Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 16

1

Conceptual Knowledge

 Knowledge refers to an acquaintance with facts, truths, or principles coming from study or
investigation
o It refers to what one knows.
 Conceptual knowledge: it is a sub-section of knowledge, and it refers to knowledge that
enables a person to recognize objects and events and to make inferences about their
properties
o How to tell a Labrador from a German shepherd etc. which we generally know the
answers to without thinking about. This type of knowledge comes in the form of
concepts that have been identified as “categories of objects, events, and abstract
ideas”
 A category includes all possible examples of a concept.
 Concepts provide rules for sorting objects into categories so they are often discussed together
and a lot of research has been focused on a process known as categorization which is the
process by which things are placed into categories
o This process allows us to better understand the world and it enables us to take action
in it
o Example: how to drive, you must know the basic mechanisms, rules of driving, and
road you need to take..)
o This process also allows us to better understand other people’s behaviors such as
seeing a person in a custom and then realizing it is Halloween so they must be
heading to a custom party
 There are various points are all to highlight the importance of using categories and without
this capacity, it would be difficult to manage the world as seeing a white cat for the first time,
for example, would be processed as a completely new and individual kind of data as opposed
to having access to one’s previously collected knowledge about cats and adding onto to that
they can be white. However, when dealing with unfamiliar objects, categorization is difficult
to rely on
 Unfamiliarity and brain damage can make categorization more difficult
 Basic Properties of Concepts and Categories
2

o How are objects placed into categories?


 A classic approach known as the definitional approach states that an object is
placed in a category if it meets the definition of said category which cognitive
psychologists have shown is fallible, thus we consider another approach
known as the prototype approach.
o Why definitions don’t work for categories?
 According to the definitional approach to categorization, one decides whether
an item belongs to a category based on its definition
 It has been observed that definitions work well for some items such as
geometric shapes (squares, triangles, etc.) but for most natural objects (trees,
plants, etc.) or human-made objects (chairs, tables, etc.) definitions don’t
work as well.
 To clarify, every item in a category may have different features. So by
definition, a chair is a piece of furniture made of a back, seat, legs, and often
arms designed to accommodate one person. Based on this logic, a chair
without a back and the rock wouldn’t belong to the chair category yet clearly
both can serve as a chair
 To solve this problem, philosopher Ludwig Wittgenstein (1953) proposed the
notion of family resemblance which refers to the idea that items in a particular
category resemble one another in several ways
 This allows for some variety in a category instead of having a set of
criteria for a category
 In the 1970s, Eleanor Rosch et al. applied this notion in a series of
experiments investigating the basic nature of categories which resulted in the
emergence of the idea of prototype
 The Prototype Approach: Finding the Average Case
o According to the prototype approach to categorization, belonging to a category is
determined by comparing to a prototype of said category.
o A prototype refers to a “typical” member of a category.
o Eleanor Rosch (1973) proposed that the typical prototype is based on an average of
members of a category that are commonly experienced
3

 such as the prototype of a dog which might be based on some usually seen
dogs like Labradors, huskies or golden retrievers but the prototype of a dog
that one may have in mind doesn’t necessarily have to look like any of these
dogs)
 thus a prototype is an “average” representation of a category
o variations with categories are described as differences in typicality in the sense that:
 high typicality: resembles prototype
 low typicality: doesn’t closely resemble prototype
o Rosch examined this by presenting participants with a category title such as “birds”
along with a list of 50 members of the category.
 Participants were asked to rate each item on a scale of 1 to 7, 1 referring to
closely resembling prototype while 7 meant that the item was a poor fit of the
category
 Results: on average one holds a certain prototype in mind as a sparrow was
rated as highly close to a prototype of a bird whereas a penguin was rated as
not a good example of a bird
 Prototypical Objects Have High Family Resemblance
o In 1975, another experiment by Rosch and Carolyn Mervis sought to examine how
good and bad examples of a category compare to other items within the category
o In this design participants were shown a list of items and were asked to list as many
features in common between these objects (word list example: chair, mirror, sofa,
telephone)
 Results showed that when an item’s characteristics overlap with the features
of many items in a category then the family resemblance of these two items is
high (so here this would apply to chair and sofa) as opposed to when there is
less overlap which means the family resemblance is low (this would apply to
mirror and telephone)
o These results led to the conclusion that there is a strong correlation between
prototypicality and family resemblance so good examples of a category would share
many features with other members of this category)
 Statements about Prototypical Objects Are Verified Rapidly
4

o In 1974, Smith et al. used a procedure known as the sentence verification technique to
determine how fast one could answer questions about an object’s category with either
yes, if the statement is true, or no
o Result showed that participants responded faster for objects that are high in
prototypicality as opposed to those that are low in prototypicality effect
o Conclusion: the prototype approach provided a lot of evidence showing that all items
within a category are not the same. Another approach that shows this is the exemplar
approach.

 The Exemplar Approach: Thinking About Examples


o This approach involves determining whether an object is similar to other objects
based on exemplars that refer to actual members of a category that a person has
encountered in the past, an exemplar of turtles would be your pet tortoise for instance
o Which approach works better?
 Each one has its advantages and it has been concluded that each provides its
kind of data which tends to work together to produce one’s conceptual
knowledge allowing for each kind of data to explain the task it’s most suited
for. A prototype would be a general idea (a dog) but an exemplar would be
specific to the person (my dog)
5

o Is there a psychologically basic level of categories?


 Categories can include different sub-categories that are more specific and
organized in a hierarchical organization
 Rosch’s Approach: What’s Special About Basics Level Categories?
o Rosch proposed that there are different levels of categorization ranging from general
to specific and that when one employs the use of categories there is a tendency to
focus on one of the following levels:
 The superordinate or global level
 The basic level
 The subordinate or specific level

o Her design included the following task-demonstration


6

o Results showed that participants listed a few features common to furniture but several
shared features for table and kitchen table
o Based on these results, Rosch proposed that the basic level is psychologically special
because going above it to the global level results in a loss of information whereas
going below it results in little gain

 How Knowledge Can Affect Categorization


o Research shows that on average, one is more likely to name things based on the basic
level but when one is an expert (such as a musician) then one can name items on a
specific level
o This led to the conclusion that to understand how one categorizes objects, one must
consider both the properties of an object as well as the learning and experience of the
person perceiving said object
 Network Models of Categorization
o Representing relationships among categories: semantic networks
 In this section we will be looking into how concepts or categories are
organized in the mind and we start by describing the semantic network
approach which proposes that concepts are organized in networks
 Introduction to semantic networks: Collins and Quillian’s hierarchical model:
o Fig.9.12 illustrates a simplified version of the model proposed by Collins and
Quillian (1969). It represents a network made up of nodes that are connected by
links in the sense that each node represents a concept/category and related
concepts are connected. The networks also represents properties linked to each
concept and the links connecting the concepts show how they are related in the
mind.
7

o The model is a hierarchical model that consists of levels ranging from global
concepts placed at high levels (“living things”) and more specific concepts found
at the bottom (“salmon”).
o According to Collins and Quillian, if we were to move across the network from
the node “canary”, we would have access to some information about it but to
access more data, one would have to go back to higher-level nodes. By moving up
a level from “canary”, we learn that it is a bird so it can fly and it’s also an animal
and a living thing.
o If not all birds can fly, then why is the property of flight added at the node of bird
and not canary? According to Collins and Quillian, adding the property of flight
to each bird that can would be inefficient and would take up a lot of space as most
birds can fly, hence why this characteristic is a better fit for the node “bird” This
technique of storing shared properties once at higher level nodes is identified as
cognitive economy.
o It is important to note that cognitive economy still allowed for exceptions which
would be found at lower level nodes such as penguins which would have the
added characteristic “can’t fly” thus making it different than most.
o Their model doesn’t mirror physiology (as in nodes and links don’t match cortical
areas), but rather it shows how concepts and their properties are linked mentally
and how one can retrieve properties associated with a concept.
8

o Their model predicts that the time it takes a person to retrieve information about a
concept is determined by the distance that must be travelled throughout the network.
And they tested this by using the sentence verification technique (asking participants
to respond yes or no to different statements) and measuring participants’ reaction time
to these statements such as a canary is a bird vs. a canary is an animal.
o Results showed that statements that required further travel from “canary” (so from
canary to animal), resulted in longer reaction times.
9

o Another prediction posited by their theory is spread activation which refers to activity
that spreads out along any link that is connected to an activated node.
o Due to spreading activation, other concepts that are indirectly activated become
“primed” and thus can be retrieved more easily from memory. So the word Robin can
activate the node bird which can activate other concepts linked to the category bird
and animal hence making that information more accessible

 Criticism of the Collins and Quillian Model


o Though their theory received a fair share of support, it also received its fair share of
critiques.
o It was observed that their theory does not account for the typicality effect (such as
longer RTs for “a penguin is a bird” compared to “a pigeon is a bird” even though
they are both at equal distance from the node of the bird).
10

o Other researchers put in doubt the notion of cognitive economy as it was shown that
some people do place specific properties of a concept (like has wings) right at the
node of that concept (so at the level of canary or ostrich etc.).
o Furthermore, it was shown through a sentence verification design experiment that
longer RT predictions were not consistent as the model suggested that it ought to be
quicker for someone to ascertain that a dolphin is a mammal vs. a dolphin is an
animal and yet people showed faster RTs for a dolphin is an animal.
o All these critiques eventually led to the proposal of a new theory in the 1980s known
as connectionism which we explore next.
 The Connectionist Approach
o In 1986, James Mclelland and David Rumelhart proposed a new approach known as
connectionism and it was well favored by researchers for two main reasons: because
it is inspired by how information is represented in the brain and because it can explain
several findings such as how concepts are learned and how damage to the brain
affects people’s knowledge about concepts
o What is a connectionist approach?
 Connectionism refers to a network model of mental operations that proposes
that concepts are represented in networks that are modeled after neural
networks. This approach relies on creating computer models for representing
cognitive processes.
 These models are also known as parallel distributed processing (PDP) model
as they propose that concepts are represented by activity that is distributed
across a network
 Fig.9.17 illustrates a simple connectionist network.
 The circles represent units that are inspired by neurons. Concepts and their
properties are represented in the network by the pattern of activity across these
units. The lines represent connections that transfer data between units (similar
to axons).
 There are 3 main types of units; input unit (which are activated by stimuli
from the environment) which sends signals to hidden units which then
transmit signals to output units.
11

 Another property of this model is connection weights which determine how


signals sent from 1 unit either increase or decreases the activity of the next
unit
 These represent what happens at a synapse that sends signals from 1
neuron to another.
 As we’ve seen, some synapses can transmit signals more strongly than others
and this results in a high firing rate in the receiving neuron (chap.7, LTP);
others can cause a decrease in the firing rate of the next neuron. Connection
weights function similarly; high connection weight strong tendency to excite
the next neuron, a low connection weight has less tendency to excite, and
negative weight decreases excitation or inhibit the activity of the receiving
neuron.
 To summarise the activation of units in a network depends on the signal that
originates in the inputs units and the connection weights throughout the
network.
12

 How are Concepts Represented in a Connectionist Network?


o Training a network:
 According to connectionism, the answer to “a canary can…” is represented in
a network by activation of the network’s representation and hidden units and
this had to be trained to achieve this
 a network might respond before training wherein stimulating “canary” and the
relation unit “can” would send out activity to the rest of the network-based
 if the connection weight between all units is 1.0 then the activity would spread
throughout the network and activate property nodes unrelated to “canary”
such as “flower” or “bark”. Thus for the network to operate efficiently, the
connection weights have to be adjusted in a way that ensures that activating
the concept only specific property units such as move, grow, and fly..
13

 This learning process takes place when a wrong response unit activated (such
as “flower”) sends an error signal back through the network through a process
known as back propagation which refers to a process in which an error signal
is transmitted backward through the network which provides the information
needed to adjust the connection weights in a network, to achieve the correct
output signal for a stimulus  hence learning.
 Basically connectionist networks are created by a learning process that shapes
the networks in a way that information about each concept is contained in the
distributed pattern of activity across several units, which is very different from
the hierarchical model.
 There are several results which support connectionism but two main ones are
the following:
 It was observed that brain damage does not completely disrupt the operation
of a connectionist network which is due to how information is distributed
amongst different units within a network  This disruption of performance
which only occurs gradually as parts of the system are damaged is known as
graceful degradation and it is similar to cases of brain damage wherein there is
partial loss of function.
 Researchers also showed that connectionist networks can explain the
generalization of learning as similar concepts have similar patterns so training
a system to recognize the properties of one concept also provides information
about other similar concepts.
 How Concepts Are Represented in the Brain
o Four proposals about how concepts are organized in the brain:
 The first approach proposed was known as the sensory-functional hypothesis
and it was based in studying brain damage patients who had lost the ability to
understand specific types of concepts
o The sensory-functional hypothesis:
 in a classic paper in neuropsychology, in 1984, four patients were reported to
have memory impairment due to brain damage caused by encephalitis;
specifically these patients suffered from category-specific memory
14

impairment which refers to an impairment in which they had lost the ability to
identify 1 type of object (animals) but were still able to identify others
(flowers, tools etc.).
 Researchers suggested that the difference between perceiving living things
(animals) and non-living things (tools, plants, etc.) relied on the sensory
features as non-living things are better recognized by their function instead of
their sensory features
 This observation led to the proposal of the sensory-functional (S-F)
hypothesis which states that our ability to differentiate between living
things and artifacts depends on a memory system that distinguishes
sensory attributes and a system that distinguishes function.
 However, it was observed that several cases of people with brain damage
could not be explained by the S-F hypothesis by simply distinguishing
between sensory and function as some people were able to identify animals
instead of artifacts while others were able to distinguish large artifacts but not
small ones for instance which led to the proposal of other theories.
o The Multiple Factor Approach
 The notion of a distributed representation is central to this approach as it seeks
to describe how concepts are represented in the brain by searching for
multiple factors that determine how concepts are divided within a category
 so grouping objects in terms of similarity should be based on more than one or
two features.
 Researchers observed that unlike mechanical devices/artifacts, animals tend to
share many properties which was termed as crowding and according to this
idea, patients with category-specific impairment have difficulty distinguishing
between items that share similar features which is why it’s harder for them to
recognize animals for example. Other results showed that some artifacts can
involve both sensory properties as well as function.
o The semantic category approach:
15

 It proposes that there are specific neural circuits for some specific categories.
It was posited that there are a limited number of categories that are innately
determined because of their importance for survival (recall FFA, PPA, EBA).
 According to this approach, there are specific brain areas that respond to
specific types of stimuli, and the brain’s response to items from a particular
category is distributed over several different cortical areas  So, identifying a
face is based on activity in the FFA and activity in other areas which respond
to emotion, facial expressions, attractiveness etc.
o The Embodied Approach
 This approach states that our knowledge of concepts is based on the
reactivation of sensory and motor processes that occur when we interact with
the object.
 Based on this approach, when one uses a hammer, sensory areas are active in
response to its size, shape and color, as well as motor areas that are involved
in carrying out actions using a hammer. Later, if one were to see a hammer or
read the word hammer, the same sensory and motor areas are activated as
when one was interacting with the hammer, and this pattern of activation is
said to represent the hammer (recall perception action pathways and mirror
neurons, chap.3)
 thus thinking about a concept or interacting with it, activates both
perceptual and action pathways.
 Studies exploring this have shown a similar cortical activation pattern between
doing actual movement and reading action words which demonstrated a
phenomenon known as semantic somatotopy
 this refers to a correspondence between words related to specific parts of the
body and the location of brain activity associated with that part of the body.
 Other researchers have observed that the ability to represent motor activity
associated with an action is not necessary for recognizing objects and that this
approach does not explain one’s knowledge of abstract notions such as truth.

Summary
16

• One thing that all of the approaches agree on is that information about concepts is
distributed across many structures in the brain, with each approach emphasizing
different types of information. The multiple-factor approach emphasizes the role of
many different features and properties. The category-specific approach emphasizes
specialized areas of the brain and networks connecting these areas, and the
embodied approach emphasizes activity caused by the sensory and motor properties
of objects. It is likely that, as research on concepts in the brain continues, the final
answer will contain elements of each of these approaches as this is still an active
debate.

You might also like