine, medium or coarse task-breakpoint, researchers also used psychophysiological sensors [102,103] or the sensors of an off-the-shelf virtual reality device [14] to predict the interruptibility of users.Finally, a combination of task and physical data seems to provide the best interruptibility predictions [103].
A large corpus of HCI work addresses the negotiation of interrupting tasks to understand when interruptions are less disruptive and more acceptable.For example, laboratory studies showed that users prefer to monotask and accept interruptions more often in phases of low workload [43,80], but that time constraints, distance to the next good breakpoint [10] and task properties [33,93] also influence the decision to accept an interruption.
Finally, immediate interruptions have been addressed due to the difficulty of resuming the task at hand after handling the interruption [65].In healthcare, immediate interruptions are the most prevalent interruption coordination method [34,75] and definitions of interruptions in the healthcare context have been characterized by an unexpected and immediate onset that results in a temporary break in the task at hand [11,32].For example, in the emergency department, 75.4% of all interruptions were immediately attended to, whereas 22.2% were coordinated via concurrent multitasking and only 2% were negotiated or rejected.

Quantifying and Mitigating the Effect of Immediate Interruptions
To quantify the effect of immediate interruptions on primary task resumption, Trafton et al. [91] suggested measuring the so-called resumption lag.The resumption lag is the time from the end of an interruption to the continuation of the primary task.Longer resumption lags indicate a more difficult task resumption compared to shorter resumption lags.The idea that the resumption lag is sensitive to interruptions is built on the activation-based Memory for Goals theory [2].According to the Memory for Goals theory, the most active goal at any moment directs a person's behavior.Furthermore, the level of activation of a suspended (i.e., interrupted) goal decays over time.The higher the activation level of the suspended goal, the easier it is to resume the suspended task, which is indicated by short resumption lags.The activation level of a goal can be raised by providing environmental cues that are associated with the goal.In this context, a cue can be any perceptual signal that is directly associated with the goal or guides the individual towards something in the environment that is associated with the goal [92].
The Memory for Goals theory and the resumption lag construct have received empirical support in various studies.Studies in laboratory [52,54,61] and field settings such as intensive care nursing [31] showed that longer interruptions (i.e., more goal decay of the to-be-resumed primary task) resulted in longer resumption lags.Furthermore, longer interruptions were associated with forgetting to resume the interrupted tasks in emergency departments [24].Interruptions at moments of low workload resulted in smaller resumption lags [43,102].The positive effects of environmental cues on the resumption lag have shown that visibility of the primary task during the interruption reduced the resumption lag compared to occluded primary tasks [40,74].Furthermore, providing visual cues at the end of the resumption in the form of mouse cursors at the to-be-resumed location also reduced the resumption lag in some studies [5], while it did not in others [90].Besides general environmental cues, additional artifacts can be used as cues to facilitate primary task resumption.Such cues might be artifacts that are relevant for reaching the goal (e.g., healthcare workers holding on to a syringe during the interruption [31]), as well as additional cues such as sticky notes [28] or reminder systems designed for specific tasks such as unmanned aerial vehicle supervision [84].
Recommendations to diminish the negative effects of interruptions can be summarized in two general strategies [65,75].The first recommendation is cognitive rehearsal, which increases the activation of the to-be-resumed task's memory representation.Although effective in laboratory settings [4,66], cognitive rehearsal has not yet been observed in field settings [31].Furthermore, the nature of immediate interruptions may restrict users from conducting cognitive rehearsal before attending to the interruption or during the interruption.
The second recommendation is to maintain the activation of the to-be-resumed task's memory representation high via environmental cues.This may include visual access to the primary task during the interruption [74] or post-interruption cues, such as artifacts in the environment [31], a mouse cursor [5], marked text passages [17,18,90] or open software in the foreground [45].Placing environmental cues in form of mouse cursors or marked text passages in desktop-based tasks at the end of the interruption has been shown to be effective in assisting task resumption in some [13,90] but not all cases [17,18,90].Salient cues that provided spatial information about the to-be-resumed tasks seems to be the most effective method [13,90].
However, using physical artifacts or computer screen-based cues has drawbacks.These cues require that the user has to either stay at the location of the primary task, maintaining the primary task in their field of view for visual perception, or carry the cue along during the interruption.Furthermore, such cues may be only used for the specific task or sub-task that they have been design for.AR cues do not face these disadvantages, as they can be displayed irrespective of location, are task independent, and can be operated hands-free.These represent distinct advantage, particularly for environments where interruptions often require the user to change the physical location or work on different task that may include a computer screen or are tangible tasks.For example, in intensive care units [31] or emergency departments, nurse or doctors need to be mobile, need technology that can be used hands-free and satisfies a bare below the elbows policy and have computer-based tasks such as charting but also manual tasks such as setting up infusions.

Augmented Reality and Interruptions
In recent years, there has been a surge in research dedicated to utilizing the seamless integration of digital information via AR in fields like education [8,21], architecture [59] and remote assistance [36,69].The inherent adaptability of head-mounted AR displays yields potential for application in interruption-rich environments, particularly in domains such as healthcare [30,86], aviation [97] manufacturing [72,100,101] and assembly tasks [41,42,55,79].Researchers started to investigate AR systems for patient monitoring in intensive care units [49,78].These developments enable real-time data access and visualization, providing healthcare professionals with location-independent and hands-free information.However, providing information on head-mounted displays also may be a further source of interruptions.For example, supervising anesthesiologist who used a head-mounted display with information of six ongoing operations felt an obligation to consider vital sign changes in the operating rooms and interpret the situation.This resulted in looking at the head-mounted display with a "frozen" head posture and 10-25 second-pauses of the ongoing task [53].
In manual assembly tasks, instructions presented via AR reduced cognitive load and received higher usability ratings compared to print instructions [55].Similarly, for warehouse order picking, workload and performance showed benefits for head-mounted display compared to all other methods [35].Because interruptions are more disruptive during periods of higher cognitive load [43], the potential improvement in task resumption by transferring tasks onto the AR glasses warrants further investigation.Studies have demonstrated the effectiveness of AR-based task guidance in supporting car repair, particularly for workers facing novel tasks [41,42] and performance was better for central presentation compared to peripheral presentation on the head-mounted display [99].In general, task guidance via AR offers potential benefits for reorienting workers after interruptions, but does not offer explicit support for task resumption by leveraging AR cues.Unlike previous research, such as that by Rukubayihunga et al. [79] on assembly step recognition in AR, which lays the groundwork for automatic deployment of cues, our study focuses on utilizing manually or automatically set cues in a more versatile setting.
Finally, recent research has also addressed how one should interrupt head-mounted display users in virtual reality [14,26,70].For example, researchers have collected sensor data of VR devices to predict interruptability to schedule interruptions [14].Despite this interest in interruptions, the current research is rather limited regarding task resumption support via AR cues.More related to immediate interruptions and the need for task resumption, a study investigated the effect of interruptions on opaque or transparent smart glasses on the resumption lag [52].Based on the Memory for Goals theory, researchers expected that transparent smart glasses would provide environmental cues of the to-be-resumed task while attending to the interruption on the smart glass.However, compared to opaque smart glasses or a tablet control condition, transparent smart glasses did not reduce the resumption lag.It may be that the focus on the smart glass in order to work on the interrupting task resulted in inattentional blindness for the environmental cues [51], and as a result, the cuing was not effective.Consistent with the findings of Trafton et al. [90], who showed that subtle cues such as placing the mouse cursor at the to-be-resumed task step did not improve task resumption, the mere visibility of the environment might have been too subtle a cue to facilitate task resumption.
Following Trafton et al. [90], the design of our AR cue as a salient red arrow aims to resolve this issue by providing a blatant cue.In addition to noticing the cue, task resumption is also a spatial memory problem [21,77].Ratwani and Trafton [77] demonstrated the effect of intact or compromised mental spatial representations on task resumption and suggested that memory for spatial location may guide task resumption.Different to, for example, verbal information or semantic information about the to-be-resumed task step, a red AR arrow can be placed at the exact location and support spatial memory during task resumption.Finally, in relation to above described artifact cuing were cues are most of the time only working for a specific task, we also expect the red AR arrow to be a generic cue design that is task-independent and can work for computer-based and tangible tasks.Wolff et al. [97] showed that such salient cues can improve task resumption by circling the to-be-resumed task step in a scanflow task (i.e., participants had to confirm the color of various shapes in a 48-shape array).While their approach involved a Wizard of Oz paradigm and mock-up AR cues instead of circles on a monitor, making it less versatile compared to our paradigm, they still demonstrated that AR cues can reduce the number of errors and mental load resulting from interruptions.Applying Wolff et al. [97] findings to an actual AR setup would enable us to leverage the task-and location-independent characteristics of AR cues and empirically test whether the benefits translate to actual AR applications for computer-based and tangible tasks.
Despite a broad interest in interruptions in recent AR and virtual reality research and the possible benefits of AR cues in form of location and task independent cueing of to-be-resumed tasks, theory-based empirical research on the effectiveness of AR cues in task resumption is missing.Because the current empirical evidence for the effectiveness of environmental cuing in relation to opaque or transparent smart glasses [52] and subtle cues [90] is not fully conclusive and the only AR study included mock-up AR and Wizard of Oz [97], we decided to conduct experiments in which we compared a no cue condition with the different AR cue conditions.Our study's contributions lie in investigating the effectiveness of manually or automatically set AR cues at the point of task resumption in both computer-based and tangible tasks.

EXPERIMENT 1: COMPUTER-BASED PRIMARY TASK
Experiment 1 aimed to examine the potential of AR cues in mitigating the negative effects of interruptions on the resumption lag.Additionally, we sought to determine whether the mode of cue presentation, specifically whether the cue is automatically placed by task-tracking software or manually placed by the user, impacts the effectiveness of the cues.The factor cue was manipulated between subjects and included the three groups: automatic cue (i.e., automatically set cues), manual cue (i.e., cues set via a hand gesture at the start of the interruption) and no cues (i.e., a control group with no cues at all).To examine the potential influence of interruption length on the effectiveness of cues, the study varied the duration of interruptions between 15 and 45 seconds within-subject.Due to the fact that, to the best of our knowledge, this is the first experiment to address the effect of AR cues on the resumption lag, we used a computer-based primary task with a structure that is regularly used to investigate the resumption lag [52,76].In this task, participants had to transfer patient information from tables in the center of the screen into their corresponding input fields above and below the tables.In line with resumption lag research [52,76], there is only a single task to-be-resumed and the focus is on task resumption times.Based on previous research on the resumption lag [67, 90], we hypothesized that a shorter resumption lag would occur with (1) an AR cue (automatic and manual) compared to no cue and (2) short interruptions compared to long interruptions.Additionally, based on the Memory for Goals theory, because the availability of a cue at the end of the interruption should render the activation loss during the interruption irrelevant, we expected that (3) longer interruptions would have a smaller effect when an AR cue (automatic and manual) was presented.Finally, setting the cue manually increases the interruption lag and includes a physical gesture.As longer interruption lags [40] and tangible tasks [25] may increase memory encoding of the to-be-resumed task, we further hypothesized that (4) a manual cue reduces the resumption lag more than an automatic cue.To assess the subjective experience of using a manual cue or an automatic cue, we used the NASA Task Load Index (NASA TLX) [38] to investigate workload differences between the cue types, which is a commonly used metric in interruption research [e.g.43,48,58,62,83] and AR research [e.g.35,42].The research question, empirical hypothesis, research design, sample size, and analysis plan for Experiment 1 have been preregistered (https://doi.org/10.17605/OSF.IO/GBX2V).3.1.2Design.We used a mixed 3 (cue) x 2 (interruption length) design.Participants were randomly assigned to three groups: manual cue, automatic cue and no cue.The within-subject factor interruption length varied between 15 and 45 seconds.We conducted a prospective power analysis to determine the required number of participants for this study.Based on a previous unpublished study, we estimated the effect size of the cue × interruption length to be f = .175(power = .80,alpha = .05,two-sided test).The power analysis indicated a sample size of 3 × 28 participants.In the manual cue condition of Experiment 1, the interruptions started after the participants marked a position on the primary task via a hand gesture.The resumption lag was measured from the end of the interruption, i.e., the tablet turning black after the last arithmetic task, to the first interaction with the respective primary task.The red AR cue was only present in the cue condition and the arrow was visible during the resumption lag.Note that we enhanced the outline of the AR cues for better visibility in the figure .on the "OK" button (Figure 1, number 3) of the respective module when they have entered the patient data.Previous data got cleared after clicking on "OK" to not give the participant any cue about the current progress, thereby forcing them to remember their current progress.Clicking on the "OK" button may also have triggered an interruption, indicated by an acoustic signal and the mouse rendering immobile at the center of the screen.The interruption consists of either three or nine simple arithmetic problems with results ranging from 0 to 30 (Figure 3) which were presented on a tablet that was located on a table behind the participant.Each problem was presented for 5 s.The problems, their answers and their order were the same for all participants.When the interruption ended, an audio cue sounded, and the tablet turned black.The participant had to turn around again and resume the primary task by activating the next module.For the resumption lag, we measured the time between the end of the interruption and the first correct click in the primary task (= pressing the next "Edit" button).Once the participant had finished all modules (i.e., a trial), Figure 3: The primary task (left) with the superimposed cue from the participants' perspective.The cue was pointing to the edit button of the next module and disappeared as soon as the participant clicked on the edit button.The interrupting arithmetic problems task (right) was presented in the center of the tablet, which was placed on a table behind the participants.Four solutions to the problem were presented at the bottom, of which only one is correct.they needed to progress by clicking on the "Process" button on the right-hand side of the screen (Figure 1, number 4).The timeline of the interruptions is shown in Figure 2.

Method
In the automatic cue and manual cue groups, participants wore Microsoft HoloLens.In the manual cue group, participants had to place the cue themselves at the beginning of the interruption by using an air tap gesture (tapping the index finger onto the thumb and raising it again while pointing onto the primary task).Participants in the manual cue group also practiced the air tap gesture for setting the cue.At the end of the interruption, a three-dimensional, red arrow was superimposed via the HoloLens as a cue in both cue groups (Figure 3).The arrow pointed to the "Edit" button of the module that the participant needed to resume or to the "Process" button.Since participants in the no cue group did not benefit from the HoloLens and the display might be distracting when not in use, they did not wear the HoloLens.
Following this, the main part of the experiment commenced.Participants were tasked with inputting patient data for a total of eight unique cases, with interruptions introduced in four of these cases.Within each of the interrupted patient data sets, three interruptions of varying lengths were incorporated.In total, participants were interrupted six times for 15 s and six times for 45 s.The time of the interruptions was defined in four configurations, which were equal across the cue groups and were assigned randomly.The configurations allowed no more than two interruptions in succession within a patient data set, no more than three consecutive data sets with interruptions and balanced the order of the interruption lengths within the data sets.
Lastly, when participants completed the main part, they were asked to complete a post-survey.Participants completed the NASA TLX [38], provided demographic data and were asked to describe if they used any particular strategy during the experiment.
3.1.4Analysis.During data preprocessing, interruption lags shorter than 1 second were highlighted and manually investigated.The analysis of these interruption lags indicated that two resumption lags of two participants were invalid because a bug in an early version of the experiment software sometimes automatically set the cue, preventing participants from manually setting it with the gesture.We therefore removed these two resumption lags.Additionally, 31 (3.1 %) resumption lags were removed due to resumption errors, i.e. when the first click after the interruption was not on the edit button of the correct module.A total of 963 (96.7 %) interruptions were included in the resumption lag analysis.
We performed a 3 (cue) × 2 (interruption length) repeated measures ANOVA.In three of the six experimental conditions, the values of the resumption lag were not normally distributed (Shapiro-Wilk; all p-values < .05).Because reaction times are most of the time positively skewed [95] we planned to transform the data using the common logarithm, but the transformation did not improve all distributions.As we could assume homogeneity of covariances (Box-Test, p = .055)and homogeneity of the error variances (Levene-Test, all p-values > .05),we assumed the ANOVA to be reliable without transformation of the data [9,85].To test our hypothesis that AR cues (automatic or manual) would reduce the resumption lag, we conducted two Helmert contrasts.For our three cue groups, the first Helmert contrast compared the effect of any cue (automatic or manual) vs. no cue.The second Helmert contrast compared the effectiveness of automatic cues vs. manual cues.

Results
As a general manipulation check to see whether the interruptions were disruptive, we compared the resumption lag with the interaction interval (i.e., the time that is used for the same task step without an interruption).A Wilcoxon signed-rank test showed that the time between the end of an interruption and the next click on the correct module (resumption lag M = 4072 ms, SD = 1221 ms) was significantly longer than the time between a click on "OK" and the next correct click on the next module when no interruption occurred (interaction interval M = 1669 ms, SD = 660 ms), (Z = 3483.000,p < .001).The main effect of cueing was significant, F(2,80) = 4.249, p = .018,partial η2 = .058(Figure 4 and Table 1).Supporting our hypothesis, the first Helmert contrast of any cue (automatic or manual) vs. no cue indicated a significantly shorter resumption lag when a cue was present, t(80) = 2.893, p = .005.However, contrary to our expectations, the second Helmert contrast of automatic cue vs. manual cue showed no advantage of the manual cue over the automatic cue, t(80) = 0.973, p = .333.The main effect of interruption length was significant F(1,80) = 16.727,p < .001,partial η2 = .173.As expected, longer interruptions resulted in longer resumption lags.However, there was a significant cue × interruption length interaction, F(2,80) = 3.135, p = .049,partial η2 = .073.Consistent with our hypothesis, a Bonferroni-adjusted post-hoc analysis showed that, when no cue was provided, the resumption lag was significantly longer after the 45s interruption than after the 15s interruption (p < .001),whereas there was no difference between the 45s and the 15s interruptions when a cue was provided (automatic cue: p = 1.000; manual cue p = 1.000).
Resumption errors showed the same result pattern as the resumption lag.The cue had a significant main effect, F(2,80) = 6.159, p = .003,partial η2 = .072.The first Helmert contrast indicated that participants exhibited fewer resumption errors when presented with an AR cue compared to no cue, t(80) = 3.408, p < .001.d = .536.
Considering the subjective mental workload, we calculated the NASA TLX raw score by averaging the six sub-scales [37].A oneway ANOVA showed no difference between the cue groups, F(1,80) = 1.008, p = .370,partial η2 = .025.More specifically, there was no significant difference between the manual and the automatic cue regarding the NASA TLX raw, t(80) = 0.770, p = 1.000, d = 0.222.
In order to explore the temporal dynamics of reorientation towards the primary task when a cue was presented, we compared the first interaction interval following task resumption, the so-called edit lag [71], between the groups.This analysis aimed to determine whether the participants simply clicked where the cue indicated but needed more time to reorient and retrieve task information after making the first click.When a cue was presented (M = 1673 ms, SD = 632 ms), the following interaction interval was not prolonged compared to when no cue was presented (M = 1659 ms, SD = 725), t(80) = 0.105, p = .917.

Discussion
The objectives of Experiment 1 were to assess the potential of AR cues in facilitating task resumption following interruptions and to examine the influence of interruption length on the effect of AR cues and cue-setting methods.We focus our discussion on the important results for Experiment 2 and provide a broader examination of the results in the general discussion section.The findings only partially support the expected benefits of task resumption facilitated by AR cues.After long interruptions, both the manually and automatically set cues demonstrated superiority over the absence of cues.However, when considering short interruptions, no significant differences were observed between the cue conditions.This lack of differentiation after the 15-second interruptions may be attributed to a floor effect.While we could demonstrate that the interruptions were disruptive, the primary task was still fairly simple, rendering it challenging to detect a facilitation of the task's resumption.Task resumption in the tangible task in Experiment 2 maybe cognitively more demanding and thus might provide more insights on AR cue task resumption facilitation for shorter interruptions.
Importantly, the resumption lag did not significantly increase between 15-second and 45-second interruptions when any type of cue was provided.These findings align with the Memory for Goals theory [2], as the priming constraint suggests that a cue can elevate the activation of the goal, regardless of the duration of the interruption.
In contrast to our expectations, the manual cue provided no additional facilitation of task resumption compared to the automatic cue.The large effect of the presence of any AR cue might lead to a floor effect, precluding any further reduction of the resumption lag through the manual setting.Based on this result, we focused on automatic cues in Experiment 2 only.

EXPERIMENT 2: TANGIBLE PRIMARY TASK
In Experiment 2, we used a tangible primary task instead of a computer-based task to further explore the generalizability and applicability of the observed effects in Experiment 1.Few studies investigate the resumption lag with tangible tasks [25,31,62] and, to the best of our knowledge, no studies investigate the effects of AR cues on tangible task resumption.We constructed a tangible primary task that resembles the act of sorting medication into a 7-day pill organizer.Medication preparation is a cognitively demanding [98] and error-prone task [46].Interruptions have been identified as a contributing factor to medication preparation errors [23,46].
Based on the observation that there were no discernible objective or subjective differences between manually and automatically setting the cue in Experiment 1, we chose to exclusively test the automatic cue in Experiment 2 due to its user convenience.Furthermore, we used the red arrow as cue again.The cue was effective in Experiment 1 and we deemed its salience and spatial memory support [13,77] even more important for the tangible task because of more background movement, differently colored backgrounds, and a presumably even larger demand on spatial memory due to the larger and three-dimensional working area compared to the computer-based task in Experiment 1.In Experiment 2, both the factor cue (automatic cue vs. no cue) and the factor interruption length (15 s vs. 45 s) were manipulated within-subject.Drawing upon the Memory for Goals theory and the results of Experiment 1, we expected that (1) the presence of an AR cue would result in a shorter resumption lag compared to no cue, and (2) long interruptions would cause longer resumption lags compared to short interruptions.Furthermore, we anticipated that (3) the negative impact of longer interruptions would be mitigated when an AR cue was presented.As Experiment 1, Experiment 2 has been preregistered (https://doi.org/10.17605/OSF.IO/KY9HZ).

Method
4.1.1Participants.A total of 38 undergraduate students participated in exchange for course credits or € 10.Participants were recruited via the same system as in Experiment 1 but participants from Experiment 1 were excluded from signing up for Experiment 2. No participant was excluded, and the final data set included 28 female and 10 male participants with an average age of 22.1 years (SD = 3.0).All participants had correct or corrected-to-normal vision.This research was approved by the institutional review board and informed written consent was obtained from each participant.
4.1.2Design.We used a 2 (cue) x 2 (interruption length) withindesign.The factor cue included the conditions automatic cue vs. no cue, and the factor interruption length included 15 s interruptions vs. 45 s interruptions.We conducted a prospective power analysis to determine the required number of participants.Based on the slightly below medium effect size of the main effect of the cue in Experiment 1, we estimated the effect size for cue to be f = .200(power = .80,alpha = .05,two-sided test).The power analysis indicated a sample size of 36 participants.To account for potential data loss, we included a slightly larger participant sample.4.1.3Procedure and Material.Experiment 2 adopted the same general procedure of training, execution and post-survey as Experiment 1 with a different primary task.For the primary task, participants engaged in a pill sorting task, for which they were required to transfer colored beads, symbolizing medical pills, from six distinct bowls into a 7-day pill box.The instructions, including which pills were supposed to be placed in each compartment, were presented in AR above the bowls to guide the participants throughout the task.Participants were instructed to individually pick up each pill using their dominant hand and sequentially place it into the pill box in the order provided by the instructions.Upon completing the pills for a given day, participants advanced the instructions to the subsequent day by interacting with an AR-button located alongside the instructions.Following the placement of pills for Sunday, the last day of the week, participants replaced the current pill box with an empty one.The setup and AR view are illustrated in Figures 5  and 6 below.Each day of the pill sorting task involved the placement of six to eight pills and featured either no interruptions or one or two interruptions, which resulted in a total of seven interruptions for each pill box.An interruption was triggered when a pill was placed in the pill box, indicated by both an acoustic signal and the disappearance of the instructions until the end of the interruption.The interruption task was the same as in Experiment 1.In the cue condition, a cue in the form of a red arrow was presented when the interruption ended, directing participants' attention to the specific bowl containing the next required pill color.The cue disappeared once the participant picked up the pill.We ensured that the color of the pill immediately following the interruption did not occur more than once during that day, so that the cue was unambiguous.Although participants were informed that a cue could occur, no cue was presented during the training week.To counterbalance the cue manipulation, the cue was introduced in the second and third weeks or in the fourth and fifth weeks of the experiment.A total of five pill boxes were used, with the initial box serving as a training session.
To measure the interaction interval, the resumption lag and possible resumption errors, we used the HoloLens hand tracking feature.Each of the six bowls and the pill box were placed in AR shapes that were invisible to participants but enabled us to track when a pill was taken out of a bowl and dropped in the pill box.This tracking also enabled us to provide feedback about the pill color that was picked up by showing a small round shape in the respective color at the fingertips of the participant.Furthermore, we could provide feedback if the wrong pill was about to be picked up by providing a short audio sound.The technical details of the tracking system are provided in the supplementary materials.

Analysis.
For the resumption lag analysis, only correct resumptions without resumption errors can be analyzed.Resumption errors occurred, when a participant reached into the pill box or any bowl, but the correct next one.A total of 1013 (95.2 %) interruptions were included in the resumption lag analysis.We performed a 2 (cue) × 2 (interruption length) repeated measures ANOVA to test the hypotheses.In all four experimental conditions, the resumption lag values were not normally distributed (Shapiro-Wilk; all p-values < .05).Analogous to the first experiment, we assumed that the ANOVA is robust to the violation of the normal distribution.Therefore, we report the results of the analysis of the untransformed data.

Results
A paired samples t-test showed that the time between the end of an interruption and the participant reaching into the next correct bowl (resumption lag M = 3804 ms, SD = 550 ms) was significantly longer than the time between sorting a pill and reaching into the next bowl when no interruption occurred (interruption interval, M = 1255 ms, SD = 304 ms), (t(37) = 27.862,p < .001,d = 4.520).
The main effect cue was significant, F(1,37) = 150.766,p < .001,partial η2 = .803(Figure 7 and Table 2).As expected, the resumption lags with no cues were longer than with cues.The main effect of interruption length was significant, F(1,37) = 76.689,p < .001,partial η2 = .675.As expected, long interruptions resulted in longer resumption lags than short interruptions.Finally, consistent with our hypothesis, there was a significant cue × interruption length interaction, F(1,37) = 60.104,p < .001,partial η2 = .619.A Bonferroniadjusted post-hoc analysis revealed that, when no cue was provided, the resumption lag was significantly longer after the 45 s interruption than after the 15s interruption (p < .001),whereas there was no difference between the 15 s and the 45 s condition when a cue was provided (p = 1.000).
As in Experiment 1, we compared the edit lag between the experimental conditions.For this comparison, we considered only the time between placing one pill into the pill box and picking up the next pill for the interaction intervals, excluding the time taken between picking up a pill and placing it, as placing a pill into the pill box does not require the participant to reorient themselves on the primary task.A repeated measures ANOVA showed no difference in the edit lag between the cueing conditions, F(1,37) = 0.165, p = .687,the interruption length, F(1,37) = 3.424, p = 0.072, and no interaction, F(1,37) = 0.821, p = .371.
The comparison of the resumption errors between the experimental conditions revealed significant main effects and interaction.Participants exhibited fewer resumption errors when presented with an AR cue compared to no cue, F(1,37) = 34.946,p < .001,partial η2 = .203.Following short interruptions, fewer resumption errors occurred, F(1,37) = 5.908, p = .020,partial η2 = .042.The observed significant interaction, F(1,37) = 8.963, p = .005,partial  η2 = .054,was further explored using a Bonferroni-adjusted posthoc analysis.This analysis revealed that in the absence of a cue, resumption errors were higher after a 45 s interruption than a 15 s interruption (p = .002),whereas with an AR cue, no such difference was observed (p = 1.000).

Discussion
The primary objectives of Experiment 2 were to assess the potential of AR cues in facilitating task resumption following interruptions and evaluate the mediating effect of interruption length on a tangible primary task.The results support the expected benefits of task resumption facilitated by AR cues.AR cues not only facilitated task resumption after both short and long interruptions, but also negated the adverse effects of increased interruption duration.All comparisons showed a large effect size for the AR cue condition.The resumption errors showed the same pattern of results as the resumption lag.The latter result appears to be due to the fact that almost no errors were made when the cue was presented.Similar to Experiment 1, we were able to demonstrate that the interruptions in this experiment were disruptive.However, the comparison between the resumption lag and the interaction interval warrants careful consideration due to the physical movement required for participants to transition between the interrupting task and the primary task.Due to the relatively small movement involved and given that the resumption lag exceeded the interaction interval by more than threefold, we still deem it plausible to characterize the interruptions in this experiment as disruptive.Additionally, we were able to demonstrate that the reorientation process occurs within the resumption lag, since the interaction directly after the task resumption was not prolonged.Consequently, our chosen primary task proves suitable for further research regarding interruptions and the implementation of visual cues to support task resumption.

GENERAL DISCUSSION
Through the two experiments, we successfully demonstrated the facilitative effects of AR cues on task resumption times and errors and these benefits did not translate to compromised performance in subsequent tasks steps (i.e., edit lag).Furthermore, these effects were observed in a computer-based and a tangible task.Finally, we observed that manually set cues by the user were not superior in relation to resumption times and errors than an automatically set cue by the system.Different to a previous head-mounted display study in which the mere presence of surrounding environmental cues due to the opaque or transparent smart glass design did not support task resumption times [52] or subtle cues on computer screens [17,18,90] our red AR arrow cue showed a positive effect on task resumption times.This result is in line with previous research on computer-based tasks that showed that salient non-AR cues that provide spatial information improved task resumption time [13,90].Going beyond previous research on cues to support task resumption after interruptions, we showed that the AR cues not only improved task resumption time but also reduced task resumption errors.This is noteworthy because in context of the Memory for Goals research, the focus is on task resumption times and effects on the resumption lag, while effects on resumption errors are seldomly reported [e.g.13,17,18,52,90].In addition, we showed the beneficial effects on resumption times errors as well as the unaffected edit lag in a computer-based task as well as in a tangible task.Previous research on tangible tasks were quantitative observational field studies in the hospital with no experimental control [25,34] or only investigated the resumption lag [62].Finally, while Wolff et al. [97] already indicated the potential use of mock-up AR cues in a Wizard of Oz approach, our experimental setup demonstrated the effect of AR cues with real-time augmentations.
One unexpected finding was related to the interruption length effect.Specifically, for the computer-based task, the AR cue only exhibited significant benefits after longer, more disruptive interruptions.In contrast, the AR cue for the tangible primary task exhibited very large effects on both short and long interruptions.The disparity related to interruption length may be attributed to differences in primary task difficulty.In Experiment 1, the modules had distinctive titles and were arranged in two rows on the screen, while in Experiment 2, the pills differed solely in color and were presented in a continuous row.This distinction may have rendered the to-beresumed task in Experiment 1 easier to remember, because spatial cues are known to enhance task resumption [76].Consequently, the lower difficulty of task resumption in Experiment 1 could have resulted in a floor effect in the resumption lag following short interruptions, thereby explaining the lack of additional advantage provided by the cue.
Alternatively, the disparity in results between the primary tasks could potentially be attributed to the presence of distinct resumption strategies.In the case of the computer-based task, participants' failure to recall the correct module after the interruption led to guessing, resulting in either a relatively swift task resumption or a resumption error.In contrast, the pill sorting task allowed participants to adopt a counting strategy by tallying the previously sorted pills in the pill box.However, in the no cue group, only four participants reported utilizing this counting approach during the experiment.Interestingly, 17 participants reported that they actively looked at and remembered the next required pill before placing the current one in the pill box.
Furthermore, our study reveals that the rehearsal acquired from manually placing the AR cue did not yield any significant advantages for task resumption.This was surprising, as we expected the increased time for encoding when placing the cue (i.e., interruption lag) to reduce the resumption lag [40] and frustration [44].The lack of a positive effect might stem from the unfamiliar hand gesture that users needed to perform correctly in order to place the cue.Performing the gesture might have caused excessive extraneous cognitive load, leaving not enough resources for the encoding of the contextual cues [88].Alternatively, this lack of difference could be explained by soft constraints [29].According to the Memory for Goals theory, the process of manually setting the cue should raise the activation level of the next goal and thereby facilitate task resumption.If participants perceive following the AR cue as less demanding than retrieving the goal from memory, they will opt for this choice, rendering the initial difference in goal activation irrelevant.A final explanation may stem from the repetitive nature of the hand gesture for placing the AR cue.Wilson and Emmorey [96] demonstrated that repetitive hand movements decreased working memory performance.In our task, this negative impact on working memory might have offset any potential positive effects on additional encoding of the primary task goal.However, users in our study only used the gesture a total of 14 times compared to one more than one hand signs per seconds in Wilson and Emmorey's study [96] which makes this explanation less likely.
Based on our performance and subjective mental workload results, an automatic cue would be preferable, as it seems to offer greater convenience for users with the same performance benefits and workload demands.However, it is crucial to consider the potential challenges linked to implementing an automatic cue system in actual work or everyday environments, such as technical complexities and compatibility issues.These challenges might necessitate the consideration of a manually set cue as a viable alternative in certain practical contexts.Future research could explore the possibility of combining both cue types to optimize task resumption in complex work scenarios.Finally, in both experiments, we found no prolonged time to execute the first action after task resumption (i.e., edit lag [71,102] with an AR cue compared to no cue.Within the constraints of null hypothesis testing, this result indicates that the participants reoriented and restabilized the task context during the resumption lag rather than passively following the AR cue and simply executing the indicated action.This is importation because if reorientation had merely been delayed, the advantage of a visual cue in facilitating task resumption would have been limited.In terms of the Memory for Goals theory [2], the AR cue appears to have facilitated task resumption.The resumed task step, in turn, aided in cueing the next task step through associated links.
Future research will need to investigate whether simple AR cues also work for more complex tasks such as resuming an assembly task in manufacturing [50].For more complex tasks, enhancing cues with additional information such as providing multiple next steps instead of only indicating the immediate next step might be necessary.Furthermore, in our case, supplementing the AR cue with pertinent details, such as the name of the medication being administered or the name of the patient, may prove beneficial for safety.Exploring the existing domain-specific task resumption strategies can guide the decision on what information to incorporate.
We now turn to the contribution of our results in relation to the larger HCI research on interruption and interruption coordination [65,82] and head-mounted display research in general.Our benefitable findings of AR cues on resumption times and errors and specifically the edit lag demonstrate that the cue-based approach is an effective strategy to compromise the detrimental effects of immediate interruptions and might be a viable alternative or addition to mediated or scheduled interruptions.That is, in our study, participants were effective in dealing with the interruption and the AR cue minimized detrimental effects of the interruption.The benefits of AR cues may be particularly pronounced for environments in which users are mobile and have many different primary tasks whereas the benefits might be less pronounced for computer-based office work.
Moreover, AR cues may also affect how users address interruptions with a negotiable starting point [64].For example, Bogunovich and Salvucci [10] showed that workload and time constraints affect negotiation strategies and participants delayed to attend interruptions during times of higher workload.The availability of AR cues may lead to users changing their strategy regarding the timing of interruptions.Users might become more willing to choose points of high mental load when a cue facilitates task resumption.Future research will need to address the how AR cues affect interruption negotiation strategies.
Furthermore, in our study, we focused on sequential multitasking, which has characterized by fewer task switches and longer periods of attention devoted to, most of the time, two tasks (i.e., primary and interrupting task).Concurrent multitasking includes more task switches and shorter actions on the different tasks.The latter situations are addressed in Salvucci and Taatgen [81] Threaded Cognition theory.The general idea of AR cues may also work in environment with multiple ongoing or paused tasks; however, such environments will pose technical challenges in relation to a tracking system for automatic cues and design challenges in relation to designing AR cues for multiple tasks.
Additionally, AR cues offer a potential alternative for managing interruptions in guided assembly tasks.Rather than relying on a smart interruption system that schedules interruptions at predefined task breakpoints [50], the task guidance system could enhance task resumption through the integration of an AR cue.Future research will need to compare the effectiveness of both approaches.
Beyond this, the use of digital cues may also be a viable solution for interruptions in virtual reality.In VR settings, bystanders often hesitate to interrupt users, perceiving interruptions of a VR user as especially disruptive and the user as unwilling to be disturbed [70].The knowledge that task resumption is facilitated in VR could potentially lower the inhibition to interrupt VR users, facilitating interactions between VR users and bystanders.
Finally, our study showed no disadvantage of placing an AR cue via a hand gesture in regards to mental load or task resumption.Since this approach requires only gesture recognition instead of permanent task tracking, it is a more practical solution for many work environments.Work environments, that require the user to be mobile, e.g.intensive care units [31], profit from the task-independent nature of the manually set cue.The manual setup also allows the use of less powerful AR glasses that would not be able to perform task tracking.

Limitation and Strength
Our study has limitations.First, even though we used a tangible and more complex task in Experiment 2 and in most of the resumption lag research [e.g. 12, 92], the task in the present study does not have the complexity inherit to those in actual work environments.However, by applying the Memory for Goals theory [2], our research benefits from a well-established cognitive framework that enhances the credibility of our findings.This theory-driven approach boosts the generalizability of our results, particularly when considering the impact of interruptions in more complex, dynamic environments.Studies in the emergency department and intensive care unit confirmed the effect of interruption length [24,34] and the general concept of environmental cueing [31].Moreover, the inclusion of a tangible task in Experiment 2 represents a significant strength of our study.While prior research on resumption lags has primarily focused on computer-based tasks [e.g.52, 73], the tangible task carried out in Experiment 2 may be more representative of tasks in interruption-rich domains such as healthcare [32] and aviation [63].
Second, the interruptions in both experiments were intentionally structured to allow for the indication of the next correct step through a simple arrow cue.However, actual work tasks may involve subtasks or items that may occur more than once within a task (e.g., two pills of the same color in our pill sorting tasks).These aspects must be considered when transferring our findings to practical work environments by, for example, using differently designed cues to inform of varying colors and shapes or providing additional information that can effectively support those carrying out recurring tasks.
Third, the cues in the experimental setup were 100% correct and therefore fully reliable, which might be impossible to achieve in field settings.Falsely recognized interruptions or a small percentage of incorrectly placed cues could reduce the users' trust in the cues and limit their usefulness [60].The manually set cues could circumvent this limitation until reliable ways for detecting interruptions and placing cues become available.Indeed, with the limitations of null hypothesis testing, our results show that manual cues had no disadvantages in terms of perceived workload or task resumption performance.
Fourth, interruptions in both experiments were positioned solely at discrete sub-task boundaries such as completing a module (Experiment 1) or dropping a pill in the box (Experiment 2).This may limit the transferability to everyday tasks in which interruptions may occur at any time including more cognitively demanding mid-task interruptions [7,66].However, research showed that users frequently avoid such mid-task interruptions in field studies [31,102] or if the study setting in the laboratory allows such discretionary behavior [7,93].Furthermore, in many tangible everyday tasks, users use artifacts that are associated with a task goal.Research showed that such artifacts make task resumption easy [31].In Experiment 2, for example, being interrupted when holding a pill in the hand and keeping the pill in the hand during the interruption might have acted as such an artifact.Finally, based on the Memory for Goals theory [91], our approach of cueing the location of the to-be-resumed task-step should also work for mid-task interruptions.However, such cues would require a more sophisticated tracking system and some tasks may lack clear sub-tasks boundaries at all [22,56].Future research is needed to investigate the effects of AR cues on the resumption of mid-task interruptions and more continuous tasks such as driving.

Design Implications
This study provides several design recommendations.First, AR cues should be displayed automatically as automatic cues help task resumption as well as manual cues while saving the user time.Furthermore, automatic cues are likely more convenient.Manually set cues might be a viable option when interruption detection or correct placement of the AR cue is not possible.Indeed, we found no drawbacks to manual cues in relation to task resumption performance (i.e., resumption lag, resumption error, subjective workload); manual cues reduce the computational effort while providing increased flexibility.
Second, simple cues such as salient red arrows are not only suitable for finding the to-be-resumed location [13,77] but also seem to be enough to recall and establish task context.Based on our results, the cue can be used for computer-based tasks and tangible tasks.For more complex tasks, the cue may also be enhanced with additional information.However, it has been argued that additional information or more complex cues might distract and diminish performance on the interrupting task if they are in the field of view of the user [68].When moving towards more specific tasks contexts and actual users such as nurses sorting medication in elderly homes in our case, cue design may become more specific, and designers can carefully evaluate the benefits of more complex cues.When working with actual users and tasks, we would also recommend using qualitative measures to assess the experience of users and their feedback on the appropriateness of the technology and implementation for the intended use.
Third, the incorporation of highly visible AR cues may serve as a preventive measure against users losing track of the to-be-resumed task entirely.It is possible to design AR cues in a manner that remains perceptible even when the user leaves the room during an interruption.This could be particularly helpful for addressing interruptions that occur in healthcare settings, as they often require staff to change location.Fong and Ratwani [25] showed that such interruptions can result in unfinished tasks.In addition to being advantageous for interruptions that require the user to leave the location of the primary task, these visible AR cues may benefit those who experience nested interruptions (i.e., interrupting tasks that are interrupted again).For unfinished tasks and nested interruptions, the red arrow may need to be replaced or enhanced by icons, that allow users to differentiate between the interrupted tasks.
Fourth, the second experiment demonstrated a new paradigm for task resumption research.The task paradigm enables the tracking of task resumption metrics errors, resumption lag and edit lag reliably.Subsequent experiments with this paradigm have the flexibility to adapt the difficulty levels of both the primary task and the interruption task to accommodate specific research needs.Additionally, experimental designs with multiple secondary tasks or nested interruptions could easily be implemented.
Fifth, due to the simplicity of the AR cue, it could be easily combined with AR task guidance systems.Task guidance via AR facilitates repair work [41,42].Since the red arrow is only presented temporarily, it could blend in well with the interface of a task guidance system.Safety-critical environments, such as aviation, where checklists are commonly employed to manage interruptions [20], could potentially benefit from the combination of AR task guidance and AR resumption cues.

CONCLUSION
Interruptions are unavoidable in many work domains and in particular immediate interruptions can result in unfinished tasks and contribute to errors [65].The presented research shows that AR cues can reduce resumption times and resumption errors and reinstate the task context when compared to the absence of cues.AR cues seem to be a viable solution to compromise the negative effects of immediate interruptions, especially in domains where tangible tasks or staff need to change location, such as healthcare [32], aviation [63] or manufacturing [50].Moreover, AR offers a more streamlined alternative to conventional task resumption methods, like relying on sticky notes [28] carrying artifacts such as a syringe [31] or building task-specific systems [84] or reminders.
Beyond work settings, translating our findings to virtual reality appears promising, as interruptions within virtual reality environments can significantly detract from the immersive experience, similar to the impact of interruptions in real-world settings [27].Exiting the virtual environment to attend to a real-world interruption may yield a level of disruption similar to that induced by interruptions that require the user to leave the physical workspace.The potential transferability of our insights underscores the relevance of our research in addressing the challenges posed by interruptions across virtual, augmented and actual reality.Finally, as head-mounted AR displays, such as the upcoming Apple Vision Pro, become more ubiquitous, user interactions with interruptions are likely to become increasingly frequent and potentially disruptive.Given its simplicity and versatility, the manual setting of cues offers a viable solution that could be integrated seamlessly into a multitude of everyday tasks, such as cooking, to improve user comfort and reduce the risk of forgetting.

Figure 1 :
Figure1: Screenshot of the primary task interface.Central to the interface is a table containing six columns of different categories regarding the monitoring of an anesthesia.The participants needed to transfer the information from each column into the corresponding surrounding modules.The primary task was developed by Kruse et al.[52].

Figure 2 :
Figure 2: Structure of interruption (top row) and timeline of cue conditions in Experiment 1 and 2. Acoustic signals indicatedthe start and the end of the interruptions along with task specific means to prevent a continuation of the primary task.In the manual cue condition of Experiment 1, the interruptions started after the participants marked a position on the primary task via a hand gesture.The resumption lag was measured from the end of the interruption, i.e., the tablet turning black after the last arithmetic task, to the first interaction with the respective primary task.The red AR cue was only present in the cue condition and the arrow was visible during the resumption lag.Note that we enhanced the outline of the AR cues for better visibility in the figure.

Figure 4 :
Figure 4: Results of the mixed ANOVA of Experiment 1.Both manual and automatic cues negate the negative effects of longer interruptions compared to no cue.Error bars represent the standard errors, since the standard error was too small to be displayed clearly.

Figure 5 :
Figure 5: The setup for Experiment 2 as seen from above.The tablet with the interruption task (left) was placed outside of the field of view during the primary task (right) but was easily reachable by turning around on the office chair.

Figure 6 :
Figure 6: The left figure shows the AR shapes that were used to track which pill was picked up and placed into the pill box to measure resumption performance metrics.The right figure shows the primary task from the participants' point of view.The instructions were superimposed by the HoloLens above the bowls with the pills.

Figure 7 :
Figure 7: Comparison of resumption lags between the cue type and interruption length conditions.The error bars represent the standard deviation, since the standard error was too small to be displayed clearly.
3.1.1Participants.A total of 87 undergraduate university students were recruited via the departments participant recruitment and study management system.Students participated in exchange for course credits or € 12.50.Four participants were excluded (two data sets were lost due to server issues, one participant did not follow the instructions and for another the gesture for setting the manual cue did not function correctly).The final data set consisted of 28 participants in the automatic cue group (22 female, average age 26.6 years), 27 participants in the manual cue group (17 female, average age 23.2 years) and 28 participants in the no cue group (20 female, average age 24.2 years).All participants had correct or correctedto-normal vision.This research was approved by the institutional review board, and informed written consent was obtained from each participant.

Table 1 :
Descriptive results of Experiment 1. Values indicate the M (SD) of the resumption lag in ms and the M (SD) of the resumption errors in percentages.

Table 2 :
Descriptive results of Experiment 2. Values indicate the M (SD) of the resumption lag in ms and the M (SD) of the resumption errors in percentages.