Semantic Memory

1
Semantic Memory Knowledge memory Main
questions How do we gain knowledge? How is
our knowledge represented and organised in the
mind-brain? What happens when we access
information? (Note 2nd and 3rd questions are
strongly related.)
2
Semantic Memory Knowledge memory Important
task lexical decision task make a
word-nonword judgement for a letter string
3
higgle
4
murget
5
beer
6
stout
7
Semantic Memory Knowledge memory Main
questions How do we gain knowledge? Repetitio
n memorisation of lists (Ebbinghaus)
consider lexical decisions across
multiple presentations
8
Lexical Decision RT for Words and Nonwords As a
Function of Number of Trials  
700
Nonword
RT (ms)
Word
400
1 2 4 6 8 10 . . . .
30
Number of Trials
9
Semantic Memory How do we gain knowledge?
Repetition Drop in lexical decision RT
across repetitions, especially for
nonwords After many reps, nonword RT as
low as word RT
10
Lexical Decision Threshold for Words and Nonwords
As a Function of Number of Trials  
100
Nonword
Threshold (ms)
Word
0.0
1 3 6 . . . .
30
Number of Trials
11
Semantic Memory How do we gain knowledge?
Repetition Drop in lexical decision
thresholds across repetitions, especially
for nonwords After roughly 6
presentations, nonword decision threshold
as low as word threshold
12
Semantic Memory Knowledge memory Main
questions How is our knowledge represented and
organised in the mind-brain? What happens when
we access information? (These questions are
strongly related.)
13
Semantic Memory Organisation Semantic network
(Collins Quillian,1969 ) hierarchical
organisation categories within categories
properties of items (nodes) represented once
at highest category level possible cognitive
economy some nodes connected to each
other properties connected to nodes
14
Node (a representation)
Animal
15
properties
Breathes
node
Animal
Skin
16
p
Breathes
node
Animal
p
Skin
17
p
Breathes
Animal
p
Skin
is a
Fish
18
p
Breathes
superordinate
Animal
p
Skin
is a
subordinate
p
Gills
Fish
p
Fins
p
Swims
19
p
Breathes
Animal
p
Skin
is a
Swims
p
Gills
Fish
p
Fins
is a
p
Pink flesh
Salmon
p
Cold water
20
Spreading activation activation of a node
spreads through the network spread of
activation is automatic the strength of
activation dissipates across nodes farther
nodes receive less activation activation
decreases with time
21
Breathes
Animal
Skin
Swims
Gills
Fish
Cold blooded
Pink flesh
Salmon
Cold water
22
Breathes
Animal
Skin
Swims
Gills
Fish
Cold blooded
Pink flesh
Salmon
Cold water
23
Breathes
Animal
Skin
Swims
Gills
Fish
Cold blooded
Pink flesh
Salmon
Cold water
24
Breathes
Animal
Skin
Swims
Gills
Fish
Cold blooded
Pink flesh
Salmon
Cold water
25
Breathes
Animal
Skin
Swims
Gills
Fish
Cold blooded
Pink flesh
Salmon
Cold water
26
Evidence Sentence verification task (measure
RT) A salmon is a salmon. A salmon is a
fish. A salmon is an animal. Prediction The
manner in which activation spreads means that RT
should be fastest for the 1st sentence, slower
for the 2nd sentence, slowest for the 3rd
sentence.
27
Evidence Sentence verification task (measure
RT) A salmon is a salmon. ( links 0) A
salmon is a fish. ( links 1) A salmon is an
animal. ( links 2) Prediction The manner in
which activation spreads means that RT should be
fastest for the 1st sentence, slower for the 2nd
sentence, slowest for the 3rd sentence.
28
Verification Time as a Function of the Number of
Links from the Activated Node
1500
RT (ms)
1000
0 1 2
Number of Links
29
Evidence Sentence verification task (measure
RT) use properties A salmon needs cold
water. ( links 0) A salmon has gills.
( links 1) A salmon can breathe. (
links 2) Prediction The manner in which
activation spreads means that RT should be
fastest for the 1st sentence, slower for the 2nd
sentence, slowest for the 3rd sentence.
30
Verification Time for Properties as a Function of
the Number of Links from the Activated Node
1500
RT (ms)
1000
0 1 2
Number of Links
31
Evidence Sentence verification task (measure
RT) Prediction The manner in which
activation spreads means that RT should be
fastest for the 1st sentence, slower for the 2nd
sentence, slowest for the 3rd sentence. Predictio
n upheld support for the semantic network theory
32
Breathes
Animal
Skin
Swims
Gills
Fish
Cold blooded
Salmon
Eel
33
Breathes
Animal
Skin
Swims
Gills
Fish
Cold blooded
Eel
Salmon
Original semantic network predicted similar RTs
for all members of a category. (Prediction A
salmon is a fish An eel is a fish)
34
Different theory Feature list model or
Attribute list model (Smith, Rips, Shoben,
1974) Idea Each concept has a list of features
or attributes
35
Different theory Feature list model or
Attribute list model Idea Each concept has a
list of features or attributes Fish
Salmon Eel breathes breathes
breathes skin skin
skin gills gills gills cold
blooded cold blooded cold
blooded swims swims swims pink
flesh long and narrow cold water
no pectoral fins colourful can be
in warm water
36
Different theory Feature list model or
Attribute list model Idea Each concept has a
list of features or attributes To make
verifications First stage one compares the
global features of the two concepts
(e.g., living vs. nonliving). Get a value
(score) for amount of overlap. Low value
quick rejection (no) High value quick
acceptance (yes) Middle value not sure
37
Different theory Feature list model or
Attribute list model Idea Each concept has a
list of features or attributes To make
verifications 1st stage Compare the global
features of the two concepts.
Middle value not sure Go to 2nd stage
Compare defining features of the
concepts. End up with a slow response for
match or mismatch. (Slow yes an eel is a
fish or slow no a dolphin is a fish)
38
Different theory Feature list model or
Attribute list model Predicts fast RTs for
typical members of a category Predicts slow RTs
for atypical members of a category (e.g. A
perch is a fish lt A salmon is a fish lt An eel is
a fish)
39
Verification Time as a Function of Category
Typicality

RT (ms)

High Medium Low (perch)
(salmon) (eel)
Typicality
40
Feature list model good for isa questions,
but not very good with property
questions Typicality is important Cognitive
economy may not be so important (also Conrad,
1972)
41
Revised semantic network model (Collins
Loftus, 1975) connection between typical
category members and the superordinate are
shorter (closer) than the connections between
atypical category members and the
superordinate properties can be represented
more than once (no more cognitive
economy) captures idea of semantic relatedness
42
Breathes
Animal
Skin
Swims
Gills
Fish
Tail fin
Cod
Gills
Trout
Gills
Eel
Gills
43
Breathes
p
Animal
p
Skin
isa
Swims
p
p
Gills
Fish
p
Cold blooded
isa
Perch
isa
isa
Gills
p
p
Salmon
Gills
Eel
p
Gills
44
Semantic Memory Knowledge memory Main
questions How do we gain knowledge? repetition
(form a node?) How is our knowledge represented
and organised in the mind-brain? semantic
network What happens when we access
information? spreading activation