Chapter 9 Knowledge

what are the differences between a connectionist model and the Collins and Quillian model?

-Connectionist networks can learn through the adjustment of the weights of the connections; the C & Q model can't learn - knowledge is already set. -Concepts on connectionist network are represented by a distributed pattern of activation across the hidden units - in the C & Q model concepts are represented by a single node. -Connectionist networks operate more like the brain operates

How does expertise affect categorization?

But a person's knowledge can also affect how we categorize people with more expertise tend to focus on the specific level. This shows that learning and experience can influence how we categorize objects

What does it mean that there is a psychologically privileged level of categorization? What level does Rosch think that is and why?

Categories can be organized into levels or hierarchies. Global level - animal Basic level - fish Specific level - trout There is a level of categories that is more special to us than the others - the basic level. Rosch think the basic level is probably the most psychologically important because going above it( to global) results in a large loss of information and going below it(to specific) results in little gain of information.

In the Rogers & McClelland (2003, 2004) connectionist model, how does the activity of the hidden units change over the course of learning trials?

Different hidden units are activated over the course of the learning trials Concepts on connectionist network are represented by a distributed pattern of activation across the hidden units, that hidden units will increase the reaction time if we use or see that object more often and reverse.

What improvements do the Collins and Loftus model make to the Collins and Quillian model? What are the problems with the Collins and Loftus model?

Longer and shorter links based on prototypicality - explains prototypicality effects. Problems: No hierarchical structure,Abandoned hierarchical structure - can explain why we're faster verifying "A dog is animal" than "A dog is a mammal" Problem - can explain any result of categorization experiments -can't falsify this model (Collins and Quillian)

What does it mean that the boundaries of categories are fuzzy?

The problem with definitions is that not all members of a category have the same features

What was the Prototypes CogLab trying to test? What were the dependent and independent variables and what are the predictions? why did we categorize the prototypes faster than the new variants?

This experiment investigates whether we use prototypes to categorize concepts. Independent variable Which pattern you saw in the test phase - the prototype or the new variant. Dependent variable RT at classifying the new variants vs. prototypes Prediction If we store prototypes in memory, then we should be faster at classifying the never-before-seen prototypes than at classifying the never-before-seen new variants. We categorize the prototypes faster than the new variants because we have store the prototypes in our head so we can define them faster.

What is prototypicality, and how has it been shown to affect behavior?

Variations within a category, High prototypical items share more features in common with other members of the category than low prototypical items. -protoypicality affect by priming -supports the idea that we have stored prototypes in memory.

What is the definitional approach to categorization, and why is it unlikely that we categorize in this way?

We may decide whether something is a member of a category by determining whether a particular object meets the definition of the category. We are unlikely to categorize in this way because a lot of things in the same categorize doesn't always have a particular object or feature. The problem with definitions is that not all members of a category have the same features

What do the weights of the connections represent in a connectionist model? How could you "damage" a connectionist network to study how human behavior changes with brain damage?

Weights of the connections represent in a connectionist model as how signals sent from one unit either increase or decrease the activity of the next unit. If the connection weight between the various units in the network. -Patients who can identify animate but not inanimate objects, and vice versa. -Dyslexia if we damage one of the units or neurons then we will have some sort of brain damage where we can't do the normal thing like identifying if a item is what, however it doesn't mess up everything because it has to go to many units so it will only be partially damage.

what is a prototype? how does people use prototypes to categorize objects? what is the prototype approach to categorization?

a typical member of a category Prototypes are not actual members of a category - they are an average based on the members of a category. Membership in a category is determined by comparing the object to a prototype that represents the category.Efficient because you don't need to compare an object to every object in that category

What is cognitive economy and spreading activation in the Collins and Quillian model?

cognitive economy - shared properties are stored at the highest possible level Spreading activation is activity that spreads out along any link that is connected to an activated node.

What is Collin and Quillian's semantic network model? how does it work? what are its feature and predictions? does these predictions hold true? what are the problems with this model?

is a system that is used to graphically depict how concepts are organized in the brain using interconnected nodes. Features: nodes - categories or concepts links - connects nodes properties - features hierarchical - concepts are ranked - they go from more general to more specific( like canary on the bottom node, bird in the middle node, and animal is on the top node) cognitive economy - shared properties are stored at the highest possible level(animal node) Predication: -The amount of time it takes to retrieve info about a concept should be determined by the distance traveled through network. -Activation will spread throughout concepts that are linked. Spreading activation is activity that spreads out along any link that is connected to an activated node. The predication does hold true. The problems with this model is that Hierarchy, Can't explain prototypicality effects,

How does a connectionist model work? how are they similar to the brain?

is an approach which hopes to explain human cognition using artificial neural networks. Each object has it's own unique neural "chorus" of firing. These patterns of neural firing are the neural code for a particular stimulus. This is the idea of distributed coding. Similar to the brain: -Each object has it's own unique neural "chorus" of firing. -These patterns of neural firing are the neural code for a particular stimulus. -This is the idea of distributed coding. -They have neuron-like units. -They have synaptic-like connections between nodes that communicate excitatory and inhibitory messages. -Like neural synapses, the network is plastic and the weights of the connections get adjusted through learning.

What is the exemplar approach to categorization? how is it different from the prototype approach? what is the difference between prototypes and exemplars? what are the advantages of using exemplars instead of prototypes to categorize?

like the prototype approach, involves determining whether an object is similar to a standard object. But the standard involves many examples (or exemplars) that are actual members of the category.The advantages of using exemplars are exemplar approach to categorization can explain our behavior just like the prototype approach, and it can account for being able to categorize an object that is atypical.