While intimately related to issues of knowledge representation discussed previously, people have come up with some generic models of how knowledge might be organized in the mind.

The intuition exists that we process the world using concepts as the most basic division of knowledge.

Concepts are relatively discrete ideas, such as "apple", "red", or "liberty."

The first level of organization is to collect related concepts into categories.

Categories are how related concepts get linked together, such as "fruit", "color", or "dog."

The obvious question to ask is how do we decide what concept a given object belongs to?

We talked about them with regards to perception, and the ideas with regards to concepts are analogous.

Feature-based categorization: Categories are represented as lists of features, such as the concept STOP SIGN being represented as "red; octogon; at street corner" etc.

Prototype theory: Categories are represented based on "typical" features. Another way to think of this as that prototypes consist more of lists of probable features.

Prototypes with exceptions:

Normal categorization still uses prototypes, but we also maintain a list of exceptions of objects that deviate too far from the prototype. A similar concept to exemplar theory.

The Core-prototype model:

Whereas a prototype does not necessarily contain required features, the notion of a concept core is that the core contains the features that are absolutely required for category membership.

You can think of the core as containing the necessary features, and the prototype as containing the sufficient features.

One most popular alternative put forth has been the semantic network model, which posits that concepts are stored in a hierarchy.

Each concept of the hierarchy has a set of properties associated with it.

These properties are inherited by all the concepts attached below it in the network.

(See Sternberg, Figure 8.2)

Each class of features, such as SIZE or COLOR, is a dimension in a geometric space. An object is categorized by where it falls in the space.

(See Sternberg, Figure 8.3)

These ideas have also been discussed earlier, but there are two main ways that people envision concepts being related: schemas and scripts.

Schemas: To review, schemas are high level knowledge constructs that tie together multiple concepts into a larger structure, encapsulating the relationships among concepts.

Scripts: Knowledge structures that represent how we interact with the environment in various situations.

How do we represent our knowledge of how to do things?

The most common way is through the production system. The production system contains two main components:

• Memory elements containing information about the current state
• Productions that manipulate the memory elements to get to a new state.

Productions have two important parts:

• The Left-hand side (LHS): This is a set of if conditions that must be satisfied by the memory elements of the system before the production will "fire."

• The Right-hand side (RHS): This is a set of actions that will be executed when the production fires. These actions can add, delete, or modify memory elements or other productions.

(LHS)

        IF (I am driving) AND (I see a stop sign)

(RHS)

        THEN (Press on brake with right foot)

Anderson’s ACT model (c.f. Sternberg, Fig. 8.4):

Three components —

• Working memory — where information needed for current processing is maintained

• Declarative memory — Uses a network model where related concepts are connected together with varying strengths depending on the strength of the relationship. The more a connection is used, the stronger it gets. Uses spreading activation across connections.

• Production memory — where procedural knowledge is stored.

Anderson hypothesizes three stages in the proceduraliztion of productions:

1. Cognitive: We explicitly think about every step in the process.
2. Associative: We practice the steps to associate them with each other.
3. Autonomous: The procedure has essentially become an automatic process.