Semantic Memory

1
Semantic Memory

• Psychology 3717

2
Introduction

• This is our memory for facts about the world
• How do we know that the capital of Viet Nam is
  Hanoi
• How is this type of information stored?
• Is there any difference between natural and
  logical concepts?
• How the heck could we figure this out?

3
TLC

• No, not that lame TV channel ful of trading
  spaces and makeovers.
• The Teachable Language Comprehender
• Hierarchical associative network of concepts that
  began as a computer simulation

4
bacground

• Quinlan and his colleagues tried to make a
  computer program that would simulate how a person
  learns language
• Realized that concepts, not words, were the
  building blocks of knowledge
• Example the policeman held up his hand and the
  cars stopped (Collins Quinlan, 1973)

5

• You know exactly what that phrase entails
• Cop is directing traffic
• People push pedals etc
• Tacit knowledge if you will
• Three types of elements of semantic memory

6
The three elements

• Units, properties and pointers
• Unit is a thing (the cop, his hand, the car etc)
• Properties are conceptual (raising the hand that
  sort of thing)
• Pointers denote specific associations

7
So where does this get us?

• Semantic memory then is a HUGE hierarchical
  network of relationships between elements
• Is statements then are relationships between a
  superordinate element and a subordinate
• A bird is a fish will not produce a yes
• A fish is an animal and a bird is an animal will

8
Collins and Quinlan (1969)

• A _____ is a ______ statements
• Sentence verification
• RT longer when number of associative links was
  greater
• Less relevant property relationships, longer RT
• (a canary is yellow is longer than a canary can
  breather)

9
However

• Ok, by this theory, the satement a bear is a
  mammal should be quicker than a bear is an
  animal
• Ummm, no.
• So the semantic distance effect does not show
  up here

10
More buts

• Does not deal with the typicality effect at all
• Does not explain why a robin is a shark is more
  quickly rejected than a robin is a salmon
• Conrad (1972) said that it is the typicality of
  the statement itself that is the issue

11
Feature set theory

• Smith, Shoben and Rips (1974)
• Concepts stored as sets of attributes
• RT then depends on comparing features of exemplar
  with stored concept
• Predicts the symbolic distance effect (very
  similar items take long RTs
• Predicts the category size effect (concept to
  superordinate)

12
But.

• You knew there would be a but
• What are the defining features of a concept
• What is a dog for example
• Your idea is different than mine
• But we gt the same effects with subjects

13
Spreading activation

• Collins and Loftus (1975)
• Network like TLC
• NOT as rigidly hierarchical
• Still have units and properties, but not so
  hierarchical
• Distance is important

14
But..

• This one sounds better but there is a but
• Distantly related concepts should give longer RTs
• But a canary is a shark takes like no time

15
Propositional network theory

• Anderson (1983)
• ACT and ACT
• For example Kurts mother sent him a package
  last week
• Three propositions
• More propositions, sentences take longer, even
  with fewer words

16
Neural network models

• Units or nodes that are like neurons, on and off
• But, units have a threshold and once acticavted
  past this threshold you get firing
• Units are connected either excitatorily or
  inhibtorily
• There are activation rules
• Output rules, how a unit sends info to next unit
• Learning rules
• Groups of units or modules, are devoted to
  specific cognitive functions

17
properties

• Content addressable memory
• Network makes guesses
• Network makes spontaneous generalizations

18
conclusions

• I dont want to get too bogged down in this but,
  I think that we can make a few statements
• All the theories are about connections
• Probably NOT hierarchical, completely
• Activation probably spreads
• This stuff can get way hard.