Chapter 7 (Knowledge Representation)

How does Collins and Loftus' SPREADING ACTIVATION MODEL differ from Hierarchical models?

Collins and Loftus (1975) presented an elaboration of the Collins and Quillian (1969) hierarchical network model that they called spreading activation theory. In general, these authors sought both to clarify and to extend the assumptions made about the manner in which people process semantic information. They again conceived of semantic memory as a network, with nodes in the network corresponding to concepts. They also saw related concepts as connected by paths in the network. They further asserted that when one node is activated, the excitation of that node spreads down the paths or links to related nodes. They believed that as activation spreads outward, it decreases in strength, activating very related concepts a great deal but activating distantly related nodes only a little bit. - different nodes are weighted differently based on there importance and relevance to the activated node. They dispensed with the assumptions of cog- nitive economy and hierarchical organization, helping their model avoid the problems of the Collins and Quillian (1969) model. - it is hard to think of data that would falsify the model. The proposal is therefore regarded more as a descriptive framework than as a specific model. spreading activation, the idea that excitation spreads along the connections of nodes in a semantic network. If the word above bread was butter, participants were quicker to respond than to other unrelated words. • Got rid of hierarchy • Got rid of cognitive economy • Allowed links to vary in length to account for typicality effects • Spreading activation - Activation is the arousal level of a node - Spreads down links - Used to extract information from network

What is an example of your PROCEDURAL MEMORY system according to the ACT model?

For example, when you ride a bicycle, swim, or swing a golf club, you are thought to be drawing on your procedural memory. Here's another example of procedural memory. Right now, almost all the telephones I use have touch-tone pads for dialing. I "know" many phone numbers only by the sequence of moves I make to enter the number on the keypad. If someone asks me for one of these phone numbers (a task that requires me to state information in words), I often find myself at a loss; then I start "dialing" on an imaginary keypad, watching where my finger goes and "reading off" the phone number based on the motions of my finger.

What is the TYPICALITY EFFECT?

Rips et al. (1973) found that responses to sentences such as "A robin is a bird" were faster than responses to "A turkey is a bird," even though these sentences should have taken an equivalent amount of time to verify. In general, typical instances of a concept are responded to more quickly than atypical instances; robins are typical birds, and turkeys are not. The hier- archical network model did not predict typicality effects; instead, it predicted that all instances of a concept should be processed similarly.

What are the important aspects of the EXEMPLAR view of concepts?

The exemplar view of concepts sserts that concepts include representations of at least some actual individual instances. The exemplar approach assumes that people categorize new instances by comparing them to representations of pre- viously stored instances, called exemplars. it thus explains people's inability to state necessary and defining features: There are none to be stated. It also explains why people may have difficulty categorizing unclear, atypical instances: Such instances are similar to exemplars from different categories (for example, tomato is similar both to fruit exemplars, such as oranges or apples, and to vegetable exemplars, such as beets or squash) or are not similar enough to any known exemplars It fails to specify, for example, which instances will eventually be stored as exemplars and which will not. It also does not explain how different exemplars are "called to mind" at the time of categorization. Exemplar view of concepts • Concepts include representations of at least some specific instances (not just an idealized prototype). • People categorize new instances by comparing them to specific instances of categories (exemplars). • Does not require necessary and sufficient features of a concept When do we store information about individual exemplars? • When it is useful to distnguish among members of a category (pet vs. aeack dog) • If the learning situation allows us to get to know individual instances well • If the stimuli lend themselves to hypothesis testing • If instances do not belong to a large number of different categories at once

Compare and Contrast the THREE MODELS described to Organize and Store Knowledge Hierarchical Network

The model was called a hierarchical semantic network model of semantic memory, because researchers thought the nodes were organized in hierarchies. Most nodes in the network have superordinate and subordinate nodes. A superordinate node corresponds to the name of the category of which the thing corresponding to the subordinate node is a member. So, for example, a node for "cat" would have the superordinate node of "animal" and perhaps several subordinate nodes, such as "Persian," "tabby," and "calico."

Understand the key theory underlying connectionist models

[let] go of the idea that items are stored in memory as such. Instead the funda- mental idea is that what is stored in memory is a set of changes in the instruc- tions neurons send to each other, affecting what pattern of activity can be constructed from given inputs. When an event is experienced, on this view, it creates a pattern of activity over a set of processing units. This pattern of activ- ity is considered to be the representation of the event. The formation of this pattern of activity provides the trigger for the creation of the instructions. The set of instructions is then stored in the connections among the units, where it is available for use in the construction of subsequent patterns of activity. Under some circumstances—for example, when the constructive process takes place in response to a recall cue—the cue may result in the construction of a pattern of activation that can be viewed as an attempted reconstruction of the pattern that represented the previously experienced event. Such a reconstructed repre- sentation corresponds to a recollection. The patterns themselves are not stored, and hence are not really "retrieved"; recall amounts not to retrieval but to reconstruction." Connectionist models • Model "learns" to develop patterns of activation through many trials with training examples. • Initially, connections between nodes are set at random strengths (weights); experience leads these connections to be activated more or less strong.

What is a Script

a schema for a routine event, such as going to the dentist, or eating at a restaurant - Allows us to make inferences about facts not stated - Central concepts to a script are better remembered than concepts of less importance to the script -fill in details, don't give more information than is necessary - no need to say pick up the fork.- can rearrange order to correct order.

Be familiar with the three types of memory systems defined by the ACT model.

adaptive control of thought (ACT) model of memory, Based on analogies to computers, ACT has given rise to several computer simulations of cognitive processing of different tasks. ACT models do not make the semantic/episodic distinction described earlier, but distinguish among three kinds of memory systems. The first is working memory, thought to contain information the system is currently using. The other two kinds are declarative memory and procedural memory. Declarative memory contains knowledge, facts, information, ideas—basically, anything that can be recalled and described in words, pictures, or symbols. In contrast, procedural memory holds information concerning action and sequences of actions its implicit.

Basic-level of categorizes.

basic level of categories A hypothesized type of concept thought to be at a psychologically fundamental level of abstraction. Although concepts exist at many different levels of a hierarchy (for example, "Bernese mountain dog," "dog," "canine," "mammal," "animal"), one level of abstraction appears psy- chologically fundamental. They called this the "basic" level and distinguished it from both higher-level (superordinate) and lower-level (subordinate) concepts. "Piano" and "guitar" are examples of two basic-level categories.

What are the important aspects of the PROTOTYPE view of concepts?

prototype view of concepts: The idea that all concepts are organized around idealized mental representations of examples. Members of a concept share "family resemblance" rather than a specific set of necessary features. the more features a member possesses, the more typical it is. =typicality effects. Robin more features of prototypical bird than penguin. Prototypes are often thought of as mental "summaries" or "averages" of all the instances

What is a schema

schema An organized framework for representing knowledge that typically includes characters, plots, and settings, and incorporates both general knowledge about the world and information about particular events. • Schema: a large unit of organized informa8on used for represen8ng concepts, situa8ons, and events

Subordinate Level Of Categorization

subordinate level of categories; these categories are less distinct than are two basic-level categories, such as "piano" and "guitar." subordinate level of categories A hypothesized type of concept thought to make fewer distinctions than does a basic level concept.- e.g.. classical guitar, vs. folk guitar. is less of a distinction than is Guitar vs. Drum.

superordinate levels of categories

superordinate level of categories A level of categorization broader than the basic level, including exemplars that can be quite dissimilar from one another. e.g.,musical instruments. or fruit.

Understand the principle of COGNITIVE ECONOMY

• Facts or properties are stored at the highest appropriate level and assumed to be true for all lower nodes. • "Does a dog have live young?" requires us to travel up the network to "mammal," where the appropriate fact is stored. • "Does a dog have a tail?" can be answered more quickly, because tail is stored with the "dog" node. This way you don't have to remember, for example, has live young for every single mamal, but rather can learn that the overarching category of mama entails all mammals have live young.