Knowledge - UNCW Faculty and Staff Web Pagespeople.uncw.edu/tothj/PSY410/PSY410-08-Knowledge-color.pdf · knowledge examines (a) knowledge ... things to say to scared child, ... categories

Knowledge1. Semantic (Associative) Memory. Measuring Semantic Memory.

Hebb's Law & Hebbian Circuits.

The Hierarchical Semantic Activation Model.

2. Concepts & Categories. C&C's defined.

Four key observations. The process of categorization: Four theories.

Why are C&C's important?

The Spreading Activation Model. Connectionist Models.

Implicit Learning.

Semantic vs. Episodic Memory

Semantic & Episodic Memory are both forms of __________ knowledge.declarative

Episodic - Autobiographic events; includes info about place & time; details; "source".

• Is a butterfly a bird?

Semantic - Facts about the world that are not tied to specific events. Aka. General knowledge; Associative memory*.

• Was "butterfly" on the study list?

Semantic Memory Three major questions define research on

semantic memory (knowledge):

1. How is knowledge represented & organized?

2. What processes are involved in the acquisition and application of knowledge?

3. How is knowledge stored in the brain?

• Representation: Words? Sensations? Actions?• Organization: Hierarchical? Associative Strength?

• Activation/Retrieval & Decision Processes.

• Areas involved? Local or Distributed? Category vs. Property-based? Embodied?

Measuring Semantic Memory

Naming & Lexical Decision Tasks (RT).• Picture Naming. Attribute Generation.

• Semantic Priming in LD: e.g., Doctor > Nurse.

Sentence-/Category-Verification Tasks (RT).• Category: Is a cat a mammal? • Property: Does a cat have claws?

Association* Tasks.

• Category Association: Fruit - ?• Free-association (cf. Freud): Party - ?

• Word gen. tasks: e.g., Verbal Fluency (FAS) test.

• Word Naming: "hint" vs. "pint"; "flag" vs. "fugue".

Sentence Verification DataCollins & Quillian (1969)

Hierarchical Semantic Network ModelCollins & Quillian (1969)

• Note how this model exhibits cognitive economy.

• Note how this model is hierarchical.

Problems with a strict hierarchy

Frequency effects.

Typicality effects.• A robin is a bird < A penguin is a bird.

• Verification faster for more typical exemplars.

Negative-judgment effects.• A canary is a robin > A canary is a tulip.

• Verification faster for more distant associations.

• Faster RT despite higher level in hierarchy.• A dog is an animal < A dog is a mammal.

[ < & > refer to Reaction Time (RT) results]

Spreading Activation ModelCollins & Loftus (1975)

Node: A representation of a concept; a pattern of neural activity associated with a thing or idea.

Connections: Associations between concepts, with distance = strength of association.

Spreading Activation: Activity in one node spreads to others, increasing their activation.

Assumptions: Associative, not hierarchical; Activation decreases with time & distance.

Long-Term Potentiation (LTP).• Selective strengthening of synapses.

Hebb's (1949) Law.• if Neuron A repeatedly helps fire Neuron B, the

connection between them will be strengthened.

Neural Basis of Associative Memory

neurons that fire together wire together

• Increased excitability of pre & post-synaptic neuron.• Growth of new dendritic connections.

Woof“Doggie”

Hebbian Circuits

The Neural Basis of Associative/Semantic Memoryneurons that fire together wire together

nodes, connections, distance, spreading activation Spreading Activation Model

In addition to semantic priming, this model can nicely explain frequency, typicality, & negative-judgment effects.

Connectionist ModelsUnits: Neuron-like processing nodes that take on values.

(aka. Neural Networks; Parallel Distributed Processing)

Connections: The links between units (cf. axons); connections have (+/-) strngths or "weights" that determine how a unit is affected by activation.

Hidden Units: Units that have no connection with the outside world (cf. interneurons).

Key features of PDP networks are that knowledge is (a) distributed rather then local; and (b) embedded in the links between nodes, rather the being directly represented. As well, PDP nets are good at (c) completing partial/messy patterns, (d) satisfying multiple constraints, and (d) generalizing to new (untrained) stimuli. Finally, (e) they show "graceful degradation" – the ability to still function when damaged (when units are removed or "injured").

The Neural Bases of Semantic Memory What does it mean to "know" something?

Research on the neural bases of knowledge examines (a) knowledge deficits in brain-damaged patients; & (b) neural activity during knowledge tasks.

Much of this research suggests that knowledge is widely distributed across different areas of the brain.

The Neural Bases of Semantic Memory

Silently naming pictures of tools vs. animals produces very distinct patterns of neural activation (suggesting distinct semantic representations).

Similar results (distinctive pattern of activity) have been found with faces (the "FFA"), houses ("PPA"), musical instruments, etc.

Amnesic performing semantic task

Left prefrontal cortex is often highly active during verbal semantic tasks.

The Neural Bases of Semantic Memory








Implicit Learning.

Categories as the primary groupings by which we think, communicate, & learn.

Concepts and Categories Defined Category - A class or collection of things

grouped together on the basis of one or more common properties.• objects (animals, dogs, collies, cars, etc.)

• events (games, parties, arguments, wars, etc.)

• categories can also be "ad-hoc" (hairy things, fast things, etc.) or "goal-driven" (things to take on a camping trip, things to say to scared child, etc.).

Categories can be based on perceptual, biological, or functional properties.

Semantic/Associative Network

• concepts as the fundamental units of semantic/ associative memory (i.e., general knowledge).

Concept – A general idea or understanding about something; the mental representation of a category.

Concepts and Categories Defined

Hebbian circuit for "Dog"

The primary evidence that a person has (understands) a concept is that they can (a) positively classify (include) instances as members of a category; and (b) negatively classify (exclude) instances as not members of a category.

What does it mean to "have" a category?

Pattern recognition is an act of categorization

Types of Categories 1) Nominal categories are artificial (man-made),

logical, and well-defined (e.g., letters of the alphabet, geometric shapes, positive integers).

2) Natural categories are those that occur in everyday life. They are often, but not always, "given" by the world and are ill-defined (e.g., dogs, bachelors, games, furniture, running, crying, etc.).

3) Ad-hoc categories are a special case of natural categories and refer to things that satisfy a particular purpose or goal (e.g., things to take on a camping trip, things to save if your house is on fire, things to talk about on a first date, green things).

Why Are C&C's Important?

2. Allow for rapid identification & action.

3. Reduce the need for learning.

4. Provide the foundation for new learning.

"I'll bet that shark has sharp teeth!"

"although technically a fish, many sharks use a reproductive strategy highly similar to mammals".

1. Reduce complexity. 50 birds become "a flock";

50 people, "a crowd".

"A shark! Get out of the water".

Implicit LearningThe incidental learning of complex information (such as

statistical regularities) in the environment.

Two major paradigms for studying implicit leaning:

1. Serial Pattern Learning.

2. Artificial Grammar Learning.

"The process by which knowledge about the rule-governed complexities of the stimulus environment is

acquired independently of conscious attempts to do so" (Reber, 1989, p. 219).

Serial Pattern Learning• Subjects respond to

a series of lights by pressing matching buttons.

• Unbeknownst to the subjects, there is an underlying pattern to the sequence of lights (and thus to their responses).

• Additional evidence for implicit learning is provided by including random sequences* after learning has occurred; evidence for implicit learning is shown by a slow-down on this block of trials.

• Example: 3, 1, 4, 1, 3, 2, 3, 1, 4, 1, 3, 2, 2, 4, 3, 4, 1, 3, 2, etc.

Implicit serial pattern learning in dyslexic (circles) and control children (squares).

• As the task progresses, subjects respond faster to the items in the sequence, despite not knowing there was a sequence, or not being able to explicitly reproduce it.

1 2 3 4

Artificial Grammar Learning• Subjects are exposed to a

set of letters strings created by a finite state grammar.

• After study, subjects are told that the letter strings were based on an underlying set of rules (a "grammar").

• Subjects can discriminate between letter strings that do vs. do not correspond to the underlying grammar, even if the specific test strings were never shown in the exposure phase.

• Examples: TPTS, TTS, VVPS, VXVS, TTXXVS, VXVPXVS, TPTXVPS, etc.

A typical finite state grammar.

• When asked to describe the rules, few can. However…








Implicit Learning.

The Difficulty of Defining Common Things

• What is a game?

• As first noted by Wittgenstein, some things are less defined by definition than by family resemblance.

• What makes hide-and-seek a game?(a) played by children; (b) done for fun; (c) has rules; (d) involves more than one person; (e) is in some ways competitive; (d) done during periods of leisure.

The Ease of Defining Uncommon Things

• Consider the category… Things to save from your burning house

• As first noted by Basalou, many categories are "ad hoc" – defined by goals or temporary needs.

• Which of the following go into that category?

The finding, in numerous paradigms, that more "typical" members of a category hold an advantage over less typical members.

Typicality Effects

3501

4,447

The Flexibility & Stubbornness of Categories

• 1. Is it possible to transform a toaster into a coffee maker?

?

• 2. Is it possible to transform a skunk into a raccoon?

?

Four Key Observations about Concepts and Categories

1. Despite categorizing thousands of concepts, people find it difficult to say how they do that, or describe the info that defines a category.

2. People easily & rapidly create new (ad hoc) categories and use them appropriately.

3. People see some members of a category as better or more typical than other members, even when the category is defined "by rule".

4. People believe that (some) categories have an underlying essence (deep structure) that is more important than any perceptual features .








Implicit Learning.

1. Defining Features (rule-based categorization).

How do we categorize objects and events?

• only works well for well-defined/artificial categories (e.g., triangle; an "ace" in tennis).

Do you know what a "bachelor" is?

Categorization The fuzzy nature of categories...

• Bob's an unmarried man, but has been living with the same girl for 20 years. They are happy and Bob has no intention of changing his living situation. Is Bob a bachelor?

• James is an unmarried man with no girlfriend (or boyfriend). He's also a monk, living in a monastery, and has taken a vow of celibacy. Is James a bachelor?

• Robert's a married man, but has been separated from his wife for 10 years. He's actively dating, with no intention of ever remarrying. Is Robert a bachelor?

Categorization The fuzzy nature of categories...

• Table: coffee or display table bed, stool, counter

That would include:

Identify features of the following categories...

But not include:

• Bottle: pill or baby bottle jar, glass, carton

• Dog: Chihuahua, Greyhound wolf, coyote

• Furniture: phone, beanbag porch swing, car seat

• Fruit: Strawberry, Tomato olive, almond, squash



2. Resemblance (similarity-based categorization).

(b) Exemplars - specific instances of a category.

(a) Prototypes - an average of (abstraction across) many different instances (exemplars) of a category.

• No specific feature(s) that allmembers have; however, all members share a subset of relevant features, creating a “family resemblance”.

Concepts are represented as prototypes –a summary or average of the features that make up instances of the category (based on family resemblance).

Prototype View

• Prototypes have "fuzzy boundaries", thus making categorization somewhat ambiguous ("graded membership").

Typicality effectsEvidence for Prototypes

• Sentence verification (Smith, Rips, & Shoben, 1974).

• Shown a series of sentences and have to indicate whether they are true or false.

• An apple is a fruit.

• A fig is a fruit.• A German Shepherd is a dog.

• A potato is a fruit.• A bat is a bird.

• A St. Bernard is a dog.

Items close to the prototype are more quickly identified as members –categorizing as judging similarity to prototype.

• Picture identification (Rosch, Mervis, et al., 1976).


• Exemplar generation (Barsalou, 1983; 1985).

• Name 5 articles of clothing.• Name 5 examples of furniture.

• Name 5 types of weapons.

• Items close to the prototype are the earliest and most likely to be mentioned in these production tasks.







• Explicit judgments of cat. membership.






• Explicit judgments of cat. membership.

• Induction (extrapolation to novel instances).

X

Name these items.Prototypes

Prototypes have a hierarchical structure:(Eleanor Rosch)

How are they organized in semantic memory?

• “Basic” level categories have special status: (1) best balance of discriminability and info overload; (2) used most often in discourse; (3) develop first in children; (4) come to mind 1st & fastest; (5) central in memory errors.

Superordinate:

Subordinate:

Basic:

Fruit

Banana Grape Apple Melon

Granny Smith Golden Fuji etc...

Prototypes

Concepts are not abstract averages, but are made up of memory for prior encounters with specific objects and events.

Exemplar View

• Emphasis on basic memory mechanisms (e.g., ES, TAP) and the variability with which people process things for specific purposes.

• Abstract knowledge is computed "on-line",. rather then pre-computed (as with prototypes).

Evidence for Exemplars• Typicality effects.

• Availability of info about variability.

• Availability of info about correlation.

• Context-sensitivity of categorization.

• Ad-hoc & goal-driven categories.

• Effects of specific exemplars on categorization.



2. Resemblance (similarity-based categorization).

• only works well for well-defined/artificial categories (e.g., triangle; an "ace" in tennis).

• Exemplars - specific examples of a category.

• Prototypes - an average across many different instances ('exemplars') of a category.

3. Explanation (knowledge-based categorization).• used to explain the deep structure of concepts

or categories, based on essential features and implicit theories about how things are related.

Concepts as implicit theories, created by "naïve scientists" (us!) trying to understand the world.

Explanation View

• Concepts and categories as based on a complex web (neural network) of knowledge.

• For this view, exemplars are to concepts as data are to scientific theories.

• Knowledge of purpose & history as critical.

• Psychological essentialism: The (naïve?) assumption that things have an underlying nature that make them what they are.

Evidence for Explanation• The ease with which we deal with

bachelors, drunkenness, etc..

• Transformed skunks and toasters.

• Counterfeits.

• Mutilated lemons.

Categorization

Defining features.

Exemplars.

Prototypes.

Explanation.

All four theories are probably correct...

• Artificial categories; categorization "by rule".

• Heuristic strategy for novel & distinct things.

• Heuristic strategy for well-learned things.

• Analytic strategy for complex things/situations.

Knowledge - UNCW Faculty and Staff Web Pagespeople.uncw.edu/tothj/PSY410/PSY410-08-Knowledge-color.pdf · knowledge examines (a) knowledge ... things to say to scared child, ... categories

Documents