Attentional capture by the onset and offset of motion signals outside the spatial focus of attention

Abstract

Abstract

Abstract:

Abstract  The present study examined whether participants were able to ignore a task-irrelevant commencement or cessation of optic flow while they were engaging in a letter-identification task, as claimed by adherents of the view that attentional set determines deployment of attention, or whether irrelevant events would capture attention regardless of observers' attentional set, as claimed by adherents of a broad range of views emphasizing the behavioral urgency of stimulus motion. Observers identified a green letter in a central rapid stream of heterogeneously colored nontargets. A completely task-irrelevant optic flow occurred in the periphery. If attentional deployment were governed by a top-down attentional set, the letter identification would be unaffected by the temporal change in the optic flow. The results reflected attentional capture by commencement or cessation of optic flow, which is inconsistent with the top-down view. When the peripheral dots expanded at various speeds before onset of the target, identification was impaired relative to when no motion occurred. Mere commencement or cessation of motion was sufficient to produce the capture effect. Qualitative (commencement or cessation) rather than quantitative changes (acceleration or deceleration) of the motion display were critical for the occurrence of attentional capture. We conclude that salient discontinuities in optic flow induce attentional capture when observers search for a feature in a different stimulus domain, an idea implying a unique role for of expanding global motion in the deployment of visual attention.

Introduction

 

It was a chilly afternoon. Two participants showed up at our lab. After signing several documents, they were given instructions and engaged in an experiment intended to examine the role of task set in attentional capture. Then, IT came. It started with small tremors in the windows and was followed by the furniture shaking. The rumblings of the ground grew louder. Articles on tables fell and scattered all over the floor. Upon hearing a loud sound of glass shattering, all of us, including the participants engaged in the attentional capture experiment, realized that it was time to leave the room (Tsukuba, Japan, March 11, 2011, the day of the Great East Japan Earthquake.)

 

The establishment of a task set is a cognitive function necessary for focusing on relevant stimuli and avoiding the diversion of attention to irrelevant distracting stimuli. Thus, the ability to maintain a task set is clearly advantageous when attending to critical aspects of a task and attempting to achieve an optimal behavioral goal (Eriksen, 1995; Kahneman & Treisman, 1984). Nonetheless, it is also true that excessive adherence to a task set may be problematic for survival, such as in emergency situations like the earthquake described in the above anecdote. Specifically, some types of dynamic environmental signals, such as luminance changes, abrupt onsets, the appearance of new objects (Franconeri, Hollingworth, & Simons, 2005; Hillstrom & Yantis, 1994; Jonides & Yantis, 1988), or the sudden initiation of motion (Franconeri & Simons, 2003; von Mühlenen & Lleras, 2007), inform observers of an event that may require urgent action. In these cases, attention is allocated to salient stimuli in a bottom-up, stimulus-driven way regardless of observers' current task set. 

Whether attention is governed by a top-down attentional set or in a bottom-up fashion determined by physical stimulus saliency has been examined over the course of more than two decades (Bacon & Egeth, 1994; Egeth & Yantis, 1997; Theeuwes, 2010). Although early studies suggested a pure form of stimulus-driven control (e.g., Hillstrom & Yantis, 1994; Theeuwes, 1992; Yantis & Jonides, 1996), most recent studies have provided evidence for top-down control (Einhauser, Rutishauser, & Koch, 2008; Lamy, Leber, & Egeth, 2004). Specifically, when a task was designed to prompt observers to adopt an attentional set to search for a particular color, it was possible to eliminate attentional capture by new objects that served as irrelevant distractors (Ansorge & Heumann, 2003; Chen & Mordkoff, 2007; Folk & Remington, 1998; Folk, Remington, & Johnston, 1992; Folk, Remington, & Wright, 1994; Lien, Ruthruff, Goodin, & Remington, 2008; Pratt & McAuliffe, 2002). The principle that a top-down task set for a color target prevents attention from being deployed to newly appearing objects seems applicable to cases in which irrelevant objects start moving as distractors because both the appearance of new objects and the onset of stimulus motion involve transient signals and thus should be coded by the common dorsal pathways (e.g., Livingstone & Hubel, 1988). However, the story is not that simple. 

Recent studies have repeatedly demonstrated that stimulus motion captures attention regardless of observers' task set. For example, Abrams and Christ (2003, 2005) found that visual search for a target letter improved when the target coincided with an object that started moving upon the appearance of the search display relative to when the target remained static throughout the trial. Using a different type of task, von Mühlenen and Lleras (2007) provided evidence in support of this observation. They presented a random motion display so that motion in one hemifield gradually became coherent. Probe-detection and letter-discrimination performance at the center of either hemifield was enhanced when the side on which the target (probe or letter) was presented coincided with the hemifield in which the coherent motion occurred. These findings suggest that the motion of the stimulus captured attention even when the motion was uninformative in terms of the observers' task. Recently, the notion that stimulus motion breaches observers' attentional set was tested more rigorously by Al-Aidroos, Guo, and Pratt (2010). In their study, the observers' task was to detect whether one of two white triangles turned red (target). Shortly before the color change, one of the two frames surrounding the two triangles began to either rotate or change its color from white to red. Under this condition, the change in the color of the frame was associated with a cueing effect. That is, reaction times for detecting the color change in the target were faster when the frame surrounding the target changed color relative to when the change occurred in the other frame. Importantly, a comparable cueing effect was obtained by the motion of the frame. The reaction times for detecting the change in the target color were also faster when the target appeared in the frame that rotated than when it appeared in the frame that remained stationary. Moreover, this cueing effect did not occur when the frames changed to a task-irrelevant color, confirming that subjects had adopted a control set that was specific to the target color. In summary, these results clearly demonstrate that the initiation of movement by objects attracted attention regardless of the observers' task set. 

Nonetheless, our current knowledge regarding whether stimulus motion causes stimulus-driven attentional capture is not entirely coherent. The traditional finding reported by Folk et al. (1994) demonstrated that attentional capture derived from stimulus motion was eliminated when observers were set to search for a color target (contingent involuntary orienting hypothesis; Folk et al., 1992). The cueing effect was obtained only when the properties determining the cue matched those determining the target and when the information supplied by the motion was not crucial. Although other studies (Abrams & Christ, 2003, 2005; Al-Aidroos et al., 2010; von Mühlenen & Lleras, 2007) have shown that stimulus-driven attentional capture can occur in response to various types of motion, these studies, which include Folket al.'s (1994) original study, used tasks in which the distractors spatially overlapped with (or were located adjacent to) potential targets. The inconclusiveness regarding whether motion captures attention may be related to this spatial proximity between target and distractors. Therefore, the present study adopted a different strategy: to test the notion that stimulus motion penetrates observers' attentional set. 

We argue that if the effectiveness of attentional set for color over stimulus motion were not limited to the original condition used by Folk et al. (1994), an equivalent avoidance of attentional capture by stimulus motion would be observed when an attentional set for color was adopted, regardless of the paradigm. Alternatively, if stimulus motion were imperative, then attentional capture would occur regardless of the adoption of an attentional set to search for color, even in a task that has been established to demonstrate the strong contingency effect of task set. Specifically, we used a task involving identification of a colored letter embedded in a rapid stream of nontargets, heterogeneously colored letters (Folk, Leber, & Egeth, 2002) and added a global optic flow in the periphery as a distracting event. In this paradigm, the effectiveness of attentional set for a specific color has been demonstrated by inserting a distractor frame in which critical item(s) appeared in the periphery shortly before the target display appeared. Typically, the accuracy with which a (e.g., green) target is identified is impaired when the distractor frame contains an item of the same color as the target (green) relative to when it contains an item of another color (e.g., red). This paradigm has been established as a way to induce observers to search for a specific color (Du & Abrams, 2008; Folk, Ester, & Troemel, 2009; Lamy, Leber, & Egeth, 2004; Leber, Kawahara, & Gabari, 2009). Importantly, participants had no incentives for diverting their attention to the periphery during performance of this task. This aspect of the task renders the present procedure more suitable for testing the question of whether task-irrelevant motion captures attention than were the cueing paradigms used in previous studies (Abrams & Christ, 2003, 2005; Al-Aidroos et al., 2010; von Mühlenen & Lleras, 2007), in which potential target locations overlapped (or were grouped into one object because of proximity) with the moving distractors. 

In the present study, the attentional set was to search for and identify a green letter embedded in a stream of heterogeneously colored nontargets. Therefore, the global motion flow in the periphery was completely irrelevant to the letter-identification task. If the motion signal occurring briefly before the target impaired identification accuracy, this finding of attentional capture would support the notion that motion is an important determinant of attention deployment (Al-Aidroos et al., 2010; von Mühlenen & Lleras, 2007). In contrast, if any changes in peripheral global motion had no effect, such a null finding with respect to attentional capture would strongly support the hypothesis of contingent involuntary orienting (Folk et al., 1992; Folk et al., 1994). Moreover, the present study included different types of global optic flows (e.g., expansion, cessation, and contraction) to examine whether the effect of motion, if any, would be modulated by these types of manipulations. We hypothesized that the effect of motion would differ across types because several studies have found that the expansion/contraction of local motion stimuli has asymmetric effects (Abrams & Christ, 2005; Franconeri & Simons, 2003). Thus, similar asymmetries may be observed in the context of stimuli involving global motion display. 

Experiment 1: Interference from task-irrelevant commencement of optic flow

In the first experiment, participants viewed a field of random dots that could expand, contract, or remain static while a central stream of letters was presented rapidly. We predicted that the accuracy with which targets were identified would not be affected by the saliency of the commencement of optic flow if the deployment of attention were governed exclusively by top-down control (i.e., an attentional set that involves searching for the target color), as the contingent involuntary orienting hypothesis posits. Alternatively, if attentional deployment were susceptible to the saliency of the stimulus appearing in the display, an abrupt change in the optic flow shortly before the appearance of the target would cause a diversion of attention, resulting in impaired identification accuracy for the target. Because we did not know when the change should be inserted into the optic flow, we varied the temporal lags between the onset of the motion signal and the onset of the target. 

Method

Participants

Fourteen undergraduate and graduate students (three females and 11 males, between 21 and 24 years of age) participated; none were informed of the aim of the experiment. All participants in this and the following experiments provided written informed consent and had normal or corrected-to-normal vision. 

Stimuli and apparatus

Stimuli were generated by Microsoft Visual Studio 2008 with OpenGL and displayed at the center of a monitor (Mitsubishi RDF223H, 1280 × 960 pixels, 60 Hz) controlled by a computer that also collected responses. The stimuli consisted of a rapid central stream of 20 heterogeneously colored uppercase letters and a peripheral field of dots. The letters subtended approximately 1° of the visual angle in height at a viewing distance of 45 cm. 

The target letter, which appeared at a position of 11 to 15 in the stream, was green, and nontarget letters were red, yellow, gray, or light blue. Each letter was displayed for 67 ms and was separated from each other letter by inter-stimulus intervals of 33 ms, yielding a presentation rate of 10 items per second. These temporal parameters were maintained throughout the study (Experiments 1–4). 

We presented an optic flow consisting of a cloud of 21,320 dots around the letter stimuli (see Figure 1); these were randomly placed in a virtual space (100 cm width × 100 cm height × 200 cm depth) with the constraint that the dots would not be located in the “neutral zone” (approximately 1° in radius) at the center of the screen, which prevented dots from overlapping with letters. The dots were white (71 cd/m²) and were presented on a black background (0.001 cd/m²). The viewpoint was initially set at the horizontal and vertical center of the front end of the virtual space, and the dots immediately in front of observers were always visible to 100 cm in depth (depth clipping: 0 to 100 cm). Approximately 2,000 dots were always in the field of vision. The viewpoint of observers moved forward and backward in the space in a linear fashion, causing an expanding and a contracting optic flow, respectively. The speed of viewpoint motion was 120 cm/s and the duration was 100 ms. If not otherwise noted, the apparatus, the parameters of the optic flow, and the central letter task were the same throughout the present study. 

Figure 1

 

Schematic stimuli and sequence of a trial under the expansion condition. The letter “Y” was actually green and target in this example. The other letters randomly appeared as red, yellow, gray, or light blue (see the text for detail).

Design and procedure

Two within-subject variables were examined: type of motion (expansion, contraction, or no-motion conditions) and temporal lag between the target and distractor (−300, −100, +100, or +300; negative values mean that the distractor preceded the target). These two factors were randomly intermixed within an experimental session. 

Before starting the session, instructions were read to participants by the experimenter and were also displayed on the computer screen. At the beginning of each trial, a fixation cross was displayed at the center of the screen. Participants initiated each trial by pressing the spacebar. After the key press, a rapid stream of letters was presented at the center of the screen. The target was preceded by 10–14 randomly selected nontarget letters. A distracting event involving motion (i.e., expansion or contraction) occurred 300 ms or 100 ms before (or after) the appearance of the target during trials including a distractor. The onset and offset of the motion event were synchronized with the onset and offset of a central letter. Therefore, the duration of the motion event was 100 ms under the no-motion condition: the peripheral dots remained stationary throughout the trial. The task of the participant was to identify the green letter by pressing the corresponding keys on the keyboard after the stream ended. The colors of nontarget letters were randomly chosen in every trial, with the constraint that consecutive letters would not be the same color. A brief feedback tone was provided following an incorrect response. After each response, the screen was blacked out for a 1000 ms inter-trial interval, during which only a fixation point appeared. Participants were tested individually in a dark room for three 15 min sessions consisting of 10 practice trials followed by 120 experimental trials. 

Results

Mean percentages of correct target identifications are plotted separately in Figure 2 as a function of target-distractor lag for the three motion conditions. An analysis of variance (ANOVA) was conducted on these accuracy scores with two within-subject variables (motion: expansion, contraction, or no-motion; lag: −300, −100, +100, or +300 ms). The analysis revealed a significant interaction between motion and lag, F(6, 78) = 2.37, η_p². = 0.15, p < 0.05. A post hoc analysis (Ryan's method, alpha = 0.05) revealed that the mean accuracy under the expansion condition was lower than that under the no-motion [t(13) = 4.01, r = 0.74, p < 0.001] and the contraction [t(13) = 3.32, r = 0.68, p = 0.001] conditions at a lag of −300 ms. No other statistical differences were found between other lags. In summary, the identification accuracy was impaired by the commencement of expanding motion that occurred 300 ms before target onset. No such effect was found under the contraction and no-motion conditions. 

Figure 2

 

Mean percentages of correct responses as a function of target-distractor lag in Experiment 1 for the expansion, contraction, and no-motion conditions. Negative lags indicate conditions under which the distractor preceded the target. Error bars indicate standard errors.

Discussion

Experiment 1 produced two important findings. First, the results indicated a clear hallmark of attentional capture. When the flow of peripheral dots expanded for 100 ms at 300 ms before the appearance of the target in the central stream, target identification was impaired relative to when no such motion occurred. Importantly, attentional capture occurred even when the commencement of the peripheral optic flow was unrelated to the central task. In other words, the participants established an attentional set, known as the feature search mode (Bacon & Egeth, 1994; Folk et al., 2002), to search for a green item. Nonetheless, this attentional set was ineffective for avoiding attentional capture by a task-irrelevant stimulus, the commencement of expanding optic flow in the periphery. It was highly unlikely that participants had adopted the singleton-detection mode in that they successfully identified green targets among nontarget letters in heterogeneous colors. In this respect, the present result is inconsistent with the contingent involuntary orienting hypothesis (Folk et al., 2002; Folk et al., 1992; Folk et al., 1994) and suggests that exogenous stimulus motion can break through an endogenously established attentional set. 

It should be noted that the attentional capture observed here could be attributed to a spatial diversion of attentional focus (e.g., Folk et al., 2002; Theeuwes, 1992), and/or to a nonspatial effect (e.g., Folk, Leber, & Egeth, 2008). Determining the involvement of spatial shift would require specific investigations and thus go beyond the scope of the present study. A critical difference from the previous studies (Folk et al., 2002, 2008; Theeuwes, 1992) is that the present study included no overlap between the locations of the target and the distracting event or between the task set used to search for the target and the definition of the distracting event. Specifically, target locations were adjacent to or overlapped with the distractor locations in the studies conducted by Folk et al. (2002, 2008) and Theeuwes (1992). The task set overlapped at least partly with the definition of the distracting event in the studies by Folk et al. (2002, 2008) and Theeuwes (1992). The present experiment demonstrated that an attentional capture effect could occur even when a distracting event occurred in completely task-irrelevant spatial locations as task-irrelevant stimulus attributes. 

Second, the attentional capture effect was observed only when the static peripheral dots started to expand. No such capture was observed when the peripheral dots started to contract. This pattern of the results is consistent with previous findings that the visual system was more sensitive to expanding than to contracting motion (e.g., Abrams & Christ, 2005b; Franconeri & Simons, 2003; Takeuchi, 1997; von Mühlenen & Lleras, 2007). According to the behavioral urgency hypothesis, the onset of expanding motion signals that something may be approaching the observer, which leads attention to be diverted to the motion to enable immediate action. The finding that the onset of contracting motion signal did not cause attentional capture is important because it means that a merely transient event in the display, such as a luminance change, was insufficient to producing attentional capture. Rather, the present results suggest that the onset of stimulus motion involving expanding optic flow was critical for producing attentional capture. 

Experiment 2: Temporal changes in optic flow capture attention regardless of speed

Experiment 1 demonstrated that a 100 ms expansion in optic flow captured attention even when participants adopted an attentional set designed to search for a stimulus value (i.e., green) in a different stimulus domain (i.e., color) and to exclude other stimuli in task-irrelevant domains (e.g., motion). This finding strongly supports previous findings that demonstrated attentional capture in visual search and cueing paradigms (Abrams & Christ, 2005b; Franconeri & Simons, 2003; Takeuchi, 1997; von Mühlenen & Lleras, 2007). In Experiment 2, we replicated the finding of Experiment 1 by using slightly different speeds of optic flow. 

Method

In the present experiment, we manipulated the speed of optic flow, which could abruptly expand at speeds of 60, 120 (identical to Experiment 1), or 240 cm/s. We also included a no-motion condition as a control condition. We omitted the lag manipulation and instead designed the task so that the motion of the optic flow, if any, could occur for 100 ms at 300 ms before the onset of the target letter. The observers' task was to identify a letter of a specific color (green) in a stream of the nontarget letters in heterogeneous colors. Ten undergraduate and graduate students (two females and eight males, between 21 and 24 years of age) participated; all were naive to the purpose of the present experiment. Six of the subjects had participated in Experiment 1 before participating in this experiment, four subjects participated in Experiments 3b and 4 (see below), and three subjects participated in Experiment 5 (see below). Participants received a few practice trials and 120 testing trials consisting of 30 randomly ordered repetitions of the four motion conditions. 

Results and discussion

Mean percentages of correct target identifications under four motion conditions (three expansion and the no-motion control conditions) are shown in Figure 3. A one-way ANOVA, conducted on the accuracy scores for the four motion conditions revealed a significant main effect of motion, F(3, 27) = 6.12, η_p². = 0.40, p = 0.003. The same multiple comparisons used in Experiment 1 revealed that the mean accuracy under the three expansion conditions was lower than that under the no-motion condition, ts(9) > 2.44, ps < 0.05. No other statistically significant differences among the three expansion conditions were found. Thus, the present results indicate that attentional capture occurred under the expansion conditions, regardless of speed, confirming that the attentional capture by the onset of task-irrelevant expanding motion signal observed in this study is a robust effect. 

Figure 3

 

Mean percentages of correct responses as a function of the motion condition in Experiment 2. Error bars indicate standard errors.

Experiment 3: Does commencement or cessation capture attention?

In the present experiment, we examined whether the abrupt onset and/or offset of motion signal was critical for a specific type of attentional capture. Specifically, the temporal changes in optic flow in Experiments 1 and 2 consisted of abrupt commencement and cessation after 100 ms. Therefore, it was unclear which components were critical to induce the attentional capture. To clarify this issue, Experiment 3a included two conditions: the commencement condition, in which the optic flow started to expand, and the control condition, in which the dots remained stationary. In Experiment 3b, two different conditions were used: the cessation condition, in which a trial started with an expanding optic flow that then ceased to move, and the control condition, in which the optic flow continued to expand. 

We examined this issue because we have previously demonstrated (Inukai, Kumada, & Kawahara, 2010) that attentional capture by the onset of a static distractor character in the periphery was reduced when the distractor disappeared after 172 ms or remained visible compared to when the distractor disappeared after 43 ms. This effect of distractor duration suggested that transient signals arising from the disappearance of distractors directly following their onset plays a critical role in the attentional capture by peripheral events. If the same rule applies here, we would expect the duration of the motion event to affect attentional capture. Our prediction is as follows: if abrupt commencement and cessation of motion are necessary, no capture effect would be obtained in the present experiment. If either commencement or cessation of motion were sufficient, the capture effect would be obtained in either or both of the Experiments (3a and 3b). 

Method

Experiment 3a: We manipulated the types of motion. Specifically, under the commencement condition, a trial started with a static field of dots that abruptly started to expand and then continued its expansion. Under the control (i.e., no-motion) condition, the dots remained stationary. The change in motion, if any, occurred 300 ms before the onset of the target. Participants identified the green letter (target) in the central stream. Thirteen undergraduate and graduate students aged between 21 and 24 years participated (four females and nine males). Seven subjects had participated in Experiment 1, six subjects had participated in Experiment 2, seven subjects participated in Experiment 3b, four subjects participated in Experiment 4, and four subjects participated in Experiment 5. Participants received a few practice trials and then engaged in 60 testing trials consisting of 30 repetitions of the two motion conditions presented in random order. 

Experiment 3b: This experiment was identical to Experiment 3a except for the types of motion. Under the cessation condition, a trial started with an expanding optic flow of dots, with abruptly ceased to move. Under the control (i.e., continuous-motion) condition, a trial started with an expanding optic flow, which continued to expand. Sixteen undergraduate students aged between 21 and 24 years participated (seven females and nine males). Twelve subjects had participated in Experiment 1 before participating in this experiment and nine subjects participated in Experiment 5. 

Results and discussion

The mean percentages of correct target identifications in Experiment 3a are shown as a function of the motion condition in Figure 4a. The data were analyzed with t tests. The analysis indicated that accuracy under the commencement condition was marginally significantly lower than that under the no-cessation condition, t (12) = 2.06, r = 0.51, p = 0.062. Thus, attentional capture can be caused by merely initiating optic flow. Figure 4b presents the mean percentages of correct target identifications in Experiment 3b as a function of motion condition. A t test on these scores revealed that accuracy under the cessation condition was significantly lower than that under the no-cessation condition, t(15) = 2.62, r = 0.56, p = 0.019. 

Figure 4

 

Mean percentages of correct responses as a function of the motion condition in Experiment 3 (a) for the contrast between the commencement and the no-motion conditions and (b) for the contrast between the cessation and continuous motion conditions. Error bars indicate standard errors.

These results suggest that any change in the expanding optic flow (i.e., commencement or cessation) was sufficient to produce attentional capture. Unlike the case with regard to static distractors (Inukai et al., 2010), mere initiation or cessation of a task-irrelevant optic flow caused attentional capture. Thus far, the present results have been consistent with the behavioral urgency hypothesis, which posits that expanding motion signals may contain important clues about objects that may be approaching observers, thus attracting attention and enabling immediate action. 

Experiment 4: Does acceleration or deceleration capture attention?

Experiment 3 demonstrated that mere commencement or cessation of optic flow caused attentional capture. Nonetheless, it was unclear whether initiating of optic flow or changing its speed was critical for attentional. To examine this issue, we compared three (acceleration, deceleration, and control) motion conditions. If any change in speed were sufficient for attentional capture, this phenomenon would be observed in the present experiment. 

Method

This experiment differed from Experiment 3 in terms of changes in the speed of the motion. Three motion conditions were used. Under the control (no-motion) condition, a trial started with an optic flow expanding at a speed of 120 cm/s. As the central stream of letters proceeded, the peripheral optic flow continued. Under the acceleration condition, the peripheral optic flow accelerated from 120 cm/s to 240 cm/s 300 ms before the onset of the target and maintained this speed until the end of the trial, whereas under the deceleration condition, the optic flow decelerated from 120 cm/s to 60 cm/s 300 ms before the onset of the target and maintained that speed until the end of the trial. These speed changes occurred abruptly as a step function. Ten undergraduate and graduate students aged between 21 and 24 years participated in the present experiment (one female and nine males). All subjects had participated in Experiments 1 and 3b before participating in this experiment, and nine subjects participated in Experiment 5. Participants, who were instructed to identify a green target in a letter stream received a few practice trials and then engaged in 90 testing trials, which consisted of 30 repetitions of the three motion conditions, intermixed in a random order. 

Results and discussion

Mean percentages of correct target identifications under the three motion conditions (acceleration, deceleration, and no-change) are shown in Figure 5. A one-way ANOVA on accuracy scores indicated that the main effect was not significant, F(2, 18) = 0.217, p = 0.807. Thus, the change in the speed of the optic flow failed to capture attention. We suggest that qualitative changes in the expanding optic flow, such as the onset and cessation of motion, rather than quantitative changes, such as the doubling of acceleration or the halving of deceleration, are critical for the occurrence of task-irrelevant attentional capture. These results are consistent with findings indicating that the ability to detect an acceleration produced by an instantaneous change in velocity is less sensible than is velocity discrimination (Gottsdanker, 1956; Snowden & Braddick, 1991; Werkhoven, Snippe, & Toet, 1992). 

Figure 5

 

Mean percentage of correct responses as a function of motion condition in Experiment 4. No change means that the optic flow remained expanding throughout the trial. Error bars indicate standard errors.

Experiment 5: The effect of the focus of expansion

In the experiments described thus far, the focus of expansion (FOE) of optic flow coincided with the location at which target and nontarget letters appeared. Given that the incoming field of a moving random-dot display induces a spatial shift of attention toward that location (Watanabe, 2001), this overlap between the critical location (i.e., the display center, where the central letter-identification task occurred) and the incoming field of optic flow may have caused artifacts. For example, participants may have integrated the central letter-identification task with the viewing of task-irrelevant peripheral motion into a single whole display. Thus, it is unclear whether the superimposition of the FOE on the target location was critical for the attentional capture by the task-irrelevant optic flow. In other words, attentional capture may have been caused by the occasional sharing of the FOE with the critical location at which target identification occurred. We tested this hypothesis by misaligning the FOE and the target location. If this integration hypothesis were confirmed, optic flow display would not be associated with attentional capture under such a misaligned condition. Otherwise, we would observe attentional capture as similar to that found in previous experiments (Experiments 1, 2, and 3). 

Method

The procedures were same as those used in Experiments 1, 2, and 3 except that the FOE of the optic flow was randomly misaligned by 5° either to the right or to the left from the center of the display. The stream of the target and nontarget letters was presented at the center of the display. Under the expansion condition, the optic flow started to expand 300 ms before the onset of the target. Under the no-motion condition, the field of random dots remained static throughout the trial. Eleven undergraduate and graduate students aged between 21 and 24 years participated (two females and nine males). All subjects had participated in Experiment 1 before participating in this experiment. Participants received a few practice trials and then engaged in 60 experimental trials consisting of 30 randomly ordered repetitions of the two motion conditions. 

Results and discussion

Mean percentages of correct target identifications under the two motion conditions are shown in Figure 6. A t test revealed significantly lower rates of accuracy under the expansion condition than under the no-motion condition [t(10) = 3.28, r = 0.72, p = 0.008], indicating that attentional capture occurred when the FOE and the location of the target were misaligned. The present results suggest that the coincidence of the FOE and the critical location of the primary task (i.e., letter identification in the present study) were unrelated. Attentional capture by the onset of optic flow occurred regardless of the FOE. 

Figure 6

 

Mean percentage of correct responses as a function of motion condition in Experiment 5. Error bars indicate standard errors.

General discussion

The present study examined whether a task-irrelevant event, the commencement of optic flow, can be ignored while performing a letter identification task, as claimed by adherents of the contingent involuntary orienting hypothesis (Folk et al., 2002; Folk et al., 1992; Folk et al., 1994), or whether such an irrelevant event captures visual attention regardless of observers' attentional set, as claimed by those who endorse a broad range of views that emphasize the behavioral urgency of moving items (e.g., Abrams & Christ, 2005b; Franconeri & Simons, 2003; Takeuchi, 1997; von Mühlenen & Lleras, 2007). 

Moving beyond the previous studies, the present study used a task in which the motion event was completely unrelated to the critical task that served as the index of the deployment of attention (i.e., to the task or to a distracting motion event). Specifically, observers searched for and identified a green letter embedded in a rapid stream of nontarget heterogeneously colored letters presented at the center of the display. This critical task was combined with an abrupt change in the peripheral distracting optic flow. Under this condition, observers needed to devote attention to the center of the screen because the targets never appeared in the periphery. This condition contrasted sharply with those of previous studies (Abrams & Christ, 2005b; Al-Aidroos et al., 2010; Folk et al., 1992; Folk et al., 1994; Franconeri & Simons, 2003; Takeuchi, 1997; von Mühlenen & Lleras, 2007), most of which located potential targets at the same place as allegedly task-irrelevant distractors. For example, in the task used by Franconeri and Simons (2003), participants searched for a target item among spatially distributed nontargets and the potential locations of the target and distractor could coincide. In the task used by von Mühlenen and Lleras (2007), the targets could appear either on the left or on the right of the screen, and the moving dots also appeared in those areas. Although the side of the motion did not provide any useful information for detecting the target, observers needed to monitor both sides of the screen, and motion events occurred at both locations. Similarly, in the task used by Al-Aidroos et al. (2010), participants identified one of two items presented on the left or the right of the screen, and the distracting motion was created by rotating one of the frames outlining each of these two target locations. Therefore, the locations of potential targets and nontargets were very close in space in terms of coarse spatial coordinates (Atkinson & Braddick, 1989) although, strictly speaking, their locations differed slightly. Unlike in these cases, in the present study the peripheral area was occupied exclusively by the task-irrelevant optic flow. The targets never appeared in the periphery. Therefore, the present study was a more stringent test of whether task-irrelevant motion interferes with the central letter identification. We reasoned that if deployment of attention were governed by a top-down strategy, as the contingent involuntary hypothesis assumes, the accuracy of letter identification would be unaffected by optic flow. 

The results of the five experiments in this study are clearly inconsistent with the contingent involuntary hypothesis. Indeed, we found attentional capture caused by the commencement or cessation of peripheral optic flow when participants were engaged in a feature search (i.e., searching for a green letter at the center of the display; Bacon & Egeth, 1994; Folk et al., 2002). When the peripheral dots expanded for 100 ms and preceded the appearance of the target in the central stream by 300 ms, the target identification was impaired relative to when no such motion occurred (Experiment 1). Additionally, attentional capture by task-irrelevant optic flow was observed in response to the cessation of different speeds of motion (Experiment 2). Moreover, the focus of expansion was unrelated to the capture effect (Experiment 5). Therefore, under the present circumstances, an attentional set configured to detect and identify a letter of a specific color was ineffective for avoiding attentional capture by a task-irrelevant onset of motion signals in expanding optic flow appearing in the periphery. 

The present results cannot be explained by the notion that the emergence of an object requires attention. Specifically, this explanation argues that when an object appears, visual attention is directed to create an object file to represent the properties of such an object (Hillstrom & Yantis, 1994; Kahneman, Treisman, & Gibbs, 1992). This explanation cannot easily be reconciled with the present findings showing that contraction did not cause capture (Experiment 1) but that cessation did cause capture (Experiment 3b). If the emergence of an object created by the commencement of motion mattered, the capture effect would have been observed regardless of motion direction (i.e., expansion or contraction), and no capture would have been associated with cessation of motion. The present results are consistent with the view that behaviorally urgent moving items capture attention (e.g., Abrams & Christ, 2005b; Franconeri & Simons, 2003; Takeuchi, 1997; von Mühlenen & Lleras, 2007) and suggest that stimulus motion can penetrate an endogenously established attentional set. 

The results of the present five experiments delineate the properties that constitute behaviorally urgent motion. That is, expansion, rather than contraction, of the motion display captures attention regardless of the attentional control setting (Experiment 1). This finding is consistent with those of previous studies suggesting the behavioral significance of looming motion (Franconeri & Simons, 2003; Takeuchi, 1997; von Mühlenen & Lleras, 2007). The present study also reveals that mere commencement or cessation of motion was sufficient to produce the capture effect (Experiment 3). At a glance, this finding seems inconsistent with the results reported by Abrams and Christ (2003), which indicated that cessation of motion did not attract attention in a visual search task. We assume, however, that Abrams and Christ's (2003) study differed from the present study in many ways (e.g., type of motion, task, and spatial arrangement of displays) and that these differences may have contributed to the apparent inconsistency. Finally, the present study demonstrated that qualitative differences with respect to commencement, rather than quantitative changes in motion, were critical for producing the attentional capture effect (Experiment 4). This aspect has never been examined in previous studies. In a nutshell, the present results demonstrated that onset and offset motion signals are capable of automatically capturing attention. However, the present study does not deny the possibility that other types of motion also produce the same stimulus-driven capture, especially given previous findings showing that forms of local motion can also be particularly salient (e.g., Al-Aidroos et al., 2010; Folk et al., 1992). Thus, it is possible that task-irrelevant local motion outside of the spatial focus of attention can capture attention. 

Why does this qualitative/quantitative difference matter? We argue that human perception is more sensitive to qualitative (i.e., commencement or cessation) than to quantitative (i.e., velocity) changes. This asymmetry may be attributed to the fact that the minimum motion threshold or D_min is considerably lower than are the thresholds for velocity discrimination and visual acuity (Legge & Campbell, 1981; McKee, 1981; McKee & Nakayama, 1984; Nakayama, 1981). Moreover, the ability to detect an acceleration produced by an instantaneous change in velocity is less sensitive than is velocity discrimination (Gottsdanker, 1956; Snowden & Braddick, 1991; Werkhoven et al., 1992). Therefore, these differences in sensitivity in motion perception may be at least partly reflected in our findings. Another reason for the aforementioned asymmetry may relate the ecological importance of the qualitative commencement of stimulus motion. An abrupt commencement or cessation of motion can be caused by a crucial but unexpected incident, whereas a quantitative change in velocity usually reflects a variation in ongoing events. Thus, from an ecological perspective, it seems reasonable to neglect quantitative velocity changes but to be alert to qualitative motion changes even while attending to other objects or engaging in other tasks. 

One might argue that the stimulus-driven attentional capture observed in the present study may be explained by stimulus onset. Specifically, participants needed to identify a target letter in a rapid central stream of letters in which every letter had an onset. Therefore, detection of stimulus onset may have been a part of the attentional sets of participants. Given that stimulus onset and motion could serve as equivalent media to attract attention (Folk et al., 1994; Experiment 5), the participants may have been attuned to searching for any dynamic stimulus. Thus, the optic flow in the periphery would result in attentional capture because it belonged to the same category of dynamic stimuli. However, we do not believe that this condition is likely for the following reasons. First, if this were the case, any nonspecific dynamic stimulus would cause attentional capture. In fact, the present study showed attentional capture was selective for motion that expanded from a static display. Second, if this logic of shared dynamic features between the central stimulus and the peripheral optic flow were true, attentional capture would not have been contingent on the observers' use of a set attuned to color in the original task used by Folk et al. (2002), in which peripheral distractors also had an onset. In that task, peripheral distractors in a nontarget color did not capture attention even though the both central letters and peripheral distractors had an onset. Therefore, the interpretation relying on shared dynamic features is untenable. Instead, we suggest that the visual system is biased to divert attention to behaviorally urgent stimuli, such as an expanding optic flow (i.e., approaching motion toward observers). 

Another concern to be dismissed was that subject might actually include a kind of “looming” feature in their top-down set and thus letters that appear one after another at the same location may indeed be encoded as a stream of information that is moving forward in time. Given that Gibson and Amelio (2000) showed that even when participants adopt an attentional set for color, they may also include a display wide set for onsets, it seems possible that in the present experiments, in addition to the top-down set for color, subjects may have adopted a more general set for a type of “looming information.”¹ Although this is an intriguing idea, we believe this interpretation is untenable on the basis of the results of Experiment 3b. Specifically if participants adopt a top-down set for the “looming” feature, no capture should have been expected from the cessation of the optic flow. Inconsistent with this prediction, we found a capture effect in that experiment. Therefore, we can dismiss the concern of attentional set for looming, illusorily created from the search in the central rapid sequence. 

The present study extended the behavioral urgency hypothesis of moving stimuli (e.g., Al-Aidroos et al. 2010; Franconeri & Simons, 2003; Takeuchi, 1997; von Mühlenen & Lleras, 2007) in that observers could not prevent their attention from being diverted by task-irrelevant events occurring at locations that did not overlap with the current critical task. Al-Aidroos et al. (2010) offered an intriguing interpretation of stimulus-driven capture by irrelevant motion: they argued that attentional set cannot avoid capture from moving distractors in that attentional set is always attuned to new motion because such information carries behaviorally relevant signals. This interpretation would fit with the anecdote of the earthquake described in the Introduction. If the motion in the environment was relatively rare (Folk & Remington, 2006; Neo & Chua, 2006) and/or highly salient (Forster & Lavie, 2008; Theeuwes, 2010), attention would be captured by such an event. In other words, less urgent behavioral information would no longer capture attention. As Al-Aidroos et al. (2010) suggested, participants would be able to resist attentional capture by a motion event if that event occurred very frequently and was therefore highly predictable. Our laboratory is currently examining the effect of frequency on attentional capture by optic flow. 

Acknowledgments

This work was supported by Grant-in-Aid for Scientific Research (C) 11014283. 

Commercial relationships: none. 

Corresponding author: Jun Kawahara. 

Email: [email protected]. 

Address: Department of Psychology, Chukyo University, 101-2 Yagoto, Showa, Nagoya, Japan. 

Footnotes

¹  We thank an anonymous reviewer for pointing out this possibility.