Flash-lag effect: complicating motion extrapolation of the moving reference-stimulus paradoxically augments the effect
- 186 Downloads
One fundamental property of the perceptual and cognitive systems is their capacity for prediction in the dynamic environment; the flash-lag effect has been considered as a particularly suggestive example of this capacity (Nijhawan in Nature 370:256–257, 1994, Behav Brain Sci 31:179–239, 2008). Thus, because of involvement of the mechanisms of extrapolation and visual prediction, the moving object is perceived ahead of the simultaneously flashed static object objectively aligned with the moving one. In the present study we introduce a new method and report experimental results inconsistent with at least some versions of the prediction/extrapolation theory. We show that a stimulus moving in the opposite direction to the reference stimulus by approaching it before the flash does not diminish the flash-lag effect, but rather augments it. In addition, alternative theories (in)capable of explaining this paradoxical result are discussed.
Being able to predict the future values of the environmental stimuli and cues is a fundamental capacity of an organism and an individual both in natural and social environment. When we foresee how environmental circumstances and changing objects will appear in the near future, our chances for survival and practical success increase. For example, catching-, avoiding- or escaping behavior benefits from this considerably (Nijhawan, 2008; Nijhawan & Khurana, 2010). This adaptively important ability is thought to be implemented by some dedicated information processing routines used by neural sensory and motor mechanisms in the brain in order to cope with the fast-changing environmental stimulation. Interestingly, as a result of these routines, in special perceiving circumstances perceptual phenomena emerge where spatial and/or temporal values of the coordinates of objects represented in visual awareness are not veridical, but appear distorted and thus form illusory experiences (Bachmann, Breitmeyer, & Öğmen, 2007; Nijhawan & Khurana, 2010). It remains a matter of hot debate whether the flash-lag effect (FLE) also results from the workings of the extrapolative/predictive mechanisms or not or some other mechanisms are needed to explain it (Alais & Burr, 2003; Arnold, Ong, & Roseboom, 2009; Bachmann, Luiga, Põder, & Kalev, 2003; Baldo & Klein, 2008; Brenner & Smeets, 2000; Eagleman, 2008; Eagleman & Sejnowski, 2000, 2007; Enns, Lleras, & Moore, 2010; Gauch & Kerzel, 2009; Kanai, Carlson, Verstraten, & Walsh, 2009; Kirschfeld, 2006; Kreegipuu & Allik, 2004; Krekelberg & Lappe, 1999; Linares, López-Moliner, & Johnston, 2007; Murakami, 2001; Nijhawan, 1994; Nijhawan & Khurana, 2010; Patel, Öğmen, Bedell, & Sampath, 2000; Shi & Nijhawan, 2008; Watanabe & Yokoi, 2008; Whitney & Cavanagh, 2000).
In FLE, an object briefly flashed alongside a continuously moving reference object appears to lag behind the moving one. Because FLE appears as a kind of “microcosm” of many different problems in research on perception, sensorimotor coordination, attention, cognition and consciousness, the solution to this controversy is quite important.
Among the several interpretations of the FLE, the most influential one says that because of involvement of the mechanisms of extrapolation and visual prediction, the moving object is perceived ahead of the static one (Nijhawan, 1994, 2008). This is carried out by a modular system of interacting internal models set for predictive or anticipatory control (Nijhawan, 2008). Thus, when we consider extrapolation/prediction mechanisms as they are described in the literature it soon appears that there may be a variety of them (Nijhawan & Khurana, 2010), including the mechanisms shifting peaks of neural population responses, shifting receptive fields, spreading neural facilitation laterally, integrating information from multiple sources to feed higher level internal models, etc. (Nijhawan, 2008). Unfortunately, the works developing the prediction account of FLE are not specific enough about the mechanisms involved. This makes experimental testing of the extrapolation/prediction account of FLE very difficult. But we have to start from somewhere and we will do that without pretending to be able to test all possible explanatory mechanisms of FLE belonging to the family of the extrapolation/prediction account in a single study. For example, in one case the putative mechanism of extrapolation may belong to the ones typical for motion-detecting mechanisms with opponent coding of motion directions and lateral-inhibitory interaction (Mather, Verstraten, & Anstis, 1998). (The opponency of the motion-sensing mechanisms is robustly evidenced by the well-known motion aftereffect.) If signals or commands for extrapolation come from the motion analyzer, value of extrapolation should depend on motion vectors. In this case the computed spatial increment leading to the advancement of the future predicted position of an object moving along a trajectory would be canceled (“nullified”) if another object having an opposite motion vector moves towards the first moving object by approaching it. This should result in the absence of FLE in perception, provided that the extrapolation theory of FLE would stick to this mechanism as its explanatory resource. Let us term this explanatory resource—the mechanism of a low-level, mutually inhibitory opponent coding of motion vectors—as M1. This M1 computes extrapolation by using signals received from the mechanisms that encode motion direction and if these signals are mutually incompatible, extrapolation cannot be effectively executed. In order to test the validity of M1 as used for explaining FLE, an experiment should be run where extrapolation by M1 before flash is canceled (nullified) by an oppositely moving stimulus—a potential other reference. (In studying FLE, the position of the flash has to be evaluated against a reference object that is moving in a certain direction.)
In the other case two mutually opposite extrapolatory computations might be run in parallel and independently by some low-level mechanism void of opponency relations, each leading to the extrapolatory effect of its own. If this kind of low-level, non-interacting directionally sensitive mechanism is involved, an important question is whether the putative effect of prediction on the perception of the spatial position of a moving stimulus, even if not counteracted at the same low level, could be sensitive to motion direction uncertainty because of an influence by a higher level processes related to the cognitive inference mechanisms at the decision-making level. (In the present context, the inference about the future direction of motion of the already moving reference stimulus is aided by an internal cognitive model of the dynamic environment. Such models produce good fit with experimental data when they are built to be strongly dependent on the level of uncertainty arising at different stages of sensorimotor and perceptual processing (Orbán & Wolpert, 2011; Sturz & Bodily, 2010). One typical source of uncertainty is ambiguity present in the environmental cues. Thus, an a priori directional uncertainty refers to the number of potential remaining post-flash directions of reference motion indicated by the pre-flash motion vectors and, as such, this is a variable influencing the effects of the internal model. If there are two pre-flash directions, uncertainty is twice as large compared with when there is only one pre-flash direction. Uncertainty is involved in the model ability to predict future motion of the reference stimulus.)
For example, if higher level internal models are involved, is the effect of extrapolation (a) equally strong with only one and with more than one different non-interacting motion directions (i.e., extrapolation being insensitive to the uncertainty with which the decision mechanism operating on motion sensors’ signals has to cope with) or (b) does its magnitude depend on this uncertainty? In other words, does a higher-order uncertainty-sensitive perceptual decision mechanism appear to participate in the extrapolative prediction or not? If yes, then the magnitude of extrapolation should decrease with the uncertainty level with which the decision mechanism is confronted. Let us term this mechanism M3. In this case an experimental manipulation introducing uncertainty to the pre-flash motion directions of the moving reference stimuli should diminish FLE magnitude. If not, FLE magnitude should not depend on the level of motion uncertainty specified by the number of pre-flash motion trajectories. In this latter case we term the mechanism involved in perceptual evaluation of the events in a FLE display as M2.
In the present study we address these problems by reporting experimental results contradicting two of the three possibilities described above—the extrapolation/prediction theory that assumes low-level perceptual extrapolation based on opponent mutually inhibitory mechanisms of neural computation (M1) and the higher level decisional theory assuming the effect of uncertainty of motion vectors on the value of perceptual extrapolation mediated by an internal cognitive model (M3).
In addition to the main aim of the present study—testing the validity of the three putative versions of the extrapolation account—we also want to see how the other well-known FLE theories would succeed in predicting and explaining the results obtained by the new experimental method. Before moving on to the experimental part, let us briefly describe these other theories and their predictions.
The postdiction theory of the FLE (Eagleman, 2008; Eagleman & Sejnowski, 2000, 2007) says that novel sudden stimuli, acting as a kind of instigators of query for what happened in the visual field reset the integrative computations the brain uses for building up a veridical perceptual model of the external visual events. Because it takes time to collect different features of the environment (including motion cues) the complete perceptual model becomes available only with a delay. The delayed percept is attributed postdictively to the time of the flash, but it is a function of events that happen following the flash, for example representing the advanced position of the moving stimulus. In the trials when flash arrives simultaneously with or after the disappearance of the additional moving object, the two conditions should be identical for the post-diction model, and there should be no difference between them. For the trials with flash appearing before disappearance of the second object, it seems difficult to make any predictions.
The attentional theory (Baldo & Klein, 2008) explains FLE as follows. The flash captures attention which is needed to aid binding the flash and the moving reference into a unitary percept. As focusing of attention on the moving reference takes time, the unitary percept with information about both the flash and the moving stimulus will be fully attended only when the moving reference has changed its position to an advanced spatial location. Basically, the magnitude of the FLE depends on how long does it take for attention to catch the post-flash moving reference for perception. If the speed of attention deployment depended on load on attention or level of uncertainty, FLE should be larger in the condition with an additional object which may require the use of more attentional resources. However, provided that the pre-flash differences in the display events do not influence attentional focusing and the flash is the sole and main cause of capturing attention and influencing its focusing, the attentional theory predicts equal FLE between these conditions because the post-flash reference motion is equal.
Another theory explains FLE as a result of shorter visual latency to stimulus-in-motion compared to the static stimulus (Patel et al., 2000; Whitney, Murakami, & Cavanagh, 2000). Consequently, the moving stimulus becomes perceptually represented faster than a static stimulus and as a corollary to this, its perceived spatial position will be shifted along the motion trajectory away from the static stimulus. Because motion parameters of the reference stimulus are invariant between our two experimental conditions, differential latency theory predicts equal magnitude of the FLE for these conditions.
The positional sampling and averaging accounts of FLE (Brenner & Smeets, 2000; Krekelberg & Lappe, 1999) explain the perceptual relative advancement of the moving reference as a result of processing by a position computing mechanism: the continuously changing net result of the time-consuming computation about the precise location of a moving stimulus should indicate a position shifted forward along the motion trajectory. Provided that direction of motion is not a critical factor for the position-sampling device in this theory, this theory predicts that FLE should be stronger in the oppositely moving pre-flash references condition compared to the standard condition because with two oppositely moving stimuli one of the pre-flash instances of the moving stimulus acts as a pre-cue for the later-remaining sole moving stimulus and it may add to the signals that are used by the positional averaging processes. As a result, the moving reference stimulus appears as being shifted more ahead along its own motion direction when an oppositely moving other stimulus pre-dates its future position.
According to the perceptual retouch theory (Bachmann et al., 2003; Bachmann, 2010), consciousness-level perception of a stimulus is a result of the two interacting processes, each of them necessary, but individually insufficient for explicit perception—(a) the pre-consciously operating fast process responsible for formation of the object representation by feature integration in the cortical specific modules tuned to specific contents of the stimuli; (b) a slower operating non-specific thalamo-cortical modulation process acting upon specific pre-conscious representations built up by (a). The retouch theory predicts stronger FLE when an oppositely moving pre-flash stimulus antedates the future position of the remaining post-flash reference compared with the standard condition where there is no such a preliminary prime. This is because the opposite mover initiates the process (b) for the to-be-remaining reference ahead in time and from an advanced spatial position.
The updating theory states that the perceived timing of the appearance of a new object is delayed compared to the updating of perception of the changing properties of an already present object (Enns et al., 2010; Kanai et al., 2009). Construction of a new object representation (e.g., for the flash) requires additional time to establish a stable neuronal representation compared to the updating of the already present object (e.g., a reference stimulus in motion). This leads to the FLE. However, in our oppositely moving objects condition most of the features of the both moving pre-flash objects are the same and updating only the position indicated by the opposite mover from a location far ahead along the trajectory of the moving reference takes very little time. Thus, this theory predicts that FLE should have a larger magnitude in the condition of oppositely moving pre-flash objects.
Seven observers participated in the experiment (5 females, 2 males; age range 26–52 years). All of them reported normal or corrected to normal vision. The study received ethical approval from the University of Tartu Research Ethics Committee and the procedures used were in accordance with the Declaration of Helsinki.
Stimuli were presented on the computer monitor (Eizo Flex Scan, refresh rate 80 Hz). The viewing distance to the monitor was 60 cm. All stimuli were of positive contrast with luminance at 24.8 cd/m2; they were presented within a large area as background having luminance equal to 0.4 cd/m2.
In one spatiotemporal condition, two oppositely moving stimuli (pairs of horizontal rectangular bars with a small gap between them) started to move from peripheral locations horizontally towards the center of the visual field approaching each other until they became superimposed at the fixation, but only one of them continued its motion in the same direction as before (Fig. 1a). One pair of bars (green) discontinued its motion after fully overlapping with the other pair of bars (pink); the pair of bars (pink) that remained in motion after overlap was used as a reference stimulus for evaluating flash position. (When the green and pink pairs of bars underwent ever increasing spatial overlapping after their first contact, the to-be-remaining in motion, stimulus,—the pink one—remained always fully visible while the other one became gradually more occluded by it until full overlap with only the pink one being visible. While piloting the experiment, subjects were easily able to discriminate the green and pink objects and evaluate the flash relative to the pink one as a distinct separate object without fusing it with the green object. Subjectively, they described pink versus green objects as separate objects and not as one fused object. See also Fig. 3a. Moreover, there is strong evidence that subjects are able to attend to the target colors if task set requires so—e.g., Ansorge, Kiss, Worschech, & Eimer, 2011.) The choice of the post-flash direction of continued motion was chosen randomly for each trial. A single flashed object was a white bar centered with regard to central fixation and positioned so as to fit in (along the vertical dimension) between the two bars of the pairs of moving bar stimuli. Flash was presented either simultaneously (SOA = 0 ms) or 24, 48, or 72 ms before or after the moving stimuli overlapped. The SOA for each trial was chosen randomly. In the spatiotemporal control condition only one pink-colored moving stimulus was used similarly to the standard FLE displays.
The size of the object formed from a pair of horizontal bars was 1.1o (height) × 0.8o (width), and the size of the flashed bar was 0.3o × 0.8o. The distance from fixation to the first position of the moving stimulus at its motion onset was 15o. All stimuli were of positive contrast with luminance at 13.5 cd/m2 for the moving stimuli and 24.8 cd/m2 for the white flash; they were presented within a large area as background having luminance equal to 0.6 cd/m2. The time of excursion of the moving stimulus from its onset to the center of display at fixation was 3,000 ms, which equaled also to the time it took to complete the motion after the fixation point. The stimuli moved with constant velocity (approximately 5o/s).
The main experimental condition with oppositely moving stimuli and the control condition with one moving reference stimulus were run in blocks of trials in counterbalanced order between subjects. The task in both conditions was to evaluate whether the flashed object lagged behind the moving reference-colored (pink) object in space or was presented before the reference reached central position. Two-alternative forced choice responses were required. (With SOA = 0 ms between the flashed object and the moving object(s) (when at the central location) the 0.5/0.5 proportion of the “lagging behind”, and the “flashed before”, responses would be theoretically expected if neither flash-lag nor flash-lead percepts would dominate and reporting bias is absent.)
Each observer performed 2 blocks—one in the main condition and one in the control condition. Trials were initiated by observers clicking the trial-start icon on screen, which started visible motion of the moving pairs of bars from periphery towards the central fixation. Because FLE is considerably diminished when subjects track the moving stimulus with smooth pursuit eye movements (Nijhawan, 1997), the observers were instructed to carefully keep central fixation throughout each trial. Also, visual sensitivity is known to increase during pursuit (Schütz, Braun, Kerzel, & Gegenfurtner, 2008; Schütz, Braun, & Gegenfurtner, 2009) and thus may indirectly impact FLE, which means that fixation should be well kept. In piloting the experiment, we asked observers either to track the moving stimulus with known color (and known to remain in motion after the flash) or to carefully keep stationary fixation and compared the magnitudes of FLE. This preliminary procedure showed that FLE almost disappeared with pursuit. As in our formal experiment we obtained strong FLE, we conclude that subjects were successful in keeping their visual fixation. After stimuli presentation, subjects entered the response using computer keyboard by indicating whether flash was shifted to the left or right from the position of the reference. These responses were later coded as flash-lag or flash-lead for the flashed object depending on the direction of reference motion in each particular trial. Next trials were initiated by a mouse click again; a fixation dot (formed from the small area sized as 4 screen pixels) appeared in the center of the monitor, followed after 700 ms by the stimuli. 30 practice trials were administered to each observer before the experiment begun. There were 700 experimental trials for each subject, 4,900 trials all in all. The overall actual number of trials per each randomly chosen SOA value varied between 675 and 734. Thus, on average, there were 100 measurements per one SOA value for each subject and 700 measurements per one SOA value for the subjects group as a whole.
Results and discussion
The method was essentially the same as in Experiment 1, main condition, except for the following differences.
Four subjects with normal or corrected-to-normal visual acuity participated (females, age range 26—30).
The background, the bar-shaped elementary stimuli, luminances and movement speed parameters were a replication of what was used in Experiment 1, main condition. However, there were two basic sets of changes in terms of how stimulation elements were arranged and/or moving in space. In the condition opposite motion/continuous, the pair of bars with non-target color remained in motion after spatial overlap with the target pair of bars and moved until the target bars were switched off, reaching the lateral mirror image position of the last position of the target bars (see Fig. 3a). In the condition opposite motion/adjacent, the distance between the bars in pairs was increased so that it would be possible for the target pairs and non-target pairs to move along the movement vectors that are spatially adjacent, permitting the bars to avoid spatial overlap along their motion trajectory (see Fig. 3b). Similarly to the main condition of Experiment 1, the target pair (pink) remained in motion after the flash, but the other pair was switched off.
Procedure was essentially the same as in Experiment 1. In this experiment, the replicated conditions from Experiment 1 (no opposite motion; opposite motion/offset) together with new experimental conditions termed opposite motion/continuous and opposite motion/adjacent were run in blocks, with order of blocks counterbalanced between the subjects.
Our results show that the theory of automatic extrapolation/prediction based on the signals from low-level motion-coding mechanisms with opponent (mutually inhibitory) connections and assuming that two mutually incompatible path-prediction hypotheses cannot be simultaneously processed by M1 is inconsistent with FLE obtained with oppositely moving stimuli. However, if according to the mechanism M2 we assume that more than one mutually incompatible motion-paths hypotheses can be processed simultaneously (without causing a decrease in extrapolation) and applied only later (when only one of the oppositely moving objects remains in view) the FLE can be predicted by the extrapolation/prediction theory (Nijhawan, 1994, 2008) also with our novel experimental design. This is when we exclude the participation of the higher level internal models in influencing the mechanism M3 whose operation is sensitive to uncertainty about the motion vectors signaled by the lower level mechanism. However, if the prediction/extrapolation theory assumes involvement of M3, including the higher level integrating mechanisms (Nijhawan, 2008), then complicating its operation by increasing uncertainty about motion direction should nevertheless decrease FLE. This should happen even when the lower level non-interacting motion channels without mutual inhibitory interactions are the basis for sampling the motion signals. The initial conclusion from our data showing that oppositely moving pre-flash reference objects do not eliminate or decrease FLE reveals that mechanism M2 can be used for explaining FLE while mechanisms M1 and M3 cannot.
Yet, in the conditions of oppositely moving stimuli the effect was stronger compared to when only one unambiguously moving stimulus was used, which cannot be explained by this theory; thus a theory based on the mechanism M2 seems also insufficient. Either the followers of the extrapolation/prediction theory should provide a detailed description of some mechanism other than the ones listed here as M1, M2 and M3 and the predictions of which would be consistent with our results or it should be admitted that extrapolation/prediction is not an exhaustive, sufficient theory of the FLE.
It can be argued that although the automatic low-level extrapolation theory or higher level uncertainty-sensitive internal model theory (M1 or M3) does not work in our present main experimental conditions, some other higher-level cognitive mechanism having a top-down influence might have its effect in combination with some as yet unspecified influences that render FLE weaker in the control condition. For example, we see that the slope of the FLE function is steeper in the control condition (Fig. 2). The latter condition has less uncertainty to be solved by the perceiver compared to the main condition with higher uncertainty. This may simply reflect differences in task difficulty without any relation to the motion analyzing mechanisms. However, in this case it would be difficult to understand why the generally more flat shape of the FLE function obtained in the condition with oppositely moving stimuli compared with the condition with one moving reference stimulus is present only with negative SOA values and not present with positive SOA values. Task difficulty should have an additive effect on the FLE performance in all SOA conditions.
In the supplementary experiment to Experiment 1 where pre-flash color knowledge was controlled, the FLE was considerably stronger than in the main experiment. At first this difference may appear as resulting from the fusion of two white moving peri-flash stimuli that become an “extended moving target” and the spatial location of the leading edge of the moving stimulus should undergo a kind of sudden “jump” in the motion direction, thus artifactually leading to the FLE. (In the main experiment of Experiment 1 color distinctly differentiated the moving objects and there was no fusion.) Yet, this explanation is doubtful because this kind of sudden advance spatial extension takes place in both directions of motion and thus it is indeterminate, mutually opposing and direction-wise ambiguous. More importantly, in Experiment 2, opposite motion/adjacent condition we had a control for this putative artifact by laterally shifting the two oppositely moving pairs of bars in space so that fusion was not possible (see Fig. 3b). Because a robust FLE was nevertheless obtained, the fusion explanation cannot be used.
Noticeably, there is a conspicuously large absolute value of the FLE errors in the supplementary experiments compared to the main one. What does FLE literature say about finding large FLE effects? For example, extrapolatory errors in localizing a moving stimulus may be larger when measured by pointing movements compared to perceptual comparisons (Kerzel & Gegenfurtner, 2003). But as the dependent measure in our study was strictly perceptual, we could not use this explanation.
At some stage of the FLE-debate the fact that in the flash-initiated conditions the effect is well present has been used against the extrapolation theory (for review see Nijhawan & Khurana, 2010). Subsequent to this criticism Nijhawan (2008) presented data and arguments to show that extrapolation can take effect within a very short time and, combined with the inevitable delay it takes to the final explicit perception, FLE can be explained by prediction theory also in the flash-initiated conditions. Yet this position was weakened again by Gauch and Kerzel (2008) who compared FLE when moving probes were used and when static flashed probes were used, with probes moving opposite to the direction of target motion along a motion path below the motion path of the target. (In our Experiment 2, opposite motion/adjacent condition a similar method was also successfully used.) While the latest versions of extrapolation theory predicted no FLE during motion offset (Maus & Nijhawan, 2009), an FLE was found at motion offset, too (Gauch & Kerzel, 2008). (Notably, Gauch and Kerzel (2008) found that FLE was augmented when the probe moved in the opposite direction from the target stimulus-in-motion, compared to the static flash condition. This finding resembles our present results because opposite motion vectors were employed. However, any finalized conclusions with regard to the present study cannot be made from this because our flashed stimulus was always static.)
The fact that with positive SOAs (where flash comes only after the moving stimulus has reached or moving stimuli have reached the fixation point) there are virtually always flash-lag responses in all spatiotemporal experimental conditions seems to support the postdiction theory (Eagleman, 2008; Eagleman & Sejnowski, 2000, 2007). It states that a flash as a novel sudden stimulus or a kind of query for what happened in the visual field resets the integrative computations the brain uses for building up a veridical representation of the external visual events. But the brain continues to collect information after the flash and because it takes time to collect different features of the environment (including motion signals) the perceptual model becomes available only with a delay. Consequently, the percept attributed postdictively to the time of the flash is a function of events that happen following the flash, including the indication of the advanced position of the moving stimulus which is different from the previous static invariant position of the flash. However, the postdiction theory cannot account for our results because the oppositely moving stimuli condition leads to the considerably stronger FLE with negative SOA values (i.e., when the flash precedes the moment of stimuli in motion arriving at fixation point) compared to the condition with one moving stimulus. This means that—as the post-flash events are identical between the conditions—just the events preceding the moment when stimuli-in-motion arrive at the fixation point produce this difference. Somehow the stimulus moving oppositely to the reference before the flash determines how the post-flash events are perceived in awareness. Postdiction theory would have also predicted differences in the FLE between the conditions with continuous post-flash motion of both of the moving stimuli and with discontinued motion of one of the moving stimuli (e.g., Experiment 2). Actually no such differences were detected. In order to explain our results, the post-diction theory should be complemented with some mechanism dependent on uncertainty created by the additional moving object.
The attentional theory (Baldo & Klein, 2008) predicts dissociation of the spatial coordinates of flash and the moving reference-stimulus at the time when the flash-initiated attentional focusing is completed with some delay (during this delay the moving stimulus has changed its position, but the flash position in memory is invariant). With two motion directions against only one, the load on the attention-focusing system could be higher and consequently attentional focusing in the process of binding the flash and the reference motion together in a unitary percept takes longer. This leads to the percept where FLE value is higher. This account also explains the difference between the negative and positive SOA conditions (asymmetry of psychometric curves) in the two-object condition. Note that the additional uncertainty disappears at zero SOA. It may be still difficult to explain some results of the supplementary experiment. Namely, why were the flash-lag responses equal when SOA = 0 ms between the conditions with oppositely moving stimuli on the one hand and the conditions in the no opposite motion condition (with white bars as the stimuli) on the other hand (both roughly at 90%, within same confidence limits)? Uncertainty should be unequal here, but FLE frequency is equal.
Some researchers explain FLE as a result of shorter visual latency to stimulus-in-motion compared to the static stimulus (Patel et al., 2000; Whitney et al. 2000). It is not easy to use this theory to explain the FLE difference between the oppositely moving and single moving stimuli conditions. Motion parameters of the reference stimulus are invariant between our two experimental conditions and the differential latency theory predicts equal magnitude of the FLE for these conditions. It is likely that motion-detecting channels are arranged and tuned so that the ones responsible for opposite motion inhibit each other (e.g., motion after-effect) and thus FLE should have had smaller magnitude in our two-stimulus opposite motion conditions if differential reaction time would be the case. On the other hand, it can be argued that in the oppositely moving stimuli condition the spatial vector of motion is not important and some general rectified response to the absolute values of motion-detector activity is what matters for determining the FLE value. Consequently, with two moving stimuli the abstracted signals about absolute motion may be stronger and thus send a message about a higher speed attributed to each of the separate moving stimuli. Thus the percept is formed faster and FLE increases in the main condition. But this somewhat construed ad hoc explanation is also suspect: while it has been found that simple visual latency to the change in the single stimuli in motion increases with speed of motion (Kreegipuu, Murd, & Allik, 2006), when more complex tasks of comparing the changes in different stimuli is used—as is the case with our FLE displays—the effect of motion speed on perceptual latencies disappears (Murd, Kreegipuu, & Allik, 2009). Moreover, there are strong arguments against the possibility that differential latency could be the only or main cause of FLE (Arnold, Durant, & Johnston, 2003; Arnold et al., 2009; Bachmann & Põder, 2001). For example, apparent timing of moving and static color changes prove to be both veridical and FLE can be robustly obtained also when the reference stimulus appears in a stream of static objects, with no motion involved at all.
How do the positional sampling and averaging accounts (Brenner & Smeets, 2000; Krekelberg & Lappe, 1999) comply with our findings? The continuously changing net result of the time-consuming computation about the precise location of a moving stimulus should indicate a position shifted forward along the motion trajectory. Because with two oppositely moving stimuli one of the pre-flash instances of the moving stimulus acts as a pre-cue for the later-remaining moving stimulus, it may add to the signals that feed averaging processes and bias the moving stimulus to appear more as being shifted towards its own motion direction. However, this theory needs to postulate that the averaging operator is not sensitive to the spatiotemporal direction of the change of the moving stimulus. This account also needs to explain why in our Experiment 2 with laterally distanced motion paths (condition opposite motion/adjacent) the effect is conspicuously present and has even a larger magnitude.
Perceptual retouch theory (Bachmann et al., 2003; Bachmann, 2010) explains that for a stimulus to become consciously perceived, two interacting processes are necessary. First, the pre-consciously operating fast process responsible for formation of the object representation by feature integration, executed by the cortical specific modules tuned to specific contents of stimuli. This system is not sufficient for explicit perception because perceptual contents remain pre-conscious unless this system activity becomes modulated by a slower non-specific thalamic afferents. For the continuously moving stimulus there is a “headstart” for the non-specific process because this stimulus has been in view already earlier. Consequently, the moving stimulus becomes represented in awareness faster than the newly flashed one. In case of the oppositely moving stimuli, the already perceptually speeded-up moving stimulus wins not only due to having been presented earlier (as is the case in the standard FLE conditions), but also because the oppositely moving stimulus was preparing the modulating operation also from the spatial position where the stimulus that remains in view arrives later. In this theory, the direction of motion is not important. What matters is the presence of the signals from the advanced position ahead in time, regardless of the motion vector. As soon as the signals from the moving reference stimulus arrive at the “prepared” position it becomes represented in awareness at once. We emphasize that for the retouch theory the direction of motion of the priming stimulus is irrelevant because this attribute is not critical for evoking the non-specific modulation. This conjecture of the retouch theory is supported also by the results from Maiche, Budelli and Gómez-Sena (2007) who showed that a primer moving orthogonally towards the future position of the moving target ring augmented the FLE between target and the reference flash. Similarly, the results from our Experiment 2 with spatially shifted trajectories of the moving stimuli showing a robust FLE are consistent with retouch theory because the nonspecific modulation effect is spatially coarse.
The updating theory accepts that perceived timing of the appearance of a new object is delayed compared to the updating of perception of the changing properties of an already present object (Enns et al., 2010; Kanai et al., 2009). Construction of a new object representation requires additional time to establish a stable neuronal representation. As in our oppositely moving objects condition all the features of the moving objects, except their position, are the same, updating only the position takes less time than formation of all the features of the flashed object. In this theory, again, direction of motion or motion at all is not important because attributes of an individuated object (e.g., the distractor stimulus which is similar to the reference object) are largely invariant to motion. Nevertheless, the updating theory has its own caveats and constraints. First, its mechanism should be immune to the uncertainty about which object is relevant among the two moving pre-flash objects and two parallel updating operations should be executed as effectively as only one updating operation. Second, the updating process should be equally proficient when oppositely moving objects occupy the same path of motion and when the paths are adjacent. Third, updating of one moving reference object should not be influenced by how long another alternative moving object is presented; updating operations should be largely independent.
To conclude, we showed that neither the extrapolation/prediction theory (in its three versions presented here) nor the post-diction theory can account for our results. On the other hand, our results can be explained by the few theories that show why the conscious-level perception of the continuously changing stimulus is temporally facilitated in awareness compared to the perception of a newly appearing stimulus. Follow-up research should examine whether these theories are mutually compatible or whether some of them have higher explanatory potential than the others. Also, we should not ignore the possibility that the differences found between the two conditions in this study are caused by mechanisms separate from the main mechanism of FLE. Another important task would be to test whether the regularities found here apply only to the allocentric frame of reference of the FLE or also to egocentric frame (e.g., Becker et al., 2009). Perhaps the main contribution of our paper is the introduction of a new experimental method which led to some unexpected results thus motivating further research on FLE. We can therefore expect more detailed descriptions of the theories capable of explaining the effect.
We thank the Estonian Scientific Competency Council for their support via targeted financing research theme SF0182717s06, “Mechanisms of Visual Attention”.
- Bachmann, T. (2010). Priming and retouch in flash-lag and other phenomena of the streaming perceptual input. In R. Nijhawan & B. Khurana, B. (Eds.), Space and time in perception and action (pp. 536–557). Cambridge: Cambridge University Press.Google Scholar
- Bachmann, T., Breitmeyer, B. G., & Öğmen, H. (2007). The experimental phenomena of consciousness. A brief dictionary. New York: Oxford University Press.Google Scholar
- Enns, J.T., Lleras, A., & Moore, C.M. (2010). Object updating: a force for perceptual continuity and scene stability in human vision. In R. Nijhawan, B. Khurana, (Eds.) Space and time in perception and action (pp. 503–520). Cambridge: Cambridge University Press.Google Scholar
- Mather, G., Verstraten, F., & Anstis, S. (1998). The motion after-effect: a modern perspective. Cambridge: MIT Press.Google Scholar
- Nijhawan, R., & Khurana, B. (2000). Conscious registration of continuous and discrete visual events. In T. Metzinger (Ed.), Neural correlates of consciousness: empirical and conceptual problems (pp. 203–219). Cambridge: MIT Press.Google Scholar
- Nijhawan, R., & Khurana, B. (Eds.). (2010). Space and time in perception and action. Cambridge: Cambridge University Press.Google Scholar