Video Module 19: Knowledge Networks

Question 1

“theory” theory of categorisation

Answer 1

proposes that categories provide explanations for how things work in the world, in the same way that theories provide explanations for scientific phenomena
- suggests that we know more about categories than a list of their features or their values in dimensional space (dimensions = height, size, colour, etc.)
- suggests that categories centre on causal relations between entities in the world
- theories guide perception by leading us to believe that particular features of an object are interesting or relevant while others may not be

Question 2

What evidence supports the “theory” theory of categorisation?

Answer 2

A study by Springer and Keil (1989) presented children with a description of an object belonging to a natural kind (e.g. “This is an animal that looks, acts, and sounds like a horse”). They then gave the children new defining features of the object for a different natural kind (e.g. “The animal has the inside parts of a cow, cow parents, and cow babies.”)
- researchers found that children <6 yrs thought these animals to be horses, but 7< thought of them as cows; adults thought of them as cows
- revealed that category membership of these objects is determined by our concepts/theories of biology (e.g. internal structures, parentage)

Question 3

What are some theories we have of natural kinds? Of artifacts?

Answer 3

1. natural kinds have essential properties: no matter how much we change something, it still is going to be the original thing it was
2. artifacts do not have essential properties as natural kinds do: you can change artifacts into new objects
   Researchers found that natural kinds and artifacts are represented in different regions of the brain.

Question 4

Does theory theory contradict probabilistic theories of categorisation?

Answer 4

No; theory theory does not directly contradict probabilistic theories. It acknowledges that we do use object similarity to make judgements about category membership. However, TT considers our understanding of a category beyond its dimensional qualities.

Question 5

What are three models for how we organise concepts?

Answer 5

1. Collins & Quillian’s hierarchical model
2. Propositional networks
3. Connectionist networks

Question 6

Collins & Quillian hierarchical network

Answer 6

suggests that we organise concepts in a hierarchical structure, in which concepts are stored at the highest possible level.
- employs the concept of cognitive economy, in which our knowledge of concepts is efficient in its storage
- Proposes the concept of inheritance, in which categories at lower levels (subordinate categories) inherit the properties of the higher level categories (superordinate categories) to which they belong
- Suggests that processing time—the time it takes to retrieve information—depends on the number of links between concepts/features within the structure

Question 7

What evidence supports C&Q’s hierarchical network theory for concept organisation?

Answer 7

C&Q conducted a study in which participants had to do a sentence verification task. C&Q found that participants’ reaction times increased when associate paths became longer
- e.g. participants were faster to answer “A canary can sing” than “A canary has skin” because the associative path (# of levels) between “canary” and “skin” is longer

Question 8

propositional networks

Answer 8

propose that we store concepts through propositions, which are the smallest unit of a true or false statement.
- Nodes in propositional networks are concepts in our mind
- Nodes can be linked to multiple propositions: e.g. “Dog” can be linked to “Dogs like food” and “Dogs are animals”
- Links between nodes represent relations and associations between concepts
- Relies on local representations: each node = 1 concept

Question 9

connectionist networks

Answer 9

proposes that our ideas are represented by a pattern of activation across our network of knowledge
- relies on the idea of distributed processing
- relies on the idea of parallel processing, in which the processing of different info occurs simultaneously
- connection weights/strengths of connections depends on how often two nodes fire together; error signals (misfires) cause a node to decrease its connections
- AKA parallel distributed processing (PDP) networks

Question 10

parallel distributed processing (PDP) networks

Answer 10

• AKA connectionist networks
• learning takes place by changes in the connection weights or strength of connections between concepts in the network
• connections strengthen/weaken when two nodes fire together
• rely on back propagation: error signals (misfires) cause nodes to decrease their connections to the input nodes that led to the error