BChap4

advertisement
+
Chapter 4
Learning
+
2
Herman Ebbinghaus

First to propose an experimental study of memory, inspiring a
new science of learning and memory.

Tested a single subject: himself … for over 2 years.

To minimize the effects of previous learning, he:


Studied, tested, and relearned nonsense syllables comprising
consonant–vowel–consonant combinations, e.g. WUX.
Avoided making associations with real words.

He always tested himself under similar conditions, including
time of day.

He aimed to determine the rate of learning and forgetting.
+
3
Testing the Rate of Learning

The Total Time Hypothesis:


The Experiment:



The amount learned depends on the time
spent learning
Each day, Ebbinghaus learned a new list of
syllables, reciting the list at a constant rate
for varying numbers of repetitions.
24 hours later, he determined how many
additional trials he needed to relearn the list
(called savings).
 The more trials necessary, the worse his
memory was.
The Results:

Learning was linearly related to amount of
study.
From Ebbinghaus (1885).
+
4
The Total Time Hypothesis
 The
total time hypothesis has held up as a
rule of thumb for over a century.
 The
“practice makes perfect” approach applies not
only to word learning, but also to skills like writing,
chess, typing, and music.
 However,
there are ways that the
effectiveness of studying can be maximized
to get more learning from the time spent.
+
5
Distributed Practice
 The

Distributed Practice Effect:
It is better to space out learning trials sparsely (thinly) across
a period of time than to mass them together into a single
learning block.
 This leads to faster improvement rates and more lasting
retention.
 As distributed practice takes longer in absolute terms (i.e.
less actual training, but more days), it is not always
practical or convenient.
 Individuals using distributed practice often paradoxically
feel as though they’re being less efficient.
 The benefit is not related to fatigue with denser learning.
+
Massed Practice is a Relatively
Inefficient Way to Study
Rate of learning a typing skill for a range of training schedules: 1 × 1 equals one session of 1
hour per day, 2 × 1 equals two such sessions, 1 × 2 is one session of 2 hours and 2 × 2 two 2hour sessions. From Baddeley and Longman (1978). Copyright © 1978 Taylor & Francis.
Reproduced by permission (http://www.tandf.co.uk/journals).
+
7
The Distributed Practice Effect
Microdistribution (Landauer & Bjork, 1978)

Is it better to repeatedly study and test items on a list (e.g. word
pairs) in close succession or spread apart? There are
competing factors:
Spacing Effect

Spaced presentation (i.e. study)
enhances memory for a variety
of materials

Based on this alone, study
and test should be separated
as much as possible, but …
Generation Effect

Successfully testing yourself
strengthens memories more
than passively studying items

The sooner an item is tested
after initial presentation, the
more likely it will be recalled
and strengthened
The Solution: The Expanding Retrieval Method
+
8
Expanding Retrieval
(Landauer & Bjork, 1978)


A flexible strategy, in which:

A new item is initially tested after a short delay to ensure that it is
recallable.

As the item becomes better learned, the practice–test interval is
gradually extended.
Each item should be tested at the longest interval at which it
can be reliably recalled.

Recall failure indicates that it should be presented after a shorter
delay. Successful recall indicates that the delay should be
increased.

Optimum is about 10-20% of study time before testing. For testing
after 10 days, spacing between presentations should be 1-2 days. In
general, longer inter-trial delays are better than short ones.
+
9
The Distribution of Practice Effect
Finding the Optimal Spacing (Pashler et al., 2007)

Example: Pimsleur language learning samples:
http://www.cateeslanguageworld.com/pimsleur/samples.php

The optimal spacing depends on the length of the delay
between learning and testing

The optimum interval between learning episodes should be 10–20%
of the test delay

In general, longer inter-trial delays are preferable to shorter
ones

Receiving corrective feedback after test trials is important,
though it can occur after a short delay without consequence
+
Computing Optimum Spacing of
Study Sessions
Chap 1
Chap 2,
Review
Chap 1
Chap 3, Chap 4
Chap 5,
Review Review
Review
Chap 1-2 Chap 2-3 Chap 2-4
Chap 6,
Review
Ch 1, 3-5
Chap 7,
Review
Ch 2, 4-6
Chap 8,
Review all
chapters
Midterm
First day of
class
Four months (28 days)
Optimum spacing is 10-20%, so .1 x 28 = 2.8 and .2 x 28 = 5.6 so
optimal study time of each piece of information is every 2.8 to 5.6
days. This includes review of previous material not just new
presentations.
+
11
The Importance of Testing
Karpicke and Roediger (2008)

Karpicke and Roediger (2008) wanted to establish the
importance of testing, with four groups who studied and were
tested on Swahili–English word pairs over the course of a
week:

Group 1 (ST): All the word pairs were repeatedly studied and tested.

Group 2 (SNTN): After a word was successfully recalled, it was not
studied or tested further.

Group 3 (STN): After a word pair was successfully recalled, it
dropped out of additional test rotations (they continued to be
studied).

Group 4 (SNT): After a word pair was successfully recalled, it
dropped out of additional study rotations (they continued to be
tested).
+
12
The Importance of Testing
Karpicke and Roediger (2008)

Then, after a week’s delay (e.g., at the end of two weeks), the
groups were tested on all the word pairs

Results:


The rate of learning did not differ across groups during week 1

Retention over the second week, did differ, however:

Groups 1 and 4 (those that continued to be tested) recalled 80%
after the delay

Groups 2 and 3 (those that did not continue testing of learned
pairs) only recalled 30% after the delay
The Message:

Continue to test yourself on vocabulary items you already know!
+
Importance of Testing
SNT tests instead of
presenting another
study trial – STN
presents a study trial
instead of testing.
Notice the difference.
The importance of
testing for later
remembering. The
pattern of learning and
test trials had no effect
on rate of learning, but
the presence of tests
had a major effect on
what was remembered 1
week later. From
Karpicke and Roediger
(2008). Copyright ©
1980 AAAS. Reprinted
with permission.
+
14
The Importance of Testing
What Type of Test is Most Effective?
 The
generation effect shows that having to come up
with an answer, rather than having it provided
simultaneously, leads to greater retention.



This is true even if the test and the answer are only
separated by a very brief delay (Metcalf & Kornell, 2007).
Multiple-choice tests enhance subsequent long-term recall
(Marsh et al., 2007).
However, short-answer tests are more effective than multiplechoice in enhancing subsequent recall.
+
15
The Importance of Feedback
 Recall
errors made early in training persist unless
corrected with feedback.
 Instructing
normal, healthy participants to guess
when they know that they will receive feedback
yields equivalent performance on multiple-choice
tests compared to those instructed NOT to guess.
 Amnesic
patients don’t benefit from feedback (since
they can’t remember it); instead they benefit more
from learning procedures that avoid errors in the first
place.
+
16
Motivation to Learn
 Unlike
in animal learning studies, human participants
are often intrinsically motivated to please the
experimenter, prove their proficiency at the task, and
to maintain interest.

Nilsson (1987) demonstrated that adding cash rewards or
emphasizing social competition to further motivate
participants did not affect the results of a memory task.
 Motivation
does have an indirect effect on learning, it
helps determine:


The amount of time devoted to the material.
The degree of attention devoted to the material.
+
17
Repetition and Learning

Simple repetition of learning
material is not sufficient for
learning:

Especially if the information is
complex or if it is not
perceived as immediately
useful.

Having an existing, wellformed schema for the new
material allows for its easier
integration into memory.

Housewives recalled the date
of a radio frequency change
but not the frequencies.
From Rubin and Kontis (1983). Copyright © The Psychonomic Society.
Reproduced with permission.
+ Change Blindness
http://viscog.beckman.illinois.edu/flashmovie/23.php
 Change
Blindness:
 The
phenomenon in which some prominent feature
of the visual environment is dramatically changed
without the perceiver apparently noticing.
 This is NOT due to:
 A failure to take notice of the feature that is
eventually changed.
 The fragility of visual STM.
 LTM
for complex scenes is not very detailed.
18
+
19
Implicit Learning
 Non-Declarative
 Is
or Implicit Learning:
evident in changes in behavior, rather than
explicitly remembering information.
 Can be divided into three broad categories:
 Classical Conditioning
 Priming
 Procedural Learning
+
20
Implicit Learning
Classical Conditioning

Classical Conditioning:


An associative learning process in which an unconditioned
stimulus (UCS) automatically produces the unconditioned
response (UCR); after repeatedly pairing a neutral, conditioned
stimulus (CS) with the US, the CS eventually comes to elicit a
conditioned response (CR), often similar to the UR, on its own
First noted by Twitmyer (1902) in a knee reflex

Pavlov (1927) was credited with its discovery in the salivary reflex in
dogs:

Meat powder (UCS) causes dogs to reflexively salivate (UCR)

Repeatedly presenting a bell (CS) before the meat powder (US)
results in the dogs salivating (CR) to the bell (CS) alone
+
Pavlov’s Studies
Classical Conditioning
Prior to conditioning
Neutral stimulus
(tone)
(Orientation to sound
but no response)
UCS
(food powder in mouth)
UCR
(salivation)
Conditioning
Neutral stimulus
CS (tone)
+
CR
(salivation)
UCS
(food powder)
After conditioning
CS
(tone)
CR
(salivation)
+
23
Implicit Learning
Classical Conditioning
Features/Types of Conditioning
Description
Extinction
If the CS is repeatedly presented
without the UCS, the production of
the CR gradually decreases
Backward (Trace) Conditioning
When the CS follows the UCS
during training, much less
conditioning occurs (very weak)
Latent Inhibition
Repeatedly presenting the CS
alone before forming its association
with the UCS impairs subsequent
conditioning of the CS
Mere-Exposure Effect
Simply increasing one’s exposure
to a novel stimulus will increase its
rated pleasantness (though not
necessarily memory for detail)
+
Conditioning in Advertising
Conditioned attitude to a novel brand of toothpaste as a function of the number of
conditioning trials. Participants rated the likelihood that they would choose the
positively conditioned brand over the randomly associated control brand.
C, conditioning; RC, random control. Data from Stewart et al. (1987).
+
25
Implicit Learning
Priming
 Priming:



The ability to implicitly influence the subsequent perception or
processing of material by presenting it or a related stimulus
beforehand. Involves non-conscious processing.
Occurs across all senses, as well as for verbal and visualspatial memories – a coherent memory system.
Amnesic patients demonstrate normal priming (Warrington &
Weiskrantz, 1968) – visually degraded words were
recognized when subjects were asked to “guess.”
 Conversely, explicitly asking amnesics to remember what
they saw previously (a declarative task), results in highly
impaired performance.
+
26
Implicit Learning
Two Priming Tasks
Priming Procedure
Description
Example
Stem Completion
Provide the first few
letters of a previously
seen word and ask them
to come up with any
word that fits*
Present: STAMP
Test: ST_____
Result: more likely to
generate “STAMP”
Word Fragment
Completion
Provide only some of
the letters of a
previously presented
word and ask what
might fit*
Present: ELEPHANT
Test: _L_P_A_T
Result: More likely to
generate “ELEPHANT”
*Note: To be an implicit task, no reference/requirement to remember what
they saw previously is made.
+
27
Implicit Learning
Durability of priming
 Priming
is more durable over the long term,
but explicit memory is better over the short
term (Tulving, Schacter, Stark, 1982).
 For
short delays (e.g. 1 hour):
 Explicit recall produces better performance than
word fragment completion.
 For long delays (e.g. 1 week):
 Word fragment completion remains effective,
whereas free recall reveals substantial forgetting.
+
28
Implicit Learning
Priming improves when the physical conditions
present during encoding are reinstated at testing.

Types of priming:


Perceptual priming
 The prime and the target share physical properties
 Example prime: NURSE
 Example target: N_R_E
 Answer: NURSE
 Sensitive to the modality and form of the stimuli
Conceptual priming
 The prime and the target are related semantically
 Example prime: NURSE
 Example target: D_C_O_
 Answer: DOCTOR
+
Encoding Specificity & Priming
Influence of encoding semantically
or physically on explicit cued recall
versus implicit stem completion.
Performance improves when the
test conditions match the physical
conditions present during encoding.
Based on Graf and Mandler (1984)
Experiment 3.
+
30
Implicit Learning
Procedural Learning

Procedural Memory:


Long-term memory for skills and procedures – a form of learning that
doesn’t rely on conscious strategy.
Masters’s (1992) “Choking” Study:



Training: Golf putting
 Half also had a concurrent demanding attentional task during training.
Testing:
 Half of each group was tested under stressful conditions.
Results:
 Learning is somewhat impaired by the concurrent task.
 Those trained with the concurrent task were more resistant to stress.
 The concurrent task reduced reliance on explicit putting strategies,
which are prone to disruption.
+
Implicit Learning (with Concurrent
Task) is More Resistant to Stress
Mean number of putting shots holed
as a function of skill acquisition
phase (sessions 1–4) and a test
phase (session 5) in the study by
Masters (1992). Implicit learning led
to lower performance but was more
resistant to the effects of stress at
test.
+
32
Implicit Learning
Learning Artificial Grammars

Natural grammar (for one’s native language) appears to be
learned implicitly, without formal training.

Grammars are generative:


Grammar permits the items of the language to be combined and
recombined to form nearly an infinite number of statements.
Artificial grammars are designed with arbitrary rules allowing certain
combinations but not others – used to study language learning.

People can learn to reliably determine whether a sequence is
grammatical, but they generally can’t explicitly state the rules of
the artificial grammar (Reber, 1967).

Learning appears to generalize, suggesting that subjects have
acquired the underlying rules, though there is some question
about what exactly has been learned.
+
Sample Artificial Grammar
A finite state grammar of the type used by Reber to study implicit learning.
Three examples of grammatically permissible sequences are shown.
+ Methods of Second Language Learning
Explicit Training

Explicitly learning grammatical
rules of a language

e.g. going through a grammar
workbook in Spanish class
Immersion Method

Implicit learning in which
grammatical rules are acquired
simply by being surrounded by
the language


34
e.g. spending a year abroad
in Spain
Ellis (1993; 1994) tested whether teaching grammar by example or by
explicit rules is better

By example: Rapid learning but no generalization

Explicit rules: Good explicit knowledge of rules but no generalization

Applying explicit rules to samples: Slow learning but showed generalization
Thus, the immersion method may not be a sure bet for generalized learning,
although this has not been tested with exposure and learning conditions
equivalent to native language learning in early childhood.
+
35
Implicit Learning
Teach Back Method
 Could
an apparent lack of explicit knowledge
simply reflect difficulty in verbalizing a
nonverbal task?
 The
Teach Back Method:
 After training, the learner is asked to instruct a
naive participant how to perform the task.
 This method appears to produce some level of
explicit information transmission, but only after
numerous trials.
+
36
Learning and Consciousness
 Core

A continuum ranging from being fully awake and alert to deep
sleep/coma.
 Is



Consciousness (Damasio, 1994):
learning possible without normal consciousness?
Learning tends to be better when awake and attentive.
Claims for learning during sleep are unsubstantiated.
 However, sleep may help consolidate previous learning.
Some implicit learning under anesthesia has been reported.
 Little evidence exists for explicit learning when anesthesia
is correctly applied.
 Thus, conscious awareness may be less necessary for
implicit learning than for explicit learning.
+
37
Explaining the Dissociation between
Implicit and Explicit Memories
 Hypothesis
1: The different types of implicit learning
all share a common learning mechanism.
 Hypothesis
2: The only real commonality between
the different types of implicit learning is the absence
of episodic learning – Baddeley likes this view.


Episodic learning: a system that glues together events that
we experience at the same time.
This view is supported by neuroimaging evidence, which
reveals distinct brain regions underlying the various forms of
implicit learning.
+
38
Explicit Learning and the Brain
Hebbian Learning
 Hebb
(1949) proposed that long-term learning is
based on cell assemblies:



When two separate neurons are repeatedly excited in
synchrony (at the same time), the chemistry of the synapse
(gap) between the neurons changes.
This makes each one more likely to have an action potential
when the other does.
“Neurons that fire together wire together.”
 Hebb
believed that short-term learning was reflected
in temporary electrical changes within existing cell
assemblies.
+
39
Explicit Learning and the Brain
Long-Term Potentiation
 Neurobiological
 Bliss
evidence for Hebb’s idea:
and Lomo (1973) repeatedly stimulated
axonal pathways, which led to lasting increases in
the electrical potentials generated by post-synaptic
neurons, called long-term potentiation (LTP).
 These changes (LTPs) are strongly represented
in the hippocampus and surrounding regions
associated with long-term memory.
 They also occur in the amygdala supporting
emotion-based learning (classical conditioning).
+
40
Explicit Learning and the Brain
Is LTP really the mechanism behind learning?
 Drugs
that enhance synaptic transmission
enhance learning, whereas drugs that block
LTP tend to reduce learning.
 Hippocampal
LTPs are blocked by a substance
known as AP5.
 Administering AP5 impairs rats’ spatial learning in
the Morris water maze.
 Hippocampal lesions (as opposed to drugs that
block LTP) cause a similar deficit.
+
Morris Water Maze
Typical swimming paths shown by rats within a Morris water
maze. Normal rats (a) rapidly acquire a direct path, as do rats
with cortical lesions (b), whereas hippocampal lesions result in
a failure to learn (c). Data from Morris et al. (1982). Similarly
rats with AP5 blocking LTP performed like (c) rats with
hippocampal lesions, increasing with size of dose of AP5.
+
42
Explicit Learning and the Brain
Neurotransmitters
 The
neurotransmitter N-methyl-D-aspartate
(NMDA)
 While
NMDA receptors are necessary for LTP to
occur, it may not be sufficient for learning.
 This leaves a role for other psychological
processes and underlying brain regions, such as
attention.

Genetic factors also play a role.
+
43
Explicit Learning and the Brain
Consolidation
 Consolidation:


A process whereby a temporary memory trace becomes
established, either on a molecular level or a systems level (i.e.
transfer from one brain region to another brain region).
Classically thought to be interrupted by electrical brain stimulation
or drugs, damaging unconsolidated memories, but:
 Memory traces are not necessarily lost but are just temporarily
inaccessible.
 Consolidated memories again become vulnerable to disruption
whenever they are recalled (reconsolidation).
 Such complications prompted Nadel and Moscovitch’s
multiple-trace theory, which assumes that each retrieval sets
up new episodic and semantic traces.
+
44
Implicit Learning and the Brain
No Single System can Account for Learning
Procedure: One of the presented colors is paired with a
loud sound, which induces an anxiety response (increased
skin conductance) – Classical conditioning
Bechara et al.
Participant Group
Tests: Participants are shown the colors alone to test for a
conditioned anxiety response and asked to recall the
colors (which one was paired with the sound?)
Results
Normal Controls
Normal fear conditioning and explicit memory for
the colors
Amygdala Damaged
No fear conditioning but normal explicit memory for
the colors
Hippocampus Damaged
Normal fear conditioning but no explicit memory for
the colors
Damage to Both Areas
No fear conditioning and no explicit memory
Conclusion: The amygdala is important in human fear conditioning
More recent neuroimaging studies support this conclusion
+
Bechara et al.’s Findings
SM = bilateral amygdala damage
WC = hippocampal damage but intact
amygdala
RH = damage to both hippocampus and
amygdala
Conditioning of anxiety to the blue slide
accompanied by loud noise is affected by
damage to the amydala (left). Declarative
learning (explicit) for colors is affected by
damage to the hippocampus (right). Control
stimuli are shown in the center. Control
subjects (top) show both forms of learned
response.
Control data and performance of three
contrasted patients (SM, WC, and RH)
studied by Bechara et al. (1995). Copyright
© 1995 AAAS. Reprinted with permission.
+
46
Implicit Learning and the Brain
Different Components are Associated with Priming
 Schott
et al.’s (2005) fMRI study found different patterns
of activity with explicit vs implicit processing in a stem
completion task.
 During
learning, subjects were asked to count the
syllables in 160 words.



Under explicit instructions (try to remember the words) they
found increased blood flow in both hemispheres in the parietal,
temporal, and frontal regions.
Under implicit instructions (say the first word that comes to mind)
they found decreased blood flow in the left fusiform gyrus and in
the frontal and occipital regions.
This reduction likely reflects easier processing of primed words.
+
47
Implicit Learning and the Brain
Procedural Learning
 Serial



Reaction Time Task:
A well-studied task of procedural, implicit learning
Task – Four buttons and four lights:
 Press the button that corresponds to the light that is on.
 A pattern is introduced to the button pressing sequence,
leading to faster performance, then the pattern is changed.
Results:
 Both normal and amnesic patients learn the patterns and
both slow down when the pattern is changed (not relevant).
 Simultaneously performing an attentionally demanding task
eliminates conscious pattern awareness (explicit) without
eliminating the learning effect. The task becomes implicit.
+
48
Implicit Learning and the Brain
Procedural Learning
 Neuroimaging
the Serial Reaction Time Task
(Grafton et al., 1995):
 Implicit
condition (dual-tasking):
 Learning-related changes occur in the left motor
and supplementary motor cortex.
 Single-task condition (permitting greater pattern
awareness):
 Learning-related changes occur in the right
prefrontal cortex, premotor cortex, and the right
temporal lobe, as in other previous studies.
Download