From MEXICA to MEXICA-Impro: The Evolution of a Computer Model for Plot Generation - Computational Creativity Research: Towards Creative Machines - page 272

axis indicates the number of emotional links. I employ coordinates of type (Number
of tensions, number of emotional links) to point out specific positions. For instance,
(4, 3) refers to the position (4 tensions, 3 emotional links). Each entrance in the map
represents the number of CK-Structures in that position.
The following lines provide a hypothetical example to illustrate how to utilise
the map. In Fig. 13.3 the blue circles represent the Contextual-Structures generated
after processing the Previous Stories. They are situated close to each other within
the Map. As explained earlier, the reason of this proximity is the alikeness in the
content of the Previous Stories; as a consequence, the Contextual-Structures have
similar characteristics and therefore they are located in the same or adjacent areas.
The similarity between narratives also provokes that some of such structures include
several possible next actions to be performed. These two characteristics provide an
important flexibility to the system during plot generation.
The green circles in Fig. 13.3 represent the new knowledge structures produced
after incorporating into the system a narrative generated by MEXICA-impro; as
a main characteristic, such a tale is similar to those in the Previous Stories. As
expected, only few new Contextual-Structures are created and they tend to be located
in populated blue areas within the K-Map. In the same way, some new possible next
actions to be performed are added to existing structures; this is represented in Fig. 13.3
by a cross on top of some blue circles.
The red circles represent the new knowledge structures generated after incorpo-
rating into the system a second narrative generated by MEXICA-impro; as a main
characteristic, such a tale is dissimilar to those in the Previous Stories. When a new
tale is unrelated to the Previous Stories several new Contextual Structures are created
and they tend to be located in isolated regions within the map. In most cases, they
have associated only one next possible action to be performed.
Intuitively, one might wish that our computer agent generates narratives as differ-
ent as possible to those in the Previous Stories. After all, one of the main character-
istics of creativity is novelty. However, that is not necessarily the best scenario in a
long term. It is true that the isolated Contextual-Structures that emerge from singular
narratives allow MEXICA-impro to register in its knowledge base unknown situa-
tions. That is, the agent is increasing its knowledge-base. Certainly, that is a good
thing to do. However, the lack of different possible next actions to be performed and
the absence of similar Contextual-Structures around it make the use of isolated struc-
tures very rigid and predictable. Some experiments have shown that the use of one
isolated structure pushes the system to employ other isolated structures producing
as a final result the reproduction of tales that already exist in the Previous Stories.
Thus, in order to improve its capacity to produce original outputs, MEXICA-impro
requires mechanisms that allow locating new structures in isolated areas within its
K-Maps and then populate them. I refer to this process as Full Expansion of Knowl-
edge . The interaction of agents with different knowledge bases seems useful for
this purpose. Such expansion brings to mind what some artists report when they
are preparing themselves to produce a new piece of art. For example, Pablo Picasso
generated several similar sketches in preparation to paint the Guernica; Gabriel Gar-
cía Márquez reports that he spent several years researching about dictators in Latin