PSYC 317 Ch. 9: Knowledge

Prototype

"Typical" example of a category - An average representation of "typical" member of a category - Encompassed characteristic features that describe what members of that concept are like - An average of category members encountered in the past

Global, basic, specific levels of hierarchical organization

*Global:* superordinate; ex. furniture *Basic:* chair, table, bed *Specific:* subordinate; kitchen, dining room, single, double

High v. low prototypicality (Rosch)

*High:* category member closely resembles category prototype - "typical" member ex. bird = robin, sparrow, cardinal *Low:* category member does not closely resemble category prototype ex. bird = penguin, bat, ostrich

How is Exemplar approach like/unlike the Prototype approach?

*Similarities:* - Representing a category is not defining it - Can explain Rosch's prototype experiments *Difference:* - Representation is not an average, but rather, specific examples

Exemplar Approach to categorization

- Actual category members - To categorize, compare the new item to stored exemplars that have been encountered before - Gameboy mug example: you're comparing the mug to other graphic mugs you've encountered, rather than a definitional or prototype approach, so it's much easier/faster to determine that it's a mug

Describe Tanaka & Taylor's (1991) experiment on the role of experience in how people categorize

- Asked bird experts & non-experts to name objects in images from various categories - In response to bird images, experts named the specific (species) bird & non-experts responded with "bird" Experts have more knowledge of the details that differentiate birds & many more exemplars of the category, which shows how experience affects individuals' categories.

Explain Rosch's (1975) study on prototypicality

- Gave participants lists of several examples of a category & asked them to list characteristics that pairs of the objects had in common *Results:* those that have a large # of overlapping characteristics have high family resemblance -> strong positive relationship btwn prototypicality & family resemblance - Low overlap = low family resemblance - Low prototypicality = low family resemblance

According to Rosch's (1976) experiment on hierarchical organization, why is the basic level "special"/more important?

- Going above the basic level (global) results in a large loss of info. - Going below basic level (specific) results in little gain of info. - People tend to generate basic level names of objects rather than global or specific *Experiment:* - Participants listed features common to each level of category (furniture, table, kitchen table) - Greatest # of features in common were for the specific level, but very few additional features over the basic level

Describe Collins & Quillian's (1969) study on RT and semantic networks

- RT for someone to respond should correspond to the distance btwn nodes in the semantic network - Participants responded "yes/no" to statements; those that required a greater distance of "traveling" over semantic network will result in longer RTs

Semantic category approach (category representation in the brain)

- Specific neural circuits in the brain to process & identify specific semantic categories (living v. nonliving) - While some regions are specialized to process certain categories/information (e.g. faces), identifying objects also depends on networks of activity across cortical regions - Innate neural circuits evolved to be specialized over time to aid survival

Meyer & Schvaneveldt's (1971) Lexical decision task (semantic networks)

- Tested spreading activation theory w/ this - Participants read stimulus & are asked to say as quickly as possible whether the stimulus is a word or not - Presented pairs of words/non-words - "yes" if both were words, "no" if not - Some pairs were closely associated - RT was faster for words that were closer nodes (so more closely related concepts) in the semantic network

List the criticisms of Collins & Quillian's Semantic Network

1. Cannot explain typicality effects - Semantic network model would predict equally fast responses for "a canary is a bird" & "an ostrich is a bird" - both 1 node from bird - Some sentence-verification results are problematic 2. Cognitive economy - Some evidence that properties of object are stored @ specific nodes rather than at higher levels - Led to the proposal of the connectionist approach

Pros of Exemplar Approach

1. Explains typicality effect 2. Easily considers atypical cases - compares penguin to other examples of birds that don't fly, rather than an averaged prototype like a robin 3. Easily deals w/ categories that have a lot of variation

List pros of the connectionist approach

1. Graceful degradation 2. Slow learning process that creates a network capable of handling a wide range of inputs 3. Learning can be generalized - learning about 1 concept often provides info about related topics 4. More complex than semantic network models, but more accurate to what is occurring in the brain - can simulate cognitive activity for processes such as learning & memory

Why are categories useful?

1. Help to understand individual cases not previously encountered 2. "Pointers to knowledge" - Provide a wealth of general info about an item - Allow us to identify the special characteristics of an item 3. Share many "global" characteristics, but are also different for people w/ different experiences We are constantly relying on background info about objects/events, which enables us to understand everyday situations.

What 2 things have to be considered to fully understand how people categorize objects? (Hierarchical organization)

1. Object properties 2. Learning & experience of perceivers

Typicality Effect (Prototype Approach)

1. Prototypical objects are processed preferentially - High prototypical objects judged more rapidly (Smith et al. 1974) 2. Prototypical objects are named 1st when producing a list of objects in a category 3. Prototypical category members are more affected/"activated" by a prime - Rosch (1975b): hearing "green" primes a highly prototypical "green"

Describe Collins and Quillian's (1969) semantic network model for how concepts & properties are associated in the mind

1. Semantic memory is organized into networks of nodes & pathways - networks list additional features/characteristics of objects/concepts - Can be listed from broad (global) categories to more specific (in accordance w/ hierarchical model) 2. Activation of a given concept/category *(node)* spreads to related nodes 3. Each link represents a proposition of the concept - diff. links can represent diff. pathways 4. Spreading activation can prime activation of another related concept - Concepts are linked together w/ lines; related concepts are linked - Goal was to create computer model of human memory.

Sentence Verification Technique (Prototype Approach) - Smith et al. (1974)

A technique in which the participant is asked to indicate whether a particular sentence is true or false. Ex. "An apple is a fruit" = fast response; "A cucumber is a fruit" = slower response

Exemplar

An example or model, especially an ideal one

Patients with some impairment may have a hard time distinguishing objects within a certain category because of crowding. Based on the idea of "crowding," which category can be hard for some patients to distinguish objects within that category?

Animals

Connectionist approach/connectionism

Based on how info is represented in the brain - Explains how concepts are learned & how brain damage affects concept knowledge - Proposed by Rumelhart & McClelland *Connectionism:* create computer models to represent cognitive processes

Hierarchical organization

Broader, more general categories are divided into smaller, more specific categories 3 Levels of Organization: 1. global 2. basic 3. specific

Prototype Approach to categorization

Category membership determined by comparing objects to prototype that represents the category - Instead of using definitions/key functions, we use "typical" examples (AKA prototypes) - Still does not account for the problem w/ characterizing different types of birds that don't look alike/have the same functional properties (flying)

Semantic networks approach

Concepts can be arranged in networks that represent the way concepts are organized in the mind

Parallel distributed processing (connectionist approach)

Connectionist model that proposes that knowledge is represented in the distributed activity of many units - Concepts are represented by the pattern of activity across units (similar to population & sparse coding)

Which kind of organization is best represented in the brain?

Connectionist network

Semantic somatotopy (embodied approach of category representation in the brain)

Correspondence between words related to specific body parts & location of brain activation ex. reading/hearing about someone eating can cause activation in the "mouth" region of the somatotopic map - *fMRI evidence:* activation that occurs during executing a movement occurs in the same area when reading action words related to that movement

Category-specific memory impairment

Damage caused inability to identify 1 type of object (category) but retained the ability to identify other objects - Double dissociation for categories "living things" & "nonliving things" - this has guided theories about category representation

Connection weight (connectionist approach)

Determine at each connection how strongly a signal will activate the next unit & whether activity @ next unit is increased or decreased - Corresponds to synaptic activity/neurotransmitters

Definitional Approach to categorization

Determine category membership based on whether the object meets the definition of the category - Works fine for simple categories (ex. shapes) - *Overall doesn't work well* - not all members of everyday categories have the same defining features ex. 4 chair example

Error-signal (connectionist approach)

Difference btwn actual activity of each output unit & the correct activity - Cause corrections to the connection weights & in the resulting activation pattern - continues until correct

Graceful degradation (a pro of the connectionist approach)

Disruption of performance occurs only gradually when parts of the system are damaged - Bc representation is distributed across many units in the network, it continues to operate w/ damaged regions - Similar to cases of human brain damage - it's not a total breakdown of operation, just a loss in parts of function. Rehab can allow the creation of new connections & improve certain functions

Back-propagation (connectionist approach)

Error signal transmitted back through the circuit - Indicates how weights should be changed to allow the output signal to match the correct signal - The process repeats until the error signal is zero, & the correct activation is achieved

Mirror neurons

Frontal lobe neurons that fire when performing certain actions or when observing another doing so. The brain's mirroring of another's action may enable imitation, language learning, and empathy.

Multiple factors approach (category representation in the brain)

How are concepts further divided up within a category? - Degree to which various features (color, motion, function) & their associated networks are important to object categories

Within the connectionist network approach, pattern across activation of units may be similar across unfamiliar items or objects. As learning progresses, what allows for the distinguished changes in the appropriate pattern of activations? (pick the best answer)

Incorrect responses send error signals back to hidden units, leading to adjusted connection weights.

Units (connectionist approach)

Inspired by neurons; lines connecting units correspond to axons *Input units:* activated by stimulation from environment *Hidden units:* receive input from input units *Output units:* receive input from hidden units

Which is the best model for categorization?

It depends on the size of the category, but exemplar and prototype

Which of the following is NOT true about using definitional approach to categorization?

It works very well for complex objects

Conceptual knowledge

Knowledge that enables us to: 1. Recognize objects & events 2. Make inferences about their properties

Inheritance (semantic networks)

Lower-level items share properties of higher-level items - There are exceptions for which the higher node doesn't accurately describe the lower node (ex. not all birds can fly)

Concept

Mental representation; the meaning of objects, events, or abstract ideas

Connectionist network (connectionist approach)

Must be "trained" for a pattern of activation to occur in response to a stimulus *How learning occurs:* 1. Network responds to stimulus 2. Some correct & incorrect unit activation 3. Provided w/ correct response 4. Weights are adjusted to match correct response - As more info is learned about a concept, more units will become active, & others we learn are not related & will be corrected to not be activated - Accounts for interactions between objects - some objects' activation patterns look similar bc the objects are similar

Sensory-functional hypothesis (category representation in the brain)

Our ability to differentiate living things & artifacts/objects depends on a semantic memory system that distinguishes sensory attributes & a system that distinguishes function - alternate explanation for double dissociation of objects & living things - Living things tend to be distinguished by sensory features (appearance) - Objects distinguished by sensory function/uses - Imperfect evidence: while some case studies offer evidence, others don't support this hypothesis

Embodied approach (category representation in the brain)

Our knowledge of concepts is based on reactivation of sensory & motor processes that occur when we interact w/ the object - mirror neurons: neurons in premotor cortex that fire when we do a task/when we observe another doing that same task - semantic somatotopy: correspondence btwn words related to specific body parts & location of brain activation

How does a priming stimulus affect RT for prototypical members vs. non-prototypical members? - Rosch (1975b)

Priming effect allows you to develop a mental representation/concept before being shown the object; so having a mental representation/concept readily available in WM allows you to respond quicker when the object matches that concept - Rosch (1975b): hearing "green" primes a highly prototypical "green"

Cognitive economy (semantic networks)

Shared properties are only stored @ higher-level nodes - Follow the lines to see which concepts are connected: more specific objects contain the qualities contained along higher up along the links - Including characteristics @ each specific node (rather than basic-level node) would take up too much storage to work efficiently - There are exceptions for which the higher node doesn't accurately describe the lower node (ex. not all birds can fly)

Family resemblance effect

The more similar a specific exemplar is to a known category member, the faster it will be categorized

Categorization

The process by which objects are grouped - Occurs often & helps us understand our surroundings - Gives us additional info about objects that may be new to us - Often occurs w/o us realizing

Family Resemblance Approach to categorization

Things (definitions) in a category resemble one another in several ways - Similarities & relationships used to categorize - Proposed as the "solution" to definitional approach problem (doesn't include all members in a category) - allows for variation within a category

True/False: it's possible that people use both prototypes & exemplars to categorize

True: exemplars may work best for *small* categories; prototypes may work best for *large* categories

Spreading activation (semantic networks)

When a node is activated, activity spreads out along all connected links - Activation is the "arousal level" of a node - Concepts that receive activation are primed & more easily accessed from memory ex. if someone hears/sees "robin", then "bird" node is activated bc it's a closely linked concept. This activation then spreads along the links to other related concepts, & "primes" connected nodes for later use. People will now respond faster to any of these activated nodes

Crowding (multiple factors approach of category representation in the brain)

When different concepts within a category share many properties - ex. "animals" all share "eyes", "legs", and "the ability to move" - Category-specific memory impairment (naming animals) may be result of inability to differentiate btwn objects w/ similar features