Hypermnesia: Why does memory improve over repeated tests

1
Hypermnesia Why does memory improve over
repeated tests?

• Hajime Otani, Central Michigan University
• Robert L. Widner, Jr., University of Colorado at
  Colorado Springs

2
What is hypermnesia?

• Improvement in memory performance over repeated
  testing.
• Erdelyi and Becker (1974)
• Presented a mixed list of words and pictures
• Administered three tests - T1 T2 T3
• Found - memory improved over three tests for
  pictures but not for words.
• Important - The stimulus materials were not
  presented between tests.

3
Big Question

• How does memory improve over repeated test?

4
Interesting

• Counter-intuitive based on Ebbinghaus forgetting
  curve.
• Test 1 should show the best performance.

5
So, why does memory improve?

• Erdelyi and colleagues
• Imagery hypothesis
• Memory improves only when the materials are
  encoded as pictorial images.
• Erdelyi and Becker (1974) and others -
  hypermnesia occurs only when pictures or words
  with imagery instructions are used.

6
But, soon people discovered...

• Hypermnesia occurs without pictures or words with
  imagery instructions.
• Good example - Roedgier, Payne, Gillespie, Lean
  (1982) Exp. 2
• used nonsense syllables.

7
New hypothesis

• Roediger et al (1982)
• Cumulative recall level hypothesis
• Hypermnesia occurs when performance on the first
  test does not reach asymptote.
• Prediction - the higher the asymptote, the
  greater the hypermnesic effect, because it takes
  more time to reach higher asymptote.
• Studies generally have supported this.

8
But

• This is not complete - Roediger and Challis
  (1989)
• This is only a functional account. So, how does
  performance increase toward the asymptotic
  performance?
• It considers only one component of hypermnesia.

9
In fact.

• Two components of hypermnesia.
• Reminiscence (item gain) - retrieval of new
  items.
• Inter-test forgetting (item loss) - forgetting of
  previously recalled items.
• So, on the second test
• Gain 20 items (wow!)
• But, if you lose 20 items or more, performance
  does not increase.

10
But, to be fair.

• It is common to find similar forgetting across
  conditions.
• But still, you cannot forget this component.

11
Good example...

• Payne (1986)
• equated the asymptotic performance of pictures
  and words.
• Prediction - this should equate the amount of
  hypermnesia.
• Wrong - Pictures still showed better hypermnesia.
• Why? - less inter-test forgetting

12
So, how do you explain hypermnesia?

• My approach - use different retrieval tests.
• Assumption - different tests require different
  processing.
• Others were already testing hypermnesia using
  different tests.
• Also, manipulate the encoding strategies.
  Because processing at encoding may interact with
  processing at retrieval.

13
My selection

• Retrieval tests
• Recognition Cued recall
• Encoding strategies
• Relational and item-specific processing (Hunt
  colleagues)
• Relational - relate items together. Create
  organization.
• Item-specific - make each item distinctive.

14
My selection was based on...

• Recognition - others tried and failed.
• Cued recall - prediction was negative.
• Relational item-specific processing - Hunt and
  colleagues were trying to explain recall and
  recognition using these notions.

15
How did it go?

• Recognition - No luck what-so-ever.
• Otani Hodge (1991)
• Otani Stimson (1994)

16
Cued recall

• Turned out to be somewhat interesting
• Otani Hodge (1991) Experiment 2
• Presented with 36 word pairs - cue - target
• Categorized list - targets belong to 6
  categories.
• Relational - Sort the targets into categories.
• Item-specific - Rate the pleasantness of the
  targets.
• Three cued recall test - 7 minutes each.

17
Results.

• Relational - hypermnesia
• Item-specific - no hypermnesia
• Intentional - no hypermnesia
• The difference - Greater reminiscence in the
  relational condition.
• Only marginally significant difference in
  forgetting.

18
But.

• Performance was much lower in the relational
  condition.
• So, we decided to use a loosely categorized list
  (no obvious categories).
• Prediction - The use of such a list would reduce
  the hypermnesic effect.

19
Results

• No overall difference in performance.
• There was a slight improvement in the
  item-specific processing condition.
• But, the greatest improvement occurred in the
  relational processing condition.
• Reason - Greater reminiscence.
• No difference in inter-test forgetting.

20
Why is it?

• My initial reaction was - the category labels
  created extra retrieval cues.
• E.g., So, what was it? It was a fruit, red
  one maybe I got it. It was apple.
• Similar to Roediger and colleagues idea.

21
Elaborated by Payne, Hembrooke, Anasatasi,
(1993)

• Two mechanisms based on the SAM model.
• Incrementing - memory trace and its connections
  to other items become stronger and retrieval
  becomes faster.
• Alternative retrieval routes - use different
  retrieval paths to find new items.

22
Otani, Widner, Whiteman, St. Louis (in press)

• Tested the alternative retrieval routes
  hypothesis.
• Approach - give participants more than one
  retrieval cue.

23
Experiment 1

• Presented 36 word pairs - cue - target
• Single cue condition - no additional cues.
• E.g., war - bomb
• Multiple cue given - two additional cues.
• E.g, war, kill, loud - bomb
• Multiple cue Generate - Asked to generate two
  additional cues
• E.g., war, ___, ____, - bomb

24

• Three cued recall tests - 7 minutes each
• Presented only one cue for all the conditions.
• E.g., war - ______
• Assumption - Participants would use additional
  cues to retrieve previously unretrieved items.

25
Results

• Performance improved for both the multiple cue
  given and multiple cue generate conditions.
• No improvement for the single cue condition.
• Reason - Greater reminiscence.
• No difference in inter-test forgetting.

26
Experiment 2

• Did they really use the additional cues?
• Made extra cues explicit at retrieval.
• Presented 60 word pairs.
• Three cued recall tests - 7 minutes each.

27

• Three conditions
• Single - Single - single cue at study and single
  cue at test.
• Multiple - Single - multiple cue at study and
  single cue at test.
• Multiple - Multiple - multiple cue at study and
  multiple cue at test.

28
Results

• Performance improved for all conditions.
• Reason - categorized list.
• But, improvement was greatest in the multiple -
  multiple condition.
• Reason - Greater reminiscence.
• No difference in inter-test forgetting
• Improvement was similar between the single -
  single and the multiple - single.

29
Experiment 3

• What about free recall?
• Without explicit cueing at test, multiple cues
  may not work.
• Presented 60 word pairs - cue - target
• Three free recall tests - 7 minutes eac

30

• Two conditions
• Single cue - no additional cue at study and no
  cue at test
• Multiple cue - two additional cues at study and
  no cue at test

31
Results

• Performance improved for both conditions.
• But, the improvement was greater in the multiple
  cue than in the single cue condition.
• Reason - Greater reminiscence.
• No difference in inter-test forgetting.

32
Our initial conclusion was.

• Multiple cues produce greater hypermnesia.

33
But..

• Is it really the multiple retrieval cues?
• Hunt and colleagues
• Hypermnesia is based on relative contributions of
  relational and item-specific processing.
• Relational processing - reduced inter-test
  forgetting.
• Item-specific processing - increase reminiscence.

34
What are we doing?

• Asking participants to attach three cues that are
  related to one target.
• E.g., war, kill, loud - bomb.
• These cues may direct participants attention to
  unique characteristics of the target word.
• This is the definition of item-specific processing

35
Reinterpretation

• Multiple cues produce greater hypermnesia because
  multiple cues induce greater item-specific
  processing.
• How to test this?

36
Our approach (Otani,Widner, Throne)

• Manipulate both the number of cues and processing
  condition.
• Item specific Relational
• Single
• Multiple cue-given
• Multiple cue-Generate

37
Prediction

• If the number of cues is important - Multiple
  cues would produce greater hypermnesia regardless
  of processing condition.
• If processing condition is important - There
  would be an interaction.

38
Nature of interaction

• If cues are inducing item-specific processing,
  they are redundant with the item-specific
  processing task participants are already
  performing.
• Redundant processing - no additional benefit.
• No effect of the number of cues.

39

• But, for the relational processing condition, the
  cues would induce non-redundant processing.
• Non-redundant processing - additive effect.
• The greatest benefit of non-redundant processing
  should occur in the Multiple-cue generate
  condition (i.e., greatest hypermnesia).

40
Results

• T1 T2 T3 T3-T1
• Intentional
• Single 19.85 20.40 20.85 1.00 5.00
• M. Given 17.00 18.75 18.80 1.80 11.00
• M. Generate 22.85 24.70 25.80 2.95 13.00
• Item-specific
• Single 20.60 21.45 23.45 2.85 14.00
• M. Given 15.00 15.85 16.85 1.85 12.00
• M. Generate 20.00 20.65 21.55 1.55 8.00
• Relational
• Single 12.30 13.10 13.35 1.05 8.00
• M. Given 12.75 14.05 15.30 2.55 20.00
• M. Generate 20.45 23.05 24.45 4.00 20.00

41
Significant three-way interaction

• Separate ANOVA on each processing
• Found
• The number of cues interacted with test in the
  intentional and relational processing condition.
• The number of cues DID NOT interact with test in
  the item-specific processing condition.

42
Why?

• Possible - its not the number of cues, but the
  processing that is important.
• What else can we look at?

43
Reminiscence and Forgetting

• Int Isp Rp
• Reminiscence
• Single 1.85 3.05 3.05
• M. Given 2.70 3.00 3.75
• M. Gen 3.60 2.90 5.65
• Forgetting
• Single 1.05 0.85 2.25
• M. Given 1.05 1.55 1.55
• M. Gen 0.80 1.95 1.85

44
Well .

• Reminiscence - Should mirror the net recall. It
  does. But, the Processing x Number of cue is
  only marginally significant.
• Forgetting - The greater the relational
  processing, the lower the forgetting should be.
  The greater the item-specific processing, the
  higher the forgetting should be.

45
So.

• The results indicate that the type of processing
  is important.
• But, results are not completely consistent with
  Hunt and colleagues prediction.
• Problem - maybe due to retrieval test. Cued
  recall versus Free recall.

46
Conclusion

• Currently two theories are viable as the
  explanation of hypermnesia.
• One is Roediger and colleagues explanation based
  on the notions of incrementing and alternative
  retrieval routes.
• The other is Hunt and colleagues explanation
  based on the notions of relational and
  item-specific processing.