Transcript Kein Folientitel

Situation Models
and Embodied Language Processes
Franz Schmalhofer
University of Osnabrück / Germany
1)
Memory and Situation Models
2)
Computational Modeling of Inference Processes
3)
What Memory and Language are for
4)
Neural Correlates
5)
Integration of Behavioral Experiments and Neural Correlates (ERP;
fMRI) by Formal Models
Northwest of Germany
Schloß (University Building Osnabrück)
Cognitive Science in Osnabrück
Bachelor Degree:
Cognitive Psychology,
Computational Linguistics
Computer Science and
Artificial Intelligence
Neurobiology
Neuroinformatics
Mathematics
Philosophy of Mind
Masters Degree:
PhD Program:
Townhall of Osnabrück (early 17-th century
negotiations for a European peace)
Central Europe: 1618 - 1648
Arbitrary and Perceptual Symbols
Steckenpferd
Hobby Horse
Franz Schmalhofer
EDUCATION
• 1978 Diplom Psychology
Universität Regensburg
• 1982 PhD
University of Colorado
• 1996 Habilitation
Universität Heidelberg
EMPLOYMENT
• Universität Heidelberg und
Freiburg, 1982-1987
• McGill University/Montreal,
1987/88
• University of Colorado/USA,
visiting, 1989
• German Research Center for
Artificial Intelligence (DFKI)
Kaiserslautern, 1989-2000
• Cognitive Science, University
of Osnabrück; (Cognitive
Psychology), since 2000
Introduction
• Please introduce yourself:
Name, University, City, Country
• What you have studied so far
• What do you expect from this course
Situation models and Embodied Language
Processes
Lectures
1.
2.
3.
4.
5.
Memory and Situation models
Computational Modeling of
Inference Processes
What memory and language are
for
Neural correlates of language
processes
Integration of behavioral
experiments, computational
models, neural correlates (ERP;
fMRI)
Small group meetings
•
•
•
•
•
Analyze data from memory
experiments
Do a C-I model
Embodied language processes; extend
C-I model (marker passing);
Discuss ERP and fMRI
Levels approach, material structuring,
analysis of dependent measures
Memory and Situation Models
1.
2.
3.
4.
5.
Osnabrück, Maps,
Cognitive Science
Ebbinghaus and serial
position effects
Serial position curves ,
Aktinson & Shiffrin as
cognitive architecture
False memory, Loftus,
Bransford & Franks
Referenced situation
important, not so much the
stimuli
Analyze false memory;
• Enter data; one participant
per line
• Merge data
• Calculate means and standard
deviations
Herrmann Ebbinghaus (1850 - 1909)
• Born in Barmen (near Bonn)
• studied in Bonn, Halle,
Berlin,
• went abroad and tutored in
France and England
• 1880 „Privatdozent“ at the
University of Berlin
• 1885 „Über das Gedächtnis“
• then professor at Breslau and
Halle
• What made Ebbinghaus
famous?
Nonsense Syllables and Learning to
Criterion
• Ebbinghaus used only
himself as subject
• but strictly followed his
experimental prescriptions
• commit CVC-trigrams
(nonsense syllables) such as
„gid, var, mon“ to memory
• in comparison to such
syllables, memorizing a poem
was approximately nine times
faster
Number of
syllables in
a series
Number of
Probable
repetitions
error
required for
errorless
reproduction
7
1
12
16.6
16
24
36
30.0
44.0
55.0
+
+
+
+
-
1.1
0.4
1.7
2.8
A natural science with statistical laws
Ebbinghaus, (1885/1913):
• „When in repeated cases I memorised series of
syllables of a certain length to the point of their first
possible reproduction, the times (or number of
repetitions) necessary differed greatly from each other,
but the mean values derived from them had the
character of genuine constants of natural science (p. 52)
• logarithmic function of forgetting
Rubin & Wenzel (1996) One hundred years of forgetting.
Psychological Bulletin.
• Ebbinghaus‘ results confirmed by many data sets
Human memory
• Human intelligence as
adaptive behavior
– for computational reasons,
memory is a necessary
condition for being adaptive
• Memory system
– Is it possible to identify
subsystems of memory?
– How do they interact?
• Memory contents
– what information is stored?
– How is the information
organized?
– How is the information
retrieved
• Performance factors
– Under what circumstances do
we remember less? more?
Memory in common experience
Memory content durable?
yes
no
yes
A well-learned
phone number we
are currently
thinking about
A phone number
remembered just
for dialing
no
Well-learned phone
numbers we are
not currently
thinking about
A phone number
that we have
forgotten
Memory
content
currently
available?
 Discuss 
• Do the introspectively compelling properties of
availability and durability provide an argument for
postulating the existence of memory subsystems?
• What type of evidence would be needed to establish the
existence of functionally different memory subsystems?
Classroom demonstration
• Instruction
– 14 words are going to be
presented to you each for 1 s.
– try to remember as many as
possible
– after the presentation, write
down the words you remember
in any order you like
dog
pants
chair
sofa
coat
mouse
tree
horse
mail
video
table
salad
flower
shoe

Serial position effect
idealized diagram!
recall
80%
60%
40%
20%
3
6
9
12
15
serial position
Interpretation
idealized diagram!
recall
80%
short-term
memory
effect
60%
40%
long-term
memory
effect
20%
3
6
9
12
15
serial position
Variation A of the experiment
• Instruction
– 14 words are going to be
presented to you
– try to remember as many as
possible
– after the presentation, count
backward from 315 for 20 sec
– then write down the words you
remember
nice
sunny
rainy
green
fast
high
bright
cheap
noble
small
round
soft
blond
tired
tire

Serial position effect: exp. A
idealized diagram!
recall
80%
60%
40%
20%
3
6
9
12
15
serial position
Cognitive Architecture: Memory model of
Atkinson & Shiffrin (1971)
Enviromental
input
Sensory
registers
visual
.
.
auditory
Short term
store
Tempory
working
memory
(limited
capacity)
Response
output
Long term store
unlimited capacity
Multi-component working memory
• Baddeley (1992)
– Phonological loop
– Visuo-spatial sketchpad
– Central executive
Central executive
Visuo-spatial sketch pad
Phonological loop
Experimental/Empirical Research Strategy
in Cognitive Science
Human cognition
as an object of
investigation
experiment
modeling
Computational
theory
revise
derivation
Hypothesis about
specific situations
Data
prediction
Forgetting
• Underwood (1957) reviewed forgetting over 24-hour
retention interval
– 80% forgotten, if 15 or more lists had been previously learned
– 20- 25% if no earlier list had been learned
• Der Spiegel 14/1993: Wissenszwerge unter Druck:
„Und von dem was bei Hörern, Lesern und Sehern
ankommt, vergessen sie innerhalb von 24 Stunden mehr
als 80 Prozent“. P. 150
Two-process theory of memory
• Recall = Search + Decision (i.e. retrieval + recognition)
• Recognition = Decision
• Whereas Recall involves two fallible stages,
recognition involves only one
• Bahrick (1970): Pr (recall) = Pr (retrieval) * Pr (recog.)
yielded good prediction of empirical results
Recognition and Recall Experiment
Four Stages
• 1) Pairs of words are presented (cue + to-beremembered word); study second word (e.g. BLACK of
word-pair; the cues (train) need not be remembered but
might be helpful
– train - BLACK
• 2) Freely associate to the cues which will be presented;
write the words on the left side of a sheet of paper
– bed - ?
• 3) Recognition task
• 4) Recall task
Experiment by Tulving & Thomson (1973)
• 1) Study a word list with weak associates:
pretty - BLUE
• 2) Write down what comes to mind from strong
associates of the target word:
– sky - ? (BLUE)
• 3) Recognition test (recall in the presence of the target):
– BLUE; did it occur on list 1?
Only 24 % yes answers
• 4) Cued recall test:
– pretty - ? BLUE; 63 % of the words are correctly recalled
Classroom demonstration
• Instruction
– You will be presented with
three lists of English words
– Afterwards you will perform
numerical tasks
– and your memory will be
tested.
BED
BLANKET
REST
DOZE
AWAKE SLUMBER
TIRED
SNORE
DREAM NAP
WAKE
PEACE
SNOOZE YAWN
DROWSY
Classroom demonstration
TABLE
STEAL
JAIL
ROBBER
GUN
SIT
CUSHION
CROOK
VILLAIN
LEGS
SWIVEL
BURGLAR
CRIME
MONEY
COP
BANK
BAD
BANDIT
ROB
CRIMINAL
SEAT
COUCH
DESK
RECLINER
SOFA
WOOD
STOOL
SITTING
ROCKING
BENCH
Solve arithmetic problems
• 83 * 20 =
• 70 * 19 =
• 29 * 31 =
• Memory test
Recognition test on April, 21, 2002 in class; 45
participants
NO YES
How confident are you:
(1= not at all, 7=very)
objectively
DREAM
5
40
7
yes
FORK
45
0
7
no
WEATHER
45
0
6
no
BRACELET 45
0
5
no
CHAIR
10
35
4
lure
ROBBER
20
25
7
yes
STOOL
5
40
7
yes
TRAFFIC
45
0
6
no
Recognition test (continued)
NO YES
How confident?
objectively
SNOOZE
0
45
7
yes
COUCH
5
40
6
yes
RADIO
43
2
6
no
JAIL
10
35
6
yes
SLEEP
5
40
6
lure
SAND
45
0
7
no
BLANKET
25
20
6
yes
THIEF
42
3
5
lure
Bransford, Barclay & Franks (1972)
• 1a. Three turtles rested on a
floating log, and a fish swam
beneath them.
• 1b. Three turtles rested on a
floating log, and a fish swam
beneath it.
• 2a. Three turtles rested beside
a floating log, and a fish
swam beneath them.
• 2b. Three turtles rested
beside a floating log, and a
fish swam beneath it.
• Recognition test:
– When 1a was heard, false
alarms on 1b
– When 2a was heard, only rare
false alarms on 2b
Practical problems of memory
• Eyewitness testify that they
recognize a person as having
caused the accident or having
committed the crime
• Can we trust their memory,
when the event occurred a
day, a month, a year, many
years ago?
• adults recover „repressed
memories“ of sexual and/or
physical abuse they suffered
in childhood
• Are these so called recovered
memories genuine or are they
false memories (i.e. the event
never happened)
Influence of post-event information
Loftus & Palmer (1974)
• Memory for an incidence can be systematically distorted by
questioning that occurs subsequently
• Film with multiple car accidents is shown.
• Participants describe what has happened.
• Questions
– A) No question about car speed (control group).
– B) About how fast were the cars going when they hit each other?
– C) About how fast were the cars going when they smashed into each
other?
• One week later all participants were asked: Did you see broken
glass?
Estimated speed of car
Incorrect memories
Memory is suggestible
• Source misattribution:
Believing sth that never
happened
• Misinformation
acceptance: additional
information becoming
part of a genuine
experience
• Overconfidence in
memory: Misremember
what we have
experienced
•Repeated exposure to
misinformation
•Imagining that sth
happened increases
memory that it did happen
•Misinformation effects
occur even when
participants are warned
beforehand
Individual differences (Tomes & Katz 1997)
• People accepting misinformation, tend to have
– Poor general memory
– High scores on imagery vividness
– High empathy scores
Bransford & Johnson (1972)
• The procedure is actually quite simple. First you
arrange items into different groups. Of course one pile
may be sufficient depending on how much there is to
do. If you have to go somewhere else due to lack of
facilities, that is the next step; otherwise, you are pretty
well set. It is better not to overdo things. That is, it is
better to do too few things at once than too many. In the
short run this may not seem important, but
complications can easily arise. A mistake can be
expensive as well.
Recall and comprehension of „washing clothes“
story (Bransford & Johnson, 1972)
Summary of lecture (Memory)
• Empirical Precison: Forgetting and the differentiation betweeen short term
and long term effects
• Formal Specification: The use and usefulness of computational models
• Experimental Demonstrations:
– Distinction between explicit (verbal tasks) and implicit memories (behavioral
tasks)
– Context effects (encoding specificity, mood congruency, transfer appropriate
processing) in recall (action oriented) and recognition (perception oriented)
– „False memories“ and the distortions of memories, but also the improvement of
memory by topical information
• References:
– Any recent text book on Cognitive Psychology (e.g. Eysenck & Keane, 2000,
Kellogg, 2003, or Sternberg, 2003)