Conceptual Knowledge

Learning Outcomes

1. How are natural categories different from the classical approach?

2. Describe the components of the following approaches to knowledge representation:

- prototypes

- exemplars

- feature comparison model

- semantic network models

- connectionist approach

3. Explain the pros and cons of each model.

Categorization

________: class of objects, associated on the basis of some relationship

_______: mental representations of a category

- depends on ______________: identifying features common to all members of a conceptual class

- ______________: notice differences between conceptual categories

Bruner, Goodnow, & Austin (1956): benefits of categorization:

• reduces __________ of the environment

• means by which objects of the world are __________

• reduces need for constant ________

• allows us to decide what constitutes an appropriate ______

• enables us to order and relate _______ of objects and events

Classical Approach

Assumptions:

• categorization based on lists of ________ ________: those that are necessary to the meaning of the item

e.g., a triangle is a closed, three-sided figure

• features are individually necessary and collectively sufficient

Pros & Cons:

distinction between different categories is clear and logical

works for _______ categories:

e.g., geometric shapes, prime numbers

implies that all members are created _____

(Note: The textbook refers to the classical approach as the “definitional approach.”)

Natural Categories

Assumptions:

• groupings or clustering of objects or concepts that occur naturally in the real world

• _____ borders; membership may overlap

e.g., what are the defining features for the concept _______?

The Prototype Approach

Eleanor Rosch [née Heider] (1973)

Assumptions:

• items are comprised of a list of features or attributes

• concept organization based on _________: abstract, idealized item that is most typical category member

• ______________ features: describe the prototype, but are not necessary

• categorization of items is based on similarity to the prototype

• concepts are ______________ organized:

	animals
birds		fish
robin	bluejay

▸ _____________ level (e.g., fruit, animals)

- largest categories

- members of a category have few attributes in common

e.g., musical instruments

- tends to be abstract

▸ _____ level (e.g., grapes, apples)

- prototypes formed to represent the category

- members of a category share many attributes, and are highly differentiated from other basic categories

e.g., guitar, piano

- shows _______ ______: response is faster if item is preceded by a similar item

e.g., judgments about specific apples are faster if preceded by apple, than by fruit

- in general, people prefer to use basic-level names for things

- ________ learn basic level objects first (dogs), before they categorize in superordinate (animal) or subordinate (collie, hound)

▸ ___________ level (e.g., Thompson seedless, Concord)

- narrowest categories

- less ________ than basic categories

e.g., classical guitar, folk guitar

- _______ prefer to use subordinate-level terms; greater expertise greater use of sub-subordinate terms (Johnson & Mervis, 1997)

Evidence:

• __________ effect: items differ in how well they represent a category

e.g., rank order these items in their category:

vehicles: car, elevator, sled, tractor, train

clothes: jacket, mittens, necklace, pajamas, pants

• ______ ___________: each item has at least one shared attribute with another item in the category; is a kind of typicality

- good members of a category share many attributes with members of the same category

- but share few attributes with members of other categories

Rosch & Mervis (1975):

- generated list of category members

e.g., fruit: apple, banana, coconut, olive, orange, tomato

- asked participants to rate typicality of each member within category

e.g., apple as a member of fruit

- others listed attributes of category members

e.g., apple: red, sweet, crunchy, round

- strong correlations found between typicality rating and number of attributes shared by other category members

e.g., r = .85 for fruit (also, r = .88 for furniture, r = .91 for clothing)

- ______ ___________ is important to typicality

• brain activity

Kosslyn, Alpert, & Thompson (1995):

- participants placed in PET scanner

- saw picture of an item, and heard a word (e.g., “toy,” “doll,” or “rag doll”

- superordinate terms more likely to activate part of the __________ cortex (language, associative memory)

- subordinate terms more likely to activate visual attention areas

Pros & Cons:

accounts for concepts representing _____ groups

e.g., games--merely share a family resemblance

allows for ___________

explains how information is _______ to a single, idealized abstraction

but we also store ________ information about individual examples

number of potential ________ is very large (infinite?)

what determines feature weights?

how does expertise change categories?

counterintuitively implies categories have fuzzy boundaries

The Exemplar Approach

Assumptions:

• _________: members of a category that you have previously encountered (vs. prototypes, which are idealized)

e.g., dog concept is based on actual dogs you’ve seen

• first you learn specific exemplars of a category, then you classify new items based on their similarity to the exemplars

Heit & Barsalou (1996):

- asked participants for exemplars of various categories of animals

- measured exemplars’ typicality, and how typical the categories were of animals

- strongly correlated: r = .92

(_______ are most typical; ______________ the least)

- implication: our concepts are based on the most typical (i.e., exemplary) items

Pros & Cons:

accounts for typicality effect

no lists of features needed

no abstraction process needed

allows for “__________” to categories

e.g., penguins as birds

good for categories with ___ members

requires vast storage for individual members of large categories

Feature Comparison Model

(Smith, Shoben, & Rips, 1978)

Assumptions:

• concepts represented as a set of features:

• ________ features: essential, required features of a concept; are at the top of the feature list

• ______________ features: descriptive, but not essential (“loosely speaking”); are at the bottom of the list

e.g., birds:

- defining features = wings, feathers,...

- characteristic features = fly, sing,...

• ________ ____________ task: measure RT to correctly respond, “A robin is a bird.”

• relations between concepts are computed based on shared features; more features slower RTs

• two-stage model:

	Stage 1: compare ___ features (fast comparison)
low overlap		medium overlap				high overlap
	Stage 2: compare ________ features (slow comparison)
“false”		mismatch		match		“true”

Pros & Cons:

typicality effects: faster RT when item is a typical member of a category

e.g., “A robin is a bird.” (faster than) “A penguin is a bird.”

____/_____ effect: quick rejection of false sentences

e.g., “A pencil is a bird.” (faster than) “A bat is a bird.”

doesn’t explain ________ ____ effect: faster RT when item is member of a small category

e.g., “A robin is a bird.” (faster than) “A robin is an animal.”

- small categories have more defining features more stage 2 processing slower

can account for __________ of category size effect:

e.g., “Scotch is a liquor.” (slower than) “Scotch is a drink.”

not all defining features of a category are necessary

e.g., “Is a robin with no wings still a bird?”

features poorly _______; characteristic features have circular definition

Semantic Network Models

Common assumptions:

• concepts represented in network of interconnected _____

• concept defined in terms of ___________ to other concepts

Hierarchical-Network Model (Collins & Quillian, 1969)

Assumptions:

• ____ represents a single concept

• concepts organized hierarchically

• pathways represent associations between concepts

- “___” pathways: express category membership

- “____” pathways: express properties

• properties stored at the most general (“highest”) level possible, with no redundancy (cognitive economy)

• sentence verification via ____________ ______:

- “A canary is a bird.”

- “canary” and “bird” activated; activity spreads to neighbours

- both spreads eventually intersect, allowing answer to be made

Pros & Cons:

cognitive economy:

e.g., “A bird has feathers.” (faster than) “A bird has skin.”

corresponds well with category size effect:

e.g. “A canary...”

doesn’t explain violations of category size effect:

e.g., “A dog is a mammal.” (slower than) “A dog is an animal.”

typicality effects (model predicts they should not be obtained):

e.g., “A robin is a bird.” (faster than) “An ostrich is a bird.”

Spreading Activation Model (Collins & Loftus, 1975)

Assumptions:

• not hierarchical

• link length represents degree of relatedness; search time depends on link length

• passive concepts not in working memory; active ones are

• _________ __________: activation of one node leads to (partial) activation of connected nodes

• degree of activation decreases over distance and time

Pros & Cons:

accounts for: category size effects (& violations), typicality effects

Meyer & Schvaneveldt (1976):

- _______ ________ task (word/nonword):

- closely related concepts have shorter RTs

- ________ _______ effect: activation of a conceptual node facilitates retrieval of associated concepts or words

difficult to falsify: what determines link length?

Connectionist Approach

- a.k.a. Parallel Distributed Processing (PDP) or artificial neural network models

- classical approach: based on information & rules (serial approach)

e.g., mind is like a computer: data & programs

• ________:

- many (simple) processors working simultaneously

- models are artificial neural networks, analogous to human brain

• ___________:

- memory and information processing occur in the connections, not in storage locations (“connectionism”)

- information is distributed across the entire network, not localized

• __________:

- processing units = neurons; are interconnected

- active unit may pass along its activity via connections,  excitatory vs. inhibitory

- strength of connections may be modified--learning!

- damage resistant: partial network may still solve problem

Jets & Sharks Example (McClelland, 1981):

- information can be stored in a table:

- or distributed in a network:

Components:

• neuronally inspired

• nodes: processing units like neurons

• excitatory/inhibitory connections among nodes

• __________ rules: specify conditions for activating a node

• ________ rule: describes how connections can be changed, to improve performance

Pros & Cons:

biologically _________

so far, fairly ______ (but increasing in complexity)

as a model of a complex system becomes more complete, it becomes less understandable (______’s Paradox, 1963)

This document copyright © 2000-2024 Karsten A. Loepelmann. All rights reserved. Viewing this page is taken as acceptance of the copyright agreement.