5 Semantic Memory

Learning Objectives
1. Describe the associative nature of semantic memory.
2. Demonstrate an understanding of concepts, categories, and theories of
their representation.
3. Explain schemas and scripts and their role in semantic memory.
4. Interpret the nature of visual representation in semantic memory.

Think of the many things that you have learned in school. In

economics, you learned that prices are determined by supply and
demand. In physics, you learned that electricity and magnetism are
related. In literature, you learned that T. S. Elliot was an important
poet of the 20th century. In music history, you learned that Debussy
was an impressionist composer. Also think of the things you know
that you have learned elsewhere. You have learned the function of
your car’s carburetor; you have learned that in double knitting, two or
more yarns are alternated; and you have learned that hybrid cars
use less gasoline and emit less greenhouse gases than normal cars.
You may know the name of the current prime minister of Canada
(Justin Trudeau, in 2020). You may also know who is dating Taylor
Swift, and that Christie Brinkley posed for bikini photographs at age
61. These facts about the world are not things that you are born
knowing. We learn them as we go and need a memory system to
store them until we need them.

Our memory for knowledge of the world—from literature to knitting to

politics to gossip—is called semantic memory. Semantic memory is
broadly defined as general knowledge of the world. Semantic
memory includes our knowledge of history, sports, ideas, geography,
pop culture, and music. As such, it covers a broad range of topics,
only some of which we will be able to address here. What unites
these topics is that the memory involved is stored in a common
representational system. Semantic memory is thought of as
supported by a separate neurocognitive system than is episodic
memory, which we discussed in the last chapter. Recall that episodic
memory has a temporal dynamic—we attribute retrieval from
episodic memory to the past. In semantic memory, we attribute
retrieval of information to truth about the world. In this chapter, we
will consider how people represent information into semantic
memory as well as issues of encoding and retrieval.

Lexical memory (also known as the lexicon) is our mental

dictionary, a representational system for the words of our language.
Lexical memory is also considered a separate neurocognitive system
from semantic (and episodic memory). There is much overlap
between what is meant by semantic memory and what is meant by
lexical memory. However, semantic memory is usually studied by
researchers interested in memory, and lexical memory is usually
studied by researchers interested in language.

Psycholinguists are interested in the structure of the representational

system involved in lexical memory for a number of reasons. Lexical
memory must maintain a huge number of items (words and rules of
grammar). Access to these items must be extremely fast in order to
support normal speech and normal speech comprehension. Given
that human beings start life knowing no words, but children acquire
vocabulary at astonishing rates, the encoding of words into lexical
memory is also of great interest. We will consider lexical memory
alongside semantic memory, but we also note that there are
important differences.

Semantic memory: The neurocognitive memory system that encodes,

stores, and retrieves information concerning knowledge of the world.

Lexical memory (lexicon): Our mental dictionary; a representational

system for the words of our language.
Associative Structures in Semantic
Memory
Each of us knows countless facts about the world. One goal for
memory researchers is to try to model how all this information is
represented by one’s cognitive systems (not to mention how it is
physically stored by the brain). To achieve this goal, most theorists
have focused on associative models of semantic memory. An
associative model means that we represent information in semantic
memory in terms of connections among units of information (see
Figure 5.1). A node is the unit of memory, which is then connected to
other nodes. For example, you might have a node for the 18th-
century British philosopher George Berkeley, who championed the
ideas of associationism. The node for the memory of this philosopher
might be linked to the node for John Locke (another associationist
philosopher), to the node for the University of California at Berkeley,
and to the node for Charles Barkley, the star basketball player for the
Philadelphia 76ers in the 1980s and 1990s. Nodes may be strongly
connected, as in the Berkeley–Locke connection, or less strongly
connected, as in the Berkeley–UC-Berkeley connection and the
Berkeley–Barkley connection. Eventually, as the activation spreads
to multiple nodes, it fades, and we can no longer see the trail of
activation. Thus, in association models of semantic memory, our
memory is a web of interconnected ideas and facts.
Description

Figure 5.1 ■ An associative network in semantic memory. Arrows

show spreading activation. Actual associations are bidirectional.
Source: Based on theory by Collins and Loftus (1975).

Semantic network models posit that retrieval takes place when one
node is activated based on the input from a cue. Once a node is
activated, this activation will spread to existing associated nodes
(Collins & Loftus, 1975). The term spreading activation refers to
the transfer of activation from one node to an associated node. For
example, if you are asked, “Who was the British philosopher most
noted for his theories of associationism and his theory that the mind
starts as a blank slate?” this should activate the node in memory for
“John Locke.” Activating the node for John Locke can lead to the
spoken response, “Of course, I know that one, it’s that bloke from
Somerset, England—John Locke.” The current thinking in cognitive
psychology is that activation spreads—that is, as soon as the node
for John Locke is activated, some, but a lesser amount of activation,
spreads to the node for George Berkeley because the nodes for
Berkeley and Locke are connected. In turn, some activation spreads
to the node for Charles Barkley. Thus, activation spreads from one
node to the next when one node is activated. However, an important
component of these models is that activation weakens as it spreads
from node to node. Thus, “Charles Barkley” will be less activated
than “George Berkeley.” It is also likely that the activation will not
spread further than “Barkley” (e.g., failing to activate any other
basketball knowledge).

Associative model: We represent information in semantic memory in

terms of connections among units of information. A node is the unit of
memory, which is then connected to other nodes.

Spreading activation: Refers to the transfer of activation from one

node to an associated node.

Collins and Loftus’s (1975) spreading activation model focuses on

the idea of a web of connections among nodes (or concepts) within
the model. Activation spreads along this web, even when the web is
not necessarily logical or organized. Thus, for example, the
connection between the philosopher Berkeley and the basketball
player Barkley is circumstantial. There is no other reason to group
these two people together other than the similarity in their names. To
some extent, the Collins and Loftus (1975) model was an elaboration
(and a criticism) of an earlier model, the hierarchical network
model developed by Collins and Quillian (1969) some years earlier
(see Figure 5.2).

Description
Figure 5.2 ■ Collins and Quillian’s (1969) model of semantic
memory: the hierarchical network.

In the hierarchical network model, semantic memory is organized

hierarchically. Thus, if we take the concept “people,” you would find
that concept associated with general characteristics of people—that
is, they have faces, they talk, and they come in all different ages and
sizes. “People” is a higher-level concept over particular kinds of
people, such as musicians, marines, and mermaids. Each of these
kinds of people is then associated with particular characteristics
associated with their calling. Thus, “musicians” might be associated
with “plays music,” “drinks coffee in cafés,” and “teaches others to
play piano.” A “marine” might be associated with “big and strong,”
“willing to defend country,” and “wears hair short.” A mermaid might
be associated with “has fins,” “lives in ocean,” and “not real.” The
important aspect of this model is that if one has to “find” in memory
whether a mermaid can talk, one has to move up the hierarchy, as
“talking” is associated with people, not with mermaids. Similarly, if
one has to find whether marines have bank accounts, one has to
move up the hierarchy, as “having bank accounts” is associated with
people in general rather than with marines specifically. This leads to
the prediction that when a person is asked to make decisions about
aspects of a concept that are stored at a different level of the
hierarchy, they will be slower than when asked about aspects of a
concept stored at the same level. We will consider the implications of
this model when we examine sentence verification tasks.

Hierarchical network model: A model in which semantic memory is

organized by levels or hierarchies in which particular nodes are
associated with characteristics associated with that level, but not with
higher or lower levels.

Semantic Priming and Lexical

Decision Tasks
This theoretical architecture has been repeatedly put through
experimental testing and, by and large, is generally supported by the
data. One of the main methods of testing this model is semantic
priming. The basic idea of semantic priming is as follows: When a
particular node is activated (e.g., by hearing “John Locke”), that
activation will spread to associated nodes, including the one for
George Berkeley. Similarly, if one reads the word tiger, activation
should spread to related nodes for concepts such as “striped” and
“lion.” Activation of these additional nodes should make it easier for
the person to process the associated items. Thus, we should
respond faster to the name George Berkeley if we have just heard
the name John Locke, and we should respond faster to the word lion
if we have just heard the word tiger. Thus, priming refers to the effect
of the first presented word or name on the response to the later word
or name. Priming will generally work better when the two items are
also presented closely in time.

In experimental situations, semantic priming is typically measured by

use of a lexical decision task. In a lexical decision task, participants
must judge as quickly as they can whether a string of letters is a
word or not (Meyer & Schvaneveldt, 1976). For example, the words
doctor and cranberry should elicit a response of yes, but the
nonsense string xffxere should elicit a response of no. Possible
words that are not used in English, such as pluckban or scrawps,
should also elicit a response of no. Participants are asked to make
such judgments as quickly as they can so that reaction times can be
measured. Typically, a young adult in college can make these
decisions in under half a second. In most experiments, nonsense
strings (e.g., xffxere) are easier and therefore quicker to reject than
possible, but not employed nonwords (e.g., scrawps). More common
words (e.g., doctor) are easier and therefore quicker to verify than
obscure words (e.g., jejeune).

Semantic priming: The effect of one word or idea on the processing of

a related word or idea. A related word will activate a target item and
allow it to be processed more quickly.
 Lexical decision task: A cognitive task in which participants judge
whether a string of letters is a word as quickly as they can.

Semantic priming can then be added to the lexical decision task.

Participants still have to judge if a string of letters is a word, but now
that string is preceded by a prime, which is also usually a word but
may be a picture as well. Primes may be related or unrelated to the
to-be-judged words. For example, if the target word is doctor, a
related prime is nurse, and an unrelated prime is gumball. Research
consistently shows that related primes produce faster and easier
processing of the target word. This translates into faster reaction
times in the lexical decision task for words preceded by related
primes than those preceded by unrelated primes (McNamara, 2005;
Meyer & Schvaneveldt, 1971). Semantic priming occurs because the
participant processes the prime and the prime enters semantic
memory, and activation from the prime spreads to adjacent nodes in
memory. Thus, nurse activates nearby concepts, such as doctor, and
this activation allows one to respond faster to whether or not doctor
is a word. Indeed, the closer the prime is connected to the target
word, the greater the semantic priming effect (Moldovan, Ferré,
Demestre, & Sánchez-Casas, 2015). There will be a bigger semantic
priming effect if nurse primes doctor than if heart primes doctor.

Moreover, this priming can cross languages. Poort, Warren, and

Rodd (2016) showed that priming with sentences in one language
(Dutch) was sufficient to prime words in a lexical decision task in
English. In another recent experiment, Schröter and Schroeder
(2016) found that beginning readers in third grade showed robust
semantic priming effects. Thus, shortly after one becomes a
competent reader, semantic priming effects are apparent. Semantic
priming can also occur when the participant is not consciously aware
of the prime, as in subliminal priming (Marcel, 1983). Subliminal
priming may be sufficient to activate the network and allow that
activation to spread.

Semantic priming can also serve as a marker of certain neurological

impairments (Howells & Cardell, 2015). For example, anomic
aphasia is a condition in which patients have difficulties finding the
correct word in a given context. Their speech is characterized by
pauses and attempts to compensate for their inability to recall a word
clearly in their lexical memory. Howells and Cardell compared
patients with anomic aphasia to control patients on a semantic
priming task. Interestingly, despite their word retrieval problems,
patients with anomic aphasia showed strong semantic priming
effects, even if their overall reaction times were slower than the
control patients. Thus, even though they have difficulty retrieving
words, the nodes exist within their semantic memory, and priming
existing nodes will spread to related nodes, allowing these patients
with anomic aphasia to respond faster during semantic priming.

Spreading activation states that activation will move from one node
to the next. Thus, nurse will prime doctor, which is associated with
the word lawyer. Thus, the associative model predicts that a word
like nurse will also show some small priming effect on a second-
order association. The smaller amount of priming occurs because
activation spreads and dissipates. Anticipating how much priming
occurs is important for developing quantitative (mathematical)
models of how priming occurs. This kind of priming is called
mediated priming. Mediated priming occurs when the prime word is
related to a word, which is also related to the target word. For
example, if the target word is lion, then the word stripes will prime it.
The reason is that stripes is associated with tiger, and tiger is
associated with lion. Therefore, preceding lion with stripes allows for
faster processing (L. L. Jones, 2012; McNamara & Altarriba, 1988).
In an interesting study on mediated priming, Duñabeitia, Carreiras,
and Perea (2008) showed that mediated priming could also cross
over from semantic characteristics to phonological characteristics.
The first link in priming is semantic (lion to tiger), but the second link
is phonological—that is, sound related (tiger to Geiger). In a similar
way, John Locke can prime Charles Barkley through George
Berkeley. Duñabeitia’s stimuli were in Spanish, but their example is
that cup is semantically related to coffee, but cup also primes toffee,
which is related to coffee in sound as well as the visual appearance
of the letters but does not have the same meaning overlap.
Sentence Verification Tasks
Another manner in which spreading activation of association and the
hierarchical model have been experimentally tested is by looking at
sentence verification tasks. In a sentence verification task,
participants are asked to decide as quickly as possible if a sentence
is true or false. Thus, a sentence such as “All men are mortal,”
should be judged to be true, whereas a sentence such as “Hawaii is
located in the Atlantic Ocean,” should be judged to be false.
According to the model of spreading activation, sentences reflecting
closely linked nodes will be verified faster than sentences reflecting
distantly linked nodes. This is likely because activation can spread
from one node to a nearby node faster than it can spread to a more
distant node. For example, the sentence “Tigers have stripes,”
should be verified faster than an equally true sentence that reflects a
more distant association, such as “Tigers have lungs.” Collins and
Quillian (1969) reasoned that characteristics that are unique to a
particular item will be closely linked in associative space, whereas
characteristics that are shared with other items will be more distantly
linked. Stripes will be stored closely to tiger, but lungs will be linked
through a common link to “land animals.” Thus, Collins and Quillian
had participants judge the truth value of sentences, and participants
were instructed to do so as quickly as possible. Returning to our
example from earlier, sentences like “Marines wear their hair short”
were verified as being true faster than sentences like “Marines have
faces.” Both sentences are true, but “short hair” is closely linked to
the concept of a marine. In their experiment, this prediction turned
out to be correct; closer associations led to faster verification times.

Sentence verification task: Participants are asked to decide as quickly

as possible if a sentence is true or false.

Neuroimaging work employing sentence verification tasks supports

the basic ideas that associative networks exist. In one study,
Raposo, Mendes, and Marques (2012) looked at semantic
verification tasks while participants were being scanned by fMRI
technology. They found that areas of the left prefrontal lobe and
areas of the left medial temporal lobe were particularly active during
sentence verification. However, left temporal lobe activity was
equally high regardless of the associative distance between the
subject clause and verb clause of the sentences used in sentence
verification, whereas greater frontal lobe activity was needed to do
sentence verification when the associative distance between the
subject clause and verb clause was greater. Thus, for example,
“Limes are sour,” requires less frontal involvement than an equally
true statement such as “Limes are fruit.” This supports the notion of
spreading activation because the closer associates (lime–sour)
require less recruitment of brain regions than the more distant, but
equally true, associations (lime–fruit).

Therefore, it is a reasonable conclusion that semantic memory is

organized into an associative network. In this associative network,
information is linked with varying strengths and at varying distances.
Strong connective links or close links lead to much shared activation,
whereas weaker links or more distant links lead to less shared
activation. When a particular node is activated, that activation
spreads to other, adjacent links. In the next section, we will expand
on the notion of associative structure in semantic memory and
discuss how items in semantic memory are bound together into
categories and concepts and how the mind creates these.

Bilingual Representation
Lexical memory is a close cousin of semantic memory but focuses
on how language is represented in memory. As with semantic
memory, huge stores of information must be stored so as to be
easily accessed when that knowledge is called for. A debated issue
is whether bilingual or multilingual people have independent
representation of concepts in semantic/lexical memory (which we
call simply semantic memory here) for all the languages that they
speak or each language has its own lexical memory system within
the cognitive architecture of that individual (Cai, Pickering, Yan, &
Branigan, 2011). One view is called the single-store view of
bilingual representation. This view argues that there is a common
semantic level of representation—that is, meaning is shared. The
shared semantic representational system then connects to a system
that produces the right way of saying that concept in each language.
Thus, a concept such as truth is represented once and then
connects to two (or more) phonological systems, one for each
language. Thus, the concept of truth is associated with “truth” in
English but vérité in French. The alternate view is the dual-store
view (or separate store) of bilingual representation, which
postulates that meaning is represented separately for each
language. In this view, there are two separate conceptual
representations, one for truth and one for vérité. Much of the recent
research on the topic supports the single-store view (Declerck,
Meade, & Grainger, 2018; J. F. Kroll, Bobb, Misra, & Guo, 2008).

Think of a person you know who is bilingual. This person has two
phonological forms that may represent any particular concept, one in
one language and one in the other. Thus, an English–Spanish
bilingual will know both the English word clock and the Spanish word
el reloj. Obviously, these words will have different phonological
(sound) representations, given that there is no overlap in sound.
However, is the semantic representation stored jointly, or is there a
separate representation for the meaning in each language? The
single-store view states that each concept is stored in a single
system and is linked to the separate phonology in each language.
The dual-store view states that your friend has two separate
semantic stores, one for each language (Gianico & Altarriba, 2008;
X. Wang & Forster, 2015).

The view that semantic representations are shared across a

multilingual’s languages is supported by cross-language semantic
priming. Cross-language semantic priming is similar to the semantic
priming discussed earlier, except that the prime and the target word
are in different languages. Consider a lexical decision task in which a
participant has to determine whether a string of letters is a word. The
prime word is presented in Spanish (e.g., el libro, meaning “book”).
Shortly thereafter, the target word appears in English (e.g.,
magazine). The single-store view of lexical memory predicts that el
libro should prime magazine. However, the dual-store view argues
that meaning is represented separately for both languages and,
therefore, the Spanish word should not prime the English word.
Studies with cross-language semantic priming typically show that
cross-language priming occurs, supporting the single-store view
(Guasch, Sánchez-Casas, Ferré, & Garcia-Albea, 2011). Poort et al.
(2016) found that presenting sentences in one language (Dutch)
primed participants on a lexical decision task in another (English)
even when there were nearly 20 minutes between the prime and the
target, suggesting that some forms of spreading activation can last
much more than fleeting seconds.

Single-store view of bilingual representation: There is a common

semantic level of representation in bilinguals.

Dual-store view (or separate store) of bilingual representation:

Meaning is represented separately for each language in the lexical
memory of a bilingual.

Cross-language semantic priming: The effect of priming a word in one

language has on a related word in another language in bilinguals or
multilinguals.
Section Summary and Quiz
Semantic memory is our memory for knowledge. Lexical memory is
our knowledge of the words in the languages we speak. Given the
size that each memory system must be in any human being, memory
science has focused on how all this information is represented. For
semantic memory, the representational system is thought of as a
system of interlocking nodes. Activation spreads from node to node
when a cue is given to elicit retrieval. We can test this spread of
activation by doing semantic priming experiments on lexical decision
tasks. Semantic priming means the effect of one word or idea on the
processing of a related word or idea. A related word will activate a
target item and allow it to be processed more quickly, thus allowing
faster response times in a lexical decision task. A lexical decision
task is a cognitive task in which participants judge whether a string of
letters is a word as quickly as they can. Semantic verification tasks
assess how participants are asked to decide as quickly as possible if
a sentence is true or false. Results show some evidence for
hierarchical organization. Closely associated information is verified
faster than more distant information. Cross-language semantic
priming also supports the single-store view of bilingual
representation. The single-store view of bilingual representation
states that there is a common semantic level of representation in
bilinguals.

Quiz

1. Mediated priming means that

1. The priming only occurs in the presence of a masking stimuli
2. Episodic representations are primed so that people can make
semantic responses
3. An intermediate associative link is present between the prime
word and the target word
4. In some instances, spreading activation does not apply to
lexical decision tasks
2. In a sentence verification task, the participant is required to
1. Determine as quickly as possible whether a group of words is
a grammatical sentence
 2. Decide as quickly as possible if the content of the sentence is
true or false
3. Determine as quickly as possible if the sentence possesses
words that were given earlier in the priming phase
4. All of the above
3. The single-store view of bilingual representation means that
1. A bilingual represents meaning in one representational
system across his or her two languages
2. A bilingual represents meaning in a representational system
for each language
3. Bilinguals will respond faster in a lexical decision task
because of multiple association routes
4. Bilingual memory storage is organized in a substantially
different way than monolinguals
4. Which finding is best explained by the hierarchical network model?
1. That we are faster to verify that “sharks have big teeth” than
“sharks have blood”
2. That nurse primes doctor
3. That gato (Spanish for “cat”) primes cat
4. That we are faster to verify sentences presented in our native
language

Answers

1. c
2. b
3. a
4. a

Concepts and Categories

Concepts and categories are two terms borrowed from everyday
usage, but their definitions within cognitive psychology are quite
specific. A concept is a mental construct that contains information
associated with a specific idea. A concept is a mental representation
of an idea. Concepts can be representations of concrete objects
such as “automobiles” and “chili peppers,” and they can also be
abstract, such as the representation of “peace” and “prosperity.” A
category is a mental construct referring to a set of objects or ideas
that are grouped together or are associated with each other. That is,
a category usually refers to things that are similar in some respect.
Categories include items that are like each other in some way, for
example, “tools,” “furniture,” “flowering plants,” “forms of
government,” and “Mexican Americans.” Note the flexibility of a
category. A hammer has little in common with a pressure cleaner,
except that they are both used to maintain order in a household. The
category “Mexican American” means all people in the United States
who share a common origin in Mexico. It says nothing about how old
they are, what they look like, what language they speak, where they
live, their socioeconomic status, or which political parties they prefer.

Concepts can refer to our mental representations of categories. For

example, the concept of a “tool” is our mental representation of the
characteristics that tools share. However, concept is a broader term
than category. For example, concepts such as truth, love, and
literacy are abstract ideas not easily translated into categories.
Categories tend to be more concrete, such as “things that are true,”
“the reasons we love,” and “literacy rates among Western countries.”
It is your semantic memory system that stores concepts and
categories and allows you to access them when you need to. It is
important for cognitive systems to be able to categorize material so
that we can organize information and understand it.

Because of the complexity of concepts and the need for spreading

activation networks to account for them, cognitive neuroscientists
expected that representation of concepts would be widely distributed
throughout the brain. In fact, activating a concept may draw on a
great many different brain regions. Kuchinke, van der Meer, and
Krueger (2009) examined decisions made about whether two objects
belonged to the same category while monitoring participants using
functional magnetic resonance imaging (fMRI). They found a wide
distribution of brain areas became active, including areas in the
frontal lobes and in the temporal lobes. Because they were dealing
with linguistic concepts, there was a decided emphasis on the left
hemisphere, but areas in the right hemisphere were activated as
well.
 Concept: Mental construct that contains information associated with a
specific idea.

Category: Mental construct referring to a set of objects or ideas that are

grouped together or are associated with each other.

Categories Are Fuzzy

At first, a category seems like a concrete entity. “Tools,” “furniture,”
“literacy rates,” and “Mexican Americans” all seem like fairly
straightforward categories. But each category has fuzzy boundaries.
Consider the category of “tools.” Clearly, hammers and pressure
cleaners are tools, despite differences in what they are used for and
how they operate. But is a pickle jar a tool? Looking up the word tool
in a dictionary might yield something like this: a device that helps us
accomplish a task. Using this definition, a pickle jar certainly is a tool.
It is a human-made object whose function is to store pickles,
something that would be much more difficult to do without a pickle
jar. Yet to most of us, a pickle jar does not quite fit into our mental
category of “tools.” It is too passive—it just sits there holding pickles.
Tools need to be wielded. Tools are entities like axes, hammers, and
blowtorches, and a pickle jar is not quite one of them.

Consider the category of “Mexican American.” More straightforward?

Clearly, a U.S. citizen whose parents were born in Guadalajara, grew
up speaking Spanish, and has a last name of García fits into our
category of “Mexican American.” But others may be a bit more
confusing. What about Californians whose ancestors lived in
California while it was still part of Mexico (until 1847) but whose
ancestors have been American for over 170 years? What about Sen.
Mitt Romney, whose father was born in Mexico to American parents
temporarily living there? Is he “Mexican American”?

There are other ways of looking at boundary conditions as well. For

example, function is an important defining feature of many concepts.
Thus, for tools, function—that they aid humans in tasks—is at the
heart of what a tool is (see Figure 5.3). What happens when an
object is no longer able to serve that function? What about broken
tools? Is a pressure cleaner that is out of gas a tool? Is a broken
hammer a tool? They cannot serve the function of being a tool even
though they still resemble tools. Which is more important for the
definition of a tool—its function or its appearance? Consider the
concept “bachelor,” which refers to an unmarried adult male.
Certainly, an unmarried sexually active heterosexual male at the age
of 40 who lives alone is a bachelor. But there are unmarried
heterosexual adult males who would not be considered bachelors—
Catholic priests, for example. One might choose to revise the
definition of “bachelor” to mean heterosexual unmarried adult men
who have not taken vows of celibacy. But then there are widowers
and divorcees. Do they revert to bachelorhood? Without belaboring
this example, suffice it to say that boundary conditions are important
in defining categories.

Figure 5.3 ■ All these implements are a part of a common category,

tools. What does an object have to be in order to be considered a
tool? Is an aspirin a tool? Is a computer a tool?
Thinkstock

Because of these boundary conditions, researchers think of

categories as complex and related to each other in multiple ways.
The complexity of categories seems to be clear in the way memory
is organized for categories. An important variable that has
psychological reality is called levels of categorization—that is,
categories are nested structures in which the level of organization is
important in defining the category. There are three such levels:
basic, subordinate, and superordinate (Rosch & Mervis, 1975). The
basic level is the one we are most likely to invoke. For example,
“tools” is a basic-level category for most people. The subordinate
level allows for more specific or specialized categories to be formed
(e.g., “power tools”). Superordinate categories are higher-order
categories that include general information and include many more
common categories (e.g., “human implements”).

Levels of categorization are fluid, depending on a person’s expertise.

For example, for most of us, “cat” is the basic-level category and
“Scottish fold” is the subordinate category. But for a cat fancier who
takes her cats to shows across the country, “cat” might assume a
superordinate classification, as the person endlessly discusses the
variants among various types of purebred cats. For the cat fancier,
“Scottish fold” takes on the basic-level classification, and the
differences between American short-hair Scottish folds and
European piebald Scottish folds become the subordinate category
(see Figure 5.4).
Figure 5.4 ■ A Scottish fold. At a basic level, we call this animal a
“cat.” For most of us, “Scottish fold” is a subordinate category, and
“feline” is a superordinate category.
Thinkstock

Levels of categorization have a psychological reality. Research has

demonstrated that basic-level information is retrieved faster than
subordinate or superordinate information (T. T. Rogers & McClelland,
2004; Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976).
Indeed, when we see an object, we tend to use the basic level first.
For example, we call the object on your desk generally a “computer,”
not a “MacBook Air laptop computer.” Furthermore, different levels of
categorization are associated with different areas of the brain.
Kosslyn, Alpert, and Thompson (1995) used positron emission
tomography (PET) technology to examine categorization. They did
so by asking participants to match pictures to words. Thus, a picture
of a dog was accompanied by three words: animal (superordinate),
dog (basic), and miniature poodle (subordinate). They showed that
superordinate levels were more likely to activate the prefrontal cortex
than basic-level categories. In contrast, when people chose
subordinate categories, there was more activity in the parietal lobe of
the brain. It is possible that superordinate categories cause people to
reason about the category or have to remember more features,
taxing the prefrontal lobe, whereas it is likely that the subordinate
level draws on more sensory representations, thus drawing on the
parietal lobe.

Levels of categorization: Categories are nested structures in which the

level of organization is important in defining the category. There are
three such levels: basic, subordinate, and superordinate.

Family Resemblance
We’ve seen that categories are ambiguous, depending on the
viewpoint of the person conceptualizing them. Moreover, nearly
every natural category will defy a definition and will have members
that differ from the mainstream. Think of the category “Mexican
American.” A U.S. citizen living in San Diego, named García, and
fluent in Spanish is a typical Mexican American. However, the
grandson of a German immigrant who lived briefly in Mexico before
moving to the United States is a less typical Mexican American.
Family resemblance means that membership in a category may be
defined by each item’s general similarity to other members in the
category rather than by a specific list of features (Johansen, Savage,
Fouquet, & Shanks, 2015; Rosch & Mervis, 1975). Family
resemblance takes into account that many categories are
multimodal, with many characteristics that may apply. For example,
think of the category “bird.” When we consider birds from a purely
zoological understanding, both penguins and ostriches are every bit
as much of a bird as is a sparrow, a robin, or an eagle. Yet because
penguins and ostriches vary from most birds in many respects (most
notably, flying), they are often considered less central members of
the category.

Because of the complexity of categories and their fuzzy boundary

conditions, cognitive psychologists have found it necessary to create
models to account for how concepts and categories are represented
in semantic memory. The following sections describe the major
models that have been advanced.

Family resemblance: Membership in a category may be defined by

each item’s general similarity to other members in the category rather
than by a specific list of features.

Exemplar Theory and Feature Comparison

Theory

Exemplar Theory.
According to exemplar theory, categories are classified by
maintaining a large number of specific instances of a category that
are associated with each other in semantic memory (J. D. Smith &
Ell, 2015). When we encounter a new item, we compare it to each
and every exemplar in that category to determine whether it matches
or not. The individual items that are stored in semantic memory as
evidence of a particular category are called exemplars. As we will
see, this contrasts with prototype theory in that prototype theory
suggests that we draw an average of all the instances of a category
and use that as a comparison. Exemplar theory contends that the
category is all of the associated items (Medin & Rips, 2005).

Exemplar theory: Categories are classified by maintaining a large

number of specific instances of a category (exemplars) that are
associated with each other in semantic memory.

Exemplar theory is successful in accounting for the multidimensional

nature of a category. Consider the category “tools.” It is difficult to
imagine what an average or prototypical tool would look like.
Consider the difference between hand tools and power tools. Thus,
prototype theory is challenged here. However, if each and every
exemplar is represented in an associated structure in semantic
memory, any new item can be compared against the appropriate
match.

What evidence exists for the reality of the exemplar theory? It turns
out that priming can be specific to the particular aspect of the
exemplar that is being primed (Gagné & Shoben, 2002). If the prime
focuses on one aspect of the category (e.g., color), it will better
prime those members of the category for which that aspect is salient.
Thus, it will be easier to verify what the term television show means
if it is preceded by reality show than if it is preceded by trade show,
because both television show and reality show share a common
meaning of the word show.

Some have criticized the exemplar approach because it necessitates

a large memory system that can store hundreds, if not thousands, of
exemplars for each and every different category. However, the
capacity of long-term memory to store information is seemingly
limitless, so large demands on capacity alone are not enough to
disqualify the theory. It is also possible that both exemplars and
prototypes may be used in the formation of categories and in
comparing new instances to existing categories.

Feature Comparison Theory.

Another theory as to how categories are represented in semantic
memory is the feature comparison theory. Feature comparison
theory states that we maintain a list of features for each category.
According to this view, each category is represented by an
associated list of the kinds of characteristics that make up the
category. Thus, for example, the category of “books” might include
the following features: (a) contains written material; (b) is intended to
be read; (c) contains a set number of pages, usually enclosed in
covers of paper or cardboard; and (d) is intended to last a long time.

New objects are compared neither to a constructed prototype nor to

a set of exemplars. Rather, new objects are matched to the list of
features. Thus, when one encounters a magazine, it fails to fit into
the category of “books,” because it does not fulfill the fourth feature
of the category “books.” Your textbook, however, does fulfill the
features and would be classified as a “book.”

E. E. Smith, Shoben, and Rips (1974) distinguished between

defining features and characteristic features. Defining features are
required for any example of a particular category. For example, the
defining feature of a book is that it stores words. Characteristic
features generally accompany an instance of the category but are
not required. Characteristic features of books include that they are
meant to be read, have covers, are made of paper, and have an
author. Most of us would still classify books-on-tape as books,
because they store words, which can transmit meaning even if they
are not printed, not made of paper, and not meant to be read.

Feature comparison theory is essentially flawed because so few

categories have clear defining features. Consider the example of the
category “books.” The defining feature of a book is that it stores
words. But is the defining feature really accurate? What about a
children’s book that only has pictures but not words? What does that
leave as a defining feature of the category “books” when it includes
picture books, audiobooks, electronic books, and so on? The
category “books” becomes multidimensional, and a single set of
features no longer defines it. Another example of this problem is to
think of what the defining feature is of the category “cat.” Is it that it is
a living thing, that it meows, and that it has cat genes? What about a
dead cat, a mute cat, and a stuffed animal cat? All of these
exceptions propelled theorists to consider the prototype as being
better able to explain the structure of semantic memory.

Feature comparison theory: We define our categories by maintaining

a list of features for any particular category.

Defining features: According to feature comparison theory, defining

features are required for any example of a particular category.

Characteristic features: According to feature comparison theory,

characteristic features generally accompany an instance of the category
 but are not required.

Prototype Theory
A prototype is defined as the most typical member of a particular
category. Prototype theory states that prototypes form the central
characteristics in our representation of categories. In prototype
theory, we compare examples that we encounter in the real world to
a constructed mental prototype. If the example matches the
prototype, we decide it is an example of the category. If it does not
match, we reject the category label for that item. For example, we
compare an example of a tool to our prototype of a tool. The
prototype itself need not be a real member of the category. In fact, it
is likely to be an averaged mental representation of the concept
(Rosch, 1975). Thus, the prototype may not exist at all in the real
world; it is an abstraction. Consider your prototype of a bird. If you
live in the Northeast of the United States, your prototype may look
something like a robin, but without the specific coloring. If you live
farther south or closer to the ocean, your “bird” may obtain longer
legs and a longer beak but is not yet an egret. If you live in a big city,
your prototype may resemble a pigeon. When you encounter a new
animal, you match the animal to your prototype of a bird. If it is a
close enough match, you assign it to the category “bird.” If it fails to
match, you dismiss that example as a bird.

What evidence is there that we represent categories in terms of

prototypes? We will consider two main lines of research. The first
demonstrates that prototypes are rapidly remembered and named
when a category is supplied. The second shows the effects of
semantic priming on prototypical members of a category and
nonprototypical members of a category. Each of these points to a
strong role for prototypes in our semantic memory.

• Prototypes are easily named. In studies done by Eleanor Rosch

and her colleagues, participants were asked to judge how
prototypical a given instance was of a particular category. For
example, “raccoon” might be judged highly prototypical of the
category “mammal,” but “platypus” might be judged less prototypical.
Similarly, “carrot” may be judged more prototypical of the category
“vegetable” than “kale.” A second group of participants was then
asked to generate the names of examples from categories, such as
“mammal” and “vegetable.” The researchers found that the items
named first and more frequently by the participants were the ones
that were usually judged as more prototypical (Mervis, Catlin, &
Rosch, 1976). Thus, the first and most common vegetables
generated included prototypical vegetables like carrots and peas and
only later, if at all, did the participants produce items like kale and
bok choy.

• Prototypes and semantic priming. The more prototypical an

example is, the more likely it is to be judged more quickly after
semantic priming. Semantic priming works better if the target words
are prototypical rather than nonprototypical. For example, the prime
word feathers will create faster response speed if it precedes robin
than if it precedes cormorant. Thus, prototypical examples are more
locked into the associative structure that creates categories. For
example, using subliminal priming, in which participants did not
consciously see the prime word, Miles and Minda (2012) found that
priming led to faster judgments for category members and also led to
more errors in which non–category members were judged as being a
member of the category.

Prototype theory: States that prototypes form the central characteristic

in our representation of categories. A prototype is defined as the most
typical member of a particular category.

Situated Simulation Theory

The situated simulation theory states that the representation of a
particular concept varies as a function of the situation or context
(Barsalou, 2003). In this view, the context in which a category is
encountered calls upon different aspects of its representation. For
example, we think about the concept of “hydrogen” very differently
when we encounter it in the context of hydrogen-cell batteries than
when we encounter it in the context of theories of the early universe.
Similarly, we draw on different aspects of the concept “pumpkin”
when we are thinking of preparing soup and when we are thinking of
preparing for a Halloween party. This idea makes intuitive sense, but
are there data that support the situated simulation model?

Wu and Barsalou (2009) examined the situated simulation theory by

asking participants to regard specific concepts from different
perspectives. The experimenters gave participants a phrase, such as
“split coconut” or “chipped coconut,” and asked participants to
generate as many properties as possible for each phrase. Though
both stimuli call one’s attention to the concept of a coconut, the
adjective served as context, and people focused on different aspects
of the concept depending on that context. Thus, a “split coconut”
elicited phrases relating to the opening of a coconut, whereas
“chipped coconut” elicited phrases related to the drinking of coconut
milk. In this way, Wu and Barsalou demonstrated that context is
important when considering our semantic memory for concepts.

Situated simulation theory: The situated simulation theory states that

the representation of a particular concept varies as a function of the
situation or context.

Schemas and Scripts

To this point, our discussion of semantic memory has concerned the
topic of concepts, categories, and the connections among them.
Semantic memory also includes more general aspects of memory—
that is, units of memory that are more complex, which are commonly
called schemas or scripts. For example, think of the concept of
“parking”—that is, the idea that a car can be turned off and left for a
while in a safe location where you can return to it later. In U.S.
society, parking one’s car is a well-learned ritual. The script for
parking a car involves (a) arriving at the parking lot, (b) searching for
a spot—get there early or the spots will all be taken, and (c)
choosing to swing around the lot an extra amount of time in the
hopes that someone with a really good spot is just leaving and you
can take that one, (d) once you have your spot, position your car so
that other cars cannot sneak in front of you, (e) place your car in the
spot—and maximize the difference from nearby cars so as to avoid
“door dings,” and (f) note the location of your car so you can return to
it later. Many of us follow this script many times a week. The point is,
we all have this script memorized already. We know what to expect
when we enter the parking lot. The script of “how to park your car” is
another form of information stored in semantic memory.

Fundamental to this area is the idea of a schema. A schema is

generalized knowledge about an event, a person, or a situation (the
plural of schema can be either schemas or schemata). Schemas are
patterns of connected information about a particular topic. We have
schemas, for example, as to what you find in a car. We expect seats,
beverage holders, controls, and transparent windows. Imagine your
surprise if you entered your friend’s car and there were no seats, but
there was a sink. Think of a car in which all the windows were
rendered completely opaque. These cars would violate your schema
of a car (and be quite dangerous). We also have schemas about
what to expect during college exams. This particular schema may be
all too familiar to college students. But imagine coming to an exam to
find your professor dancing and playing loud disco music from the
1970s on a phonograph player while you were trying to take the test.
This would violate our schema, as included in our schema of taking
an exam is that the professor will be quiet. From the point of view of
memory representation, schemas and scripts are well-learned
patterns that guide our behavior and organize information in memory
(Abelson, 1981).

Scripts refer to a particular kind of schema. Scripts are well-learned

sequences of events associated with common activities. For
example, we have scripts for such common activities as “going to a
restaurant.” Without having to recall specific instances of restaurant-
going, you can describe the common process. A person shows you
to a table and gives you menus. You spend several minutes deciding
what to eat. Then a waiter comes by, and you give him your order
(politely, hopefully). Several minutes later, the waiter returns bearing
your meal. You eat your meal. And so on and so forth. Our memories
abound in such scripts—waiting to board an airplane, checking out at
the grocery store, and your morning “getting ready” ritual are just a
few of the many scripts you have stored in your semantic memory.
Violations of scripts often turn out to be memorable events. Consider
an airplane flight in which the flight attendants decided to forgo
boarding by seat numbers. The ensuing crush to board the airplane
may prompt you to avoid that airline in the future. Consider a
restaurant without menus. This puts a maximal demand on the
waiter, who must tell you all of the options for the day’s meal. It also
puts memory demands on the patron, as she must remember all of
these options. However, you may also consider a restaurant without
menus an interesting novelty and recommend the restaurant to
friends.

Schema: Generalized knowledge about an event, a person, or a

situation.

Scripts: Well-learned sequences of events associated with common

activities.

Consider a schema for a dentist’s office. There is a waiting room with

relatively comfortable chairs. Magazines are available for patients
while they wait. These magazines usually include Time, Sports
Illustrated, and National Geographic. A few children’s toys lie unused
under a television showing Oprah. A woman sits behind a glass
panel and checks you in. The office is decorated with hokey signs
(“A bright smile shines on everyone’s day”). An anthropomorphic
tooth, encouraging kids to brush, stands to one side. Beyond a door
is a maze of rooms with dental chairs, X-ray machines, and
occasionally, an actual dentist. These are items we come to expect
after a lifetime of visiting dentists’ offices. However, consider when
an event or item is inconsistent with the schema. Consider sitting
down in the waiting room and finding pornographic magazines
instead of the expected National Geographic. Consider coming to a
waiting room that had yoga mats instead of chairs. Consider a
dentist’s office covered in Marxist slogans instead of the usual
platitudes. Such variations from the dentist schema would not go
unnoticed. Indeed, many of us would find these deviations from the
schema sufficient reason to find a new dentist.

A classic study by W. F. Brewer and Treyens (1981) illustrates points

about schemas and our expectations very nicely. Participants arrived
for an experiment and were asked to wait in a lab room (Figure 5.5).
The participants waited in the room a short time and then entered
another room for testing. To their surprise, the participants were then
asked to recall all of the objects that they could remember from the
waiting room.

Description
Figure 5.5 ■ What can you remember from this room?
Source: Brewers and Treyens (1981).

Participants remembered many items from the office that were

consistent with the concept of a waiting room. For example, people
remembered the desk, the chair, and the typewriter (this was 1981,
after all). Some people falsely remembered schema-consistent items
not present in the room (e.g., books). However, several items in the
room are generally not found in offices but did not strongly violate
the schema for a waiting room. For example, wine bottles and picnic
baskets are seldom found in offices but their presence here should
not have been particularly surprising, and few participants
remembered these items. However, items that violate the office
schema were well recalled; most participants recalled the skull, even
though this is seldom found in academic offices. Thus, we can
conclude that schemas help us notice and recall items and events
that are consistent with the schema and help us remember violations
of the schema. However, items irrelevant to the schema may go
unnoticed. To summarize Brewer and Treyens’s (1981) results:
Schema-consistent items were well remembered, schema-violating
items were also well remembered, but schema-irrelevant items were
less well remembered.

The reconstructive nature of maintaining schemas also has a dark

side. Stereotypes function as schemas for our semantic memory
representations of people (Jackson & Rose, 2013). If you think about
stereotypes, they are bodies of knowledge (however false) that make
generalizations about specific groups of people. Thus, like our
schema for the dentist office, a stereotype provides us with a
roadmap about what to expect from a particular kind of person.
Regardless of whether the stereotype is benevolent or malignant, it
promotes the encoding of stereotype-consistent information. It can
also lead us to falsely remember something that is stereotype-
consistent but not actually true about the specific individual.

In normal circumstances, however, once a schema is formed, it is

useful in encoding new information. The schema forms the backbone
upon which new information can be associated. For example,
Bransford and Johnson (1972) presented participants with confusing
passages to read, such as the one shown in Box 5.1. Some
participants were given schematic information to assist them in
understanding the passage. For example, in the passage below,
some participants had the title “Washing Clothes” to aid their
understanding. Those provided with the title recalled more
information from the passage than those who had not read the title.

As you can see, the text is written in an ambiguous way. Once you
know, however, that it is about washing clothes, most of the
ambiguous words should make sense. Therefore, having a schema
in mind (in this case, for washing clothes) can help an individual
encode new information into memory.

Box 5.1 Washing Clothes

The procedure is actually quite simple. First arrange items into different
groups. Of course one pile may be sufficient depending on how much
there is to do. If you have to go somewhere else due to lack of facilities,
that is the next step; otherwise, you are pretty well set. It is important not
to overdo things. That is, it is better to do too few things at once than too
many. In the short run this may not seem important, but complications
can easily arise. A mistake can be expensive as well. At first, the whole
procedure will seem complicated. Soon, however, it will become just
another facet of life. It is difficult to foresee any end to necessity for this
task in the immediate future, but then, one can never tell. After the
procedure is completed one arranges the material into different groups
again. Then they can be put into their appropriate places. Eventually
they will be used once more and the whole cycle will then have to be
repeated. However, that is part of life.

Source: Bransford and Johnson (1972).

In some situations, using schemas may be counterproductive to

learning new information. Think about the typical clothing shop at the
mall—after a few minutes of browsing, a nosy salesperson will
engage you and try to get you to buy the most expensive clothes.
Having a schema for this form of shopping may not help you when
you shop at the Salvation Army store. Thus, it will not help you to
invoke the mall schema when trying to remember your trip to buy
clothes for charity. Indeed, if the goal is to learn new episodic
information, using schema information will lead to the kind of errors
we just saw in the Brewer and Treyens (1981) experiment in which
participants often recalled items from the room that were consistent
with the schema but not actually present in the room. Badham and
Maylor (2016) asked participants to learn weakly related word pairs,
such as “barbell–saucepan.” In one condition, participants were
given a schema that helped them organize the items (“metal,” in the
example), whereas in other conditions they relied on their own
idiosyncratic encoding techniques. Interestingly, in this experiment,
providing the schema did not improve memory performance for
younger adults and hurt performance in older adults. Presumably, in
this case, providing schema-type information led participants to use
less than optimal encoding techniques, thereby lowering the
performance in the schema condition relative to the control.

Reconstruction of Events
Schemas are also useful in providing structure for retrieving longer
events and information, such as stories. Consider describing a
recent trip to a restaurant. The distinctive feature about this lunch
was getting together with an old friend whom you had not seen in a
long time, and you dined at a restaurant with food you are not
accustomed to, perhaps a Middle Eastern restaurant. Thus, what
you remember is catching up with your friend, on the one hand, and
the interesting tastes, on the other—such as hummus and baba
ghanoush. Now consider describing the event. Do you really
remember the maître d’ bringing you to the table? Do you remember
whether the maître d’ or the waiter brought you the menus? Did the
same waiter who took your order bring you your food? In many
cases, we may use the schema to help us fill in this information. In
telling the story of your visit to the restaurant, what is important is
communicating to your listener the gist of the story: the good food
and the good friendship. The listener is not expecting you to get the
details of the service accurate, but it may be necessary to insert
such information to maintain the narrative flow. For example, you
might say, “By the time the waiter came to take our order, we were
already drinking the most interesting tea,” when in fact, you ordered
before the waiter brought the tea.

Indeed, a classic experiment concerns the use of schemas in

storytelling. Bartlett (1932) asked his British college students to play
a “telephone game.” In the experiment, he asked one student to read
a story derived from a Native American folktale from the Pacific
Northwest. The story was called “The War of the Ghosts” and is
shown in Box 5.2. Sometime after hearing the story, the student was
asked to tell another student the story, who then had to tell the story
to yet another person. After several such retellings, Bartlett had the
final student write down the story. He found that changes in the story
from the original had to do with making it more consistent with the
schema for a story from someone from British culture. For example,
“hunting seals” became “going fishing,” and often an explanation for
the sudden and unexplained death at the end of the story was
included. In short, the British participants transformed the story
according to their own schemas. In another experiment, Bartlett had
students read the story, and then each student wrote it down four
months later. As in the telephone game, the errors that students
intruded into the story made the story more consistent with British
story schemas.

Box 5.2 The War of the Ghosts

One night two young men from Egulac went down to the river to hunt
seals, and while they were there it became foggy and calm. Then they
heard war-cries, and they thought: “Maybe this is a war party.” They
escaped to the shore and hid behind a log. Now canoes came up, and
they heard the noise of paddles, and saw one canoe coming up to them.
There were five men in the canoe, and they said:

“What do you think? We wish to take you along. We are going up the
river to make war on the people.”

One of the young men said, “I have no arrows.”

“Arrows are in the canoe,” they said.

“I will not go along. I might be killed. My relatives do not know where I
have gone. But you,” he said turning to the other, “may go with them.”

So one of the young men went, but the other returned home.

And the warriors went up the river to a town on the other side of
Kalama. The people came down to the water, and they began to fight,
and many were killed. But presently the young man heard one of the
warriors say: “Quick, let us go home: that Indian has been hit.” Now he
thought: “Oh, they are ghosts.” He did not feel sick, but they said he had
been shot.

So the canoes went back to Egulac, and the young man went ashore to
his house, and made a fire. And he told everybody and said: “Behold I
accompanied the ghosts, and we went to fight. Many of our fellows were
killed, and many of those who attacked us were killed. They said I was
hit, but I did not feel sick.”

He told it all, and then he became quiet. When the sun rose he fell
down. Something black came out of his mouth. His face became
contorted. The people jumped up and cried.

He was dead.

Source: Bartlett (1932).

Section Summary and Quiz
A concept is a mental construct that contains information associated
with a specific idea. A concept is a mental representation of an idea.
A category is a mental construct referring to a set of objects or ideas
that are grouped together or are associated with each other. A
number of models seek to account for how we represent categories
in our semantic memory. These include the feature comparison,
exemplar, prototype, and the situated simulation model. The most
successful theory is prototype theory. Prototype theory successfully
predicts that prototypes will be more easily named than other
examples of the category. Prototypical examples also show greater
semantic priming effects than do less prototypical examples.
Schemas and scripts are higher-order semantic memory
representations. Schemas are generalized knowledge about an
event, person, or situation, whereas scripts are well-learned
sequences of events associated with common activities. In a classic
experiment, Brewer and Treyens (1981) showed how important
schemas are to the organization of semantic memory. In another
classic experiment, Bartlett showed that culture has an important
role in schemas. We also reviewed evidence of the importance of
semantic memory in music.

Quiz

1. A mental construct referring to a set of objects or ideas that are

grouped together or are associated with each other is known as a
1. Lemma
2. Prototype
3. Exemplar
4. Category
2. In prototype theory, our mental representation of a particular
concept is
1. A very specific example that we use to compare other
instances of that category to
2. A hierarchical model in which category confusion is limited
through semantic priming
3. Represented by a series of exemplars
 4. A generalized or averaged idea of that category
3. In which theory of concept representation do we define our
categories by maintaining a list of features for any particular
category?
1. Feature comparison theory
2. Prototype theory
3. Exemplar theory
4. Situated simulation theory
4. That membership in a category may be defined by each item’s
general similarity to other members in the category rather than by
a specific list of features is associated with which aspect of
concept representation?
1. Family resemblance
2. Situated cognition
3. Exemplar specification
4. Prototype priming
5. After getting his car serviced, André remembers that he saw a man
covered with car grease outside the garage. It turns out that he got
there just as the shop was opening and employees were just
coming to work. Why might André remember the man with car
grease?
1. André has severe amnesia—this is obvious from the example
2. André may be relying on his schema for what to expect at the
garage
3. Because of poor attention, André is more subject to false
memories early in the morning
4. André is likely suffering from prototype justification
6. Brewer and Treyens (1981) asked participants to recall the waiting
room they had just been in. This experiment showed that
1. People have outstanding visual memories
2. Lemmas are necessary for the retrieval of speech errors
3. Schemas guide our recall, even if that leads to an occasional
schema-based error
4. All of the above are true
7. Bartlett’s (1932) experiment on “The War of the Ghosts” showed
that
1. Cultural schemas are important in memory
2. Scripts have no effect on memory when narrative recall is
required
3. All scripts are schemas, but not all schemas are scripts
4. None of the above are true

Answers
 1. d
2. d
3. a
4. a
5. b
6. c
7. a

Representation and Imagery

An important question is: How does semantic memory store visual
images? One of the oldest debates in visual imagery is about exactly
how, at both cognitive and neural levels, we code the images we
store and remember. The next section on representation of visual
images is one of the more difficult ones in the text. Keep in mind that
the goal of the research is to infer—from people’s performance—
what the exact nature of the representational code is. This
representational code is not the visual image itself that we call up in
visual working memory. Rather, the code is the manner in which that
image is stored when not in use—that is, when it is being stored for
future use. With that in mind, we will jump into this topic.

Representation means the storage of information in long-term

memory when that information is not in use. When we need to
retrieve information, we activate it from this long-term
representational system. In the case of visual imagery, activating that
information means creating a visual image. Once the image is
activated, we can inspect it and make decisions about it. For
example, you may have stored in semantic memory the layout of
your grandmother’s house. When you retrieve the pattern of her
house, you are retrieving visual information from semantic memory.
When we retrieve these representations, we then experience mental
imagery.

The problem of studying representation is that scientists cannot

directly examine mental images or their stored representations. We
cannot pull them out of a person’s head and inspect them. They exist
only in mental space. Therefore, cognitive psychologists must devise
experiments so as to infer what mental representations are like
based on visible behaviors or reports of the person doing the
imaging. This restriction presented some serious challenges to
cognitive psychologists when they first began to investigate visual
imagery.

There are two main classes of theories as to how we represent

visual information in memory. The first is called analog
representation, and the second is called propositional
representation. Analog representation means we store visual
images in a manner a lot like pictures. When we retrieve a
representation of the basketball player LeBron James, we retrieve
the neural equivalent of a photograph. Propositional representation
means that we store visual images in terms of a language-like code.
When we retrieve an image of the president, what we get is
equivalent to a list of features—visual descriptions of the president.
Our imagery system then recreates an image in visual working
memory (i.e., the visuospatial sketchpad).

Note that the propositional view does not deny that our images
appear to us as pictures. However, the propositional view states that
we do not store images in a pictorial format. Our representation is in
an abstract code. At the time of retrieval, we reconstruct a visual
image based on that representation.

To put this issue in perspective, when researchers first began

debating issues of image representation in the 1960s, most
researchers thought that the propositional view would turn out to be
correct. One reason for this view was that these researchers argued
that analog visual memory would take up huge amounts of memory
within the brain. Just a few pictures stored in analog fashion, they
argued, would eat up the brain’s entire memory capacity. This
reasoning emanated from the realization that in computers, text
requires much less memory storage space than visual images, and
back then computers had much less overall memory storage. Storing
information in a descriptive text-like representation requires much
less memory on your computer and presumably would require much
less memory in your brain. However, just as modern personal
computers are more than able to handle many digital pictures, it
turns out that the brain too has a lot more capacity for storing
pictures than had been originally thought. The propositional view,
however, does allow more information to be stored in a less-
demanding format, a model that still appeals to some theorists today
(see Pylyshyn, 2003).

However, most researchers today are convinced by the vast amount

of empirical data that support the analog view (Provost & Heathcote,
2015). It may be memory intensive, but the brain appears to be able
to handle the load. Studies using behavioral and neuroscientific
methodologies all support the analog view. One of the earliest
studies to support the analog view is also one of the most famous
studies in psychology. It is Shepard and Metzler’s (1971) mental
rotation experiment.

Representation: The storage of information in memory when that

information is not in use.

Analog representation: The theory that we store visual images in a

manner similar to actual pictures.

Propositional representation: The theory that we store visual images

in terms of a language-like code.

Shepard and Metzler’s (1971) mental rotation experiment: An

experiment on visual imagery that showed that representation is analog.

Shepard and Metzler’s Mental Rotation

Experiment
Shepard and Metzler (1971) set out to show that, at least in some
circumstances, imagery was best explained by the analog view. To
demonstrate this, Shepard and Metzler wanted to show that mental
images behave like real pictures, at least in some ways. They
thought that mental images ought to respond to experimental
manipulations in the same or similar way as actual perception of
those objects would. If inspecting a visual image is like inspecting a
real picture, then we might be able to answer at least some novel
questions about it. If, however, inspecting images is not like
inspecting a real picture, then, as was commonly believed in 1970,
the propositional view is the default position. Thus, the goal of
Shepard and Metzler was to determine whether we can use mental
images as we would real images. To test this notion, participants
were asked to mentally rotate objects that were shown to them in
pictures (see Figure 5.6). The task was relatively simple: Decide
whether two figures were geometrically the same or different as
quickly as possible.

Imagery: The experience of retrieving a memory that is mostly visual or

experienced primarily as a sensory experience. Imagery can also refer
to the representation of those memories.
Figure 5.6 ■ Objects to be mentally rotated.
Source: From Shepard, R. N., & Metzler, J. (1971). Mental rotation of thee-
dimensional objects. Science, 171, 701–703. Reprinted with permission from
AAAS.

Consider the participants’ task: For each trial, participants saw a pair
of objects displayed in pictorial format. Half of the pictures were
identical, and half were not. One of the objects was rotated to be in a
different orientation than its partner. Participants were asked to
determine whether the two objects were identical or different. For
participants to determine whether the two figures were identical, they
needed to mentally rotate one of the figures to see if it matched its
pair. If, in a participant’s mind’s eye, the figures matched up, he or
she was expected to say “same.” If they did not, he or she was
expected to say “different.” Most people who do this task describe
imagining one of the figures and “moving” it in their mind’s eye to see
whether, when rotated, it fits perfectly onto the other figure.

The empirical question asked in this experiment was whether the

participants take longer to mentally rotate a figure more. Shepard
and Metzler (1971) reasoned that if images were stored in an analog
manner, then it would take a finite amount of time for the participant
to rotate one figure to match the other. If one figure was rotated only
a small amount to the left or right, it should take less time to mentally
rotate it than if it had been rotated closer to 180 degrees. This is
analogous to how we would do this task if we were given toy-size
objects and asked if the objects were the same shape or different.
We would rotate one until it matched the other. If it did, we would say
they were identical. If we could not rotate them until they matched,
we would say they were different. Shepard and Metzler’s goal was to
see if our imagery systems function in the same way. Shepard and
Metzler varied the extent to which the two figures were rotated
relative to each other. In some cases, one was rotated just 15
degrees to the left or right of its partner. In other cases, one figure
was rotated as much as 180 degrees to the left or right of its partner.
This was true for both pairs that were identical and pairs that were
different. The extent to which the two figures were rotated relative to
each other was the main independent variable.

Therefore, Shepard and Metzler (1971) measured the amount of

time it took participants to come to a decision of same versus
different as a function of the degrees of difference between the two
figures in a pair. They found that the more the two figures were
rotated away from each other, the more time it took participants to
make their decisions. In fact, there was a clear linear relation
between the degrees of difference between the two figures in a pair
and how long the decision took, as can be seen in Figure 5.7. It took
—on average—just about two seconds to determine that the images
were the same or different when the two figures were rotated 20
degrees off each other but nearly five seconds when the figures
were 160 degrees different. The Shepard and Metzler data support
the analog view; people’s imagery systems appeared to be actually
manipulating images or pictures and not abstract, code-like
representations.

Description

Figure 5.7 ■ Results of mental rotation experiment.

Source: From Shepard, R. N., & Metzler, J. (1971). Mental rotation of thee-
dimensional objects. Science, 171, 701–703. Reprinted with permission from
AAAS.

To review, here is a short summary of the points to keep in mind both

to understand the experiment and its importance. Keep in mind that
the participants only see two images per trial—that is, they do not
see the figures actually move. They must rotate them mentally—that
is, to align them, they must use visual imagery. Because it takes
more time to mentally rotate a greater degree than a smaller degree,
there must be something similar between mental rotation and actual
physical rotation. Because of this, we conclude that the visual
images are being used in an analog, vision-like representation (but
see Liesefeld & Zimmer, 2013, for a different view).

Shepard and Metzler (1971) were not the first to demonstrate the
nature of imagery as being a vision-like analog system. One
demonstration came shortly before from the work of Lee Brooks
(reviewed in Chapter 3, because of its relevance to the visuospatial
sketchpad). In Brooks’s (1968) experiment, participants retrieved a
mental letter—that is, they imagined a letter, such as T, in their
mind’s eye. Then they were asked to make decisions about the
angles in that letter. Brooks found that responding by pointing was
more difficult than responding by speaking. The visual aspects of
pointing apparently interfered with the vision-like processes going on
during imagery. Thus, this experiment also supports the analog view.
If imagery were propositional, it would not require an overlap
between real visual tasks and imagery tasks.

Many other researchers subsequently became intrigued by the

notion that they could study the nature of imagery. One such
prominent researcher was Stephen Kosslyn, who invented many
creative ways of studying imagery (Kosslyn, 2010). Like Shepard, he
was restricted to measuring behavior, but also like Shepard, he
came up with ingenious ways of inferring how the imagery
representation system worked from seemingly ordinary behaviors. In
Kosslyn’s studies, people were asked to make simple judgments on
images but to make them as fast as possible. Participants studied a
picture and later had to make decisions based on the image they
remembered from the picture. In other studies, people were simply
asked to make images of familiar objects. In one such study, it took
participants less time to verify that a rabbit has soft fur when they
imagined the rabbit next to a fly than when they imagined the rabbit
next to an elephant (see Figure 5.8). Kosslyn (1975) argued that this
occurs because the image of the rabbit is likely to be bigger and
more detailed in our imagery system when we contrast it to
something smaller, like a fly, than when we contrast it to something
bigger, like an elephant. Therefore, details such as the quality of the
fur will be more salient in the mentally bigger image than in the
mentally smaller one. This only matters if we are using a picture-like
representation of each of the animals.
Figure 5.8 ■ Picture of big rabbit/small rabbit.
Source: Kosslyn, S. M. (1975). Information representation in visual images.
Cognitive Psychology, 7, 341–370. Published by Elsevier, Ltd.

In another study, Kosslyn, Ball, and Reiser (1978) showed

participants a map of an imaginary island (see Figure 5.9). The
participants were given ample time to memorize the features of the
map. When the participants were satisfied that they had memorized
the map, the map was removed. They were then asked to imagine
the map and make several decisions based on their mental map.
They were asked to imagine a dot moving around the island from
various locations to various locations. Kosslyn et al. (1978) found
that response time for getting the imaginary dot from Place A to
Place B was longer if the two locations were further away on the
map; the response times were faster if the two locations were closer
together. Thus, as in the Shepard and Metzler (1971) experiment,
mental imagery followed the same principles as the real world does
—longer distances take longer amounts of time to travel. This
suggests that the mental image people made of the island had visual
and spatial properties, much as a real picture does (Kosslyn, 2010).
Figure 5.9 ■ Briefly look at this map. Then look away and see if you
can make an image.
Source: Kosslyn, S. M., Ball, T. M., & Reiser, B. J. (1978). Visual images preserve
metric spatial information: Evidence from studies of mental scanning. Journal of
Experimental Psychology: Human Perception and Performance, 4, 47–60.
Published by the American Psychological Association.

Theorists who argue for the propositional view have criticized much
of the data presented here (Pylyshyn, 2003). They argue that we do
not need analog visual imagery and that mental propositions or
sentences can account for much of the data. For example, when you
are asked about the map of the island, you may experience a visual
image of the map, but according to the propositional view, that
experience is not causal. Rather, that image is a by-product of the
proposition that stores the information.

To explain this view in a bit more depth, in Pylyshyn’s version of the

propositional view, images are not represented as pictures. Rather,
they are abstract, language-like descriptions. As such, what we
retrieve when we access a visual memory is this description.
Therefore, it is at retrieval that we reconstruct the image that we
“see” in our mind’s eye. According to this model, the visual image is
a by-product of a retrieval process and not the result of encoding and
storage. As such, the image does not have a direct causal role in our
behavior, because the image does not become accessible until after
all the internal decisions are made on the language-like description.
To explain Kosslyn’s data, Pylyshyn argued that we slow our
response time because we think it must take longer to get from one
point on the island to the other but that the actual image of the island
is not being used to determine that distance.

Neuroimaging and the Analog View

With neuroimaging, we can look at the neural correlates of imagery
in the brain directly while people are engaged in imagery tasks
without having to measure a behavior. With the aid of neuroimaging,
a number of researchers have been able to determine the regions of
the brain involved in mental imagery. These neuroimaging studies
support the view that imagery is analog (Kosslyn, 2005; Slotnick,
Thompson, & Kosslyn, 2012). Kosslyn and his colleagues have
found that areas of the visual cortex are excited while people are
engaged in visual imagery tasks. In many cases, this activation
extends to the primary visual cortex.

The primary areas of vision are located in the back of the brain in the
occipital lobe. The primary visual cortex (also known as striate
cortex, V1, or Area 17) is the first area in the cerebral cortex that
receives input from the retina of the eye (see Figure 5.10). Countless
studies have shown it to be involved in basic visual processing.
Indeed, damage to the primary visual cortex results in blindness.
Surrounding the primary visual cortex are other areas in the occipital
lobe that process visual information (named V2, V3, etc.). All of
these areas become activated when people engage in visual
imagery (Kosslyn, Ganis, & Thompson, 2006; Slotnick et al., 2012).
In fact, the more similar the imagery task is to an actual visual task,
the more activation is seen in visual areas of the brain. Finally, those
individuals who do better in imagery tasks also show more activity in
the primary visual cortex than those who are less good at imagery
tasks (Kosslyn, 2005). Thus, it is clear that imagery is housed in
many of the same areas of the brain that house visual perception.

Primary visual cortex: The first area in the occipital cortex that
processes visual images.

Figure 5.10 ■ The occipital lobe is in the back of the brain.

Henry Gray via Wikipedia

To demonstrate the importance of visual regions of the brain in visual

imagery, consider the patient studied by Boucard et al. (2016). At the
time of the study, S. H. was a 70-year-old woman. At the age of 23,
both of her eyes were severely damaged in an auto accident, and
she lost sight in both eyes. Despite the 47 years of blindness, patient
S. H. still reports strong visual imagery. Boucard and her colleagues
asked S. H. to engage in visual imagery while her brain was being
monitored by fMRI. The brain scans revealed strong activity in V1
when S. H. was engaged in visual imagery, consistent with that of
sighted controls. Thus, visual areas of the brain, such as V1, appear
to be involved in visual imagery as well as visual perception, in this
case, even after years of not being able to actually see.
However, V1 is not essential for visual imagery. Other areas of the
visual cortex seem to be sufficient to support visual imagery. Another
recent study looked at visual imagery in a blind patient, but a blind
patient whose etiology is very different from S. H. (de Gelder,
Tamietto, Pegna, & Van den Stock, 2015). Patient T. N. became blind
following two successive strokes that completely destroyed the V1
regions on both the left and right sides of his occipital lobe. Thus, his
blindness is caused not by damage to the eyes but by damage to the
primary visual cortex. T. N. also claims to have visual imagery, and
his visual imagery is associated with other regions of the occipital
cortex (and activation in other regions of the brain). Thus, V1
activation may cause greater and more intense visual imagery, but
other areas of the occipital lobe can support it as well.

In another striking demonstration of visual imagery and brain injury,

Bisiach and Luzzatti (1978) studied a patient with hemifield neglect
(also known as unilateral neglect). Hemifield neglect describes a
condition in which patients ignore one half of the visual world. When
forced to, they will report that they see what is in that half of the
visual world but simply pay it no heed. For example, male hemifield
patients will not shave half of their face, because they ignore it in the
mirror. Hemifield neglect is caused by damage to the parietal lobe,
usually on the right side of the brain. Luckily, the severe effects of
hemifield neglect often wear off relatively quickly after the brain
damage, leaving the patient with a mild tendency to ignore
information in that half of the visual world.

Hemifield neglect (unilateral neglect): A condition in which patients

ignore one half of the visual world. It occurs because of damage to the
right parietal lobe.

The patient studied by Bisiach and Luzzatti (1978) had suffered from
a stroke, which resulted in hemifield neglect. The neglect wore off to
the point that the patient, a prominent newspaper editor, could return
to his professional duties. The researchers had him return to the lab
periodically. In the study, they tested his visual imagery, which
showed evidence of imagery-based hemifield neglect. Asked to
imagine himself walking down the main square from north to south in
his native city in Italy, the patient described the monuments and
buildings only on the west side (his right). When asked to imagine
himself walking up the main square from south to north, the patient
described the sites only on the east side (now on his right). Thus,
there was nothing wrong with his visual memory. He successfully
retrieved the buildings on either side of the main square during one
of the tests. However, he always ignored the left side of the square.
Thus, the imagery deficit mirrored the perceptual deficit. Both
neuroimaging and neuropsychology point to the role of visual areas
of the brain as being critical to visual imagery.

Returning to the issue of representation of imagery, neuroscience

data support the analog view. Remember that the analog view
advances the idea that images are stored as “pictures” rather than
descriptions. Pictures are most logically stored in the visual areas of
the brain, which is, in fact, activated during visual imagery. If
representations were more like descriptions, one might expect to see
the representation of them in verbal areas of the brain, which does
not occur. Thus, the neuroimaging and neuropsychological data are
consistent with the analog position.

Photographic Memory: Reality or

Fantasy?
As a professor who teaches courses in memory, I frequently hear
from students about someone they know who possesses
photographic memory. Belief in photographic memory is quite
common in our society. The questioner usually wants to know how
that person acquired this photographic-like memory and whether the
skill is learnable. It is often thought that a photographic memory
provides a huge advantage in the academic world, as it allows the
person with it to memorize vast amounts of materials without effort.
That is, massive amounts of information are immediately deposited
in semantic memory for visual information. I usually reply that self-
professed “photographic memory” may be correlated with a good
visual imagery system but is not photographic in the sense that it
creates a literal representation of what the person saw. I also add
that although photographic memory may seem to help that person, it
is no substitute for hard work and understanding.

Photographic memory: Very strong visual memories that have a strong

feeling of being images.

It is likely that those who claim to possess “photographic memory”

(or eidetic imagery, in technical terms) do have extraordinarily good
visual imagery systems that allow them to form strong visual images
of the material that they learn. However, these images are not truly
photographs. This can be demonstrated empirically. It has been
shown that people who have “photographic” memories do remember
visual information better than those who do not make this claim.
However, they still make errors in memory, and even when they
claim to be “reading” off their photograph, their errors tend to be
related to meaning, as in normal semantic memory, rather than
pictorial (Crowder, 1992).

That normal human memory is not photographic is easy to

demonstrate empirically. Consider an experiment conducted by
Nickerson and Adams (1979). They asked people to recognize the
“real” penny (that is, how an actual penny appears) from a number of
close distractors (see Figure 5.11). They found that, despite
familiarity with pennies, the participants were surprisingly poor at
picking out the real penny from the distractors. Try it for yourself in
the demonstration. You may have to find a penny in your change jar
to be sure of the answer.
Description

Figure 5.11 ■ Which is the real penny?

Source: Nickerson, R. S., and Adams, J. J. (1979). Long-term memory for a
common object. Cognitive Psychology, 11, 287–307. Published by Elsevier, Ltd.

In a similar, but perhaps more contemporary version of this idea,

Blake, Nazarian, and Castel (2015) looked at people’s memory for
the Apple logo, another image that most people may see on a daily
basis (see Figure 5.12). When people had to pick out the real Apple
logo from a number of similar images, only half of the participants
could pick out the correct image from the distractors, despite high
confidence in their ability to do so. Indeed, when people were asked
to draw the Apple logo, only one out of 85 participants got the bite
and the stem in the correct direction (see Figure 5.12). In another
study, Vendetti, Castel, and Holyoak (2013) showed that longtime
workers in an office building could not accurately recognize elevator
panels that they saw several times a day, in some cases for years.
These studies demonstrate that we typically extract meaning from
images rather than their literal visual representation.
Figure 5.12 ■ Which is the real Apple logo?
Source: Blake, Nazarian, and Castel (2015).

Thus, contrary to popular belief, even those who claim to have

photographic memory employ meaning-based representations as
well as visual ones. Indeed, this is adaptive, as in most situations, it
is meaning that is relevant rather than literal visual representation.
The “penny” and the “Apple logo” studies are illustrative examples of
this: Despite our frequent exposure to the stimuli in these
experiments, our literal representation of a penny or the Apple logo
is likely to be inaccurate.

Cognitive Maps
An important feature of semantic memory for visual material is that it
allows us to learn and remember our physical landscape. We all
have “cognitive maps” stored in our semantic memory. Cognitive
maps are mental representations of the external world (Tversky,
2000). These cognitive maps usually refer to the familiar
environments we inhabit. We may have a cognitive map of our own
homes, our neighborhood, our dorm complex, the campus, and
perhaps familiar towns and cities. We may have cognitive maps of
roads that we use frequently. After returning from a visit abroad, you
may have many new cognitive maps, which you learned as you
traveled in a new country. We may also have cognitive maps based
on drawn maps. You may never have been to India, for example, but
you may be able to create a mental image of a map of India.

Cognitive maps: Mental representations of the external world. Based

on our spatial representation of the world.

Consider the following example. You are asked to give directions to

the student center just as you leave your class in the business
school. You might consult your cognitive map and then give the
following directions: “Go over the bridge on the pond. Then continue
on the path past the library. Then make a right, and you’ll be at the
student center. It should take you about 10 minutes at most.” To
provide these directions, you must have some mental representation
of what the campus looks like, the relative distances between
buildings, and how long it takes to walk these distances. You may
also have more specific information, such as the fact that the student
center is due east of the business school, but providing this
information would probably be of little use to a stranger on campus.

Unlike ordinary maps, our cognitive maps are affected by semantic

categories. For example, consider the borders between towns,
states, and countries. In some cases, such borders may be physical
boundaries as well as semantic ones, as rivers or coastlines may
define such borders. But in almost all cases, such borders are
semantic in nature—that is, they define the human-defined
differences between two states. We judge places on our cognitive
maps that are grouped together to be closer than places that are not
grouped together, even though the items in different categories may
be closer (Friedman, Montello, & Burte, 2012; Huttenlocher, Hedges,
& Duncan, 1991). For example, most Americans would likely judge
that the city of Wayne, Michigan, is closer to Detroit, Michigan, than
Windsor, Ontario, is. However, in reality, Windsor and Detroit are
separated by a thin river and are much closer than are Detroit and
Wayne. Semantic categories also affect our judgments of cardinal
directions. Our semantic category “the United States” is placed south
of our semantic category for “Canada.” Thus, most of us would judge
Windsor to be north of Detroit. Not so! Detroit—in the United States
—is north of Windsor, in Canada. One of the most famous examples
of this derives from our semantic categories of the “Atlantic Ocean”
and the “Pacific Ocean.” Think of your cognitive map. It is pretty easy
to determine from that map that the Atlantic Ocean is east relative to
the Pacific Ocean. Thus, most people would judge the
Atlantic/Caribbean entrance to the Panama Canal as being east of
the Pacific entrance to the Panama Canal. Close inspection of a real
map reveals the surprising fact that the Pacific entrance is further
east than the Atlantic entrance. The point here is that we use
categories, such as the general locations of the Pacific and Atlantic
Oceans, to create cognitive maps, and these may lead to
inaccuracies.

To illustrate this concept, Friedman et al. (2012) had American and

Canadian college students examine maps of California and Alberta.
The maps were either marked with only dots and not with city names
or marked with dots and the corresponding city names. Participants
were directed to attend to the maps. A short time after seeing the
map, participants were asked to judge the distance between various
dots based on their cognitive maps formed by viewing the physical
maps. Friedman et al. were interested in how the inclusion of city
names would influence these judgments, given that the city names
provided semantic information. In the condition in which the names
were supplied, estimates of distance between dots were heavily
influenced by perceived proximity to nearby cities. That is, the
semantic information influenced the judgments. Thus, cities
perceived as related, such as Los Angeles and Long Beach, were
often judged to be closer than they actually are, whereas cities
perceived to be unrelated, such as Los Angeles and Bakersfield,
were judged to be further apart than they actually are. Indeed, the
distance estimates were more accurate for those participants who
had not had the names supplied originally. This experiment,
therefore, supports the idea that category information is used when
making cognitive maps.
Semantic categories (maps): Meaning affects our cognitive maps. For
example, people tend to think of borders between countries as being
straighter than they often are.
Section Summary and Quiz
One of the major issues in visual imagery is the nature of
representation. Analog representation essentially means we store
visual images like pictures. Propositional representation essentially
means that we store visual images in terms of a language-like code.
Much of the early cognitive research on this topic was done to
differentiate these two views. At present, almost all data strongly
support the analog view. Shepard and Metzler (1971) conducted an
experiment on mental rotation in which participants who mentally
rotated objects took longer to match objects that were at greater
angles relative to each other, thereby supporting the analog view.
Kosslyn’s (2005) research has also shown that it takes longer to
obtain information from small images than from large images,
consistent with the analog view. It has been found through
neuroimaging studies that visual areas of the brain are active during
visual imagery. Even in neuropsychological patients, there is
evidence that imagery is analog; a patient suffering from hemifield
neglect has deficits in both visual attention and attention in visual
imagery. Photographic memory is a myth. Even people with very
strong visual memories make errors that are based on meaning.
Studies such as the “penny” study and the “Apple logo” study show
how meaning typically trumps visual realism in representation.
Cognitive maps are visual representations of space that guide our
navigation through the world. As with most issues in this chapter,
cognitive maps are also heavily influenced by meaning, in this case,
categories such as countries, states, towns, and borders.

Section Quiz

1. In cognitive science, representation means

1. The storage of information in neurocognitive systems
2. Interpolation between analog and propositional imagery
3. The encoding of visual information
4. The transformation of visual rotation
2. In Shepard and Metzler’s (1971) experiment, the participants were
asked to
 1. Rotate objects until they matched the visual image stored in
their memory
2. Compare two presented pictures to determine if their shape
was the same
3. Rotate mental objects to determine the depth of their visual
imagery
4. Recall the number of locations marked on a map seen earlier
3. In Shepard and Metzler’s (1971) experiment, the researchers
found that
1. Visual imagery is propositional
2. If you ask participants to rotate visual objects, they will
3. The greater the angle of rotation, the more time it took to
determine if two figures were the same
4. When participants were given mental rotation instructions,
they rotated the objects faster than when they were not given
these instructions
4. Boucard et al. (2016) tested a patient who had been blinded as an
adult. Their study showed that
1. The blinded patient also lost her ability to do visual imagery
2. Area V1 of the brain was active when this patient engaged in
visual imagery
3. Only nonoccipital regions of the brain were active when the
patient was given imagery instructions
4. None of the above are true
5. In a study in which people were asked to recognize the pattern of
floors on the panel of an elevator, the researchers showed that
1. Recall of the pattern was at floor (0%), but recognition
approached ceiling (100%)
2. Even for people who had worked in the building for many
years, recognition was still quite low
3. Participants preferred the penny study and the Apple-logo
study to the elevator study
4. Participants refused to enter the elevator, as the weight load
had been exceeded
6. Research on cognitive maps shows that
1. People use meaningful landmarks to shape their cognitive
maps
2. People with eidetic imagery use more semantic organization
in their cognitive maps
3. People judge distances to be closer if they cross meaningful
boundaries
4. None of the above are true
 Answers

1. a
2. b
3. c
4. b
5. b
6. a

Language, Lexical Memory, and

Semantic Memory
Although often treated as separate topics in separate courses,
language and memory are closely intertwined. Language is heavily
dependent on the proper functioning of the memory systems that
serve it. We need to remember tens of thousands of words,
thousands of aspects of grammar, and the way in which to
pronounce certain words, and in languages such as English, we also
need to memorize the spelling of thousands of irregular words. In
other languages, such as French and Spanish, every noun has a
gender, and we must memorize that la mesa (the table) is feminine
but el corazón (the heart) is masculine. For this reason, we introduce
the concept of lexical memory. Lexical memory is closely related to
the concept of semantic memory, but lexical memory is specific to
the representation of the words that we use to express ideas—and
the use of words involves language. Thus, in the section, we are not
interested in psycholinguistics per se, but rather the memory
representations that make language possible.

Lexical Memory
Episodic memory refers to our memory for individual events from our
lives, and semantic memory refers to our knowledge of the world.
Both of these memory systems are considered declarative
memory, because we can verbalize the contents of each kind of
memory. It is possible to talk about your past experiences and what
you know about the world. Thus, both of these memory systems are
based on our ability to manipulate and use words. However, to use
words, we must have those words represented in memory and easy
to access. Lexical memory refers to the memory system that stores
words and other linguistic entities in memory. It is also called the
lexicon.

Declarative memory: Any memory that can be verbalized; includes

both episodic and semantic memory.

In theory, lexical memory is a representational system that stores

words. But words are not that simple. Words have multiple meanings
and many associations in different contexts. They may be
pronounced in different ways depending on the context (e.g., “You
say tomato, and I say tomahto”). In bilinguals, it is important to
address the relation in memory between translation equivalents
(e.g., tomato/tomate or book/libro in English and Spanish).

What makes models of lexical memory even more complicated is the

speed of access with which speakers must be able to access words.
First, consider how fast most of us speak: We usually produce three
to four words per second. In order to speak, we must be able to
rapidly access our lexical memory. We must also be able to rapidly
choose the right word from as many as 100,000 words in our lexicon.
In addition, there must be representation space that stores our basic
knowledge of the grammatical rules of our language. Most models of
lexical memory focus on three levels of representation—one at the
level of meaning, one at the level of syntax (that is, grammar), and a
third at the level that contains information about the phonology (that
is, sound) of the word (see Levelt, 1989; Oppenheim, Wu, & Thierry,
2018; see Figure 5.13). The assumption in these models is that a
network of semantic associations carries meanings and relating
concepts from one association to another—in essence, a semantic
memory system. In this conceptual system, the meaning of tiger may
be represented as an image of a big cat (associated with other
concepts such as “lion,” “leopard,” etc.), an image of a baseball
player from Detroit and its associations (e.g., “the baseball team,”
“the Detroit Tigers”), and any other secondary meanings of the word
tiger (e.g., “a sexually charged person”). This conceptual-level
representation informs a level of representation known as the
lemma. A lemma is a hypothetical entity containing only meaning-
based and syntactical information without any information
concerning the phonology of the word (Chang, Bauman, Pappert, &
Fitz, 2015). But unlike the conceptual level, the lemma is lexical in
nature in that it contains information about language usage, such as
the part of speech (noun, verb, etc.), how to make it plural, its
grammatical gender (in many languages other than English), and
other grammatical features. The lemma informs the next highest
level, known as the lexeme. The lexeme is the level of
representation that stores the phonology of a word—that is, how the
word sounds. When the lexeme is retrieved, the word can then be
sent to the motor system (Caramazza & Miozzo, 1997; Levelt, 1989).

Lemma: A hypothetical entity containing only semantic and syntactical

information without any information concerning the phonology of the
word.

Lexeme: The level of representation that stores the phonology of a word

—that is, how the word sounds.
Description

Figure 5.13 ■ A model of lexical memory.

Courtesy of Sarina D. Schwartz

Let’s see how this hypothetical system works. Say a person asks
you a question: “What is the word that means ‘second to last item on
any list?’” The question starts by inducing the lexical retrieval
process in the person who must answer the question. According to
the model outlined above, first the respondent must access
information at the semantic level—information such as “toward the
end,” “part of a list,” “fancy word for showing off your vocabulary,”
and so on. This initial response results in the retrieval of the lemma,
which contains both semantic and syntactical information. Here,
grammar information might be added—noun or adjective, derived
from Latin—as well as the semantic information from the previous
level. However, the lemma does not contain any phonological
information about how to say the word.

Once the lemma representation has been retrieved, the next stage
begins: the retrieval of the lexeme representation. In retrieving at the
lexeme level, the person retrieves the phonological information for
the word penultimate. The lexeme is still an abstract representation,
but it is a representation of sound. The phonological entry
penultimate is then sent to the motor system so that the person can
articulate the answer. Thus, according to this view, every time we
retrieve a word (even in normal speech), it passes through these
three stages of representation: semantics, lemma, and lexeme. Most
researchers in psycholinguistics agree on this general model,
although, of course, they argue over the specifics (see Levelt, 1989).

Much of the research that supports this model comes from errors in
ordinary speech called speech errors (M. Garrett, 1992; Warker &
Dell, 2015). Research focuses on two kinds of speech errors. The
first are called word exchange errors. Word exchange errors occur
when we substitute a word with a similar meaning for another word,
regardless of how the words sound. For example, if the word one
was trying to retrieve was sandals, one might accidentally say heels
instead. People often exchange the names of family members. You
know your older sister’s name but accidentally call her by your
younger sister’s name instead. These errors occur at the level of the
lemma and seldom cross syntactical categories. If the word you want
to say is a noun, the substitution will also be a noun, showing that
the correct meaning has been retrieved as well as the correct
syntactical information (e.g., noun, proper noun, plural or singular),
but the wrong lexeme is then activated.

Completely different are errors called sound substitutions (M.

Garrett, 1992; Martin & Dell, 2007). Sound substitutions occur when
a similar but incorrect phonological structure is retrieved but the word
retrieved has no obvious semantic or syntactical connection to the
desired word. For example, if the word you are seeking is tiger, you
might accidentally retrieve tirade. These errors involve mistakes in
sound but not in meaning and therefore are thought to occur at the
last level of retrieval in which sound is retrieved through the lexeme.

It is possible to have an error that is both semantic and phonological.

In this case, it is difficult to tell whether the error is the result of a
mistake at the level of the lemma or the lexeme.

Speech errors: Errors in ordinary speech, including exchange errors

and sound substitutions.
Section Summary and Quiz
Language is also dependent on memory. Lemmas are hypothetical
entities containing only meaning and grammatical information, whereas
lexemes are the level of representation that stores the sounds of words.
Studies on speech errors show the theoretical advantage of both the
concept of the lemma and the lexeme.

Section Quiz

1. A lemma contains information about

1. The way a word sounds
2. The meaning and grammar of a word
3. Only phonological information
4. Subphonic nonsyntactical information
2. In the representation of words in language, the representation at
the level of sound is called the
1. Lemma
2. Schema
3. Lexeme
4. Lexicon
3. When retrieving a word, if we accidentally retrieve blanket when
we meant to retrieve comforter, we are making a
1. Sound substitution
2. Exchange error
3. Lexeme-reversal error
4. Lemma-retrieval error
4. When retrieving a word, if we accidentally retrieve tesseract when
we meant to retrieve cataract we are making a
1. Sound substitution
2. Exchange error
3. Lexeme-reversal error
4. Lemma-retrieval error

1. b
2. c
3. b
4. a
Key Terms
analog representation 153
associative model 132
category 139
characteristic features 144
cognitive maps 163
concept 139
cross-language semantic priming 138
declarative memory 166
defining features 144
dual-store view (or separate store) of bilingual representation
137
exemplar theory 143
family resemblance 143
feature comparison theory 144
hemifield neglect (unilateral neglect) 160
hierarchical network model 133
imagery 154
lemma 167
levels of categorization 141
lexeme 167
lexical decision task 135
lexical memory (lexicon) 132
photographic memory 161
primary visual cortex 159
propositional representation 153
prototype theory 144
representation 153
schema 146
scripts 146
semantic categories (maps) 163
semantic memory 131
semantic priming 135
sentence verification task 136
Shepard and Metzler’s (1971) mental rotation experiment 154
single-store view of bilingual representation 137
 situated simulation theory 145
speech errors 168
spreading activation 132

Review Questions
1. What is semantic memory? How does it differ from episodic memory
and lexical memory?
2. What is meant by the term spreading activation? How do experiments
on semantic priming support the idea of spreading activation?
3. What is meant by the term family resemblance for categories? How
does it relate to the idea that categories have fuzzy boundaries?
4. What are the differences among prototype theory, exemplar theory, and
feature comparison theory?
5. What were the results of Brewer and Treyens’s classic experiment on
students’ memory for a waiting room? What do these results tell us
about schemas?
6. What is the difference between the single-store and dual-store view of a
bilingual’s lexical memory? Which view do the data on cross-language
priming support?
7. When does relying on schemas lead to errors in retrieval from memory?
Can you give an example of this from the discussion of music and
memory?
8. What are the theoretical differences between a visual memory system
that uses analog representation and one that uses propositional
representation?
9. Describe the Shepard-Metzler experiment. What was their hypothesis?
How was the experiment conducted? What results did they get? And
how were these results interpreted?
10. What is the difference between a lemma and a lexeme? How do
speech errors support this conceptual difference?

Online Resources

1. For more on the life of George Berkeley, go to

[Link]
2. For more information on Frederic Bartlett, go to
[Link]
 3. For a different view of language in nonhuman primates, go to the
following website: [Link]
4. For more information on bilingual representation, go to
[Link]

Descriptions of Images and Figures

Back to Figure

The chart has three rows. The bottom row shows “George Berkeley” in
the middle with arrows leading to “UC-Berkeley” on the left and to
“Charles Barkley” on the right. Three upward arrows lead up to elements
in the middle row as follows:

UC-Berkeley to UCLA
George Berkeley to John Locke
Charles Barkley to Michael Jordan

An upward arrow leads from John Locke to “a locksmith” shown in the

top row.

Back to Figure

The chart shows “Living thing” at the top represented by a circle. In the
hierarchy, at all levels below “Living thing”, concepts such as “tree”,
“bird”, “fish”, and “animal” are indicated by circles. Color-coded lines
branching out from each concept indicate “Is a” relationships and
“feature” relationships.

Feature arrows leading out from “Living thing” are labeled “can grow”
and “is living.”

The concept “Living thing” branches out into two concepts with “is a”
arrows – “plant’ and “animal” shown at the next lower level.

A feature arrow leading out from “plant” is labeled “has roots.” Feature
arrows leading out from “animal” are labeled “has skin” and “can move.”

The concept “plant” branches out into two concepts with “is a” arrows –
“tree” and “flower” shown at the next lower level. Feature arrows leading
out from “tree” are labeled “is big” and “has bark.” Feature arrows
leading out from “flower” are labeled “is pretty” and “has petals.”
The concept “animal” branches out into two concepts with “is a” arrows –
“bird” and “fish” shown at the next lower level. Feature arrows leading
out from “bird” are labeled “can fly” and “has feathers.” Feature arrows
leading out from “fish” are labeled “can swim”, “has scales” and “has
gills.”

The concept “tree” branches out into two concepts with “is a” arrows –
“pine” and “oak” shown at the next lower level. A feature arrow leading
out from “pine” is labeled “is green.” A feature arrow leading out from
“oak” is labeled “is tall.”

The concept “flower” branches out into two concepts with “is a” arrows –
“rose” and “daisy” shown at the next lower level. A feature arrow leading
out from “rose” is labeled “is red.” A feature arrow leading out from
“daisy” is labeled “is yellow.”

The concept “bird” branches out into two concepts with “is a” arrows –
“robin” and “canary” shown at the next lower level. A feature arrow
leading out from “robin” is labeled “is red.” A feature arrow leading out
from “canary” is labeled “is red.”

The concept “fish” branches out into two concepts with “is a” arrows –
“trout” and “koi” shown at the next lower level. A feature arrow leading
out from “trout” is labeled “is silver.” A feature arrow leading out from
“koi” is labeled “is orange.”

Back to Figure

The right side of the room has a table and two chairs – one along the
longer side of the table and the other along the width of the table. A
typewriter is placed on the table. The adjoining wall has three open
shelves that house items such as a globe, sheets of paper, a cane
basket, a bottle of wine, an umbrella, and some electrical equipment. In
the wall that is to the left is a felt board onto which a few charts are
pinned. A swivel chair is seen along this wall. Next to the chair is a
tabletop that is fixed to the wall. A kettle and flasks are kept on the
tabletop. A painting is fixed to the wall. Another set of table and a chair
are in the foreground. A few tools are placed on the table. The floor has
a small rug.

Back to Figure
The left graph shows the results of Rotation in picture plane. The
horizontal axis shows “Angle of rotation (degrees)” from 0 to 180, in
increments of 20. The vertical axis shows “Reaction time (seconds) from
1 to 5. The dashed line-of-best-fit begins at 1.4 seconds for a 0-degree
rotation and shows a rising trend, touching 2.5 seconds for a 60-degree
rotation, 3.8 seconds for a 120-degree rotation, and ending at 4.6
seconds for a 180-degree rotation. The reaction times (seconds), by
angles of rotation are as follows:

0 degrees: 1.4
20: 1.7
40: 2.0
60: 2.5
80: 2.8
100: 3.0
120: 3.8
140: 4.0
160: 4.4
180: 4.6

The right graph shows the results of Rotation around vertical axis. The
horizontal axis shows “Angle of rotation (degrees)” from 0 to 180, in
increments of 20. The vertical axis shows “Reaction time (seconds) from
1 to 5. The dashed line-of-best-fit begins at 1.2 seconds for a 0-degree
rotation and shows a rising trend, touching 2.3 seconds for a 60-degree
rotation, 3.1 seconds for a 120-degree rotation, and ending at 4.3
seconds for a 180-degree rotation. The reaction times (seconds), by
angles of rotation are as follows:

0 degrees: 1.2

20: 1.6

40: 2.0

60: 2.3

80: 2.7

100: 3.0

120: 3.1

140: 3.5
160: 4.0

180: 4.3

Back to Figure

The pennies have the text “In God We Trust”, “1978” and “Liberty”
placed in different positions. In a few, the image of Lincoln is shown
facing right, while in a few others, it is facing left. One of the pennies
also shows the text “One cent” below Lincoln’s image.

Back to Figure

It shows three levels of representation—one at the “Conceptual level”,

one at the “Lemma” level, and a third at the “Lexeme” level.

At the “Conceptual level” shown at the top, the associated words are
“Sweet” and “Fruit.” Below the words, a picture of a strawberry on a
plate is shown. The associated “Idea” is “Strawberry.” At the “Lemma”
level that is the next lower level, the associated words “Noun” and
“Singular” are shown. At the next lower “Lexeme” level, the word is split
into “Straw” and “berry.”