Coltheart argues that the existence of an unrecoded internal representation after stimulus termination can be interpreted in three ways. First, visible persistence might reflect neural persistence, that is, persistence of neural activity associated with a visual stimulus beyond its physical duration. Second, visible persistence is associated with a particular phenomenological experience. That is, one can "see" a rapidly fading image of an image after offset. Finally, an observer might only have knowledge of the visual characteristics of a stimulus when in the presence of an unrecoded representation of the stimulus. Coltheart calls this latter interpretation iconic memory; it is evidenced by the fact that information about a physical stimulus is available to an individual after its physical termination. The goal of the current paper is to integrate conceptions of neural persistence, visible persistence, and iconic memory.
Sperling showed that, following a brief presentation of a visual array of stimuli, subjects can report a greater percentage of items in partial report than in full report recall conditions. That is, subjects can recall all, or nearly all, elements when cued to recall one row of a stimulus array, but only about one-third of elements when cued to recall the entire array. This finding can be interpreted as evidence for two forms of memory, one that is fleeting and one more durable. The memory storage mechanisms differ quantitatively. They also differ qualitatively because manipulations that affect iconic memory (e.g., backward masking), but not the more durable form.
The contents of iconic memory must be transferred to the more durable storage before they can be reported. This is the case because initiation of a verbal report exceeds the estimated duration of the iconic trace—about 300 ms. That only about 4 or 5 items from a display can be reported, whether in partial or full report conditions, might reflect a capacity limitation of durable storage. However, the estimated capacity of short-term memory is usually estimated at about 7 items, but the number is in the range. An alternative explanation is a limitation in transfer capacity. More specifically, it could be that only about 4 to 5 items can be transferred at any given time. An final possibility is that, while items are being transferred, the remaining items fade. This explanation is untenable because varying the presentation rate between 15 and about 500 ms has little effect on recall performance. Moreover, while presenting the display on a dark background lengthens the duration of the trace, recall performance remains constant.
What happens in the interval between display offset and cue presentation? If no transfer takes place, then one would expect no items to be recalled if the presentation-cue interval exceeds the duration of iconic memory. This is occasionally found, but more often recall is at about FR level. This suggests non-selective transfer—that is, subjects transfer information in the absence of cue information. The problem with this interpretation is that cued recall performance is 4-5 items, even though nonselective transfer may have occurred. Apparently, nonselective transfer does not interfere with selective transfer and readout. Some evidence is relevant to this issue. For example, Sperling found that subjects used a non-selective transfer strategy (in the form of encoding a particular line of the stimulus array) only for longer cue delay intervals. For shorter intervals, they waited on the cue. This suggests that there is some cost associated with non-selective coding, perhaps in the form of overtaxing the transfer mechanism or overcrowding durable storage.
What kind of information represented in iconic memory? This has been investigated by using different types of partial-report cues. Typically, the cue indicates which line to recall. This indicates that spatial position is represented. In addition, however, PR superiority is found for brightness, shape, and color cues. For example, the usual PR superiority is found when subjects are cued to recall bright, as opposed to dim, items. Higher level information is not represented. For example, PR superiority is not found when subjects are cued to recall letters instead of numbers, or letters containing a particular phoneme. Thus, iconic images seem to be represented in terms of precategorical, visual information.
The basis of research on iconic memory is visual information: "these experiments do not require that this information be visible (though this may be so), or that its neural basis be persisting activity in the visual system (though this may be so too)" (p. 188).
Visible persistence refers to the persistence of an visual image of a stimulus beyond its physical duration. Ways in which the duration of visible persistence can be measured are described below.
Judgment of synchrony
One way to estimate the duration of visual persistence is to instruct subjects to adjust the interval between offset of a test stimulus and onset of a probe stimulus until the two events appear simultaneous.
| | |
test stop probe on (perceived point of simultaneity)
0 200 250
VP = point of probe on – physical duration
VP = 250 - 200
Estimates using this method indicate that the duration of visible persistence varies according to the duration of the test stimulus presentation. Persistence duration is inversely related to stimulus duration and intensity when stimulus durations are less than 130 ms. The relationship between persistence duration and stimulus duration seems to be constant when the stimulus duration exceeds 130 ms, but there is some evidence that longer stimuli show zero persistence when the test and probe stimuli are in different modalities. Overall, it remains unclear whether longer stimuli show any persistence.
Using this approach, the subject makes a reaction time response to both the onset and offset of the test stimulus. The duration of visible persistence is then calculated by subtracting the onset RT from the offset RT (offset RT – onset RT). The problem with this approach is that because visible persistence decays exponentially, the criterion about when to respond will be somewhat arbitrary. Relative measures of VP duration can nevertheless be obtained. Results indicate that longer stimuli produce visible persistence when they are presented peripherally, but not foveally. Consistent with other results, visible persistence for foveally presented shorter stimuli is inversely related to stimulus duration. These conclusions notwithstanding, some have argued that an important limitation of RT methodology in visible persistence research is that even stimuli rendered invisible can trigger an RT response. Thus, as Coltheart (1980) notes, "Detection of offset stimulus may trigger and offset RT, even though the stimulus remains visible" (p. 194). With this in mind, failure to show visible persistence for longer, foveally presented stimuli might be explained short-latency off-detection responses in the fovea. In sum, the point is that off-stimulus responses might be triggered by nonconscious detection of stimulus termination, rather than by visual persistence.
X X . . . . (vp) . . . .
onset RT offset
Stroboscopic illumination of a moving stimulus
In this approach, stimulus flashes occur in rapid succession across spatially adjacent locations. For example, if a 100 ms flash appears in location 1, and if visual persistence for that flash is 200 ms, then, the subject will see two flashes at location 2—the visual persistence of the first flash and the second flash. In turn, at location 3, the subject will see the visual persistence from the second flash plus the third flash. What the subjec will perceive is a flash moving from left to right. The following illustrates:
1 2 3
The duration of visible persistence is determined by varying the flash rate. For example, if a 50 ms flash rate results in longer visual persistence—say 150 ms--then visual persistence from the first flash will appear not only in location 2, but also in location 3. The perception will thus be simultaneous flashes.
The moving-slit technique
A slit is oscillated across a stimulus. The duration of visible persistence equals the minimum oscillation rate at which the entire stimulus is visible. Results obtained using this technique are consistent with results obtained using other techniques: an inverse relationship between stimulus intensity and visual persistence, and comparable estimates of visual persistence duration.
In a slight variation of this procedure, the phenomenal-continuity technique, Haber and Standing (1969) alternated presentation of a black circle and a blank field. The on-time for the circle was twenty times greater than the off-time, and a cycle was defined as the on-time plus the off-time. Visible persistence was estimated by varying the cycle time. When the cycle time exceeded the duration of visible persistence, subjects perceived blinking; below the duration, they perceived a continuous image. In a subsequent experiment, Haber and Standing found evidence for visual persistence either when the stimulus was presented to one eye, or when presentations alternated between eyes. They interpreted this as evidence for a central rather than peripheral locus of visual persistence by suggesting that, if visual persistence were mediated by a peripheral mechanism, then a monocularly presented stimulus would appear discontinuous unless the stimulus was represented to the same eye at the precise moment that visible persistence terminated. But, as Coltheart explains, this reasoning is fallacious because the results indicate that visible persistence occurs when the stimulus is re-presented to either eye. The results of Haber and Standing can therefore not be interpreted as evidence against peripheral mediation of visual persistence.
Temporal integration of form parts
In yet another approach, elements of a nonsense syllable are divided into two sub-stimuli, such that the stimulus can be identified only when the two parts are superimposed. The interval between the successive presentations indicates the duration of visual persistence. That is, if the interval exceeds the duration of visual persistence, then the subject sees the stimuli successively rather than as superimposed. In a similar technique, elements of a stimulus are presented successively in cells within a matrix. One cell is left blank, and the subjects task is to identify the coordinates of that cell. The task is only possible when the entire display duration—that is, the time to display all cells in succession—is less than VP. Consistent with previous results, results obtained from this technique indicate an inverse relationship between stimulus intensity and VP duration, and shorter VP for longer stimuli.
Summary of visible persistence
How are visible persistence and iconic memory related?
Iconic memory and visible persistence are often treated as the same. However, whether these two phenomenon are the same is an empirical question. The earlier review established that two properties of visible persistence are 1) an inverse relationship with stimulus duration and 2) an inverse relationship with stimulus luminance, or intensity. Iconic memory must exhibit these two properties before it can be equated with visible persistence. There is no evidence that the duration of iconic memory and luminance are inversely related. Furthermore, although there is some evidence that increasing stimulus duration has an effect on iconic memory persistence, this effect is direct rather than inverse. Thus, visible persistence—which is very sensitive to physical display features—and iconic memory seem to behave differently.
The neural basis of iconic memory and visible persistence
Visible persistence and iconic memory behavior differently. Thus, the neural bases of these phenomena must be considered separately. Before considering these issues, Coltheart discusses afterimages. Briefly, Sakitt has distinguished among three types of afterimages: ultraweak, weak, and strong. She equates the ultraweak afterimage with visible persistence, and the weak with iconic memory. Rod saturation is the basis for the different types of afterimages. Ultraweak afterimages require rod saturation, whereas weak afterimages occur at sub rod saturation. Coltheart argues that data on visible persistence are inconsistent with this view because visible persistence can occur for stimuli presented at luminance levels not sufficient to produce rod saturation. Moreover, he points out that increasing stimulus intensity seems to increase afterimages, while decreasing visible persistence duration. Nevertheless, Coltheart acknowledges that afterimages can mediate partial report superiority. But, "circumstances exist in which iconic memory effects . . . can be observed without their being produced by afterimages" (p. 212).
Are iconic memory and visible persistence mediated by rods or cones, or both? Coltheart reports that similar partial report superiority effects are found for stimuli visible to rods, cones, and rods and cones. Visible persistence, by contrast, is stronger for stimuli visible to rods than to cones. Once again, iconic memory and visible persistence are shown to be distinct phenomena. Based on these results, Coltheart concludes that visible persistence is mediated peripherally, whereas iconic memory is a more centrally located phenomenon. That is, "visible persistence is intimately tied to processes going on in the visual pathways from retina to visual cortex, while iconic memory is a property of some much higher stage in the information-processing sequence" (p. 213).
Persistence in the photoreceptors
Persistence of photoreceptor activity may be the neural basis of visible persistence. Consistent with this idea, rods and cones continue to output signals after stimulus offset, but cone activity wanes before rod activity. Recall that visible persistence is somewhat longer for stimuli presented to rods than to cones. There is also evidence that the inverse relationship between stimulus intensity and visible persistence duration is attributable to inhibition of rods by cones at high luminance levels. The inverse relationship between stimulus duration and visible persistence might also be explained by the inhibition of rods by cones. For example, cone inhibition might be more intense for longer stimuli. However, the effects of stimulus duration and intensity on visible persistence are additive, not interactive. That is, the effect of luminance on visible persistence is constant across stimulus duration. Another explanation for the inverse persistence-duration relationship is thus required, and it appears that visible persistence is not entirely mediated at the photoreceptor level. With this in mind, Coltheart states the following: "I will take the view that photoreceptor persistence is responsible for some, but certainly not all, visible persistence effects" (p. 215).
What is the neural basis for the inverse persistence-duration relationship? It is possible that visible persistence in part reflects the activity of cells with a more central locus, for example, cells at the level of the laternal geniculate nucleus or visual cortex. As one possibility, Coltheart proposes that visible persistence is in part mediated by the activity of sustained cells. That visible persistence declines as intensity increases might be explained by greater inhibitory action of transient cells on sustained cells at higher intensity levels.
In summary, "it is inappropriate to ask where the neural locus of visible persistence is: the loci are multiple, as all stages of the visual system from retina to cortex are involved.
What is iconic memory?
Coltheart has thus far specified what iconic memory is not: it is not visible persistence, it is not an afterimage, and it is not neural activity at any particular locus. Moreover, partial report superiority is neither necessary nor sufficient for iconic memory. An afterimage is sufficient to produce partial-report superiority, and iconic memory can investigated using other techniques. What, then, is iconic memory? Coltheart proposes as a sufficiency condition partial report superiority with luminance levels to low to produce an afterimage. Another problem, however, is output interference. That is, is partial report superiority attributable to iconic decay or output interference in full report. To illustrate, with a 3x4 display, performance of 75% in partial report would require output of nine items in full-report. A strong effect of cue delay on partial report, but not on full report is therefore required.
A common assertion is that pre-categorical, featural information is stored in iconic memory. Consistent with this view, partial report declines when cued by spatial location and color, but not when cued by semantic information. But according to Coltheart, this might also be interpreted to suggest that while information about physical attributes decays, information about identity is more durable. For example, using a single stimulus cue technique, Averbach and Coriell (1961) found that while performance declined as cue delay increased, the number of intrusion errors committed did not. Other research indicates that the number of transposition errors also increases, while intrusion errors remain constant. Based on these and other results, Coltheart argues that "The identity of an item is stored rapidly and in a stable form early in the lifetime of a display, while physical attributes of the item are registered with more difficulty and in an unstable decaying form" (p. 222). This view of differential information decline is inconsistent with the notion that an icon fades as a whole.
Based on the above suggestions, Coltheart argues that stimulus presentation activates a relatively permanent entry in an internal lexicon, irrespective of the stimulus’s physical features. For example, if an A is presented, then some prototypical representation of A is activated. Iconic memory consists of attaching physical, featural information to that representation. The attachment is rapid and automatic, but fleeting unless transformed to a more durable form. From this view, the typical full-report limitation is attributable to rapid decay of activation. By contrast, cued items can be stabilized; this stabilization process acts on the basis of physical features. As Coltheart explains, "The lexical monitor can respond to the cue by stabilizing that subset of items specified by the cue . . . " (p. 223). The relative stability of identify information relative to physical information is explained by slower decay rate for lexical activation than for physical information. Coltheart’s model of memory also addresses the backward masking phenomenon. Briefly, he argues that a backward mask interferes with the lexical monitor, which is responsible for lexical stablization.
Coltheart proposes a "postlexical" model of iconic memory in which iconic memory is conceptualized as the process of attaching information about physical features to lexical activation. Physical information fades away quickly, unless it is stabilized by the "lexical monitor." Can the lexical monitor be interpreted as some form of covert attention—an attentional spotlight (see Cowan, 1988, for the parallel)? This view has interesting parallels with activation models of memory in which attention is required to prevent activation decay.