Semantic organization
• Rosch and others have argued that our
categorization of the world is not an arbitrary
historical accident, but reflects our
psychological makeup, and hence is subject
to investigation
Semantic organization
– Berlin & Kay (1969) investigated colour names
across 100 different languages
– order and frequency of colours used is consistent
across cultures:
black white
red
green yellow
blue
brown
purple pink orange grey
Semantic organization
I.e. If there are two words for colours they tend to
be black and white; three black, white, and red;
etc.
Semantic organization
– Rosch-Heider (1972) experiment with Americanspeaking subjects and members of the Dani, a
stone-age New Guinea tribe
– Dani only had words for black and white
Semantic organization
– Rosch built on a study by Brown & Lenneberg
showing that the Zuni, whose language
categorizes colours different from Englishspeakers, and North American English speakers
tend to remember focal colours (e.g., pure green)
better than nonfocal (e.g., purple)
– Focal colours are colours that are selected to be
good exemplars of particular colours
Semantic organization
– Experiment 1.
showed a single coloured chip, and were required
to recognize it from a set of 160 chips
both U.S. and Dani subjects performed better with
focal colours
Semantic organization
– Experiment 2
Dani required to associate different colours with
clan names; did better with focal colours than with
non-focal colours
– Conclusion
same colours were focal for Dani as for US
subjects
therefore it is not language that makes certain
colours easier to remember, but their perceptual
salience
Semantic organization
• Why do people form categories?
(Rosch,1978)
– Cognitive economy; want to obtain as much
information from the environment as possible
with the least effort
– the perceived world is a structured world; our
perceptions shape the concepts that we form
Semantic organization
• Structure of categories
– categories have a horizontal and a vertical
dimension
horizontal--segmentation of categories at the same
level of inclusiveness (e.g., sugar maple, silver
maple)
vertical--different levels of inclusiveness
Semantic organization
Superordinate
Basic
Subordinate
Furniture
Chair
Kitchen chair
Office chair
Lamp
Floor lamp
Office lamp
Tree
Maple
Sugar maple
Silver maple
Semantic organization
– Rosch argues that the basic level of organization
is the most useful level for many purposes
because it provides the most information for the
least effort
Semantic organization
– evidence to support hierarchical distinction
– common attributes experiment
presented 9 taxonomies (e.g., tree, bird, fish, fruit,
musical instruments, furniture, vehicle) at 3 levels
participants were instructed to list all of the
attributes they could think of that were true of the
items listed
few attributes at the superordinate level;
significantly more at the basic and subordinate
levels
Semantic organization
– evidence to support hierarchical distinction
– motor movements
subjects were presented same materials as in
previous study, and were asked to describe motor
movements
basic objects were the most general classes to
have motor sequences in common
– similarity of shape and identifiability of averaged
shape were other lines of evidence to support
hierarchical distinction
Semantic organization
• How should concepts be represented?
• Classical theory
– what specifies a concept is some combination of
semantic features (e.g., bird -- has feathers,
wings, lays eggs, has a beak, etc.)
– this model has been formally developed by
Collins & Quillian, and Smith
Semantic organization
• How should concepts be represented?
• Classical theory
– problem is that many naturalistic concepts (birds,
fruits, games, tools, etc.) are not rigidly defined
– not all birds fly, not all games involve more than
one person, are competitive etc.
– Wittgenstein argued that family resemblance may
be a more useful way to think about category
membership
Semantic organization
• How should concepts be represented?
– The idea of family resemblance leads to the idea
that category membership is not determined by
rigidly defined categories but by resemblance to
a typical member
– Rosch asked subjects to rate basic level words
as being typical or atypical of a category (e.g.,
robin, ostrich, chicken)
Semantic organization
• How should concepts be represented?
– results: subjects were very consistent in their
responses (i.e, robin rated as typical)
– subsequent study showed that verification was
faster as well (robin is a bird is faster than
chicken is a bird) for typical than for atypical
categories
– Rosch showed that typical instances had many
features in common with other members of the
category
Semantic organization
• Semantic relatedness is a general finding in
this literature
– prototypical members of a category are verified
quickly
– related negative instances of a category are
verified more slowly (e.g., potato is a tree takes
longer to verify than does rifle is a tree) Kintsch,
1980
– comparison process seems to be critical; not a
simple category search
Semantic organization
True-False latency as a function of relatedness
1250
1200
1150
Mean RT 1100
(msec) 1050
1000
True
False
950
900
Related
Unrelated
Semantic organization
• Feature comparison models--Smith, Shoben,
& Ripps (1974)
– model assumes that concepts are represented by
bundles of features, separated into those that are
defining, and those that are characteristic
e.g., bird -- defining -- feathers, lays eggs
characteristic -- flies, two legs, migrates
Semantic organization
• Verify a sentence e.g., a robin is a bird
– model postulates that subject retrieves features
associated with robin and with bird; if there is a
high degree of overlap respond yes
– if there is less overlap begin a second slower
stage in which the defining features are
compared; if there is overlap respond yes; if
there is a mismatch respond no
Semantic organization
• Semantic network theories
– Collins & Quillian
– hierarchical memory structure model (see page
261 Reed)
– critical assumptions: cognitive economy and a
hierarchical model
features that are true of all animals such as eating
and breathing are stored at the highest level
Semantic organization
• Semantic network theories
– prediction: takes longer to respond to a true false
question the further away the two types of
information are stored
Collins & Quillian 1969
Type of
judgement
Level 0
Level 1
Level 2
Property
A canary can
sing
A canary can
fly
A canary has
skin
Category
A canary is a
canary
A canary is a
bird
A canary is an
animal
Collins & Quillian, 1969
Reaction time to verify property and category
statements
Mean RT msec
1500
1400
1300
1200
Property
Category
l
1100
1000
900
Level 0
Level 1
Level 2
Collins & Quillian, 1969
• Results were consistent with the hierarchical
model with cognitive economy
• However, Conrad (1972) showed that if you
control for relatedness, the level effect
disappears
• also model has difficulty accounting for
typicality effects of Rosch
More recent semantic networks
• Spreading activation model of Collins and
Loftus
– see your text
More recent semantic
representations
• Schemas, frames, and scripts
– in 1932 Bartlett proposed that people remember
new material in terms of existing structures of
knowledge that he dubbed schemas or schemata
– schemas represent some aspect of the
environment, or our experience, or beliefs
– learning was conceptualized as an active
process in which people attempted to make
sense of what they had experienced
– effort after meaning
More recent semantic
representations
• Schemas, frames, and scripts
– Bartlett studied effects of schemas on memory by
investigating memory for a North American folk
tale (structured but unfamiliar material)
– showed that the students tended omit material
that was strange to them or to distort it in ways
that fit their expectations
– criticism--model too vague to be testable
More recent semantic
representations
• Schemas, frames, and scripts
– with the advent of computers and the cognitive
approach to psychology scientists have begun to
actively investigate these knowledge structures
– Minsky, Rumelhart, Schank, Abelson, Kintsch,
Anderson
More recent semantic
representations
• Schemas, frames, and scripts
– characteristics of this approach
this type of knowledge structure enables people to
make sense of partially observed or described
situations
e.g., the man bought a candy bar. People typically
would infer that in money was given in exchange
for the candy bar
e.g., the man drove in a nail
More recent semantic
representations
• Schemas, frames, and scripts
– characteristics of this approach
schemas have variables: buying something in a
store; knowledge structure represents that it
entails an exchange of money for a good; however,
the amount of money or the good is left
unspecified
hammering: there is a tool (hammer), an object or
recipient of the action (nail), an action (hammering
motion), and an agent or person
More recent semantic
representations
• Schemas, frames, and scripts
– characteristics of this approach
schemas can embed within each other
schemas operate at many levels of abstraction
schemas represent knowledge of belief
schemas are active recognition devices
More recent semantic
representations
• Schemas, frames, and scripts
Thorndyke (1977) studied the role of story
structure on recall
original version had a theme and then a narrative
that elaborated the theme
version 2: narrative then theme (after theme)
version 3: narrative no theme
version 4: randomly ordered
Thorndyke 1977
Recall as a function of organizational structure
100
% Recall
80
story
after-theme
no theme
random
60
40
20
0
1
2
3
Level in hierarchy
4
Thorndyke (1977)
• Conclusions
– level of recall depends upon
degree of structure provided in the story
level of importance of the information (hierarchy
level)
these two factors interact. Importance of
information is evident only in structured stories
Schank: scripts
• Schank and Abelson hypothesized that we
have developed scripts that represent
commonly experienced social events
– e.g., going to a restaurant
– e.g., going to a bank, taking a bus
Schank: scripts
• Restaurant script
Props: restaurant, tables, menu, food, bill, money,
tip
Agents: customer, waiter, cook, cashier, owner
Entry conditions: customer hungry, customer has
money
Results: customer has less money, owner has
more money, customer is not hungry
Schank: scripts
• Restaurant script
– Scene 1: entering
customer enters restaurant
customer looks for table
customer decides where to sit
customer goes to table
customer sits down
– Scene 2: ordering
– Scene 3: eating
– Scene 4: exiting
Neuropsychology of semantic
memory
• Visual agnosia (Lissauer, 1888)
– GL sustained a blow to the head
– complained of difficulty seeing
examination showed normal visual acuity
normal ability to copy objects
recognition of objects was severely impaired; but
it was not a general deficit; e.g., unable to
recognize a whistle when presented visually, but
able to recognize a whistle from its sound
Neuropsychology of semantic
memory
• Tactile agnosia (Beauvais, 1978)
– patient unable to recognize objects to touch, but
could recognize objects when they were
presented visually
– also patient was able to use objects appropriately
• these results suggest that semantic memory
is not a single unitary system, but has a
number of subcomponents associated with
the modality of input
Neuropsychology of semantic
memory
– Warrington & Taylor (1978) showed that subjects
with brain injury made two types of semantic
errors in the visual modality
access disorder--some subjects had difficulty
recognizing a picture of an object (e.g., tennis
racquet)
degraded semantic store--other subjects
recognized the object, but had difficulty
recognizing which object was commonly
associated with the object (e.g., a tennis ball)
Neuropsychology of semantic
memory
• Warrington & Shallice (1979) proposed the
following criteria to distinguish access versus
degraded semantic store impairments
 consistency--if deficit is degraded semantic store, there
should be consistency across test sessions (and type of test,
Bayles)
 On the other hand if the problem is one of access, then one
might expect that different types of retrieval cues might lead
to retrieval of the item
 priming--patient should not show priming effects if there is a
degraded store; however, certain primes might facilitate
access to items if the problem is one of access
Neuropsychology of semantic
memory
• Structure of semantic memory: modality
specificity or a single semantic store?
– one view holds that semantic memory consists of
a single amodal system
– second view hypothesizes that there are
separate systems for verbal, visual, and other
types of information
– the evidence at this point is not yet entirely clear
on this point
Neuropsychology of semantic
memory
• How are other types of information
represented in semantic memory?
– Some evidence suggests that evaluative
information is processed and stored in a different
location than denotative information
Case Description of AM
 Successful businessman prior to TBI
 Average to very superior general intellectual
functioning
 Normal academic, attention, and executive
function abilities
 Generally intact memory abilities
 Poor social judgment; everything is positive
Park et al. (2001) Neuropsychologia
R. Temporal
Temporal
a
L. Amygdala
b
Amygdala
Frontal
c
d
Attitude Priming Study of AM
 Purpose: to investigate AM’s evaluative rating
of words
 Hypothesis: impaired automatic evaluation of
negative but not positive evaluative stimuli
Park et al. (2001) Neuropsychologia
Attitude Priming (continued)
 Method: attitude priming paradigm
– Participants: AM and 8 age - and education matched controls
 Procedure:
– Phase 1: rate single words as “good” or “bad”
– hypothesized positivity bias
Park et al. (2001) Neuropsychologia
Rating of Words in Phase 1
Number Rated "Good"
80
70
60
50
Controls
AM
40
30
20
10
0
Controls
AM
Response Latency to Phase 1
Words
3000
2500
Msec
2000
Negative
Positive
1500
1000
500
0
AM
Control
Phase 2
prime
(pos or neg)
250 ms
blank
screen
50 ms
target
(pos or neg)
Task: rate target as good or bad as quickly as possible
Control Priming Results Phase 2
800
Msec
760
Pos. Prime
Neg. Prime
720
680
640
Positive Target
Negative Target
AM Priming Results Phase 2
4000
3500
Msec
3000
2500
Pos. Prime
Neg. Prime
2000
1500
1000
500
0
Positive Target
Negative Target
Summary of Attitude Priming
 Positivity bias in rating single words
 Slowed responses only to words rated as bad
 Priming in positive valence condition only
 Conclusion: AM can automatically access
positive but not negative evaluative
information
Park et al. (2001) Neuropsychologia
Connotation Generation Study of AM
 Purpose: to determine whether AM could
access negative evaluative information when
directed
 Task: describe two positive and two negative
features of single words (e.g., coffee)
 Same 92 words used as primes in
Experiment 1
Park et al. (2001) Neuropsychologia
Acceptable Good and Bad
Connotations
Acceptable Connotations
0.9
0.8
0.7
0.6
0.5
Good
Bad
0.4
0.3
0.2
0.1
0
AM
Control
Semantic priming and AM
• Purpose of experiment
– to determine whether AM would show normal
semantic priming
– prior research has shown that the latency to
respond to a target is facilitated when the prime
preceding the target is semantically related
compared to when it is unrelated
• Method
– similar to Phase 2 of the first study
Semantic priming and AM
• Method
– similar to Phase 2 of the first study
– task: show prime-then target; make a lexical
decision about target item (word/nonword)
Semantic priming and AM
Mean response latency
(ms)
Mean response latency (ms) to semantically related
and unrelated word pairs
1000
900
800
700
600
500
400
300
Related
Unrelated
AM
Control
Conclusions
• Conclusions
– AM impaired in his automatic processing of
negative evaluative information
positivity bias
no priming for negative evaluative words
– AM not impaired in his denotative or semantic
processing of words
– suggests a dissociation between these two
aspects of semantic memory