p.790 The Hearsay model [Red73Mo] has been developed for problem-solving in domains which must use large amounts of diverse, errorful, and incomplete knowledge in order to search in a large space.

p.791 The major units on the blackboard are hypotheses... A KS can create new hypotheses, specifying values for attributes of the new hypotheses... Each KS has two major components: a precondition and an action. The purpose of the precondition is to find a subset of hypotheses that are appropriate for action by the KS and to invoke the KS on that subset; the subset is called the stimulus frame of the KS instantiation.

p.794 The asynchronous, data-directed control structure used in Hearsay-II was designed to permit:

1. The quick refocusing of attention to appropriate hypotheses in the blackboard.
2. The flexible reconfiguration of the system with different sets of independent (and possibly competing) KSs, and different global control strategies.
3. The exploration of parallel processing.

p.797 The paradigm of viewing problem solving in terms of hypothesize-and-test actions distributed among distinct representations of the problem (where these representations form a hierarchy of abstractions) has been shown to be a computationally feasible approach to solving knowledge-intensive tasks. This paradigm also provides a convenient framework for structuring and applying knowledge. This has been demonstrated both by the successful application of the Hearsay-II architecture to the speech understanding task and also its adoption as an approach to problem-solving in a diverse set of other domains such as image understanding [Pra77Se], reading comprehension [Rum76To], protein-crystallographic analysis [Eng77Kn], signal understanding [Nii77Ru], and complex learning [Sol77Kn].

p.797 The overhead costs involved in implementing this type of control structure are acceptable for KSs which do moderate amounts of internal computation at each invocation... This control structure is not appropriate for domains in which the hypothesis credibility ratings are not selective enough to suggest strongly good paths to search.

p.797 The problem of focus of attention in this type of control environment, though not completely solved, is now understood well enough so that it no longer represents a major obstacle to the effective use of the architecture. The integrated representation of alternatives on the blackboard, which permits a global view of the current state of problem solution, and the data-directed control structure make it possible to quickly refocus attention to the appropriate places in the blackboard.