Short Term/Working Memory

1
Short Term/Working Memory

• Langston, PSY 4040
• Cognitive Psychology
• Notes 4

2
What do these have in common?

• You can still remember details of your tenth
  birthday party (which you dont need), but you
  have trouble remembering a definition long enough
  to write it down.
• Pizza I You look up the number of a pizza
  delivery place and someone asks you a question
  before you can make the call. When you go to
  dial, the number is gone.
• Youre trying to get the lunch order straight.
  Three people tell you what they dont want on
  their hamburger but you can only remember part of
  the information.
• Pizza II. Why cant you remember a number and
  talk to someone, but you can remember a number
  while you look around the room?

3
What do these have in common?

• Short-term memory.
• Two kinds of memory, short and long.
• The duration is short.
• The capacity is small.
• There are different resources available for
  different tasks.

4
Architecture

• Recall our box model

Sensory Store
LTM
STM
Filter
Pattern Recognition
Selection
Input (Environment)
Response
5
Short-Term Memory

• A brief memory store with a limited capacity that
  helps you to hold information as you process it.

6
Themes

• One STM or many?
• Is STM really different from Long-Term Memory?
  (Later in the class, but the foundation will come
  tonight.)

7
Two Kinds of Memory

• The phenomenological evidence is very strong.
  Everyone has experienced the phenomenon of having
  some memories that dont last long and some that
  do. What is the evidence for two kinds of memory
  store?

8
Two Kinds of Memory

• Evidence
• The serial position curve.
• The task I present you with a list and you
  recall it. You can recall the words in any order
  and try to recall as many as you can (called a
  free recall task).
• We graph the frequency of recall by serial
  position in the list (first word, second word,
  etc.).
• Looking at that curve can tell us something about
  memory stores.

9
Two Kinds of Memory

• Try the free recall task here

10
Two Kinds of Memory

• Glanzer and Cunitz (1966) tested a two-store
  explanation of the curve
• There are two parts to the curve. The first part
  is called primacy (its the earlier words) and
  the last part is called recency (its the most
  recent words).
• People start by writing down the last words they
  heard. Recency is high because people just dump
  out the contents of STM.

11
Two Kinds of Memory

• Glanzer and Cunitz contd.
• When people go back to words they have to try to
  remember, they produce the recalls that will go
  into the primacy part. This part is coming from
  LTM.
• In other words, even though it looks like one
  curve, it actually reflects two kinds of memory.

12
Two Kinds of Memory

• Glanzer and Cunitz contd.
• We can test this by thinking of variables that
  should affect each kind of memory differently.
• What should affect recency (STM) but not primacy
  (LTM)?
• Whether or not people can recall right away. If
  STM doesnt last long then having to wait will
  allow it to go away and there wont be anything
  for recency. Since you wait for primacy anyway,
  it wont matter.
• GC Make people count backwards before they get
  to recall.

13
Two Kinds of Memory

• Glanzer and Cunitz contd.
• We can see the effect of counting backwards in
  the first graph (see the overhead).

14
Two Kinds of Memory

• Glanzer and Cunitz contd.
• We can test the two store explanation of the
  curve by thinking of variables that should affect
  each kind of memory differently.
• What should affect primacy (LTM) but not recency
  (STM)?
• How much time people have between each item.
  With more time, theres more time to rehearse,
  and more stuff should get into LTM. Since
  recency isnt based on how much you rehearse, it
  shouldnt be affected.
• GC Space out the words in the list.

15
Two Kinds of Memory

• Glanzer and Cunitz contd.
• We can see the effect of spacing on the second
  graph (see the overhead).

16
Two Kinds of Memory

• Research like Glanzer and Cunitz (1966) also
  represents an important tool in cognitive
  psychology called the double dissociation.
• The basic idea is that if different parts of a
  task use different processes, then different
  variables will affect those parts differently.

17
Two Kinds of Memory

• This makes it more clear

18
Two Kinds of Memory

• Also neuropsychological evidence
• HM Damage to hippocampus when having corpus
  collosum severed. Can remember old stuff, but
  cannot acquire new memories. Appears to have STM
  deficit and transfer deficit (anterograde
  amnesia).
• Also people with retrograde amnesia who can learn
  new things but forget parts of their past.
• Suggests another type of double dissociation.

19
Connection

• We can address our first question
• You can still remember details of your tenth
  birthday party (which you dont need), but you
  have trouble remembering a definition long enough
  to write it down.
• Why?

20
Properties of STM

• Now that we have decided that STM and LTM are
  separate, what are the properties of STM?
• Duration.
• Capacity.
• Mechanism of forgetting.
• Representation (code).
• Search.

21
Properties of STM

• Duration
• Peterson and Peterson (1959) had people learn a
  list of three letters, count backwards, and
  recall it.
• The counting could go from 0 to 18 seconds.
• What they found was that after 18 seconds, STM
  recall was virtually zero.
• This was the inspiration for Glanzer and Cunitzs
  (1966) counting backwards task.
• Look at the data graph

22
Properties of STM

• Duration

Peterson Peterson (1959, p. 195)
23
Connection

• Duration
• We can answer Pizza I
• Pizza I You look up the number of a pizza
  delivery place and someone asks you a question
  before you can make the call. When you go to
  dial, the number is gone.
• Why?

24
Properties of STM

• Capacity
• Measured using span tasks.
• I present you a list of information (e.g., s, r,
  d, g, n, v, p), and you repeat it back.
• We make the lists longer until you cant do it.

25
Properties of STM

• Capacity
• Miller (1956) noted that over a variety of span
  tasks (letters, digits, words, binary numbers)
  people came out with a capacity of 7 plus or
  minus 2. Thats the capacity (since the task is
  clearly using STM).

26
Properties of STM
Miller (1990, p. 349 from Hayes, 1952)
27
Properties of STM

• Capacity
• As Miller (1990) puts it Absolute judgment is
  limited by the amount of information. Immediate
  memory is limited by the number of items. (p.
  349)
• We did a span task in CogLab, we can look at the
  results

28
Properties of STM

• Capacity
• Note that a process called chunking messes up our
  measure of span.
• A chunk is an integrated unit of information.
  You could remember seven digits, or you could
  call it my phone number and then its just one
  thing.
• The capacity is really 7 plus or minus 2 chunks.

29
Properties of STM

• Capacity
• Consider this from Miller (1990)

Miller (1990, p. 350)
30
Properties of STM

• Capacity
• Chunking is like putting it in your own words.
  We recode our experience into verbal descriptions
  all the time.

31
Properties of STM

• Capacity
• To the extent that you have LTM knowledge to use
  to make chunks, you can have an incredible span.
• Chase and Simon (1973) found that chess masters
  could remember more than 7 plus or minus 2 pieces
  on a board. But, they only did about 8 chunks.
  They had 10,000 to 100,000 chunks memorized.
• When the board was arranged at random, they
  werent nearly as good.
• Learning curves for master, class A player, and
  beginner

32
Properties of STM
Chase Simon (1973, p. 61)
33
Properties of STM

• Capacity
• Lets do a chunking example
• Remember
• A A M L J Y K V C D S F R T E

34
Properties of STM

• Capacity
• Now try
• A A M L J Y K V C D S F R T E

35
Properties of STM

• Capacity
• Now try
• YMCA JFK TV LSD ERA
• You can go way beyond your capacity with
  chunking.

36
Connection

• Capacity
• We can answer our third question
• Youre trying to get the lunch order straight.
  Three people tell you what they dont want on
  their hamburger but you can only remember part of
  the information.
• Why?
• How could you do better?

37
Properties of STM

• Mechanism of forgetting
• Decay The passage of time causes it to fade
  out. (Analogous to rusting.) But, theres a
  mechanism for rusting, shouldnt there be a
  mechanism for forgetting?
• Interference New stuff coming in makes it hard
  to keep what you have.

38
Properties of STM

• Mechanism of forgetting
• Waugh and Norman (1965) manipulated two things
• Rate How fast the material was presented.
• Number of intervening items How much material
  came between the critical item and the chance to
  recall.
• Comparing decay and forgetting
• If its decay, more time equals more loss. So,
  slower vs. faster should have a big effect.
• If its interference, more material equals more
  loss, so amount should be the big variable.
• They found that number of interfering items was
  the important variable.

39
Properties of STM

• Mechanism of forgetting
• Interference. Two kinds
• Retroactive What weve been discussing. Trying
  to put in new stuff messes up existing stuff.
• Proactive All of the old stuff you know is
  making it hard to fit in new stuff.
• Learning a new language is an example of this.
  Trying to learn by working through your old
  language makes it very difficult. Your existing
  language interferes.
• We can try a proactive interference demonstration

40
Properties of STM

• Mechanism of forgetting
• Some important points about proactive
  interference.
• It makes it hard to learn too much of the same
  type of stuff at the same time. This has
  implications for cramming.
• Release from PI shows the benefit of mixing up
  the materials that youre studying.
• Keppel and Underwood (1962) showed that Peterson
  and Petersons (1959) results were mostly
  proactive interference. If people learn just one
  list, count, and recall, you dont get the
  forgetting. (Note that it is still interference.)

41
Properties of STM

• Code
• What is the format of the information?
• Conrad (1964) had lists of letters that were
  auditorially confusable (BCPTV and FMNSX).
  Peoples memory confusions with these lists
  showed that the letters were much more likely to
  be confused based on sound than on appearance.
• Wickelgren (1965) would present span tasks like
  4NF9G27Z. When people recalled, their mistakes
  were based on sound.
• So, auditory code.

42
Properties of STM

• Code
• Posner and Keele (1967) presented pairs of
  letters like A-a or A-A. Participants made a
  same-different judgment.
• If the letters were less than 1.5 seconds apart,
  the appearance mattered (A-A was easier than
  A-a).
• After 1.5 seconds, appearance didnt matter.
• So, it looks like an early visual code is recoded
  into an auditory code within 1.5 seconds.

43
Properties of STM

• Code
• Note that since you get release from PI in a STM
  task, and release from PI is a semantic task
  (based on meaning), there must be some
  representation of semantic information in STM as
  well.

44
Properties of STM

• Search
• Weve been treating STM as a static storage
  place. We know it doesnt last long, it doesnt
  hold much, and interference is what causes
  forgetting. We also know it has a variety of
  information formats.
• Now lets think about processing. If you have
  something in STM and you are asked a question
  about it, how do you search for it?

45
Properties of STM

• Search
• Search of STM was the topic for Sternberg (1972).
• The task was simple
• Present a span list of 1-7 items (e.g., 3, 2, 6,
  9, 5, 7).
• Present a test digit (e.g., 2).
• Participant says whether or not the test digit
  was on the list.

46
Properties of STM

• Search
• Sternberg did two things
• Improved reaction time methodology by developing
  something called the additive factors method.
• Learned about short term memory search.
• We will digress for a moment to look at the
  additive factors method to help us interpret
  Sternbergs results.

47
Properties of STM

• Search
• The original method was the subtractive method.
• If there are different stages, find tasks that
  have different amounts of those stages and
  subtract them.
• For example, a task that requires you to respond
  when a light comes on differs from a task that
  requires you to respond one way to one light and
  a different way to a different light. (At least
  a decision stage differs.)

48
Properties of STM

• Search
• The additive factors method is to manipulate
  variables that affect different stages and look
  at how that affects time.
• You can tell if stages are independent and what
  goes on inside the stages.

49
Properties of STM

• Search
• Sternberg broke search up into four stages
  (starting after the test digit is presented)
• Encoding.
• Search.
• Decision.
• Response.

50
Properties of STM

• Search
• Different stages should be influenced by
  different variables
• Encoding Degraded vs. intact stimulus.
• Search How many items are in the memory set.
• Decision Yes or no answers.
• Response Probability of a particular response.

51
Properties of STM

• Search Obviously, we will be considering the
  search stage. How do people search STM?
• Search in parallel Search all items at once.
• Serial search, self-terminating Search items one
  at a time, stop when you find it.
• Serial search, exhaustive Search items one at a
  time, search them all regardless of where the
  item is in the list.
• Lets consider each in turn

52
Properties of STM

• Parallel search

53
Properties of STM

• Serial search, self-terminating

54
Properties of STM

• Serial search, exhaustive

55
Properties of STM

• Sternberg found that the search was serial and
  exhaustive. It seems counterintuitive, but it
  makes sense if the search is an automatic
  process.
• The function RT 38n 397 (ms)
• What do we know from that?
• Each comparison takes 38 ms.
• Stages 1, 3, and 4 take 397 ms together.
• Lets check our CogLab result

56
Properties of STM

• You could also look at other stages using the
  same technique.

57
Working Memory

• Lets make a transition. We now have the
  properties of short term memory, but weve been
  looking at it as a static storage device. What if
  we thought about its dual role as a storage
  device and a place where information is
  manipulated and transformed? Thats working
  memory.

58
Working Memory

• One big implication is in how we measure
  capacity. We need a task that involves both
  memory and processing.
• Reading span Read a set of sentences, hold the
  last word of each sentence in memory. After the
  set, recall. Start with sets of two, then three
  The average span is low compared to the regular
  span tasks (2-5.5).
• Operation span Another way of getting at span.
  See the CogLab results

59
Working Memory

• These measures correlate well with reading
  comprehension, SAT score, GPA, etc. In fact, they
  are better at estimating these kinds of variables
  than simple span tasks.

60
Working Memory

• Changing to working memory also has implications
  for the structure of the STM box (Baddeley, 1985)

Visuo-spatial sketchpad
Articulatory loop
Central executive
61
Working Memory

• Central executive Kind of the controller for the
  system, scheduling tasks, allocating resources,
  monitoring performance.
• Generate 100 random letters at one letter per
  second. It should be tough because the executive
  must monitor the output (which is automatic, but
  not favorable to randomness) and the executive
  must intervene to make those random, plus
  remember what was recently produced. At a slower
  rate, this isnt so tough.

62
Working Memory

• Central executive Evidence
• The evidence comes from neuropsychology patients
  with frontal lobe damage who have difficulties
  with executive function.

63
Working Memory

• Articulatory loop A slave system for storing
  verbal information temporarily.
• Traditional memory span tests could be seen as
  operating here. A lot of what weve said so far
  about STM could apply to the loop.
• One observation of the loop is that it seems to
  have a trace decay forgetting function. The
  duration it takes to say words is more important
  than the length of the words in determining
  forgetting (hence Welsh digit spans).

64
Working Memory

• Articulatory loop Evidence
• Conrad showed that letters that were auditorially
  more confusable (D, C, E) were harder to remember
  than lists of letters that were visually
  confusable, but not auditorially similar (B, K,
  R). (Connect to pattern recognition.)
• Articulatory suppression (saying the, the,
  the,) makes verbal tasks harder.
• Duration of materials affects performance.

65
Working Memory

• Visuo-spatial sketchpad A slave system for
  holding image-type information.
• People show similar limits on holding
  visuo-spatial information as they show for lists.
• A lab task similar to trying to count the number
  of windows in the house or apartment where you
  live interferes with image memory tasks.

66
Working Memory

• Visuo-spatial sketchpad Evidence
• Scanning time for images is similar to scanning
  time in the real world.
• Picture a rabbit by an elephant. Zoom in on the
  rabbits eyelash.
• Picture a rabbit by a fly. Zoom in on the
  rabbits eyelash.
• Mentally rotating objects takes longer the
  farther they have to rotate.
• It takes longer to imagine walking home carrying
  a cannonball than a balloon.
• Well see more of this in the imagery unit.

67
Working Memory

• How do we know the different systems are
  independent? Another double dissociation.
• Participants are in a dual task paradigm
• Primary Either a verbal memory task or a visual
  memory task.
• Secondary Either articulatory suppression or
  tapping.

68
Working Memory
69
Working Memory

• A pattern like that in the table would suggest
  that the capacities are independent.
• We could do something similar with the executive.
  Generating a string of random letters should
  interfere with decision-making tasks, we could
  probably work out double-dissociations.

70
Connection

• We can address our final question
• Pizza II. Why cant you remember a number and
  talk to someone, but you can remember a number
  while you look around the room?
• Why?

71
Working Memory Applications

• Ashcraft and Krause (2007)
• Working memory is essential for math performance.
• Problem-size effect Larger operands are more
  difficult to work with (9 x 6 vs. 4 x 5). Smaller
  ones more frequent in practice, more
  retrieval-based (automatic). Larger ones strategy
  (therefore working memory) driven.

72
Working Memory Applications

• Ashcraft and Krause (2007)
• Problem-size effect
• Example Larger minuends take longer, more
  errors, more strategy driven.

73
Working Memory Applications
Ashcraft Krause (2007, p. 244)
74
Working Memory Applications

• Ashcraft and Krause (2007)
• Problem-size effect
• People with low capacities or given working
  memory loads more strongly affected.

75
Working Memory Applications

• Ashcraft and Krause (2007)
• Working memory is essential for math performance.
• The number of steps in a problems solution is
  affected by working memory.
• Carry problems significantly harder.

76
Working Memory Applications
Ashcraft Kirk (2001, p. 230)
77
Working Memory Applications

• Ashcraft and Krause (2007)
• How does math anxiety affect math performance?
• Higher math anxiety goes with lower math learning
  and motivation (overall r -.31).

78
Working Memory Applications

• Ashcraft and Krause (2007)
• How does math anxiety affect math performance?
• Anxiety is associated with decreases on more
  advanced math.

79
Working Memory Applications
Ashcraft Krause (2007, p. 245)
80
Working Memory Applications

• Ashcraft and Krause (2007)
• How does math anxiety affect math performance?
• The value of psychology After cognitive
  behavioral interventions to reduce anxiety, math
  scores reach the normal range. This is with NO
  new instruction in math.

81
Working Memory Applications

• Ashcraft and Krause (2007)
• How does math anxiety affect working memory?
• Capacity is compromised when the task activates
  anxiety.

82
Working Memory Applications

• Ashcraft and Krause (2007)
• How does math anxiety affect working memory?
• 2-letter load

Ashcraft Krause (2007, p. 246)
83
Working Memory Applications

• Ashcraft and Krause (2007)
• How does math anxiety affect working memory?
• 6-letter load

Ashcraft Krause (2007, p. 246)
84
Working Memory Applications

• Ashcraft and Krause (2007)
• Education
• Not a lot of research on working memory and math.
• Math-anxious individuals will be more strongly
  impacted the farther they go.
• Anxiety leads to avoidance, closing off options.
• Teachers play a role in developing anxiety and
  its maintenance.

85
Working Memory Applications

• Ashcraft and Krause (2007)
• Education
• Attitudes (like youre either good or bad at math
  regardless of work) support anxiety.
• College majors with the highest levels of math
  anxiety Future elementary school teachers.

86
Working Memory Applications

• Where else might working memory capacity be an
  important predictor of performance?

87
End of Short Term/Working Memory Show