Knowing Semantic memory
Semantic Memory Memory of the general knowledge of the world While episodic memory is personal – events that happened to you – semantic memory is more general – information that everyone can learn about the world
Two basic questions asked 1. What is the structure and content of semantic memory? Current perspective is that semantic memory is a network of nodes each representing a basic concept and nodes are linked together 2. How do we access the information in semantic memory? Accessing or retrieving information from the network involves spreading activation
Semantic memory models Quillen and Collins network model Smith’s feature comparison model
Collin and Quillian Model A network model – interrelated concepts or nodes are organized into an interconnected network – these connections can be direct or indirect Memory is the activation of a node which can spread to other nodes activating other memories Two forms of connections or propositions: Category membership “is a” Property statements “has”
Collin and Quillian Model
Collin and Quillian Model
Collin and Quillian Model
Smith’s feature overlap model Showed significant problems of the Quillen and Collins model Used lists of characteristics instead of a network Concepts are defined by a list of features. These features are stored in a redundant manner The decision of whether one concept is an example of an another depends upon the level of overlap
Smith’s feature overlap model
Smith’s feature overlap model Feature comparison Where features of two concepts overlap a great deal or very little, the decision is made quickly If some features overlap and others do not, then a stage 2 comparison has to be made and the decision is slower
Smith’s feature overlap model
Empirical Tests of Semantic Memory Models Sentence Verification Task: Simple sentences are presented for the subjects’ yes/no decisions. Most early tests of semantic memory models adopted the sentence verification task.
Challenges to Collin and Quillian Model Support for Collin and Quillian was cognitive economy – only nonredundant facts stored in memory. Conrad (1972) found that high frequency properties were stored in a redundant fashion
Challenges to Collin and Quillian Model Conrad (1972) found that high frequency properties were more highly associated with the concepts and are verified faster than low frequency properties – not shown in network model
Challenges to Collin and Quillian Model
Challenges to Collin and Quillian Model Typicality: The degree to which items are viewed as typical, central members of a category. Typicality Effect: Typical members of a category can be judged more rapidly than atypical members.
Challenges to Collin and Quillian Model
Modified Collin and Quillian Model
Semantic Relatedness Semantic Relatedness Effect: Concepts that are more highly interrelated can be retrieved and judged true more rapidly than those with a lower degree of relatedness. Resulted in a third revision of the model which required a 3-dimensional model
Categorization, classification, and prototypes Knowing Categorization, classification, and prototypes
Knowledge Knowledge is the acquisition of concepts and categories – your mental representations that contain information about objects, events, etc.
Categorization Concepts usually involve the creation of categories Categories – grouping things into groups based upon similar characteristics Categories help organize information so that you do not have learn about every new thing you expereince
Concepts and Categories Two basic questions: What is the nature of concepts? How do we form concepts and categories? Three approaches to these questions, classical, prototype, and exemplar
Classical Approach - Aristotle Categories have defining features – semantic features that are necessary and sufficient to define the category Necessary – features have to be present for inclusion Sufficient – if these features are present no other features are necessary for inclusion Problem – most members of a category do not have the same defining features
Prototypes A prototype of the category is developed The prototype has the semantic features that are most typical of the members of the category New objects compared to different prototypes of different categories, and are included in category with the most similar prototype Members of a category that are less similar to the prototype require longer to verify their inclusion
Prototypes (cont) Nonmembers of a category can be seen as members if they are similar to the prototype and the differences are not known When asked to name members of a category, those members most like the prototype are named first Priming most effected by prototypes
Exemplars Identification of examples or exemplars of the category New objects are compared to to other objects you have seen in the past – your exemplars Advantage of the use of exemplars – it uses actual examples not just a constructed prototype – atypical members can be exemplars of a category
Prototypes and Exemplars Evidence supports both models of categorization One possibility is that we use prototypes in large categories and exemplars in defining smaller categories
Feature comparison theory of determining category membership This model focuses on the strategy used to decide whether an exemplar (i.e. a canary) is a member of a larger category (i.e. bird) This strategy consists of two rules: If the feature associated with the exemplar (canary has feathers) is found to be associated with the larger category (birds have feathers), it provides positive proof the exemplar is a member of the larger category If the feature is not associated with the category (bats have fur), they are not members of the category (a bat is not a bird)
Support for Feature comparison model Consistent with typicality effects – typical exemplars have extensive overlap of features; atypical exemplars have less overlap and require more time to determine their membership Consistent with the false relatedness effect- subjects respond faster when the exemplar is unrelated to the category than when it is somewhat related Also consistent with levels effects
Level effects Categories are organized in a hierarchy – one category is part of a larger category which is part of an even larger category Superordinate category – largest and most abstract – animal Subordinate category – smallest and least level of abstraction – a canary Base level category – in the middle and at an intermediate level of abstraction - bird
Base level categories Most useful and most likely to come to mind and tend to be the most important Children develop base categories before superordinate or subordinate categories When asked to identify pictures, people more likely to give base level category
Category levels When asked for common attributes of superordinate category, people give very few (vehicle) When asked about attributes of base level categories, many more given (car) When asked about attributes of base level categories, not many more than those given at the base level are added (SUV) Movement from a superordinate category to a base level category results in a great increase in information, but movement to a subordinate category adds very little information
Base level thinking Humans prefer to think a the base level of categorization because it provides the most useful information for predicting membership in a category Superordinate members of a category maybe very different with few similarities – fruit Base level share many common features – apples Subordinate categories are more informative , but are poor discriminators – McIntosh apples share many features of other apples Subordinate level thinking most important in areas of expertise. Choosing wine for dinner
Knowing Connectionism
Importance of context Context can act as a prime to understanding correct meaning I saw a man eating fish. Visiting relatives can be boring Context can activate the meaning meant to be conveyed By understanding the context of a communication, you can understand and remember the material better
Connectionist model of semantic memory Involves a network of interconnected nodes each node connected with specific information The connections between nodes vary in strength – referred to as connection weights Nodes that are more strongly connected have greater connection weights Learning involves strengthening the connection by increasing connection weights
A neural network
A neural network example