Q2: Lecture 12 Flashcards
similarity-based theories
offer different suggestions about how the mind might
represent concept knowledge. These theories are considered similarity-based because each one
suggests that category membership is determined by judging how similar an instance is to a relevant
concept representation.
feature set theory
an account of conceptual memory.
According to feature set theory, concepts are represented as sets of semantic features that
collectively define concepts.
semantic features
meaningful properties like MADE OF METAL,
CAN FLY, and IS ALIVE. By this view, a concept like BOOKSTORE would be composed of a set of
features that includes HAS WALLS, SITE OF RETAIL SALES, and CONTAINS BOOKS
defining feature
necessary and, therefore, are common to every instance of the concept. In other words, if something does not exhibit a defining feature, then it is not an example of the corresponding
concept. For example, it is necessary for any triangle to have three sides, so three-sidedness would be considered a defining feature of the concept triangle
characteristics feature
are typical of concepts, but they are not necessary and, as a result, not all instances of the concept exhibit the characteristic features. For example, consider the
concept MUSIC. Many instances of the concept MUSIC exhibit the feature MELODIC, but not all
music is melodic. MUSIC is typically MELODIC, but it doesn’t have to be. Thus, MELODIC is a characteristic feature of the concept MUSIC rather than a defining one
Stage 1 processing
involves a random selection of the features from both the to-be-identified object and a concept representation stored in memory. The two feature subsets are mentally compared and if there is a high degree of similarity between the two subsets, then the to-be-identified object is considered a member of the category defined by the concept representation; viewed as a very fast process and is described as “global” because it involves sampling from both feature types-defining and characteristic
Stage 2 processing
involves a thorough comparison of features of a to-be-identified object with the defining features of the concept representation. Compared to Stage 1, Stage 2 is a controlled mechanism because characteristic features are ignored, and defining features are processed in a slow, deliberate manner.
Smith, Shoben, & Rips (1973)
made use of a statement verification task in which people read a series of statements about category membership; reasoned that verification times should be relatively short if judgments require only Stage 1 processing, but relatively long if both stages of processing are required.
prototype theory
are not feature-based and the distinction between defining and characteristic
features is, therefore, irrelevant. According to prototype theory, concepts are represented in abstract
terms as idealized forms.
Rips (1975)
found was that responses to the question were easily predicted by the nature of the species that was specified as infected. When people were told that all the DUCKS were infected, they
almost invariably suggested that GEESE might also become diseased. When told that all the HAWKS
were diseased, they concluded that the EAGLES were most likely to also become infected. Rips
interpreted these results as consistent with prototype theory by suggesting that people assumed the
species with the most similar prototypes were likely to be most similar in terms of susceptibility to the
imaginary disease.
Posner & Keele (1968)
wanted to study concept formation from the ground up. That is, they were
interested in studying the acquisition of new concepts – concepts about which research participants
had no prior knowledge. For this reason, they studied a rather strange but easy to manipulate class of
stimuli – dot patterns.
prototype pattern
There was nothing inherently special about the prototype pattern, it was only a prototype because of
the relationship that Posner and Keele (1968) created between it and other dot patterns used in the
study.
low distortion pattern
based on the prototype pattern, but each was somewhat different from the prototype. For every low-distortion pattern, the location for every dot was changed so that it shifted a little bit in a random direction. Thus, the low-distortion patterns were always different from the prototype, but also substantially similar.
high distortion pattern
were variations on the
prototype. Again, every dot of the prototype pattern was changed, but this time it was moved in a random direction by a greater distance than was used for the low-distortion patterns. Thus, the high-distortion patterns were somewhat similar to the prototype, but less similar than the low-distortion patterns were.
endorsement rate
increased as dot patterns became more like the prototype, with the prototype being endorsed most reliably even though people had never seen it before
exemplar theory
be viewed as the exact opposite to prototype theory. According to exemplar theory, a concept is not stored in memory as a single set of features or a single abstract representation. Instead, our knowledge of any concept is composed of the entire collection of specific examples of that concept we have ever encountered
Allen & Brooks (1991)
One of the best examples of research supportive of the exemplar perspective on concept knowledge; Their study was like the Posner and Keele (1968) study, in
that it involved a study phase and a test phase in which participants were assessed for their abilities
to categorize stimuli. The similarity between the two studies, however, pretty much ends there; participants were given a simple rule to memorize
categorization rule
If an animal exhibits at least two of the following
features, that animal is a Constructor: Long Legs, Angular Body and/or Spots. By contrast, any cartoon animal that does not exhibit at least two of those characteristics is an Excavator. Thus, an animal with long legs and angular body is a Constructor, an animal with long legs and spots is a Constructor, an animal with an angular body and spots is a Constructor, and an animal with long legs, an angular body, and spots is a Constructor. All other animals are Excavators
positive match
the test drawings looks very much like study drawings in which the same type of animals were seen before. In this example, the Constructors are seen in a woodland environment both times.
negative match
the test drawings look like the study drawings in
which a different type of animal was seen before. In this example, the Constructors are seen in the
arctic environment in which Excavators were previously seen. By using these two different types of critical test trials, Allen and Brooks (1991) were able to separate two different influences on
categorization task performance: performance based on rule use, and performance based on
exemplar similarity
concept node
The meaning of each concept node is defined by that node’s relationship to the other nodes in the network. In this way the meaning of any specific concept is determined by its relationships with many other concepts within the network. This means that concept meanings become relative and dependent upon one another, rather than being independent
representations as was the case for the non-network theories we’ve reviewed
network models
TLC model & spreading activation model
TLC model (Collins & Quillian, 1969)
viewed as less valid, it nevertheless served as a precursor to the more successful spreading
activation model; TLC stands for Teachable Language Comprehender. It is named the TLC model because Collins and Quillian (1969) were interested in developing an artificial system that could understand language. At the outset, however, they realized that a language comprehending system would need to have access to an extensive and organized knowledge base. For this reason, Collins and Quillian (1969) established a theory of conceptual knowledge that could be accessed by their system, and today this is seen as the primary legacy of the model they proposed.
hierarchical organization
requires that concepts be arranged so that the most general ones are at the highest levels in the network, and more and more specific concepts
are arranged at appropriate levels underneath. In other words, superordinate concepts are placed above subordinate concepts.
property tag
identified an important attribute of the concept. In addition, property tags were
attached only to the most general concepts to which they applied, and all subordinate concepts
inherited the properties of the superordinate concepts.
principle of inheritance
allowed the semantic network to exhibit a property called cognitive economy. This is because the same information did not need to be represented over and over for every subordinate concept, but instead the information could be represented more economically by including it only once
cognitive economy
This principle of inheritance allowed the
semantic network to exhibit a property called cognitive economy.
category size effect
the expectation
that verification times will always be longer for statements about larger categories. This prediction follows from the idea that longer search pathways will be traced when statements concern larger categories, and because searching takes time, the larger search
will also be longer
typicality effect
people reliably verify A CANARY IS A BIRD more quickly than they verify AN OSTRICH IS A BIRD. This is called a typicality effect because the more typical category member (CANARY) is verified more rapidly than the less typical category member (OSTRICH). Typicality effects are very
common and robust, but the TLC model cannot explain them. Furthermore, measurements of typicality are much better predictors of statement verification times than category size.
production tasks
The most common way of quantifying the typicality of a category member is to use normative data from production tasks. A production task simply involves presenting people with the name of a category and asking them to name category members. Data collected from large samples is then tallied to establish a rank order of the responses from most common to the least common. This process shows that CANARY is generated much more often OSTRICH in response to the category BIRD.
production norms
Production norms determined in this way are very strong predictors of statement verification
times.
spreading activation (collins & Loftus, 1975)
The spreading activation model was proposed by Collins and Loftus (1975) and is a revision of the
TLC model that Collins had co-authored with Quillian. Like the TLC model, the spreading activation model is a network model of semantic memory in which concepts are represented as nodes within an expansive interconnected network. Unlike the TLC model, however, the concept nodes of the spreading activation model are not organized hierarchically. Instead, concept nodes are organized according to semantic relatedness. This means that concepts that are highly meaningfully related are located close together in the network, whereas concepts that are less related are located farther apart.
semantic distance
the degree of semantic relatedness is inversely related to the semantic distance between concepts within the network; The model also makes use of indirect concept connections to represent greater semantic distances.
