From MEXICA to MEXICA-Impro: The Evolution of a Computer Model for Plot Generation - Computational Creativity Research: Towards Creative Machines - page 267

the model are connected. These relationships, together with the use of emotional
links and tension between characters to represent knowledge structures, give the
MEXICA project its singularity.
One of the most intriguing experiences associated to the development of computer
models like the one described in this document is what I refer to as the unforeseen
dynamics of the system . Characteristics that become visible only during the imple-
mentation of the model and that, at the least, force you to reflect on aspects that were
not considered during the design of the model but play essential roles in it.
An example is the instantiation of characters. The selection of the number of
characters participating in a tale and the decision about how they conduct in the
story-world affect the coherence, interestingness and novelty of a plot. For instance,
some experiments have shown that if during plot generation the system starts to incor-
porate more and more characters the narrative becomes incoherent. Here is another
interesting example. One of my favorite plots generated by MEXICA describes how
Princess is kidnapped by Enemy and then she is rescued by Eagle Knight. As a
consequence she falls in love with the knight and then, surprisingly, she decides to
kill him. This plot is interesting because none of the Previous Stories describes a
situation where a character kills her lover; this action provides its singularity to the
tale. It emerges because of the way MEXICA instantiates characters. The following
lines explain it. Sometimes, during plot generation, MEXICA retrieves actions that
are not fully instantiated (an explanation of how this situation arises can be found
in Pérez y Pérez [ 10 ]). The agent has some routines to sort out this problem. One of
them, inspired by the work on improvisation by Johnstone [ 11 ], instructs the system
to reincorporate characters. So, if a not instantiated action is retrieved from memory,
the agent attempts to instantiate it using one of the characters that already exists in the
narrative. In this example, during engagement, after the princess falls in love with
the knight, the agent gets as the next action to perform “Princess kills Someone”
(the variable Someone represents an undefined character); the system selects who
will participate in the action by reincorporating the only character available in the
tale: the knight. This is how Princess ends killing Jaguar Knight. Notice that if the
system had introduced a new character to be murdered, the story would have taken
a completely different twist and probably would not have been as interesting as it is.
There are thousands of pages written about emotions; however, it is hard to find
works that focus on studying the role of emotions as a way to link ideas during the
creative process (c.f. Gelernter [ 12 ]). The MEXICA project offers a computational
version of this activity that illustrates the potential and flexibility of such represen-
tation. For instance, a sequence of actions can be converted into one Contextual-
Structure. In this way, the events that occurred previously in the tale influence what
happen next, a fundamental requirement in a model of creativity. It is possible to
associate to one Contextual-Structure several possible next actions to be performed
that make sense but drive a story into completely different directions.