Knowledge and Neuro Networks 3

Question 1

Conceptual Knowledge

Answer 1

Bodies of knowledge that are stored in long term memory AND used by cognitive processes

Question 2

Knowledge Organisation

Answer 2

Hierarchy VS Network

Question 3

Category Membership Models

Answer 3

1. Feature Based Model
   - concepts are stored as list of necessary features
   - Defining features: attributes necessary to meaning of object
   - Characteristic features: Attributes descriptive but are not essential

Typicality effect: People reach decision faster when an item is a typical member of a category

2. Prototype Theory
   - abstract and idealized example of a concept

Question 4

Relational Models

Answer 4

Collins & Quilian’s Model:
- Hierarchical structure
Subordinate, Basic, Superordinate
- Spreading Activation: Time to retrieve infomration based on number of links
- Cognitive Economy: Properties of a concept stored at highest possible level
- Inheritance: Lower level items share properties of higher level items

Support for Model: Sentence verification task

Issues: anomalies, such as “lion is a mammal” takes longer then “lion is an animal”

Collins & Loftus Semantic Network:
- Got rid of hierarchy
- Relations between concepts, “is a”, “has a”
- Type of Connectionist Network

Question 5

ACT-R

Answer 5

Adaptive Control of Thought - Rational
- propositional semantic network
- using agents, objects, and relational links

Question 6

Artificial Neural Networks

Answer 6

• Good at classification, recognition, control problems
  -> can filter noisy data
  -> learn via backpropagation
  (modifying weights based on error signal)
  (can correct errors using graceful degradation (aka: correction when certain systems aren’t working))
  (can generalize, apply a learned set of connections to new situations)

Question 7

Top Down Effects in Recognition

Answer 7

1. Word Superiority Effect
   - Letters are recognised more accurately when presented within a meaningful word, compared to a non existing word
   - Easier to detect letter presented in word vs presented alone
2. Pseudo-word superiority effect
   - Easier to detect letter presented in pronounceable word vs not pronounceable.
3. Semantic dependency
4. Degraded Letter correction