Feature Comparison Model

FEATURE COMPARISON MODEL “The part of long-term memory dealing with words, their symbols, and meanings is semantic memory.” Semantic memory allows humans to communicate with language. In semantic memory, the brain stores information about words, what they look like and represent, and how they are used in an organized way. It is unusual for a person to forget the meaning of the word "dictionary," or to be unable to conjure up a visual image of a refrigerator when the word is heard or read. Semantic memory contrasts with episodic memory, where memories are dependent upon a relationship in time. An example of an episodic memory is "I played in a piano recital at the end of my senior year in high school." Models of semantic memory make assumptions about both the structures of knowledge and the processes that operate on these structures. Direct-storage models, such as Collins and Quillian’s, tend to make elaborate assumptions about the structuring of knowledge and explain differences in verification times in terms of

differences

differences

in

in

underlying

the

number

knowledge of

levels

in

structures the

(e.g.,

hierarchy).

Computational models such as Smith, Shoben, and Rips’, on the other hand, make minimal assumptions about the structuring of knowledge; instead, they elaborate on the processes that operate in semantic memory and explain differences in verification times in terms of differences in processing. In examining Smith, Shoben, and Rips’ model, we will first look at

how they assume semantic memory is structured. Then we will examine the detailed processing model they proposed for operating on these structures. Smith, Shoben, and Rips proposed a model called a feature comparison model of semantic memory. The assumption behind this model is that the meaning of any word or concept consists of a set of elements called features. Features come in two types: Defining, meaning that the feature must be present in every example of the concept, and Characteristics, meaning the feature are usually, but not necessarily present. According to Smith et al., concepts are stored in semantic memory as sets of attributes, called semantic features. The following sets of features illustrate how the concepts robin and bird might be represented in someone’s memory. ROBIN = {has wings, lays eggs, has feathers, can fly, is redbreasted, eats worms} BIRD = {has wings, lays eggs, has feathers, can fly, eats worms} The features associated with a given concept vary in the degree to which they are central to defining the concept. Those features which are essential to defining the concept are called defining features; those features which are often associated with a concept but which are not essential to its definition are called characteristic features.

Defining features are attributes that are shared by all members of a category. For example, the features has wings, lays eggs, and has feathers, are all defining features of the concept bird because all birds have these attributes. Characteristic features are attributes that are shared by many, but not all, members of a category. The feature can fly, for example, is a characteristic rather than a defining feature of the concept bird because most, but not all, birds can fly. Assuming the semantic memory is organized in terms of feature list, question arises, how is it knowledge retrieved and used? According to Smith et al. model verification of sentences as “robin is a bird” is carried out in two stages. In the first stage, the

feature

list

(containing

both

the

defining

and

the

characteristics features) for the two terms are accessed, and a quick scan and comparison is performed. If the two lists show a great deal of overlap, the response ‘true’ is made very quickly, if the overlap is very small, then the response ‘false’ is made, also very quickly. If the degree of overlap in the two feature list is neither extremely high nor extremely two low, then a second stage of processing occurs. In this stage the, a comparison is made between the sets of defining features only. If the list match the person respond ‘true’; if the list do not match, the person respond ‘false’. This process can be easily understood through following flowchart-

The feature comparison model can explain many finding that the hierarchical network model could not. One finding it explain is the typicality effect: sentences such as “ a robin is a bird” are verified more quickly then sentences such as “a turkey is bird” because robin being more typical examples of birds, are thought to share more characteristics feature then ‘bird’ then do turkeys. The feature comparison model also explains fast rejection of false sentences, such as “a table is a fruit.” In this case the list of feature for ‘table’ and the list of ‘fruit’ presumably share very few entries. The feature comparison model also provides an explanation for a finding known as the category size effect. This term refers to the fact that if one term is a subcategory of another term people will generally be faster to verify the sentence with the smaller category. That is people are faster to verify the sentence “a collie is a dog” than to verify “a collie is a animal,” because a set of dog is a part of the set of animals. The feature comparison model explain this effect as follows, it assumes that as category grows larger (for example from robin, to bird, to animal, to living thing)

they

also

become

more

abstract.

With

increased

abstractness, there are fewer defining features. Thus in first stage of processing there is less overlap between the feature list of the term and the feature list of an abstract category. The model also explains how “hedges” such as “a bat are sort of like of bird” are processed. Most of us know even though bat fly and eat insects, they are really mammals. The feature comparison model explains that the processing of hedges consist

of a comparison of a characteristic features but not the defining features. Because bat share some characteristics features with birds (namely flying and eating insects), we agree they are “sort of like” birds. We recognize, however that bat are not really birds, presumably because they don’t share the same defining features.

Depiction of Smith et al. Feature Comparison Model Read sentence Retrieve feature lists of subjects and predicate nouns. Compare both lists.

Degree of similarity? High

Low Medium

Respond “true”

Compare defining features only

Respond “false”

Lists match?

Respond “true”

Respond “false”