Page 91 - AC/E Digital Culture Annual Report

The most popular AI technique used to generate new stories has been AI planning. This is a technique that constructs plans to achieve a desired conclusion based on a given starting point. The path the plan takes is built as a sequence of linked actions that lead from the initial set up to the desired outcome. When
the actions in this sequence are interpreted as events and the sequence as a narrative thread, these plans constitute a good approximation of what is expected of a story. By being solutions to a planning problem, all the events in the resulting narrative are, by construction, linked by cause and effect, which provides coherence and thematic unity.
Plans obtained in this way are very lineal and have little variation, allowing only for the cre- ation of very simple and monotonous stories. Because of this, these solutions have been more popular in the field of interactive narration, where a story is generated in steps, offering the user to intervene after each step is taken. The planning-based solution allows the system to re- act when the user does something unexpected, going back to the drawing board if the user’s unexpected event threatens to derail the story, in order to absorb the event and continue driving the story towards the same conclusion.
The structure of narratives was studied as
a finished product (in narratology), as were the cognitive processes involved in writing in general.
The other obvious option for constructing story generators has been to take advantage of knowl- edge gathered from centuries of literary studies on narrative. Of the many narrative theories developed in the humanities, the morphology
of the Russian folktale developed by Vladímir Propp is the most widely recognized. Propp analyzed a hundred stories from the anthology of Afanásiev’s folktales and proposed an official common framework that explained how they were constructed. This framework is based
on the adventures of a hero who confronts a
villain to resolve the initial dilemma and go on to triumph. Propp’s formula is simple and clear, offering as it does a basic framework for con- structing stories, which has been used by various AI systems to generate stories that are both sequenced and interactive.
Turning something we have seen into a story involves composing a narrative based on a given number of events. This is a fundamental task
for the human brain. It’s also the type of basic narrative that people use in their every day lives to communicate with each other, convince, inform, recall the past, interpret the present
and plan the future. Curiously, this task has only been subject to investigation recently. In the past, the structure of narratives was studied as
a finished product (in narratology), as were the cognitive processes involved in writing in general and the understanding of narrative in particular (in cognitive science) and the generation of text (in AI).
The task of telling stories involves all these ingredients, combining them in a complex mech- anism that is now starting to be studied in detail. Current efforts are still focused on very specific aspects of the task, such as the mechanisms involved in generating suspense in the reader or in specific cases of little complexity, such as the narration of a baseball game.
Examples of current projects: computational creativity
as a means of exploring human creativity
This section describes a series of the author’s personal projects, which illustrate the proce- dures that are being investigated in the area of the computational model of literary creation.
It is important to understand that we are in no way presuming that any one of these processes operating in isolation could produce results comparable to the endeavors of human authors.
AC/E DIGITAL CULTURE ANNUAL REPORT 2018
91
Digital Trends in Culture