Limits to the Use of Iconic Memory

Transcription

1 Limits to Iconic Memory 1 Limits to the Use of Iconic Memory Ronald A. Rensink Departments of Psychology and Computer Science University of British Columbia Vancouver, BC V6T 1Z4 Canada Running Head: Limits to Iconic Memory Abstract: 147 Word count (body): 3420 Word count (notes + Acknowledgement) 581 Author Notes Most of this work was done while the author was with Cambridge Basic Research, Nissan Research & Development, Inc., Cambridge, MA USA. Many thanks to Duncan Bryce, Puishan Lam, Nayantara Santhi, and Monica Strauss for their help in running the experiments. Correspondence should be addressed to R.A. Rensink, Department of Psychology, University of British Columbia, 2136 West Mall, Vancouver BC V6T 1Z4, Canada. ( rensink@psych.ubc.ca or rensink@cs.ubc.ca).

2 Limits to Iconic Memory 2 Abstract Human vision can retain a trace of a stimulus for several hundred milliseconds after it disappears. This trace iconic memory is widely believed to be a surrogate for the original stimulus, i.e., its contents can be used as if the stimulus were still visible. This belief was tested here by a full-scan technique based on visual search. Results show that iconic memory can be a surrogate for some processes, with little cost for switching between visible and iconic representations. However, they also show that different limits exist on the duration usable for different tasks. It is proposed that the relevant factor is the establishment of re-entrant (feedback) connections from higher to lower levels of visual processing. In this view, the purpose of iconic memory is to establish and maintain such connections, with the usable duration reflecting the extent to which re-entry is involved in a given task.

3 Limits to Iconic Memory 1 Limits to the Use of Iconic Memory It has long been known that human vision can retain a trace of a stimulus for several hundred milliseconds after its disappearance. This trace, often referred to as iconic memory, has been extensively studied for decades (e.g., Coltheart, 1980; Becker et al., 2000; Sperling, 1960). It is widely believed to be a visual echo that can act as a surrogate, i.e., that as long as it lasts, its contents can be used in the same way as if the stimulus were still visible. However, there is little consensus as what role if any iconic memory plays. It has sometimes been considered a simple side-effect, with potentially deleterious effects on processing (Haber, 1983). On the other hand, it might be a common resource for visual processing in general, potentially supporting many kinds of basic operations. Traditionally, iconic memory has been studied via partial report, in which observers are briefly presented with an array of items and then asked to report a subset that is cued after the presentation disappears (Averbach & Coriell, 1961; Sperling, 1960). Other studies have examined the extent to which iconic and visual representations can be swapped in memorization and recognition tasks (e.g., Keysers et al., 2005; Loftus et al., 1985). All of these assume that iconic memory is equally available to any process that uses it. However, given its potential involvement in all basic aspects of vision, it is important to determine the extent to which this is true and in particular, the extent to which iconic memory supports various kinds of visual processes. To determine this, a full-scan technique was developed which minimizes the effects of higher-level processes such as memory consolidation or transfer. This is based on visual search, where the observer must determine as quickly as possible the presence or absence of a given target among a set of

4 Limits to Iconic Memory 2 nontarget items (or distractors) in a display; a variety of visual operations can be tested by appropriate choice of target and distractors. Here, observers seach displays that appear intermittently: after a given display time (on-time), the display is blanked for a brief interval (ISI or off-time), with this sequence repeated until the observer responds or times out (Fig. 1). For most stimuli, search requires scanning by focused attention (e.g., Treisman and Gormican, 1988; Wolfe & Horowitz, 2004), with time needed proportional to the number of items. If their scanning speed is sufficiently slow, observers must make at least one full scan of both the visible representation and the corresponding iconic memory i.e., they must scan during both on- and off-times 1. The question is then whether for a given task, scanning iconic memory is the same as scanning the stimulus that gave rise to it. This can be determined by comparing performance when iconic memory is used for different fractions of a display cycle. GENERAL METHOD In the experiments here, each condition used three timing patterns, or cadences: a base cadence of 80/120 (80 ms on; 120 ms off), and longer cadences of 80/240 and 200/120, created by increasing the off-time and on-time by 120 ms respectively. Twelve observers participated in each condition, with the order of cadences counterbalanced. Lighting level was low, but sufficient to allow colour to be easily seen. Each display subtended 11.5 x 8.5 and contained 2, 6, or 10 items, with the target present on a randomly-selected half of trials. The blank fields and display backgrounds were both medium gray, resulting in a continual flickering of the items on a static background. Observers were instructed to maintain fixation between trials and were encouraged to keep error rates low. Responses were given by pressing one of

5 Limits to Iconic Memory 3 two response keys. All observers completed 4 sets of 60 trials in each condition; performance was generally measured in terms of the speed of search, as given by the average target-present slope 2. A trial was timed out (and counted as an incorrect response) if more than 5 seconds was needed. EXPERIMENT 1 The target in this condition was a black vertical line of length 0.8, with the distractors (nontargets) similar lines oriented ±30 to the vertical (Fig. 1a). Observers were asked to detect whether the flickering display contained a vertical line, and do this as quickly as possible, keeping error rates below 5%. In a completely static display, this type of search typically has targetpresent slopes of ms/item (e.g., Treisman & Gormican, 1988). Similar speeds were encountered here (Fig. 1b). Importantly, no effect of cadence on slopes was found (F(2,10) = 0.006; p >.9). Analysis of baselines (Fig. 1c) also failed to show an effect of cadence (F(2,10) = 0.29; p >.7). Error rates were also similar for the different cadences, indicating that no speed-accuracy trade-offs had occurred. These results suggest that the information in iconic memory can survive without serious degradation for at least 240 ms, consistent with conclusions obtained elsewhere (e.g., Sperling, 1960). Attentional scanning was also found to have much the same speed in both, supporting the proposal that attentional selection and iconic memory involve common representations (Ruff et al., 2007), and indicating that the iconic representation can be used as easily and effectively as the visual one. In addition, these results indicate that switching between visible and iconic representations requires little or no time. Taken together, these results support the view that iconic memory can act as a surrogate of the visible

6 Limits to Iconic Memory 4 input at least over the spatial extents tested and that it can be used in this capacity for at least 240 ms Place Figure 1 about here EXPERIMENT 2 To examine whether these conclusions hold for other tasks, Experiment 2 used the same items and timing cadences, with the target now an item that changed orientation. Each display contained approximately equal numbers of verticals and lines tilted counterclockwise by 30 ; the target was the item that changed its orientation between displays (Fig. 2a). In contrast with Experiment 1, a strong effect of cadence was now found (F(2,11) = 33.1; p <.0001; Fig. 2b). In particular, search slowed with increased offtime (p <.001), but not increased on-time (p >.3). Recasting slopes in terms of the number of items held across the temporal gap (Rensink, 2000), the effect of cadence was still evident (F(2,11) = 22.6; p <.0001; Fig. 2c). Interestingly, there was no significant difference in hold for increased off-time (p >.05), but a significant increase with greater on-time (p <.003). This supports the proposal that the speed of this task reflects the loading of visual short-term memory and its subsequent comparison (Rensink, 2000); indeed, it indicates that these operations take place only during on-time, plus some additional duration. More precisely, a comparison of slopes 3 shows that speed is a function of on-time plus a usable duration u=112 ms ±19 ms. Past this limit, iconic memory apparently cannot be used, even though Experiment 1 shows that it still contains sufficient information Place Figure 2 about here

7 Limits to Iconic Memory 5 EXPERIMENT 3 To further assess the generality of this effect, Experiment 3 investigated other kinds of change (Fig. 3). Condition 3A examined orientation changes larger than those of Experiment 2. Items were rectangular outlines 0.4 x 1.2, with targets changing orientation 90 between vertical and horizontal (Fig. 3A). Again, search speeds depended strongly on cadence (F(2,11) = 16.7; p <.0001), with search slowing for increased off-time (p <.0005) but not for increased ontime (p >.05). Usable duration u was 112 ms ±26 ms, much the same as before Place Figure 3 about here Condition 3B examined location change. Here, the target moved back and forth 1.2 each alternation, with distractors remaining stationary. Slopes again depended on cadence (F(2,10) = 12.2; p <.0005), with search slowing for increased off-time (p <.001) but not increased on-time (p >.2). Duration u was 118 ms ± 34 ms, similar to previous values. Condition 3C looked at shape change, with the target alternating between a circle and a square. Although more difficult than the other conditions, similar results were still found: slope depended on cadence (F(2,11) = 11.2; p <.0005), slowing for increased off-time (p <.006) but not increased on-time (p >.7). Duration u was 136 ms ±36 ms, compatible with previous values. Condition 3D examined changes in contrast polarity (black vs. white). Slopes again depended on cadence (F(2,11) = 7.9; p <.003). Search was not affected by increased off-time (p >.05), but sped up with increased on-time, although significance was marginal (p =.047). This latter effect has been found elsewhere, where it was taken to indicate a grouping process based on contrast polarity that occurs over several hundred milliseconds (Rensink, 2000). Comparing performance for the 80/240 and 200/120 cadences (which equates

8 Limits to Iconic Memory 6 time per alternation) shows the cadence with greater on-time to be reliably faster (p <.005); relative speeds yield u = 140 ms ± 45 ms, comparable with the results for other kinds of change. Thus, all change-detection tasks appeared to show the same kind of behavior, with approximately the same useable duration u. EXPERIMENT 4 Experiment 4 investigated how different tasks affect u. To determine if task difficulty might be relevant, Condition 4A had observers detect a fixed target with a horizontal bar located only slightly higher than those of the distractors. Speeds were comparable to those encountered in Experiments 2, 3A, 3B, and 3D (Fig. 4a). However, no dependence on cadence was found (F(2,11) =0.79; p >.4), indicating that difficulty per se was not the key factor Place Figure 4 about here Condition 4B asked observers to report the orientation of a T-shaped target (left or right) among L-shaped distractors (Fig. 4b). A dependence on cadence now reappeared (F(2,11) =11.9; p <.0003), with search slowing for increased off-time ( p <.001) but not increased on-time (p >.6). Usable duration u was 175 ms ± 26 ms, larger than the values found for the change-detection tasks, but still less than the full duration of 240 ms. To test whether the key factor in Condition 4B was the existence of multiple kinds of target, Condition 4C asked observers to detect a T-shaped target among L-shaped distractors; all items targets as well as distractors could be in any of 4 orientations (Fig. 4c). No dependence on cadence was found (F(2,11) =1.22; p >.3). Finally, to determine the left-right distinction was relevant, Condition 4D asked observers to detect an L-shaped figure with the stem facing left among

9 Limits to Iconic Memory 7 similar figures with stems facing right (Fig. 4d). No reliable dependence on cadence was found ( F(2,11) =0.74; p >.4). Taken together, these results suggest that the key factor in Condition 4B was the general type of the task, viz., ascertaining some property (direction) about a detected target. GENERAL DISCUSSION These experiments confirm that for some tasks iconic memory can act as a surrogate of the visual input for at least 240 ms: during this time it can be used as easily and effectively as if the stimulus were visible, at least for the part of the visual field examined here. They also indicate little or no cost of switching between a visual representation and its corresponding memory. However, these results also show that for some tasks iconic information cannot be used beyond a short duration. The key factor appears not to be the difficulty of the task or the information involved, but the type of task itself. Earlier work showed that iconic memory has at least two distinct components. The first is a high-density, retinotopic visible persistence that supports visual integration at relatively peripheral levels; this lasts about 100 ms from stimulus onset (Eriksen & Collins, 1967; Di Lollo & Bischof, 1995). The second is a longer-lasting, more spatiotopic informational persistence mediated more centrally (Coltheart, 1980, Loftus & Irwin, 1998). For the experiments here, the difference in onset times between displays was far greater than 100 ms, ruling out visible persistence as a major factor. (In accord with this, items were always seen as flickering and not as fused.) Meanwhile, informational persistence can last for several hundred milliseconds, and so was likely the type of persistence involved. Keysers et al. (2005) found that visible persistence can be a surrogate of visual input for recognition tasks. The results here show that informational

10 Limits to Iconic Memory 8 persistence can be a surrogate as well, and can act as such for at least 240 ms for some tasks. The usable duration of 120 ms encountered here may relate to the finding that iconic memory can provide the equivalent of 110 ms of visual input for memorization tasks (Loftus et al., 1985). Given that the tests here were based on visual scanning rather than memory consolidation, it is unclear whether the same mechanisms are involved 4. In any event, these results show that the usable amount of iconic memory is not something indicative of its overall quality or general accessibility, but is different for different kinds of task: For some (e.g., change detection) it is about 120 ms, for others (e.g., identification) it is about 180 ms, and for yet others (e.g., static detection) it is at least 240 ms. How can these different limits be explained? An interesting possibility involves re-entrant (or feedback 5 ) connections from higher-level visual areas to lower-level ones. Static patterns can be detected by neurons in areas such as temporal cortex; cells here have a considerable degree of spatial invariance, responding to much of the visual field (Felleman & Van Essen, 1991; Bullier. 2004). However, to see an item change or to identify it after detecting it requires that it be individuated, i.e., treated as a particular individual at a particular location (Pylyshyn & Storm, 1988; Smith, 1998). This must be done by linking its spatially-invariant representation at the higher level with its lowest-level retinotopic location. Such registration can be established by correlating downward, spatially-diffuse signals from higher levels with upward, spatiallyprecise signals from striate cortex (Di Lollo et al., 2000, MacLean & Tsotsos, 2000). (More generally, the representation of any item is likely cascaded, i.e., distributed over several levels, each with an increasing amount of spatial invariance.) But such correlation can be carried out only while reasonably precise information exists at the lower levels; once this has dissipated, the process must be suspended. A process requiring precise registration will

11 Limits to Iconic Memory 9 therefore exhibit photo-gating it will operate only while visible input exists. In this view, usable duration reflects the extent to which a given process is dominated by the establishment and maintenance of such re-entrant connections. Results from other lines of research support this possibility. Neurophysiological studies suggests that iconic memory in macaque monkeys involves neurons in area STSa of temporal cortex, with both spatial invariance as well as the ability to sustain firing for a few hundred milliseconds past stimulus offset (Keysers et al., 2005). Anatomically, massive re-entrant connections are known to exist between the cortical areas involved in visual perception (e.g., Felleman and Van Essen, 1991). Behaviorally, such connections have been used to explain phenomenon such as common-onset masking (Di Lollo et al., 2000) and context effects in recognition (Weisstein & Harris, 1974); indeed, models based on feedback suggest that it could be involved in a wide variety of visual processes (Fukushima et al, 1991; Tsotsos et al., 2008). Meanwhile, work on iconic memory indicates the existence of two forms of localization: within-item localization (local relationships for item shape) and absolute localization (registration with lower levels). For example, partial report errors increase when a mask is shown after stimulus disappearance. Two distinct types of error are found: (i) identification errors, which arise only if the mask is shown within the first 150 ms of stimulus onset, and (ii) localization errors, which can be induced when the mask is present much later (Mewhort et al., 1981). These are compatible with the two main types of feedback possible: (i) horizontal (or lateral) connections with adjacent cells at the same level, which converge quickly and could support rapid establishment of shape, and (ii) vertical connections linking corresponding cells at different levels, which are longer, and take more time to converge.

12 Limits to Iconic Memory 10 This proposal also explains why performance on iconic and visible representations is so similar, and why there is little or no cost for switching between them: iconic representation is simply a higher-level, more spatiallyinvariant component through which normal visual perception proceeds, and which briefly remains in use after the lower-level representations have dissipated. In this view, iconic memory has a clear purpose: A relatively longlasting memory organized in terms of relative location is exactly what is needed to establish and maintain links between the spatially-invariant representation of an item at higher levels (e.g., in temporal cortex) and the ever-changing retinotopic representation that exists at lower visual levels (e.g., striate cortex). Indeed, iconic representations may be closely related to and perhaps even identified as those at preattentive levels. It has been proposed that the representation of any item in this form of storage [i.e., iconic memory] is achieved by creating a temporary file of information (Coltheart, 1983 p. 291), with relatively complex contents (such as characters) created in parallel across the visual field, and which are susceptible to overwriting by subsequent stimuli (Coltheart, 1983; Mewhort et al., 1981). This is similar to the proposal of protoobjects in visual search (Rensink & Enns, 1998), which are relatively complex pieces of limited extent formed rapidly and in parallel, and which are temporary, either fading away within a few hundred milliseconds, or being overwritten by any new item that appears at their location (Rensink et al., 1997). The appeal to rapid horizontal and slower vertical connections could also help explain why some forms of within-item localization can be achieved rapidly and without attention (e.g., Enns & Rensink, 1990; Rensink & Enns, 1998), while other kinds of localization require more time and attention (e.g., Treisman & Gormican, 1988; Wolfe & Horowitz, 2004).

13 Limits to Iconic Memory 11 What might the value of usable duration indicate? If photo-gating occurs only while the stimulus is visible, usable duration would reflect the extent to which the process involves feedforward alone. However, part of usable duration may reflect a third form of persistence, a photo-persistence at the retinotopic level that lasts for some time perhaps 100 ms or so past stimulus offset. If so, photo-gating would also occur while photo-persistence lasts. Note that this kind of persistence would not be encountered in most tests of iconic memory, being subsumed in the longer-lasting informational persistence; it would only be manifest in tests involving re-entrant processes. Some physiological support exists for this possibility: cat retinal ganglion cells can fire for about 60 ms past stimulus offset (Levick and Sacks, 1970), while some (sustained) cells in macaque striate cortex fire for about 100 ms after the offset of a 250 ms stimulus (Maunsell & Gibson, 1992). Given that within-item localization requires little time, the representations of the simple items used here would likely be complete by the end of the first display. In Experiments 1, 4A, 4C, and 4D, no operations are required beyond static detection (which does not require individuation), and so usable duration would simply be the extent of informational persistence. In contrast, the detection of change needed in Experiments 2 and 3 needs each item to be individuated, requiring the creation of re-entrant connections. If there is a photo persistence of about 100 ms, a usable duration of 120 ms would indicate an operation heavily dependent on re-entry. Meanwhile, the usable duration of about 175 ms in Experiment 4B would indicate that identification of a detected item involves a different combination of feedforward and feedback processes. In that case, it may be that the detection component can proceed past 120 ms, but then needs to wait whatever the result for individuation to occur so that identification can then be carried out.

14 Limits to Iconic Memory 12 In summary, then, limits exist to the extent that a process can use iconic memory: beyond some duration, it cannot use the information contained there, even if this is still accessible by other processes. It is proposed that the extent of this usable duration reflects the degree to which a process is dominated by reentrant links, with iconic memory needed to drive the establishment of those links. According to such an analysis, re-entrant links do not play a large role in the detection of static patterns, but dominate the detection of change. More generally, by determining the usable durations for various visual processes, insight may be gained as to their component operations, and the extent to which they involve re-entrant connections.

15 Limits to Iconic Memory 13 References Averbach, E., & Coriell, A.S. (1961). Short-term memory in vision. Bell Systems Technical Journal, 40: Becker, M.W., Pashler, H., & Anstis, S.M. (2000). The role of iconic memory in change-detection tasks. Perception, 29: Bullier, J. (2004). Communications between cortical areas of the visual system. In LM Chalupa and JS Werner, eds, The Visual Neurosciences, Vol 1. p Cambridge MA: MIT Press. Coltheart, M. (1980). The persistences of vision. Philosophical Transactions of the Royal Society B: Biological Sciences 290: Di Lollo, V. & Bischof, W.F. (1995). Inverse-intensity effect in duration of visible persistence. Psychological Bulletin 118: Di Lollo, V., Enns, J.T., & Rensink, R.A. (2000). Competition for consciousness among visual events: The psychophysics of reentrant visual processes. Journal of Experimental Psychology: General 129: Enns, J.T., & Rensink, R.A. (1990). Sensitivity to three-dimensional orientation in visual search. Psychological Science, 1: Eriksen, C.W., & Collins, J.F. (1967). Some temporal characteristics of visual pattern perception. Journal of Experimental Psychology, 74: Felleman, D.J. & Van Essen, D.C. (1991). Distributed hierarchical processing in primate visual cortex. Cerebral Cortex 1: Fukushima,K.,Imagawa,T.,Ashida,E. (1991). Character recognition with selective attention. In International Joint Conference on Neural Networks.Vol. 1., Seattle WA; pp Haber, R.N. (1983). The impending demise of the icon: A critique the concept of iconic storage in visual information processing. Behavioral and Brain Sciences 6: Keysers, C., Xiao, D.-K., Földiak P., Perrett, D. (2005). Out of sight but not out of mind: The neurophysiology of iconic memory in the superior temporal sulcus. Cognitive Neuropsychology, 22: Levick, W.R. & Sacks, J.L. (1970). Responses of the cat retinal ganglion cells to brief flashes of light. Journal of Physiology, 206: Loftus, G.R., & Irwin, D.E. (1998). On the relations among different measures of visible and informational persistence. Cognitive Psychology 35:

16 Limits to Iconic Memory 14 Loftus, G.R., Johnson, C.A., & Shimamura, A.P. (1985). How much is an icon worth? Journal of Experimental Psychology: Human Perception and Performance, 11: MacLean, W.J., & Tsotsos, J.K. (2000). Fast Pattern Recognition Using Gradient-Descent Search in an Image Pyramid. Proc. International Conference on Pattern Recognition, Barcelona, Spain. pp Maunsell J.H.R., & Gibson, J.R. (1992). Visual response latencies in striate cortex of macaque monkey. Journal of Neurophysiology 68: Mewhort, D.J.K, Campbell, A.J., Marchetti, F.M., & Campbell, J.I.D. (1981). Identification, localization, and iconic memory : An evaluation of the barprobe task. Memory & Cognition, 9: Pylyshyn, Z.W., & Storm, R.W (1988). Tracking multiple independent objects: Evidence for a parallel tracking mechanism. Spatial Vision, 3: Rensink, R.A. (2000). Visual search for change: A probe into the nature of attentional processing. Visual Cognition 7: Rensink R.A., & Enns, J.T. (1998). Early completion of occluded objects. Vision Research, 38: Rensink, R.A., O'Regan, J.K., & Clark, J.J. (1997). To see or not to see: The need for attention to perceive changes in scenes. Psychological Science, 8: Smith, B.C. (1998). On the Origin of Objects. Cambridge, MA: MIT Press. Sperling, G. (1960). The information available in brief visual presentations. Psychological Monographs 74: Treisman, A., & Gormican, S. (1988). Feature analysis in early vision: Evidence from search asymmetries. Psychological Review, 95: Tsotsos, J.K., Rodriguez-Sanchez, A.J., Rothenstein, A.L., & Simine, E. (2008). The different stages of visual recognition need different attentional binding strategies. Brain Research, 1225: Weisstein, N., & Harris, C. S. (1974). Visual detection of line segments: An object-superiority effect. Science, 186: Wolfe, J.M., & Horowitz, T.S. (2004). What attributes guide the deployment of visual attention and how do they do it? Nature Reviews Neuroscience, 5:

17 Limits to Iconic Memory 15 Notes 1. Observers were dropped from the analysis if search was over before a full scan of the first iconic representation could be made. The criterion used was that search should be slow enough to allow a complete scan of a single ontime and off-time in the fastest cadence. Note that this does not assume an item-by-item scan of the display; the process could proceed in parallel. The main point here is the extent to which the process serial or parallel can use iconic memory. On the basis of this criterion, two observers were removed: one from Experiment 1, and one from Experiment 3B. More severe criteria did not significantly change the overall pattern of results. 2. Slopes for each observer were calculated by determining mean response time for each set size, and calculating a least-squares fit through these points. Analysis used repeated-measures ANOVAs, and paired, two-tailed t-tests. Target-present slopes were used; target-absent slopes either followed the same pattern or showed no strong effects. Error rates in the target-absent condition were generally low (below 2%) and did not vary much over different conditions. Errors for target-present conditions either followed the pattern of the slopes or showed no strong effects, indicating that speedaccuracy trade-off was not a factor. 3. Usable memory duration u was calculated in the following way. The total usable time in each alteration is the duration of the visible component plus the usable duration of the iconic component. Assuming the usable duration in the 80/120 and 200/120 cadences is at least approximately 120 ms (supported by the results here), and assuming the same speed for visible and iconic inputs (supported by the results of Experiment 1A), scanning speed can be estimated by averaging the slopes of these two cadences to get slope s V. As such, the usable fraction f = s V /s L, where s L is the slope in the 80/240 cadence. For this cadence, this fraction is (80+u) / 320; rewriting, u=320f 80. The standard error of the mean of u can be determined from this formula, via the standard errors of the slopes. Note that usable time might have reflected a single duration d that began at stimulus onset. If so, the 80/120 and 80/240 cadences would have an effective on-time of d=80 ms ms=192, and search in the 200/120 cadence would take place only during on-time (200 ms). Slopes for the

18 Limits to Iconic Memory /120 cadence would then be 320/200 = 1.6 times those for the 80/120 cadence, and nearly the same as for the 80/240 cadence. However, the opposite pattern was found. Thus, usable duration u would appear to begin at stimulus offset. 4. The results of Loftus et al. (1985) do not indicate whether the use of iconic memory is largely during the first 110 ms after stimulus offset, or is spread throughout a longer interval, reflecting a more degraded form of information pickup. If similar mechanisms are involved here, the former is the more likely possibility. The results of all detection conditions (Experiments 1, 4A, 4C, and 4D) show no serious degradation in speed over 240 ms the effect of 240 ms of iconic memory equals 240 ms of visible input. 5. As used here, the term re-entry denotes a particular type of feedback, viz., that in which density of back connections is similar to or exceeds the density of forward connections.

19 Limits to Iconic Memory 17 Acknowledgements This research was supported by Nissan Motor Co., Ltd., and the Natural Sciences and Engineering Research Council of Canada (NSERC). No competing financial interests.

20 Limits to Iconic Memory 18 Figure Captions FIG. 1. Experiment 1. (a). Stimuli used. Target is a vertical line; distractors were lines tilted ±30. (b) Slope estimates. Slope for base cadence (24.3 ms/item) is unaffected by either an increase in off-time (24.4 ms/item) or an increase in ontime (24.8 ms/item). Note that since these are target-present slopes from a selfterminating search, the true search speed is obtained by multiplying by 2. The resultant speeds are about 50 ms/item, similar to estimates found elsewhere. (c) Baseline estimates. Baseline is the estimated response time for when the display contains a single item. Baseline for the base cadence (581 ms) is unaffected by an increase in off-time (575 ms) or on-time (593 ms). FIG. 2. Experiment 2. (a) Stimuli used. Approximately 50% of lines in each display are vertical, and 50% are tilted by 30 counterclockwise. Target is the item that changes between vertical and tilted; distractors are those items that maintain a constant orientation. (b) Slopes. Slope for base cadence (53.0 ms/item) is strongly affected by an increase in off-time (93.5 ms/item) but not by an increase in on-time (59.3 ms/item). Error bars indicate standard error of the mean. (c) Hold estimates. The hold is the number of items held across the temporal gap in visual short-term memory and then compared with the subsequent display. Hold for the base cadence (2.1 items) is unaffected by an increase in off-time (1.8 items), but does increase with increased on-time (3.0 items).

21 Limits to Iconic Memory 19 FIG. 3. Experiment 3. (a). Detection of changing orientation. Slope for base cadence (42.7 ms/item) is strongly affected by an increase in off-time (78.1 ms/item) but not an increase in on-time (51.1 ms/item). (b). Detection of changing location. Slope for base cadence (36.0 ms/item) is strongly affected by an increase in off-time (62.9 ms/item) but not an increase in on-time (41.7 ms/item). (c). Detection of changing shape. Slope for base cadence (75.2 ms/item) is strongly affected by an increase in off-time (113.1 ms/item) but not an increase in on-time (77.3 ms/item). (d). Detection of changing polarity. Search in base cadence (45.1 ms/item) is not significantly affected by an increase in offtime (57.0 ms/item) but is sped up marginally by an increase in on-time (39.2 ms/item). Error bars indicate standard error of the mean. FIG. 4. Experiment 4. (a). Detection of offset horizontal line in target. Slope for base cadence (52.1 ms/item) is unaffected by an increase in off-time (57.7 ms/item) or on-time (52.3 ms/item). (b). Identification of T-shaped item. Slope for base cadence (47.4 ms/item) is affected by an increase in off-time (59.8 ms/item) but not an increase in on-time (48.1 ms/item). (c). Detection of T- shaped item. Slope for base cadence (31.3 ms/item) is unaffected by an increase in off-time (34.3 ms/item) or on-time (30.7 ms/item). (d). Detection of left vs right. Slope for base cadence (38.8 ms/item) is unaffected by an increase in offtime (46.3 ms/item) or on-time (41.6 ms/item). Error bars indicate standard error of the mean.

22 T: D: time (a) Search task slope (ms/item) / / / 120 baseline (ms) SOA (ms) (b) Slopes 10 error (%) SOA (ms) (c) Baselines Figure 1

23 T: D: time (a) Search task / / slope (ms/item) / error (%) hold (items) SOA (ms) (b) Slopes SOA (ms) (c) Holds Figure 2

24 T: D: T: D: T: D: T: D: slope (ms/item) error (%) (a) Changing Orientation (90 ) (b) Changing Location (1.2 ) (c) Changing Shape (d) Changing Polarity Figure 3

25 T: D: T: D: T: D: T: D: slope (ms/item) error (%) (a) Detection: Horizontal Offset (b) Identification: T vs L (c) Detection: T vs L (d) Detection: Left-Right Figure 4