It is unreasonable to allow any character in the story to match to a current pronoun; in longer stories, this will quickly become too inaccurate of a heuristic. As a result, the task actually considers only the characters in the story which are within the current scope of the pronoun. A scope is determined by certain structural elements of the story. For example, in a novel, one does not wish to make pronoun references across chapter boundaries; therefore, a new set of characters is maintained for each chapter in order to do the reference. In the short stories I am dealing with, chapters do not exist; instead, I utilize a smaller scope--the paragraph. It might be the case that the pronoun is a pure forward referencing one. So, if no characters exist which can be referenced to the pronoun, the supertask will generate an anticipation that one should be seen in the upcoming text.
The non-word portions of the sentence also carry information concerning the proper interpretation of the text. For instance, the punctuation contained in the sentence is a source of a great deal of information; as a result, another task of the sentence processing supertask is that of punctuation analysis. Consider Lycanthrope again--the first sentence of the story ends with an exclamation point and is bounded by a set of double quotations. Both of these punctuation markers carry important information to the reader. Later, story structure comprehension will be able to make use of the fact that quotation marks exist to anticipate that a conversation is occurring between characters. When the punctuation analysis task recognizes the quotation marks, an anticipation will be generated that a speaker and a recipient should be discovered in the text. The story structure comprehension supertask uses this information when characters are discovered. If the characters it discovers can act as the speaker and the recipient, then the anticipation is fulfilled. Similarly, the exclamation point will be used by the scenario comprehension supertask to indicate a level of excitement should be attached to the speaker of this first sentence. Again, an anticipation is generated which is used by the other supertask; specifically, the task of agent modeling will use the anticipation to modify its view of the agent which is discovered to be the speaker of the utterance.