Control

  The high-level task structure which forms the interior of the control supertask is shown in Figure 30. The phrase packer grabs words and markers from the augmented text created by SARTrE. When a phrase boundary is discovered (such as a period, a question mark, or a comma), the packer sends the phrase to the sentence processor supertask. That supertask returns a conceptual entity which represents the meaning of the original phrase. This is then passed on to the other two primary supertasks, scenario comprehension and story structure comprehension.   In addition, the tasks of focus control and suspension of disbelief are given the conceptual entity. These two tasks manage the control of processing which the system is undertaking. They make decisions based on the current state of comprehension and then inform the other supertasks via         suggestions, anticipations, expectations, and requests. They also take in these four control messages from the other supertasks in order to manage some of the inter-supertask   communication. In addition, the interest management task acts to modify the depth of reading based on the current interest the ISAAC system has in the text being presented to it. Finally, the inspector task is responsible for keeping track of the activity of the system, including which supertasks are being called upon, what information is being discovered by the elements of the system, and what control messages are being passed. This information contributes to the meta-reasoning trace being built by ISAAC.

  

Figure 30: Control supertask breakdown

Not all of these tasks are built to the same level of sophistication. In particular, the interest management is primitive when compared to the others. Although the theory does make some claims about interest-driven comprehension, this aspect of the process was not the focus of my research. As a result, the interest management task is limited--ISAAC is primarily interested in things which it has recently been reasoning about. It is also interested in answering the questions which arise during processing a story; i.e., the control messages which are generated. In terms of the ISAAC model, then, interest represents a high-level goal which the system creates and then acts towards satisfying. The remaining tasks are implemented to a much higher degree of ability. However, note that the modular design of the system would allow a ``better'' interest management system to be substituted into the model if one was developed, in much the same fashion as COMPERE and MOORE were used in my work.