Categories & Concepts Flashcards
Define categories, concepts and categorisation.
Category: Sets of things in the world that we represent as alike in some way, or treat as equivalent for some purpose
e.g. birds, plants, odd numbers, victims
Concept: Mental representation of a category/Mental category.
- Not interchangeable
Categorisation: Act of recognising commonalities amongst these sets of things, or that the commonality applies to a new thing, and builds/adds to the concept
What does categorisation of objects or forming concept change our experience? (5)
- Allows treatment of new things as familiar, via induction (the inference of a general rule from particular instances)
- Allows prediction based on classifications (e.g. Dogs bark. There’s a new dog - probs barks)
- Gives common reference in communication
- Category labels = words assigned to represent categories - Building blocks of more complex knowledge structures/reasoning
- Cognitive economy: Summary representations more efficient to reason with than memory of every word
FEATURE-BASED
What assumption underlies feature-based approaches of category representation?
Categories are represented by unstructured collections of features, describing the properties of individual objects
(e. g. Dog: four legged, furry, mammal, barks, friendly)
- similar to assumptions of Tversky (1997)’s Feature model of similarity
Feature-based approaches: Classical rule-based view
- Categories are represented by set of defining necessary and sufficient features that distinguish members from non-members
e. g. Bachelor: Unmarried man - People learn categories by having hypotheses/candidate rules and testing the strength of different rules in predicting membership, using inductive classification method
e. g. Snarg/Blicket
- Feature-based approaches: How did the Prototype theory criticise the classical view?
- Often no necessary and sufficient conditions
e. g. Bachelor: pope, widower - Wittgenstein’s “game”: No single feature is common to every example
- What does the Prototype Theory posit about the representation of categories?
Rosch: Prototypes are the collection of the average features across examples
- Graded category membership: similarity/typicality of example to prototype vs Y/N
- Classification is not just testing rules, but seeing similarity of new exemplar to prototype
How do the two experiments conducted by Rosch and Mervis (1975) on natural and artificial categories support the Prototype theory of category representation?
Experiment 1: Natural categories
Participants either
- Rated typicality of given examples of categories “How typical is X and Y to categories A?”
- List properties of exemplars of categories and contrast between two categories
FOUND: Higher typicality ratings when exemplars had a) many features in common with other category exemplars, and b) fewer features common with contrast categories
Typicality as function of overall “cue validity”
Experiment 2: Artificial categories
Participants learnt to classify 6-letter strings as members of two categories
- Category exemplars had different letters in common with its own kind and with the other category
High cue validity –> better learning and accuracy
Not all/nothing - no singular defining features of categories. Therefore supports prototype theory.
How does Ponser and Keele’s (1968) experiment support the prototype theory for category representation?
Participants told to categorise dot patterns distorted from prototype. However, during learning, participants never saw the prototype
FOUND: After learning, participants were just as fast and accurate (if not more so) at classifying prototype than the other exemplars.
–> discovered prototype from learning?
- Were able to deduce the rules for the category without seeing the prototype
Supports more graded categorisation?
Feature-based approach of representation: How does Exemplar theory respond to Prototype theory?
Agrees with prototype theory - graded membership, classification based on similarities and not rules
Doesn’t agree with abstraction: Categories are represented as the collections of encoded exemplars
Posner and Keele’s (1968) unseen-prototype adv can be explained as the generalised collective similarity across all exemplars
- How does the Exemplar theory suggest that we use features to represent categories?
People generalise to things that are superficially quite similar to past experiences - which often includes irrelevant info.
E.g. Doctors’ generalisation of diagnoses are aided by past cases, even when similarity is based on attributes irrelevant for the diagnosis
Novel atypical stimuli - classified as members if similar to even a single encoded exemplar (e.g. Ostrich = weird bird –> helps classifying Emu)
“Why would a system be designed like this? Why not just store what is useful?”
But not sure what will be useful later - Storing lots of info may allow for greater variety of info to be used if it becomes important (no storage costs, because LTM)
- Cluster models (Love, Medin & Gureckis, 2004)
We make abstractions AND store exemplars.
As exemplars are encoded, system predicts their category membership
This forms prototype-esque summary representations of highly similar exemplars that all lead to the same accurate classification
If a new exemplar is dissimilar enough from summary of previous exemplars (but is part of same category) system is surprised –> forms new cluster
Then if more exemplars are like the odd one in the new cluster, it becomes its own local summary representation. If not, exemplar remains an exemplar.
- Category boundaries theory
Focuses more on importance of borders between categories
- Less focused on summary representation from the middle of the category
Many categories are represented in opposition to each other - easy to highlight border (e.g. conservative/liberal, fruit/veg)
“Caricatures” are important - exaggerates features away from the category boundary
- predicted by error-based mech
Same error-based learning can lead to new cluster recruitment
What did Daris and Love’s (2010) experiment reveal about feature-based categorisation?
- Participants learnt 4 categories of energy sources (or political supporters), which differed on 2 dimensions: Cost vs By-product pollution.
- Learning categories during trials: - Rated average cost of sources
FOUND:
- Contrasted dimension: “average” value is shifted/idealised away from the category boundary.
- Not contrasted dimension: average value remained average
e.g. Coal vs wind plant: pollution is idealised, while cost remains average
Category learning distorts our understanding/memory
Summary of Feature-based models (If you’re not super lost by this stage)
- Classical view: Categories are represented by necessary and sufficient conditions
- People learn categories by testing hyp of category-defining rules - Prototype theory: “family resemblance structures”? graded membership, for abstract representations of category average. Classification via similarity to prototype
- Exemplar theory: No abstraction. Classification via summed similarity from all exemplars, or similarity to individual exemplars
- Concepts composed of multiple clusters that code for smaller-order generalisations or even exemplars
- Boundaries: Focus on the dividing lines between categories –> Ideals/caricatures, as predicted by the error-driven learning of cluster models
STRUCTURED KNOWLEDGE APPROACHES: RELATIONAL
What do the Structured knowledge approaches posit about the way concepts are mentally organised?
We represent knowledge as relational structures - knowledge elements are bound by how we relate
Concepts not isolated from each other, but are parts of large knowledge structures that make them coherent and systematic.
e.g. Bachelor: not just “unmarried man” but is a particular phase within a male hetero-normative life trajectory
- Pope doesn’t violate the definition, he is just not on the same trajectory
