The Collins and Quillan model of Semantic Memory (69, 72,
75)
An extensive theory of semantic memory, comprehension, and meaning.
Basic Architecture
Semantic concepts are represented as nodes : a point or location in semantic space.
Nodes are connected together with pathways : labeled, directional associations between concepts.
These labeled pathways specify the relationship between two concepts.
Proposition : a relationship between two concepts in a network.
Related nodes are connected by pathways to form semantic networks : An interrelated set of concepts, or body of knowledge.
How is meaning accessed in the C & Q model ?
Spreading Activation : the mental activity of accessing and retrieving information from a semantic network.
Activation spreads along the pathways through the network once a single node is activated (by hearing an auditory or visual
code which represents that node).
Related conceptual nodes receive increased activation by a process known as Priming
Spreading Activation is an automatic process.
Activation patterns within networks allow us to make Sentence Verification judgements by finding (or not finding) connected
pathways between nodes.
Sentence Verification Task : simple sentences are presented to the subject who has to make a yes/no decision concerning the
truth of the stated proposition.
p1 A robin is a bird
p2 A robin is a mammal
p3 A robin is an animal
Semantic Distance : The idea that related concepts are stored "closer" together in semantic space than unrelated concepts.
The shorter the semantic distance between two concepts, the quicker we can make a sentence verification decision.
RT p1 < RT p2 or p3
Superordinate Relationships and Property statements
In addition to the superordinate relationships specified by the previous sentences :
p1 A robin is a bird :
Superordinate (or categorical) Relationship
We can also ask about specific properties of a node in sentence verification tasks.
p4 A robin has wings : Property Statement
Verification judgements of Superordinate relationships is faster than judgements of property statements.
Superordinate relationships and the typicality effect.
Typicality effect : typical members of category can be judged more quickly than atypical members.
Consider the following two sentences :
p1 : A robin is a bird
p2 : A penguin is a bird
RT p1 < RT p2
If we ask a sample of individuals to "name some birds" , Robin will be a much more common response (higher frequency) than
chicken, penguin, ostrich, or other atypical members of the bird category.
The most typical member of a category is called the prototype.
Their studies led Collin & Quillan & Loftus to theorize that our network of semantic knowledge is organized according to
Semantic Relatedness
The semantic relatedness effect : concepts that are highly interrelated can be judge "true’ more rapidly than concepts which are
only slightly related.
The semantic relatedness effect is synonomous with the semantic distance effect.
One aspect of C & Q model which was not supported is the idea of strict cognitive economy.
Concepts can be stored more than once within semantic networks.
Semantic Relatedness and Priming
The Lexical Decision Task (LDT)
Subjects must decide whether a letter string represents a proper English word.
Girl : Yes, a word
Girk : No, not a word (nonword)
A typical visual trial proceeds as following :
Fixation + (50 ms)
Prime word nurse (100 ms)
Masking Pattern xxxxx (400 ms)
Target doctor (appears until response
The stimulus onset asynchrony (SOA) is 500 ms
We can vary many variables in the LDT
Prime - target relationship :
Unrelated : Nurse Cabin
Semantic : Nurse Doctor
Nurse Stethoscope
Orthographic : bleed bleep
& Phonological sound round
Nonword : Nurse girk
The specific relationship between the prime and target will determine whether we see
Facilitation (positive priming) : decreased reaction time to make a lexical decision, or
Inhibition : increased reaction time to make a lexical decision.
The second key manipulation in LDT
By varying the time between the presentation of the prime and the presentation of the target (stimulus onset asynchrony or
SOA)
We can examine both automatic and strategic processing.
Spreading activation is the automatic and fast acting process responsible for priming at very short SOA’s
Positive priming (facilitation) with SOA’s from 33 ms to about 400 ms is almost exclusively due to the automatic spread of
activation within the semantic network.
However, what about conceptually driven (strategic, top-down) processing which we constantly engage in with language
comprehension and production ?
Strategic Priming or Expectancy
Human beings are active information processors
When the prime word appears, people may use that prime word to generate an
Expectancy Set : a small list of words related to the prime which the subjects believe might appear as a target word.
How the expectancy set is generated depends on list context, and previously noted relationships between primes and targets.
Let’s consider Two different LDT lists with 100 prime target pairs.
List A : 50 word targets, 50 nonword targets, 40 of the word targets are semantically related to their prime
List B : 50 word targets, 50 nonword targets, but only 5 of the word targets are semantically related to their prime.
List A will create a greater Expectancy within the subject that will cause them to generate an expectancy set of words related to
the prime, because they notice very quickly the relationship between primes and word targets.
List B will probably produce less expectancy, and therefore less positive priming than list A, bc subjects will be much slower to
pick up on the possible meaning relationship between prime and target due to the small relatedness proportion.
How can we experimentally distinguish automatic spreading activation (ASA) and expectancy ?
Neely (1977) : Manipulated expectancy by telling subjects beforehand what type of expectancy set they should generate.
No Shift Group : Told to expect members of the same category:
Given Bird, generate types of birds in your expectancy set (no category shift)
Category Shift group : Told that
When you see "Body" as a prime, generate an expectancy set of Building parts (doors, windows, walls)
When you see "Building" as a prime, generate an expectancy set of Body parts (arm, leg, hand)
Neely (1977) (continued pg. 2)
By manipulating the SOA and the creation of the expectancy sets, he hoped to examine the time course of semantic priming.
For the No category shift group (Bird, expect sparrow or robin) , we saw
Facilitation for related targets at all SOA’s, both short and long (250 ms, 400ms, 700ms, and 2000ms)
Also Inhibition was observed for unrelated targets (i.e. BIRD – rake) at the longer SOA’s of 400ms, 700ms, and 2000ms
Neely (1977) (continued pg. 3)
For the category shift expectancy group, a much more complicated pattern of priming results appeared.
At the Short SOA’s of 250 and 400 ms :
We saw positive priming within categories (Body – heart), despite being told to expect a category shift.
ASA is the mechanism responsible for this short SOA priming.
At longer SOA’s of 700 and 2000ms, we now see significant inhibition to within category targets (Body – heart)
This is evidence that ASA does not last longer than 400 ms. ASA is a rapidly decaying phenomenon.
Neely (1977) (continued pg. 4)
When there was the expected category shift, a different priming pattern also appeared.
At short SOA’s of 250 and 400 ms, there was only the slightest hint of facilitation.
The expectancy set takes time to generate, and those two SOA’s were too short for the subject to take advantage of their
generated expectancy set.
At the longer SOA’s of 700 and 2000ms, now we see significant facilitation when the expected category shift occurred. (i.e. Body
–door)
Neely (1977) (continued, pg. 5)
Results from both experimental groups allow us to more precisely state the relationship between ASA, expectancy, and priming.
ASA is the fast acting and rapidly decaying activation of related concepts which is responsible for priming at very short intervals.
Expectancy is the strategic and deliberate memory search which is responsible for priming at longer SOA’s (in particular, any
SOA greater than 500ms)
Distinguishing Facilitation from Inhibition
Prime Target RT Result
disk nurse 674ms 32ms Inhibition
xxxx nurse 642ms Baseline
doctor nurse 604ms 38ms Facilitation
In many priming studies, neutral primes are not used, and
priming effects are determined by subtracting related RT
from the unrelated RT
Unrelated : 674 ms
Related : 604 ms
Result : 70 ms of positive priming
Lexical Ambiguity and Priming
Many words have multiple meanings :
Bug : insect or spying device
Count : a verb or a royal title
The more frequent meaning is referred to as the dominant
meaning, while less frequent meaning is the subordinate
meaning.
When activation starts, it spreads to all meanings of the
words, but very quickly, the dominant meaning shows
greater priming.
The more definitive the surrounding context (therefore, the
more the person is relying on expectancy), the more likely it
is that only one meaning will be highly activated.