Brain and memory (LONG)

John Reynolds reynolds at park.bu.edu
Mon Mar 23 17:43:37 EST 1992

 The shortest duration visual memory mechanism appears to be the icon,
which is estimated to last between 100 and 300 msec.  (This estimate
is approximate.  Loftus has shown that duration appears to depend on
stimulus complexity.)  It could be argued that the icon is not really
a mechanism of visual memory because, as Haber (1983) pointed out,
under normal viewing conditions it stores no information because it is
always masked by light coming from the visual environment.  Under the
right conditions it clearly can store visual information, so I will
call it a visual memory mechanism.

 Sperling (1960) originally discovered the icon when he showed that
subjects could briefly retain a great deal of information about a
stochastically presented stimulus array.  Letters arranged in a grid
were followed by a cue either (1) to list the items in one row or (2)
to list items from the whole array.  In the whole array condition
subjects on average recalled only about 38% (4.5/12) of the items from
the whole grid.  When cued to a specific row, they recalled over 82%
(3.3/4) of the items.  The cue was presented after the array
disappeared and in the row condition subjects did better than just
naming the items they happened to remember from that row.  This
suggests that subjects could retain the whole array in a short term
buffer until the cue arrived, and then access the items in the cued
location.  Varying the delay between array onset and cue onset showed
that performance dropped to the whole array level (38%) at delays of
about 250 msec., indicating that some information was available in the
iconic buffer for about a quarter of a second after the onset of the

 Davidson, Fox and Dick have found evidence that the icon is fixed in
a retinotopic coordinate frame.  Subjects fixated on one position and
a row of stimuli was presented.  After the stimulus disappeared,
subjects made a horizontal saccade to a new location and a mask was
presented at the new location.  Subjects' ability to identify the
stimulus that originally fell at the mask position was unimpaired, but
they were unable to identify the stimulus which, if shifted by the
distance between the two fixation points, would fall on the mask
position.  In other words, the mask affected the stimulus that had
appeared at its position in retinal coordinates, not on the display
screen.  This indicates that the icon lives in a retinal coordinate

 Even at this end of the duration spectrum we encounter two
dissociable mechanisms.  Micheal Turvey has shown that two types of
masks can be discerned, pattern and noise masks, and that they appear
to operate at different stages of the visual system.  Noise masks,
such as a blast of bright light, operate only if the energy of the
mask exceeds the energy of the stimulus and do not work if they are
presented to one eye while after the stimulus is presented to the
other eye.  In contrast, pattern masks, which are composed of visual
features similar to those appearing in the stimulus, do not depend
strongly on the relative energy of the mask, are dichoptic, and
operate over longer delays than noise masks.  These facts suggest that
pattern masks are affecting mechanisms located at a more central
location, in a stage of processing which receives binocular input and
after the stage at which stimulus energy is discounted.  It appears,
therefore, that there are at least two independent iconic mechanisms.

 In counting icons we encounter an example of how the operating
characteristics of the visual system could spuriously give rise to the
appearance of multiple systems.  Engaging in a mentally demanding task
such as counting backwards results in a much briefer estimate of
iconic duration using the partial report paradigm.  If we separate out
memory mechanisms according to duration alone we might, on the
strength of this result, conclude that there is a very short duration
icon, which is only activated when we are engaged in a mentally
demanding task!  A much simpler explanation is that processes which
access information from the icon are slowed down by counting

 At durations far in excess of the persistence of the hardiest icon we
find evidence of a short term memory store.  Irwin, for example,
varied the number of elements in two successive arrays which were
separated by a six to eight second delay.  Subjects were able to
perform perfectly when four elements were present in each display,
indicating that they could maintain some visual information in a
memory store for several seconds.  However, as the number of elements
increased to nine, subjects performed at chance, indicating that this
memory store, unlike the icon, is strictly limited in capacity.
Varying the duration of the delay by a few seconds resulted in only
minor decreases of performance, so apparently this form of memory does
not decay like the icon.  In a related experiment, Irwin shifted the
stimuli up to one degree of visual arc during subjects' saccades, with
no decrement in performance.  Since identical images shifted on the
retina and shifted in space could still be perceived to be identical,
this form of memory doesn't reside in a retinal coordinate frame or a
world centered coordinate frame, but rather, appears to reside in an
image-centered or object-centered frame.

 This form of memory can be dissociate from a longer term visual
memory, as shown by Phillips and Christie (1977), who asked subjects
to judge whether a test pattern matched one of a series of one to
eight patterns.  They found that all elements in the series were
recognized at better than chance (60%) even when a long delay
intervened between the last item and the test pattern.  This
recognition was uninhibited when subjects performed mental arithmetic
during the retention interval, and improved if each pattern in the
series was presented for a longer period of time.  In contrast, if the
test pattern matched the last item in the list, it could be recognized
with extremely high accuracy (better than 90%) unless subjects engaged
in mental arithmetic during retention, in which case the score dropped
to 60%.  The 90% rate was unaffected by the presentation duration, but
decayed slowly with a long delay.

 Phillips and Christie concluded from these results that two separate
memory mechanisms were being tapped in this task.  They claimed that
the last item in the list was recognized very well because it was
stored in an accurate short term memory buffer, which was subject to
interference when mental arithmetic was carried out.  The 60%
performance on the other items is assumed to be due to the storage of
the other items in a larger capacity, lower grade, long term memory
mechanism which is unaffected by mental arithmetic.

 It is not completely clear that Phillips and Christie's short term
buffer can be identified with Irwin's mechanism.  Irwin did not, for
example, test whether his subjects' performance would drop if they
engaged in mental arithmetic or if patterns intervened between the
comparison stimuli.  In contrast to Irwin's finding that arrays with
nine elements could not be stored in memory, Phillips and Christie's
subjects appear to have stored patterns that could differ in any of up
to sixteen positions.  However, Phillips and Christie used
considerably simpler arrays, so the stimuli are not directly
comparable.  Since I see no way to dissociate the mechanisms suggested
by Irwin and Phillips & Christie, and the time scales of memory decay
are comparable, I will assume that their data were driven by the same

 Walker and Marshall have shown that under the right conditions short
term visual memory is not limited to the last item that was presented.
Their experiment was a modification of Rabbit and Vyas' procedure to
test the ability of an item in short term visual memory to enhance
responding if the identical stimulus is presented again.  Rabbit and
Vyas asked subjects to judge whether successive digits were even or
not.  If two visually identical items appeared one after the other,
reaction time was enhanced.  This enhancement disappeared if a third
item intervened between them, suggesting that the item had been wiped
out of short term memory by the presentation of the intervening item.

 Using a similar procedure, Walker and Marshall found that the effect
was unaffected by the passage of times up to 2 seconds, which is
consistent with the interpretation that the item was being stored in
the same short term memory mechanism investigated by Phillips and
Christie.  They then had subjects perform two tasks intermittently,
both of which required them to distinguish between two sets of
letters.  For example, in one task, subjects had to hit a button if
the letter A, R, a or r appeared, but not if N, E, n or e appeared.
The other task was similar but used a different set of letters.
Letters from task one were interdigitated with letters from task two
throughout each trial.  Surprisingly, this enhancement was unaffected
by the fact that a letter from one task always intervened between
letters from the other task.

 At first it might appear that this implies that two independent short
term memory systems are available, making it possible to keep a letter
from one task in memory without allowing the stimuli of the other task
to replace the item in memory.  Walker and Marshall suggest an
alternative hypothesis, which does not require us to posit the
existence of a different short term memory system for each task.  They
believe that response speed is enhanced if the stimulus is expected by
a visual matching process, and that normally, the matching process
expects the next letter to be the same as the last one.  When the two
tasks are interdigitated, this default expectation is replaced by the
expectation that the last letter from the task will appear.

 The long term visual memory Phillips and Christie explored in humans
appears to be related to mechanisms discovered by Miller, Li, and
Desimone (1991), who measured cells in the inferior temporal cortex of
macaque monkeys trained to perform a task that required the retention
of a visual stimulus in memory while successive patterns were
presented.  Monkeys were presented with a sample stimulus followed by
a series of one to eight test patterns and rewarded if they released a
lever when a test stimulus matched the sample stimulus.  The tasks are
not strictly comparable because Phillips and Christie's human subjects
had to maintain a series of patterns in memory while waiting for the
test pattern, whereas the monkeys had to retain the sample pattern
until the right test pattern appeared However, the tasks both required
subjects to simultaneously store and attend 

More information about the Neur-sci mailing list

Send comments to us at biosci-help [At] net.bio.net