LPI Topic 7 Flashcards
Concept Learning
The process of learning to classify objects, events, or ideas based on shared features.
Categorisation
Grouping similar objects or ideas based on shared characteristics to simplify thinking.
Category
A set of objects that share common properties, e.g., “birds” or “vehicles”.
Concept
A mental representation of a category, used for reasoning and communication.
Defining Features Approach
Concepts are defined by a set of necessary and sufficient features.
Hierarchical Network Model (Quillian, 1969)
Concepts are stored in a hierarchy from general (superordinate) to specific (subordinate).
Superordinate Level
The most general category (e.g., “animal”).
Basic Level
Intermediate level, often the most used in daily life (e.g., “dog”).
Subordinate Level
The most specific category (e.g., “Labrador”).
Economy of Storage
A cognitive benefit of the hierarchical model, reducing redundancy in knowledge storage.
Typicality Effect (Smith et al., 1974)
Some category members are recognised faster than others (e.g., “robin” faster than “penguin” for “bird”).
intransitivity of Categorisation (Hampton, 1982)
People may categorise inconsistently (e.g., “car seat is a chair” but not “car seat is furniture”).
Issues with Atypical Examples
Some members do not fit neatly into categories (e.g., “penguin” as a bird).
Challenges from Semantic Dementia
General knowledge is retained longer than specific details, contradicting the hierarchy model.
Prototype Approach (Rosch, 1973)
Concepts are represented by an ideal or most typical example.
Prototype Example
A robin is a more typical “bird” than a penguin.
Typicality Gradient
More typical examples are categorised faster.
Flexibility of Prototypes
Prototypes can change with experience.
Abstract Concepts
No clear prototype for ideas like “crime” or “justice”.
Prototype Combination Problem
How do we combine prototypes for complex ideas? (e.g., “killer firework”).
Exemplar Approach (Medin & Schaffer, 1978)
Concepts are represented by stored examples rather than a single prototype.
How Categorisation Works
A new object is compared to known exemplars.
Preserves Variability
Unlike the prototype approach, it retains category differences.
Strength Over Prototype Approach
Accounts for real-world variability, not just an “average” example.
Criticisms of the Exemplar Approach
Cognitive Load
- Storing all exemplars requires large memory capacity.
Abstract Concept Limitation
- Difficult to use with non-physical concepts like “justice”.
Connectionist Model
Knowledge is represented by patterns of activation in neural networks.
Parallel Distributed Processing (PDP)
Cognitive processing happens in parallel across multiple connections.
Pattern Completion
Concepts are reconstructed from past experience, not stored as fixed entities.
Training the Network
Adjusts connection weights based on feedback to improve accuracy.
Rumelhart’s Connectionist Model
Input Units
- Represent concepts at the basic level (e.g., “canary”).
Hidden Units
- Mediate learning by forming connections between input and output.
Output Units
- Represent category attributes (e.g., “canary can sing”).
Error-Driven Learning
- Learning occurs by adjusting incorrect predictions.
Comparison to Rescorla-Wagner Model
- Both involve learning through error correction.
Progressive Differentiation
Broader categories (e.g., “animal” vs. “plant”) are learned before finer distinctions.
Coherent Covariation
Features that frequently co-occur help category learning (e.g., “feathers” and “wings” for birds).
