Concepts! Categories!

• A concept is simply an idea in your • A category is a collection of things

mind.! in the world, things that correspond
• Concrete ideas like “dog” and “fur”! to a concept in your head.!
• Abstract ideas like “space travel” • All the actual dogs in the world form
and “freedom”! the category of dogs.!
• Actions like “walking” and “rolling”! • All the instances of space travel.!
• The key is that concepts are in your • All instances of walking!
head, not things in the world.!

Plato (ca. 428-348 B.C.)! Aristotle (ca. 384-322 B.C.)! Exemplars/Instances!

Founder of Rationalism! Founder of Empiricism!

• Plato: objects in the world are imperfect versions of the

perfect ideas in our heads!

• Aristotle: objects in the world are the true reality?!

• A concept is an idea, and a category is the set of imperfect

things in the real world that correspond to that concept.!

• So this is a combination of Plato’s ideas and Aristotle’s ideas!

• “Exemplars” and “instances” are terms used to mean the
specific members or examples of a category.!

Networks of Concepts! Networks of Concepts!

Fire Engine!
Truck!

Lips! Red! Apple!

Mouth!
Sweet!

• The concepts in your mind are organized into a

network in which related concepts are connected
• A dictionary defines a word by linking it with other words. !
together. !
• This creates a problem: If you don’t know any of the
• If one concept is activated, this spreads to other
words, you can’t understand any of the definitions.!
related concepts.!
• So how can our concepts actually mean something?!

© S. J. Luck

All rights reserved
1

 Networks of Concepts! Networks of Concepts!

• This problem is solved by linking some concepts

House!
to sensory and motor representations. !
Fire!

Fire Engine!
• For example, when you see the word APPLE, this
Truck! leads to: !
!-activation of the color red in visual cortex !
!-the taste of an apple in the gustatory
Lips! Red! Apple! !cortex !
!-the representation of chewing in motor
!cortex.!
Mouth!
Sweet!

Brain in a Vat!

• This same idea is at the heart of the Matrix movies,

where people believe they are in a normal world when
in fact they’re all in a state of suspended animation,
• Philosophers have developed the idea that our brains connected to computers that create their sense of
could be pulled out of our bodies and put into a vat. ! reality!

• With the right stimulation, we would think that we

were walking and talking.!

Brain in a Vat!

• On the other hand, researchers are hard at work

• In practice, this would be much harder than people creating brain-computer interfaces, which allow people
think, because our thoughts are very closely tied to to control computers and robotic arms via electrodes
our sensory inputs and motor outputs. ! implanted in the brain.!

• If the inputs and outputs aren’t just perfect, we’d

notice it right away.!

© S. J. Luck

All rights reserved
2

 Brain in a Vat! Networks of Concepts!
• Key idea: thought isn’t as abstract as philosophers like
Plato thought. !

• Our thoughts are intimately tied to our sensory

experiences and our ability to act on the world.!

• The general idea that thinking is closely tied to our

sensory inputs and motor outputs is called “embodied
cognition.”!

• It’s not actually enough just to have links between concepts.

The links need to specify relationships.!

• Bear is related to Mammal and Fur, but that a bear is a mammal

and has fur.!

Networks of Concepts!
T
Contains Contains

Vertical Horizontal
Line Line

Supports

Bisects
• The idea of a conceptual network was developed by
• This is analogous to structural description theories of cognitive psychologists and artificial intelligence
object perception, which specify the parts of an object researchers back in the 1960s.!
and the relationships between the parts.!
• Google has recently taken this idea and used it to help
• But with a conceptual network, we have relationships organize the mountains of information they extract
between concepts rather than relationships between from the internet.!
parts.!

Categorization: The Classical View!

The Process of Categorization!
• 4 types of theories about how humans perform
categorization:!

§ The classical view: Necessary and sufficient features!

§ Prototype-based categorization!
§ Exemplar-based categorization!
§ Theory-based categorization!
• Classical view goes all the way back to Aristotle.!

• A category is defined by a set of features that are “singly

necessary and jointly sufficient” to constitute category
membership!

• A classic example in philosophy is a bachelor!

© S. J. Luck

All rights reserved
3

 Categorization: The Classical View!

• Bachelor: Unmarried, adult, male, human!

• If someone has all of these features, then that person is

a bachelor.! • Problem: many categories can’t easily be defined by a
set of necessary and sufficient features.!
• If someone lacks one or more of these features, then
that person is not
Bachelor: a bachelor.!
Unmarried, adult, male, human! • Classic example: the category of “game.”!

Necessary features: Must have all of these

features or it’s not a game!
• Hard to find a set
Sufficient of necessary
features: and
If it has sufficient
these features
features,
that includes all then
games and excludes
it must everything
be a game ! else.!

• What would be the set of features that are both

necessary and sufficient to include Candy Land?! • These same features seem to fit hide-
and-seek…!
• Fun!
• Clear Rules!
• Sudoku is a single player game that does
• Players compete against each other! not require competition.!

• Are all of these features necessary for it to be a game?! • Some people think math is “fun” and it
! also has clear rules, but it is not a game.!
• And are they sufficient, so that something will definitely
be a game if it has these features?! • Riding a roller coaster is fun, and there
are rules like keeping the seat belt on,
but it’s not a game.!

• Another example: the category of sports.!

• Volleyball is clearly a sport. !

• It has teams, a ball, requires athleticism, and involves • Car racing is also a sport, but doesn’t have many
competition. ! features in common with volleyball.!

© S. J. Luck

All rights reserved
4

 • Rock climbing is also a sport. !

• It involves athleticism, but not teams or competition.!

• Fishing is a sport but doesn’t require teams, doesn’t use a
• Rhythmic gymnastics is a sport that’s done at the
ball, doesn’t require much athleticism, and doesn’t usually
Olympics. !
involve competition.!
• Like volleyball, it involves teams, athleticism, and
• Still a sport when it’s done for fun.!
competition. !

• Main point: it’s difficult to define a set of necessary

and sufficient features for many categories, such as
games and sports.!
• However, dancing is not a sport, even though it has
many features in common with rhythmic gymnastics.!
• This is a shortcoming of the classical view of
categories.!

Categorization: Prototype Theory! Categorization: Prototype Theory!

• In the 1960s, cognitive psychologists proposed a

different way of thinking about categories. ! • Average of birds you’ve experienced might look like this !

• They realized that the boundaries between categories are • Not any particular bird, and not an average of all birds
often hard to define, and they proposed that we instead that exist. !
represent the center or average of a category. !
• It’s an average of the birds that you’ve encountered.!
• The center of a category is represented by what’s called
a “prototype” of that category.!

© S. J. Luck

All rights reserved
5

 Prototype! Prototype! Exemplar 1! Exemplar 2! Exemplar 3! Exemplar 4!

Category Category
A! A!

Category Category
B! B!
Mike Posner! Steve Keele! • Next step was to create exemplars for
each category.!
!
Category
C!
• Began with 4 prototypes, one for each Category
C! • Each exemplar was made by taking
category. ! each dot from the prototype and
shifting it by a small random amount.!
• Random sets of dots, so the subjects
Category
D!
wouldn’t have any prior experience with Category
D!
the specific categories.!

Posner & Keele (1968)! Posner & Keele (1968)!

Prototype! Exemplar 1! Exemplar 2! Exemplar 3! Exemplar 4! Prototype! Exemplar 1! Exemplar 2! Exemplar 3! Exemplar 4!

Category Category
A! A!

• The two circled dots from the

Category
B! prototype are slightly shifted in each Category
B!
of the exemplars.!

• In the experiment, each dot in the

Category Category
C! prototype was at the mathematical C!
average of the corresponding dot in
each of the four exemplars.!
Category Category
D! • The prototype was truly an average of D!
the exemplars.!

Posner & Keele (1968)! Posner & Keele (1968)!

Prototype! Exemplar 1! Exemplar 2! Exemplar 3! Exemplar 4! Learning Phase:!

Category
Which category is it (A, B, C, or D)?!
A!

Category • During the learning phase, subjects would see one of the
B! exemplars and indicate whether it belonged to category
A, B, C, or D.!

Category • Then they were given feedback indicating whether their

C!
response was correct or incorrect.!

• They saw each exemplar many times so that they would

Category learn the categories.!
D!

• Initially, subjects just guessed. But eventually they

started learning and got up to about 80% correct. !
Posner & Keele (1968)!

© S. J. Luck

All rights reserved
6

 Test Phase: Original exemplars, new exemplars, and prototypes! Test Phase: Original exemplars, new exemplars, and prototypes!
Which category is it (A, B, C, or D)?! Which category is it (A, B, C, or D)?!

• During the test phase, subjects were shown:!

• the original exemplars!
• new exemplars!
Original exemplars: 80% correct!
• the prototypes!
New exemplars: 50% correct!

Posner & Keele (1968)!

Prototypes: 68% correct! Posner & Keele (1968)!

Prototypes! Prototype! Exemplar 1! Exemplar 2! Exemplar 3! Exemplar 4!

• When the subjects were tested with the Category

A!
prototypes, they were 68% correct, even
though they had never seen the prototypes
before.!
Category
B!
• This was significantly better than the 50%
correct for the new exemplars.!
Category
C!
• Even though they had never seen the
prototypes or the new exemplars, they did
better at categorizing the prototypes than at
categorizing the new exemplars.! Category
D!

What they
actually learned! What they saw during the learning phase!

Prototype! New Exemplar! Prototype!

Category C Prototype!

Category Category
A! A!

Category Category
B! B!

Category • When subjects were shown a new pattern, Category • If they were shown the prototype for
C! they compared this pattern with each of C!
Category C, this would match their
their 4 prototypes.! mental prototype even better.!
• If it was a new exemplar of Category C, it
Category Category • Consequently, subjects were better at
D! would be somewhat similar to the prototype D!
categorizing the prototypes than at
for that category, so they were above
chance at saying that it belonged to categorizing new exemplars!
Category C.!

© S. J. Luck

All rights reserved
7

 Reed (1972)!
Exemplars! Prototypes!

Class A! • Same basic design as Posner & Keele, but with

artificial face stimuli.!

• Each exemplar was generated by making slight

Class B!
changes to the eyes and nose of the
prototype.!

Learning phase: ! • Result: Subjects scored above change when

Subjects indicated Class A or Class B for the exemplars tested on new exemplars, but they did even
and were given feedback! better when tested on the prototypes, even
! though they had never seen the prototypes
Test phase: !
Accuracy for new exemplars = 61%!
before (just like Posner & Keele’s results)!
Accuracy for the prototypes = 90%!
Reed (1972)!

Categorization: Instance Theory!

Exemplar 1! Exemplar 2! Exemplar 3! Exemplar 4!

New Exemplar! Category

A!

Category
B!

Category
C!

Average of 32 female German faces! Average of 32 male German faces!

Category
D!

Categorization: Instance Theory! Categorization: Instance Theory!

Exemplar 1! Exemplar 2! Exemplar 3! Exemplar 4!

New Exemplar! Category

• When the subjects are tested on a new dot A!

pattern that they had never seen before, they

can just compare the new pattern against all 16
of the exemplars.! Category
B!

Prototype A!
• For example, this new exemplar is fairly similar to
the exemplars of Category A, and it’s not similar
to the exemplars of the other categories, so the Category
C!
subject would categorize it as an A.!

Category
D!

© S. J. Luck

All rights reserved
8

 Categorization: Instance Theory! Categorization: Instance Theory!
• This can also explain why subjects do better when tested
Exemplar A1! Exemplar A2!
on the prototype than when tested on new exemplars,
even though they’ve never seen the prototype before.!
Prototype A!

• The prototype is the average of all the exemplars within a

category, so it’s fairly similar to all 4 of of the exemplars
from this category.!
Exemplar A3! Exemplar A4!

• The new exemplar won’t be quite as similar to the other

exemplars as the prototype is.! • The prototype dot, by definition, is at the average of
the four trained exemplar dots.!

• Even if subjects don’t use prototypes for categorization, • This means it is reasonably close to all four of them!
they should be able to categorize the prototype better
than the new exemplar even though they’ve never seen
either of them before.!

Categorization: Instance Theory! Categorization: Instance Theory!

Exemplar A1! Exemplar A2!

Exemplar A1! Exemplar A2!

Prototype A! New Exemplar!

Exemplar B1!

Exemplar A3! Exemplar A4!

New Exemplar!
Prototype A!
• A new exemplar may be closer to an exemplar of another category
(Exemplar B1 above), so it will be more easily miscategorized than
will the prototype.!

• The prototype is guaranteed to be reasonably close to all of the

exemplars of the category, but the exemplars are not.!

Exemplar A3! Exemplar A4! • This is how the instance theory can explain why people categorize
the prototype better than they categorize new exemplars, even
though this theory proposes that subjects don’t actually use
prototypes during categorization.!

Exemplar 1! Exemplar 2! Exemplar 3! Exemplar 4! Application to Real-World

Categories!
• Imagine that you were given a new animal and asked whether it’s a
bird, a dog, a cat, or a horse.!

• You would compare it against all of your memories of birds, dogs,

cats, and horses, and you’d determine which set of exemplars it
matches best.!

• It looks a lot more like the cat examples than the bird, dog, or horse
examples, so you would categorize it as a cat.!

• You don’t need some kind of abstract prototype of each category to

categorize this new animal. The instances are enough.!

© S. J. Luck

All rights reserved
9

 • Both the prototype and Feature Weighting!
instance theories help us
understand how we form
categories like sports.!

• You don’t need to come up

with a set of necessary and
sufficient conditions. !

• Instead, you can take your • This also helps us understand

experience with things that are why volleyball seems like a good
example of a sport, whereas
• Another key issue in categorization is the weighting or
called sports and either come
up with a prototype of sports fishing is a somewhat odd importance of different features.!
or use your memories of example.!
individual instances when • When you’re categorizing fruit, color can be an important
you’re trying to decide if • You frequently encounter sports feature. For example, color helps us determine whether
something is a sport.! that share many features with
volleyball, like basketball, so
something is a lemon or an orange.!
volleyball will be similar to both
your instances and your • And many birds are distinguished by their coloring, like
prototype.! this bluebird.!

Feature Weighting! Theories!

• But color is not important in categorizing cars. !

• For example, pink is not a common car color, but you

don’t have any trouble telling that this is a car.!

• And the color doesn’t provide much information about • How do we decide which features are important?!
what make or model a car is. !
• For example, usually something that has wings, flies, and eats insects
would be a bird.!

• However, these features are also present in bats, which are not birds.!

• This has led to the idea that we have implicit theories about things
like birds and mammals, and we use them when making category
judgments.!

Theories! Concepts & Categories:

A Synthesis!

Instances!
Classical View!

• These theories include things like the idea of evolution.

Because bats evolved from earlier mammals, they are not
birds.!

• In addition, we’ve been taught that feathers and eggs are

more important for being a bird than is flying, !
Theories!
Prototypes!
• So the fact that bats have fur rather than feathers helps
All four theories appear to be correct. We use different
us realize that bats are not birds.! approaches to categorization in different situations.!

© S. J. Luck

All rights reserved
10

 Concepts & Categories: Concepts & Categories:
A Synthesis! Prototypes! A Synthesis!

Instances!

• We use instances when the number of items in the

category is small, ! • We rely on prototypes if the number of instances is huge, or
we haven’t had enough time to learn all the instances.!
• Or when we’ve had lots of time to learn about each
individual instance.! • We use both prototypes and instances to define the center of
a category.!
• If the number of instances is large, it would be hard to • If I ask you to name a typical bird, you’ll name one that is
use all of the instances when we categorize.! prototypical or common.!

Concepts & Categories:

A Synthesis!
• If you’re trying to find the
boundaries of a category, or if
you’re faced with conflicting
Classical!
information, you may rely on a
View! theory, and this may involve
necessary and sufficient
conditions, as in the classic view. !

• For example, your underlying

theory of birds will help you
Theories!
determine that a bat is not a bird,
even though it flies, and that an
ostrich is a bird, even though it
doesn’t fly.!

© S. J. Luck

All rights reserved
11