Cog5 lecppt chapter08

© 2010 by W. W. Norton & Co., Inc.
Concepts and Generic Knowledge
Chapter 8
Lecture Outline

Chapter 8: Concepts and Generic Knowledge
 Lecture Outline
 Definitions
 Prototypes and Typicality Effects
 Exemplars
 Difficulties with Categorizing via Resemblance
 Concepts as Theories

Definitions
 Concepts like dogs or
chairs
 Building blocks
 Simple but complex to
explain

Definitions
 Dog
 Definition
 A mammal with four legs that barks and wags its
tail
 Exceptions
 Dog that does not bark or that lost a leg
 For any definition, we can always find such
exceptions

Definitions
 Philosopher Ludwig Wittgenstein (1953)
 Simple concepts have no definition
 Consider a “game”
 Played by children
 Engaged in for fun
 Has rules
 Involves multiple people
 Is competitive
 Is played during leisure
 For any set of definitive features, we can think
of exceptions that are still considered games.

Definitions
Definition Exception
Played by children Gambling?
Engaged in for fun Professional sports
Has rules Playing with Legos
Involves multiple people Solitaire
Is competitive Tea party
Is played during leisure Flying simulators
Games

Definitions
 Family resemblance
 members of a category
have a family
resemblance to each
other
Ideal member
Atypical member
In the example, dark hair, glasses, a
mustache, and a big nose are typical for
this family but do not define the family.

Definitions
 A dog probably has four legs, probably
barks, and probably wags its tail
 A creature without these features is
unlikely to be a dog

Definitions
 There may be no features that are shared
by all dogs or all games, just as there are
no features shared by every member of a
family
 The more characteristic features an object
has, the more likely we are to believe it is
part of the category

Prototypes and Typicality Effects
 Rosch’s prototype
theory,
 Prototypes
 Rather than thinking
about definitions that
define the boundaries
of a category
 One that possesses all
the characteristic
features

 Prototype
 An average of various category members that
have been encountered
 Differ across individuals (depending on their
experiences)
 May differ across countries
 For example, the prototypical house in the United States
compared to Japan

 Prototypes
 Graded membership
 Some members are closer to the prototype
 Fuzzy boundaries
 No clear dividing line for membership

 Which is the best red?

Evidence Favoring the Network Approach
 The sentence-verification task. 
typicality effects
 True or false?
 Robins ( 知更鳥 ) are birds
 Penguins are birds
 This is because robins share more features
with the prototypical “bird” than penguins do.

 using production tasks
 Typicality effects
 Name as many fruits as possible
 Name as many birds as possible
 If we ask people to name as many birds as
they can, they typically start with category
members that are closest to the prototype
(e.g., robin). For fruit they are likely to start
with bananas, apples, or oranges.

Does this picture show you a bird?
[Insert typical bird]
[Insert a penguin]
Faster
Slower
picture-identification tasks

 The more
prototypical
category
members are
also “privileged”
in rating tasks

Birds in a tree?
Not thisImagine this

 Typicality also influences judgments about
attractiveness. Which fish is the most attractive?

 Just as certain
category members
seem to be privileged,
so are certain types of
category
 For example, what is
this object?

Furniture
Chair
Upholstered armchair
Too general
Just right
Too specific
DetailExample
Rosch argued that there is a basic level of categorization that is neither too
general nor too specific, which we tend to use in speaking and reasoning about
categories
Here, “chair” is the basic-level category, as opposed to “furniture” (more
general, or superordinate) or “wooden desk chair” (more specific, or
subordinate)

 Basic-level categories
 Single word.
 The default for basic level
 Easy-to-explain commonalities

 Basic categories are learned first
 Used by children to describe most objects

Exemplars
 Exemplar
 What is this?
 An alternative to prototype theory is exemplar-
based reasoning—drawing on knowledge of
specific category members rather than on more
general, prototypical information about the
category.

Exemplars
Theory Prototype Exemplar
Typicality Average of a category Encountered more often
Graded membership Less similar to average How often it is encountered
Illustration Ideal fruit (apple)
vs. less ideal (fig 無花果 )
Apples (often)
vs. figs (not as often)
Both prototype theory and the exemplar view can explain the typicality and graded-
membership effects that we have discussed.

Exemplars
 Prototypes
 Economical but less flexible
 Exemplars
 More flexible but less economical
 Chinese versus American Birds
 A gift for a 4-year-old who recently broke her wrist
 Our ability to “tune” our concepts to match
circumstances may also fit better with the
exemplar view than with prototype theory.

Exemplars
 Kermit the Frog
 Prototypical features
 Is green, eats flies
 Exemplar (unique)
 Sings, loves a pig
Both prototype and exemplar provide information
In sum, the evidence seems to suggest
that we use a combination of prototypes
and exemplars.

Exemplars
 Every concept is a mix of exemplar and
prototype
 Early learning involves exemplars
 Experience involves averaging exemplars to
get prototypes
 With more experience, we can use both

Difficulties with Categorizing via Resemblance
 Category membership and typicality
 Prototypes that are based on averaged
exemplars
 A process of triggering memories
 This is because both judgments should be
based in resemblance between the test case
and the prototype or exemplar.

 Typicality and
category membership
sometimes dissociate
 Moby Dick （白鯨）
was a whale （鯨
魚） but not a typical
one

The category is clear and yet typicality goes down

 Atypical features do not exclude category
members
 For example, a lemon that is painted with red
and white stripes, injected with sugar to make
it sweet, and then run over with a truck is still
a lemon

 All the typical features
but not category
members
 For example, a perfect
counterfeit bill.

 Similar examples come from studies with
children (Keil, 1986)
 A skunk （臭鼬） cannot be turned into a
raccoon （狸）
 It has a raccoon mommy and daddy…
 A toaster can be turned into a coffeepot
 Just need to poke some holes in it…

 Essential properties
 Those that define a category
 Which are those?
 some categories are reasoned about in terms
of essential properties and not superficial
attributes; for example, the abused lemon still
has lemon DNA; it still has seeds that would
grow into lemon trees

Concepts as Theories
 Resemblance
 Prototypes and exemplars work
 categorization is based in comparing the
resemblance of the test case to prototypes and
exemplars
 Not enough
 Perfect counterfeit bill resembles a bill but is
not

 Heuristic （捷思）
 A reasonably efficient strategy that works
most of the time
 the resemblance of more superficial features is
compared
 Prototypes and exemplars
 Heuristics allow some degree of error in
exchange for efficiency

 When heuristics fail, may need a more
complete view
 Concept-as-theory

whipped cream airplanesReal airplanes resemble

 Concepts are like schemas
 They allow people to form generalizations
 Related to typicality
 Generalizations more likely from typical cases
 Robins are more likely to be like all birds
 Penguins are less likely
 Research in this area shows that people are willing
to make inferences from a typical case (e.g.,
robins) to an entire category (e.g., birds) but not
from an atypical case (e.g., ducks).

 Theories also explain cause and effect
Lion Gazelle （羚羊）
EnzymeEnzyme
For instance, if told that gazelles have a particular enzyme, people
conclude that lions have it as well. But they are not willing to make the
reverse inference, given what they know about the food chain.

 Natural kinds and artifacts are reasoned
about differently
 Natural kinds (e.g., the skunk and raccoon)
have essential properties
 These principles do not apply to artifacts (e.g.,
toaster and coffeepot)

43
Categories represented
in different brain areas
different sites are activated
when people are thinking
about living things than
when they are thinking
about nonliving things (e.g.,
Chao et al., 2002).

Knowledge Network
 Knowledge is represented via a vast
network of connections and associations
between all of the information you know

Knowledge Network
 Other evidence for the knowledge
representation in a network comes from
the sentence-verification task
 Participants must quickly decide whether
sentences like the following are true:
 Robins are birds.
 Robins are animals.
 Cats have hearts.
 Cats are birds.

Knowledge Network
 “Cats have hearts” requires two links
 “Cats have claws” requires one link

Knowledge Network
Reaction time goes up for
longer associative paths
The time to answer these questions
depends on the length of the associative
path between the pieces of information
(Collins & Quillian, 1969).

Knowledge Network
 Nodes can represent concepts
 Links such as hasa or isa can associate
each concept

Knowledge Network
Proposition = smallest unit that can be true or false
Four propositions about dogs
A more complex network (Anderson’s ACT)
is designed around the notion of
propositions—the smallest units of
knowledge that can be true or false.

Knowledge Network
Abstract knowledge represented via time and location nodes

Knowledge Network
 Propositional networks
 Localist representations—each node is
equivalent to one concept
 Connectionist networks (parallel
distributed processing, PDP)
 Distributed processing—information involves
a pattern of activation
 Parallel processing of information occurs at
the same time

Knowledge Network
 How does learning take place in a
connectionist or parallel distributed
processing (PDP) network?
 Changes in the connection weights or
strength of connections

Knowledge Network
 Learning algorithms—how weights are
changed
 Both nodes firing together strengthen their
connection
 Error signals cause a node to decrease its
connections to input nodes that led to the
error (back propagation)

Concepts
 In sum, concepts are central to human
reasoning, but are complex
 We often reason about concepts using
prototypes and exemplars, particularly in cases
where fast judgments are required
 However, for more sophisticated judgments, we
also employ theories, represented by networks
of interrelated conceptual knowledge
 Finally, various computational networks have
attempted to capture this complexity

1. According to Wittgenstein,
a) we have no real general concept for each
category we know but instead learn each
category member individually.
b) we assess category membership
probabilistically, by family resemblance.
c) we can find rigid features that define a
category but only after intensive study.
d) we first encounter the prototypical
member of a category, and then we
compare all other potential members to it.

2. Which of the following facts fits with the claims of
prototype theory?
a) Pictures of items similar to the prototype are
identified as category members more quickly than
pictures of items less similar to the prototype.
b) Items close to the prototype are not the earliest
(and most likely) to be mentioned in a production
task.
c) When making up sentences about a category,
people tend to create sentences most appropriate
for the prototype of that category, as opposed to a
more peripheral member.
d) all of the above

3. Which of the following claims is TRUE?
a) Reliance on prototypes is likely to emerge
gradually as a participant’s experience with
a category grows.
b) People are likely to rely strongly on
prototypes early in their exposure to a
particular category.
c) People only rely on prototypes when they
have time to make a decision.
d) With exposure to many instances of a
particular category, it becomes easier to
remember each particular instance, and this
contributes to the emergence of a prototype.

4. Which of the following is true?
a) People only use prototypes when there
are no clear definitions to fall back on.
b) Just because people use prototypes does
not mean that is the only information
available to them.
c) People use exemplars rather than
prototypes whenever possible.
d) Clearly defined category boundaries are
necessary for deciding category
membership.

5. Which of the following is true about
heuristics?
a) One way to ensure error-free decisions is
to use the typicality heuristic.
b) One example of a heuristic is determining
cause and effect.
c) The categorization heuristic emphasizes
superficial characteristics.
d) Using heuristics is an inefficient way to
get things done.

6. In a production task, the ___ category
members that a person mentions are the
category members that produce the
slowest reaction times in a sentence-
verification task.
a) first
b) last
c) loudest
d) slowest

7. The idea that we categorize objects
based on their similarity to previously
stored instances is known as
a) geometric theory.
b) prototype theory.
c) feature theory.
d) exemplar theory.

Cog5 lecppt chapter08

More Related Content

What's hot (20)

Viewers also liked (14)

Similar to Cog5 lecppt chapter08 (20)

More from Visual Cognition and Modeling Lab (20)

Recently uploaded (20)

Cog5 lecppt chapter08

Editor's Notes