27 Apr 2014

Sperling (1967) ‘Successive approximations to a model for short term memory’, notes

Corry Shores
[Search Blog Here. Index-tags are found on the bottom of the left column.]

[Central Entry Directory]

[Neuro-Science, entry directory]

[The following is summary and quotation except for my bracketed commentary. All underlining and boldface are mine.]

George Sperling

‘Successive approximations to a model for short term memory’

Brief Summary:

By working through possible models for testing visual sensory memory, we learn that a sensory stimulus with information persists in our awareness even after it physically disappears. We read (‘scan’) parts of that image, and store them temporarily by repeating them vocally or subvocally (‘rehearsal’). But the parts that were never scanned disappear from awareness and seem to be lost forever to consciousness. [Whether or not they are registered subconsciously or unconsciousness would be another matter.]

Abstract [quoting]:

Experimental data are considered from a simple task in which an observer looks at letters and then writes them down. Three models are proposed. Model 1 consists of only two components: a visual memory for the letters and a motor translation component to enable copying a visual memory onto paper. Model 1 is inadequate because the visual image is shown not to persist until the time of reproduction. Model 2 corrects this deficiency by incorporating the possibility of subvocal rehearsal of the stimulus letters and an auditory memory for the rehearsal. However, Model 2 cannot account for performance with extremely short duration images because of the limit on the maximum rehearsal rate. The critical improvement in Model 3 is a more detailed specification of scanning, recognition and rehearsal, including a form of memory which is inherent in the process of recognition itself. Model 3 accounts for these data and incidently gives rise to some interesting inferences about the nature of consciousness. [285]


1. Introduction

Sperling proposes some simple memory tasks involving seeing random letters and writing them down from memory. On the basis of these simple methods, more complex ones might be derived.

2. Models

2.1 Model 1

The subject briefly sees a series of letters, and then from memory must immediately transcribe them from what they see in their ‘mind’s eye’. “While the subject is writing, the contents of his visual memory  are decaying, so that when he finally comes to write the fifth or sixth letter his visual memory of the stimulus no longer is legible.” [286]

Sperling. 1967. fig1

The problem with this model is that just as soon as the subject begins writing the letters, the persisting image has already faded away.

2.2 Model 2

In various such experiments, Sperling recorded the voice of the subject, and found that often they speak the letters while writing. If the delay before reporting is extended to about 20 seconds or so, often times the subject will repeat the series and then when time comes to report them, speaks the letters while writing them. This could be a memory mechanism which refreshes the fading image.

Occasionally a subject, when he is writing down letters, can be heard to mumble the letters as he is writing them. His tendency to say the letters aloud can be emphasized by playing loud noise into his ears. Noise itself not seem to alter performance in any other significant way. We have this technique, together with a microphone placed near the subject’s mouth, to record the actual letters the subject is saying. We also recorded automatically whenever the subject was writing. The most interesting results with this technique are obtained when the subject is required to wait (e.g., | for 20 sec) after the stimulus exposure before writing the letters. He repeats (rehearses) the entire letter sequence several times with a pause between each repetition during the interval. Then, at the time of writing each letter, he also may speak it simultaneously.

Rehearsal suggests an obvious memory mechanism. The subject says a letter, hears himself saying it, and then remembers the auditory image. As the auditory image fades, he repeats it to refresh it. Most of our subjects do not vocalize during recall, but they all concur in stating that they rehearse subvocally. Therefore, we assume that the sound-image of a letter lean enter auditory memory directly from subvocal rehearsal without the necessity of actually being converted into sound and passing into the external world. These relations are illustrated in fig. 2.

Sperling. 1967. fig2

According to Model 2, stimulus letters first are retained in visual storage. They are rehearsed, one at a time (i.e., converted from a visual to an auditory form), and then remembered in auditory storage. Subsequently they may be rehearsed again and again as required until they are written down. The limits on performance may arise either from the limited duration visual storage (so that some letters decay before they can be rehearsed) or from the limited capacity of the rehearsal-auditory storage loop, depending on the stimulating conditions. [288]

The problem with this method is the following. [Someone can recollect three letters from a retained image (having been given or lasting in the mind) for 0.1 sec. This means that in order to have pulled those letters out by reading them (vocally or subvocally), they had to do so at a speed of 30 letters a second.]

Attractive as Model 2 seems, it is inadequate for the following reason: it is possible to generate an image in visual storage which has a duration of definitely less than .1 sec and from which 3 letters can be reported. This would require a rehearsal rate of over 30 letters per second, which clearly is completely beyond the capabilities of the rehearsal processes described for Model 2.  [288]

2.3 Short duration visual images

[When testing recollection after many letters, the subject may only store four or five items in their memory. But it is also possible that the image of more of these items was there, but faded before they could report them. The solution to this is partial reporting. The idea is that the subject does not know until after which portion of the whole array they need to report. In

Sperling 1960, for example, there were grid arrays of letters, and subjects were asked after a flash of them to report on just one row. The idea here seems to be that if they can remember 4 or 5 items from one row, which was requested after the image faded, then they must have remembered that many in each of the other rows. For, were any of those other rows requested instead, they would have likewise reported just as many items. That would mean, that if a subject could recall all of one row a second or so after the image disappeared, then probably the whole image remained in sensory memory for that amount of time.] Partial reporting has shown that for a visual image of 18 letters given at 1/20 of a second, up to 10 items remain for as long as 2 seconds after the exposure.

There were also ‘letter-noise’ stimulus sequences, where visual noise was given after the stimulus. Then, by using clicks, the subject subjectively determined the onset of the visual image with its disappearance. This experiment found that

The apparent image duration of the letters in a letter-noise sequence is zero for extremely brief exposures (e.g., less than 10 msec) and then increases linearly with increasing exposure duration for durations exceeding about 20 msec.

The results also showed that each subject had a particular order where they were most correct. In one case it was left-to-right, but more scrambled in others. However, the fact that all positions are reported better than chance indicates that the retentional images are not placed in the mind serially but rather in parallel [and then only afterward ‘read’ in some idiosyncratic order. Please see the first full paragraph on p.290 to be sure this is a correct interpretation.] [290]

Here is quotation for the above summarized parts:

In a letter-noise stimulus sequence, a second, interfering, stimulus (visual 'noise') is exposed immediately on termination of the letter stimulus. The duration of the letter images can be estimated by comparing them to an | auditory signal. Two different methods were used. In the first method two clicks were produced at the ears of the subject. He then adjusted the interval between the clicks until the auditory interval was judged equal to the visual duration. In the second method, the subject heard only one click at a time. We adjusted this click to occur so that it coincided subjectively with the onset of the visual image. After this judgment was complete, he made | another adjustment of the click to coincide with the termination of the visual image. The measured interval between clicks – taken to be the duration of the visual image – was the same by both methods. The apparent image duration of the letters in a letter-noise sequence is zero for extremely brief exposures (e.g., less than 10 msec) and then increases linearly with increasing exposure duration for durations exceeding about 20 msec (fig. 3a).

When stimuli of 5 letters, followed by noise, are exposed for various durations, the accuracy of report increases with exposure duration as shown in fig. 3b. The most interesting aspect of these data is revealed by analyzing separately the accuracy of report at each of the 5 locations (fig. 3c). The accuracy of report at each location reported increases continuously as a function of exposure duration. For this subject, the order of the successive locations which are reported correctly is generally left-to-right (I to V), except that location V is reported correctly at shorter exposures than location IV. Other subjects have different idiosyncratic orders, e.g., I, V, III, II, IV. By definition, in a purely serial process the nth location is not reported better than chance until the exposure duration at which the n-1th location is reported with maximum accuracy is exceeded. The observation that all locations begin to be reported at better than chance levels even at the briefest exposures, may be interpreted as evidence of an essentially parallel process for letter-recognition. This process gives the illusion of being serial because the different locations mature at different rates (cf. GLEZER and NEVSKAIA, 1964; SPERLING 1963). These findings are taken into account in Model 3 (fig 4).



2.4 Model 3

Model three has a scan-rehearsal component like Model 2.
But this part is subdivided into three other components. The first is the scan component. It “determines – within a limited range – the sequence of locations from which information is entered into subsequent components.” [291] [It seems the scan is reading from the persisting sensory image, and then the order of that scan is carried over into the rehearsal.]

The second different sub component is “recognition buffer-memory,” which “ converts the visual image of a letter provided by the scanner into a ‘program of motor-instructions’, and stores these instructions.” [191] And this program is then executed by the rehearsal component. [So the ordering of the scan is then carried into a program for repeating that order.]  What is important is that creating this program can be carried out in relatively short time (about 50 msec for 3 letters, for example) compared to the time it will take to execute that program of rehearsal (500 msec for 3 letters, for example.) [290]

The third new sub-component then is the rehearsal, in which the visual information is now stored in auditory cycles governed by motor-instructions. This then translates back to the visual image when reporting the letters on paper. [290]

3. Consciousness in the Memory Models

We can infer that there is an act of consciousness involved in the scan component of the process. Contents of visual memory which are not scanned fade away. But also, parts that were never scanned we never conscious. This is strange because it means neither subjectively nor objectively can these unscanned elements be observed.

Sperling, G. "Successive approximations to a model for short term memory." Acta psychologica 27 (1967): 285-292.



No comments:

Post a Comment