Categorical Induction: Conceptual Representation & Probabilistic Properties

Categorical Induction and Conceptual Representation

Categorical induction

Definitional Properties Focus

Definitional properties approach: An exemplar is a category member when it meets the defining properties of that category. For example, an object is considered a member of category 1A if it presents the defining properties of category 1A. Definitional approaches treat category membership as determined by a set of necessary and sufficient properties.

Probabilistic Properties

Probabilistic properties: Typicality can be measured by similarity. An exemplar's typicality is its similarity to a prototype or to other category members; similarity orders the members of a category. Categories are often not homogeneous and membership can be graded rather than strictly binary.

Exemplar Approach

Exemplar approach: Concepts can be represented structurally by collections of concrete exemplars rather than by a single abstract prototype. Membership of an exemplar in a category depends on its similarity to stored members of that category. When relational information is available, the exemplar representation preferentially retrieves exemplars relevant to the current goals or tasks.

Selective Modification

Selective modification: The noun (the main concept) serves as the anchor or frame for interpretation; adjectival selection then highlights properties that correspond to that frame. This process determines which adjectives emphasize prominent properties and assesses their diagnostic value for categorization.

Conceptual Specialization Model

Conceptual specialization model: Adjective values can specify the main noun's concept or property, but they do not automatically encode the consulting expertise of the subject. Specialization refines the concept by specifying values or ranges for salient properties without implying expert knowledge.

Substantive Combinations

Winiewski identifies three types of substantive combinations:

• Modifier as relational basis: The modifier defines a relationship that constrains the base concept (e.g., "golden retriever" — type of dog defined by relation to breed).
• Property application to base: The modifier denotes a property that applies to the basic concept (e.g., "hammerhead shark" — a shark with a specialized head property).
• Hybrid combinations: Combinations that mix relational and property-based effects, creating a hybrid concept (e.g., "apart-hotel").

Conceptual Specialization in Schemata

The first type corresponds to a representation by schemata in which the main concept takes the relation of the modifier (for example, constructions like "partridge dog"). Schemata capture how a primary concept can adopt roles or relations introduced by a modifier.

Larger Combination Types

The other two types explain how combinations occur when a property applies to the base concept or when a hybrid combination of properties emerges. These combinations create conceptual structure through comparison and constructive processes; they do not always add entirely new properties but reorganize or highlight existing ones.

Pragmatic Reasoning

Pragmatic reasoning: The system applies rules distributed across schemata that are grouped into different structures and processed in parallel. Some rules or schemata are inhibited while others are enhanced depending on the context; the system imposes constraints on processing to achieve coherent interpretation.

Distributed Activation

Distributed activation: The cognitive system operates not solely with discrete symbols or explicit rules, but with activation patterns across a network. Information is conveyed by these activation patterns; representations are dynamic rather than static. This is consistent with a PDP (Parallel Distributed Processing) perspective in which processing emerges from distributed activations rather than from isolated symbolic manipulations.