Semantic Memory

Declarative knowledge, many facts, and other information clustered in categories (class of information that belong together). A category is formed from abstract knowledge named concept. A concept is for example our mental representation of a dog.
Our semantic memory is often thought of as a list of features: fur, four-legs, pricked or droopy ears, sharp white teeth, braks, fetch for bones, etc. Check the list and see if it fits and you use the Feature Comparison model.

If we use ideal features of that four-legged, furry, little (or larger) animal that barks (but we don't have in mind one specific one of these) then we use the Prototype approach of a Dog. If instead we think of Taylor our family little white one going nuts when the waste management truck comes around, barks at the Mail person because he/she takes something out off our mail box, and hates the School bus because that is a monster that swallows the kids in the morning and spits them out in the afternoon (hopefully kids are a little smarter by then), then we use the Exemplar approach.

There are three big levels that usually prototypes are categorized.
Taylor, our little JRT family dog is just an exemplar at the subordinate level.
When we talk in general about a dog, that concept is at the basic level.
When we think of a dog as being a animal, we talk at the superordinate level.

Experts such as judges at the JRTCA and AKC trials use subordinate levels of thinking.
The judge is at the subordinate level since she/he has detailed knowledge about each breed and has such a refined semantic /conceptual memory that is able to separate the perfect prototypical dog from the many. As compared to me who is a novice; my knowledge is mostly at basic level and I also (like the judge) have some superordinate level knowledge that helps me separate the different breeds, and the different types of dogs. But I don't have a too developed subordinate level knowledge. So, the expert has a large knowledge at ALL three levels as opposed to a novice that has mostly basic level knowledge.
That means (if we look at the next model: Network model) that the judge will have a large and dense web of knowledge, compared to mine which will be "lighter" not as many connections, not as much detail.

Collins and Loftus (1975) developed the Network model that proposes that semantic memory is a structure similar to a net. In the nodes are the concepts that are linked, and all that forms a network. When one node is activated the activation will spread through the links to other nodes.
If certain specific links are more often used the activation will be faster through those links. (Like the "use it or loose it" process I explain in class, the more use, the "shinier" the path, the faster to slide on it).

According to ACT (Adaptive Control of Thought) we think in propositions and try to make sense of our declarative memory (factual memory that responds to the question "What?"). Propositions form a network also. There are links between the propositions and practice will increase our ability to make the links between the propositions.

PDP (Parallel Distributed Processing ,we learned in perceptual processing about this) another network-like model where the nods are like little neurons/units, that is why is called connectionist model. Activation of one little nod/neuron-like unit will activate in parallel many other linked units. From here the name of parallel distribution.

Retain that there are different approaches and some explain best certain processes while other ones explain best different other processes. We use what best fits for the needs of the process our brain is involved at that moment.

The beauty of semantic memory is that is factual, language, and concepts. All the "dictionary" we have in our mind. There are many useful ways to develop our semantic memory. One thing is very important that we expand it. As you know there is the Network model, in earlier chapters we learned about the connectionist model, and PDP.

Once a concept (or a label -- such as "cat") is activated, that information will connect to many other ones stored in our LTM (all the features and details about how a cat looks and behaves, what eats and what is not). Now, at times too much of those connections will give us the bottle neck effect, since we know that LTM is very large capacity, but the WM is only 7+/-2. So if there are too many links activated then they get stuck and we cannot retrieve a certain detail.

Also when we talk about retrieval of information from LTM we have two kinds:
- Recognition (such as when we are asked to respond multiple choice questions), and

- Recall ( such as when we are asked to respond a question, or we have to write a short essay on a given topic).

In this process of course we do lots of top-down processing.

Sometimes we cluster the information that is linked to a concept and we develop an abstract description; a prototype. When we encounter a new information with a quick top-down processing we use that prototype (which at times is a stereotype-- often a prototype with a negative connotation ). This use of information in a block (using the stereotype) gives space to bias in judgment. We will learn more in the next part when we learn about schema and scripts.