PSY246 Study Guide - Final Guide: Music Therapy, Frontal Lobe, Change Detection
PSY246 Exam Notes (Week 6 onwards):
Semantic memory: Week 7
Lecture outline;
• Episodic vs. semantic memory
• Theories of semantic memory organization
o Word meanings (lexical semantics): led to computational investigations of natural
language processing (machine translation; text corpus analysis)
o Sentence verification experiments
o Network models
▪ Hierarchical network model
▪ Spreading activation model
o Feature comparison model: distributed representation
• Disorders of semantic memory:
o Neuropsychological studies (category-specific deficits)
▪ Perceptual-functional theory
▪ Distributed-plus-hub (‘hub and spoke’ model) theory
Division of long term memory –
• Declarative memory (facts)
• Procedural memory (actions, motor skills)
• UNDER declarative memory;
o Semantic memory (general knowledge)
o Episodic memory (dated recollections of events) → events tied to temporal and spatial
context
Semantic memory (also called conceptual knowledge) is the aspect of human memory that corresponds
to general knowledge of objects, word meanings, facts and people, without connection to any
particular time or place. Knowing that you ate Szechuan scallops at the Peking Restaurant in
Cambridge last Thursday evening is episodic, not semantic, memory. Knowing that Szechuan refers to
a province of China, that food from this region tends to be spicy and that scallops are sea-creatures
that live in brittle bivalve shells are all forms of conceptual knowledge. Memory for episodic events is
not only specific to times and places, it is also largely specific to an individual. Conceptual knowledge
on the other hand, is mostly shared across individuals in a given culture, although its precise scope
depends on the individual’s experience.
Episodic vs. semantic memory –
• Semantic memory = conceptual knowledge, linguistic knowledge, memories for general facts
• E.g. dogs have fur, Ottawa capital of Canada
• Impairments in semantic memory = cannot comprehend meanings of words or pictures or
express ideas
Experiments on semantic memory organisation –
• Early studies used the sentence verification task
• DV = reaction time
• Subject-predicate: ‘a canary is a bird’
• Sentence types
find more resources at oneclass.com
find more resources at oneclass.com
o Set inclusion; describe the subject being a member of a category = a canary is a bird
(true), a whale is a fruit (false)
o Property-attribute = a canary has feathers (true), a whale has seeds (false)
Network models –
• Concepts are represented by nodes (localist representation)
• Relationships between concepts are represented by links
o Set inclusion (canary is a bird): isa
o Property attribution (canary can sing): has
Hierarchical network model (Collins and Quillian, 1969)
• Concepts are organised in a hierarchy
• Cognitive economy
o Property attribute is stored non-redundantly at the highest (most general) level → e.g.
“Leonardo da Vinci had knees”; da Vinci is a human being – human beings have knees
• Sentence verification RT is a function of levels – in order to verify info like ‘canary has skin’ –
you need to access the canary node AND the animal node
• In order to verify info, you need to traverse two links
• ‘A canary can sing’ is verified quicker than ‘a canary can fly’ or ‘a canary has skin’
• Hierarchy e.g. canary to bird to animal
Problems with the hierarchical network model:
• Challenge to the cognitive economy
• Conrad (1972)
o Argued that RT data better explained in terms of frequency of co-occurrence of concept
and property rather than levels
▪ RT varied with the subject-property frequency (associative strength: determined
from norms) within a level
▪ E.g. within Level-1 sentence: birds have feathers
▪ High frequency property = peacock has feathers
▪ Low frequency property = canary has feathers
o Results showed RT is a function of frequency of co-occurrence (associative strength)
Other problems with the hierarchical network model:
• Hierarchical network model does not predict –
o RTs do not always mirror hierarchical relationship e.g. a dog is an animal < a dog is a
mammal
o Within-category; typicality effects e.g. a canary is a bird < an ostrich is a bird
o Negative (‘false’) judgements are not faster for closer concepts; in fact the opposite –
e.g. a canary is a salmon < a canary is an ostrich → people took longer to verify canary
is an ostrich, compared to canary is a salmon
Spreading activation model: (Collins and Loftus, 1975)
• Network model
o Concepts are organised non-hierarchically
▪ Explains lack of hierarchical effect
find more resources at oneclass.com
find more resources at oneclass.com
o Links vary in associative strength
▪ Explains typicality effect (bird-canary < bird-ostrich)
▪ Still does not explain well how negative judgements are made
o Activation of a concept spreads to other concepts linked to it
▪ Explains semantic priming effect
Semantic priming effect:
• Response to a word is faster following a semantically related word e.g. lexical decision task –
is ‘truck’ a word? Car-truck < flower-truck
• Semantic priming effect is used to study:
o Organisation of semantic memory
o Automaticity (e.g. unconscious processing/masked priming)
Feature comparison model:
• Network models assume knowledge is represented within a concept node (localist
representation)
o e.g. canary: is yellow, wings, sings
o does not explain well how knowledge of a concept can be partially degraded e.g. ‘who
is the actor that appeared in the “Pirates of Carribbean” and tried to smuggle cats into
Australia?’; how many animals each did Moses take on the ark?
• Feature comparison model assumes concept is represented as distributed features in semantic
space
Distributed semantic features –
• Multidimensional scaling (based on similarity ratings)
o Participants rate how similar a pair of concepts are e.g. sheep-goat; sheep-lion
o Data are represented in semantic space and the underlying dimensions (Features) are
extracted e.g. domesticity (wild-domesticated); size (small-large) etc.
Explaining the sentence verification data with the feature comparison model –
• Two-stage decision model
o Assumes that decisions are made by comparing the similarity of features of subject and
predicate terms
▪ E.g. canary (subject) is a bird (predicate)
o Assumes that features can be two types
▪ Defining features
▪ Characteristic features
o Defining vs. characteristic features
▪ Defining features = essential features e.g. bird: wings, has a beak
▪ Characteristic = less important features e.g. bird: can fly, builds nests
Feature comparison model: Two stage decision process –
• (first stage) Compare ALL features of subject and predicate terms to determine overall
similarity
• (second stage) Compare DEFINING features of subject and predicate terms
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Division of long term memory : declarative memory (facts, procedural memory (actions, motor skills, under declarative memory, semantic memory (general knowledge, episodic memory (dated recollections of events) events tied to temporal and spatial context. Semantic memory (also called conceptual knowledge) is the aspect of human memory that corresponds to general knowledge of objects, word meanings, facts and people, without connection to any particular time or place. Knowing that you ate szechuan scallops at the peking restaurant in. Cambridge last thursday evening is episodic, not semantic, memory. Knowing that szechuan refers to a province of china, that food from this region tends to be spicy and that scallops are sea-creatures that live in brittle bivalve shells are all forms of conceptual knowledge. Memory for episodic events is not only specific to times and places, it is also largely specific to an individual.