PSYCH 240 LEC 12 (CATEGORIZATION)

Semantic vs. episodic

Semantics

• General Knowledge

• Memory of Facts

  • not tied to specific time or place

  • Pittsburgh in PA, 2+2=4, birds have wings

Episodic

• Memory of Personal episodes

  • Tied to a specific time and place

  • Personal perspective

  • First kiss, walk to class this morning, game vs. Wisconsin

Explicit vs. Implicit

Explicit

• Conscious recollection

• Declarative Knowledge

  • recall

  • recognition

Implicit

• Unconscious change

• Procedural Knowledge i.e. skills

• Priming

Categorization: Use by pigeons

• Pigeons peck 1 out of 4 keys depending on the picture stimulus in the middle

• Goal is for it to peck the correct hole corresponding to picture

• Results:

  • ABLE TO ABSTRACT CATEGORY FROM EXAMPLES

  • 2/3 accuracy with new exemplars/pictures they generate with old pics they’ve seen before

Categorization: Experiment on physical similarity vs. conceptual knowledge (milk vs. mashed-up food experiment)

• Hawk sim to flamingo bc both birds but visually similar to bats

• Exp:

  • Chn say shadow of hawk will feed young mashed-up food like flamingo, not milk like the bat

• Results:

  • Debunks how category not just similarity-based

  • Suggest chn can match using the essentials (bird-like characteristics)

Modern probabilistic view: characteristic properties

• Defining properties of given categories: Necessary and Sufficient

  • properties/features are characteristic, but not defining (not guaranteed to specify if something is in category, more common characteristics)

Modern probabilistic view: similarity

• objects with common characteristics (something similar in-between them then belongs into one category

• some members have more characteristic properties than others

Classical view: defining properties

• Category borders boundaries are fuzzy or probabillistic

  • i.e. Monk could be on the boundary of the bachelor category

Classical view: Problem: what defines “game”

• Some terms do not have necessary + sufficient conditions

• Basketball game vs Chess Game

• i.e. is monk a bachelor

• i.e. is a game-show = reality TV

Typicality evidence: Ratings

• evidence for fuzzy categories

• rate characteristic properties as being most typical to atypical of a specific category (not defining)

• takeaway

  • some fruits are more typical than others

  • reflects similarity of concept to category via contrast model

Typicality evidence: Sentence verification

• Judge sentence and verify if they are T or F

• Result:

  • Faster to verify more typical exemplars vs less typical exemplers

    • i.e. Robin more typical vs chicken less typical

Typicality evidence: Hedges (linguistic)

• cannot categorize with definitions (bc both live on land), judge their similarities

• only necessary when classical categorization ( i.e. biologists say mammals is an animal)

  • more black and white

Categorization based on similarity: Exemplar Theories

• Def: Example of a category

• Multiple exemplars are stored in memory

• Categorize new things based on similarity to stored exemplars

  • Diff. types of dog breeds in head and match it with the one you see dog in visual

Categorization based on similarity: Prototype Theories

• Def: Best, ideal, or average example

• Only a prototype is stored in memory

• Categorization based on sim. to prototype

Geometric approach: Similarity-rating task

• Subject-ratings from 1-6 how sim. is apple to a plum?

• Results:

  • ratings used to place concepts in geometric space (w/ help of “multi-dimensional scaling”)

• Scaling:

  • items that are judged super-similar then they will be close (vise versa for not-similar then far apart)

• not accurate reflection of rep. categorize in semantic memory

• only accurate if fulfills all three metric axioms

3 Metric Axioms

• If concepts represented in geometric space is true, then should satisfy certain prop (axioms) of (geo)metric space

3 Metric Axioms: Minimality

• Dissimilarity between concept and itself must ALWAYS be the smallest possible

• Highly familiar concept rated more similar to itself than one that is less familiar

  • i.e. apple-apple distance is 0/ more similar than pomegranate-pomegranate (more unfamiliar)

3 Metric Axioms: Symmetry

• Similarity between 2 concepts must be the same regardless of the order (Going from A→B = Going from B→A)

  • i.e.

    • how similar is an apple to a plum?

    • how similar is a plum to an apple?

3 Metric Axioms: Triangle inequality

• 1st concept sim. to 2cd concept

• 2cd concept sim. to 3rd concept

• THEN 1st + 3rd must be reasonably similar

  • i.e. plum, date, grape

• Hypotenuse of triangle cannot be longer than the sum of 2 sides

Examples of violations of metric axioms

• Minimality: Pomegranate is diff. to itself is not reflected

• Symmetry:

  • Human similarity judgement with similarity going from apple → plum DOESNT equal distance from plum → apple

  • Since unfamiliar category is judged more similar to an unfamiliar category

    • Pomegranate judged more sim. to apple than apple is to pomegranate

• Triangle Inequality:

  • 1st and 3rd concept are not similar

Feature-based measure/approach

• ppl give similarity ratings inconsistent with geometric space/axioms

• proposed feature-based similarity approaches don’t require metric axioms

• feature-based approach look at features in common + diff. features

Tversky’s feature comparison (contrast) model

• L = Instance you need to categorize

• O = Example from category

• Judges similarity of L (lemon) and O (orange)

  • Add up their common features

  • Minus features unique to L has that O doesn’t

  • Minus features unique to O has that L doesn’t

    
    

Tversky’s weighted function of features common to L and O

Similarity of (L,O) = a x f(shared) - b x f(L but not O) - c x f(O but not L)

• a, b, c are the weights

  • vary depending on individual

• can use typicality rating to use in this equation

How violations are accounted for

• Minimality

  • familiar object have more feature vs unfamiliar ( when u compare concept to itself apple-apple)

  • increased a(shared)

• Symmetry

  • b + c weights are different

  • order of comparison makes a difference

• Triangle Inequality

  • 2 concept sim. to 3rd for diff reasons, but have little in common themselves