Transcript working memory

Short-term working memory
• Students of memory (e.g., James, Galton)
have long considered that there is a memory
system that keeps in consciousness a small
number of ideas
• William James referred to this system as
primary memory
• the primary memory is probably more closely
related to working memory than to STM; this
model will be discussed later on today
Short-term working memory
• Capacity of short-term memory is traditionally
measured using a memory-span procedure
• Procedure.
– participant is presented a sequence of items, and
is required to repeat them back
– start with one item, increasing the number of
items by 1 until the participant begins to make
mistakes
Short-term working memory
• Scoring.
– point at which the participant is able to recall all
items correctly 50% of the time is designated as
her/his memory span
Short-term working memory
• Factors affecting memory span
– auditory presentation leads to larger memory
span estimates than visual presentation
– rhythmic presentation is better than non-rhythmic
presentation
Short-term memory
– The next slide contains a series of digits. The
digits are presented in pairs. Read the pairs of
digits rhythmically aloud. Pause between each
pair. For example, suppose the digits were
24 89 17 14 29 12 3
– After you have read the pairs aloud, I want you to
write down as many digits as you can remember.
Any questions?
Short-term memory
– Write down as many digits as you can remember.
Read aloud these digits
• 41 64 00 40 11 49 2
Short-term memory
– Write down as many digits as you can remember.
Read aloud these digits
• 416 400 401 1492
Short-term memory
– Write down as many digits as you can remember.
Short-term working memory
• factors affecting memory span (cont’d)
– recoding or chunking information; George Miller
showed in his classic paper (1956) that memory
span is determined by the number of ‘chunks’ or
integrated items you need to recall, not the
number of items presented
Inducing rapid forgetting
• Brown-Peterson paradigm
– Brown (1958) and Peterson & Peterson (1959)
showed that it is possible to induce very rapid
forgetting if you distract person
– paradigm
study: present a small number of items followed
by a number such as 632. Participant is required to
count backward by threes until given a recall
signal. Then he/she attempts to recall studied
items
Inducing rapid forgetting
Percent correct
recall
Peterson & Peterson (1958) Recall of three
consonants
100
80
60
40
20
0
0
3
6
9
12
Retention interval (sec)
15
18
Inducing rapid forgetting
– Note: Murdock (1961) showed that performance
is about the same for 3 consonants as it is for 3
words, illustrating the importance of chunking
Inducing rapid forgetting
• Potential accounts for why information
forgotten in the Brown-Peterson paradigm?
– trace decay: automatic fading of memory
– interference: memory is disrupted by other
memory traces
proactive interference: effects of prior items on
recall of subsequent items
retroactive interference: effects of subsequent
items on recall of previous items
Inducing rapid forgetting
• why is information forgotten in the BrownPeterson paradigm?
– Petersons argued that it must be trace decay; it
couldn’t be retroactive interference because
numbers are very different from consonants
– Keppel & Underwood (1962) showed that
proactive interference seemed to be responsible
because if performance on the first trial only is
examined there is little decline in performance
over the retention interval
Inducing rapid forgetting
• Further evidence for the importance of
proactive interference (PI)
– release from PI
– numerous studies have established that if you
present several lists of items using a BrownPeterson procedure (Study: present list of 3
items; count backwards by 3s for 15 sec, then
attempt recall of the studied items. Results show
that performance declines across lists
Inducing rapid forgetting
• Results show that performance declines
across lists (build up of PI)
• If you change categories, then performance
increases (release from PI)
One or two memory systems?
• The theoretical question underlying much of
this research had to do with whether there
was evidence for the STM/LTM distinction
• One approach to investigating this question
involves determining whether certain tasks
have separable components
• One task is free recall
Percentage correct recall
Free Recall performance
(Craik, 1970)
100
80
60
Immediate
Delayed
40
20
0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Serial position
Interpretation of free recall study
 Primacy and intermediate components of the
serial position curve are lower in the delayed
compared to immediate condition; recency
portion of the curve is differentially lower in
the delayed condition
 interpretation: delayed condition has a
stronger influence on recency portion of
curve because recency reflects STM
performance
Neuropsychological Evidence for
separation of STM and LTM
 Data from amnesics support the viability of
the distinction between STM and LTM
because amnesics have normal digit span,
which is mediated by STM, but are impaired
in their ability to acquire and retain LTM
memories

Neuropsychological Evidence for
separation of STM and LTM
 Free recall data in amnesics also supports
this distinction. Given your understanding of
free recall I want you to predict performance
of amnesics (Baddeley & Warrington, 1970)
 In immediate free recall, how should amnesics
perform on the recency portion of the curve?
 What about the primacy portion of the curve?
Short-term working memory
• Atkinson-Shiffrin model of memory (1968)
– distinguishes between two types of memory:
short-term and long-term memory
– short-term memory (STM): a temporary storage
system capable of holding a small amount of
information (e.g., telephone number)
– information in STM is forgotten quickly unless it is
rehearsed or transferred into LTM
– Long-term memory (LTM): a permanent memory
store with no capacity limitations
Atkinson-Shiffrin Model
(Atkinson & Shiffrin, 1968)
Problems with modal model
• Modal model assumes that STS plays a
critical role in the transfer of information into
LTS
– Specifically, this model suggests that the capacity
of the STS should determine the probability that
an item enters LTS and
– The amount of exposure in STS should affect the
likelihood that an item enters into LTS
Problems with modal model
• Both these implications are incorrect
– several studies have shown that under some
conditions the number of times material is
rehearsed is a poor predictor that it will be
recalled subsequently (shallow rehearsal)
Problems with modal model
– Shallice and Warrington (1970) and others have
established that at least some people with poor
memory span (this suggests that STS is
damaged) have normal long-term memory
KF memory span WAIS score = 2, Mean = 10,
Standard deviation = 3
established that KF understood spoken words by
presenting a list of spoken words; task was to tap
table when words were from a given category
KF also was impaired when RN STM test
administered
Summary
• Evidence supporting STM vs LTM distinction
– tasks such as free recall seem to have both STM
and LTM components
– Neuropsychological evidence suggests that both
components can be damaged
amnesics have damaged LTM component, but
intact STM component
KF (and others) have damaged STM but intact LTM
Summary
• However, the modal model (Atkinson-Shiffrin)
does have problems accounting for
– the finding that patients with STM deficits appear
to have intact LTM
– maintaining an item in STM does not ensure its
transfer to LTM
Working memory model of
Baddeley
• Baddeley’s early work focused on testing the
hypothesis that STS is important because it
acts as a working memory, a system that is
important for holding and manipulating
information, and it is needed for a broad
range of cognitive tasks
Working memory model of
Baddeley
• Experimental paradigm (dual task paradigm)
– primary task: grammatical reasoning
Determine whether sentences are true/false
e.g., A follows B -- BA (true)
e.g., B is not preceded by A - AB (false)
– secondary task: concurrent digit task: remember
number sequences ranging in length from 0 to 8
Baddeley (1986) cont’d
• Results
– as shown in the accompanying figure, reasoning
time increased with concurrent digit load.
However, performance remained high, and errors
remained low (about 4% and did not vary with
digit load)
– thus, overall performance remains quite good,
even when the overall digit load is 8 (memory
span capacity)
Baddeley (1986)
Speed of reasoning by concurrent digit load
Reasoning time
(seconds)
3.1
2.9
2.7
Reasoning time
2.5
2.3
2.1
0
2
4
6
Concurrent digit load
8
Other important results
• Baddeley, Lewis, Eldridge, & Thomson
(1984) showed that:
a concurrent digit span task had a strong effect on
encoding and remembering new material
however, it had no effect on accuracy of
performance when the concurrent digit span task
was performed during retrieval (although retrieval
latency was slowed)
this suggests that the system responsible for
holding digits does not play a critical role in
retrieval as suggested by previous models of
memory
Conclusions
• These findings and others are difficult to
reconcile with a model in which overloading
the short-term store leads to a complete
breakdown of performance on the primary
task
Working memory model of
Baddeley
– Baddeley proposed to account for these results
by postulating that the digit span limitations are
set by one system, leaving other components of
working memory relatively unimpaired
– Basic model of working memory consists of a
controlling attentional system (called the central
executive) and two slave systems, an articulatory
or phonological loop system and a visuo-spatial
sketch pad
Baddeley’s working memory
model
Visuo-spatial
sketchpad
Phonological
loop
Central Executive
Working memory
• Phonological loop characteristics
– consists of a phonological store (codes speechbased information), and maintains information for
about 2 seconds
– articulatory control process that refreshes items
in store by means of subvocal rehearsal
Working memory
• Phonological loop
– appears to play an important role in reading
poor readers tend to have poor short-term memory
span
– also appears to play a role in the comprehension
of language and in the acquisition of vocabulary
Visuo-spatial sketchpad
• Information can enter the sketchpad visually
or through the generation of a visual image
• access to this store by visual information is
obligatory
• the information in this store may be visual or
spatial or both
Central Executive
• The central executive plays an important role
in controlling attention. Our discussion of the
central executive will begin with a discussion
of the interplay of attention and memory
Brain and working memory
– Psychological theory and data suggest that there
is an important distinction between maintaining
information in consciousness (James, Galton,
Miller idea) and manipulating information
(Baddeley idea)
Brain and working memory
– Neuroimaging data suggest dorsal regions of
prefrontal cortex (PFC) necessary manipulation
in addition to maintenance, whereas ventral PFC
needed for maintenance
– Conclusion different cognitive processes are
mediated by different brain regions, consistent
with psych data
–
Brain and working memory
– Different pattern of findings has been obtained
from neural brain region recordings of monkeys
– This research showed that dorsal regions of PFC
were activated for spatial memory, whereas
ventral regions of PFC were specialized for
object working memory
– Conclusion: this finding suggests that brain
regions are organized on the basis of the content
of information being processed
– Issue is not yet resolved
Central Executive
• Vigilance
– recall vigilance refers to sustained attention
Parasuraman (1979) showed that vigilance
performance decreases if the vigilance task has a
short-term memory component involving storage
and manipulation of information. For example, if
the participant has to detect three consecutive odd
numbers from a stream of digits or must judge
whether adjacent items are of the same hue,
performance declines
Central Executive
• Vigilance
however, if the participant must evaluate each item
on its own (e.g., detect whether a product such as
a frying pan) has flaws, then performance tends to
remain stable
• Dual task performance
as discussed in a prior lecture, it is difficult to
perform two tasks at the same time. However, the
degree of difficulty depends upon the tasks being
performed and the expertise of the person
A model of the Central Executive
Supervisory Attentional System
SAS
• Norman and Shallice developed a model of
the control of action called the Supervisory
Attentional System
– this model was developed by considering our
knowledge of action slips and frontal lobe
function
A model of the Central Executive
Supervisory Attentional System
SAS
• Action slips
– probably all of us have had the experience of
performing some unintended action
e.g., driving home from York in your car and
forgetting to make a detour to pick up your clothes
from the dry cleaners
e.g., William James… going upstairs and ending
up in bed
Reason (1979) has studied action slips and
showed that these errors tend to occur when you
are pre-occupied with some other thought
A model of the Central Executive
Supervisory Attentional System
SAS
• Action slips are actions that are inappropriate
for the goals of the participant. However, the
actions themselves are meaningful, and
reasonably well performed
– my driving is safe, I obey traffic rules etc.
• This suggests that some actions, once they
are initiated, can be accurately performed
with little conscious attention being paid to
them
A model of the Central Executive
Supervisory Attentional System
SAS
– Other actions and other types of behaviour seem
to require a central system and performance
declines if such a system is not in place
research with damaged frontal lobe patients and
monkeys suggests that performance is impaired if
it requires
coordination of different elements of a complex
activity
focused attention
focusing on the whole of a task
working on new situations
A model of the Central Executive
Supervisory Attentional System
SAS
– It is well established that patients with frontal lobe
damage may have relatively intact performance
on IQ tests
– Luria (1966) proposed that the frontal lobes are
involved in programming, regulation, and
verification of activity
A model of the Central Executive
Supervisory Attentional System
SAS
– Sample problem given to pt with frontal damage
There were 18 books on two shelves, and there
were twice as many books on one shelf than on
the other. How many books were on each shelf?
Pt. Response
Step 1. 18/2 = 9 (Clause 1)
Step 2. 18 x 2 = 36 (Clause 2)
A model of the Central Executive
Supervisory Attentional System
SAS
• Sample problem 2
– Bat and ball cost $1.20
– Bat costs $1.00 more than ball
– How much does the bat cost?
A model of the Central Executive
Supervisory Attentional System
SAS
– For problems such as these Shallice, Norman,
and others have proposed that a central
executive is needed
– their model is presented in the next slide
Supervisory
Attentional
System
Perceptual
Structures
Trigger
Data
Base
Effector
System
Contention
Scheduling
SAS system
• According to this model the control of actions
is influenced by a variety of different inputs
– Perceptual inputs
– These can be modulated by a relatively passive
stored routine knowledge, dubbed contention
scheduling
– A more flexible system called the supervisory
attentional system (SAS)
SAS system
• According to this system routine actions run
off relatively automatically
– perceptual information comes into the system
and it makes contact with stored information and
that information triggers certain responses.
These responses eventually result in actions that
are produced by the effector system
– e.g., walking on a country road
SAS system
• At any given moment this model postulates
that our behaviour is controlled by schemata,
that control lower-level programs
– for example the schema that controls our driving
requires visual spatial and motor control systems,
and may call particular component schema in
well-defined circumstances (e.g., if light turns
orange, and you are well away from the
intersection, start braking)
SAS system
– Schemata activated by triggering inputs; selected
if the level of activation exceeds a threshold
– they also tend to be mutually inhibitory; i.e.,
activation of one schemata will inhibit another
– once a schema is selected, the component
schema associated with a given schema become
activated (e.g., component schema for braking)
– routine selection between alternative actions is
called contention scheduling
SAS system
– the process of routine selection between
alternative actions is called contention
scheduling; see Figure
e.g., light is orange and you are close to
intersection, do you brake, accelerate, or maintain
speed and continue through intersection
SAS system
– in addition, this model assumes that there is an
additional system, the supervisory attentional
system
this system has access to the environment and to
the organism’s intentions
it does not directly control behavior, but instead
modulates the lower level contention-scheduling
system by activating or inhibiting particular
schemata
SAS system
– the supervisory attentional system is involved in
initiating willed actions, and in working in
situations in which routine actions are not
satisfactory--e.g., dealing with novelty,
overcoming temptation, etc.
Episodic buffer of working
memory (Baddeley’s new model)
• Overview
– Baddeley’s 3-component model of working
memory has been updated by Baddeley
– It proposes a 4th component, an episodic buffer
It has limited capacity
Stores information in a multimodal code
Binds information from subsidiary perceptual
systems and LTM into episodic memory
Information is consciously retrieved
Episodic buffer of working
memory (Baddeley’s new model)
• Background
– 3 component model of working memory consists
of central executive and two slave systems, the
phonological loop and the visuo-spatial
sketchpad
– Central executive is an attention controller
– Phonological loop stores speech-based info
– Visuospatial sketchpad stores visual info
Episodic buffer of working
memory (Baddeley’s new model)
• Problems with 3-component model of WM
– Articulatory suppression
Saying ‘the’ repetitively (occupying the
phonological loop) does not have a devastating
effect on recall of visually presented numbers
Recall drops from 7 to 5 digits
One might expect recall to drop dramatically
because Phonological loop is occupied and VSS is
not very good at storing this type of information
Episodic buffer of working
memory (Baddeley’s new model)
• Problems with 3-component model of WM
– Prose recall of a patient (PV) with word-span of 1
word is 5 words. This is less than the span of 15
words, but much more than 1 word
– Possible accounts
1. Sentences are stored in PV’s LTM. Implausible
because PV has normal LTM. Also amnesic px
have normal memory span
Episodic buffer of working
memory (Baddeley’s new model)
• Possible accounts
– 2. PV stores information in phonological loop
Implausible because her capacity is 1 word
Episodic buffer of working
memory (Baddeley’s new model)
• Possible accounts
– 3. Already existing information in LTM is
activated
Implausible because such a system could not
store novel information

Episodic buffer of working
memory
• Possible accounts
– 4. information is stored in an episodic memory buffer
separate from LTM
 Accounts for this result
 Also accounts for finding that amnesics can retain relatively
large amounts of complex information briefly (e.g., sentence
span, info about a bridge game)
 People integrate information across modalities (note: may be
two types of integration; automatic and controlled; episodic
integration is controlled integration); see binding problem
discussion

Episodic buffer of working
memory
• Binding problem
– Information that is processed independently by separate
cognitive processes must be bound together because our
experience of the world (and our memory of it as well) is
coherent
– People can also retrieve information about an episode
when give part of an episode (e.g., given a spatial cue,
state what object was stored there)
– Episodic buffer is one way in which the binding problem
can be solved
4-component model of WM (see
Fig.1)
Central
Exec
visspat
Episodic Buff
Episodic LTM
Phon.
Properties of Model
•
See previous notes for description of
– Central Executive Function
– Phonological Loop
– Visual spatial sketchpad
Properties of Model
•
Episodic buffer
– Integrates information across modalities and
from different sources
– Integrates information across time
– Has limited capacity
– Is capable of manipulating information
– Is consciously accessible from Central
Executive