Lecture 19 - Models of Memory

Question 1

Adaptive Control of Thought (ACT)

Answer 1

took some of the shortcomings of the hierarchical model and tried to build on top of them

to address the short-comings and salvage that type of approach

Question 2

Parallel distributed processing (PDP)

Answer 2

new direction, new architecture

scrapped hierarchical model

Question 3

What are the criticisms of the hierarchical model of long-term memory?

Answer 3

People verify typical instances of a category faster than atypical examples (typicality effect).

Some activations seem to go around the hierarchical order

People consider similarities when evaluation sentences.

Question 4

advantage of hierarchical system

Answer 4

It forces you to make natural inferences based on limited information.

these inferences would explain why sometimes a partial cue would be useful in bringing up an entire concept

little activation tags going from node to node

Question 5

why do we want models in the first place?

Answer 5

we still need models of

memory in order to make testable predictions about how the system operates .

Question 6

The distinctions suggested by the types of long-term

memory attempt to establish

Answer 6

categories of knowledge that can be learned over time.

Question 7

Adaptive control of thought (ACT) theory

Answer 7

− Information is discrete (concepts held within individual nodes) and modular

concepts held within individual nodes in a very tight modular sense activate a concept and have discrete links to other concepts

• fix hierarchical model

Question 8

Parallel distributed processing (PDP) models

Answer 8

− Information is distributed and has more plasticity (links can change in their nature)

• hiercharcy is based on a computer model and we need something biologically based
• there are no individual nodes that hold an entire concept: concepts distributed over many different units

Question 9

Each model has its own form of

Answer 9

mental representations and

processing of those representations.

Question 10

The adaptive control of thought (ACT)
theory is a general model of human
cognition (Andersen 1976, 1983, 1991).

Answer 10

• Based on a hierarchical model of memory.

• Built as a computational /symbolic model (including software), but more recently
includes data from cognitive neuroscience.

• The representation of declarative memory includes episodic and semantic elements.

• The basic representational unit is the proposition [originally chunks], and
processing occurs through spreading activation (makes more active) of different propositions

Question 11

Propositions

Answer 11

are the smallest unit of
meaning that can be true or false.

They encode declarative facts.

some chunk of information that can be falsified

e.g. The canary is yellow.

(propositions that are false are still propositions)

Question 12

Propositions must have two elements:

Answer 12

1. Nodes: A concept or idea (e.g. bird, person,
   canary, rhinoceros).
2. Links: The semantic association between
   nodes (agent, relation, or object).

Question 13

A single sentence or thought can contain many propositions.

Answer 13

e.g. Fiona is pretty and on parole. (True, Yes a proposition, 2 propositions that each can have their own truth value)

Question 14

A single sentence or thought can contain many propositions.

The propositions may be embedded (or nested).

Answer 14

Ned believes [that winter is coming].

object link: indicates that another proposition is being acted on

Catelyn doubts [that Ned believes [that winter is coming]].

Question 15

Both semantic and episodic memories are part of the ACT model.

This requires a type-token distinction.

Answer 15

• A type refers to a general concept (i.e.. node, canary) or class of objects. [Semantic memory: what is a canary?]
=> is all of them

• A token refers to a specific instance of a type (a specific canary). [Episodic memory]
=> is one of them

Question 16

“The chair is blue”

Answer 16

chair is a node: prior semantic/type knowledge of chairs

a token, a specific chair is blue

replace chair with variable: Token X that has a link

Question 17

Processing in the ACT model is done through spreading

activation.

Answer 17

• Each node has a level of activation. [zero]
• Nodes with the most activation are made available in working memory - the thing you are explicitly, consciously aware of.

• When processing a sentence, the appropriate nodes are activated.
Activation then spreads along the links, with stronger links passing
more activation.

Question 18

To test the ACT model, you can

Answer 18

look at the speed at which
propositions are verified.

if there’s really that spread of activation it then should start to activate other propositions, and you would be faster to recognize that they (are) were present

Question 19

Recall from Meyer & Schvanevedlt (1971),

Answer 19

how nurse => doctor. Priming words spread activation to related nodes and decreased reaction times.

• Have you seen this sentence? (yes/no)

faster RTs:
− The chair is blue.
− The sky is blue.

Slower RTs:
− The dog is hungry.

faster reaction time if those statements are linked because of spreading activation!

nodes are being manipulated: the node activation and what becomes available to working memory
=> it’s the association through the links that’s doing the manipulation: how you raise things off of zero into working memory

Question 20

what’s being manipulated?

Answer 20

nodes are being manipulated: the node activation and what becomes available to working memory
=> it’s the association through the links that’s doing the manipulation: how you raise things off of zero into working memory

Question 21

the ACT model assumes that all nodes start at zero but with activation it believes you can

Answer 21

strengthen links through constant association