One of the major ways in which we deal with the bewildering array of information in the world is to form categories. Our language mirrors these categories‑the words grandfather, data, bird, psychology, red, dog, and man each represent a category meaningful to most of us. Concepts are the mental structures by which we represent these categories. Particular objects or events are grouped together based on perceived similarities; those that "fit" the category are examples or instances of the concept; those that do not are non examples. The similar features across examples of a concept (e.g., all oceans contain water and are large) are called attributes; features essential to defining the concept are called defining attributes. Learning concepts involves discovering the defining attributes along with discovering the rule or rules that relate the attributes to one another.
The work of Bruner and others has shown that individuals typically solve concept identification problems by trying to discover the rules relating the concept attributes. In general, concepts that have more difficult rules are more difficult to learn. The simplest rules involve affirmation (e.g., any green object) and negation (e.g., any object that is not green), which apply if there is only one attribute being considered. Most concepts, however, involve more than one relevant attribute and hence more complex rules. Among the most common are conjunctive rules, in which two or more attributes must be present (e.g., any triangle that is green), and disjunctive rules, in which an object is an example of a concept if it has one or the other attribute (e.g., either a triangle OR a green object).
In recent years, Bourne's work has represented the clearest statement of rule‑governed conceptual structure. In his view, concepts are differentiated from one another on the basis of rules such as the above. These rules provide the means for classifying new instances as either linked to a concept or not. According to Bourne, membership in a conceptual class (e.g., grandfathers, data, and birds) is determined by applying a set of rules. These rules can be learned either through instruction or through experience with instances that either are members of the class (positive instances) or are not (negative instances). Thus, one learns to classify a set of animals as birds or nor birds on the basis of instruction or experience that leads to acquiring rules for combining characteristic attributes of birds (e.g., wings, bills, feathers). Instruction, according to Bourne, should involve presentation of both positive and negative instances (e.g., for birds, pigeons versus bats) so that critical attributes clearly can be linked to the concept. Presumably, use of these rules unambiguously classifies a new instance as either a bird or nonbird. Note, however, that this classification is a very simple one‑a new instance either is a bird or is something else, a nonbird!
Although a rule‑based conceptual system works to organize information for many concepts, it is inadequate for others. Most natural or "real‑world" concepts are more "fuzzy" and differ qualitatively from those studied in the laboratory. For instance, consider the concept furniture. Our past experience would let all of us quickly agree that furniture includes tables, chairs, sofas, and floor lamps. Furthermore, we can describe many rules that differentiate articles of furniture from other objects. But some of our attempts at rule formation quickly run into trouble. Presence of legs? But what about some floor lamps? A seating surface? But what about tables or a desk? Is rug furniture? Some would say that it is, but would wish to include a Qualifying statement or "hedge"‑it is like furniture, but not exactly like it. What is the set of rules that unambiguously determine which objects are members of the concept class furniture? Logical efforts to determine such sets of rules mostly have been unsuccessful, especially with ambiguous examples such as Rug. Rosch and Mervi, dissatisfied both with the artificiality of laboratory work on concept formation and with the difficulties of classifying concepts with rule‑governed approaches, proposed an alternative view based on‑degree of family resemblance" to a highly typical instance of the concept, a prototype.