Reflection

  The control supertask fills one final role. It is often necessary for a reader to reason about their actions during the reading process; this task is known as reflection. Remember the last mystery story you read? When did you realize who the killer was? What clues tipped you off? Did your opinion of the person change as a result; if so, why did it change? These are all the kinds of questions which the metareasoning task will enable the reader to answer. Metareasoning results from keeping explicit representations of the reasoning processes occurring during the cognitive activity. These reasoning traces can then be reasoned about, much like any other piece of knowledge within a cognitive agent ([#!learn:cox1!#]). This functionality enables a reader to reflect on their own reasoning, thus enabling them to learn more from the experience. For example, if the reader discovers that they were ``led down the garden path'' at some point in the story, the meta-analysis will enable the reader to learn this and therefore (potentially) avoid a similar problem in the future.

The reasoning trace is a list of conceptual nodes representing the history of the reader's actions which were undertaken to arrive at a comprehension of the story. In particular, the specific actions which were controlled by the control supertask are maintained in this reasoning trace. Another possibility is that all actions performed by the reader could be recorded. However, the reasoning trace represents the level of behavior which the reader can actively reflect on; the control supertask and the actions it controls best captures this level of description. Future work and experimentation with the model and with human performance data should be able to reveal how accurate this claim is.

It is also possible to view reflection as an ends as well as a means as in the previous examples. For example, the enjoyment in reading most mystery novels is not derived from the conclusion; otherwise, the author could simply inform the reader upfront as to ``who-done-it.'' Rather, the enjoyment comes from the twists and turns that the reader is forced to endure during the process. The final discovery of the true criminal results in a sense of supreme satisfaction, as one can look back over their reasoning patterns and see exactly how the conclusion was achieved.

    Part of the role of reflection is also to determine whether the reader is believing the story being presented to them, in the sense that it should be coherent. Suspension of disbelief ([#!general:coleridge1!#]) enables a reader to accept, at least temporarily, a text which violates their world view. This last function is particularly important in the case of reading a story containing unfamiliar concepts. The normal approach taken in a number of AI models is to allow the background knowledge of the reasoner to act as a check against noisy input--if the reasoner is presented with something which is a violation of known facts, it can be discarded, or there can be different sections of facts which are held to different levels of ``belief.'' This cannot occur, however, in a creative reading system where the assumption is that novel concepts will be presented to the reasoner on a regular basis.   A rational reader of Lycanthrope knows that the story cannot be true. Witchcraft does not allow the transmutation of humans into any sort of were-creature, no matter how elaborate the ceremony. In order to read, comprehend, and enjoy the story, however, the reader must be willing to accept the unfamiliar ideas for the duration of the reading experience. The way in which the suspension of   disbelief is handled is by allowing disbelief of background knowledge to continue as long as the concepts being considered are providing a satisfactory comprehension of the story. Currently, the ISAAC model does not possess the ability to completely abandon a story which is not making sense; I assume all the stories given to it will eventually be comprehended.