Introspection about backward crosstalk in dual-task performance

A long tradition of dual-task research has shown that multitasking usually comes at a cost (for reviews see Koch, Poljac, Müller, & Kiesel, 2018; Pashler, 1994). These costs are frequently investigated with psychological refractory period (PRP) experiments (e.g., Pashler, 1994), where two tasks (Task 1 and Task 2) are presented with varying temporal overlap (stimulus onset asynchrony, SOA). Typically, reaction time (RT) for Task 2 (RT2) increases with increasing overlap (i.e., with decreasing SOA), the so-called PRP effect (Telford, 1931). The prominent central bottleneck model attributes this effect to a resource-limited central processing stage, which can serve only one task at a time (Welford, 1952; Pashler, 1994). Accordingly, central processing in Task 2 cannot start until the central processor is freed from Task 1 central processing. At short SOAs, when there is a high temporal overlap between the two tasks, this leads to a waiting time in Task 2 processing and a corresponding prolongation of RT2.

Numerous studies have demonstrated that introspection about dual-task costs is severely limited, since participants are usually not aware of the PRP effect (Bratzke & Bryce, 2016; Bryce & Bratzke, 2014, 2015, 2017; Bratzke, Bryce, & Seifried-Dübon, 2014; Corallo, Sackur, Dehaene, & Sigman, 2008; Marti, Sackur, Sigman, Dehaene, 2010; for a diverging pattern regarding task switch costs, see Bratzke & Bryce, 2019). To assess people’s introspection, most of these studies used the method of quantified introspection (Corallo et al., 2008), in which participants provide estimates of their RTs (introspective reaction times, IRTs) after each trial. Consistent across several studies, the PRP effect was not reflected in IRTs (Bryce & Bratzke, 2014, 2015; 2017; Corallo et al., 2008; Marti et al., 2010). This result has been interpreted as evidence for a “unified attentional bottleneck” encompassing not only response selection, but also conscious perception (Corallo et al., 2008; Marti et al., 2010; see also Arnell & Jolicoeur, 1999; Marti, Sigman, & Dehaene, 2012; Ruthruff & Pashler, 2001; Tombu et al., 2011).

In light of the intriguing introspective blind spot of the PRP effect, it is noteworthy that introspective abilities for other manipulations of task difficulty are often preserved. For example, Corallo et al. (2008) manipulated the difficulty of perceptual and central processing in a numerical comparison Task 2 (i.e., notation and numerical distance) and observed that both effects were reflected in IRTs. Bryce and Bratzke (2014) used different degrees of stimulus degradation to manipulate perceptual processing in Task 2, and made a similar observation. Thus, so far it seems that the effects of difficulty manipulations, irrespective of whether they are supposed to affect central or perceptual processing, are not subject to the same introspective limitations as the PRP effect.

In the present study, we focused on a particularly interesting case of a difficulty manipulation in Task 1, where compatibility relations between the responses in both tasks affect not only processing of Task 2, but also of Task 1. Such a compatibility-based backward crosstalk effect (BCE) was first demonstrated by Hommel (1998). For example, in Experiment 1 of his study, a colored letter served as the stimulus and participants responded in Task 1 with a manual left or right key press to the color of the letter, and in Task 2 with a vocal “left” or “right” utterance to the identity of the letter. Even RTs in Task 1 (RT1) were shorter in trials with compatible responses (e.g., left key press and vocal “left” utterance) compared with incompatible responses. Similar observations were reported by subsequent studies, some of which extended the result to other responses such as pedal responses in Task 2 (e.g., Ellenbogen & Meiran, 2008; Janczyk, Pfister, Hommel, & Kunde, 2014; Janczyk, Renas, & Durst, 2018; Lien & Proctor, 2000; Miller & Durst, 2014).

A characteristic of the BCE, which is interesting for the present study as well, is its sequential modulation. Such sequential modulations were reported first for the Eriksen flanker task (Eriksen & Eriksen, 1974), where the respective congruency effect is larger following congruent than following incongruent trials (Gratton, Coles, & Donchin, 1992). Similar effects were also reported for other tasks (e.g., Janczyk & Leuthold, 2018; Leuthold & Schröter, 2006; Stürmer, Leuthold, Soetens, Schröter, & Sommer, 2002; Wühr, 2004, 2006). According to a prominent explanation, the cognitive system registers cognitive conflict in incongruent trials, and adjusts subsequent processing to deal with this conflict experience in goal-directed ways (Botvinick, Braver, Barch, Carter, & Cohen, 2001). Similar sequential modulations were also reported for the BCE in several studies (Durst & Janczyk, 2019; Janczyk, 2016; Renas, Durst, & Janczyk, 2018; Scherbaum, Gottschalk, Dshemuchadse, & Fischer, 2015): The BCE is much smaller, sometimes inverted, following incompatible trials compared with following compatible trials. Conceivably, this sequential modulation is an example of a rather subtle effect on RT1s that results from the co-occurrence of several task features. In the present study, we utilize this effect to explore possible limits of introspection regarding difficulty effects observed in Task 1 of a dual task in a trial-by-trial manner.

Experiment 1 combined the approach of Janczyk’s (2016) study with the method of quantified introspection. Accordingly, participants performed a dual task with manual responses in Task 1 and pedal responses in Task 2. After each trial, participants were asked to provide IRTs for Task 1 and Task 2.

Experiment 2 was basically the same as Experiment 1, but we collected IRTs only after each second trial. That is, each trial was essentially a pair of a prime and a test trial, and participants were asked to provide IRTs only for the test trial. With this procedure, we expected a stronger sequential modulation of the BCE than in Experiment 1.

The present experiments replicate the standard BCE and its sequential modulation in dual-task performance. Accordingly, response compatibility between the two tasks affected not only Task 2 but also Task 1 performance, and this effect was stronger after compatible than after incompatible trials. Our main interest was how participants would introspect about the objective result pattern in this quite complex dual-task situation. Regarding the BCE and its sequential modulation, introspective RTs largely reflected the objective pattern. This was especially the case in Experiment 2 with the pairwise trial procedure, where participants reported in their IRTs not only the BCE but also the SOA-dependent sequential modulation of the BCE. Remarkably, in Experiment 1, the PRP effect was reflected in IRTs. The results of Experiment 2, however, showed the typically observed absence of the PRP effect in IRTs. Thus, Experiment 1 appears to be one of the rare exceptions in which participants report a PRP effect in their IRTs (see also Experiment 2 of Bryce & Bratzke, 2014).

Another interesting aspect of the present results is the apparent longevity of the sequential modulation of the BCE. Previous results on sequential modulations in other conflict tasks have shown that these effects rapidly decay when more time elapses between subsequent trials (Duthoo et al., 2014; Egner et al., 2010). Note that in Experiment 1 of the present study, there was not only a prolongation of the ITI, but participants had to indicate their IRTs during this interval, a procedure that usually takes up to a few seconds. Importantly, we still observed a significant sequential modulation of the BCE, even though the effect was smaller than in Experiment 2, where no additional delay due to the assessment of IRTs occurred. This longevity might be conceived as a problem for a conceptualization of the sequential modulation of the BCE as conflict adaptation (see, e.g., Janczyk, 2016). However, it remains an open question if and how the assessment of IRTs plays a crucial role in the persistence of the sequential modulation over time. It is, for example, conceivable that the demand to provide IRTs, which may require maintaining a short-term representation of RTs, strengthens a possible experience of conflict, which in turn counteracts the dissipation of the sequential modulation.

Previous conceptualization of IRTs in dual tasks in light of the unified attentional bottleneck model imply that IRTs reflect relatively veridical time estimates of the consciously accessible internal processing times (Corallo et al., 2008; Marti et al., 2010). Others have proposed that introspective RTs in dual-task situations reflect retrospective inferences based on a variety of cues (Bratzke & Bryce, 2016, 2019; Bratzke et al., 2014; Bryce & Bratzke, 2014, 2017; see also Klein & Stolz, 2018). In line with the latter assumption, one could argue that in the present study, participants inferred their IRTs from, for example, their experience of the between-trial sequence of R1–R2 compatibility or of the sequence of required and performed responses (e.g., 50% of compatible–compatible and incompatible–incompatible sequences include full response repetitions). It is unlikely, however, that participants inferred the BCE from these experiences, because then they should have been insensitive to the SOA dependency of the BCE and its sequential modulation.

Other potential sources of information for IRTs may be the feeling of difficulty (Bryce & Bratzke, 2014) and the feeling of conflict (or the “urge-to-err”, e.g., Morsella, Wilson, Berger, Honhongva, Gazzaley, & Bargh, 2009; Questienne, Atas, Burle, & Gevers, 2018). Even though these subjective aspects of cognitive control are certainly highly related, if not indistinguishable (but see Questienne, van Dijck, & Gevers, 2018), they have been differently conceptualized. In an introspective PRP study, Bryce and Bratzke (2014) observed a close relationship between IRTs and the feeling of difficulty, and a better match of these variables with error rates than with objective RTs. From these results, the authors concluded that IRTs are strongly influenced by the feeling of difficulty and that error performance can serve as a proxy for the subjective feeling. In the present experiments, error rates largely mirrored RT performance, with the exception of the PRP effect. An exclusive use of this information, however, seems also unlikely because participants did not reliably indicate a sequential modulation of the BCE in Experiment 1, although both error rates as well as RTs showed such a modulation. Additionally, in Experiment 1, participants reported a PRP effect in their IRTs although there was no such effect on error rate.

In the standard PRP paradigm, potential input and output conflicts are usually minimized to investigate the “pure” central (or attentional) costs of dual tasking (e.g., Pashler, 1994). In this situation, the rather unspecific feeling of difficulty may be an appropriate aspect of subjective experience to ask for. In conflict tasks (as, e.g., the BCE task in the present study, or the flanker, the Simon, and the Stroop Task), however, the more specific feeling of conflict (or the “urge-to-err”) may better describe the participants’ subjective experience of their performance. Based on our intuitive understanding of conflict tasks, the feeling of conflict and objective RTs should be highly related. A recent study by Questienne, Atas et al. (2018) indeed provided evidence for such a relationship between RT and the urge-to-err in a simple conflict task (masked priming task). A second important variable related to the urge-of-err in this study was response competition (indexed by EMG activity on the wrong response hand), which modulated the relationship between RT and the urge-to-err, with a much steeper slope of the RT‒urge-to-err function in case of apparent response conflict. Bratzke and Bryce (2019) observed a similar linear relationship between objective and introspective RT in an introspective task-switching study. Consequently, IRTs and the feeling of conflict are probably highly related. The precise relationship of these two measures, however, can only be speculated upon; that they simply reflect confounded experiences seems to be just as possible as that one of them forms the basis for the other.

It is conceivable that the experience of conflict still plays an important role in the awareness of the BCE. According to the influential conflict monitoring theory by Botvinick et al. (e.g., Botvinick et al., 2001), the anterior cingulate cortex (ACC) monitors conflict and triggers cognitive control adjustments in case of conflict detection. Even though this theory does not make any explicit assumptions about the awareness or unawareness of conflict detection and the subsequent control adjustments, these internal signals are probably accessible to introspection (for the role of ACC activity in consciousness, see, e.g., Dehaene et al., 2004; Mayr, 2004; Qin et al., 2010). This consideration is also of interest with respect to the dissociation between the unawareness of the PRP effect in the present and previous studies, and the awareness of switch costs in Bratzke and Bryce (2019) and compatibility (or conflict) effects in the present study. Activation of the ACC has been consistently reported for task switching (e.g., Braver, Reynolds, Donaldson, & Louis, 2003; Dove, Pollmann, Schubert, Wiggins, & von Cramon, 2000; Hyafil, Summerfield, & Koechlin, 2009) and conflict tasks (see, e.g., Botvinick et al., 2001). However, the brain regions specifically involved in dual tasking are less clear (see Wu, Liu, Hallett, Zheng, & Chan, 2014), especially regarding the PRP paradigm (e.g., Dux, Ivanoff, & Asplund, 2006; Jiang, Saxe, & Kanwisher, 2004; Szameitat, Schubert, Mu, & von Cramon, 2002). For example, a study by Jiang et al. (2004) observed no engagement of the ACC or any other brain regions usually associated with cognitive control, when they contrasted a short with a long SOA condition in the PRP paradigm. Overall, the pattern of ACC activation associated with these effects seems to match with the dissociation regarding awareness of the effects in introspective RT studies. Based on these considerations, one can speculate that conflict detection by the ACC and/or subsequent cognitive control adjustments play a role in the awareness of effects like switch costs or the BCE. However, that a non-engagement of the ACC can explain the unawareness of the PRP effect appears very unlikely to us, as the effects of other difficulty manipulations usually not associated with ACC activation or conflict adaptation (e.g., numerical distance, stimulus degradation) have been reflected in IRTs (Bryce & Bratzke, 2014; Corallo et al., 2008).

More important, in terms of mental stage models the dissociation between the unawareness of the PRP effect and the awareness of the BCE and switch costs can be explained by a passive postponement of the central processing stage in case of the PRP effect, and a prolongation of the central stage due to active task preparation in task switching (see Bratzke & Bryce, 2019) and to conflict resolution in case of the BCE. This explanation thus combines assumptions about the processing dynamics in task switching and conflict task performance with the basic assumptions of the unified bottleneck model (Corallo et al., 2008; Marti et al., 2010; Tombu et al., 2011). What are the implications of this explanation for theorizing about the nature of the BCE? According to the conscious perception bottleneck model, conscious perception of Task 2 is delayed as long as the central processor is engaged in response selection for Task 1 (Corallo et al., 2008; Marti et al., 2010). Additionally, conscious access in dual tasks seems to be restricted to central processing (see Corallo et al., 2008; Marti et al., 2010). To reconcile the BCE with traditional bottleneck models, a sub-division of the central stage was suggested (Hommel, 1998; Lien & Proctor, 2002; see also Schubert, Fischer, & Stelzel, 2008). According to this idea, a first stage of response activation runs in parallel with other stages of simultaneous tasks, while only the second stage of (final) response selection is considered a bottleneck process. With sufficient temporal overlap between response activations, mutual crosstalk between the two tasks can occur. More recent studies, however, preferred a model where some automatic Task 2 response activation occurs, but directly affects the capacity-limited stage of response selection in Task 1 (Durst & Janczyk, 2019; Janczyk et al., 2018; Thomson, Danis, & Watter, 2015). While the response activation account suggests a postponement of Task 1 response selection, the latter account suggests that the duration of Task 1 response selection proper varies.

If one follows the assumption that in dual tasks conscious access is restricted to central processing, the present awareness of the BCE and its sequential modulation would indicate that the BCE arises from a prolongation rather than from a postponement of Task 1 response selection. This is line with the assumption that Task 2 response activation directly affects the duration of Task 1 response selection (Janczyk et al., 2018; Thomson et al., 2015). It is important to note, however, that previous results by Bryce and Bratzke (2014) argue against the”exclusive central access” assumption, at least for Task 1. In this study, the effect of stimulus degradation in Task 1 was reflected in IRTs, suggesting that perceptual (or pre-central) processing in Task 1 can be consciously accessible. Under this assumption, the present results would be consistent with both a locus of the BCE in Task 1 response activation as well as in Task 1 response selection.

Since the discovery of sequential conflict-modulation effects in the 90s of the last century (Gratton et al., 1992), there has been a discourse on the role of consciousness in conflict adaptation (e.g., Desender & Van den Bussche, 2012; Kunde, Reuss, & Kiesel, 2012; Mayr, 2004). Recent evidence suggests that conflict awareness plays an important role in conflict adaptation (Ansorge, Fuchs, Khalid, & Kunde, 2011; Desender, Van Opstal, & Van den Bussche, 2014; Fröber, Stürmer, Frömer, & Dreisbach, 2017, Kunde, 2003; Questienne, Van Opstal, van Dijck, & Gevers, 2018). Some of these studies used masking to manipulate the awareness of conflict information (Ansorge et al., 2011; Kunde, 2003), whereas others directly asked for ratings of conflict (Desender et al., 2014) or pleasantness (Fröber et al., 2017). The present result that the sequential modulation of the BCE was accompanied by an introspective trial-by-trial awareness of congruency effects on objective RT is certainly consistent with the suggestion of these studies. However, whether awareness of RT effects is crucial for the sequential conflict adaptation or rather an epiphenomenon remains an open question for future research.

In conclusion, the present study investigated participants’ ability to introspect about the effect of between-task crosstalk in dual tasks: while participants are typically unaware of the dominating PRP effect, here they showed awareness of the SOA-dependent sequential modulation of the BCE. This demonstrates that people’s introspection about the temporal processing demands in a rather complex dual-task situation is intriguingly accurate, but also severely limited at the same time.