One way of beginning is to set up an experimental situation in which we measure a participant’s verbal reports of some briefly flashed visual stimulus. We flash the stimulus and get the participant to say what they saw. In this context it is normal to refer to the stimulus as being contained in a stimulus display. To make the task simple, each stimulus display comprises an array of printable characters such as individual letters or individual digits (sometimes these are also known as display elements or items) and in some cases the stimulus display may also be referred to as the target display. The critical point is that the char- acters should be highly familiar and easily nameable – if the display were presented for an unlimited amount of time then participants should have no trouble in
naming all of the characters present. Should particip- ants show difficulties in reporting the display elements in our proposed experiment, then we need to be sure that this cannot be due to the fact that the characters are unfamiliar or are hard to name such as ‘¥’. Poor performance in that task might be taken as indicative of a memory limitation when in fact the problem reflects the participants’ inability to name the stimuli – how could they possibly report a stimulus that they cannot name?
In the experiments to be described, the critical unit of design is the experimental trial wherein at least one stimulus is presented and the participant makes at least one response. An experiment typically comprises one or more blocks of trials such that a block is simply a sequential series of trials (hence a block of trials). Two experiments immediately suggest themselves. For instance, we can control the length of time that the dis- play is presented for – this is known as the stimulus
duration – and vary the number of to-be-reported
characters – known as the display set size – across the trials. (Here the display set size refers to the number of characters presented in a given display.) Alternatively, we can hold the display set size constant and vary the stimulus duration across the trials. Of course we can vary both the stimulus duration and the display set size across trials, but let’s keep it simple and consider the case where display size is held constant and the stimulus duration is varied across trials.
In perhaps the simplest experimental set-up, we present a briefly flashed stimulus display to a particip- ant and get them to either write down what they saw or simply tell us what they saw. Such techniques were used for around 80 years prior to the time Neisser (1967) was writing (von Helmotz, 1894, cited by van der Heijden, 2003). On the basis of this sort of research it had been concluded, and generally accepted, that if a display was presented for up to about half a second, then participants could report up to about five items (Coltheart, 1972). So a display might look something like this:
A G J K F D E L V O P Z
and the participant might be able to report D, E, G, V, or A, G, L, or indeed just D. Such a result was initially interpreted as showing that the participant could only perceive these letters in the display (Coltheart, 1972). Despite this conclusion, though, it became apparent (Sperling, 1960) that participants were aware that the
displays contained far more items than they could report. This was a reasonable hint that verbal reports may only be a rather imperfect tool for studying per- ception! Participants reported seeing more than they could actually report.
So how might we provide evidence in support of this claim independently of the anecdotal reports from participants? Can we provide quantitative measures that might help illuminate what is going on? Well, pursuit of such questions sparked the interest in studying iconic memory. One seminal reference here is to the work of Sperling (1960). He presented participants with dis- plays of the type shown above and tested performance in, essentially, two conditions. In both conditions a trial began with the presentation of a central fixation point. The participant pressed a key and then 500 mil- liseconds (ms) later the fixation point was replaced by a stimulus display for only 50 ms which itself was then replaced by the original fixation field. (To be clear, there are 1000 milliseconds in 1 second – so 50 ms is 0.05 of a second, 500 ms is half a second.)
In the full-report condition participants simply had to report all of the letters that they were able to from the stimulus display. Under full-report conditions the average correct report was of the order of 4.5 items for displays containing 5–12 characters. This is roughly what would have been predicted on the basis of the previous work, and Sperling (1960) referred to this as the span of immediate memory. However, a quite sur- prising result obtained in the other condition. In this condition – the partial-report condition – participants were cued to report the letters from only one particu- lar row. Now what happened was that, at the offset of the stimulus display, a tone (the cue) was played. The tone varied in pitch such that a high-pitched tone signalled that the participant should report the top row of letters, a medium-pitched tone signalled that the participant report the middle row and a low-pitched tone signalled that the participant report the bottom row of letters.
Under these conditions participants were now able to report correctly about three of the four possible letters in the cued row. As Coltheart (1980a) argued, this result shows that – despite what might have been concluded on the basis of the full-report data – on average, the same number of items must have been stored in some form of memory from the other two uncued rows. The participants did not know which row to report from in advance of the cue so, unless they retained nearly all of the information in the dis- play, their partial reports of the items could not have been this good. So whereas the full-report data suggest
a very limited store of information from the display, the partial-report data suggest a memory store in which quite a lot of (perhaps all) the original stimulus information is retained.
Pinpoint question 3.2
How do you turn a full-report condition into a partial- report condition?
Partial-report superiority
What we have here are two different estimates of memory capacity: one from the full-report condition, and a dramatically larger one from the partial-report condition. The implication is that the estimate based on full report significantly falls short of what the true storage capacity is. It is not that verbal reports are inaccurate reflections of perception so much, but it does seem that the way in which such verbal reports are solicited is crucial. This conclusion must make us think hard about how best to get participants to report on their perceptions of the world. Nevertheless, the most important finding is that the capacity estimates of the memory store from the partial-report conditions are greater than those from full-report conditions. This is known as a partial-report advantage or partial-report superiority. The partial-report superiority (or partial- report advantage) was taken to show the large capac- ity of iconic memory – perhaps all of the display items are captured in the icon.
What kind of information is stored in the icon?
To understand the nature of the memory system that we have identified, we can begin to ask just what sort of information is stored in the system. Is it simply a copy or mental photograph of the display? This is where the ingenuity of researchers really does begin to show. The basic idea now is to consider other charac- teristics of the stimulus elements that might be stored alongside their location in the display. In fact, we often take for granted that we recover location information as well as item information, otherwise the act of read- ing a textbook would beocme evry dffiultc edined. Given that the tone cues were effective, this revealed that the relative position of the characters had been retained in memory. If the cues had not worked then participants would not have been able to use the tone to recall items from the cued row.
So what other sorts of information are captured by the icon? To address this we need to begin to explore different sorts of manipulations of the display items. For instance, we could vary the colour of the items, their brightness, their shape, their size, and we could even go completely crazy and mix in letters and num- bers together and hence vary the categorical identity of the items in the display. All of the above manipula- tions were dutifully carried out (Banks & Barber, 1977; Clark, 1969; Turvey & Kravetz, 1970; Sperling, 1960; von Wright, 1968, 1970) and the findings revealed that whereas a partial-report advantage was shown for all of the physical characteristics of the items, this was not so when categorical identity differentiated the cued items from the uncued items. Whereas participants could selectively report items of a particular colour, brightness, size or shape, they were unable to report selectively items from different categories when asked. Participants were unable to, for example, just report the letters and ignore the digits (Sperling, 1960). What this was taken to show was that the particular mem- ory system being tapped contained something very much like a copy (i.e., a mental photograph) of the relevant characteristics of the stimulus display. Think of it this way: a colour photograph of next door’s cat contains patches of colour and nothing else. You have to work on this information to decide that, ‘Yes, indeed, the photograph contains an image of the cat.’ To underline the visual nature of this sort of mental representation Neisser (1967) coined the phrase ‘iconic memory’.
Indeed Coltheart, Lea and Thompson (1974) went on to show that participants were unable to use the sounds of the display items as an effective selection criterion because they failed to show a partial-report advantage when asked to report those letters with an
ee sound in. Participants were unable to select out D, B, C, etc. from F, L, M, etc. The claim was that,
whereas iconic memory contained an explicit repres- entation of the visual properties of the items, further mental work was needed to extract their correspond- ing sounds. Neither categorical information (e.g., this item is a number) nor phonological information (e.g., this item contains an ee sound) appeared to be repres- ented in the icon. So iconic memory was said to have a large capacity because the partial reports indicated that nearly all of the items in the display were retained. In addition, it was also claimed that the information contained in it was said to be visual and pre-categorical – the category identity of the characters (i.e., whether a given display element was a letter or number) was not made explicit.
The time span of the icon
As a present I brought one of those Fred Flintstone cam- eras, the kind where the film canister is also the body of the camera, and I presented it to the chief. He seemed delighted and began to click off pictures. He wasn’t advancing the film between shots, but since we were told we shouldn’t speak unless spoken to, I wasn’t able to inform him that he wasn’t going to get twelve pictures, but only one, very, very complicated one.
(Anderson, 1995)
A third characteristic of iconic memory – that it has a very brief time span – was shown via the introduction of another experimental manipulation. Now a delay was introduced between the offset of the stimulus dis- play and the presentation of the tone cue. By varying the delay to the cue over trials it was then possible to measure partial reports over increasingly longer periods. Figure 3.1 shows a schematic representation of the sort of data that was obtained when a delay was
Figure 3.1 The decay of iconic memory
A schematic representation of data taken from Sperling’s (1963) early experiments on iconic memory. Two curves are shown and these relate (i) the delay between the onset of a target display and the cue to respond (on the x axis) with (ii) the amount of information reported from the target display (on the y axis). The different curves show performance according to whether the pre- and post-fields were light or dark, respectively. The life of the icon is clearly extended when dark fields are used. The small vertical pale blue bar that defines the 0 time-point signifies the timing of the stimulus display. The taller vertical dark blue bar indicates immediate-memory span under no mask conditions.
Source: Coltheart, M. (1972). Visual information processing. In
P. C. Dodwell (Ed.), New horizons in psychology 2 (fig. 1, p. 64). Harmondsworth, England: Penguin Books. Reproduced with permission.
introduced between the offset of the display and the presentation of the cue. Full-report performance was unaffected when participants were similarly asked to delay their responses for up to 1 second on each trial – the span of immediate memory was said to be invariant over these delays. Using this technique, estimates of the lifetime of the icon were gauged relative to the point at which the partial-report advantage was abolished. The duration of this very short-term memory system was taken to be the point at which estimates of the num- ber of items available were the same for partial and full report.
To understand Figure 3.1 completely, some fur- ther technical language needs to be introduced. The sequence of events on a trial is taken to comprise a sequence of display fields – for instance, an initial dis- play containing a fixation mark (the fixation field), followed by a stimulus display (the stimulus field or target field) followed by a blank screen (a blank field). Using this terminology, any field presented immedi- ately before the stimulus field is known as a pre-field and any field presented immediately after the stimulus field is known as a post-field. As can be seen from Figure 3.1, when the pre- and post-fields were dark rather than bright (a black screen rather than a white screen) the life of the icon was of an order of seconds. However, when the pre- and post-fields were bright, the life of the icon was shortened to around half a second. This particular pattern of results was taken to under- line the visual nature of the memory representation. As Sperling (1963) noted, ‘the accuracy of partial reports strongly suggests their dependence on a persisting visual image’ (p. 21). Remember, the stimulus display was only presented for 50 ms yet the lifetime of the icon extended well beyond this. So to answer the ques- tion ‘How long does the icon last?’ the data so far dis- cussed suggest that it can range from less than a second to several seconds depending on the visual context of the displays. Performance was critically dependent on whether the stimulus display was more or less intense than its surrounding context.
One interpretation of the effect of presenting a bright post-field was that it reflected a limitation of sensory encoding – somehow the encoding of the stimulus field was being influenced by the luminance of the post-field. If one brief visual event is immedi- ately followed by another, then they may become fused together in the same psychological moment. By this
integration hypothesis (Kahneman, 1968) it is as if
the perceptual system is unable to differentiate between where one briefly flashed event stops and another starts – the two are encoded as one. In this way it is as
if the two events add (or sum) together as one – think of a double exposure on a photograph (see Figure 3.2) where successive images are superimposed on one another as described in the quotation from Anderson (1995). In following this logic, the relative legibility of either event will be tied to the relative amounts of stimulus energy in the two events. So, in simple terms, if the stimulus field is brighter than the surrounding pre- and post-fields then it will be more legible than if it is darker than the surrounding fields. This is exactly what the data show.
A further implication of the results shown in Fig- ure 3.1 is that the representation of the items in iconic memory (a shorthand way of stating this is to say ‘the items that are stored in iconic memory’) will decay and be lost if they are not processed further – for instance, unless they are transformed or recoded into a more durable form then participants will fail to report them (Coltheart, 1972). ‘See ‘What have we learnt?’, page 72.
Figure 3.2 Okay, okay, it’s my turn to do the washing next week
The spooky result of coding two events as one. Old spiritualist photography was created by overlaying one ‘real’ world photograph on one of the ‘spirit’ world. In this case, the image of the apparition is integrated with its background. The mental equivalent is implicated in the integration hypothesis of masking (Kahneman, 1968).
Research focus 3.1