We used a novel experimental procedure that allowed for differentiating exogenous temporal attention and temporal expectation. We led observers to prioritize accuracy by using a separate go cue that prevented observers from responding too quickly. Hence, we were able to observe the performance effects of exogenous temporal attention. Accuracy improved with exogenous temporal attention, specifically for the first target. The accuracy improvements were more pronounced when there was less uncertainty regarding the stimulus timing. Our results provide evidence that specific moments can be selected and prioritized with exogenous cues, and this selection leads to behavioral benefits and costs, which vary with temporal uncertainty.
In our design, the temporal uncertainty window differed for T1 and T2. For T1 the stimulus could appear 1 or 2 seconds after the ready signal, whereas T2 onset varied between 250 and 350 ms after the T1 onset. This difference may have resulted in differences in the precision of temporal expectations for T1 and T2. Because we aimed to investigate the benefits and costs of attending to specific moments in time, we had a temporal window short enough that T1 and T2 would compete, that is, precluding the system from recovering resources from attending to T1 and deploying them to T2. We chose the short T1–T2 SOA based on the findings from a study in which a range of SOAs were tested and the strongest attention effects emerged with a T1–T2 SOA between 250 and 350 ms (
Denison et al., 2021).
The findings of the present study are in line with the dynamic normalization of temporal attention (
Denison et al., 2021). The effects of exogenous temporal attention were modeled as a function of the SOA between the two targets, but this component did not play a critical role in the model. In line with the previous results, we did not find any significant effect of the SOA between T1–T2 on the T2. However, we observed an effect of the cue validity on the performance; when the stimulus timing is unpredictable, as in the attentional blink paradigm simulated in the paper. Our findings show that exogenous temporal attention can affect performance in the presence of temporal uncertainty.
Previous studies investigating the effect of involuntary temporal attention reported improvements in RT, and no change in the accuracy (
Lawrence & Klein, 2013), because the performance was often at a ceiling level. In our design, to investigate whether visual performance changes, we prioritized accuracy specifically over speed by having observers wait 1,000 ms after the response cue, and allowing them unlimited time to respond. We also titrated the neutral accuracy at 75% to be able to see both the benefits and costs in performance in different conditions, and we used task-irrelevant, although temporally informative, cues. With our experimental design, we observed slight benefits and costs in accuracy, without any speed–accuracy tradeoff.
Spatial attention can be deployed endogenously or exogenously, and their similarities and differences are well characterized (e.g.,
Barbot et al., 2012;
Dugué et al., 2020;
Fernández et al., 2021;
Jigo et al., 2021). Feature-based attention, selective prioritization of specific features (for reviews,
Carrasco, 2011;
Serences & Kastner, 2014;
Liu, 2019) can be deployed endogenously across space, even when irrelevant for the task at hand (e.g.,
White & Carrasco, 2011;
Störmer & Alvarez, 2014;
Liu & Jigo, 2017). Stimulus-driven feature-based attention has been reported (
Lin, Hubert-Wallander, Murray, & Boynton, 2011;
Qian & Liu, 2015), but these exogenous effects have not been replicated in more recent experiments (
Donovan, Zhou & Carrasco, 2019). In any case, feature-based attention is primarily an endogenous process. Here, we characterize the effect of exogenous temporal attention on visual perception, and note that it is not as pronounced as that of endogenous temporal attention.
Can the effects observed here reflect endogenous temporal attention? Two studies have shown that endogenous temporal attention can be deployed quickly (
Hilkenmeier & Scharlau, 2010;
Yeshurun & Tkacz-Domb, 2021), but we contend that the pattern of results in the present study cannot be explained by such an effect. The first study showed that endogenous temporal attention can be allocated to a target in 100 ms, when its onset is temporally contingent on the first target (
Hilkenmeier & Scharlau, 2010). In our experimental design, the cue–target SOA was 100 ms, but the onset of the second target was not temporally contingent on the presentation of the first target as the T1–T2 SOA was randomized across trials. We did not find any cueing effect on performance for T2, suggesting that the cues did not drive endogenous temporal attention. The second study compared performance for stimuli preceded by temporally informative or uninformative cues presented shortly before the target (150 ms) and found a higher performance when the cue was temporally informative (
Yeshurun & Tkacz-Domb, 2021). Thus, the authors concluded that endogenous temporal attention can be deployed in a fast manner. In that experimental design, there was only one target presented after the cue that was behaviorally relevant, so endogenous attention needed to be allocated to the relevant time points preceded by the cues. In contrast, we present two targets, the cues are not informative about behavioral relevance because the response cue is equally likely for both targets, and it is presented at the end of the trial. Thus, endogenous temporal attention should be allocated equally to both targets. Furthermore, endogenous temporal attention can selectively improve performance for the first or the second target across different SOAs while impairing performance at the unattended interval (
Denison et al., 2017,
2021). Accordingly, when the first target was cued, we would have expected a benefit for the first target and a cost for the second target, and, vice versa, when the second target was cued we would have expected a benefit at the second target and a cost at the first. However, we did not observe any attention effect in T2. Overall, the pattern of results suggests that with our experimental protocol 100 ms was too short to deploy voluntary attention in time.
Both accuracy and RTs were better for T2 than T1, regardless of the cueing condition. Could the T1 onset have induced endogenous temporal attention to T2? We think this was not the case. In our experiment, endogenous temporal attention should be distributed across both intervals because both stimuli were equally likely to be the target in all conditions, as is the case in the neutral condition in endogenous temporal attention studies (
Denison et al., 2017,
2021;
Fernandez et al., 2019), for which accuracy and RT was also better for T2 than T1. In this study, the performance difference between T1 and T2 may be due to the higher uncertainty for T1 than T2.
Could the exogenous temporal attention be merely due to a decrease in uncertainty, because the cue presentation always indicates the subsequent appearance of the target? We think that this is not the case, because this decrease would play a greater role for trials with high rather than low temporal uncertainty. The precues decreased the possible stimulus presentation window from a longer time (2 seconds) to 100 ms in the high uncertainty condition, and from a shorter time (1 second) to 100 ms in the low uncertainty condition. Accordingly, we would expect either a larger, or at least the same, effect of uncertainty reduction in the high than the low uncertainty conditions, because the precues predict the target onset equivalently. However, the cueing effect was present for the low, but not for the high, uncertainty condition. Likewise, in the previous temporal attention studies with endogenous attention, the cue gives the same timing information in all cueing conditions. Thus, any contribution of reduced uncertainty would have been constant across cueing conditions and the observed effects can be attributed to attention (
Denison et al., 2017,
2021;
Fernández et al., 2019).
Can the effects observed here result from arousal, which is another mechanism that affects performance in time (
Sara & Bouret, 2012;
Petersen, Petersen, Bundesen, Vangkilde, & Habekost, 2017;
Wang et al., 2018;
Burlingham, Mirbagheri, & Heeger, 2022)? Arousal is described as nonspecific, global enhancements of biological processes (
Hebb, 1955;
Eysenck, 1976;
Robbins, 1997). It has strong physiological components (
Taylor & Epstein, 1967), and is related to overall readiness or wakefulness (
Posner & Petersen, 1990;
Robbins, 1997). Arousal levels, indexed by pupil responses, increase with temporal uncertainty, such that higher arousal levels are observed when a visual target onset is not predictable (
Shalev & Nobre, 2022), as well as when auditory targets are uncertain (
Friedman, Hakerem, Sutton, & Fleiss, 1973), and visual perception improves with high arousal (
Kim, Lokey, & Ling, 2017). Thus, had the arousal level mediated the temporal attention effects in our study, we would have expected a larger effect for high rather than low temporal uncertainty. However, in the present study, exogenous attention improved performance more under low than high temporal uncertainty. Hence, although manipulating temporal expectations may have modulated arousal levels, the exogenous attention effects cannot be explained by arousal.
Both attention and expectation extend in different dimensions: space (
Zuanazzi & Noppeney, 2018,
2020), feature (
Summerfield & Egner, 2016), and time (
Doherty et al., 2005;
Todorovic, Schoffelen, van Ede, Maris, & de Lange, 2015;
Moon et al., 2019). Attention and expectation differ in terms of the behavioral outcomes and the underlying neural mechanisms (
Summerfield & Egner, 2009;
Carrasco, 2011;
Lange, 2013;
Vangkilde, Petersen, & Bundesen, 2013;
Cheadle, Egner, Wyart, Wu, & Summerfield, 2015;
Summerfield & Egner, 2016;
Denison, Yuval-Greenberg, & Carrasco, 2019;
Zuanazzi & Noppeney, 2019;
Rungratsameetaweemana & Serences, 2019;
Wilsch, Mercier, Obleser, Schroeder, & Haegens,2020). These two mechanisms were often treated as a single cognitive function in the previous temporal attention and expectation literature, and the experimental designs and the conclusions often did not differentiate between them. However, it is important to define attention and expectation clearly and manipulate them either independently or separately to be able to characterize and differentiate their effects.
Attention is a limited resource, and the performance improvements for the selected information trade-off with impairments for the unselected information. Hence, attending to every expected or probable stimuli would neither be efficient, nor possible. In complex daily life situations, it is likely that attention and expectation work together. Given that the critical factors that drive attention are goal-driven by task relevance (endogenous attention) and stimulus-driven by a salient change (exogenous attention), whether the expected stimuli will be attended or not depends on context.
An example for the interplay between exogenous temporal attention and predictability of events is the red hands game. The first player puts their hands facing up, and the second player puts their hands on top, facing down. The first player tries to slap the second player's hands before they pull them away. From the second player's perspective, it is important to attend the moment when the opponent will hit them. However, there is no temporal information to endogenously attend to a certain moment, although the uncertainty of the event decreases in time: As the first player does not make a move, the probability of making a move in the next moment increases. And the salient exogenous cue in this scenario is the first player moving their hands, which cues the time point when to attend where the hand will be shortly to try to avoid being hit by the first player. The exogenous cue captures attention to a future time point, such that the performance at that specific moment increases, while lowering performance at the unattended moments. Temporal uncertainty affects the benefits and costs of exogenous temporal attention.
Here, we investigated the effects of exogenous temporal attention under different levels of temporal uncertainty. We showed that attentional selection in time affects performance based on the temporal predictability of the events. Overall, the results pointed out an interplay between temporal attention and temporal expectation.