Human Cognitive -Read Entirely FIRST!!

Objects and Concepts:

Identifying and Classifying Information

Identification and Classification: An Overview

 First steps in post-attentional processing

– Identifying and classifying input: object recognition

– Matching an incoming stimulus with stored

representations for the purpose of identification

– Concepts: The classification database

● Categories/Concepts exist at different levels

– Levels of Categorization

◘ Superordinate Level (e.g., animal)

 Most general

◘ Subordinate Level (e.g., blue-fairy wren)

 Most specific

◘ Basic Level (e.g., bird)

 Midpoint classification level

 Dominates everyday description and thought

 The entry point (i.e., default level for)

categorization

❖ Recognizing From the Bottom Up and the Top Down

● Bottom-Up Processing in Object Recognition

– Employs the information in the stimulus itself (i.e., the

“data”) to aid in its identification

 Top-Down Processing in Object Recognition

– Using expectations, knowledge, and/or surrounding

context to aid in recognition

❖ Object Recognition

 A complex and astoundingly efficient process, given

what’s involved

» Effects of Orientation and Perspective

 Key question for theories of object recognition:

– Is recognition viewpoint-invariant?

◘ Does recognition depend on a particular

view?

 Palmer, Rosch, and Chase (1981) found that recognition

depends on perspective

 Recognition is best when objects are presented in a

canonical perspective

rather than a non-canonical one

Biederman and Gerhardstein (1993)

➢ Investigated the effect of orientation on object recognition

 Used a priming task

♦ Priming = effect of previous exposure to a stimulus

➢ Procedure

✓ Phase 1: Presented objects

✓ Phase 2: Presented identical or different objects in the

same or different orientation

➢ RT: Priming in Phase 2

➢ Results

✓ Phase 1 provided a baseline identification reaction time

✓ Seeing a flashlight in Phase 1 primed identification of all

flashlights (semantic priming)

✓ Seeing the same object produced more priming than

seeing a different example

✓ Effects of rotation were negligible

» Effects of Context

 Most object recognition occurs in context

Palmer (1975)

➢ Briefly presented simple sketches of various everyday

scenes followed by a sketch of single object

➢ Varied the relationship between object and scene

✓ Consistent (e.g., loaf of bread in a kitchen)

✓ Inconsistent (e.g., a drum in a condition)

➢ Identification of objects was best when the objects followed a

scene that was consistent

Davenport and Potter (2004)

➢ Investigated the interactive effect of scenes and objects in

recognition

➢ Used photographs of familiar scenes

✓ Presented each scene followed by a mask

♦ Briefly presented visual pattern that serves to erase

the previously presented stimulus

▪ Ensures identification is not based on a fleeting

afterimage

➢ Task

✓ Identify either the object in the foreground or the

background scene

➢ Varied the relationship between scene and object:

✓ Consistent (e.g., quarterback on a football field)

✓ Inconsistent (e.g., quarterback in a church)

➢ Results

✓ Objects were more easily recognized than backgrounds

✓ Consistent scenes and objects were better recognized

than inconsistent ones

❖ Theories of Visual Object Recognition

 Parts-Based Approaches

– Incoming patterns are parsed into component parts

that are matched to information in memory

 Image-Based Approaches

– Whole image is compared it to representations in

memory until a match is found

» Parts-Based Approach

 Features of the encoded object are compared to a

stored representation of the object’s structure

– Stored representation included a list of features, as

well as the relationship between them

● Sometimes called feature analysis

● Viewpoint-invariant

– Identification does not depend on the particular view

we have of the object

 Recognition by Components (RBC)

– Object recognition is based on parsing an object into its

component parts

◘ “visual primitives”, termed geons

 RBC Stages of object recognition

– Edge extraction

– Search for non-accidental features

◘ Features that don’t appear to be an accident of a particular perspective

◘ Processes characterized by Gestalt principle of simplicity

– Parsing at boundary areas

– Geons determined; match with memory representations

– Object identified

 RBC proposes that perspective shouldn’t affect recognition

…but sometimes it does

Tarr and Pinker (1989)

➢ Used a priming procedure similar to Biederman and

Gerhardstein (1993)

➢ Procedure

✓ Training phase

◘ Presented three shapes with names

◘ Presented in the same orientation until memorized

✓ Test phase

◘ Recognition test

♦ Shapes from training phase presented at various

degrees of rotation

♦ Identify by pressing one of three keys

➢ Dependent variable: Recognition RT

➢ Results

✓ Recognition RT increased as a function of rotation

♦ “viewpoint dependence”

» Image-Based (IB) Approach

 Objects are recognized by comparing input with a stored

replica (i.e., “image)

– Template matching

◘ Exact match must be found

◘ Fails to account for the flexibility of object

recognition

 Multiple views approach

– Based on experience, we have multiple stored views of

objects

– Supported by physiological studies of recognition

◘ Logothetis, Pauls, and Poggio (1995) found evidence

of heavily viewpoint-dependent cortical cells

 “physiological templates”

 Comparing IB and PB views

– PB views

◘ Object representations are constructed by assembling

visual primitives via a fixed processing mechanism

◘ Visual primitives do not depend on experience

◘ Through experience we learn the configurations of the

primitives that differentiate objects

– IB views

◘ Experience determines the stored representations

◘ Recognition is visual memory

❖ Object Recognition: Parts or Images?

 Viewpoint-independent mechanisms (PB approaches)

– Used for (basic-level) categorical distinctions

◘ Bird vs. hammer

 Viewpoint-dependent mechanisms (IB approaches)

– Used for (basic-level) categorical distinctions

◘ Cardinal vs. blue jay

❖ Nonvisual recognition

» Tactile Recognition

– Gaining information through touch = haptics

– Exploratory procedures

◘ Motor patterns performed by hands in exploring

and identifying an object

◘ “haptic glance”

 Exploratory procedures include:

» Olfactory Recognition

 Olfactory-verbal gap

– Difficulty describing and identifying smells

◘ Women are slightly better than men

− Odor recognition is better than order naming

 Tip of the nose phenomenon

− Difficulty in labeling an odor despite strong feeling that

one knows what the odor is

– Specific and sensory example of tip of the tongue

phenomenon

◘ Feeling we know a piece of information, but are

unable to retrieve it

 TOT states allow for assessment of metacognition

– Typically an accurate reflection of knowledge

Johnson and Olsson (2003)

➢ Procedure

✓ Presented a variety of smells for identification

✓ If unable to name it, gave a feeling of knowing judgment

♦ Guessing probability of ability to correctly recognize the label presented after the odor

➢ Results

✓ 28% of the time, attempts were made to name the odor

✓ 16% correct

✓ Feeling of knowing was correlated with recognition performance

Face Recognition

 Prosopagnosia

– Inability to recognize familiar faces

» Face inversion

 Inversion disrupts the recognition of faces much more

than objects

» Face inversion

 Inversion disrupts the recognition of faces much more than objects

» Thatcher Illusion

 Disproportionate effect of inversion on faces

 Why is recognizing faces different from other objects

− Recognizing objects

◘ Requires first-order relational information

◘ Information about object parts and how they relate

− Recognizing faces

◘ Requires second-order relational information

◘ Comparison of first-order relational information to a

“typical” face

− Inverting a face disrupts 2nd order relational processing

❖ Holistic Processing

 Faces seem to be encoded as whole configurations

Tanaka and Farah (1993)

➢ Presented faces or houses accompanied by labels

✓ “Larry’s house” or “Larry’s face”

➢ Recognition test

✓ Conditions

♦ Isolated-parts conditions

▪ “Larry’s nose” or “Larry’s door”

♦ Whole-object condition

➢ Results

✓ For houses

♦ Part and whole identification was the same

✓ For faces

♦ Whole identification was better than part identification

✓ Supports the notion that faces are processed and

remembered holistically

❖ Is Face Recognition “Special?”

 Face recognition is akin to image-based approaches to

recognition, but is it a special mechanism?

 Special mechanism view

− Supported by dissociations between object and face

recognition

◘ Patient’s with intact object recognition, but

impaired face recognition

◘ Inversion hurts face recognition, but not object

recognition

 Expertise View

– Extensive experience with faces makes us “face experts”

– Expert recognition has a finer entry point for recognition

◘ NOT the basic level (as for objects)

◘ Individual level is the entry point

 Lower than subordinate level

− Expertise is special; faces aren’t

The debate continues….

❖ Individual Differences

» Gender

− Women > men in recognizing facial emotions, but are

they better at face recognition in general?

McBain, Norton, and Chen (2009)

➢ Shown line drawings of an object embedded within a larger

array of lines

➢ Objects

✓ upright face (a)

✓ inverted face (b)

✓ tree (c)

➢ Stimuli were shown from 13-104 milliseconds

➢ Task: indicate if face or tree was on the right or left side of

the display

➢ Dependent variable: accuracy

➢ Results

✓ Performance was better the longer the display

✓ Woman were more accurate than men in both upright

and inverted face conditions

✓ Performance for men and woman was the same in the

tree condition

✓ Demonstrates female superiority in face recognition

 Possible relation to:

 Moderating effects on social interaction in

schizophrenia?

 Higher rates of autism in males

» Culture

Blais, Jack, Scheepers, Fiset, and Caldara (2008)

➢ Independent variables

✓ Ethnicity of participant

♦ East Asian or Western Caucasian

✓ Ethnicity of face to be identified

♦ East Asian or Western Caucasian

➢ Procedure

✓ Phase 1

♦ Shown set of Western and Eastern faces to learn

♦ Dependent variable: tracked eye movements to see

where on the face participant’s fixated eyes

✓ Phase 2

♦ Facial recognition task

▪ Faces from Phase 1 and new faces

♦ Dependent variable: accuracy

➢ Results

✓ Recognition accuracy

♦ No overall difference in face recognition between participant groups or between race of faces

♦ Interaction: Participants were better at recognizing faces within their own racial group

✓ Eye movement patterns

 Western participants: fixated on eyes and surrounding areas

 Eastern participants: fixed on nose region

 Cultural differences in:

 social norms?

 global perceptual tendencies?

❖ Self-Recognition

 Some characterize recognition of one’s own face as

“extra special”

Keenan (1999)

➢ Presented participants with a “face movie”

✓ Face gradually transformed from their own to a celebrity

➢ Participants responded when the face seemed to transition

from “more them” to “less them”

✓ Responded with either left-hand (i.e., right hemisphere)

or right-hand (i.e., left-hemisphere)

➢ Results

✓ When using the left hand (right hemisphere) the “not

them” transition occurred earlier

♦ Indicates right hemisphere advantage in self-

recognition

Epley and Whitchurch (2008)

➢ Investigated a self-enhancement bias

✓ Tendency to see oneself as more attractive than in reality

➢ Session 1

✓ Picture taken of participants

➢ Faces were morphed with an attractive or an unattractive

face to create 10 face versions of the person’s face

➢ Session 2

✓ Presented with 10 versions plus the original

✓ Task: identify the picture taken in the 1st session

➢ Results

✓ Participants erred toward more attractive versions of

their own face

❖ Retrieving Names of Faces: Person Recognition

 Final stage of face recognition: Name retrieval

– A common source of tip-of-the-tongue phenomenon

and often associated with a strong feeling knowing

 Gradient of difficulty in person recognition

– RT for face recognition < RT for biographical info

– RT for biographical info < RT for retrieving name

– Found for recognition accuracy (Hanley and Cowell,

1988) an recognition RT (Young, Ellis, and Flude,

1988)

» Serial and Parallel Accounts

 Bruce and Young (1986) propose a serial account

− First, face must activate a stored representation in

memory called an FRU (face recognition unit)

◘ If activated, the person is recognized as familiar

– Next, FRU must activate the person identity node

(PIN), which stores biographical info about the person

◘ If activated, the biographical info becomes

available

– Next, PIN must activate the terminal node (person’s

name)

 Interactive activation and competition model—parallel

account

– Separate representation units for:

◘ Faces (FRU)

◘ General representations of people (PIN)

◘ Semantic representations of biographical info and

name (SIU)

– Biographical information receives activation from multiple

sources (people share biographical information)

– Name only receives activation from one, so it is

chronically “under-activated” and harder to retrieve

Brédart, Brennen, Delchambre, McNeill, and Burton (2005)

➢ Names of good friends and family should have chronically

high levels of activation, even higher than biographical info

➢ Participants: colleagues in psych department who had

known each other for years

➢ Procedure

✓ Presented with a cue

♦ Name

♦ Occupation

♦ Nationality

✓ Followed by a picture of themselves or a coworker

➢ Task

✓ Identify if the cue correctly belonged to the face

➢ Dependent variable: RT for verifying face-cue connection

➢ Results

✓ RT for names was faster than for other information

Networks and Concepts: The Classification Database

❖ Semantic Networks

 Knowledge stored in the form of associative networks

 Concepts are represented by nodes that are connected

to related concepts and features

◘ Related concepts and features are “close” to one

another in the metaphorical sense, non-neural

● Posits excitatory connections between concepts and/or

features to explain knowledge activation and retrieval

 Tasks for assessing access to categorical knowledge

– Category verification task

◘ Is a penguin a bird?

– Feature verification task

◘ Is a canary yellow?

– RT is the DV of interest

 Spreading Activation Model

– Knowledge represented as concept nodes linked in an

associative network; relationships include

◘ category membership (canary to bird)

◘ property to concept (yellow to canary)

◘ more subtle relationships (i.e., canary and cat)

– Processing assumptions

◘ When a concept is presented, its corresponding node is

activated

◘ Activation spreads to connected nodes

◘ Strength of activation decreases as a function of time,

distance, and the number of nodes activated

− Explains semantic priming

◘ Prime activates its node and activation spreads to related

concepts

 This “head start” activation means less activation will

be needed to recognize the word when it is presented

❖ Concepts and Categories

» Functions of Concepts

 Allow for understanding

 Allow us to make predictions

 Support new learning

 Important for communication

» Categories as a Concept

 Natural Kinds (e.g., fruit)

– Categories that occur naturally in the world

 Artifacts (e.g., weapons)

– Objects or conventions designed by humans to serve

particular functions

 Ad hoc categories (e.g., things you’d save in a fire)

– Categories formed “on the fly” in the service of a goal

 Metaphorical concept (e.g., emotional prison)

❖ Similarity-Based Categorization

 Categorization involves judging the similarity between a

target object and long-term memory standard

» The Classical View

 Items are grouped into categories if they have certain

critical features

− Possession of these critical features is necessary

and sufficient for membership in the category

 Problems with the Classical View

● What are the critical features?

● Model can’t account for:

− Graded structure of categories

◘ “apple” is a “better” example of fruit than “kiwi”

− Fuzzy boundaries

◘ Distinctions between categories are not absolute

 Example: Is bowling a sport or a game?

» The Prototype Approach

 Instances of a category are classified in terms of their

resemblance to other members

– Typical members = high family resemblance

 Prototype

− Standard to which other category members are

compared

 Prototype view is more flexible than the classical

approach

– Probabilistic nature of categories accounts for

graded structure and fuzzy boundaries

 Prototypes are abstracted through repeated experience

with category members

Posner, Goldsmith, and Welton (1967)

➢ Presented dot patterns to participants

✓ All patterns statistically generated from a (never

presented) prototype

➢ Recognition test

✓ Old and new dot patterns

✓ Critical item: prototype

➢ Results

✓ Prototype was frequently recognized as “old” and at a

much higher rate than other “new” items

 Problems with the Prototype Approach

 Category representations include more specific

information than prototypes would predict

– We’re sensitive to correlated information among

category members

 Category representation is sensitive to context

– “Best example” depends on the situation

» The Exemplar Approach

 We store all instances of encounters with a concept

(exemplars)

 Accounts for graded structure

– We have more encounters with typical objects, hence

more exemplars

 Accounts for the biasing effect of context

− Context makes certain exemplars more retrievable

» Problems with the Exemplar Approach

 Category representations do seem to be abstracted in

some cases

– Posner et al. dot study—never-presented prototype

recognized

 Problem of economy

– Do we really store every instance of encounter with

each concept?

❖ Essentialist Approaches: Concepts as Theories

 A “top-down” approach to categorization

 Categories have underlying “essences” that bind

category members

 Categorization not based on a similarity comparison

Rips (1989)

➢ Presented participants with a story of a bird-like creature called a sorp

➢ Conditions

✓ Accidental condition

♦ Catastrophic accident occurs and sorp now looks more like an insect, but still acts like a bird

✓ Essence condition

♦ Catastrophic accident occurs and sorp now looks more like an insect AND acts like an insect

▪ Given a new name—doon

✓ Control

♦ Only read story of bird-like sorps

➢ Question of interest

✓ How would the transformations affect categorization

➢ Two categorization tasks:

✓ Rate how well the sorp fit into the category of “bird”

✓ Rate the similarity of sorp and a “bird”

➢ Predictions

✓ According to similarity-based approaches, these two

tasks reflect the same thing

♦ Rating should be the same

➢ Results

✓ Accident condition: Change lowered similarity ratings more than categorization ratings

✓ Essence condition: Change lowered categorization ratings more than similarity ratings

✓ Categorization and similarity ratings were dissociable

♦ Strong evidence that categorization ≠ similarity judgment

Similarity rating Categorization ratingControl

» Biological Essentialism

 Folk biology

– Our everyday knowledge and intuitions about living

things

– Cross-cultural similarities exist in how we think about

natural categories

– Innate “module?”

 Is essence-based categorization universal?

– Choice of the Dalai Lama = evidence of essence?

 Tendency to essentialize is pervasive and powerful

– Is it overly powerful?

Haslam, Rothschild, and Ernst (2000)

➢ Interested in people’s tendency to essentialize human

characteristics

➢ Had participants rate 40 human characteristics on

dimensions thought to reflect essence

✓ Naturalness, stability, and mutability of category

membership

✓ Necessity of features for category membership

➢ Results

✓ People see essential differences between people who differ on these characteristics

✓ People don’t see essential differences between people who differ on these characteristics

 Problems with the Essentialist view

– Idea of an “essence” is vague, unspecified

– What’s the difference between an “essence” and a

concept’s set of associations and related concepts?

◘ Is an essence simply “what we know” about a

concept?