Mental Representation

1
Mental Representation

• Michael Ziessler

2
What is mental representation?

• Central concept in Cognitive Science
• Assumption that the outside world is represented
  in mind
• Mental representation knowledge about the world
• Includes the storage, integration and
  organisation of information in memory

3
Format of mental representation

• Mental representations are not necessarily in the
  same format as the outside world.
• Can be symbolic
• (e.g. mathematics works only with a symbol
  system)
• language as a symbol system

4
Format of mental representation

• It is sufficient if there is a defined
  relationship between the outside world and symbol
  system used for its representation.
• However, there is also evidence for an analogue
  representation imagery

5
Features of mental representation

• Semantic organisation

Clustering of groups of elements that are alike
in meaning Often described as hierarchical
organisation.
6
Hierarchical knowledge structures
EXAMPLE
Not alive
alive
animals
plants
minerals
drinks
Non alcoholic
birds
fishes
flowers
gems
metal
alcoholic
trees
salmon
sparrow
rose
oak
diamond
iron
water
beer
thrush
carp
tulip
birch
sapphire
juice
wine
copper
finch
trout
violet
beech
agate
tin
lemonade
whisky
7
Hierarchical knowledge structures
EXAMPLE
Underwood, Shaughnessy Zimmermann (1974)
Serial learning of word lists
5 - full hierarchical structure, within lowest
level most typical first 4 random order within
each of the subcategories 3 random order of
subcategories (sparrow, trout, finch ) 2
random order of categories on second level
(sparrow, rose, trout ) 1 full random order
8
Hierarchical knowledge structures
EXAMPLE
Underwood, Shaughnessy Zimmermann (1974)
Results
Learning depended on the list structure Increasin
g list structure was associated with a higher
learning rate. ? List structures corresponds to
memory structure, memory structure can be used to
reconstruct the list.
9
Mental representation of concepts

• Different models discussed in the literature
• Exemplar models
• Prototype models
• Feature-comparison models
• Network models

10
Mental representation of concepts

• Exemplar models (e.g. Medin Shaffer, 1978)
• Assumptions
• All exemplars of the category are stored in
  memory
• A new exemplar is recognised as a category member
  depending on its similarity to all category
  members
• A given animal is classified as a DOG because it
  looks similar to all other exemplars stored as
  members of the DOG category.

11
Mental representation of concepts

• Prototype models (e.g. Posner Keele, 1968)
• Assumptions
• From the known exemplars a prototype is derived.
• The prototype is the central tendency of all
  exemplars, includes all features that are
  characteristic for the category.
• - Learning of new exemplars results in an
  update of the prototype.
• A given animal is classified as a DOG because it
  looks similar to the prototype of the DOG
  category.

12
Mental representation of concepts
Experiment by Posner Keele (1968)
Two categories of dot patterns were created, here
shown for one category. Participants were shown
the distorted dot patterns only. Classification
task with feedback. Then test phase with old
items, new items and the prototype. Old items
86 correct New items 67.4 correct Prototype
85.1 correct ? Prototype like old items
13
Mental representation of concepts
Problems with similarity models (probabilistic
models)

• Similarity judgement is based on feature
  comparison
• it depends on which features are selected how
  similar two exemplars are!
• similarity also depends on the context (Tversky,
  1977)

Note only the face in the middle is different
14
Mental representation of concepts
Feature-comparison models (Rips, Shoben Smith,
1974)
Rule-based model of categorisation Defining and
characteristic features are used to decide on
category membership. Characteristic features
allow to account for typicality effects.
EXAMPLE Robin has wings, 2 legs, red breast,
perches in trees, likes worms Meaning is
represented by defining features, characteristic
aspects by characteristic features
15
Mental representation of concepts
Feature-comparison models (Rips, Shoben Smith,
1974)
Statements like a robin is a bird Or An
ostrich is a bird are verified in a 2-stage
process. First stage only characteristic
features, Second stage defining features ?
Explains that it takes longer to decide that an
ostrich a bird. (typicality effects)
16
Mental representation of concepts
Feature-comparison models (Rips, Shoben Smith,
1974)
Problems of the feature-comparison models No
single feature is absolutely necessary A
canary remains a bird even it cannot fly. People
have difficulties to in judging whether a feature
is defining or characterising.
17
Mental representation of concepts
Network models (Collins Quillian, 1969)
Hierarchical theory of concept representation Sub-
concepts have all features of super-concepts plus
differentiating features
Has skin
Can move
ANIMAL
eats
breathes
Has wings
Has fins
FISH
BIRD
Can fly
Has gills
Has feathers
Lives in water
Cant fly
Can sing
CANARY
OSTRICH
Is tall
Is yellow
18
Mental representation of concepts
Network models (Collins Quillian, 1969)
Evidence Time to verify statements like A
canary is a bird. or A canary is an animal, or
A canary breathes. depend on distance between
the nodes in the network.

• Problem
• Typicality effects cannot be explained.
  Additional assumption necessary that the
  associations have different strength.
• - Hierarchy does not always work. Participants
  are faster to decide that an Ostrich an animal
  than a bird.

19
Mental representation of concepts
Basic level concepts (Rosch et al., 1976)
Assumption of a privileged level within
conceptual hierarchies Basic level most
abstract category for which the exemplars
look similar maximum similarity within a
category, minimal similarity between
categories (family resemblance)
EXAMPLE canary BIRD animal CARROT vegetab
le food
20
Mental representation of concepts
Basic level concepts (Rosch et al., 1976)
In conceptual identification first identification
occurs on basic level. Level that is most often
used. There are common motor programs to interact
with all exemplars.
Ziessler (1983) basic level depends on quality
of features used to represent the concepts Basic
level most abstract concepts with smallest
amount of common sensory features.
Sub-concepts need more sensory
features Super-concepts need functional features
21
Mental representation of concepts
Conceptual system
Basic level concepts
VEHICLE
Functional features
Sensory features
CAR
FERRARI
? First and fastest identification of an object
at the basic level. More specific and more
abstract identification need more time.
22
Semantic networks
Concepts in memory are linked with each
other Concepts are the nodes, semantic relations
between the concepts the links of a network
Collins Loftus (1975) Spreading activation
model
23
Semantic networks

• Collins Loftus (1975)
• Spreading activation model
• Empirical evidence from priming experiments
• fragment completion
• stem completion
• lexical decision
• There are 2 types of priming
• repetition priming evidence for activation of
  the nodes
• semantic priming evidence for the spreading of
  activation
• over the relations between nodes

Problem of the model control of activation?
Specification of links?
24
Distributed representations
Parallel Distributed Processing (PDP
networks) McClelland, Rumelhart Hinton
(1986) connectionist networks Assumption A
concept is represented across multiple nodes -
activation pattern of nodes in a network
Local representation
Distributed representation
Vanilla
Vanilla
Chocolate
Chocolate
Coffee
Strawberry
Strawberry
Tutti-frutti
Coffee
25
Distributed representations
Parallel Distributed Processing (PDP networks)
OUTPUT units
hidden units
INPUT units
26
Distributed representations
Parallel Distributed Processing (PDP
networks) All units are connected with each
other. Learning changes the strength of the
connections. Learning activation pattern of the
input units has do be transformed in a required
activation pattern of the output units.

• Advantage of the system
• learning
• partial damage can be compensated
• new input patterns lead to the activation of the
  most similar output

27
Analogue representations
Assumption that visual information is coded in
form of an internal picture in memory More
general storage of information as storage of the
experienced perception

• Dual-coding hypothesis
• Conceptual-propositional hypothesis
• functional-equivalence hypothesis

28
Analogue representations

• Dual-Coding Hypothesis
• Paivio (1965)
• Some words are more imaginable than others
• easy e.g. elephant, tree, car
• difficult e.g. context, democracy, tool
• Familiar easily named object is coded by a verbal
  system and an imagery system, abstract concept
  can only be verbally coded.
• Memory better for specific words than for
  abstract words
• Imagery technique during encoding of words
  improves memory

29
Analogue representations
Conceptual-propositional hypothesis Anderson
Bower (1973) There is no internal picture who
should read it? Pictorial information is stored
in another format. Sensory facts are activated by
the concept results in a more detailed memory
trace
30
Analogue representations
Functional-equivalence hypothesis Shepard
Metzler (1971) Some processes in memory seem to
require a coding structure that is isomorphic to
physical structures! Comparison of 2 complex
figures- are they the same? One figure is
rotated. Decision time depended linearly on the
degree of rotation assumption that participants
mentally rotate the figures to bring them in the
same position.
31
Analogue representations
Functional-equivalence hypothesis Example
Wohlschläger Wohlschläger (1998)
Comparison between mental and manual rotation
32
Analogue representations
Functional-equivalence hypothesis Example
Wohlschläger Wohlschläger (1998)
Results
33
Analogue representations
Functional-equivalence hypothesis Finke (1980)
34
Analogue representations
Functional-equivalence hypothesis Finke (1980)
Perception and imagery activate the same
perceptual processing mechanisms functional
equivalence
35
Mental representation and behaviour
Function of cognition does not consist in the
generation of an objective and complete
representation of the outside world in
memory. Function is to control the
behaviour. Knowledge depends on what is necessary
for the control of behaviour. Behaviour
determines how we perceive the world. Same object
can be perceived differently in two different
behavioural contexts! Probably the structure of
behavioural programmes determines the structure
of our memory.