story generation

This thesis explores a niche field of Computer Science called Interactive Fiction, a field that utilizes the conventions of a regular story to offer multiple variations on how the story plays out. Our goal is to explore the possibility of developing a game that can generate a story file during game play that not only reads like a short story but reflects the events that transpire during a given game play. During development, we have determined that keeping track of various "states", we can simulate a narrative based on actions that transpire in the game.

We developed the game using a language called Inform 7. Inform 7 is a language developed for Interactive Fiction. It contains classes with functionality similar to real-life objects from a narrative stand-point and provides a system of rules that can be edited to simulate real-life actions and events. The language also bases its syntax on English and is thus easy to read and understand.

Using Electricity is a series of computer generated books, meant to reward reading in conventional and unconventional ways. The series title takes a line from the computer generated poem “A House of Dust,” developed by Alison Knowles with James Tenney in 1967. This work, a FORTRAN computer program and a significant early generator of poetic text, combines different lines to produce descriptions of houses. The series is edited by Nick Montfort.

This paper is a comparative reading of two works of generative literature: Scott Rettberg's Frequency Poetry Generator and J.R. Carpenter's Excerpts from the Chronicles of Pookie & JR from a structuralist perspective.
Viktor Shklovsky described the effect of literature in his 1988 article "Art as Technique", in which he describes the difference between practical and poetic language. The essence of poetic text, according to Shklovsky, is its process of "defamiliarization": The reader will see his/her familiar world in a different light due to poetic rather that practical descriptions. In generative poetry, however, the defamiliarizing effect does not stop there. Not only does one see the world differently, but the way one sees poetry itself is defamiliarized. This defamiliarizing effect does not mean that there are no rules. The formal elements of the text guide the reader, as Culler describes in his article "Literary Competence".
The aim of my paper is dual. First of all, I use Shklovky's author- and text-focused approach combined with Culler's reader-focused approach to gain insight into how generative texts build upon the readers' 'literary competence', their familiarity with 'conventional' literature, in order to understand the defamiliarizing effect of generative literature. Second, I argue that my specific analysis of generative poetry in turn gives insight in what readers expect from a text, thus helping to define the often implicit literary competence readers possess.
The output of Frequency Poetry Generator shows poems that are explicitly recognizable as poetry, thus guiding the readers' interpretation. Excerpts from the Chronicles of Pookie & JR on the other hand, situates the full text as a chronicle, implying the passing of time, and the individual texts as "excerpts", implying there might be more to the story that is not included in these excerpts, making the text into a serial narrative.
I analyze both code as well as output of these works in my analysis, utilizing Marino's framework of Critical Code Studies. The analysis of code is an integral part of the understanding of the work as it positions how the work is portrayed building on different conventional genres as well as the knowledge that one is reading a generative work of literature. Even if the reader chooses not to look at the source code, it is the potential of all possible texts that defines how the text is read as a work within the genre of generative literature. The code shows this structure, all sentences together make the potential of all different texts explicit.
A significant characteristic of generative literature is the fact that it will be a different text each reading. As I cannot analyze every single output, I invite readers of my paper, which is originally written in Scalar, to submit a reader experience of the output based on the structuralist method that I outline. This way, I offer a new type of criticism, which uses the affordances of the born digital paper to crowd source reading experiences that can be combined to specify the theory of "literary competence" further.

(Source: Author's Abstract)

In the field of artificial intelligence (AI) the automated generation of stories has been a subject of research for over fifty years. The underlying concept of "story" in story generation research is functional and does not imply any aesthetic notion. This is important because the evaluation of generated stories does not necessarily use the criterion of a readable and appealing text. Research on storytelling systems (computational systems capable of telling a story) initially arose as a part of the general trend in AI to build computational solutions that could undertake tasks that are easy for humans and difficult for machines. Some such efforts, such as computer vision and speech processing, have achieved success and given rise to commercial applications, whereas others, such as natural language understanding and story generation, still remain at the exploratory research stage.

(Source: Author's introduction)

Sina Queyras interviews J. R. Carpenter about two prose poems, "I've Died and Gone to Devon" and "A Turn for the Cold".

Most studies of digital media focus on elements familiar from traditional media. For example, studies of digital literature generally focus on surface text and audience experience. Interaction is considered only from the audience's perspective. This study argues that such approaches fail to interpret the element that defines digital media -- computational processes. An alternative is proposed here, focused on interpreting the internal operations of works. It is hoped that this will become a complement to (rather than replacement for) previous approaches. The examples considered include both processes developed as general practices and those of specific works. A detailed survey of story generation begins with James Meehan's Tale-Spin, interpreted through "possible worlds" theories of fiction (especially as employed by digital media theorists such as Marie-Laure Ryan). Previous interpretations missed important elements of Tale-Spin's fiction that are not visible in its output. Other story generation systems discussed include Minstrel, Universe, Brutus, and Terminal Time.These reveal the inevitably authored nature of simulations of human behavior. Further, the persistently anthropomorphizing approach to computational processes present in traditional artificial intelligence (and many critiques) is contrasted with authorship. Also discussed is Christopher Strachey's love letter generator for the Manchester Mark I -- likely the first work of digital literature, and arguably the first digital art of any kind. As with Tale-Spin, an interpretation of its processes offers more than output-focused approaches. In addition, this study considers works with algorithmic processes carried out by authors and audiences (rather than within the works) created by Raymond Queneau, Tristan Tzara, and Claude Shannon. Prior theoretical concepts are engaged, including Espen Aarseth's "cybertext," Michael Mateas's "expressive AI," and Chris Crawford's "process intensity." A set of concepts and vocabulary are proposed, beginning with the simple distinction between "surface," "data," and "process." Further chapters introduce the terms "implemented processes," "abstract processes," and "works of process." The most unfamiliar new term, "operational logics," names behavioral elements of systems that can be as elemental as gravity or as high-level as a quest structure. The computer game Fableembodies the strengths and weaknesses of using the same logics to drive graphical and linguistic behavior.

(Source: LABS: Leonardon ABstract Service)

What matters in understanding digital media? Is looking at the external appearance and audience experience of software enough—or should we look further? In Expressive Processing, Noah Wardrip-Fruin argues that understanding what goes on beneath the surface, the computational processes that make digital media function, is essential.Wardrip-Fruin suggests that it is the authors and artists with knowledge of these processes who will use the expressive potential of computation to define the future of fiction and games. He also explores how computational processes themselves express meanings through distinctive designs, histories, and intellectual kinships that may not be visible to audiences.Wardrip-Fruin looks at "expressive processing" by examining specific works of digital media ranging from the simulated therapist Eliza and the first major story-generation system Tale-Spin to the complex city-planning game SimCity. Digital media, he contends, offer particularly intelligible examples of things we need to understand about software in general; if we understand, for instance, the capabilities and histories of artificial intelligence techniques in the context of a computer game, we can use that understanding to judge the use of similar techniques in such higher-stakes social contexts as surveillance.

(Source: MIT Press)

The intelligence of a story-generating computer program can be assessed in terms of creativity, aesthetic awareness, and understanding. The following approaches are evaluated with respect to these three criteria: simple transition networks, grammar-driven models, simulations, algorithms based on problem-solving techniques, and algorithms driven by so-called "authorial goals." The most serious deficiency of the discussed programs resides in the domain of aesthetic awareness. In order to improve on this situation, story-generation should not follow a strictly linear, chronological order, but rather proceed from the middle outwards, starting with the episodes that bear the focus of interest. The program should select as top-evel goal the creation of climactic situations, create the preparatory events through backward logic, and take the story to the next highlight, or to an appropriate conclusion through a guided simulation. This strategy is ilustrated in a "reverse-engineering," or generative reading of Little Red Riding Hood that simulates the reasoning of an imaginary computer program.

(Source: Author's website)