Short-term working memory

1
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

2
Short-term working memory

• The capacity of short-term memory is
  traditionally measured using a memory-span
  procedure
• in this procedure a 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

3
Short-term working memory

• the point at which the participant is able to
  recall all items correctly 50 of the time is
  designated as her/his memory span
• factors affecting memory span
• auditory presentation leads to larger memory span
  estimates than visual presentation
• rhythmic presentation is better than non-rhythmic
  presentation

4
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?

5
Read aloud these digits

• 41 64 00 40 11 49 2

6
Short-term memory

• The next slide contains a series of digits. The
  digits are presented in groups. Read groups of
  digits aloud. Pause between each group. For
  example, suppose the digits were
• 248 917 142 9123
• After you have read the list aloud, I want you to
  write down as many digits as you can remember.
  Any questions?

7
Read aloud these digits

• 416 400 401 1492

8
Short-term working memory

• factors affecting memory span (contd)
• 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

9
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

10
Inducing rapid forgetting
11
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
• why is information forgotten in the
  Brown-Peterson paradigm?

12
Inducing rapid forgetting

• why is 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

13
Inducing rapid forgetting

• why is information forgotten in the
  Brown-Peterson paradigm?
• Petersons argued that it must be trace decay it
  couldnt 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

14
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
  Brown-Peterson 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

15
Inducing rapid forgetting

• Results show that performance declines across
  lists (build up of PI)
• If you change categories, then performance
  increases (release from PI)

16
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

17
Free Recall performance (Craik, 1970)
18
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

19
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

20
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?

21
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

22
Atkinson-Shiffrin model of memory
Rehearsal
Incoming information
Long-term memory
Short-term memory
Transfer
Information displaced
23
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

24
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)

25
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

26
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 selectively damaged
• amnesics have damaged LTM component, but intact
  STM component
• KF (and others) have damaged STM but intact LTM

27
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

28
Working memory model of Baddeley

• Baddeleys 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

29
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

30
Baddeley (1986) contd

• 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)

31
Baddeley (1986)
32
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

33
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

34
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

35
Baddeleys working memory model
Visuo-spatial sketchpad
Phonological loop
Central Executive
36
Working memory

• Phonological loop characteristics
• consists of a phonological store (codes
  speech-based information), and maintains
  information for about 2 seconds
• articulatory control process that refreshes items
  in store by means of subvocal rehearsal

37
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

38
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

39
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

40
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

41
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

42
Episodic buffer of working memory (Baddeleys new
model)

• Overview
• recently Baddeley updated the 3-component model
  of working memory
• 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

43
Episodic buffer of working memory (Baddeleys 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

44
Episodic buffer of working memory (Baddeleys 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

45
Episodic buffer of working memory (Baddeleys 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 words
• Possible accounts
• 1. Sentences are stored in PVs LTM. Implausible
  because PV has normal LTM. Also amnesic px have
  normal memory span

46
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

47
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

48
4-component model of WM (see Fig.1)
Central Exec
visspat
Episodic Buff
Phon.
Episodic LTM
49
Properties of Model

• See previous notes for description of
• Central Executive Function
• Phonological Loop
• Visual spatial sketchpad

50
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

51
A model of the Central ExecutiveSupervisory
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

52
A model of the Central ExecutiveSupervisory
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

53
A model of the Central ExecutiveSupervisory
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

54
A model of the Central ExecutiveSupervisory
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

55
A model of the Central ExecutiveSupervisory
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

56
A model of the Central ExecutiveSupervisory
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)

57
A model of the Central ExecutiveSupervisory
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

58
Supervisory Attentional System
Trigger Data Base
Perceptual Structures
Effector System
Contention Scheduling
59
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

60
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)

61
SAS system

• schemata are assumed to be activated by
  triggering inputs, and to be selected if the
  level of activation exceeds a threshold
• they also tend to be mutually inhibitory
• once a schema is selected, the component schema
  associated with a given schema become activated
  (e.g., component schema for braking, turning on
  lights, windshields etc.)
• the process of routine selection between
  alternative actions is called contention
  scheduling

62
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

63
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 organisms intentions
• it does not directly control behavior, but
  instead modulates the lower level
  contention-scheduling system by activating or
  inhibiting particular schemata

64
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.