Knowledge (Exam 3)

Knowledge

Knowledge

• Concept = an idea or piece of knowledge about something.

• Category = a group of related concepts.

  • Example: dining room chair and living room chair are both in the chair category.

• Categorization = putting new information into the right group.

  • This helps us…

    • Identify objects.

    • Ignore the variability between the objects in a group.

    • Reduce the need for constant attention to every individual object.

    • BUT categorization is also responsible for a lot of memory errors.

Categorization Example: Color variability

Color variability

• We can see about 7 million different colors.

• But we only use about 7–10 basic color names.

• So, we put many similar colors into the same color group.

• This helps us ignore small color differences.

Categorization Approaches

Categorization Approaches

• Definitional Approach: something must have certain required features to be in a category.

  • A triangle must have 3 sides.

• Family Resemblance: things in a category are similar in many ways, but no one feature is required.

  • Example: most birds fly, but penguins do not.

• Prototype Approach: we compare something new to the best example (Prototype in our mind) of a category and pick the closest match.

  • A sparrow is judged as a bird more easily than a penguin because it is closer to the best example of a bird.

Rosch (1975)

Rosch (1975)

• A prototype is the average best example of a category.

• The prototype is usually not a real exact member of the category. Its your ideal.

• The prototype can change as you see more examples.

• Some things are very close to the prototype:

  • high prototypicality

  • easier to put in the category

• Some things are less close to the prototype:

  • low prototypicality

  • harder to put in the category

• Example:

  • sparrow = more typical bird

  • penguin or bat = less typical bird-like example

• Main idea: we often sort things by how much they match the best/average example of a group.

Techniques for Studying Prototypicality (4)

Techniques for Studying Prototypicality

• Object naming = when people name things in a category typical items are named first.

  • Example: robin before penguin.

• Prototype priming = when someone is primed they respond faster when the item matches the prototype.

• Feature overlap = count how many features an item shares with the prototype and if there are more shared features then it is a more typical item.

• Category verification task = show a category and an object, then ask if it belongs.

  • People answer faster for more typical members.

    • Example: apple is judged as fruit faster than pomegranate.

Category levels

Category levels

• Superordinate level = very broad group

  • example: furniture, vehicle

• Basic level = middle group

  • example: chair, car

• Subordinate level = specific type

  • example: dining room chair, pickup truck

• Main idea: categories go from broad → middle → specific.

Basic Level Advantage (Rosch et al., 1976)

Basic Level Advantage

• We usually categorize things first at the basic level.

• Example:

  • furniture = superordinate

  • chair = basic

  • dining room chair = subordinate

• People were fastest at saying yes to the basic level.

• Example:

  • chair = fastest

  • furniture = slower

  • dining room chair = slower

• In naming tasks, people usually call an object by its basic level name.

• Example: they say “chair”, not “furniture” or “dining room chair.”

• Babies also tend to learn basic level words first.

• Main idea: the basic level is the easiest and most natural level to use.

What causes basic level advantage?

What causes basic level advantage?

• A good category does 2 things:

  • members in the category share many features

  • members of different categories share few features

• Superordinate level is too broad:

  • good at being different from other big groups

  • bad because members do not share many features

  • example: furniture

• Subordinate level is too specific:

  • good because members share many features

  • bad because it overlaps a lot with similar groups

  • example: dining room chair and office chair

• Basic level is the best middle point:

  • members share many features

  • but still stay different from other categories

  • Example:

    • chair works well because chairs share things like seat and back

    • but chairs are still different from tables, beds, and lamps

• Main idea: the basic level is easiest because it is the best balance between similarity within the group and difference from other groups.

Tanaka & Taylor (1991)

Tanaka & Taylor (1991)

• Study asked if experts think about categories differently than non-experts.

• They tested bird experts and non-experts with a naming task.

• Bird experts named birds at a more specific level.

• Example: “robin” or “hummingbird.”

• Non-experts used the more general basic level name.

  • Example: “bird.”

• Main idea: experts use more specific categories faster and more naturally.

• For experts, what is specific to other people can feel like the basic level to them.

Rosch et al. (1976)

Rosch et al. (1976)

• People listed shared features for categories at 3 levels:

  • superordinate

  • basic

  • subordinate

• Superordinate categories had only a few shared features.

• Basic categories had many shared features.

• Subordinate categories had a little more than basic.

• But subordinate categories overlap more with other similar groups.

• Main idea: the basic level gives the best amount of useful shared information without too much overlap.

Hierarchical model (Collins & Quillian, 1969)

Hierarchical model (Collins & Quillian, 1969)

• Knowledge is stored in a hierarchy from general to specific.

• Each concept is a node.

• Nodes are connected by links.

• Each node has properties (features).

Higher categories pass their properties down to lower categories.

• Example:

  • if birds can fly, then a canary can fly because it is a bird

Lower categories do not pass properties upward.

• Example:

  • if birds can fly, that does not mean all animals can fly

Cognitive economy = Saves space and effort by not repeating the same information again and again

• To answer a question, the mind must move through the hierarchy

• So:

  • “A canary is a bird” is faster

  • “A canary is an animal” is slower

  • “Canaries can fly” means going up to the bird level to find can fly

• Main idea: the model saves space, but longer travel in the hierarchy takes more time.

Assumptions of the hierarchical model notes

Assumptions of the hierarchical model notes

• To get information, you must go to the node where that information is stored.

• If a lower item is different from the higher rule, the lower item needs an exception listed.

• Example:

  • birds can fly

  • but ostrich = can’t fly

• Getting information takes time.

• Moving from one node to another also takes time.

• These times add together.

• Main idea: the farther you have to travel in the network, the longer it takes to answer.

Testing the hierarchical model

Testing the hierarchical model

• Collins and Quillian tested the model with a sentence verification task.

• People had to decide if a sentence was true or false.

• Category statements only need moving through the hierarchy.

  • Example: “A canary is a bird.”

• Property statements need:

  • moving through the hierarchy

  • plus finding the property

  • Example: “A canary can fly.”

• So:

  • “A canary can sing” is faster than “A canary can fly”

  • because can sing is at the canary level

  • while can fly is at the bird level

• Also:

  • “A canary is a bird” is faster than “A canary is an animal”

  • because it takes fewer steps

• Main idea:

  • more levels traveled = slower

  • needing a property search = slower

  • both times add together

Spreading Activation Model

Spreading activation model

• The old Collins & Quillian model had problems.

• It did not explain typicality well.

• It wrongly suggests:

  • “a canary is a bird”

  • and “an ostrich is a bird”

  • should take the same time

• But ostrich is a less typical bird, so it takes longer.

• It also got some results wrong.

• Example:

  • “a pig is an animal” is verified faster than

  • “a pig is a mammal”

• The old model did not predict that well.

Collins & Loftus (1975): Spreading activation model

Collins & Loftus (1975): Spreading activation model

• Like the old model, ideas are connected by links.

• Moving along links still takes time.

Main differences

• It is not hierarchical.

• Information does not have to be stored in only one place.

• Properties are treated like concepts too.

  • Example: “can sing” and “is yellow” act like concepts

How it works

• When one concept becomes active, activation spreads to linked concepts.

• Shorter links = stronger connection

• Longer links = weaker connection

• If a concept has many links, less activation goes down each one.

Final assumptions

• Activation fades over time.

• If activation gets high enough, you think of that concept.

• Then activation spreads again, so thoughts can move from one idea to another.

Main idea:
Knowledge works like a web of linked ideas, and activation spreads through the web, which helps explain typicality and why some ideas are reached faster than others.

Priming and spreading activation (4)

Priming and spreading activation

• Priming = seeing one thing helps you process something later faster.

• Repetition priming = the same thing is easier the second time.

• Associative priming = a related thing makes another thing easier.

• Main idea: the more two things are linked in memory, the more they help activate each other.

• Lexical decision task = decide if letter strings are real words or not.

  • Example:

    • BUTTER = yes

    • BUFLER = no

Meyer & Schvaneveldt (1971)

Meyer & Schvaneveldt (1971)

• Researchers changed whether the 2 words were related or not related.

• Related words:

  • bread – wheat

• Not related words:

  • chair – money

• People answered faster for related words.

• People answered slower for not related words.

• This shows associative priming.

• It also supports spreading activation:

  • one related word activates the other

  • so it is easier to recognize both

• Main idea: related words have lower reaction time than unrelated words.

Neural network models

Neural network models

• Knowledge can be shown with network models.

• These models use spreading activation.

• A concept is represented by a pattern of activity across many nodes.

• Links between concepts depend on how activation spreads from one concept to another.

• Main idea: ideas are stored as connected patterns, not just single separate spots.