The visual analog: Evidence for a preattentive representation across saccades

Abstract

Earlier research supports the idea that transsaccadic memory involves a relatively sparse and abstract representation with little detail, much like visual short-term memory (VSTM) within a fixation. We examined whether transsaccadic memory is restricted to VSTM representations or whether it also includes a maskable, short-lived, and more detailed representation, referred to as the visual analog. First, a within-fixation change detection experiment is reported, aimed at clarifying the distinction between VSTM and the visual analog, and also the relationship between the two components. We then report 4 transsaccadic change detection experiments that make use of a modified version of the blanking paradigm originally introduced by H. Deubel, W. X. Schneider, and B. Bridgeman (1996). The results support the idea that transsaccadic memory includes a VSTM representation for attended objects (i.e., the saccade target) and a visual analog component for both attended and non-attended objects.

Introduction

Because only a small fraction of the human retina is able to provide the high resolution needed for detailed visual processing, observers move their eyes about three to four times per second. With each eye movement or saccade, the retinal projection of the outside world changes position, bringing an initially peripheral region of interest into foveal vision (De Graef, Christiaens, & d'Ydewalle, 1990; Henderson & Hollingworth, 1999). During a saccade, the image moves very rapidly across the retina. Visual information extraction is therefore mainly limited to fixations or periods in between saccades when eye position is relatively stable. It can be assumed that the visual system somehow integrates the informational contents of individual fixations (e.g., Pertzov, Avidan, & Zohary, 2009; Van Eccelpoel, Germeys, De Graef, & Verfaillie, 2008). How much and what kind of information is retained and integrated across a saccade has been the topic of debate. 

An early hypothesis proposed that highly detailed point-by-point representations, produced during individual fixations, could be superimposed or fused to form an image-like representation containing information from multiple fixations (e.g., Breitmeyer, Kropfl, & Julesz, 1982; Jonides, Irwin, & Yantis, 1982; Wolf, Hauske, & Lupp, 1978, 1980). Despite the intuitive appeal of “spatiotopic fusion” (Irwin, 1992a) or the “integrative visual buffer” account of transsaccadic integration (McConkie & Rayner, 1976), substantial empirical evidence has shown it to be incorrect (e.g., Bridgeman & Mayer, 1983; Irwin, Brown, & Sun, 1988; Irwin, Yantis, Jonides, 1983; Irwin, Zacks, & Brown, 1990; O'Regan & Lévy-Schoen, 1983; Rayner & Pollatsek, 1983). According to an alternative account, visual detail is lost across a saccade; instead, visual information integration is carried out on a relatively abstract level with visual form being represented in terms of structural or relational aspects of the stimulus and its components (e.g., Carlson-Radvansky & Irwin, 1995; McConkie & Zola, 1979; Pollatsek, Rayner, & Collins, 1984; Rayner, McConkie, & Zola, 1980). Consistent with this view, maintenance of spatial position is relatively poor (e.g., Bridgeman, Hendry, & Stark, 1975; Bridgeman & Stark, 1979; Li & Matin, 1990a, 1990b; Mack, 1970; Pollatsek, Rayner, & Henderson, 1990; Stark, Kong, Schwartz, Hendry, & Bridgeman, 1976; Verfaillie, 1997; Verfaillie & De Graef, 2000; Verfaillie, De Troy, & Van Rensbergen, 1994; Wallach & Lewis, 1966), while relational information is integrated quite accurately (Carlson-Radvansky, 1999; Germeys, De Graef, Panis, Van Eccelpoel, & Verfaillie, 2004; Verfaillie, 1997). 

A similar conclusion was reached by Irwin (1991). Using a dot-matrix comparison task, Irwin showed that displacements of an entire dot pattern during an eye movement did not affect performance in detecting a change within the pattern. This suggests that visual information about the dot matrix is retained in a representation independent of absolute spatial position, while preserving information about the relative position of elements within the pattern. In the same study, Irwin also investigated the time course of information decay in transsaccadic memory. Varying the temporal interval that separated the two patterns from 1 ms to 5,000 ms had little effect on accuracy, suggesting that information in transsaccadic memory is long lasting. Further research has provided evidence that transsaccadic memory is capacity limited, that is, only a few items seem to survive the saccade (Irwin, 1991, 1992a, 1993, 1996; Irwin & Andrews, 1996). For example, using a partial report procedure, Irwin (1992a) showed that only three to four letters in a multi-letter array could be retained across an eye movement, regardless of the number of letters presented presaccadically. Furthermore, report of letters near the saccade target was much more accurate than report of other letters in the array. Following up on this observation, Irwin and Gordon (1998) showed that attention determines what information is encoded into transsaccadic memory. Because attention automatically precedes an eye movement to the saccade target location, information near the saccade target is more likely to be encoded (Deubel & Schneider, 1996; Hoffman & Subramaniam, 1995; Kowler, Anderson, Dosher, & Blaser, 1995; Shepherd, Findlay, & Hockey, 1986). In summary, a considerable body of evidence supports the idea that transsaccadic integration does not involve a global and visually detailed memory representation of the entire visual field, but rather a capacity-limited, attention-dependent, abstract representation, which is not strictly tied to spatial position and is relatively long lasting. What memory store could underlie the transsaccadic retention of this type of representation? 

To answer this question, it is useful to examine in what way transsaccadic memory differs from visual memory within fixations. There is general agreement for the existence of three main forms of visual memory (apart from short-term conceptual memory and long-term memory) within fixations: sensory or visible persistence, informational or schematic persistence, and visual short-term memory (Coltheart, 1980; Di Lollo & Dixon, 1988; Irwin, 1992b; Irwin & Yeomans, 1986). The first form, visible persistence, is a phenomenal trace of the extinguished stimulus and is believed to be a mere by-product of residual activity in the visual pathway (Coltheart, 1980). Visible persistence decreases with increasing stimulus durations (known as the inverse duration effect), being negligible for stimulus exposures exceeding 100 ms (Di Lollo & Dixon, 1988, 1992: Dixon & Di Lollo, 1994). In other words, its decay is triggered by the onset of the inducing stimulus (see Figure 1). The second form of visual memory, informational persistence, is a non-visible, maskable representation (also called the visual analog) that is precategorical and capable of coding form and location in a precise way. Its duration, estimated as ranging between 300 and 500 ms, is independent of stimulus exposure. In contrast to visible persistence, the visual analog representation starts to decay following stimulus offset (see Figure 1). Keysers, Xiao, Foldiak, and Perrett (2005) report a study on the possible neurophysiological mechanisms underlying the visual analog. The third type of memory, visual short-term memory, is a non-maskable, postcategorical memory system, which has a limited capacity and a long duration. It holds an abstract and schematic representation with little detail. There is a remarkable resemblance between those characteristics and the kind of representation that seems to be carried across saccades: a capacity-limited, attention-dependent, relatively long-lasting, abstract representation. This led Irwin (1991; also see Carlson-Radvansky & Irwin, 1995) to suggest that transsaccadic memory may in fact be the same as, or at least rely in large part on, VSTM. 

Figure 1

 

Schematic illustration of the time course of decay of visible persistence and the visual analog representation with (top) long versus (bottom) short stimulus presentations.

However, while spatiotopic visible persistence across saccades has been ruled out (e.g., Irwin, 1991; Irwin et al., 1983, 1990; Jonides, Irwin, & Yantis, 1983; Rayner & Pollatsek, 1983), there is still no definite answer as to whether informational persistence across saccades does or does not exist and whether it can play a role (in addition to VSTM) in transsaccadic integration. The duration of informational persistence, which is well above the average saccade duration (approximately 40–50 ms), would certainly allow it. A number of observations indeed indirectly suggest that, under some conditions, informational persistence may operate across saccades. 

First, it has been shown that detailed information regarding position and form in principle can be maintained transsaccadically (e.g., Hayhoe, Lachter, & Feldman, 1991; Lachter & Hayhoe, 1995; Melcher, 2005; Palmer & Ames, 1992). For example, Demeyer, De Graef, Wagemans, and Verfaillie (2009) showed that a presaccadic preview of an object facilitates its postsaccadic identification, even in a task in which participants have to discriminate between visually highly similar artificial shapes. Second, in a transsaccadic partial report task with letter arrays, Irwin (1992a) reported evidence for spatiotopic masking when a mask probe (instead of the typical bar probe) was presented for a short period of time following the saccade. Similarly, McRae, Butler, and Popiel (1987; also see Deubel, Schneider, & Bridgeman, 2002, Experiment 3) found evidence for spatiotopic masking in a full-report letter identification task. This suggests that certain information retained across a saccade is volatile (susceptible to masking), in contrast to information stored in VSTM. A third line of research supporting detailed transsaccadic representations concerns the transsaccadic blanking effect (Deubel, Bridgeman, & Schneider, 1998, 2004; Deubel, Koch, & Bridgeman, 2010; Deubel, Schneider, & Bridgeman, 1996; Deubel et al., 2002; Gysen, Verfaillie, & De Graef, 2002; also see Demeyer, De Graef, Wagemans, & Verfaillie, 2010). In the blanking paradigm, the saccade target is absent during the saccade and for a short period (50 to 300 ms) at the beginning of the postsaccadic fixation. Under these conditions, detection of intrasaccadic displacements improves drastically compared to the situation without blanking. In other words, in the direct absence of a localized stimulus following the end of a saccade, the visual system regains access to detailed position information. A similar result was obtained in an experiment with form changes of checkerboard-like stimuli (Deubel et al., 2002). Together, these findings imply that precise information about the presaccadic target object is retained in transsaccadic memory but that postsaccadic visual information affects its use. 

Consistent with these results, De Graef and Verfaillie (2002) provided evidence for improved discrimination of transsaccadic in-depth orientation changes of saccade target objects embedded within scenes when the postsaccadic image was blanked for a short period of time. De Graef and Verfaillie (2002) interpreted the blanking effect as evidence for a transsaccadic visual analog representation, that is, a high-capacity, non-selective, maskable representation of the form and location of objects, which decays within approximately 300 ms following stimulus offset. Blanking first of all postpones the presaccadic visual analog representation of a scene from being masked or overwritten by postsaccadic information. Second, attending a specific region of space following the saccade allows information within the visual analog to be selectively insulated against decay and masking by transferring it to a more durable store such as VSTM (Gegenfurtner & Sperling, 1993). 

To summarize, while a lot of research suggests that transsaccadic integration is supported by a relatively abstract and capacity-limited memory store (presumably VSTM), some findings show that more detailed location and form information can be remembered across a saccade than could be expected based on the characteristics of VSTM, possibly indicating a role for the visual analog representation. The main goal of the present study was to further assess whether transsaccadic memory is restricted to representations held in VSTM or whether it also contains a visual analog representation. In Experiment 1, we first examine information integration within fixations, trying to disentangle the relative contribution of VSTM and the visual analog representation to change detection performance (given the exposure duration of the memory display, no visible persistence could be expected). Based on a better understanding of these components, we then investigate information integration across eye movements (Experiments 2 to 5). 

Integration of information across temporal gaps (e.g., the temporary disappearance of a stimulus during fixation, or a saccadic eye movement) requires the encoding of a stimulus, maintaining a representation in one or more memory stores for some time, and subsequently combining the memory trace with new information (e.g., Brockmole & Irwin, 2005; Brockmole & Wang, 2003; Henderson & Hollingworth, 2003; Hollingworth, 2004). During each of these processes, information can get lost, causing incomplete integration (e.g., see Rensink, 2002; Simons & Rensink, 2005, for reviews of the extensive literature on blindness for changes, not only across saccades, but also across other temporal gaps). In the present experiments, integration is assessed by means of a change detection paradigm. All reported experiments involved the presentation of a circular array of letters or line drawings of everyday objects, for a short period of time. After a brief temporal interstimulus interval of variable duration, a second array was presented containing a single letter (or object) at one of the places on the circular array and participants had to decide as to whether the letter object in that location had changed across the temporal gap. Correct detection of a change or the absence of change implies (1) that a representation of the initial stimulus is maintained across the temporal gap and (2) that the participant is able to compare this representation with a subsequent stimulus. Note that the first requirement, namely the existence of a representation of the initial stimulus, is a necessary but not a sufficient condition for the detection of changes; the complete absence of a representation cannot be directly inferred from a failure to detect a change. 

Experiment 1

The purpose of Experiment 1 was threefold. In the first condition, we investigated the time course of information decay in VSTM by examining performance in the detection of changes in character displays as a function of interstimulus interval (ISI). Participants were shown a sequence of two displays separated by a blank ISI of variable duration (50, 100, 150, 200, 250, 300, 450, 700, or 1000 ms). The first display (memory display) consisted of eight letters arranged in a circular pattern and was presented for 250 ms. The second display (test display) contained only one letter and remained present until a response was made. In 50% of the trials, the single letter matched the corresponding letter in the memory display, in the remaining 50% the test letter differed from the corresponding letter in the memory display. The participant's task was to indicate whether the letter in the test display was the same as, or differed from the letter in the same position in the memory display. Previous studies (e.g., Irwin, 1991; Pashler, 1988; Phillips, 1974) using a similar procedure reported excellent change detection performance at very short ISIs. At longer ISIs however, performance dropped but remained stable with increasing ISIs. These results were interpreted as evidence for two kinds of visual memory within fixations: a rapidly decaying, high-capacity sensory memory that supports apparent motion (at the very short ISIs) and a slowly decaying, limited-capacity VSTM. In this first condition, we were mainly interested in the time course of information decay within VSTM. 

The second objective in Experiment 1 was to assess the amount of information that is still available after stimulus offset of the memory display. Becker, Pashler, and Anstis (2000) provided evidence that more information is represented in the visual system than change detection experiments (as our first condition) might reveal. They showed that cueing the location of a possible change during the blank ISI significantly improved change detection. As Becker et al. (2000) point out, this finding is not surprising in light of the iconic memory literature. Partial report experiments have consistently shown that an iconic memory trace can last for several hundreds of milliseconds after stimulus offset and that it can be read out by selective attention. We will assume that two main types of information are still available: information stored in VSTM (as assessed in the first condition) and information in iconic memory. Participants were shown an initial array of eight letters for 250 ms, directly followed by a 50-ms duration cue. The cue indicated the future location at which a change could take place. The time between the initial display and test display (ISI) was manipulated as in the first condition. 

Third, we examined the time course of decay of the visual analog: In the third condition, following offset of the initial display. ISI was held constant (1000 ms), while a cue, indicating the future location of a possible change, was presented at different moments following the offset of the memory display (0, 50, 100, 150, 200, 250, 300, or 400 ms). 

Methods

Participants

Three University of Leuven students participated in Experiment 1. They were paid 7.5 euros for each hour of participation. All participants had normal or corrected-to-normal vision. 

Stimuli

As shown in Figure 2, the stimulus display in each of the three conditions consisted of eight letters arranged in a circular pattern. Letters in the array were shown in uppercase Arial font and subtended 1° vertically. The letters were centered 4° from the center of the circular pattern and were equally spaced. The letters were displayed black (pixels off, 0.19 candela/m²) on white (pixels on, 13.90 candela/m²). For each trial, the letters in the initial (memory) display were selected randomly from a set of sixteen letters (B, C, D, F, G, H, K, L, M, N, P, R, S, T, V, and Z), subject to the constraint that no letter appeared more than once. The second (test) display contained only one letter (at one of the positions occupied by the letters in the memory display). In 50% of the trials, this test letter was identical to the corresponding letter in the memory display. In the remaining 50% of the trials, one letter in the memory array was substituted for a letter that was not present in the memory display. 

Figure 2

 

Schematic illustration of events and time course of a trial in the no-cue condition in Experiment 1.

Apparatus

The stimuli were displayed in 60-Hz mode on a Sony GDM-W900 Trinitron Color Graphic Display with a 756 × 486 resolution, which subtended 17° × 11°. Eye movements were recorded with a Generation 5.5 dual-Purkinje-image eye tracker (Crane & Steele, 1985). This system has 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 online classification algorithm enables detection of a saccade within 4 ms of the onset of the saccade. Eye state and position were fed into a second PC in control of stimulus presentation. Display changes were accomplished within 16.7 ms because they could be initiated at any moment during the screen refresh cycle (the display-change command did not wait for the vertical blank). 

Procedure

Participants were seated at 150 cm from the stimulus display, with their head stabilized by a headrest and a bite bar with dental impression compound. Once the eye tracker was successfully calibrated for nine points along the diagonals of the stimulus field, a block of practice trials was initiated followed by an experimental session. 

The events and time course of the three trial types are schematically illustrated in Figure 3. A trial in the first condition (the no-cue condition) consisted of the following events (also shown in Figure 2). First, a fixation cross appeared in the center of the display. Participants were instructed to fixate the cross, and calibration accuracy was checked. Once a 250-ms stable fixation was detected on the fixation cross (i.e., a stable fixation within an invisible square of 0.5° × 0.5° enveloping the fixation cross), it disappeared and the initial letter array was displayed for 250 ms. If, at any moment during the presentation of this memory display, participants made an eye movement that carried the eyes outside the invisible 0.5° × 0.5° square, the trial was interrupted, followed by the beginning of a new trial (less than 1% of the trials). Then, following a blank ISI of 50, 100, 150, 200, 250, 300, 450, 700, or 1000 ms, the test display appeared that contained one letter (either the same as the corresponding letter in the memory display or different from it). The test display remained on screen until the participant pressed one of two response buttons indicating the same or different response. Once the response was recorded, a feedback message was displayed in the center of the screen for 300 ms (correct or wrong) followed by the beginning of a new trial after an interval of 1,500 ms. 

Figure 3

 

Schematic illustration of events and time course in the no-cue, early-cue, and variable-cue-delay trials of Experiment 1.

The course of a trial in the second condition (the early-cue condition) was identical, except that, following the offset of the memory display, a cue was presented for 50 ms during the ISI. The cue was a red radial line segment, 3° in extent, pointing from the center of the array to a location previously occupied by a letter. Participants were informed that the probed letter in the test display would always be at the cued location. 

The third condition (the variable-cue-delay trials) differed from the previous two conditions in two respects. First, the ISI between the memory array and the test array was held constant at 1000 ms. Second, the timing of the 50-ms duration cue within the ISI was manipulated. The cue was presented for 0, 50, 100, 150, 200, 250, 300, 400, 650, or 950 ms following the offset of the memory display. 

Design

Each participant carried out one practice block of 90 trials and 36 experimental blocks of 108 trials. In total, each participant completed 3,888 trials that were a factorial combination of the three types of trials (no-cue, early-cue, and variable-cue-delay), 9 timings (ISIs of 50, 100, 150, 200, 250, 300, 450, 700, and 1000 ms for the first two types of trials, and cue delays of 0, 50, 100, 150, 200, 250, 400, 650, and 950 ms for the third type of trials), 8 possible letter change positions, 2 change conditions (same or different), and 9 repetitions. This resulted in 144 observations for each trial type × timing combination. Each block of 108 trials consisted of a factorial combination of 3 types of trials, 9 timings, and 2 change conditions with an equal distribution of the position of a possible letter change. The experiment was completed in 6 sessions that each lasted approximately 1.5 h. 

Results and discussion

Figure 4 shows the percentage correct change detection as a function of timing and trial type. Separate repeated-measures analyses of variance (ANOVAs) were carried out for each trial type with the timing variable as a within-subject variable. 

Figure 4

 

Percentage correct detection in Experiment 1 as a function of interstimulus interval (for the variable-ISI/no-cue and variable-ISI/cue conditions) or cue delay (for the fixed ISI/variable-cue-delay condition).

Although performance in the no-cue condition seemed to drop slightly with increasing ISI, the effect of ISI was not reliable, F(8, 16) = 1.44, p = 0.25, MSE = 0.0013. The present results are very similar to those observed by Pashler (1988) and others (e.g., Irwin, 1991; Luck & Vogel, 1997; Phillips, 1974). Indeed, performance did not deteriorate significantly across the entire range of ISIs reflecting the well-established property of VSTM: a relatively durable, though capacity-limited store (e.g., Irwin, 1991). (Note that, contrary to the previously mentioned studies, we did not observe a peak in performance at the very shortest ISI (50 ms). However, the absence of such peak in the present experiment is not completely at odds with previous work. The sensory storage effect, as referred to by Irwin (1991), has been shown to decay rapidly. For example, Pashler (1988) reported the presence of the effect at an ISI of 34 ms, while it was gone at 67 ms. In a similar study, Becker et al. (2000) reported no increase, even with ISIs as short as 16 ms. Hence, it seems that in Experiment 1, the rapidly decaying sensory storage is unusable at an ISI of 50 ms.) 

For the early-cue condition, the effect of ISI proved reliable, F(8, 16) = 43.10, p < 0.0001, MSE = 0.0008. Performance was lowest at an ISI of 50 ms but increased with longer ISIs. At an ISI of 300 ms, performance reached ceiling and remained stable thereafter. Our main objective in this early-cue condition was to assess the amount of information that is still available after stimulus offset of the initial display. The very accurate performance at longer ISIs (>200 ms) clearly indicates that a relatively good representation of all letters presented in the memory display was still present after stimulus offset. Comparison with the results obtained in the no-cue condition suggests that the increased performance in the early-cue condition, especially at longer ISIs, reflects the additional use of iconic information by a selective transfer process. There are two aspects of the present data that deserve further discussion. First, change detection performance did not reach ceiling right away but gradually increased with ISIs in the 50 to 300 ms range. The most obvious explanation for this effect is that it takes a certain amount of time to read out and transfer the cued item from iconic memory to a more durable store. This explanation assumes that readout and transfer from iconic memory is disrupted once the test display appears. Second, performance did not decline with increasing ISI once ceiling performance was obtained (ISIs of 300 to 1000 ms), indicating durable storage of the cued item, probably in VSTM. In sum, more information seems to be represented after stimulus offset of the memory display than is usually revealed by classical change detection experiments. Indeed, cueing the location of a possible change immediately following offset of the initial display enhances the ability to detect changes. This is consistent with Becker et al. (2000) and earlier reports (e.g., Averbach & Coriell, 1961; Di Lollo & Dixon, 1988; Gegenfurtner & Sperling, 1993; Sperling, 1960). We interpret these findings as evidence for a visual analog representation following stimulus offset. The increase in detection performance with increasing ISI duration is assumed to be caused by a selective readout and transfer process of a representation of the cued item in the visual analog to a durable memory store (VSTM). This process takes a certain amount of time and is interrupted whenever a new stimulus is presented. 

In the condition with a fixed ISI of 1000 ms but with a variable cue delay, a significant effect of cue delay on detection performance was obtained, F(8, 16) = 20.61, p < 0.0001, MSE = 0.0015. Performance was near perfect (96%) when the cue was presented immediately following the offset of the memory display and declined with increasing cue delays. It was only at the longest cue delays that change detection performance declined to the level of performance in the no-cue condition. In our opinion, these results confirm the existence of a visual analog representation, which starts to decay following stimulus offset. 

We should nevertheless add two comments to this interpretation of the data. First, within-fixation cueing effects in the blank interval between memory and test have been reported before, sometimes even when cues were presented well beyond the temporal range of iconic memory and even when observers were first cued to one side of the display and then to the other side (Griffin & Nobre, 2003; Landman, Spekreijse, & Lamme, 2003, 2004; Lepsien, Griffin, Devlin, & Nobre, 2005; Lepsien & Nobre, 2007; Makovski & Jiang, 2007; Makovski, Sussman, & Jiang, 2008; Matsukura, Luck, & Vecera, 2007). These effects have been interpreted as being based on VSTM, while for the present studies we think the visual analog provides a better explanation for the observed cueing effects. 

Second, while the present study allowed a better understanding of the time course of decay in the visual analog, we did not directly measure the maximum lifetime of information in the visual analog. Indeed, since we found that it takes time to read out information from the visual analog (early-cue condition), an estimate of the maximum lifetime of the visual analog should always allow for at least this readout time after a delayed cue is given. In the present experiments, the variable cue delay condition fixed the total duration between memory and test displays at 1000 ms, thus effectively curtailing the available readout time at the longer cue delays. In this sense, the current data do not provide an exact estimate of the maximum lifetime of the visual analog. 

Summarizing, we observed that change detection performance is partly based on a capacity-limited and durable memory representation, traditionally referred to as VSTM. In addition, evidence for a non-selective, maskable, and relatively fast decaying representation was found. Following Di Lollo and Dixon (1988), this representation is referred to as the visual analog. We have shown that selectively attending a cued location enables readout and transfer of item information from the visual analog to VSTM, where it can be stored for subsequent comparison with items in the test display. However, readout from the visual analog and transfer to VSTM is time-consuming, requires focused attention, and is interrupted by the appearance of new information or is no longer possible when decay of the visual analog representation has reached the point where it cannot be read out properly. When readout from the visual analog fails, change detection will be determined by the presence or absence of a representation of that item in VSTM. Note that in the latter case, the VSTM representation is created during presentation of the test display, whereas in the first, the VSTM representation is created after presentation of the cue, that is, following readout of the visual analog. 

As explained above, the distinction we make between VSTM and the visual analog is based on our findings of separate memory functions for the no-cue, early-cue, and variable-cue-delay conditions. Recently, Sligte, Scholte, and Lamme (2008; also see Sligte, Scholte, & Lamme, 2009) reported similar effects and interpreted their findings as evidence for a fragile form of VSTM (apart from traditional VSTM), instead of a visual analog representation. We argue that the main findings presented in the present study so far and in the remainder of the study are best captured by characteristics that seem more appropriate to the visual analog than to VSTM: the need to quickly and selectively attend to the spatiotopic memory of the probed item in order to improve change detection, the capacity to read out items that were not attended during the memory display, the quick decrease in performance when readout is delayed, and the susceptibility to masking. Nevertheless, we agree with Sligte et al. (2008) that the issue whether so-called fragile VSTM representations and the visual analog can be equated or not is unresolved. Future research may still show that what we distinguish as two forms of memory are in fact on the same continuum. 

Concluding our discussion of Experiment 1, we would like to add two final notes. First, in Experiment 1, all types of trials (no-cue, early-cue, and variable-cue-delay) were presented in random order to all participants. Although there is no a priori reason to expect that this would have an effect, we ran three separate experiments (with three independent, naive groups of participants, ranging from 4 to 6 subjects) corresponding to the three main conditions of Experiment 1. We replicated the findings of Experiment 1. Second, one could claim that letters are overlearned visual stimuli and that generalization of the present findings to vision of more complex objects is unwarranted. We therefore repeated Experiment 1 with line drawings of everyday objects (obtained from van Diepen & De Graef, 1994). An example of a memory display is shown in Figure 5. The findings of Experiment 1 were replicated. 

Figure 5

 

Example of a memory display in a replication of Experiment 1 with line drawings of objects instead of letters as stimuli.

Now that we have established the properties and the role of both VSTM and the visual analog in a within-fixation change detection paradigm, we return to the main topic of the present article, that is, transsaccadic memory. As outlined in the Introduction section, our main goal is to assess whether transsaccadic memory is restricted to representations held in VSTM or whether it also contains a visual analog representation. Experiments 2–5 examined performance in the detection of changes across eye movements. Similarly to the within-fixation Experiment 1, displays consisted of eight letters (or line drawings of everyday objects) presented in a circular pattern. During the initial memory display, participants were fixating the center of the array. Apart from the 8 objects, the initial display contained a line segment pointing to one of the objects. Participants were instructed to make an eye movement to this object as rapidly as possible. During this saccade, the memory display was erased and a postsaccadic image was presented. Depending on the specific conditions, the test display was presented either directly following the saccade or postponed for a brief period of time (blanking). In the latter conditions, a cue indicating the location of the to-be-probed item was inserted at different moments during the blank interval. As in Experiment 1, the test display contained a single item and remained on screen until the participant pressed one of two response buttons indicating the same or different response. 

To study the question whether transsaccadic memory includes a visual analog representation, we first need to set out what kind of evidence would allow for such a conclusion. First, to date, blanking effects have only been observed for the saccade target object or objects in the vicinity of the saccade target. In other words, blanking effects have only been observed for entities that most probably received focused attention prior to the saccade. A blanking effect for items that were not attended before the saccade, however, would constitute more convincing evidence for a transsaccadic visual analog representation. Second, readout and transfer of a specific item from the visual analog to VSTM requires selective attention. This implies that blanking in itself should not be sufficient to regain additional and/or more detailed information. In other words, delaying the postsaccadic reafference of a stimulus without at the same time redirecting attention to the location where information has to be read out from the visual analog should not improve change detection performance. Third, information within the visual analog is subject to decay. This implies that delaying the redirection of attention to a specific location in the visual field during the blanking interval should decrease the likelihood that an item can be read out from the visual analog. Fourth, while VSTM holds a non-maskable representation, readout from the visual analog representation can be interrupted by a mask. This implies that the presentation of a masking cue at the to-be-probed item location during a blanking interval should not affect change detection performance if detection were purely based on a VSTM representation. If, on the other hand, the presentation of a localized mask instead of a normal cue causes a decrease in performance, then selective readout from a visual analog may be prevented. Finally, if selective readout from the visual analog is responsible for the blanking effect, then readout and transfer of information should show the same properties as in the fixation experiment. That is, readout and transfer to VSTM should be relatively slow and interrupted whenever a new stimulus appears in the visual field. In the following experiments (Experiments 2–5), we will examine each of these points in detail. 

Experiment 2

Experiment 2 included four conditions, one in which the test display was present immediately following the critical saccade from the center of the display to one of the items in the memory display (no-blank condition) and three in which the display was blanked for 950 ms followed by the test display (see Figure 6). The first blanking condition contained no cue during the blanking interval (blank/no-cue condition), while the second and third contained a small red disk for 50 ms, indicating the location at which the probe item in the test display would appear. The cue could either appear 50 ms into the blanking interval (blank/early-cue condition) or 450 ms into the blanking interval (blank/late-cue condition). 

Figure 6

 

Schematic illustration of events and time course in the no-blank, blank/no-cue, blank/early-cue, and blank/late-cue trials of Experiment 2 (Sacc = saccade).

The no-blank condition served as a baseline to which the effects of blanking without cue and blanking in combination with cueing could be assessed. If the blanking effect observed in previous studies is due to readout from a transsaccadic visual analog representation, we predicted performance in the blank/early-cue condition (and possibly to some degree in the blank/late-cue condition, depending on the decay function of the visual analog) to be better than in the no-blank condition. 

In all conditions, we expected change detection performance to be best for items presented at or near the saccade target location. Because attention precedes eye movements (e.g., Deubel & Schneider, 1996; Hoffman & Subramaniam, 1995; Kowler et al., 1995), objects at these locations are preferentially coded in VSTM (e.g., Irwin & Gordon, 1998; Schneider, 1999). However, if a blanking effect is observed, we are specifically interested to see whether this effect is restricted to items in the saccade target region or whether detection at locations more distant from the saccade target also improves. Under the assumption of a non-selective visual analog representation, blanking effects should not be limited to items at or near the saccade target location. 

Methods

Participants

Two paid University of Leuven students and the first author participated in Experiment 2. All participants reported normal vision. 

Stimuli and apparatus

The stimuli and apparatus were identical to those used in Experiment 1. Transsaccadic display changes were typically achieved within 21 ms. Saccades were detected within 4 ms of the onset of the saccade by an online classification algorithm. Subsequent display changes were accomplished within 16.7 ms because they could be initiated at any moment during the screen refresh cycle (the display-change command did not wait for the vertical blank). 

Procedure

Each trial consisted of the following events. First, a fixation cross appeared in the center of the display. Participants were instructed to fixate the cross and calibration accuracy was checked. Once a 250-ms stable fixation was detected on the fixation cross (i.e., a stable fixation within an invisible square of 0.5° × 0.5° enveloping the fixation cross), it disappeared and the presaccadic letter array was displayed. In addition to the circular array of letters, the initial display contained a prompt that indicated the letter to which a saccadic eye movement had to be made. The prompt was a radial line segment, 2.5° in extent, pointing from the center of the array to a location occupied by one of the letters. Participants were instructed to saccade as rapidly as possible to the letter once the initial display was presented. If an eye movement was not initiated within 400 ms following onset of the memory display, the trial was interrupted and the viewer was warned to make more timely saccades. Contingent upon the detection of a saccade leaving the center of the screen, a change to a postsaccadic display was made intrasaccadically. If the saccade to the previously indicated letter did not land within an invisible square of 2° × 2° enveloping the presaccadic letter, and this within 100 ms after saccade onset, the trial was interrupted and the viewer was warned to make more accurate saccades. From the moment the eyes had landed within the specified region, participants were free to move their eyes anywhere within the display. 

The four conditions differed from each other in the events that occurred after the saccade (Figure 6). In the no-blank condition, the test display was present directly following the eye movement. In the blank/no-cue condition, a blank field was presented for 950 ms after saccade landing, followed by the test display. The same was true for the blank/early-cue and late-cue conditions, except that a cue was presented for 50 ms during the 950-ms blanking interval. The cue was a red disk 0.25° in diameter. It was presented centered on a location previously occupied by one of the letters. In the blank/early-cue condition, the cue appeared after 50 ms of the blanking interval (i.e., 50 ms blank/50 ms cue/850 ms blank), in the blank/late-cue condition the cue appeared after 450 ms (i.e., 450 ms blank/50 ms cue/450 ms blank). The test display remained on screen until the participant pressed one of two response buttons indicating the same or different response. Once the response was recorded, a feedback message was displayed in the center of the screen for 300 ms (correct or wrong) followed by the beginning of a new trial after an interval of 1,500 ms. 

Design

Each subject participated in three practice blocks and 16 experimental blocks of 128 trials. In total, each participant completed 2,048 experimental trials that were a factorial combination of 4 conditions (no-blank, blank/early-cue, blank/late-cue, and blank/no-cue), 8 saccade target locations, 8 possible letter change positions, 2 change conditions (same or different), and 4 repetitions. The experiment was completed in 5 sessions (one practice session and four experimental sessions) that each lasted approximately 1 h and 15 min. 

Results and discussion

Trials that were interrupted (i.e., saccade latencies above 400 ms or spatially inaccurate saccades) were excluded from subsequent analyses, as were trials with signal loss, changes that did not occur intrasaccadically, or blinks occurring before the critical saccade to the saccade target location. Of the trials, 10.5% were eliminated in total. Average duration of the saccade from the center of the circular array to the indicated saccade target location was 57 ms (range 35–80 ms) measured from saccade onset to the end of the postsaccadic lens overshoot (Deubel & Bridgeman, 1995; Van Rensbergen et al., 1993). Because saccade-contingent display changes took only 21 ms, they were all completed well before the end of the saccade. Saccade latencies were 235 ms on average. 

In this and subsequent experiments, we were specifically interested in two questions: First, does blanking affect change detection performance and second, if so, is the blanking effect limited to items at or near the saccade target object? To answer these questions, change detection performance, expressed as the percentage correct responses, was analyzed in a repeated-measures ANOVA with trial type and relative probe position as within-subject variables. As shown in Figure 7, relative probe position refers to the position of the probed item relative to the saccade target and has five levels. If the probed item was the saccade target object, relative probe position was set at 0, indicating no difference between probe and saccade target location. If the probed item was in a location adjacent to the saccade target, it was labeled 1 and so on, until level 4, which indicated that the probed item was in a location opposite to the saccade target location. 

Figure 7

 

Assignment of relative probe position. Arrow indicates saccade direction.

The analysis revealed main effects of trial type, F(3, 6) = 15.00, p < 0.004, MSE = 0.0028, and relative probe position, F(4, 8) = 174.96, p < 0.0001, MSE = 0.0020. In addition, the interaction between trial type and relative probe position was significant, F(12, 24) = 2.43, p < 0.04, MSE = 0.0014. Figure 8 shows detection performance as a function of trial type and relative probe position. The interaction was largely due to a ceiling effect at relative probe position zero, regardless of trial type. In other words, detecting a change to the saccade target item was near perfect and unaffected by trial type. 

Figure 8

 

Percentage correct detection in Experiment 2 as a function of relative probe position (T = saccade target) and trial type.

Because of this ceiling effect, the data were reanalyzed excluding trials in which the saccade target was probed. The main effects of trial type, F(3, 6) = 14.49, p < 0.004, MSE = 0.0033, and relative probe position, F(3, 6) = 19.04, p < 0.002, MSE = 0.0022, remained reliable. However, the interaction between the two factors disappeared, F(9, 18) = 1.43, p = 0.25, MSE = 0.0016. Change detection performance for non-saccade target items in the no-blank, blank/no-cue, blank/late-cue, and blank/early-cue conditions were 54.6%, 56.9%, 62.4%, and 67.9%, respectively. Planned comparisons between the no-blank condition and blanking conditions revealed no significant blanking effect for the blank/no-cue condition, F(1, 2) = 5.12, p = 0.15, MSE = 0.0011, a marginal effect for the late-cue condition, F(1, 2) = 10.77, p < 0.09, MSE = 0.0047, and a significant effect for the early-cue condition, F(1, 2) = 219.06, p < 0.005, MSE = 0.0006. Quite surprisingly, change detection performance for non-saccade target items did not decline as a function of distance to the saccade target (i.e., best for items adjacent to and worst for items contralateral to the saccade target). Instead, performance was worst when relative probe position was 2 (55.4%), somewhat better for positions 1 and 3 (60.9% and 60.3%), and best for position 4 (69.8%). This is a data pattern that we replicated in Experiments 3–5. We will return to this observation in the General discussion section. 

The results obtained in the present experiment can be summarized as follows. First, under blanking and no-blanking conditions, detection of changes in the saccade target item across saccades was almost perfect and superior to detection performance of changes to other items. In addition, change detection was extremely poor for non-saccade target items (further referred to as bystander items) when the test display was presented immediately following the saccade. Under the assumption that focused attention is needed to store an item into VSTM, this seems to indicate that attention was almost entirely restricted to the saccade target during presentation of the memory display. A second observation is that blanking the postsaccadic display without presenting a cue did not improve detection performance. Instead, focused attention to the location of a to-be-probed item (by means of a cue) was needed to obtain a blanking effect (blank/cue conditions). Third, a blanking effect was observed for non-saccade target or bystander items. Fourth, the blanking effect for bystander items did not decline as a function of eccentricity from the saccade target. Finally, when attention was not immediately directed to the location of a to-be-probed item during the blanking interval (blank/late-cue condition), the blanking effect was strongly reduced. Together these results lend support to the hypothesis that transsaccadic memory includes more information than is observed under the usual no-blanking circumstances. The question whether this additional information is related to a transsaccadic visual analog representation remains hypothetical. However, the present results do show three properties that are to be expected if a visual analog representation were involved: A blanking effect is observed for items that are less likely to be attended presaccadically (bystander items), focused attention is required to use this representation for change detection (blank/no-cue vs. blank/early-cue), and the representation shows signs of relatively fast decay (blank/early-cue vs. blank/late-cue). 

Note that analogous to the case of Experiment 1, we repeated Experiment 2 with line drawings of everyday objects instead of letters as stimuli (see Figure 5 for an example). The findings of Experiment 2 were replicated. 

Experiment 3

In Experiment 3, we further investigated the representational format that supports the blanking effect as observed in the blank/early-cue conditions of Experiment 2. More specifically, we investigated whether the transsaccadic representation that supports the blanking effect is susceptible to spatiotopic masking. It is generally assumed that VSTM holds a non-maskable representation and information can be accessed directly, while readout from the visual analog representation is maskable (e.g., Irwin, 1992b). This implies that the presentation of a masking cue at the to-be-probed item location during a blanking interval should not affect change detection performance if detection were purely based on a VSTM representation. If, on the other hand, selective readout from a visual analog is involved, presenting a localized mask instead of a non-masking, normal cue should strongly affect performance. In the latter case, the mask would prevent readout from the visual analog, presumably because it overwrites information within the visual analog. 

To assess whether the transsaccadic representation that supports the blanking effect is susceptible to spatiotopic masking, two types of cues were used. The first cue was a small red disk, identical to the cue used in Experiment 2. The second type of cue was a square box presented in such a way that a letter would just fit within its contours. Although both types of cues were presented at the spatial location of a presaccadically presented letter, we assumed that only the square cue would in principle be capable of masking the letter. The choice of this second type of cue was based on the results of partial report experiments conducted within fixations (e.g., Averbach & Coriell, 1961) and across saccades (Irwin, 1992a). For example, in a transsaccadic partial report study by Irwin (1992a), both a traditional bar probe appearing above or below an array location and a square probe that overlapped spatially with the location of a letter were used. Irwin found a masking effect of the spatially overlapping probe when the interval between the memory display and the probe (which included the time of an eye movement) was 40 ms. 

The present experiment had three conditions in which the postsaccadic display was blanked for 950 ms and one condition without blanking. The first two blanking conditions contained a cue throughout the blanking interval (see Figure 9), either a small red disk (blank/constant-cue condition) or a square box (blank/constant-mask-cue condition). Note that we could not simply use the early-cue condition of Experiment 2 as a control for the condition with a metacontrast mask. The reason is that the metacontrast mask remained present throughout the postsaccadic blank, whereas the red disk in Experiment 2 was shown for only 50 ms. To avoid this confound between cue duration and cue type, the constant-cue condition was added as a control condition. The third blanking condition was identical to the blank/early-cue condition used in Experiment 2. Here, the cue was a small red disk that appeared 50 ms into the blanking interval and remained there for 50 ms. (We will stick to the term blank/early-cue condition even though in the present experiment the term “early” is somewhat inappropriate when compared to the blanking conditions with a constant cue, because in these conditions the cue appears even earlier (50 ms).) Finally, a no-blank condition, identical to that in Experiment 2, was introduced to set a baseline to which blanking effects could be assessed. If the blanking effect is supported by a maskable visual analog representation, performance should be worse with the square-like cue than when the cue consisted of a small disk. In addition, the blank/early-cue condition and no-blank condition served as control conditions to assess blanking effects in the two blanking conditions with a constant cue. 

Figure 9

 

Schematic illustration of events and time course in the no-blank, blank/early-cue, blank/constant-cue, and blank/constant-mask-cue trials of Experiment 3 (Sacc = saccade).

Methods

Participants

Two paid University of Leuven students and the first author participated in Experiment 3. All participants reported normal vision. 

Procedure

The trial sequence prior to the critical saccade to the saccade target letter was identical to that in Experiment 2. Figure 9 shows the time course and events of a single trial for each of the four conditions. In the no-blank/no-cue condition, the test display was present directly following the eye movement. In the blanking conditions, a blank field was presented for 950 ms after saccade landing followed by the test display. In the blank/constant-cue and blank/constant-mask-cue conditions, a cue was present throughout the blanking interval. In the blank/early-cue condition, the cue appeared after 50 ms of the blanking interval and remained on screen for only 50 ms (i.e., 50 ms blank/50 ms cue/850 ms blank). In the blank/constant cue and the blank/early-cue conditions, the cue was a red disk, 0.25° in diameter, presented centered on a location previously occupied by one of the letters. In the blank/constant-mask-cue condition, the cue was a transparent box that measured 1.25° by 1.25°. Figure 10 illustrates the size of the box relative to the size of the letter stimuli. The box was presented such that the probed letter would appear centered within it. The outline of the box was black and had a line thickness comparable to that of the letter stimuli (0.10°). The trial sequence following the presentation of the test display was identical to that in Experiment 2. 

Figure 10

 

Illustration of cue type in Experiment 3. (Left) Non-masking cue. (Right) Masking cue.

Results and discussion

On the basis of the same criteria as in Experiment 2, 12.3% of the trials were eliminated from further analysis. Average latency and duration of the saccades from the center of the circular array to the saccade target location on the remaining trials was 213 ms and 57 ms (range 35–79 ms), respectively. 

Change detection performance (Figure 11) was analyzed in a repeated-measures ANOVA with trial type and relative probe position as within-subject variables. The analysis revealed a main effect of trial type, F(3, 6) = 23.43, p < 0.001, MSE = 0.0026, and relative probe position, F(4, 8) = 145.51, p < 0.0001, MSE = 0.0022. In addition, the interaction between trial type and relative probe position was significant, F(12, 24) = 3.44, p < 0.005, MSE = 0.0016. Similar to Experiment 2, detection of changes in the saccade target item showed ceiling performance, regardless of trial type. 

Figure 11

 

Percentage correct detection in Experiment 3 as a function of relative probe position (T = saccade target) and trial type.

Reanalysis of the data, excluding trials in which the saccade target was probed, revealed main effects of both trial type, F(3, 6) = 20.65, p < 0.002, MSE = 0.0035, and relative probe position, F(3, 6) = 5.27, p < 0.05, MSE = 0.0029. The interaction between the two factors disappeared, F(9, 18) = 1.97, p = 0.11, MSE = 0.0018. Change detection performance for non-saccade target items was 52.6% in the no-blank condition, 59.3% in the blank/constant-mask-cue condition, 67.6% in the blank/constant-cue condition, and 68.3% in the blank/early-cue condition. Planned comparisons between the no-blank condition and blank/early-cue condition revealed a significant blanking effect, F(1, 2) = 35.53, p < 0.03, MSE = 0.0046. Performance in the blank/early-cue condition, which served as a control condition, was better than in the blank/constant-mask-cue condition, F(1, 2) = 24.68, p < 0.04, MSE = 0.0030, but did not differ from the blank/constant-cue condition (F < 1). The difference between performance in the blank/constant-mask-cue condition and no-blank condition did not reach significance, F(1, 2) = 5.12, p = 0.15, MSE = 0.0011. 

The results can be summarized as follows. First, similar to Experiment 2, detection of changes to the saccade target item was almost perfect. The fact that the presentation of a masking cue during the 950-ms blanking interval did not affect performance for the saccade target item shows that the transsaccadic representation of a saccade target object is held in a durable, non-maskable store, presumably VSTM. Second, a blanking effect for non-saccade target items was again observed in both conditions with a non-masking cue. Third, the blanking effect severely decreased when a square-like cue instead of a small red disk was presented at the location of the to-be-probed item during the entire blanking interval. This suggests that the representation that supports the blanking effect is susceptible to masking, which is consistent with the properties of a visual analog. 

Note that masking was spatiotopic. It has indeed been suggested that informational persistence is organized spatiotopically. This seems to stand in contrast to proposals that a spatiotopic representation is not necessary for explaining transsaccadic visual stability. Instead, stability could be based on a remapping of retinotopic representations (e.g., Melcher & Colby, 2008), maybe even implemented in the brain as the remapping of the receptive fields of neurons (Colby & Goldberg, 1999; Duhamel, Colby, & Goldberg, 1992; but see Cavanagh, Hunt, Afraz, & Rolfs, 2010). However, we agree with Melcher that theories stressing remapping and theories with a spatiotopic focus “are not necessarily mutually exclusive and may, instead, capture different aspects of a process that aligns—across saccades or other body movements—the network of spatial maps in the brain that encode objects and actions” (Melcher, 2009, p. 1250). Moreover, we want to argue that, contrary to existing accounts on the basis of predictive remapping, transsaccadic representations are not restricted to objects that received attention presaccadically. 

Experiment 4

The results obtained in the previous experiments suggest that the representation that supports the blanking effect decays fairly rapidly, requires selective attention to be used, and is susceptible to masking. So far these properties are all in line with the characteristics of a transsaccadic visual analog. In Experiment 4, we examined another property of the visual analog. In the within-fixation Experiment 1, it was shown that readout from the visual analog and transfer to VSTM is a time-consuming process: Detection performance increased when more time was available for readout from the visual analog and transfer to VSTM. If selective readout from the visual analog is responsible for the blanking effect observed in the transsaccadic experiments, then readout and transfer of information should show the same properties as in the fixation experiment. That is, readout and transfer to VSTM should be relatively slow and interrupted whenever a new stimulus appears in the visual field. 

To test whether this property can also be observed in a transsaccadic change detection paradigm, Experiment 4 included four conditions in which the time available for readout and transfer was manipulated. As shown in Figure 12, in all four conditions, the test display was blanked for a certain period of time. A non-masking cue was always presented 50 ms into the blanking interval for a period of 50 ms. Following offset of the cue, the time available for readout of the visual analog representation and transfer to VSTM was manipulated. The time between offset of the cue and the appearance of the test display was 50, 250, 450, or 850 ms. 

Figure 12

 

Schematic illustration of events and time course of a trial in Experiment 4 (Sacc = saccade).

Methods

Participants

Two paid University of Leuven students, all reporting normal vision, and the first author participated in Experiment 4. 

Procedure

The trial sequence prior to the critical saccade to the saccade target letter was identical to Experiments 2 and 3. During this saccade, the screen was cleared and 50 ms after saccade landing, a non-masking cue was presented for a duration of 50 ms. The cue was again a red disk, identical to that used in Experiments 2 and 3, presented at one of the locations previously occupied by a letter. Following offset of the cue, the screen was still blanked for a variable period of time and then the test display appeared. The time between offset of the cue and the appearance of the test display, further referred to as postcue blanking time, was either 50 ms, 250 ms, 450 ms, or 850 ms. Once the test display was presented, the procedure was identical to that in the previous experiments. 

Results and discussion

On the basis of the same criteria as in Experiment 2, 11.9% of the trials were eliminated from further analysis. Average latency and duration of the saccades to the saccade target location on the remaining trials was 211 ms and 56 ms (range 36–79 ms), respectively. 

Change detection performance (Figure 13) was analyzed in a repeated-measures ANOVA with postcue blanking time and relative probe position as within-subject variables, which revealed main effects of postcue blanking time, F(3, 6) = 29.07, p < 0.0006, MSE = 0.0009, and relative probe position, F(4, 8) = 466.80, p < 0.0001, MSE = 0.0006. In addition, the interaction between postcue blanking time and relative probe position was significant, F(12, 24) = 2.36, p < 0.04, MSE = 0.0014. As expected, detection of changes in the saccade target item showed ceiling performance, regardless of postcue blanking time. 

Figure 13

 

Percentage correct detection in Experiment 4 as a function of relative probe position (T = saccade target) and postcue blanking time.

Reanalysis of the data, excluding trials in which the saccade target was probed, showed main effects of both postcue blanking time, F(3, 6) = 28.79, p < 0.0006, MSE = 0.0012, and relative probe position, F(3, 6) = 110.87, p < 0.0001, MSE = 0.0006. In addition, the interaction between the two factors disappeared (F < 1). Change detection performance for non-saccade target items was worst when postcue blanking time was limited to 50 ms (58.8% correct). Performance gradually increased with longer postcue blanking times (65.5% correct with 250 ms blanking time) and asymptoted once postcue blanking time reached 450 ms (69.9% and 70.4% correct with 450- and 850-ms blanking times, respectively). This suggests that, in order to obtain a blanking effect, a certain amount of time is needed between the presentation of a cue during the blanking interval and the onset of a new stimulus. This result is in agreement with what was found in the within-fixation Experiment 1. We interpreted this effect in terms of a relatively slow readout and transfer process from the visual analog to VSTM, which is interrupted whenever a new stimulus appears in the visual field. The strong resemblance between the within-fixation results and the present results obtained within a transsaccadic blanking paradigm suggests that the representation supporting the transsaccadic blanking effect is the same visual analog representation. 

Experiment 5

Experiment 5 aimed at exploring transsaccadic memory given a somewhat better approximation of object perception as achieved during real-world scene perception. While eye movements in scene exploration are generally initiated from a foveal object (i.e., the saccade source) to some object in peripheral vision (e.g., De Graef et al., 1990; Einhäuser, Spain, & Perona, 2008; Henderson & Hollingworth, 1998), this was not the case in Experiments 2–4. In these experiments, participants were required to saccade from an empty location to a designated saccade target object. This situation differs in terms of both attentional and oculomotor demands from one in which a to-be-identified source object is present in foveal vision prior to saccade initiation. In fact, the presence of a source object creates a foveal load that may attenuate the quality, duration, and spatial extent of extrafoveal processing (Henderson & Fereira, 1990; Ikeda & Takeuchi, 1975; Kennedy, 1998; Liversedge & Findlay, 2000). To explore possible effects of foveal load on transsaccadic memory, Experiment 5 included blanking and no-blanking conditions in which viewers either initiated a saccade from an empty location (cf. Experiments 2–4) or from a central source object to a designated target object. 

Experiment 5 contained two blanking conditions, in which a 50-ms cue was presented 50 ms following saccade landing. The first was a condition without foveal load, identical to the blank/early-cue conditions of Experiments 2–4. The second blanking condition was identical to the first with the exception that an additional letter (further referred to as the source item) was presented in the center of the presaccadic display (blank/early-cue-foveal-load). In this condition, participants were instructed to encode the source and then make an eye movement to the saccade target. In one ninth of the trials, the source was cued during the blanking interval. In addition to the blanking conditions, two no-blanking conditions, one without foveal load (no-blank/no-foveal-load) and one with foveal load (no-blank/foveal-load) were introduced. As in the previous experiments, the no-blanking conditions served as a baseline to assess the blanking effect. 

Change detection performance for the saccade target in the previous experiments (i.e., without foveal load) was almost perfect under blanking (with and without cueing) and no-blanking conditions, suggesting that the target, which receives selective attentional processing prior to saccade execution, was stored in VSTM. Introducing a foveal source may decrease the extent to which the target is processed prior to saccade execution. This implies that in the condition without blanking, which is assumed to reflect change detection based on a VSTM representation, we may observe a decline in performance when a source is present (i.e., no ceiling performance for the target as observed in the no-blanking conditions of Experiments 2, 3, and 4). If this is indeed observed, a blanking effect on the target object may reveal itself (e.g., De Graef & Verfaillie, 2002; Deubel et al., 2002). 

For the bystander objects, performance in the previous experiments (i.e., without foveal load) was almost at chance in the conditions without blanking and improved strongly when a blank together with a cue was presented following saccade landing. In Experiment 5, we were specifically interested to see whether this blanking effect would still hold in the presence of a foveal source. Under the assumption that a preattentive visual analog supports improved performance for bystander objects under blanking conditions, we expected a similar effect. 

Methods

Participants

Two paid University of Leuven students and the first author participated in Experiment 5. All participants reported normal vision. 

Procedure

The procedure was identical to previous experiments with the following exceptions. Similar to previous experiments, the initial letter display contained a circular array of eight letters. In trials without foveal load, the central location was empty, whereas on trials with foveal load, an additional letter was presented at the center of the circular array (i.e., the source). This letter was chosen from the 8 remaining letters that did not appear in the circular array. In addition to the letter stimuli, the initial display contained a 2° radial line segment that indicated the letter to which a saccadic eye movement had to be made. On blanking trials, the postsaccadic display was blanked for 950 ms and a cue (50-ms duration presented 50 ms following saccade landing) was presented indicating which location would be probed in the test display. In the blanking condition with foveal load, the cue was presented centrally in one ninth of the trials. The test display always contained a single item, presented at the cued location. For the foveal-load condition, this meant that the source was probed in one ninth of the trials. The no-blank/no-foveal-load and no-blank/foveal-load conditions were identical to their respective blanking conditions, with the exception that the test display was present directly following saccade landing. 

Design

Each subject participated in three practice blocks and 16 experimental blocks of 136 trials. Each participant completed 1,024 no-foveal-load trials, which were a factorial combination of 2 conditions (no-blank/no-foveal-load and blank/no-foveal-load), 8 saccade target locations, 8 possible letter change positions, 2 change conditions (same or different), and 4 repetitions, and 1,152 foveal load trials, which were a factorial combination of 2 conditions (no-blank/foveal-load and blank/foveal-load), 8 saccade target locations, 9 possible letter change positions (eight locations on the circular array plus the central location), 2 change conditions (same or different), and 4 observations. In total, each participant completed 2,176 experimental trials. The experiment was completed in 5 sessions (one practice session and four experimental sessions), which each lasted approximately 1 h and 20 min. 

Results and discussion

On the basis of the same criteria as in Experiment 2, 11.2% of the trials were eliminated from further analysis. Saccadic latencies were, on average, 39 ms longer with foveal load than without foveal load, with an overall mean of 226 ms. Average duration of the saccades was 57 ms (range 37–78 ms). 

Figure 14 shows detection performance as a function of blanking, foveal load, and relative probe position. Unlike the previous experiments, relative probe position now has 6 instead of 5 levels. This is due to the foveal-load conditions in which the source could be probed. Because the source cannot be probed in the no-foveal-load conditions, relative probe position in those conditions had only five levels. As shown in Figure 14, when the probed item was either the source item (in the foveal-load conditions) or the saccade target item, change detection performance was near perfect regardless of trial type. Because ceiling performance was observed for the source and the saccade target, and because relative probe position in the no-foveal-load conditions does not include data on the source, we focused the analysis on non-target and non-source items, which will be referred to as bystander items (i.e., relative probe positions 1 to 4). 

Figure 14

 

Percentage correct detection in Experiment 5 as a function of relative probe position (S = source, T = saccade target) and trial type.

Change detection performance, expressed as the percentage correct responses, was analyzed in a repeated-measures ANOVA with blanking, foveal load, and relative probe position (1 to 4) as within-subject variables. The analysis revealed main effects of both blanking, F(1, 2) = 262.50, p < 0.004, MSE = 0.0011, and relative probe position, F(3, 6) = 21.32, p < 0.002, MSE = 0.0007. There was no significant main effect of foveal load, F(1, 2) = 1.19, p = 0.39, MSE = 0.0020. In addition, the interaction between blanking and foveal load was significant, F(1, 2) = 192.74, p < 0.006, MSE = 0.0001. No further interactions involving any of the factors were observed (ps > 0.15). 

Following up on the interaction between blanking and foveal load, there was no effect of foveal load under no-blanking conditions, F(1, 2) = 2.55, p = 0.25, MSE = 0.0014. Mean proportions correct in the no-blank/no-foveal-load and no-blank/foveal-load conditions were, respectively, 51.2% and 53.8%. Under blanking conditions, foveal load caused a significant decrease in performance, F(1, 2) = 27.33, p < 0.04, MSE = 0.0006. Mean percentages correct in the blank/no-foveal-load and blank/foveal-load conditions were, respectively, 69.9% and 64.4%. Together, these results show that foveal load only affected change detection performance for bystander items when the postsaccadic display was blanked. The effect of blanking was significant under both no-foveal-load, F(1, 2) = 714.12, p < 0.002, MSE = 0.0006, and foveal-load conditions, F(1, 2) = 92.97, p < 0.02, MSE = 0.0009. 

To summarize, the present results show that items presented at fixation during the presentation of the memory display (i.e., the source item) were perfectly encoded, retained, and integrated with new information following the saccade, regardless of whether the postsaccadic display was blanked or not. The same is true for saccade target items; change detection performance was near perfect in both blanking and no-blanking conditions. Furthermore, performance for the target was not affected by the presence of a presaccadic source item. This suggests that the source and the target, which receive selective attention prior to saccade execution, were durably stored, presumably in VSTM. 

For bystander items, a completely different pattern of results was observed. Under normal circumstances, that is, no-blanking conditions, change detection performance was at chance (52.5% on average), suggesting that bystander items may not have been encoded at all. However, similarly to our previous transsaccadic experiments, change detection performance increased drastically when the postsaccadic display was blanked (reaching 67.3% on average). This result clearly indicates that bystander items were encoded presaccadically, even when change detection failed completely in the no-blanking conditions. Blanking the postsaccadic display and directing focused attention to the to-be-probed item location enables the visual system to use information, presumably in the form of a visual analog, that is otherwise inaccessible for change detection purposes. 

More important with regard to the aim of the present experiment is the observation of a blanking effect when the presaccadic display contained a foveal source object. Change detection performance in the foveal-load conditions showed a significant improvement when the postsaccadic display was blanked (53.8% vs. 64.4%). This is an important finding because it generalizes the blanking effect for bystanders to a situation that is a better approximation of normal object perception (i.e., a situation in which a saccade is initiated from a to-be-identified foveal source object instead of an empty location). However, foveal load did affect the magnitude of the blanking effect. That is, less information on bystander items seemed to be regained under blanking circumstances when the presaccadic display contained a foveal source item than when it did not. Performance in the blanking condition with foveal load was approximately 5% worse compared to the no-foveal-load condition (i.e., 64.4% vs. 69.9%). We return to this in the General discussion section. 

General discussion

The present study assessed what kind of information is retained and integrated across a saccade. Previous research suggests that transsaccadic integration is supported by a durable, relatively abstract, and capacity-limited memory store, presumably VSTM. However, recent studies seem to indicate that more detailed location and form information can be remembered across a saccade than could be expected based on the characteristics of VSTM. In particular, when a target to which an eye movement is made is absent during the saccade and for a short period at the beginning of the postsaccadic fixation (i.e., blanking), the visual system seems to regain access to more detailed information compared to a situation without blanking (De Graef & Verfaillie, 2002; Deubel et al., 1998, 1996, 2002; Gysen et al., 2002). De Graef and Verfaillie (2002) suggested that the transsaccadic blanking effect might be supported by a representation similar to the visual analog as observed in within-fixation experiments (i.e., a high-capacity, non-selective, maskable representation of the form and location of objects, which decays approximately 300 ms following stimulus offset; Di Lollo & Dixon, 1988; Irwin & Yeomans, 1986). The goal of the present study was to further assess whether transsaccadic memory is restricted to a VSTM representation or, as suggested by De Graef and Verfaillie (2002), whether it also contains a representation similar to the visual analog. 

The results of Experiment 1 confirm previous findings that, within fixations, both a durable but capacity-limited VSTM representation and a non-selective, maskable, relatively fast decaying visual analog representation play a role. Beyond this distinction, we have shown that selectively attending a cued location after stimulus offset enables readout and transfer of information from the visual analog to VSTM, where it can be stored for subsequent comparison with items in the test display. However, if the cued item is not retrieved from the visual analog, because either the readout and transfer process from the visual analog to VSTM is interrupted by the appearance of a new stimulus, or decay of the visual analog representation has reached the point where it cannot be read out properly, change detection will be determined by the presence or absence of a representation of that item in VSTM. In the latter case, the VSTM representation is created during presentation of the test display, whereas in the first, the VSTM representation is created after presentation of the cue, that is, following readout of the visual analog. 

Four transsaccadic change detection experiments were carried out to investigate whether memory across fixations is restricted to VSTM representations or whether it also includes a visual analog component. Experiment 2 served to answer three questions. First, is the blanking effect restricted to the saccade target object or can it also be observed for bystander objects? Second, is blanking in itself a sufficient condition to obtain an increase in change detection performance or is focused attention at the to-be-probed item location during the blanking interval necessary? Third, if focused attention is needed, does postponing the redirection of attention to the to-be-probed item location reduce the blanking effect? If the blanking effect is supported by a visual analog representation, we expected blanking effects for both saccade target and bystander objects, but only if the to-be-probed item location is given focused attention and this within a restricted time frame following saccade initiation. Experiment 2 confirmed these predictions. Change detection for the saccade target was almost perfect under no-blanking conditions, while detection of changes in bystander items was extremely poor, presumably indicating that presaccadic attention allocation was restricted to the target, limiting coding in VSTM largely to this object. Moreover, blanking in itself did not improve change detection for bystanders, while blanking and at same time cueing the location of the to-be-probed item did. In addition, the blanking effect was reduced if the cue, which is assumed to redirect attention, was postponed, supporting the hypothesis of a relatively fast decaying visual analog. 

In Experiment 3, we investigated whether the transsaccadic representation that supports the blanking effect is susceptible to masking, which would be consistent with the properties of a visual analog. Of main interest was a comparison between the magnitude of the blanking effect in two blanking conditions with different cues presented throughout the blanking interval (relative to a no-blanking control condition). The presentation of a masking cue (a square box presented in such a way that a letter could fit within its contours) during the blanking interval did not affect detection performance for the saccade target object, indicating that this item was stored in a durable, non-maskable store, presumably VSTM. However, for bystander items the blanking effect observed with the masking cue (the square box) was severely reduced relative to the non-masking cue (a small red disk). In fact, the blanking effect with the masking cue did not reach significance. 

The within-fixation Experiment 1 showed that it takes a certain amount of time to read out and transfer information from the visual analog to VSTM. Moreover, this readout and transfer process was shown to be interrupted by the appearance of a new stimulus. In Experiment 4, we tested whether these properties, associated with a visual analog, could also be observed in a transsaccadic context. More specifically, the time between offset of the cue and the appearance of the test display (i.e., postcue blanking time) was manipulated and ranged between 50 and 850 ms. Detection performance for the saccade target was almost perfect and unaffected by postcue blanking time. However, for bystander items a similar pattern of results was obtained as in the within-fixation experiment: Performance improved with increasing postcue blanking time and then asymptoted, suggesting an analogous readout and transfer process from the visual analog to VSTM, which is interrupted by the appearance of a new stimulus. 

Finally, Experiment 5 examined the blanking effect given a better approximation of object perception as achieved during real-world scene perception. Instead of having viewers saccade from an empty location to a designated target object (no foveal load), participants initiated a saccade from a foveal source object on half of the trials (i.e., foveal load). Change detection was almost perfect for the saccade source and the target object, indicating that selectively attended objects prior to saccade execution were durably stored, presumably in VSTM. Furthermore, a blanking effect for bystanders was observed when the presaccadic display contained a foveal source object. However, foveal load did affect the magnitude of the blanking effect. Performance in the blanking condition with foveal load was slightly reduced compared to the condition without foveal load. 

The transsaccadic visual analog

Experiments 2–5 provide evidence that transsaccadic memory includes a highly reliable representation, which seems to be restricted to objects that receive selective attentional processing prior to saccade execution (i.e., normally the saccade target and source object; also see Currie, McConkie, Carlson-Radvansky, & Irwin, 2000; De Graef, Verfaillie, & Lamote, 2001; McConkie & Currie, 1996; Verfaillie & De Graef, 2000). This representation proved to be durable (i.e., showed no signs of decay), resistant to masking, and readily accessible for change detection purposes, suggesting that transsaccadic memory includes a VSTM representation of presaccadically attended objects. The present experiments, however, indicate that transsaccadic memory is not restricted to VSTM representations. Instead, it also seems to include a more volatile representation, which, under normal circumstances, is inaccessible for change detection purposes. Consistent with previous studies (De Graef & Verfaillie, 2002; Deubel et al., 1998, 1996, 2002; Gysen et al., 2002), the visual system only appears to regain access to it if the postsaccadic image is blanked for a certain amount of time. 

On the one hand, comparison of the characteristics of the visual analog as observed in the within-fixation Experiment 1 with those of the representation supporting the transsaccadic blanking effect showed remarkable similarities—the necessity of blanking and cueing (and associated attentional orienting), the relatively fast decay, and the time-consuming readout, which can be interrupted by a mask—suggesting that the blanking effect reflects the use of a visual analog similar to that observed in within-fixation conditions. 

On the other hand, some of the results seem to question the idea that the transsaccadic visual analog is completely identical to its within-fixation version. First, despite the fact that change detection performance improved considerably under blanking conditions in the transsaccadic experiments, performance was far from perfect (e.g., approximately 72% correct in the blank/early-cue condition of Experiment 2). Yet, in the within-fixation Experiment 1, detection performance in the conditions that most closely match their transsaccadic counterparts was almost perfect (e.g., approximately 90% in the condition with a fixed ISI of 1000 ms and a cue delay of 100 ms). This result may suggest a fundamental difference between the transsaccadic visual analog and within-fixation visual analog in terms of capacity. 

A second finding that points in a similar direction was observed in Experiment 5. In this experiment, the blanking effect for bystanders was somewhat reduced when the presaccadic display contained a foveal source object. Although speculative, this reduced blanking effect may be understood in terms of a possible capacity limit in the transsaccadic visual analog. Under the assumption that the transsaccadic visual analog can only maintain a restricted number of items across a saccade, the foveally presented source item may have been maintained at the expense of other peripheral items in the array. More specifically, the number of unattended bystander items that can be maintained in the transsaccadic visual analog will then be reduced by one item. This hypothesis assumes a transsaccadic visual analog that faithfully represents a restricted amount of items. Furthermore, within this view, encoding of an object in the transsaccadic visual analog is assumed to be a necessary condition for subsequent attentional processing of an object prior to saccade execution. Thus, objects that receive selective attentional processing prior to saccade execution are then coded in both the transsaccadic visual analog and VSTM. 

This characterization of the transsaccadic visual analog may be intimately linked to the visual indexing or FINST (short for “fingers of instantiation”) theory as proposed by Pylyshyn (1989, 2001; Trick & Pylyshyn, 1994). This theory assumes an early preattentive stage in vision in which the location of properties of a small number of salient items are selected and indexed prior to any serial processes being applied to the visual array (e.g., prior to the allocation of limited attentional resources to an indexed item). In terms of transsaccadic integration, it has been argued that this preattentive, limited capacity indexing mechanism may serve to establish spatiotemporal continuity of a restricted set of objects across a saccade (Hayhoe, Karn, Magnuson, & Mruzcek, 2001; Pylyshyn, 2001). As such, encoding of items in the transsaccadic visual analog may be based on and restricted to items that are spatially indexed prior to saccade initiation. Research on the allocation of visual indexes has also shown that indexes need not be assigned to contiguous or nearby items in the visual array. Instead, indexes can be assigned to several disparate items scattered throughout the visual field (Bichot, Cave, & Pashler, 1999; Intriligator & Cavanagh, 1992; Sears & Pylyshyn, 2000; but see Dubois, Hamker, & Van Rullen, 2009). 

The notion of a limited capacity transsaccadic visual analog may explain the absence of a blanking effect for bystander objects in the study by De Graef and Verfaillie (2002). In this study, the effect of blanking on the discrimination of intrasaccadic in-depth orientation changes of saccade target and bystander objects was examined. Even though transsaccadic memory for the saccade target showed a clear benefit from blanking, memory for the bystander did not. However, because stimuli consisted of complex scenes that contained a multitude of objects and viewers did not know in advance (i.e., prior to saccade initiation) which bystander would be probed in the postsaccadic display. If only a limited set of bystanders can be held in the transsaccadic visual analog, the likelihood that the to-be-probed bystander would have been included is small in the presence of multiple objects. The saccade target, in contrast, would always be spatially indexed prior to saccade initiation, explaining the blanking effect for the saccade target. 

The distribution of the attentional field prior to the saccade

The observation that performance for other stimuli than the saccade target mostly was not at chance level suggests that, in addition to the target stimulus, a non-target stimulus could be attended (and transferred to VSTM) on a proportion of the trials. Moreover, performance was not equal for all non-target locations. Instead, change detection was best when the stimulus diametrically opposite to the saccade target was probed and worst when the probed stimulus was at an intermediate distance from the saccade target. Similar effects have been documented before, but we can only speculate on the possible underlying mechanism. 

Studies in which subjects had to decide whether two precued stimuli on a circular array of letters were identical or different (e.g., Bahcall & Kowler, 1999; Skelton & Eriksen, 1976) have shown that accuracy was highest for diametrically opposite pairs of cued targets (except for adjacent letters in Skelton & Eriksen's study). One possible explanation for this effect is that two letters in opposite locations on a circle stimulate opposite hemispheres. There is indeed evidence for superior visual processing when stimuli are presented in separate hemifields (Sereno & Kosslyn, 1991; also see Alvarez & Cavanagh, 2005; Delvenne, 2005). In the present study, letters that occupied positions that were 180° apart indeed always fell in opposite hemifields prior to saccade execution (the target was never presented in the exact top or bottom position of the circular array). However, Tse, Sheinberg, and Logothetis (2003) documented similar effects for stimuli arranged along the vertical axis and therefore not falling into separate hemifields (also see Müller & Kleinschmidt, 2004; Müller, Mollenhauer, Rösler, & Kleinschmidt, 2005). 

Of the experiments reported by Cutzu and Tsotsos (2003), the paradigm in their Experiment 4 comes closest to our paradigm. Subjects were presented with a circular array (with a radius of 4°, as in our experiments) of 12 letters. The array contained rotated versions of all Ts or all Ls (target absent) or one L among Ts or one T among Ls (target present) and subjects had to detect an odd L among Ts (or vice versa). Prior to this test image (presented for 100 ms), one letter location was precued for 180 ms. The findings were exactly the same as in our experiments. When the odd letter was at the cued location, detection accuracy was high. Away from the cue, accuracy initially decreased but then increased again, peaking when the target was diametrically opposite to the cued location. Cutzu and Tsotsos interpret their findings as evidence for the existence of an inhibitory zone, of annular shape and limited extent, surrounding an attended location, giving rise to a Mexican hat-shaped distribution of the attentional field. However, such an explanation would entail that the extent of the inhibitory annulus can be quite large (because both in Cutzu & Tsotsos' Experiment 4 and in our experiments, accuracy was lowest for letters that were in an intermediate location between the cued letter and the letter at the opposite side of the ring of letters), while most studies supporting perceptual suppression around the attentional focus suggest smaller regions of suppression (see Müller et al., 2005, for an overview; also see Müller & Kleinschmidt, 2004). 

Tse et al. (2003; also see Tse, 2004) used a change detection paradigm to map the distribution of spatial attention across the visual field in response to a task-irrelevant peripheral cue. Subjects were presented with a 25° circular window with 149 possible locations, half of which were filled with equiluminant green or red squares. A cue was then briefly flashed in one of four possible locations, followed by a 47-ms blank screen and a test display showing the original squares in addition to one new square in a previously unoccupied position. Subjects had to report the color of the new square. The rationale behind this study was that the blank masks the local transient accompanying the appearance of the new element, thereby interfering with the detection of the change, and that only attention to a location both in the pre- and the postblank images can overcome this change blindness. The map of attentional distribution showed a “hot spot” of attention elongated along the fixation–cue axis with attention pooling not only at the cued location but also at the location opposite the cued location. In this study (as well as in the studies reviewed above), subjects had to maintain fixation during a trial (or at least prior to the appearance of the test display). Tse et al. (2003) explicitly raise the question “whether opposite pooling [of attention] occurs even when observers are permitted to saccade to the cue” (p. 98). The present study for the first time shows that this is indeed the case. Apparently, an attention shift as a preparatory process for a saccadic eye movement stretches the attentional beam along an axis defined by the attentional focus prior to the shift of attention (mostly the foveal stimulus) and the target of the relocation of attention (e.g., a peripheral cue). Moreover, the data of Tse et al.'s study in which subjects had to maintain fixation, in combination with our data in which subjects were required to make a saccade, suggest that this elongation extends to locations opposite the target of the attentional (and eventually saccadic eye) shift. 

The possible functional role of the transsaccadic visual analog

Our results support the hypothesis that transsaccadic memory holds a preattentive visual analog representation that may be capacity-limited. This transsaccadic representation only revealed itself under specific blanking circumstances. Because postsaccadic blanking never occurs in natural vision, the question arises what the functional role of a transsaccadic visual analog could be in a normal viewing situation (i.e., without blanking). 

Before speculating on this issue, it is important to note that a preattentive mechanism is never observed directly. What is observed in the present blanking studies are the postattentive consequences of a preattentive representational mechanism. Specifically, when the postsaccadic display was blanked and a bystander cued, readout and transfer from the transsaccadic visual analog to VSTM, comparison between pre- and postsaccadic information, and a subsequent decision and response selection process are all likely to involve serial attentional components. Thus, exposing the presence of a transsaccadic visual analog always requires more than that preattentive representation itself, and these additional processes may be very specific to the blanking manipulation and the task of explicitly detecting intrasaccadic changes. In sum, this implies that the postattentive consequences observed by the blanking manipulation may only indirectly indicate the true functions of the preattentive representational system during normal transsaccadic object perception. 

However, why, then, could a transsaccadic visual analog be useful in everyday vision? One of the principal functions of the visual system is to identify objects in the environment. Accordingly, during normal scene perception, transsaccadic memory may help object identification. While a single foveal view of an object may in principle be sufficient to identify an object, the preferred modus operandi in the exploration of the environment may be to increase speed and reliability of object identification by transsaccadic integration of foveal and extrafoveal evidences across a saccade. Several studies investigating transsaccadic object perception have shown that transsaccadic memory representations indeed speed up identification of elements in the postsaccadic image (e.g., Demeyer et al., 2009; Germeys, De Graef, & Verfaillie, 2002; Henderson, 1994; Henderson & Anes, 1994; Pollatsek et al., 1984, 1990; Verfaillie, De Graef, Germeys, Gysen, & Van Eccelpoel, 2001). The hypothesis that we want to put forward is that a specific component of transsaccadic memory, namely the visual analog, plays an important role in this process of object identification across saccades. The visual analog can be regarded as reverberatory activity, allowing stimulus representations to remain active in the visual system after the corresponding objects have disappeared from the retina, thereby creating the possibility to integrate presaccadic and postsaccadic information in order to facilitate object identification. In other words, even though information stored in the visual analog normally remains inaccessible for conscious processes (so that it cannot be revealed except by creating artificial conditions, i.e., a specific combination of presaccadic cueing and postsaccadic blanking, to allow the visual system to store the information in a more durable VSTM), it expedites transsaccadic object identification at a preattentive level. 

This can be understood in the framework of the model of transsaccadic object identification, put forward by Germeys et al. (2002). While most accounts of transsaccadic preview benefits (i.e., facilitation by a peripheral preview on subsequent foveal object identification) have stressed the need for presaccadic allocation of attention to a peripheral object in order to obtain a preview benefit (e.g., Henderson, 1994; Henderson & Anes, 1994; Irwin, 1996), Germeys et al. (2002) observed preview benefits for both presaccadically attended (i.e., the saccade target) and non-attended objects (i.e., bystanders). Further results also showed that the representation supporting preview benefits for non-attended objects was strictly location bound. On the basis of these results, Germeys et al. (2002) proposed a model of transsaccadic object identification that includes a preattentive transsaccadic memory representation (in addition to an attentive representation). The model proposes that prior to saccade initiation, the visual system sets up a limited number of so-called “preattentive object files”. These preattentive object files are created in parallel on the basis of preattentive grouping processes and hold a loose collection of unbound features that seem to belong together. Similar to FINST theory, each object file is location indexed, thus providing the visual system with a limited set of potential targets for subsequent attention and gaze shifts. Once this preattentive stage is completed, one preattentive object file is attentionally selected for further processing (i.e., feature binding), and if required, a saccade to the attended object is executed. In summary, this model assumes that transsaccadic preview benefits for unattended objects are produced by a limited set of location-indexed preattentive object files that are maintained across a saccade. Because of the similarity between this preattentive representation and the transsaccadic visual analog put forward on the basis of the present blanking experiments, it is tempting to conclude that they are in fact the same. Accordingly, the capacity-limited transsaccadic visual analog may serve to aid transsaccadic object identification by providing memory for a limited number of preattentive object files that, if selected for further attentional processing, may expedite object identification. 

This view also fits well—albeit it speculative at present—with approaches that stress the interplay between feedforward processing, on the one hand, and recurrent or reentrant processing, on the other hand (e.g., Ahissar & Hochstein, 1997; Di Lollo, Enns, & Rensink, 2000; Hochstein & Ahissar, 2002; Lamme & Roelfsema, 2000). Based on the finding that there are massive backprojections from higher cortical areas to lower level areas, it is assumed that visual perception involves, first, a rapid, automatic, and preattentive feedforward sweep of activation through the cortical hierarchy leading to an initial perceptual hypothesis about the stimulus and, second, a slower, attention-modulated, reentrant or recurrent stream of processing. The architecture of reciprocal connectivity is eminently suited to search for a match between the descending code representing the initial perceptual hypothesis and the ongoing pattern of low-level activity. When such a match occurs, the perceptual hypothesis is confirmed and conscious awareness follows. Guided by this framework, Germeys et al.'s (2002) model of transsaccadic object identification may be rephrased as follows: (a) attentional object file reviewing and recurrent processing are equivalent, (b) the visual analog can be regarded as low-level reverberatory activity (at different levels), allowing stimulus representations to remain in the visual system (after the corresponding objects have disappeared from the retina) and enabling the comparison with reentrant perceptual hypotheses, (c) a preattentive object file is analogous to an implicit feedforward path of activation (which can be traced back by attentional reentrant processing), (d) an attentive object file (VSTM representation) is akin to a durable, explicit, integrated percept that is a consequence of iterative reentrant processing. 

The object substitution masking paradigm has proven to be an excellent tool to better understand reentrant processing (e.g., Enns & Di Lollo, 1997; Germeys, Pomianowska, De Graef, Zaenen, & Verfaillie, 2010; Jiang & Chun, 2001). In this paradigm, a non-overlapping mask is presented together with a target object. When target and mask disappear simultaneously from the visual field, discrimination of the target is excellent. In contrast, when the mask stays on after the target has been turned off, performance deteriorates as a function of the duration of the trailing mask. We assume that the observed masking of the target is a direct consequence of the object file review process being hampered by new feedforward activation of the mask. We are currently combining object substitution masking and transsaccadic paradigms to further validate this model. 

Acknowledgments

This research has been supported by Concerted Research Effort Convention GOA 98/01 and 05/03/TBA from the Research Fund K.U. Leuven and by Project 033816/EUKP6-IST from the European Community. 

Commercial relationships: none. 

Corresponding author: Karl Verfaillie. 

Email: [email protected]. 

Address: Department of Psychology, University of Leuven, Tiensestraat 102, B-3000 Leuven, Belgium.