| Abstract | Ever since the advent of computers there have been research efforts
to emulate computationally the way in which people can create stories. Artificial
intelligence (AI) and the myriad of specific technologies it brought with it gave rise to
more elaborate efforts. Yet the task seems to be a challenging frontier that even today
still defeats efforts to achieve performance comparable to that of humans. In part, the
difficulty lies in the fact that the approaches computers have been following to date
differ significantly from those applied by humans on a number of points. Whereas
computers generally apply a single specific AI technique to these problems, humans
seem to combine several cognitive abilities into the task. Computers focus on the
short end of narrative production (fairy tales and short stories) but humans produce
in a much wider range, with long examples of much greater complexity (novels, plays,
films) being very frequent. Where computers usually apply a one-pass algorithm that
produces a single output or focus on generating small fragments of an interactive
piece in co-operation with a human, the basic mode of operation for human authors
is to work alone and to proceed by iteratively revising a draft based on critical
appraisal of whether it meets very demanding criteria. The present chapter reviews
the existing approaches to computational generation of (emulations of) narrative
under the light of these differences in procedure. To inform this review, some of the
existing models of how humans address the corresponding tasks are described and
the existing computational solutions are examined for similarities and differences
with these models of human performance. The impact of these various aspects on
the quality of computer generated artifacts is considered and an analysis of current
trends for future development is made, with particular emphasis on how addressing
the aspects that have so far been less studied may help change the landscape of the
field.
|