TRANSSACCADIC MEMORY FOR THE POSITION OF STATIONARY AND TRANSLATING BIOLOGICAL- MOTION WALKERS

Previous research demonstrated an advantage for translating objects over stationary objects in transsaccadic displacement detection. However, in some studies, this benefit was absent. The current study was designed in order to clarify the basis of these contradictory findings. To this end, the procedure of an experiment with a clear motion benefit was combined with the stimuli of a study in which the motion benefit was absent. Participants saccaded towards either a stationary or a translating point-light walker and had to detect the intrasaccadic displacement of either the saccade target or the saccade flanker. Intrasaccadic displacements of the translating walker were found to be easier to detect than displacements of the stationary walker. Furthermore, displacements of the saccade target walker were better detected than displacements of the flanking walker. Implications for the previously contradictory observations are discussed and an explanation is proposed emphasising the differential importance of spatiotopic coding when a viewer is engaged in smooth object pursuit rather than having a stable fixation before making a saccade towards a translating object.


Introduction
High-acuity vision is restricted to the small central foveal and parafoveal part of the visual field (e.g., Anstis, 1974;Loschky, McConkie, Yang, & Miller, 2005).Therefore, during scene exploration, the eyes constantly alternate between short periods of relatively stable eye position (fixations) and very fast oculomotor jumps (saccades), in order to project new objects of the scene onto the high acuity foveal region of the retina (e.g., De Graef, Christiaens, & d'Ydewalle, 1990;Einhäuser, Spain, & Perona, 2008;Henderson & Hollingworth, 1998;Rayner, 1998).However, eye movements also create new problems for the visual system.Indeed, saccades shift the image on the retina, so that information enters the visual system in a temporally distributed and spatially fragmented way.Nevertheless, in everyday life, human observers perceive the world as stable and unified.Vision scientists are therefore confronted with the question whether and how transsaccadic integration is achieved: is image information gathered on presaccadic fixation n-1 carried across the saccade in transsaccadic memory and then integrated with information acquired on postsaccadic fixation n in order to achieve a stable percept (e.g., Demeyer, De Graef, Wagemans, & Verfaillie, 2009;Germeys, De Graef, Van Eccelpoel, & Verfaillie, 2010;Irwin, 1991;Martinez-Conde, Krauzlis, Miller, Morrone, Williams, & Kowler, 2008;Melcher & Colby, 2008;Van Eccelpoel, Germeys, De Graef, & Verfaillie, 2008;Verfaillie & De Graef, 2001)?
One experimental paradigm that has proven to be a useful tool for investigating the content and the spatial extent of transsaccadic memory and transsaccadic integration is the transsaccadic change detection paradigm (e.g., Henderson & Hollingworth, 1999;Rayner, McConkie, & Zola, 1980;Verfaillie, De Troy, & Van Rensbergen, 1994).In this paradigm, an aspect of the stimulus is changed during the saccade (intrasaccadically) on a proportion of the trials.The subjects' task is to judge whether or not a change occurred.Because the information entering the visual system during the saccade is not useful due to retinal 'smearing', intrasaccadic changes normally can only be noticed by comparing the postsaccadic image with information about the presaccadic image, stored in transsaccadic memory.The ability to notice a certain type of change is therefore interpreted as evidence that the type of information that was changed is actually stored in transsaccadic memory.Vice versa, if changes in an object's features go unnoticed, then these features are assumed either not to be stored in transsaccadic memory or not to be integrated with postsaccadic information.
However, Deubel, Schneider, and Bridgeman (1996) discovered that externally interrupting the visual input after the saccade (by inserting a postsaccadic blank before the reappearance of the visual stimulus) results in much more accurate detection of intrasaccadic position changes (see Deubel, Bridgeman, & Schneider, 1998;Deubel, Koch, & Bridgeman, 2010;Deubel, Schneider, & Bridgeman, 2002, for related research on the blanking effect).These findings suggest that position information is in fact stored in transsaccadic memory, but, under normal circumstances, the information is unavailable to conscious perception.One possible reason, originally advanced by Deubel and colleagues (also see Demeyer, De Graef, Wagemans, & Verfaillie, 2010), is that the visual system assumes that the visual world remains stable across saccades.This assumption is plausible in real-life vision, where the brief interruption in the input caused by a saccade is unlikely to coincide with unpredictable and sudden object displacements in the outside world.As a result of this assumption of visual stability, only large discrepancies between presaccadic and postsaccadic object locations are consciously noticed by observers.When the object is absent after the saccade (due to postsaccadic blanking), the assumption of stability is falsified and conscious processes regain access to information that initially was stored only implicitly, resulting in more accurate detection of intrasaccadic displacements.
In most experiments providing evidence that the visual system assumes a stable position of the saccade target during a saccade, the position change occurred in stationary objects.It can be expected, however, that, in the case of a saccade to a moving object, the visual system no longer holds the assumption that the position of the moving object will remain unchanged during the saccade.The position of moving objects might therefore have a special status during transsaccadic integration.This was demonstrated by Gysen, De Graef, and Verfaillie (2002;also see Gysen, Verfaillie, & De Graef, 2002a).In a study in which subjects had to make a saccade to one of two objects, one of which translated towards the other which remained stationary, it was shown that transsaccadic displacement detection was more accurate for the translating object than for the stationary object, suggesting that translating objects indeed have a special transsaccadic status (see Pollatsek & Rayner, 2002, for similar findings).
Moreover, using the same paradigm, Gysen, De Graef, and Verfaillie (2002b) replicated Deubel et al's (1996) finding that postsaccadic blanking improves displacement detection for stationary objects, but also found that blanking eliminates the transsaccadic benefit for moving objects.In fact, when the moving object was blanked after the saccade (even for as short as 60 ms), the detection of intrasaccadic displacements of the moving object was much worse than the detection of comparable displacements of station-ary objects.This suggests that, when the visual system plans a saccade to a moving object, it does not make the default assumption of visual stability.Instead, fast and precise pre-and postsaccadic spatial processing seems to be necessary.When the object is blanked postsaccadically, this information is no longer available, making transsaccadic change detection more difficult.In sum, the detection of intrasaccadic displacements of stationary objects is difficult, presumably because the visual system assumes stability across saccades, whereas displacements of translating objects are much more detectable, suggesting that, under these circumstances, the visual stability assumption is no longer retained.
Note that overall motion per se is not a sufficient condition for a transsaccadic benefit for the moving object.Gysen et al. (2002) did not find enhanced displacement detection for objects rotating in depth compared to stationary objects.Dahlstrom-Hakki and Pollatsek (2006) observed no benefit in the detection of intrasaccadic depth orientation jumps in a foveally presented rotating vs. stationary cube.Apparently, the assumption of visual stability is relaxed and even abandoned only in the case of motion with a component of translation.This seems plausible given that translation involves a change in the spatiotopic position of objects and the visual stability hypothesis mainly pertains to the spatiotopic position of objects.
There is, however, one study from our lab that seems to contradict this line of reasoning.In Experiment 2 of Verfaillie et al. (1994), participants viewed point-light walkers: stimuli in which the available information is confined to a number of point lights attached to the major joints of a human figure.At a designated time, participants had to make a saccade within the point-light walker and judge whether the walker was displaced during the saccade or not.In one condition, the walker did not translate, i.e., moved as if walking on a treadmill.In another condition, the walker translated across the screen.It was found that transsaccadic memory for the global position of the walker was very poor in both conditions (in fact, even slightly worse for a translating than for a non-translating walker).This is at odds with Gysen et al.'s (2002) finding of superior transsaccadic memory for translating objects.The main purpose of the present article is to better understand these contradictory observations and in this way help to further uncover the mechanisms underlying transsaccadic integration.
One difference between the studies showing a transsaccadic benefit for moving objects and Verfaillie et al.'s (1994) study is that in the former studies the stimuli consisted of inanimate objects, whereas in the latter an animate figure in motion was shown.It has been suggested that the perception of animate actions might be "special", in the sense that actions are processed in a qualitatively unique way, supported by a specialised architecture hard-wired in the brain (e.g., Grossman, Donnely, Price, Morgan, Pickens, Neighbor, & Blake, 2000;Lestou, Pollick, & Kourtzi, 2008;Peuskens, Vanrie, Verfaillie, & Orban, 2005;Reed, McGoldrick, Shackelford, & Fidopiastis, 2004;Vaina, Solomon, Chowdhury, Sinha, & Belliveau, 2001).However, it is hard to envision how the animacy of the moving object could moderate the transsaccadic benefit for moving objects over stationary objects (but see Orban de Xivry, Coppe, Lefèvre, & Missal, 2010, for a report on the influence of biological motion on smooth pursuit).
A more plausible explanation has to do with the dynamics of the eye movements involved.In the experiments demonstrating a transsaccadic benefit for moving objects, participants were fixating a fixation cross prior to making a saccade to the moving object.In Experiment 2 of Verfaillie et al. (1994), subjects were pursuing the translating walker both before and after making a saccade within the figure.One could argue that, in the former case -making a saccade to a moving object after a stable fixation -, fast and precise processing of the spatiotopic position of the object is necessary, to allow accurate saccade targeting and object tracking.Under these circumstances, the visual system might not stick to the assumption of visual stability.Gysen et al. (2002b) discuss several reasons why this might be the case.Moreover, as already mentioned before and in support of this hypothesis, Gysen et al. (2002b) showed that briefly blanking the moving object postsaccadically eliminated the transsaccadic benefit for moving objects.In fact, after postsaccadic blanking of the moving object, displacement detection was much worse than without blanking (whereas the opposite effect was observed for stationary objects).In contrast, when the observer is pursuing the moving object, both before and after the saccade, the precise spatiotopic location of the object is less relevant.Indeed, one of the primary aims of the visual system during ocular pursuit is to keep the eyes lagging behind the moving object to a minimum (e.g., Thier & Ilg, 2005; but see Orban de Xivry & Lefèvre, 2007, for a discussion of the commonalities between the saccade and the pursuit system).This is primarily based on the measurement of retinal slip, i.e., judgment of the retinal projection of the moving object in relation to the fovea, rather than on the spatiotopic position of the object.Less accurate coding of the objects' spatiotopic position during pursuit could therefore result in poorer detection of intrasaccadic displacements of the pursued object.
If the latter hypothesis holds (rather than something being special about animate vs. inanimate object motions), then having participants saccade from a stable fixation position either to a translating biological-motion walker or to a non-translatory walker moving on a treadmill, should result in a benefit for the translating walker.This is precisely what we investigated in the present study.We combined the paradigm used in Gysen et al. (2002) with the stimulus employed in Verfaillie et al. (1994). [2]

Subjects
Six University of Leuven students participated in the experiment.All had normal or corrected-to-normal vision.

Apparatus
Stimuli were displayed in a 60-Hz noninterlaced mode on a Sony GDM-W900 Trinitron Colour Graphic Display with a 756 × 468 resolution.Eye movements were monitored with a Generation 5.5 dual-Purkinje-image eye tracker (Fourward Technologies, San Diego, CA) with an accuracy of 1 min of arc and a 1000-Hz sampling rate.It was interfaced with a PC, storing every sample of the left eye's position.For each sample, the computer made an online decision about the eye state: fixation, saccade, blink, or signal loss.This on-line classification algorithm enables detection of a saccade within 4 ms of the onset of a saccade.Eye state and position were fed into a second PC in control of stimulus presentation.

Procedure
Participants were seated at 80 cm from the stimulus display, with their head stabilised by a headrest and a bite bar with dental impression compound.Once the eye tracker was successfully calibrated for nine points along the 2. Note that, in a way, our experimental design was incomplete, because it did not include a condition in which participants have to pursue an extrafoveal object before making a saccade.Indeed, in the present experiment subjects (after steady fixation) made a saccade to an extrafoveal object, whereas in Verfaillie et al's (1994) study subjects (during pursuit) made a saccade to the foveal object.In order to test our hypothesis, an experiment in which the manipulation of the state of the subject's eyes prior to making a saccade (fixation vs. smooth pursuit) and the manipulation of the dynamic status of the extrafoveal saccade target object (translating vs. stationary) are fully crossed in a factorial design, would be ideal.However, the main problem is that a condition in which observers pursue an extrafoveal object is extremely hard, if not impossible, to realise.It might be possible to implement an approximation of such a condition (e.g., asking participants to pursue a point moving above a walker translating in the periphery), but the resulting situation would be very artificial and probably will not advance our understanding of the dynamics of visual perception under natural circumstances.This is not surprising.After all, the main function of smooth pursuit is precisely to keep a moving (attended) object in the fovea!diagonals of the stimulus field, a block of 48 practice trials was initiated followed by an experimental session.
As illustrated in Figure 1 (p.184), each trial consisted of the following events.(Note that, for reasons of clarity, the point-light walkers in Figure 1 are depicted relatively larger than during the actual experiment, thus overrepresenting their size in relation to the distance between the walkers and the fixation cross.)First a fixation cross (0.5° × 0.5°) appeared.Participants were instructed to fixate the cross and calibration accuracy was checked.The trial proceeded once eye position had continuously been within a box of 1° × 1°s urrounding the cross for 1000 ms.Two dots (0.15° in diameter) were then presented at the locations where the point-light walkers would appear.The dots stayed on for 250 ms, after which they were replaced by two point-light walkers.On every trial, both a stationary (i.e., walking on a treadmill) and a translating walker (which always moved towards the stationary walker) were shown.1023 ms after the appearance of the point-light walkers, the fixation cross was replaced by an arrow, directed towards one of the walkers.This was the signal for the viewer to initiate an eye movement towards the indicated walker.If the gaze position moved outside the virtual fixation box before the appearance of the arrow, the trial was interrupted and a new trial began.
During the saccade towards the target point-light walker, the arrow was erased and a position change in one of the walkers could take place.In one third of the trials, the position of the stationary walker was shifted, in another third of the trials, the position of the translating walker was shifted, and in the remaining third of the trials, no position shift took place.In half of the trials with a position shift, the saccade target walker shifted position; in the remaining half, the other walker (i.e., the saccade flanker) shifted position.Displacement size was 1.2° for stationary and 0.7° for translating walkers.Pilot work showed that displacements of 0.7° for stationary walkers were almost impossible to detect.In order to avoid floor effects and similarly to Gysen et al. (2002), we therefore decided to increase the displacement size for stationary objects to 1.2°.The position change was 15.2% and 9.1% of the distance (i.e., 8°) between fixation cross and the stationary and translating walkers respectively.The displacement did not disrupt the natural walking cycle of the point-light figures.
Following the critical saccade, viewers had to indicate whether one of the point-light walkers had been displaced and if so, which walker was displaced.Viewers responded with a single left or right button press when, respectively, the walker presented on the left or right side of the screen was perceived as having been displaced.When no displacement was perceived, a combined left and right response was required.Following the response, a new trial began.
Each participant completed one practice block and 32 experimental blocks.Each block contained 48 trials which were produced by the factorial combination of displaced walker (stationary walker, static walker, and no displaced walker), position of the stationary and translating walker (i.e., stationary left / translating right vs. translating left / stationary right), saccade direction (left vs. right), orientation of the static walker (i.e., faced towards vs. away from the translating walker), and displacement (towards vs. away from other walker).The order of the trials was randomized for each subject and block separately.

Stimulus displays
The point-light figures consisted of 13 dots (0.15° in diameter) attached to the ankles, knees, hips, shoulders, elbows, wrists, and the head.The point-light figure performed a walking motion that was designed using motion capture data from a real walker and a 3D animation technique (Dekeyser, Verfaillie, & Vanrie, 2002;Vanrie & Verfaillie, 2004).Point-light walkers were 2.7° in height and were dark gray on a light gray background.On each trial, both a stationary and a translating point-light walker were present.While the station- The horizontal starting position of the translating walker differed from that of the stationary walker by 3.20°.This was done in order to have a similar horizontal position for both walkers (i.e., 5.5° left or right of centre) at the time of the saccade go-signal (i.e., this signal to initiate a saccade was presented 1023 ms after the appearance of the walkers, which corresponds to a moving distance of 3.20° for the translating walker), ensuring that saccadic amplitude and duration in the conditions with translating and stationary walkers were comparable.

Results
Unsuccessful trials were excluded from the analyses.A trial was considered to be unsuccessful when the subject lost his or her fixation before the saccade go-signal, when saccade latency (measured from arrow onset) was smaller than 80 ms or above 1430 ms (because, at 1430 ms, the distance between the translating point-light walker and the middle of the screen was only 1°), when the saccade was directed towards the flanker instead of the target, or when the intended displacement of the saccade target did not complete within the time period of the saccade because of computer processing delays or gaze position signal loss.This procedure resulted in an elimination of 28.7% of all trials.
Saccade latencies measured from central arrow-cue onset were 420 ms on average. [3]A repeated measures analysis of variance with saccade direction (left vs. right) and dynamic status of the saccade target (stationary vs. translating) as within-subject variables only yielded a marginally significant main effect of the latter variable, F(1, 5) = 6.01, p = .06,MSE = 0.020.On average, saccades towards translating walkers were initiated 8 ms faster compared to saccades towards stationary walkers.This means that, given the temporal and spatial parameters of the critical events making up a trial (especially the differential horizontal starting positions depending on the dynamic status of the walkers), translating and stationary walkers were approximately at the same retinal eccentricities at the time the eyes were launched.
First, hits (i.e., correct identifications of the object that shifted position) were combined with false alarms (i.e., false reports of a displacement of that 3. The mean saccadic latency was relatively long.Note, however, that saccadic latency was measured from the moment the fixation cross changed into an arrow and the two potential saccade targets were already present at that time.Therefore, saccade latencies included the time needed to process the directional cue and decide where to move the eyes next and the time needed to plan and launch the saccade (including attention disengagement from the arrow, the presaccadic movement of attention to the selected saccade target, and the start of the actual saccadic movement).
particular object when no object was displaced) to derive d' values.To obtain d' we followed the constant ratio rule (MacMillan & Creelman, 1991, pp. 243-245;see Gysen et al., 2002b The analysis revealed significant main effects of both dynamic status, F(1, 5) = 8.03, p < 0.04, and saccadic status, F(1, 5) = 11.41,p < 0.03.In addition, the interaction between the two variables was significant, F(1, 5) = 15.24,p < 0.02 As shown in Figure 2, for translating walkers, sensitivity was much higher when the translating walker was the saccade target than when it was the flanker (d' of 1.80 and 0.94, i.e., a significant saccade target advantage of 0.86, t(5) = 4.89, p < 0.02).For stationary walkers, however, this saccade target advantage was much smaller (d' of 0.93 and 0.76, i.e., a non-significant saccade target advantage of 0.17, t(5) = 0.93, p > 0.75).Comparison of translating and stationary walkers shows that, for saccade targets, sensitivity was much higher for translating walkers, t(5) = 4.32, p < 0.03.The same comparison for flankers, shows no difference, t(5) = 0.7, p > 0.85.However, displacements of stationary walkers were almost 60% larger than displacements of translating walkers, indicating that sensitivity for displacement was still better for translating walkers.
Second, we analysed the proportion of misattributions (trials in which participants reported that they saw a displacement, but incorrectly attributed the displacement to the walker that was not displaced; see Figure 3).There was no effect of the saccadic status (F(1, 5) = 1.15, p > 0.3).However, while detection of stationary walker displacements was almost at chance level (52% correct), subjects almost never attributed changes of a translating walker to the stationary figure (92% correct), F(1, 5) = 12.4,p < 0.02.This preference for attributing displacements to translating compared to stationary point-light walkers was also reflected in the observation that, in trials without displacement but in which subjects reported a displacement (false alarms), the reported displacement was attributed to the translating walker (68%) more often than to the stationary (32%) point-light walker, t(10) = 2.96, p < 0.02.

Discussion
Contrary to Verfaillie et al. (1994) and in line with Gysen et al. (2002) and Gysen et al. (2002aGysen et al. ( , 2002b)), intrasaccadic displacements of the translating walker were easier to detect than displacements of the stationary walker.This supports our hypothesis that the absence of the motion benefit in Experiment 2 of Verfaillie et al. (1994) is due to the fact that the critical saccade in that study occurs during pursuit of the translating figure. [4]During smooth pursuit, the visual system tries to keep the moving object of interest in the fovea as accurately as possible.To achieve this, the retinotopic position of the object is most important, at the expense of memory for the spatiotopic position of the objects.It has been shown before that the remembered location of translating objects is shifted in the direction of the motion (e.g., Hubbard, 1995) but that this mislocalisation decreases when the eyes remain stationary, instead of tracking the object's path of motion (Kerzel, 2000;Kerzel, Jordan, & Müsseler, 2001).When a displacement was detected, there was a large bias to attribute it to the translating object.This could follow from the visual system's proven assumption that stationary objects available immediately after the saccade remained spatially stable during the saccade and can be used as a reference object to recalibrate the position of the whole scene.Indeed, Deubel et al. (1998) showed that when observers saccade to one of two objects and one of the objects is blanked postsaccadically (either the target or flanker object) while one object is displaced intrasaccadically (again either the target or the flanker), the blanked object is generally perceived as being displaced and the nonblanked object is seen as spatially stable (regardless of which object was actually displaced).The present study also underlines the importance of the saccadic status for displacement detection.Displacements of the saccade target walker were better detected than displacements of the flanking walker.This is in line with many studies showing that the spatial extent of transsaccadic memory is primarily focused on the saccade target region (Currie, McConkie, Carlson-Radvansky, & Irwin., 2000;Deubel et al., 1998;McConkie & Currie, 1996).However, since performance was well above chance level even for flanker objects, despite the small displacement sizes, the data also confirmed the findings of Verfaillie & De Graef (2000;see also De Graef, Verfaillie, & Lamote, 2001;Germeys, De Graef, Panis, Van Eccelpoel, & Verfaillie, 2004) that, given appropriate control for extrafoveal preview quality, transsaccadic memory for position information is not strictly limited to the saccade target region.
In the work of Gysen et al. 2002, we controlled for the fact that the moving object might attract attention more than the stationary object.One could claim that such an attentional effect might be stronger for biological motion, given that a translating walker is more familiar and natural than a walker moving on a treadmill.Some observations are in favour of an explanation of the benefit for the translating walker in terms of attention.First, saccadic latencies were marginally longer for saccades directed towards the stationary walker.Second, when no intrassaccadic displacement occurred, false alarms were biased towards the translating walker.However, in other (unpublished) experiments with nonbiological objects, we repeatedly observed that saccadic latencies to translating objects were shorter than saccadic latencies to stationary objects.And, without intrasaccadic displacement, jump reports indeed were biased towards the translating walker, but this is again something we observed before with nonbiological objects.
In sum, we showed that the transsaccadic position displacement benefit for translating over stationary objects as reported by Gysen et al. (2002) and the absence of the motion benefit in the studies of Verfaillie et al. (1994) was most probably related to the fact that, in the latter study, observers were pursuing the figure before and after the saccade and spatiotopic position coding is less relevant during smooth pursuit.

Figure 1
Figure 1 Temporal course of a trial.Upon good fixation on a central fixation cross, two dots indicating the position of the point-light walkers appear.After 250 ms, the dots change into point-light walkers, one of them stationary (walking on a treadmill), the other one translating towards the stationary walker.After 1023 ms., the fixation cross changes into an arrow indicating which point-light walker is the saccade target.During the saccade, the fixation cross disappears and in two thirds of the trials, one of the point-light walkers is horizontally displaced.Subjects then indicate whether they noticed a displacement, and if so, in which of the two point-light walkers

Figure 2
Figure 2 Sensitivity (d') as a function of dynamic and saccadic status of the point-light walker in which the change occurred.The error bars represent standard errors

Figure 3
Figure 3 Proportion of correct displacement attributions in trials in which the displacement was detected as a function of dynamic and saccadic status of the point-light walker in which the change occurred.The error bars represent standard errors , Appendix A, for a detailed example).From the overall contingency table produced by the three stimulus types (shift translating, shift stationary, no shift) × 3 response types (translating shifted, stationary shifted, nothing shifted), we extracted two 2x2 tables, one for the translating walker (2 stimulus types, shift translating vs. no shift × 2 response types, translating shifted vs. nothing shifted) and one for the stationary walker (2 stimulus types, shift stationary vs. no shift × 2 response types, stationary shifted vs. nothing shifted).This was done separately for each subject both for the conditions in which the saccade target was displaced and the conditions in which the saccade flanker was displaced.Sensitivity estimates (d') were entered in a repeated measures analysis of variance (ANOVA) with dynamic status (stationary vs. translating) and saccadic status (saccade target vs. flanker) as within-subject variables.