When more is less: Adding action effects to reduce crosstalk between concurrently performed tasks

Abstract

Dual-task costs are thought to reflect the architecture of the cognitive processes that guide voluntary action. Thus, manipulations that affect dual-task costs can provide insight into how we represent and select behavior as well as allow us to better design machines and controls for safer, more efficient performance. This line of research has revealed that the sensory events that follow the responses (i.e., action effects) can affect dual-task performance even though the sensory events occur after the actions are produced. The present study assessed three hypotheses regarding how action effects impact dual-task performance: a monitoring bottleneck, central stage shortening, and crosstalk. Across two experiments, we manipulated the content of two concurrently-performed tasks: a visual task that used either spatial or nonspatial stimuli (Experiment 1) and an auditory task that used responses with or without experimentally-induced auditory action effects (Experiments 1 and 2). In Experiment 1, dual-task costs were reduced when experimentally-induced auditory action effects were present, independent of the content of the visual task. In Experiment 2, the dual-task costs depended on the content of the experimentally-induced action effects, such that costs were larger when action effects emphasized ordinal (number) information, which overlapped with the unmanipulated action effects from the visual spatial task. Strikingly, dual-task costs were reduced when added, post-response events supported greater separation between task representations relative to when no post-response events were added. These results support the crosstalk hypothesis, as action effects appear to alter task representations so that they emphasize different types of information, reducing the degree of crosstalk.

1. Introduction

To examine the cognitive operations that give rise to voluntary behavior, cognitive psychologists study human multitasking, where individuals must perform two or more tasks close together in time. Such conditions typically lead to performance decrements (i.e., longer response time (RT) and higher error rate (ER)) compared to when each task is performed in isolation—a phenomenon referred to as dual-task costs (for reviews, see Koch, Poljac, Müller, & Kiesel, 2018; Pashler, 1994). Although the sources of dual-task costs are controversial, a common theme is that these costs arise at the level of central operations (i.e., the cognitive processes linking perception and action), either through a content-independent central bottleneck (e.g., Pashler, 1984, 1994; Welford, 1952) or through the interaction of content-dependent central codes that integrate stimulus and response information (e.g., Göthe et al., 2016; Hazeltine et al., 2006; Wickens, Sandry, & Vidulich, 1983).

1.1. Action effects and response selection.

If dual-task costs reflect interactions between the response selection processes for the respective tasks, manipulations affecting the magnitude of the costs should reveal the underlying architecture of the mechanisms supporting voluntary behavior and specify the ways that possible actions are encoded. One important theoretical consideration for interpreting dual-task costs and understanding response selection more generally is that responses appear to be represented in terms of their outcomes (for review, see Janczyk & Kunde, 2020). For instance, pianists select and evaluate their actions (i.e., pressing keys) based on the anticipated outcomes (i.e., sounds) of such actions. These anticipated outcomes—i.e., action effects—include the sensory consequences of an action.

In typical laboratory tasks, action effects may be the auditory feedback produced by vocal responses or spatial and tactile/proprioceptive feedback from manual responses. Such feedback, especially with regard to widely used manual button-presses, is impoverished compared to many real-world tasks that have greater sensory consequences (e.g., pressing a key to make a letter appear on a screen; turning a wheel to alter the direction of movement of a car). This may be worthy of consideration when applying laboratories studies to our understanding of everyday behaviors. Thus, to examine how the intention of changing the environment may affect the control of action, researchers add manipulated (i.e., experimentally-induced) action effects that consistently follow the production of responses to examine the influence of action effects on performance. These manipulated action effects consistently follow specific responses (e.g., a tone that consistently follows a key press), although they do not affect the unmanipulated effects associated with responses (e.g., auditory effects from speech). These added effects may make the tasks more similar to typical real-world actions that are performed to produce changes in the environment and can robustly affect response selection processes (for review, see Pfister, 2019).

The influence of such manipulated action effects on response selection has been demonstrated in numerous single-task studies focusing on response-effect (RE) compatibility, in which response production is slowed when responses are followed by response-incompatible effects (e.g., forceful key press, quiet tone effect) compared to response-compatible effects (e.g., forceful key press, loud tone effect; Kunde, 2001) (see also, Ansorge, 2002; Kunde, 2003; Pfister et al., 2014). Critically, the influence of action effects on performance must originate from the anticipation of these action effects because they are presented after response production.

The observation that response selection is affected by post-response action effects is consistent with ideomotor theory, which assumes that humans select and evaluate their actions based on the intended or anticipated body/environmental effects of the action (for reviews, see Pfister, 2019; Shin et al., 2010). The core tenet of ideomotor theory is that when an action is performed, the response-related effects become associated with the motor response, so that they are integrated in a single representation. Thus, after the repeated co-occurrence of an action and its corresponding effect, a bidirectional link is established, so that activating the mental representation of the effect activates the corresponding motor program (for evidence that ideomotor-like effects may be driven by propositional knowledge of causal relations, see Sun et al., 2022). In this way, actions and their associated effects are coded in a common format (see, Frings et al., 2020; Hommel et al., 2001; Prinz, 1990). This account was developed to explain findings relating to stimulus-response (SR) compatibility in single-task conditions (e.g., Hommel, 1996) and special cases where dual-task costs are not observed (e.g., Greenwald, 2003; Greenwald & Shulman, 1973; but see Halvorson & Hazeltine, 2015, 2019). However, the broader point that response selection engages representations that include the intended action effects has far-reaching implications for dual-task performance (see, Janczyk & Kunde, 2020).

1.2. Action effects and dual-task performance

During dual-task performance, multiple actions are produced, so there are multiple action effects which must be bound with the appropriate responses. This may cause crosstalk between tasks that impairs dual-task performance. For instance, the backward crosstalk effect in dual-task performance (e.g., Hommel, 1998) depends on the relationship between the response features (e.g., spatial) of the first task and the action effects of the second task. When the second response produces an action effect that is incompatible with features of the first response (e.g., Task 1: right key press response; Task 2: left-sided light effect), performance on the first task is impaired (Janczyk et al., 2014). Likewise, when a mental rotation task is performed at the same time as a manual rotation task that entails a visual rotation action effect, performance is facilitated when the directions of the mental rotation and the visual action effect, but not the manual rotation itself, are the same (Janczyk et al., 2012). Thus, interactions between action effects may be a key determinant of the efficiency of dual-task coordination.

For example, interference between task representations that include action effects provides a possible account of the modality compatibility effect. The modality compatibility effect refers to changes in dual-task costs associated with the pairings of stimulus and response (and effect) modalities (Stephan & Koch, 2011). Specifically, dual-task costs are smaller when a visual-manual task is paired with an auditory-vocal task than when a visual-vocal task is paired with an auditory-manual task (Hazeltine et al., 2006). To test the role of action effects in the modality compatibility effect, we (Schacherer & Hazeltine, 2020) varied the compatibility between the modalities of stimuli and manipulated action effects, such that they were either compatible (e.g., visual stimulus, visual effect) or incompatible (e.g., visual stimulus, auditory effect), while holding the SR mappings constant. We also included an unmanipulated effect condition (i.e., no manipulated action effects provided) to assess how additional sensory consequences affected dual-task costs.

We observed smaller dual-task costs when stimuli and action effects were modality-compatible compared to when they were modality-incompatible. Although the SR mappings were identical across conditions, when similar stimulus and effect codes were present within a task (e.g., visual stimulus mapped to a visual effect; modality-compatible), central operations could easily bind the appropriate stimulus to the appropriate effect on dual-task trials, reducing between-task interactions, consistent with the proposal that action effects are integrated into task representation (e.g., Hommel, 1996). Moreover, in two experiments using bimanual tasks, dual-task costs were smaller when responses produced modality-compatible action effects compared to when responses produced no manipulated action effects. In sum, anticipating (modality-compatible) manipulated action effects reduced dual-task costs compared to anticipating only the unmanipulated action effects of producing manual responses (i.e., the spatial or tactile/proprioceptive feedback associated with a finger key press).

To explain the unexpected finding that the manipulated action effects reduced dual-task costs relative to when no action effects were added in the experiments with two manual responses, we proposed that participants receiving manipulated action effects may have emphasized the visual and auditory aspects of the two responses, rather than their shared manual component. This, in turn, may have reduced the degree of crosstalk between tasks, resulting in smaller dual-task costs. That is, without the manipulated action effects, responses for both tasks were primarily represented in terms of spatially differentiated tactile and proprioceptive effects, which would produce substantial crosstalk. With the manipulated action effects added, the visual task’s responses may have been represented in terms of visual action effects and the auditory task’s responses may have been represented in terms of auditory action effects. Thus, there would be less crosstalk between response representations, leading to smaller dual-task costs, consistent with the proposal that the relationship between action effects influences dual-task performance (Janczyk et al., 2012; Janczyk & Kunde, 2020).

1.3. How do action effects influence dual-task performance?

These findings demonstrate that action effects’ influence stems from the representation of the action during response selection. As such, action effects may affect dual-task costs in several ways. One possibility is that the compatibility between stimuli, responses, and action effects may shorten response selection processes. When stimulus and response codes are compatible (i.e., the task is SR compatible), the presentation of a particular stimulus is proposed to automatically activate the associated response (e.g., Kornblum, Hasbroucq, & Osman, 1990). Likewise, when response and effect codes are compatible (i.e., the task is RE compatible), anticipating a particular effect is proposed to activate the response producing that effect (e.g., Kunde, 2001; Pfister, 2019). With SR or RE compatibility, the time needed to retrieve stimulus-response or response-effect bindings is reduced, shortening the time needed to select a response. Shortening response selection processes, in turn, may reduce temporal overlap between the central operations for the two tasks, decreasing dual-task costs (stage shortening hypothesis) (Dux et al., 2009; Ruthruff et al., 2006; Strobach et al., 2013). If so, then action effects should produce reductions in single-task RTs along with the reductions in dual-task costs because the central operations are shortened. Note that Schacherer and Hazeltine (2020) observed that action effects altered dual-task costs even when they did not affect single-task RTs, inconsistent with stage shortening hypothesis. However, manipulations of action effects were not specifically compared to manipulations known to affect central processes. Moreover, the action effects were manipulated for both tasks simultaneously, making the locus of the effect difficult to determine.

Second, action effects may affect dual-task costs by adding monitoring requirements to processes shared by the two tasks. Specifically, monitoring the effects of one response may occupy central processing, and only once the consequences of that action have been evaluated can the selection of a second response proceed (monitoring hypothesis) (e.g., Jentzsch, Leuthold, & Ulrich, 2007; Kunde, Wirth, & Janczyk, 2018; Schaaf, Kunde, & Wirth, 2022; Ulrich et al., 2006; Welford, 1952; Wirth, Janczyk, & Kunde, 2018; Wirth & Kunde, 2020; Wirth, Pfister, Janczyk, & Kunde, 2015; but see, Bratzke, Rolke, & Ulrich, 2009). That is, having to select and monitor two action effects at the same time may pose high demands on central operations, producing dual-task costs (for review, see Janczyk & Kunde, 2020; see also, Janczyk et al., 2012, 2014). In this case, single-task RTs will not be affected because the monitoring process takes place after the response is produced.

The monitoring account does not explain the impact of action effects in single-task procedures, but there is evidence for its role in dual-task costs from experiments that manipulate the delay between a response and the appearance of a manipulated visual action effect. When performing two tasks sequentially and the first task involves a relatively long-lasting or delayed action effect, the selection of a response for the second task is slowed compared to when the first task involves a short-duration, undelayed action effect (e.g., Kunde, Wirth, & Janczyk, 2018; Ulrich et al., 2006). Thus, this account supposes that central operations include processes that monitor action effects, and only once this monitoring process has finished can a second response be selected (see also, Welford, 1952; Wirth & Kunde, 2020; Wirth et al., 2018). Schacherer and Hazeltine (2020) observed that in one of their experiments, the addition of manipulated action effects reduced dual-task costs compared to when no action effects were added (without differences in the single-task RTs), which is inconsistent with the monitoring account. However, this difference was only observed in a post-hoc comparison and action effects were added to both tasks, possibly changing the relative timing of competition for central operations (see Anderson et al., 2005).

Third, Schacherer and Hazeltine (2020) proposed that action effects affect central processing because they are integrated into the representations of the actions (crosstalk hypothesis). That is, all task-relevant features—e.g., stimulus, response, action effect—are bound within a shared representational format (e.g., Frings et al., 2020; Hommel et al., 2001; Schumacher & Hazeltine, 2016). Thus, interference arises between the central operations (which include representations of both stimuli and action effects) when the representations associated with one task overlap with the representations associated with the other task (e.g., when the stimuli and effects for both tasks include visuospatial information). Critically, this interference does not stem from the shortening of central stages (as in the stage shortening hypothesis) or from the requirement to evaluate action effects (as in the monitoring hypothesis). Instead, there is increased interference between the two tasks because of overlap between the contents of the representations associated with each task (Göthe et al., 2016; Halvorson & Hazeltine, 2015; Hazeltine et al., 2006; Schacherer & Hazeltine, 2021).

1.4. The present study

In the present study, we evaluated the stage shortening, monitoring bottleneck, and crosstalk hypotheses of how the consequences of actions affect multitasking performance. We tested predictions of these three accounts using a dual-task paradigm with tasks with distinct stimulus modalities (visual and auditory) but with the same response modalities (two manual responses). We chose to use two manual responses so that the responses for the two tasks should have overlapping codes, especially when there are no manipulated action effects. Thus, changing the representation of one task by adding manipulated action effects might reduce overlap across tasks, thereby reducing dual-task costs. This also eliminated the concern that participants would alter the timing of their vocal responses to avoid interference with auditory action effects.

We compared the effects of adding action affects to a manipulation of SR compatibility, which is known to affect central operations (Kornblum et al., 1990), across four groups of participants in a 2 × 2 design. The action effects were manipulated in the auditory task, in which manipulated action effects were presented following response production for two groups of participants (manipulated effect conditions) but not for the other two groups (unmanipulated effect conditions). In the unmanipulated action effect conditions, the task was to press a key in response to hearing a spoken word. In the manipulated action effect conditions, the task was to produce an action effect by pressing a key in response to hearing a spoken word. That is, the motor responses and SR mappings were identical across conditions¹; only the action effects and intended goals of the responses differed.

We predicted that by adding manipulated action effects, we would reduce crosstalk between the two tasks. That is, if the unmanipulated visual task representation emphasizes the spatial or tactile/proprioceptive action effects of the manual responses and the manipulated auditory task representation emphasizes auditory information, then responses may be coded so as to produce less crosstalk between tasks (e.g., Mechsner et al., 2001) than when both tasks are unmanipulated and their responses represented with spatial or tactile/proprioceptive action effects. Thus, this conceptualization of the source of dual-task costs draws on the bimanual coordination literature (e.g., Heuer, 1995) that highlights the role of goals over motor parameters in shaping the interactions between ongoing operations (Hazeltine, 2005). Because we manipulate the action effects of a single task, only the operations of that task are affected, making the results easier to interpret. That is, we can attribute changes in the dual-task costs specifically to the manipulation of a single task’s action effects.

We compared the manipulation of action effects to another manipulation known to affect central processes: the difficulty of SR translation (e.g., Kornblum et al., 1990). Therefore, we investigated whether task difficulty imposed by different SR rules influences the magnitude of interference between concurrently performed tasks with manual responses. To this end, we manipulated task difficulty by using two different sets of visual stimuli: spatial and nonspatial. If stimuli and responses are spatially compatible, task difficulty is presumably low, as such stimuli are theorized to automatically activate the spatially corresponding responses (Kornblum et al., 1990). If, however, stimuli and responses do not exist within a shared dimensional space (e.g., color stimuli mapped to spatial responses), task difficulty is presumably high, lengthening RTs.

2. Experiment 1

In Experiment 1, we manipulated the action effects for the auditory task and the difficulty of response selection for the visual task to determine whether these manipulations would lead to reductions in dual-task costs. Our primary goal was to assess how action effects alter dual-task performance—either via a monitoring bottleneck, shortened central stages, or crosstalk between central codes. For the action effect manipulation, in the manipulated effect condition, participants were instructed to produce the manipulated action effect that corresponded to the presented stimulus for the auditory task (e.g., “When you hear the spoken word mish, produce the spoken word mash. Make the spoken word mash occur by pressing the U key.”). In other words, participants were instructed to “complete the phrase” by pressing the key that produced the spoken word action effect that related to the spoken word auditory stimulus. In the unmanipulated effect condition, participants were instructed to respond to the auditory stimulus with the correct manual response (e.g., “When you hear the spoken word mish, press the U key.”). The same SR pairings were used as in the manipulated effect conditions, but no added action effect was present. Moreover, by adding action effects for only one of the tasks, we eliminated the possibility that participants slowed responses to avoid producing interference between action effects and the stimuli preceding those effects. That is, because the auditory action effect is only produced to auditory stimuli, there is no concern about crosstalk between stimuli and action effects (for discussion of this concern, see Experiment 3 of Schacherer & Hazeltine, 2020).

In the visual task, participants also responded with manual key presses. The use of key presses for both tasks is similar to the procedure used by Kunde et al. (2018). No action effects were added but the stimuli were varied so that the SR mappings were either spatially-compatible (spatial stimuli) or arbitrary colors (nonspatial stimuli). Both the action effects for the auditory task and the stimuli for the visual task were manipulated between participants, forming a 2 × 2 design with four groups of participants, as shown in Figure 1.

Figure 1.

Example task pairings for all four conditions in Experiment 1. For the visual task, stimuli were either Xs (e.g., X - - -) using a compatible spatial SR mapping or centrally presented colored circles (e.g., red, blue, green, yellow). For the auditory task, stimuli were spoken words (e.g., mish, riff, tic, zip) and were followed by unmanipulated or manipulated (e.g., mash, raff, tac, zap) action effects. See online for colored version of this figure.

For the action effect manipulation of the auditory task, the monitoring bottleneck hypothesis predicts larger dual-task costs in conditions with the manipulated effect, as a second response presumably cannot be selected until the manipulated effect has been monitored and evaluated. The stage shortening hypothesis predicts that any reduction in dual-task costs associated with the presentation of auditory action effects will be matched by reductions in single-task RTs in the auditory task. The crosstalk hypothesis predicts smaller dual-task costs in conditions with the manipulated effect because the manipulated effect may change how the task is mentally represented, reducing overlap and cross-task interactions between the two tasks. However, single-task RTs should not be reduced by manipulated action effects. We included the SR mapping manipulation of the visual task to compare to the action effect manipulation. This well-established manipulation that affects the duration of central processes should affect both single-task RTs (Kornblum et al., 1990) and dual-task costs (Halvorson & Hazeltine, 2015; Schumacher et al., 2001).

Of note, we did not expect to observe the complete elimination of dual-task costs in any of our conditions because 1) there still exists overlap in response modalities (both manual responses) and 2) except under special circumstances (e.g., Greenwald & Shulman, 1973; Halvorson & Hazeltine, 2015; Lyphout-Spitz, Maquestiaux, & Ruthruff, 2022), costs are generally present with limited practice (e.g., a single 1-hour session).

2.1. Method

2.1.1. Participants.

A power analysis was conducted in advance of data collection based on the effect size of ƞ² = .101 for the RT main effect of condition on dual-task costs in Experiment 3 of Schacherer and Hazeltine (2020), in which participants performed VM and AM tasks with or without added action effects in a between-subjects design. Using this effect size, we used G*Power (Faul, Erdfelder, Lang, & Buchner, 2007) to calculate the a priori sample size needed to obtain statistical power at a level of .8 (Cohen, 1988). The analysis indicated that 108 participants (27 for each of our four conditions) would be needed to obtain 80% power. We tested 32 participants per condition (128 total).

One hundred and forty-nine students from the University of Iowa participated in partial fulfillment of an introductory psychology course requirement. Data from 21 participants whose overall accuracy was less than 80% were discarded and not analyzed, leaving 128 total participants whose data were analyzed. These 128 participants were equally divided (32 per condition) into four conditions: nonspatial/manipulated effect (NS-M; 26 female, M_age = 18.50, SD_age = 0.62); nonspatial/unmanipulated effect (NS-U; 24 female, M_age = 18.63, SD_age = 1.21); spatial/manipulated effect (S-M; 26 female, M_age = 18.38, SD_age = 0.61); or spatial/unmanipulated effect (S-U; 27 female, M_age = 18.66, SD_age = 1.21). Vision and hearing were reported as normal or corrected-to-normal. Verbal consent was obtained prior to the experiment. All methods and procedures were approved by the Institutional Review Board at the University of Iowa.

2.1.2. Stimuli and apparatus.

The experiment was conducted using Microsoft VisualBasic software (version 15.0). For the spatial visual task (S), stimuli were X - - -, - X - -, - - X -, and - - - X, indicating the four possible responses for the left hand in a spatially compatible manner (e.g., - - - X required a left index finger press). For the nonspatial visual task (NS), stimuli were one of four colored circles (blue, green, red, yellow). For both visual tasks, stimuli were centrally presented within a 6.6° horizontal x 6.6° vertical black colored square on a black background. Stimuli were presented for 350 ms on a 19-inch computer monitor located approximately 57 cm from the participant. Manual responses for the visual task were the q/w/e/r keys on a standard QWERTY keyboard.

For both the manipulated and unmanipulated auditory tasks, stimuli were the spoken words mish, riff, tic, or zip. For the manipulated action effect condition (M), responses for the auditory task were followed by manipulated action effects, which were the spoken words mash, raff, tac, or zap, presented for 350 ms. Thus, for each stimulus, there was a corresponding action effect (mish-mash, riff-raff, tic-tac, zip-zap). Each action effect was contingent on the response produced, not the stimulus. For instance, if the stimulus was mish, but the participant pressed the key corresponding to the raff effect, they would hear mish-raff. This stimulus-effect semantic correspondence acted as its own form of corrective feedback. For the unmanipulated action effect task (U), participants performed the task with no additional action effect. Instead, a black screen was presented for the same duration (i.e., 350 ms) as the action effect in the manipulated effect task. Manual responses for the auditory task were the u/i/o/p keys on a standard QWERTY keyboard.

2.1.3. Design and procedure.

Four groups of participants formed a 2 × 2 design, with one factor representing the visual task (spatial, nonspatial) and the other factor representing the auditory task (manipulated, unmanipulated action effects). Thus, one group performed a spatial visual task paired with a manipulated-effect auditory task (S-M); one performed a spatial visual task paired with an unmanipulated-effect auditory task (S-U); one performed a nonspatial visual task paired with a manipulated-effect auditory task (NS-M); and the last group performed a nonspatial visual task with an unmanipulated-effect auditory task (NS-U) (Figure 1).

Verbal and written instructions were provided at the start of the experiment and additional written instructions were provided on the computer prior to the start of each block. Instructions emphasized both speed and accuracy across all block types. At the end of each block, participants were shown their overall accuracy and mean RT. Completion of the entire experiment took between 45 and 60 minutes.

All participants completed 18 blocks of 36 trials each. The first 10 blocks were homogenous single-task blocks, in which stimuli were presented indicating the appropriate response. These blocks alternated between visual and auditory tasks. The final 8 blocks were divided into pairs of alternating OR and AND blocks. The OR blocks consisted of 36 trials of either task (18 trials of each task) intermixed at random. The AND blocks consisted of 36 trials in which both stimuli were presented simultaneously and two responses were required on each trial. No explicit instructions were given regarding how to prioritize the tasks in AND blocks. The order of the OR and AND blocks was counterbalanced across participants. Participants were instructed to respond as quickly and accurately as possible.

For all tasks, each trial began with the onset of a fixation cross for 500 ms, followed by the presentation of the stimulus for up to 350 ms and a response interval that lasted up to 3000 ms. For the manipulated effect auditory task, the action effect was presented for 350 ms immediately following response production. In all other tasks, a blank field was presented for 350 ms. Therefore, the trial duration for all tasks was identical. The next trial began 500 ms after the completion of the feedback or 3000 ms response window. No error feedback was given when the response was incorrect.

For the manipulated action effect auditory task (M), instructions were organized as follows: First, participants were instructed to produce the manipulated effect that corresponded to the presented stimulus (e.g., “When you hear ‘MISH’, make ‘MASH’ occur”). Then, participants were provided instructions for linking the manipulated effect to its respective response (e.g., “To make ‘MASH’ occur, press ‘u’”). For the remaining tasks—spatial (S)/nonspatial (NS) visual task and the unmanipulated auditory task (U)—instructions were organized as such: “When you see/hear [STIMULUS], press [RESPONSE]”. Instructions were similar across both experiments.

2.1.4. Statistical analysis.

Across all conditions, the first two blocks (two homogenous single-task) were excluded from analysis and treated as practice. Thus, we analyzed data from 16 blocks (four visual single-task, four auditory single-task, four OR, four AND). Additionally, the first four trials in each block were excluded from analysis. All responses given within the first 200 ms after stimulus onset or any RTs greater than 2000 ms were excluded from analysis. Lastly, trials in which no response was detected, in which one or both responses were incorrect, and trials following an error were removed from analysis of RT. Of the analyzed blocks, we removed 15.9% of trials from our final analysis.

2.2. Results

2.2.1. Planned analyses.

We examined two dependent variables—RT and ER—and conducted two primary analyses for single-task performance and dual-task costs. The mean RTs and ERs for all conditions are shown in Table 1. In addition to null hypothesis significance testing, we also report Bayes factors (Lee & Wagenmakers, 2013) obtained using JASP (version 0.16.4).

Table 1.

Mean response times (RT) and error rates (ER) for each of the trial types in Experiment 1. Dual-task costs were calculated as the difference between the sum of RT/ERs in AND blocks minus the sum of RT/ERs in OR blocks. S-U = spatial-unmanipulated; S-M = spatial-manipulated; NS-U = nonspatial-unmanipulated; NS-M = nonspatial-manipulated; VM = visual-manual; AM = auditory-manual.

Condition	Trial Type:	Response Time (ms)	Error Rate
Spatial-Unmanipulated (S-U)	VM-Single	526	3.0
	VM-OR	682	3.3
	VM-AND	1209	5.9
	AM-Single	839	3.5
	AM-OR	996	2.6
	AM-AND	1307	8.2
	Dual-task Costs	838	8.2
Spatial-Manipulated (S-M)	VM-Single	541	3.1
	VM-OR	648	3.0
	VM-AND	1054	7.0
	AM-Single	779	5.5
	AM-OR	920	2.8
	AM-AND	1202	10.4
	Dual-task Costs	689	11.5
Nonspatial-Unmanipulated (NS-U)	VM-Single	719	4.1
	VM-OR	854	3.9
	VM-AND	1365	10.5
	AM-Single	834	3.7
	AM-OR	994	2.8
	AM-AND	1406	12.5
	Dual-task Costs	923	16.3
Nonspatial-Manipulated (NS-M)	VM-Single	698	3.9
	VM-OR	854	3.1
	VM-AND	1244	7.4
	AM-Single	787	5.2
	AM-OR	944	3.1
	AM-AND	1392	9.6
	Dual-task Costs	839	10.8

2.2.2. Single-task RTs.

To examine how the composition of the tasks affected single-task RT and ER, we conducted separate independent-samples t-tests on the manipulation associated with each task. For visual task RT, there was a significant effect of task, t(126) = 7.87, p < .001, d = 1.39, BF₁₀ = 4.43*10⁹, indicating smaller RT for the spatial task (535 ms) compared to the nonspatial task (716 ms). For ER, there were no differences across tasks (spatial: 3.0; nonspatial: 4.0), t(126) = 1.43, p = .155, d = 0.25, BF₁₀ = 0.48. For auditory task RT, there was a significant effect of task, t(126) = 2.14, p = .034, d = 0.38, BF₁₀ = 1.49, indicating smaller RT for the manipulated effect task (783 ms) compared to the unmanipulated effect (837 ms) task (Figure 2). For ER, there were differences across the tasks (manipulated: 5.4; unmanipulated: 3.6), t(126) = 2.23, p = .028, d = 0.39, BF₁₀ = 1.76, suggesting that for the auditory task, RT results may be affected by a speed-accuracy trade-off.

Figure 2.

Single-task response times (left, center) and dual-task costs (right) for each of the four conditions in Experiment 1. White bars represent the spatial-unmanipulated condition (S-U); light gray bars represent the spatial-manipulated condition (S-M); dark gray bars represent the nonspatial-unmanipulated condition (NS-U); black bars represent the nonspatial-manipulated condition (NS-M). Dual-task costs represent performance differences between AND-OR blocks. Errors bars represent standard error of the mean. VM = visual-manual; AM = auditory-manual.

2.2.3. Dual-task costs.

Because we were interested in examining the overall effect of responding to two tasks simultaneously, we summed the dual-task costs for both tasks for both RT and ER (e.g., Halvorson & Hazeltine, 2015, 2019; Hazeltine et al., 2006; Schacherer & Hazeltine, 2020). For our analysis of dual-task costs, we conducted a 2 × 2 univariate ANOVA with visual task (spatial vs. nonspatial) and auditory task (manipulated vs. unmanipulated action effect) as factors, with dual-task cost RT and ER as our dependent measures.

For RT, we observed a significant main effect of visual task, in which dual-task costs were significantly smaller in the spatial task (764 ms) compared to the nonspatial task (881 ms), F(1, 124) = 6.05, p = .015, pƞ² = .047, BF₁₀ = 2.89. Likewise, we observed a significant main effect of auditory action effects, in which dual-task costs were smaller for the manipulated effect task (764 ms) than the unmanipulated effect task (880 ms) task, F(1, 124) = 5.99, p = .016, pƞ² = .046, BF₁₀ = 2.81. That is, adding action effects decreases dual-task costs, regardless of the composition of the visual task. The interaction was not significant (NS-U: 923 ms; NS-M: 839 ms; S-U: 838 ms; S-M: 689 ms), F(1, 124) = 0.46, p = .498, pƞ² = .004, BF₁₀ = 0.30 (Figure 2). In sum, our manipulations, both of which affected single-task RT, also produced additive effects on dual-task costs.

For ER, we observed a significant main effect of visual task, with smaller ER in the spatial task (9.9) compared to the nonspatial task (13.6), F(1, 124) = 5.82, p = .0.17, pƞ² = .045, BF₁₀ = 2.25. There was no effect of auditory task on ER (manipulated: 11.2; unmanipulated: 12.3), F(1, 124) = 0.48, p = .504, pƞ² = .004, BF₁₀ = 0.23. Lastly, the interaction was significant, F(1, 124) = 8.12, p = .005, pƞ² = .061, BF₁₀ = 7.97. Follow-up tests for the two visual tasks revealed contrasting patterns in ER. When paired with the spatial task, the manipulated effect task numerically increased error rate relative to the unmanipulated effect task (S-M: 11.5; S-U: 8.2), t(62) = 1.69, p = .097, d = 0.42, BF₁₀ = 0.84. In contrast, when paired with the nonspatial task, the manipulated effect task significantly decreased error rate relative to the unmanipulated effect task (NS-M: 10.8; NS-U: 16.3), t(62) = 2.30, p = .025, d = 0.58, BF₁₀ = 2.31. This runs counter to the (non-significant) pattern observed in RT, in which the manipulated effect task reduced RT, regardless of whether it was paired with the spatial or nonspatial visual task.

In sum, ER was largest in the NS-U condition, suggesting that when tasks overlap on multiple dimensions (stimuli: verbal codes; effects: unmanipulated feedback from manual responses), ER is increased. Note that although the interaction was not significant for RT, the pattern was similar with the NS-U condition showing the longest RTs.

2.3. Discussion

The addition of a manipulated action effect reduced dual-task costs when the auditory task was paired with either a spatial or nonspatial visual task, suggesting that differences in dual-task costs were not associated with the difficulty of the visual task, but instead reflect interactions between the central processes for the two tasks. Unexpectedly, the manipulated action effects also reduced single-task RTs. To our knowledge, this is the first demonstration that correspondence between the stimuli and events occurring after the production of the response can shorten the duration of central operations, in the absence of either stimulus-response or response-effect compatibility. As expected, the manipulation of the visual task stimuli, which were either spatially compatible with the responses or arbitrary, produced similar effects on both single-task performance and dual-task costs. We relate these findings from the action effect manipulation to the three proposed hypotheses regarding how action effects may affect dual-task performance: a monitoring bottleneck, crosstalk, and central stage shortening.

2.3.1. Monitoring bottleneck.

The findings related to the action effects are inconsistent with the monitoring hypothesis because the manipulated action effect should require additional time to evaluate and monitor. That is, if central operations are engaged during the monitoring of action effects, having to monitor a later occurring (i.e., manipulated) action effect should have produced greater interference than having to monitor a shorter (i.e., unmanipulated) action effect (e.g., Kunde et al., 2018; Ulrich et al., 2006). Moreover, the RTs for the auditory task with the manipulated action effects tended to be longer than the RTs for the visual task, making it unlikely that dual-task costs increased because central operations for the visual task were delayed by processing the action effects for the auditory task.

It is possible that adding the action effects shortened the monitoring processes associated with the two tasks. Support for this proposal comes from studies demonstrating that participants apply less force when actions are followed by manipulated auditory effects compared to when manipulated effects are absent (Cao, Kunde, & Haendel, 2020; Horváth, Bíró, & Neszmélyi, 2018; Neszmélyi & Horváth, 2018), suggesting that action effects shape the motor characteristics of an action. Although the perception of unmanipulated tactile/proprioceptive effects may signal the successful completion of an action, manipulated auditory effects may provide more reliable or salient feedback regarding action success, resulting in more efficient monitoring. This would allow the second response to be selected more quickly, reducing dual-task costs. However, the response for the auditory task was typically produced after the response for the visual task, making this account less tenable. Nonetheless, we test whether the auditory feedback reduces dual-task costs by shortening the monitoring time in Experiment 2.

2.3.2. Crosstalk.

One possibility is that the smaller costs may reflect decreased crosstalk between the central operations for the two tasks when there are manipulated action effects for the auditory task. Without manipulated action effects, the central codes for the auditory task may emphasize spatial and tactile/proprioceptive effects from the manual responses, and the simultaneous activation of similar action effect information in the visual task could create crosstalk, increasing dual-task costs. In contrast, if the representation of the responses for the auditory task emphasizes the auditory action effects in the manipulated action effect conditions, this crosstalk would be reduced because the central codes for the two tasks were more distinct. While the dual-task costs are consistent with this account, it does not explain the advantage for the manipulated action effect conditions in single-task performance, where presumably there should be no crosstalk between tasks.

2.3.3. Stage shortening.

Alternatively, the reductions in dual-task costs associated with the manipulated action effects may stem from central processes being shortened. Shortening the duration of central processes for one or both tasks diminishes the potential overlap in selection processes, reducing dual-task costs (e.g., Dux et al., 2009; Ruthruff et al., 2006; Strobach et al., 2013). For the auditory task, single-task RTs and dual-task costs may have been smaller in the manipulated effect conditions because of shortened central operations. This suggests that response selection is affected by the compatibility between stimuli and action effects. When responses produced an action effect that formed a complete phrase with the stimulus (e.g., “mish-mash”), RTs were faster relative to when responses did not produce action effects that formed a complete phrase. That is, the addition of a semantically-compatible action effect made this arbitrary SR mapping compatible.

The effects of SE compatibility in the present experiment appear similar to SR in the present experiment (see also, Fitts & Deininger, 1954; Kornblum et al., 1990) and RE compatibility (e.g., Kunde, 2001). However, to our knowledge, SE compatibility, in the absence of SR and/or RE compatibility, has thus far not been reported (see Moeller, Pfister, Kunde, & Frings, 2019). However, this novel finding is generally consistent with theoretical accounts that hold that response selection involves binding of episodes that include features of the stimulus, responses, and intentions. Further examination into SE compatibility effects may provide insight into how task representations are structured (Frings et al., 2020; Hommel, 2004; Hommel et al., 2001). Here we note that the finding suggests that arbitrary SR mappings can be made compatible by adding action effects to the responses.

For the visual task, the findings were as expected. The single-task RTs and dual-task costs were smaller in conditions with the spatial task because the spatial task had greater SR compatibility (Fitts & Deininger, 1954), which shortens central operations (McCann & Johnston, 1992; Ruthruff et al., 2006).

3. Experiment 2

The results from Experiment 1 indicate that adding action effects can shorten central operations, which in turn can shorten dual-task costs. The finding that compatibility between the stimuli and the manipulated action effects can shorten central operations is novel and indicate that this may be one way by which action effects can reduce dual-task costs. Nonetheless, manipulated action effects have been shown to affect dual-task costs without affecting single-task RTs (Schacherer & Hazeltine, 2020), although the designs of these experiments were not optimal for adjudicating between crosstalk and stage shortening as the action effects for both tasks were manipulated simultaneously. Thus, we conducted a second experiment focusing on manipulated action effects that were not compatible with the stimuli.

In Experiment 2, we attempted to adjudicate between the crosstalk and (central) stage shortening accounts by adding action effects that we assumed would not affect the compatibility of the SE mappings (i.e., would not produce any SE compatibility effects). We also wished to determine whether the dual-task costs were reduced because the auditory action effects reduced the monitoring demands for the auditory task, even though RTs for the auditory task were consistently longer across conditions than RTs for the visual task. Thus, we created a condition (manipulated animal; MA) in which participants heard the same starting words as in the auditory task in Experiment 1 (e.g., “mish”, “tic”) and again responded with manual responses. However, in this condition, the responses caused the presentation of animal words, “cat”, “dog”, “pig”, and “bird”. These action effects were unrelated to the stimuli, unlike the auditory action effects used in Experiment 1, but would still allow for the responses to partly rely on auditory codes. We predicted that these auditory action effects would change the representation of the responses, thereby reducing crosstalk between the tasks, leading to a reduction in dual-task costs, without shortening RTs for the auditory task as in Experiment 1. If the RTs are not affected by the action effect manipulation, we will be able to determine whether the reductions in dual-task costs stemmed from reduced crosstalk or shortened central processes.

To further examine the role of action effects in dual-task costs, we included a condition (manipulated number; MN) in which auditory action effects were added to the responses for the auditory task, but the action effects included ordinal information that should interfere with the ordinal/spatial information associated with responses for the visual task. Thus, a third group of participants performed the auditory task with the auditory words “one”, “two”, “three”, and “four” as action effects. These action effects are highly similar to the MA condition described above but include information that is conceptually related to the manual responses for the other hand. In this way, they provide a strong test of the crosstalk account. If manipulated action effects reduce dual-task costs by drawing attention to the manipulated effect (and away from the unmanipulated feedback from a manual response) or by making it easier to monitor the response, then dual-task costs should be similar across both manipulated effect conditions. Alternatively, if ordinal relationships among the responses play a large role in their representation when no action effects are provided (see, Hazeltine, 2005), then action effects involving ordinal relationships (MN condition) will be less effective at reducing dual-task costs.

In sum, Experiment 2 followed the same procedure as Experiment 1 with two differences (see Figure 3). First, we did not manipulate the stimuli for the visual task. That is, all participants performed the same spatial task used in Experiment 1 paired with one of the three auditory tasks (see below). This was done to emphasize the separability of the two tasks, as the spatial task should not activate verbal information during response selection (see, Halvorson & Hazeltine, 2019). Second, for the auditory task, stimulus-effect pairs were no longer word pairings that completed phrases (e.g., “mish-mash”), but instead involved arbitrary auditory associations (e.g., “mish-cat” or “mish-one”). Participants were separated into three auditory action effect conditions: unmanipulated (U) effects, which was identical to that in Experiment 1; manipulated animal (MA) effects, in which responses were followed by spoken animal words (bird, cat, dog, pig); or manipulated number (MN) effects, in which responses were followed by spoken numbers (one, two, three, four). Thus, for both manipulated groups, the stimuli and effects in the auditory task were no longer semantically compatible (i.e., we eliminated the SE compatibility in Experiment 1), so there should be no differences in single-task RTs across conditions, as stimulus-effect modality compatibility typically does not affect single-task RT (Schacherer & Hazeltine, 2020).

Figure 3.

Example task pairings for the three conditions in Experiment 2. For the visual task, stimuli were spatial Xs (e.g., X - - -). For the auditory task, stimuli were spoken words (mish, riff, tic, zip) and were followed by unmanipulated effects, manipulated animal effects (cat, dog, bird, pig), or manipulated number effects (one, two, three, four).

3.1. Method

3.1.1. Participants.

One-hundred and three students from the University of Iowa who did not participate in Experiment 1 participated in partial fulfillment of an introductory psychology course requirement. Data from 7 participants whose overall accuracy was less than 80% were discarded and not analyzed, leaving 96 total participants whose data were analyzed. These 96 participants were equally divided (32 per condition) into three conditions: spatial/manipulated animal effect (S-MA; 23 female, M_age = 18.72, SD_age = 1.14); spatial/manipulated number effect (S-MN; 25 female, M_age = 18.97, SD_age = 2.88); or spatial/unmanipulated effect (S-U; 26 female, M_age = 19.06, SD_age = 1.92). Vision and hearing were reported as normal or corrected-to-normal.

3.1.2. Stimuli and apparatus.

The experiment was conducted using PsychoPy (version 3.0; Peirce et al., 2019) and data were collected using the online Pavlovia server². All conditions performed the spatial task (S) that was used in Experiment 1. For both the manipulated and unmanipulated auditory tasks, stimuli were again the spoken words mish, riff, tic, or zip. There were now two different conditions with manipulated auditory action effects. For the manipulated animal (MA) effects, responses were followed by the spoken words cat, dog, bird, or pig. For the manipulated number (MN) effects, responses were followed by the spoken words one, two, three, or four. Thus, the spoken word stimuli now corresponded to either an animal (e.g., mish-cat) or a number (e.g., mish-one) effect. The unmanipulated (U) effect condition was identical to that used in Experiment 1 (Figure 3).

3.1.3. Design and procedure.

The procedure was identical to Experiment 1, with the exception that the study was conducted in an online setting.

3.1.4. Statistical analysis.

Across all conditions, the first two blocks (two homogenous single-task) were excluded from analysis and treated as practice. Thus, we analyzed data from 16 blocks. Additionally, the first four trials in each block were excluded from analysis. All responses given within the first 200 ms after stimulus onset or any RTs greater than 2000 ms were excluded from analysis. Lastly, trials in which no response was detected, in which one or both responses were incorrect, and trials following an error were removed from analysis of RT. Of the analyzed blocks, we removed 11.9% of trials from our final analysis.

3.2. Results

The mean RTs and ERs for all conditions are shown in Table 2.

Table 2.

Mean response times (RT) and error rates (ER) for each of the trial types in Experiment 2. Dual-task costs were calculated as the difference between the sum of RT/ERs in AND blocks minus the sum of RT/ERs in OR blocks. S-U = spatial-unmanipulated; S-MA = spatial-manipulated animal; S-MN = spatial-manipulated number; VM = visual-manual; AM = auditory-manual.

Condition	Trial Type:	Response Time (ms)	Error Rate
Spatial-Unmanipulated (S-U)	VM-Single	554	3.5
	VM-OR	696	3.9
	VM-AND	1241	7.9
	AM-Single	973	3.3
	AM-OR	1067	4.3
	AM-AND	1320	10.6
	Dual-task Costs	798	5.2
Spatial-Manipulated Animal (S-MA)	VM-Single	530	4.4
	VM-OR	638	3.7
	VM-AND	1050	4.8
	AM-Single	999	5.4
	AM-OR	1115	4.1
	AM-AND	1256	9.4
	Dual-task Costs	553	3.2
Spatial-Manipulated Number (S-MN)	VM-Single	542	3.5
	VM-OR	687	3.4
	VM-AND	1204	9.2
	AM-Single	1002	7.4
	AM-OR	1095	6.3
	AM-AND	1300	13.6
	Dual-task Costs	722	7.1

3.2.1. Single-task RTs.

We conducted separate univariate ANOVAs for each task (visual, auditory) with condition as a between-subjects factor. For the visual task, there were no significant differences across conditions for both RT (S-U: 554 ms; S-MA: 530 ms; S-MN: 542 ms), F(2, 93) = 0.34, p = .711, ƞ² = .007, BF = 0.12; and ER (S-U: 3.5; S-MA: 4.4; S-MN: 3.5), F(2, 93) = 0.39, p = .678, ƞ² = .008, BF = 0.13. For the auditory task, there were no differences across conditions for RT (S-U: 973 ms; S-MA: 999 ms; S-MN: 1002 ms), F(2, 93) = 0.32, p = .726, ƞ² = .007, BF₁₀ = 0.12; but there were differences across conditions for ER (S-U: 3.3; S-MA: 5.4; S-MN: 7.4), F(2, 93) = 3.90, p = .024, ƞ² = .077, BF₁₀ = 2.08. Follow-up independent-samples t-tests revealed a significant difference between the S-U and S-MA conditions, t(62) = 2.36, p = .022, d = 0.59, BF₁₀ = 2.54; between the S-U and S-MN conditions, t(62) = 2.53, p = .014, d = 0.63, BF₁₀ = 3.58; but not between the S-MA and S-MN conditions, t(62) = 0.25, p = .252, d = 0.29, BF₁₀ = 0.45. Thus, our attempt to reduce the degree of compatibility between stimuli and manipulated action effects in the auditory task appeared to be successful (Figure 4).

Figure 4.

Single-task response times (RT) and dual-task costs for each of the three conditions in Experiment 2. Light gray bars represent the unmanipulated (UNMANIP) effect condition (S-U); medium gray bars represent the manipulated animal (ANIMAL) effect condition (S-MA); black bars represent the manipulated number (NUMBER) effect condition (S-MN). Dual-task costs represent performance differences between AND-OR blocks. Error bars represent standard error of the mean. VM = visual-manual; AM = auditory-manual.

3.2.2. Dual-task costs.

We again summed dual-task costs across the two tasks and conducted a univariate ANOVA with dual-task costs as our dependent measure and condition as a between-subjects factor. We observed a significant effect of condition (S-U: 798 ms; S-MA: 553 ms; S-MN: 722 ms), F(2, 93) = 5.67, p = .005, ƞ² = .109, BF₁₀ = 8.21. Follow-up independent-samples t-tests revealed significant differences between the S-U and S-MA conditions, t(62) = 3.25, p = .002, d = 0.81, BF₁₀ = 18.18; between the S-MA and S-MN conditions, t(62) = 2.23, p = .030, d = 0.56, BF₁₀ = 2.00; but not between the S-U and S-MN conditions, t(62) = 0.30, p = .295, d = 0.27, BF₁₀ = 0.41 (Figure 4). There were no significant differences in ER across conditions (S-U: 10.3; S-MA: 6.4; S-MN: 14.1), F(2, 93) = 2.83, p = .064, ƞ² = .057, BF₁₀ = 0.89, indicating that these results were not due to a speed-accuracy trade-off.

3.3. Discussion

There were no differences in single-task RTs across auditory tasks, so the differences in dual-task costs across conditions are attributed to central interference between the representations for the two tasks rather than shortened response selection processes. Critically, dual-task costs were smaller in the condition containing the MA task compared to the MN task, even though both tasks included manipulated auditory action effects. We propose that the use of auditory number action effects in the MN task may have engaged ordinal codes that overlapped with the response representations for the visual task (see Hazeltine, 2005). That is, the manual responses without action effects may engage ordinal codes (e.g., first, second, etc.), so when no action effects are used for either manual task (S-U condition), overlap, and therefore dual-task costs, is large. When action effects are added, they may diminish the role of ordinal codes and reduce interference, as in the S-MA condition. However, if the action effects also involve ordinal codes, then little reduction in dual-task costs will be observed, as in the S-MN condition. Thus, the findings are consistent with the hypothesis that the action effects reduce dual-task costs by de-emphasizing the ordinal coding of responses. They are not consistent with the stage shortening and monitoring hypotheses.

4. General Discussion

The purpose of this study was to characterize how action effects influence dual-task costs. Specifically, we sought to determine whether action effects affect dual-task costs in a manner consistent with a monitoring process that delays the selection of a second response (monitoring hypothesis), a reduction in the temporal overlap between central stages (stage shortening hypothesis), or a reduction in overlap between the representations engaged by central operations (crosstalk hypothesis). To this end, we added auditory action effects and observed reduced dual-task costs (cf., S-MN condition in Experiment 2) relative to when no action effects were added. This pattern was observed in two experiments, indicating that action effects either reduced the temporal overlap between central operations or were integrated into representations used by central operations (e.g., Frings et al., 2020; Hommel et al., 2001), which in turn reduced crosstalk between the two tasks. Because reduced dual-task costs were observed in the absence of reduced single-task RTs in Experiment 2, our findings are most consistent with crosstalk accounts of dual-task costs (e.g., Hazeltine et al., 2006; Logan & Gordon, 2001; Navon & Miller, 1987; Schacherer & Hazeltine, 2020, 2021). However, we also provide evidence that compatibility between stimuli and manipulated action effects can shorten central operations compared to conditions with the same stimuli and responses but without any manipulated effects.

4.1. How does the relationship between action effects affect dual-task costs?

The crosstalk account holds that dual-task costs are reduced for conditions containing the manipulated action effect because there is less crosstalk between the central operations for the two tasks (see also, Janczyk et al., 2012). Because central operations act on representations that include information about both stimuli and response-related action effects (Frings et al., 2020; Hommel et al., 2001), the size of dual-task costs should depend on how well the representations for the two tasks can be separated (e.g., Halvorson & Hazeltine, 2015). Thus, costs are small when participants represent the visual task in terms of the spatial or tactile/proprioceptive effects of the manual response and the auditory task in terms of a completed phrase (e.g., mish-mash) or word pair (e.g., mish-cat). In contrast, costs are large when the central operations for two tasks overlap (e.g., when both tasks contain verbal information, when both tasks are represented by similar action effect modalities, or when both actions are represented ordinally).

In the present experiments, the concurrently-performed tasks varied along two dimensions: 1) the presence of manipulated action effects for the auditory task (Experiments 1 and 2) and 2) the content of the manipulated auditory action effects (Experiment 2). These manipulations produced two critical findings relating to the role of action effects in dual-task costs. First, dual-task costs were significantly smaller when responses to auditory stimuli produced a manipulated action effect compared to when they did not, even though the stimuli and responses were identical for both types of tasks. This finding illustrates how the consequences of a response are integrated into the representation of a task (Frings et al., 2020; Hommel et al., 2001; Schacherer & Hazeltine, 2020). We propose that costs were smaller because participants conceptualized their responses as the second half of a phrase (as in Experiment 1) or a word (as in Experiment 2), rather than as the spatial or ordinal components of the consequences from the manual response. Conceptualizing the responses in terms of an auditory action effect increases the separability from the visual task, thereby reducing crosstalk and dual-task costs. These findings suggest that procedures that do not include salient action effects, particularly when button-press responses are used, may be skewing the pattern of dual-task costs compared to real-world situations in which the responses are associated with intended changes in the environment. That is, requiring button press responses without action effects may cause selection processes to rely heavily on spatial/ordinal representations of the button responses, which might, for example, increase costs if both tasks are reliant on such codes (c.f., Logan, 2003).

Second, conceptual overlap in the action effects played a role in the dual-task costs. In Experiment 2, dual-task costs were significantly smaller in the S-MA condition when compared to the S-MN condition, even though both conditions included a task with a manipulated auditory action effect. It is striking that auditory animal names and auditory numbers as action effects produced robust differences in the pattern of dual-task costs. This finding provides the strongest piece of evidence against the monitoring account, as the number action effects should not place greater demands on monitoring processes than the animal action effects. We contend that the use of spoken numbers as manipulated action effects in the S-MN condition may have caused the auditory task to rely on representations that included spatial/ordinal information. Thus, when both tasks involved ordinal information, crosstalk between the two tasks was greater. This finding indicates that even spatially compatible stimuli engage ordinal representations (see, Hazeltine, 2005). In short, the larger costs in the S-MN condition may reflect the unwanted interaction between the spatial codes associated with the visual task and the spatial/ordinal codes associated with the auditory task, strengthening the claim that action effects are not strictly related to their modality, but also to their content (Koch & Kunde, 2002; Schacherer & Hazeltine, 2021).

4.2. Single-task RTs

The addition of conceptually-compatible action effects in Experiment 1 reduced single-task RTs, even though SR mappings were identical across conditions with and without added action effects. This finding demonstrates that manipulating the end goal of an action via instructions (e.g., “complete the phrase”) and/or added action effects can improve overall performance (i.e., faster RTs). While it is established that manipulated action effects can impact performance by altering patterns of SR or RE compatibility (e.g., Hommel, 1993; Kunde, 2001), to our knowledge, SE compatibility—in the absence of SR or RE compatibility—has not been reported thus far (see Moeller et al., 2019). One mechanism contributing to this result may relate to the functional similarity between spoken word stimuli and spoken word effects. If the stimuli and effects are integrated into a shared representation (here, a completed phrase; e.g., Hommel, 2004), central processes can bind a spoken word stimulus to its corresponding spoken word effect, allowing these bindings to be retrieved more easily compared to when a spoken word stimulus is bound to the body-related tactile and proprioceptive effects stemming from the motor response. Future work may broaden the empirical basis of SE compatibility by applying it to dimensions outside semantically-related codes. For the present experiments, we note that action effects can be added to increase compatibility relative to when no effects are added.

This finding has broad implications for the design of human-machine interfaces. That is, we show that tasks can be made easier by simply adding post-response information. In many real-world dual-task scenarios, such as operating technology (e.g., phones, in-vehicle devices) while driving, operators possess limited processing resources (e.g., attention) that can be devoted to each task. Although automation tools such as voice control have been developed to reduce physical workloads (e.g., manipulating and holding the phone), cognitive workload is still high (e.g., Strayer & Johnston, 2001). The findings also suggest that simply adding information following response production can reduce cognitive workload, as indexed by reduced single-task RTs and dual-task costs. Together, understanding dual-task limitations as they arise in real-world settings and the factors that modulate these limitations (e.g., task overlap) has the potential to improve interface design and consequently improve user safety.

4.3. Alternative explanations

Although these findings are most consistent with crosstalk accounts (e.g., Hazeltine et al., 2006; Schacherer & Hazeltine, 2020, 2021; see also, Logan & Gordon, 2001; Navon & Miller, 1987), there are alternative explanations for these findings. For instance, these results could be explained by working memory accounts (e.g., Baddeley, 1986; see also, Hazeltine & Wifall, 2011; Maquestiaux, Ruthruff, Defer, & Ibrahime, 2018). In tasks involving the nonspatial visual task and/or the unmanipulated action effect task, the stimuli and responses across tasks may reside in similar working memory subsystems. If participants attach a verbal label to nonspatial stimuli, then this may tap the same working memory subsystem as the spoken word auditory stimuli. Likewise, when there is no manipulated action effect, both tasks may be represented in terms of the unmanipulated environmental effects of manual responses, which may tap the same working memory subsystem. In contrast, when the two tasks can be represented by different stimuli and action effects (i.e., visual task: unmanipulated action effects; auditory task: manipulated auditory action effect), the tasks may tap distinct working memory subsystems, reducing dual-task costs. However, it is unclear how the working memory account explains the findings from Experiment 2 if the auditory numbers and animal words are encoded by the same working memory subsystem.

Both the crosstalk and working memory accounts suggest that differences in dual-task costs depend on how central operations are structured by the properties of stimuli, responses, and effects. However, it is also possible that our results could reflect strategic adaptations to the task demands (e.g., Meyer & Kieras, 1997a, 1997b). In Experiment 1, for instance, in conditions involving the nonspatial visual task—which we posit requires verbal labeling—participants may have delayed their responses so the central operations selecting verbal information for that task did not interfere with selecting verbal information for the auditory task. However, in the two conditions containing the nonspatial visual task (NS-M; NS-U), the pattern of dual-task costs differed across the conditions: dual-task costs for the visual task were smaller in the NS-M condition (visual: 390 ms; auditory: 448 ms) than the NS-U condition (visual: 511 ms; auditory: 412 ms). Costs were also near-identical between the NS-M (839 ms) and S-U (838 ms) conditions, even though the central operations for both tasks in the latter condition are not operating on verbal information. Moreover, stimuli were presented simultaneously (i.e., 0 ms SOA) and we provided no explicit instructions to emphasize one task over the other. Critically, the action effect always occurred after stimuli for both tasks had been presented. Thus, if the tasks are performed correctly, the manipulated auditory effect should never overlap with the perception of either stimulus, reducing the likelihood of strategic adaptations (see Schacherer & Hazeltine, 2020). As such, the present data provide no compelling evidence in favor of strategically delaying the selection of one response.

Lastly, we make no strong claims against the proposal that central operations may be engaged for the purpose of monitoring the consequences of our actions. Actions may need to be monitored to support a second response that may be contingent on the preceding response (e.g., typing on a slow computer that only displays the typed letters after a delay; see Kunde et al., 2018). That is, a major role of central operations may be to monitor action effects to determine whether the intended consequences have been achieved. This process may be essential for forming the representations that produce the present pattern of dual-task costs. Nonetheless, the monitoring account, as currently described, does not provide a succinct explanation for the pattern of costs observed here. In particular, the findings from Experiment 2, in which dual-task costs were smaller for the animals rather than the numbers (which may have been easier to associate with the manual responses), suggest that action effects do not reduce dual-task costs by shortening monitoring processes.

4.4. Summary

Interactions between concurrently performed tasks limit our abilities across a wide array of everyday behaviors. Understanding these interactions and how they can be minimized will have far-reaching implications in both basic and applied domains. The present study aimed to evaluate three hypotheses concerning how manipulated action effects affect these interactions: monitoring, stage shortening, and crosstalk. Across two experiments, results favored the crosstalk hypothesis: when task representations, manipulated with action effects, overlap less across tasks, dual-task costs are reduced. More generally, these results revealed how action effects influence response selection processes. First, action effects can reduce single-task RT without changing the pattern of which SR pairs are compatible. Second, action effects can reduce dual-task costs compared to when no action effects are present—even when they do not reduce single-task RTs—suggesting that dual-task costs are affected by the relationship between the conceptual codes associated with the two tasks (see Hazeltine, 2005; Schacherer & Hazeltine, 2020, 2021). These findings suggest that adding action effects to responses can allow tasks to be performed more efficiently and interfere less with other tasks.

Data availability:

Data are publicly available at https://osf.io/juy7b/.