Production rules are a primary component of many contemporary computer models of
cognition (e.g., ACT, GPS, Soar). A production has the form: If
Flow of control in a production system goes through the set of productions sequentially until a condition is matched. After executing the action, the system continues with the next production or returns to the beginning of the set. This sequence is repeated until a terminal goal condition is satisfied. Thus, production systems require no executive level of control; all control is determined by the productions. Clearly, order of productions in the set is important since it determines which actions are satisfied first.
It is possible to add constraints to productions that alter the strict sequential order and hence introduce some form of higher level control. For example, preference can be given to conditions according to recency or frequency of occurence. Productions can be limited to firing only once for a given condition (rule of refractoriness). Or, goal symbols can be added to the conditions that must be satisfied in order for the production to be satisfied.
Productions map very closely onto the notion of rules found in many cognitive theories and hence are a natural representation to use when building computer models of such theories. They also resemble the S-R associations of behavioral theories, except that production rules do not normally encompass any notion of strength; they are all or none. However, some theorists have allowed individual production rules to have probabilities of executing based upon frequency of use or characteristics of the conditions.
Klahr, D., Langley, P. & Neches, R. (1987). Production System Models of Learning and Development. Cambridge, MA: MIT Press.
|Contemporary Theories of Learning: Learning Theorists in Their Own Words|