Natural Language Generation and Narrative Variation in Interactive Fiction

Interactive fiction can be understood and improved using concepts from narratology. Particularly useful is the idea that the discourse, or expression plane, can be considered separately from the underlying story, or content plane. While this sort of correspondence suggests many ways to improve IF and to achieve narrative variation, IF systems have yet to incorporate this distinction. An architecture that is based on this distinction, and that abstracts the simulated from the narrated, is presented. A preliminary system, implemented based on this architecture, is then described. Examples of some sorts of narrative variation this system can accomplish are provided. The Interactive Fiction Situation Interactive fiction (IF) was most prominent in the late 1970s and 1980s, when Zork I-III, Planetfall, The Hitchhiker’s Guide to the Galaxy, The Hobbit, and other best-selling programs were published by Infocom, Level 9, Melbourne House, Magnetic Scrolls, and other companies. But the development of IF works (also called “text adventures” or “text games”) continued after the commercial market for IF subsided. Recent IF works have been innovative in theme, in the texture of their output, and conceptually. This project’s ultimate goal is to further advance the state of the art of interactive fiction, allowing authors to create more aesthetically engaging work, to present striking new perspectives to interactors via simulation and narration. To understand this project’s focus on interactive fiction, it is useful to consider some of the particular qualities of IF. Characterizing Interactive Fiction The term “interactive fiction” is not meant to cover all sorts of digital literary art; it has a specific meaning. In the more than 30 years since Will Crowther and Don Woods created the canonical version of Adventure, interactive fiction has developed as a genre. It now has wellestablished conventions, its own flexible, powerful development platforms, and its own community of practice. IF also has important formal qualities. A textbased work of interactive fiction, also called a game, is a computer program that accepts textual input and generates textual output. Accepting textual input distinguishes IF from “Choose Your Own Adventure” books and most sorts of digital literary works, although there are other programs, such as chatterbots, that read what the user has written. A work of IF is additionally a potential narrative — a system that can generate several different narratives depending upon what user input is provided. A work of IF simulates a world, which is the basis for what it narrates. The world model is one feature that distinguishes IF from the typical chatterbot. Within this simulated world, the interactor influences what happens by commanding a character, called the player character. Finally, almost all IF can be better understood if it is considered as either a game that can be played (with some outcomes being better than others) or, perhaps more usefully, as a riddle that can be solved, something which the user will have to expend effort to figure out (Montfort 2003b). Interaction in IF is not strictly task-based, but it is also not simply exploratory: the user is seeking to understand the nature of the simulated world, the place of the player character in that world, and what must be done to reach a satisfactory conclusion. IF as Language Generator and Dialogue System An IF program can be conceptualized as a natural language generation system and as a dialogue system. Modern-day IF systems such as Inform and TADS generate language simply by printing orthographic strings when certain events are simulated in the world or when objects need to be described. Although this type of narration accomplishes the basic communicative purpose, it makes many sorts of narrative variation difficult. For instance, there is no capability for re-ordering events so that they are narrated in a sequence that is different from the one in which they occurred, and it is difficult to even select the order in which objects will be listed and described. Considering IF as a dialogue system highlights some of the important ways in which IF narration differs from that of a story-generator. An IF system will narrate particular events and describe certain objects in a reply to the user, a reply which exists in some discourse context. In producing each reply, the system must take into account its own past replies and the user’s commands, and it must update the discourse state so the interaction can continue. None of this was required in story-generation systems such as Tale-Spin (Meehan 1976) or the many systems of this sort that followed. Additionally, story-generation systems have typically made decisions exclusively about the story world itself — what existents are in the story and what events happen — rather than making decisions about how to tell these. The one intriguing exception is a system, in development, that focuses on narrative levels (Lönnecker 2005). An IF narrator, distinguished from the simulation or world model, would not be able to choose anything about what happens, only how it is related. Insights from Narratology The underlying architecture and the preliminary system that has been developed are based on several ideas from narratology, a field of study that developed in the late 1960s and early 1970s and which has resulted in a substantial body of formal and general theory about stories and how they are told (Prince 2003; Bal 1997; Prince 1982; Genette 1980; Chatman 1978), some of which has even been modeled computationally (Meister 2003). While a piece of interactive fiction is a computer program, not a narrative, some of what an IF program does is usefully considered as narration, and analogies can be drawn between what is computationally modeled in a work of IF and what narratology has identified as the story level, underlying the discourse or expression.