+
Chapter 3
Working Memory
+
Atkinson and Shiffrins (1971) Modal
Model
Environmental Input
Sensory Registers
Detect sensory input from the
various modalities
Visual, auditory, haptic, etc.
Memories are held for only a few
hundred milliseconds
Short-Term Store (STS)
Lasts a few seconds and is made
up of:
Temporary working memory
Control processes:
Rehearsal
Coding
Decisions
Retrieval strategies
Long-Term Store
Permanent memory store
+
Problems with the Modal Model
Levels of Processing Evidence
Atkinson and Shiffrins
Modal Model of Memory:
Assumes that holding items in
the short-term store is sufficient
for long-term learning
But Craik and Lockharts
(1972) data show that:
Learning depends on the way
the material is processed,
rather than time in short-term
storage (levels of processing
theory)
Neuropsychological Evidence
The Modal Model
predicts:
That STM is crucial in
transferring information into
and out of LTM.
Any STM deficit should greatly
impair LTM learning, in
addition to reasoning &
comprehension problems.
But a number of patients:
Had severe STM deficits
without apparent working
memory or LTM deficits.
+
Baddeley and Hitch (1974)
Attempted to simulate an STM deficit to see how STM
influenced reasoning, comprehension, and learning
tasks.
The Technique:
Require participants to repeat a sequence of digits out loud while
concurrently performing a variety of cognitive tasks. The digits
should fill up STM and interfere with doing the tasks.
The Test:
Increasing the STM load by adding more digits should cause
more interference in reasoning. However
The Results:
Participants were able to reason without difficulty, even when
repeating sequences of up to eight digits.
Reaction times did slightly increase with digit load but the number
of errors did not increase.
+
Sample Reasoning Task
True
A follows B
BA
B precedes A
AB
B is followed by A
BA
A is preceded by B
AB
A is not preceded by B
AB
B does not follow A
AB
False
X
X
+
STM Digit Load & Reasoning
Speed and accuracy of
grammatical reasoning as a
function of concurrent digit
load.
From Baddeley (1986). Copyright
Oxford University Press. Reproduced
with permission.
Response time increased but
the number of errors did not.
+
Baddeley and Hitch (1974):
Implications
Since
the error rate of the reasoning task
remained constant across digit loads (around 5%),
working memory does not appear to depend
entirely on the same underlying memory system
as digit span.
There does, however, appear to be some interaction between
these systems, as the processing time did increase with increasing
number of digits to be recalled.
To address these difficulties, Baddeley and Hitch (1974)
proposed a model of working memory (WM) -- a system that
supports complex cognitive activities like reasoning instead of
just storage (like STM).
+
Baddeley and Hitchs (1974)
Multi-Component Model
Central executive: an
attentionally-limited
system that selects and
manipulates material in
two slave systems:
Phonological loop: holds
sequences of acoustic or
speech-based items
Visuo-spatial sketchpad:
holds visually and/or spatially
encoded items and arrays
+
The Phonological Loop
Aids Language Learning
The Phonological Loop:
Corresponds to the previous idea of verbal STM (see Chap 2, pg 27-33)
Is the best studied component of WM
Includes a temporary store and a verbal rehearsal process.
What is it good for, besides letting people rehearse (repeat) items
mentally and thereby increase their digit span by two to three items when
repeating back numbers?
Neuropsychological Evidence suggests the loop is
important to language learning because patients with
impairments cannot learn a new language:
Patient PV had acquired a pure phonological loop deficit as an adult,
resulting in a digit span of two items.
She had only relatively minor language comprehension problems with
very long sentences but a profound inability to learn a new language.
+
PV was Unable to Learn a New
Language
Rate of learning pairs of items by patient
PV and controls. Her capacity to learn pairs
of meaningful words was unimpaired
(panel a), but she was not able to learn
foreign language vocabulary (panel b).
From Baddeley, Papagano, and Vallar (1988). Copyright
Elsevier. Reproduced with permission.
PVs performance with
new vocabulary words.
+
The Phonological Loop
What Happens with Normal Adults?
Testing
the language acquisition rationale with
normal participants
Papagno, Valentine, and Baddeley (1991) found that:
Repeating an irrelevant sound disrupted foreign language
learning without affecting learning native language word
pairings this showed the phonological loop was involved.
Papagno and Vallar (1992)
Varied either phonological similarity or the length of paired
words both known to influence the phonological loop.
When the response words were foreign, both factors
impaired performance more than when the response
words were from the native language.
+
The Phonological Loop
What Happens with Language-Impaired Children?
Gathercole
and Baddeley (1990) tested 8-year-old
children with a specific language impairment.
The Task:
The nonword repetition test: Pseudo words of increasing
length are heard and must be repeated (e.g. BALLOP,
WOOGALAMIC, VERSATRATIONAL).
Ability to do this task well correlates with vocabulary
development and is used to help diagnose dyslexia.
Studies reveal that phonological memory is crucial for early
vocabulary acquisition (Gathercole & Baddeley, 1989)
Results:
Language-disordered children were impaired on the nonword
repetition task, performing like 4-6 year old normal children.
+
Children with Specific Language
Impairments
Percent correct repetition of
nonwords by children with
a specific language
impairment (A), children of
the same age (B), and
children matched for
language level but half their
age (C). Adapted from
Gathercole and Baddeley
(1990).
+
Normal Children and Non-Word
Repetition (Table 3.1)
Measures
Correlation
coefficient
Simple
regression
(% variance)
Stepwise
regression
(% variance)
Chronological
age
0.218
5a
5a
Nonverbal
intelligence
0.388
15b
13b
Nonword
repetition
0.525
27b
15b
Sound
mimicry
0.295
9b
Total
0.578
33b
--
p < .05; bp < .01
+
The Phonological Loop and
Action Control
Vocally
or subvocally reminding oneself of the
relevant task often improves performance.
Articulatory suppression limits this ability and disrupts
performance (example alternation of addition & subtraction).
Vygotsky
(1962) and Luria (1959) emphasized
the use of verbal self-instruction to control
behavior.
Studied as a means to help brain-damaged patients
rehabilitate and to understand childhood development.
Largely underappreciated and under-investigated currently.
+
The Visuo-Spatial Sketchpad
Imagine what you had for Breakfast.
How Vivid is Your Visual Imagery?
Peoples
subjective report of the vividness
of their visual imagery is largely unrelated
to their performance on visual tasks.
People
who report more vivid imagery do not
necessarily have better memories
They tend to use vividness as a sign of accuracy,
sometimes mistakenly.
Subjective reports may reflect the way in which we
choose to categorize and describe our subjective
experiences, rather than their content/capacity.
+
The Visuo-Spatial Sketchpad
Image Manipulation
Imagine folding the shapes below to create a solid, with the
shaded area as the base. Will the arrows meet head on?
Shepard and Feng (1972) found that the time it takes to
answer the question depends on the number of folds required
Men are typically better than women on spatial manipulation
tasks, but this appears largely due to differences in strategies,
which when taught to women eliminate the gender bias.
+
Sample Spatial Manipulation Task
form an image of the capital letter
J. Then imagine capital D. Now rotate
the D through 90 degrees to the left
and place it on top of the J. What does
it look like?
First,
An umbrella
Articulatory suppression was used to disrupt the phonological loop and
spatial tapping was used to disrupt the visuo-spatial sketchpad.
+
The Visuo-Spatial Sketchpad
Concurrent Tasks
Pearson et al. (1999) used one of two concurrent tasks
while participants were attempting to spatially manipulate
components to form a novel object (umbrella) and/or
remember the individual components (e.g., J, D).
Articulatory suppression disrupted the capacity to remember the
component shapes, suggesting that the phonological loop held the
names of the components.
Tapping a series of spatial locations disrupted the capacity to create
novel objects but spared the capacity to remember the component
shapes, suggesting that the sketchpad is important for the former but not
the latter.
Concurrent spatial activity is especially disruptive to
individuals who rely heavily on visuo-spatial imagery to
perform specialized tasks (example: visualizing an abacus).
+
The Visuo-Spatial Sketchpad
Concurrent Tasks
Just
as spatial processing can interfere with
imagery, imagery can interfere with spatial
processing.
For example, imagining the sports events being reported on
a car radio can disrupt ones ability to safely navigate the
road ahead.
Visual
imagery can aid memory, but it can
also be disrupted by the presentation of
irrelevant pictures or colors which participants
are instructed to ignore.
+
Disruption of Competing Tasks
Left above: Examples of the material developed by Brooks and used to study the visuospatial sketchpad. Participants must repeat the sentences from memory but can use the
matrix to help them. Data from Brooks (1967). Right: The influence on recall of the
Brooks sentences on a concurrent visuo-spatial tracking task. Data from Baddeley et al.
(1973).
+
Vividness of Subjective Experience
Baddeley and Andrade (2000)
Hypothesis:
Vividness of visual imagery reflects the operation of the sketchpad;
auditory imagery reflects the phonological loop
Task:
Participants asked to form and judge the vividness of visual or
auditory images before being tested under one of the following:
Baseline conditions
Articulatory suppression conditionspredicted to reduce
vividness of auditory imagery
Concurrent spatial tapping conditionspredicted to reduce
vividness of visuo-spatial imagery
+
Working Memory and
Visual Imagery
Baddeley and Andrade (2000)
Results:
When images were novel, auditory imagery was less vivid
under articulatory suppression and visual imagery was less
vivid under spatial tapping, just as predicted.
When images were drawn from long-term memory, however,
there was far less of an effect.
Interpretation:
The loop and sketchpad only limit detail when the image
depends on STM fed by current experience.
When the image is based on LTM, it is much more reliant on
a fourth component of working memory, the episodic buffer.
+
Proposed Structure of the
Visuo-Spatial Sketchpad
Logie (1995) suggests that
the structure of the visuospatial sketchpad is similar to
that of the phonological loop,
in that it consists of:
The visual cache: A passive
store.
The inner scribe: An active
spatial rehearsal process
Logie suggests that the
sketchpad offers a mental
workspace for complex tasks
and (controversially) is
always fed by LTM
+
The Central Executive
The
central executive is thought to be an
attentional controller, with two main modes of
operation (Norman & Shallice, 1986):
A semi-automatic conflict-resolution system, based on
existing habits and requiring little attention.
Conflicts are routinely resolved without much conscious
awareness.
The supervisory attentional system (SAS), based on an
attentionally limited executive.
Crucial to the central executive.
Able to intervene when automatic conflict resolution is not
possible or when a new situation arises.
+The Supervisory Attentional System
(SAS)
When the SAS Fails
Occasionally, the SAS doesnt initiate when it should have.
Patients with frontal lobe damage appear to have problems
with attentional control, suggesting that SAS relies on this
brain region.
They often perseverate (repeatedly perform the same action
or mistake with difficulty stopping or changing behavior).
They often fail to focus attention, resulting in utilization
behaviors (uninhibitedly making use of whatever cues are
afforded by the environment to guide behavior).
They often fail to monitor behavior (fail to make sure they
are behaving appropriately for the situation) or recognize
when their statements dont make sense (confabulation).
+
The Central Executive
Attentional Focus
A
major function of the central executive is to
direct attention to the task at hand, such as
planning the next chess move or remembering
positions.
Robins
et al. (1996) tested expert & novice chess
players similar factors affected performance:
Performance was unaffected by articulatory suppression,
suggesting that the phonological loop wasnt involved.
A concurrent visuo-spatial tapping task impaired performance
somewhat, indicating the involvement of the sketchpad.
Greatest disruption came from an attentionally demanding
random generation task (producing a random stream of
numbers), suggesting involvement of the central executive.
+
The Central Executive
Dividing Attention
Another
major function of the Central
Executive is to divide attention between two
or more tasks.
Concurrent verbal reasoning tasks (e.g. talking on a cell
phone in the car) can impair judgment while driving, even
though driving skill remains intact.
Alzheimers patients have particular difficulties dividing
attention between simultaneous tasks, even when they are
all very simple. A three-way conversation is difficult for them.
Some aspects of switching between more than one task can
be relatively automatic, while others demand attention.
+
Problems with the ThreeComponent Model of WM
The three-component model of WM cannot explain
why memory span can sometimes exceed the
capacity of the subsystems (as much as 15 words
in a sentence).
for words in a sentence
Aspects of LTM (e.g. grammar) help to chunk the items; yet,
we lack an explanation of how LTM and WM interact
for digit span (6 or 7 items)
If the phonological loop can store only about three items,
where are the rest stored? If they are stored in visual STM,
how does this interact with phonological STM?
The model also cannot explain why images based
on LTM do not depend on the sketchpad or
phonological subsystems.
+
The Episodic Buffer
A Solution to the Problems with the WM Model?
Episodic
Buffer -- A newly proposed
fourth component of the WM system.
Originally assumed to be controlled by the central
executive.
A storage system with a capacity of around 4
chunks of information in a multidimensional code.
Multiple dimensions permit links between the
subsystems, as well as with LTM & perception.
Information is retrieved through conscious
awareness consciousness pulls info together.
Allows for the binding of previously unrelated
concepts -- disrupting the executive does not
impair binding, so it may be automatic/passive.
+
The Current Model
An elaboration of the
original three-part model
with a few major changes:
Added two links from
phonological and visuospatial subsystems to LTM
Arrow A: language
acquisition
Arrow B: visual and spatial
mapping
Added the episodic buffer
Accessed through either
subsystem and/or the
central executive
Possibly linked to
emotions
+
Individual Differences in WM
WM Span
A competing approach to WM inspired by Daneman and
Carpenters (1980) study of language comprehension.
Task:
Read a series of sentences and try to recall the last word of
each: It was a cold winter with many violent storms.
Results:
WM Span: People remember the last words of around two to
five sentences.
WM Span for this task predicts: prose comprehension, ability to
follow complex instructions and take notes, performance in
computer programming courses, IQ reasoning measures, fluid
intelligence.
+
Individual Differences in WM
Complex Span
Gathercole et al. developed a diagnostic battery that:
Tests the various WM subsystems and complex span task
(visual/verbal processing tasks involving simultaneously storing and
manipulating information, thereby exercising the central executive).
Can detect and predict learning problems in school-age children.
Indicates the separate components of WM differentially relate to
academic performance:
Phonological STM and the complex span task predict delayed
reading and arithmetic abilities in 7 to 8-year-old children.
Complex span continues to predict math and science scores
through age 14.
Her work led to a program designed to help teachers identify WMimpaired students and modify their teaching accordingly.
Related to ADHD attention deficit hyperactivity disorder.
Alternative Theories of WM Based on
Measuring Individual Differences
Why
do complex span measures predict
problems with learning and differences in
performance?
To study this, researchers break performance down into
tasks tapping various components and look at
relationships among them.
Most analyses result in conclusions that broadly agree with
Baddeley and Hitchs model of WM and include:
An attentionally-based control system, like the central executive.
Subsystems dealing with the simple storage of verbal and visuospatial material.
+
Theories of WM
Cowans Embedded Processes Theory
WM depends on activation that takes place within LTM and
is controlled by attentional processes.
Activation is temporary and decays, unless maintained
through active verbal rehearsal or continued attention.
Activated memory is multidimensional, with some
resemblance to the episodic buffer.
WM capacity is limited to around four chunks (NOT six to
seven, as suggested by Miller, 1956)
Adapted from Cowan (1988).
+
Theories of WM
Engles Inhibitory Control Theory
Turner and Engle (1989) suggest an operation span similar to
the complex span. To measure it, they use the following task.
Task:
Remember a set of words, each of which is followed by arithmetic
operations (e.g. APPLE, 7 + 2 1 = ?; HOUSE, 5 1 + 6 = ?; etc.)
Results:
Correlates highly with the original complex span (sentences).
Individuals with a low operation span are more susceptible to proactive
interference than are those with high operation spans.
An inability to resist interference also predicts ones susceptibility to the
cocktail party effect (being distracted by hearing ones own name in a
stream of noise).
Conclusion:
The ability to inhibit irrelevant material is tied to complex span/WM.
+
Theories of WM
Time-Based Resource Sharing Model
Assumes that complex span reflects the capacity to
prevent the decay of the memory trace through rehearsal.
Resource-Sharing Hypothesis (Barrouillet et al., 2004):
Rehearsal need not be subvocal, but can simply be intermittently
focusing attention on the fading trace.
Evidence: Simple articulatory suppression is not sufficient to diminish
performance on the Peterson task.
Suggests that even complex tasks allow brief gaps in which rehearsal
may occur, unless their pacing is rigidly controlled.
Task-Switching Hypothesis (Towse & Hitch, 1995):
Assumes a trace decay with participants switching between maintaining
the trace and performing the secondary task.
+
Long-Term WM
(Ericsson & Kintsch, 1995)
This
theory of long-term working memory
proposes that:
LTM is utilized to assist in temporary storage.
The ability of mnemonic strategies to vastly increase the
apparent capacity of WM is related to this integration with
LTM. Example: use of running times to encode 80+ digits or
visualizing an abacus to aid calculation.
However,
this theory only addresses this
particular use of LTM, not the entire WM
system.
+
The Neuroscience of WM
Single-Cell Recording
Single-Cell Recording
An electrode is placed in individual brain cells (typically of an awake
monkey) so that the neuronal activity can be recorded under various
experimental circumstances
This technique was pioneered by Nobel laureates Hubel and Weisel
(1979)
Funahashi, Bruce, and Goldman-Rakic (1989):
Discovered cells in the monkeys frontal lobe that were active during
the retention intervals on successful trials of a memory task (but not
on unsuccessful ones)
Suggested that these areas of the frontal lobe were important for
WM
Future work has identified similarly responsive cells outside of the
frontal lobe, indicating a more general WM network
+
Neuroimaging WM
Paulesu, Frith, and Frackowiak (1993)
Used PET to find:
A) An area between the left
parietal and temporal lobes
appeared related to
phonological storage
B) A more frontal region called
Brocas area, known to be
involved in speech production,
was linked to subvocal
rehearsal
Redrawn from Paulesu et al. (1993).
+
Similar Findings for Visuo-Spatial
PET Imaging
Illustration depicting
PET images of the four
areas activated in the
visuo-spatial working
memory study. Based on
Smith et al. (1996).
+
Neuroimaging WM
Smith and Jonides (1997)
Also used PET to directly
compare visual and verbal
WM
Verbal task: Decide whether
a letter was one of four
previously presented
Replicated the two
activations for verbal STM
in the left hemisphere
Visual task: Decide whether
a circle appeared on the
location of one of three
previously presented dots
Found right-hemispheric
areas associated with the
visuo-spatial sketchpad
Smith et al. (1996). Copyright Oxford University Press.
Reproduced with permission.
Further studies
supported the dorsal
(where) stream vs.
ventral (what)
stream distinction
Spatial memory
activates more dorsal
regions of the brain
Object/pattern memory
activates more ventral
areas
Based on Smith et al. (1996).
+
Neuroimaging WM
The N-Back Task
1-back
1
N-back task:
Press a button whenever an item
was seen N-items before
Increasing the lag increases the
difficulty and executive load
The frontal lobe is engaged during
this task and its involvement
increases with executive load
2-back
7
3-back
7
4
Butto
n
Press
