認知心理学10. Categorization: カテゴリー化

認知心理学もくじ

• 認知心理学もくじ

Cognition So Far…..

• Structures: sensory memory (250 ms) STM (7+1,2/ 10-30s) LTM (unlimited)
• Memories: explicit memory (semabtic, context independent), episodic memory (event/ context dependent)
• Encoding: levels of processing, structural coding, phonemic coding (shallow processing) , semantic coding (deep processing)
  • imagery coding: little better than semantic coding.

Today

• How does categorization occur and how does it benefit us.

Categorizing

• Benefits
  1. reduces complexity of the environment
  2. It is one way in which we identify objects
  3. Reduces the need for constant learning
     1. e.g, categolize classroom so we don't have to learn how to use the classroom
  4. Allow us to perform an appropriate action
  5. Enables us to order and relate classes of objects/events.

Categories

• “pointers to knowledge”
  • Once you know something is in a category, you know a lot of general things about it.
  • You can focus your energy on specifying what’s special about the object.
  • Help us make inferences about objects.

What are categories?

• Definitional perspective
• e.g.: Chair -“a piece of furniture consisting of a seat, legs, back, and often arms, designed to accommodate one person”

How are Classifications Made?

• concept identification:
  • deciding if an object is an example of a concept based on logical rules.
  • A task that requires deciding whether an item is an example of a concept, where concepts are typically defined by logical rules (CP 187)

• logical rule:
  • based on logical relations
  • A rule based on logical relations, such as conjunctive, disjunctive, conditional, and biconditional rules. (CP 187)
  1. conjunctive rule (small and yellow)
     1. A rule that uses the logical relation "and" to relate stimuls attributes, such as small and square. (CP 187)
  2. disjunctive rule (small or yellow)
     1. A rule that uses the logical relation "or" to relate stimulus attributes, such as small or square. (CP 187)

• rule learning:
  • Ss told the relevant attributes and must discover the rule (small, square)
  • A concept identification task in which people are told the relevant attributes (such as small, square) but have to discover the logical rule.(CP 188)

Concept Identification Paradigm

small and yellowといったら、左上のだけになる

samll or yellowといったら、左の2つ、右下も当てはまる

• attribute learning
  • A concept identification task in which people are told the logical rule (such as conjunctive) but have to discover the relevant attributes (CP 188)

• generating hypothesis and trial and error.

Problems with Concept Identification

• Artificial
• Real categories don't need to share attributes
• Real categories have continuous dimensions
• Real categories are hierarchically organized.

hierarchical organization: ''Rosch (1976)''

• an organizing strategy in which larger categories are partitioned into smaller categories (CP 188)
  1. superordinate category
     1. The largest and most general categories
     2. A large category at the top of a hierarchy, such as furniture, tools ,and vehicles (CP 190)
        1. Musical instruments
  2. basic-level category
     1. An intermediate category
     2. An intermediate category in the middle of a hierarchy, such as table, saw, and truck (CP 190)
        1. Horns, Drums, Guitars
  3. subordinate category
     1. Specific categories
     2. A small category at the bottom of a hierarchy, such as lamp table, jigsaw, and pickup truck (CP 190)
     3. Trumpet, Flute, Tuba

• Hierarchical Organization

Category	Super-ordinate	Basic Level	Subordinate
Musical Instruments	1	6	8.5
Fruit	7	12.3	14.7
Tools	3	8.3	9.7
Clothing	3	10	12
Furniture	3	9	10.3
Vehicles	4	8.7	11.2

カテゴリーが上に行けば行くほどshared attributeも減る。

• Levels of Categorization
• Ss give terms representing each level of categorization. (furniture)
  • superordinate category: purpose or function.
  • basic-level category: legs, seat, back…etc
  • subordinate category: large, soft, cushion.

basic-level categoryをclassifyするのが一番早い。

• Why is categorizing at the basic-level fastest?
• prototype:
  • represent an average pattern for an object
  • An item that typifies the members in a category and is used to represent the category (CP 193)

• But is this true for experts?
• Dog experts vs. Bird experts

犬の専門家は、鳥のbasicを分類するのはとても早かったが、鳥の専門家はsub ordinateと同じくらい時間がかかった。

Problems with Concept Identification

• Artificial
• Real categories don’t need to share attributes.
• Real categories have continuous dimension.
  • An attribute that can take on any value along a dimension (CP 188)
• Real categories are hierarchically organized.
• ''Its fundamental.'' ←new

The fundamental-ness of categories: ''Kenneth and Cuff (1989)''

• Ss 36, 9-15 month old infants
• Habituated infants to 3 kinds of animals
  • Cat, Cow, Horse
• Later present 6 new animals.

• Results:
  • Infants habituated significantly faster to previously seen category exemplars at the second presentation.

habituatedした物と同じカテゴリーに入る物は、早くhabituatedした。

• Evidence for categorization at the super-ordinate level.

The fundamental-ness of categories: ''Hull (1922)''

• Ps learned sets of items, each with a common element.
• They later categorized new characters correctly, but couldn’t describe why they made the decision.
  • きちんと分けられたが理由を理解できてない
• Making the common element distinctive didn’t improve categorization of novel characters.
• Indicates common element extracted ''subconsciously''.

Problems with Concept Identification

• Artificial
• Real categories don’t need to share attributes.
• Real categories have continuous dimensions.
• Real categories are hierarchically organized.
• Its fundamental.
• ''Member typicality varies''.←new

typicality

• A measure of how well a category member represents that category (CP 189)
• Refers to differences in how well members of a category represent that category.
• Why are some members more typical of category than other members?
• Good members will share many attributes with other members of the same category
• And they will share few attributes with members from other categories.

Typicality: Rosch & Mervis (1979)

• Ask Ss to list the attributes of members of a category.
  • Bicycle: wheels, pedals , handlebars…
• family resemblance
  • how frequently are the attributes of a category shared by other members of a category.
  • A measure of how frequently the attributes of a category member are shared by other members of the category (CP 197)

goal derived category

• Satisfies a particular goal
  • A category whose members are selected to satisfy a specific goal (CP 198)
• Family resemblance scores do not predict the typicality of goal-derived categories.
• Share an underlying principal
  • Organized around an ideal.
• e.g. I want to have fun (GOAL) many ways to achieve the goal.
• fit well in our life

• Organized around an ideal.
• ideal
  • An attribute value that relates to the goal of a goal-derived category (CP 199)
  • Those members that best satisfy a goal.
    • Give a birthday present; makes someone happy.
    • Choose weekend activities that we enjoy.
• Members share an underlying principal rather than attributes.

So far, we learned..

• Concept Identification Paradigm
  • big problem: artificial. use discrete..
• conjunctive rule and disjunctive rule
• attribute learning
  • harder than rule learning
• Hierarchical Organization
• typicality
• Family Resemblance

Natural Categories

• Use hierarchical organization
  • Some categories contain smaller categories.
• Relies on family resemblance and typicality to determine representativeness of a category

Classifying People

• stereotype:
  • an attribute believed to representative of social categories.
  • Uses hierarchical organization
  • typicality
  • An attribute value believed to be representative of social categories (CP 199)

• disadvantage:
• Often inaccurate
  • People tend to ignore dis-confirming evidence
  • Focuses on typical examples of the category
  • Overused
  • Self-perpetuation: we tend to see what we want to see
  • Automatic

Stereotyping

• perceptual confirmation:
  • tendency to perceive what you expect to perceive
• self-fulfilling prophecy:
  • when the actor brings about what other expect to perceive
  • stereotype threat

• ''Steele & Aronson (1995)''

• Ss: European and African Americans
• Half the Ss asked to list their race
• Just asking people to list their race before taking the SAT can influence performance
• 試験の前に人種を聞かれるだけで点数が変わってくる。

stereotype

• Advantages
  • Clinical diagnosis
    • Psychosis
    • Schizophrenia
    • Affective disorder

• • • Expect diversity among patients
    • Treat individual differences appropriately

Categories and Semantic Organization ''Warrington & Shallice (1984)''

• 4 patients with impairment to recognize living things.
• Identify objects in pictures, spoken definition, and examples.

Results: ''Warrington & Shallice''

		Inanimate objects
	Picture	Spoken
JBR	90%	79%
SBY	75%	52%

		Living objects
	Picture	Spoken
JBR	6%	8%
SBY	0%	0%

• What does this tell us?
  • Visual - functional distinction
    • living things identified by visual features
    • Non-living things identified by function.

• • Due to a selective loss of visual attributes detection.

''Farrah & McClelland (1991)''

• Presented lists of visual and functional features of objects used by Warringtion.

• Task: underline the visual and functional features for living and non-living objects

• Results:
  • Visual features emphasized more for living things.

''Caramazza & Shelton (1998)''

• doesn't really explain about a patient EW:
  • inability to name pictures of animals.
  • can name objects from other living categories.
  • cannot answer either visual or functional questions.
    • Does a whale have a large fin? - he couldn't answer
    • Does a whale fly? - he couldn't answer

Patient EW

• maybe not attribute, but semantic information is associated with identification of each objects.

• selective loss of information
• attribute may play a role.

Models of Categorization

• Classifying Novel Objects
  • nearest neighbor rule:
    • select the category that contains a highly similar item
      • Requires comparing the novel item with the most similar pattern in the category
    • A classification strategy that selects the category containing an item most similar to the classified item (CP 201)
  • What if the most similar pattern isn't a typical member of the category?
• この理論に基づくと、犬と猫を同じカテゴリーに分類するかも。

Models of Categorization

• average distance rule:
  • compare test pattern to all the patterns in a category
    • Looking for the best average similarity.
  • A classification strategy that selects the category containing items having the greatest average similarity to the classified item (CP 202)

• inefficient time consuming task.

• prototype rule:
  • select the category whose prototype is most similar to the novel item
    • requires one comparison with the prototype from each category
    • Must create the prototype by using the average of all other patterns in a category.
  • A classification strategy that selects the category whose prototype is the most similar to the classified item (CP 202)

• feature frequency rule:
  • selects the category with the most feature matches.
  • A classification strategy that selects the category having the most feature matches with the classified item (CP 202)

• focuses on the features. 数多くのマッチがあるものを選ぶ。which category it is belonged to..
  • e.g. nose, eye, lips.. etc

Which Model is Used?: ''Reed (1972)''

• Ss studied two categories of faces.
• Then classify the novel test faces.
• Have Ss report which strategy the used.

1. prototype rule: 58%
   1. Formed an abstract image for the faces in both categories and choose the closes match
2. nearest neighbor rule: 10%
   1. Compared test face with all the faces looking for a single best match.
3. feature frequency rule:28%
   1. I compared how many times a category feature matched with a test face.
4. average distance rule:4%
   1. Compared all the faces with the test face and chose the category whose faces were most similar.

• ''Hayes-Roth, Hayes-Roth (1977)''
• Ss classify people based on:
  • Age
  • Education level
  • Marital Status
• Found that people relied on the feature frequency rule

• ''Smith & Minda (1998)''
• Early learners: prototype rule
• Knowledgeable learners: nearest neighbor rule or average distance rule
• knowledgeがあればあるほど、prototypeに頼らなくて良くなる。

Theory-Based Categorization: ''Reed & Friedman, 1973''

• Two hypothetical suburban communities
  • Age, income, + children, education
• Age was best predicting feature

• About half the Ss used age
• Others used theories of who would live where
• The importance of features depends on the role the features play in Ss theories.

• semantic dementia
  • Progressive deterioration of knowledge about words and objects (CP 194)

• exemplar model
  • Proposes that patterns are categorized by comparing their similarity to category examples (CP 204)

認知心理学もくじ

• 認知心理学もくじ