Interfacing with Computer Narratives: Literary Possibilities for Interactive Fiction
There is literary potential in existing forms of interactive fiction, and just as interfaces are problematic for some of these forms, they work well for others, such as hypertext. It is in the unexplored domain of non-puzzle-based, hall-of-mirrors style interactive fiction, however, in which the interactor's role is one of co-author, that the necessity for a more complex communication between computer and interactor demands a close look at how best to combine qualities of existing interfaces.
The important features of a good interface, as observed in the examination of different interface methods above, include uniformity (the interactor communicates through the same medium that he or she is receiving information through) and consistency (similar actions produce similar results, and the way the computer refers to something - be it verbally or graphically - is the same as way the interactor refers to it). Although not a feature of many current interfaces, that an interface is continuous (and thus the fiction is highly interactive) is a major advantage, as this adds to the ability of the fiction to engage the interactor. Another good way to make the interface more engaging is to make it simple; the challenge in an interface where the interactor has a text-creating role is to balance this simplicity with the ability to convey meaning.
The demands of puzzle-solving interactive fictions exceed those on literary interfaces in some ways, however. Carefully directing a character to perform a series of actions requires an exact semantic interpretation of a textual command. This exactness is not necessarily a component of a literary experience of interactive fiction, particularly if the interface's role is to connect two co-authors (or two voices in a dialogue) rather than to translate a command into the action of a fictional character.
Here are a few sketches of interfaces that might facilitate this new brand of interactive fiction:
The computer screen is divided into two windows, each of which contains text that appears, character-by-character, at the bottom and scrolls upward a line at a time. In one window is a cursor; the interactor types messages to the computer character in the other window here. The two-window case described here is the simplest, but any number of computer characters could have windows.
This interface would be like the current textual interface for conversation-style interactive fiction in many ways, and it would share the advantages of this interface: uniformity, consistency, simplicity and the ability to convey large amounts of meaning. The asynchronous setup, where the computer is outputting text a character at a time and can respond as the interactor is typing (and vice versa), allows for the narrative to branch at any point. The conversationalist program Racter tells short nonsense stories that it displays all at once, with no possibility for interaction while the story was being told. If Racter had an asynchronous interface the interactor could interject questions, comments, even taunts and jeers at any point in the story - and the narrative could branch off in response. This property of being continuous makes each instant a potential branch point and engages the interactor as if he or she were conversing with an actual person.
Since the asynchronous text interface can be envisioned as a network "chat," it also can be used for computer-within-computer interactive fictions where the interface is the same for real and fictional worlds, further enhancing the way in which the narrative engages the interactor.
Here, an existing text parser would be used in a system that used either the usual method of displaying pre-written text fragments or used generative rules to produce text. In either case, however, the interactor could interrupt a block of text the computer was displaying or generating and influence the remainder of the text with some input. Current interactive fictions suffer, at points, from a lack of interruptability. In Infocom's Border Zone, for instance, although the action occurs in "real time" (as opposed to other Infocom interactive fictions in which time passes only with respect to the entry of commands), series of events transpire which the interactor cannot influence. Consider the following Border Zone text:
The train, without warning, slams on its brakes and comes to a stop at the border station. After so many trips, you'd think you were used to this jarring moment, but alas, it's always the same. Within seconds, the man in the trench coat along with two uniformed guards enter. This border search ritual has become second-nature to you, although the man in the coat is a new twist. They begin by asking about your trip, your nationality (as if they didn't know), and a host of inquiries about currency and other possible contraband. Then, as usual, they move on to the possessions. They start with the briefcase, looking it over, and finally dumping its contents onto the front facing seat. Seeing nothing of interest there, they continue with the camera bag. Seeing nothing of interest there, they continue with the camera. The guard examines it for a moment, then passes it to the trench coated man. He smiles at you, but you find it difficult to return the gesture. "You like to take pictures, then..." He pauses, massaging his upper lip between his thumb and forefinger, then looks in the back of the camera. He then notices the film you're holding, exposes that, and hands it over to one of the guards. "I do not understand why you would be holding your film. It is night, and the view... well, the view is not so exciting. We can only hope that the pictures here were not so good, eh?" "You have been through worse, I am sure." The man smiles that unnerving smile of his and continues. "And you have been very patient. Have a pleasant journey, and feel free to leave the train, if you wish. We shall be stopping here longer than usual today." He exits, along with the other guards.
This text is displayed all at once; at no point can the interactor interrupt it to change the course of action. If text scrolled onto the screen a character at a time and at a slow, user-adjustable speed, and if there were implicit "interruption points" throughout it, the sense of continuous interaction could be added to the Interlogic parser interface. Not every point need be interruptable; it would be meaningless, for instance, to interrupt the text between "He smiles at" and "you" (Note 13) - but it would be very meaningful to interrupt it after "you" and before "but you find it difficult to return the gesture"! If the interactor struck SPACE after the first three words of this sentence had scrolled onto the screen, "you" would appear and an appropriate command (such as "smile") could be entered. Even a full-fledged normal command (like "leave the room") could be allowed at an interruption point, although, of course, not all characters will let your attempts at interruption succeed - particularly not "trench coated men" in fictive Eastern-bloc countries. Even if the interruption does not achieve its aim, however, it can alter the course of the narrative, just as trying to leave after the man smiles would undoubtedly affect this narrative. For interactive fiction in which the interactor controls a character and directs his or her actions through commands, an interruptable text interface brings some of the advantages of a continuous interface to a "classic" style of interactive fiction. By allowing for more branch points, however, it also furthers the ability of this style of interactive fiction to utilize text-generation and make the experience more of an infinitely changeable process.
Merging graphics with text is difficult because the interface must cross between the two media. If the best in graphical, text-generating and text-parsing technology were combined, however, the result could be a system with stunning, engaging characteristics and much literary potential. Instead of the photorealistic rendered snapshots of Myst, consider an interactive fiction that has a fully explorable environment than can be seen from any perspective - any rendering, texture-mapping or other computation done to create a scene from an internal representation is done as the interactor moves through the environment. Technically, such a system is feasible, and in face extant: the game Doom is such an environment, although it is not a narrative environment and the character's interaction is limited to opening doors and shooting. The control of this environment could be divided into motion control and other action control. Through arrow keys, mouse, joystick or other controller the interactor could change the character's position and orientation, and the viewdisplayed on the screen. Preferably all degrees of freedom would be controlled with a combination mouse and thumbwheel device or some other single controller, which would be used only for controlling motion and perspective.
In a text window, the interactor could type commands in order to effect all other actions besides motion. The problem that arises is that objects in the environment need to be referenced by typed text, yet the "viewport" onto the character's world does not textually name these objects. One solution would be to use a controller to gesture to objects and either ask their names or simply insert their names in the text that is being typed. Preferably, to avoid problems with uniformity, this would be a different controller than is used to control the character's position and orientation. Another solution, involving fewer controllers, would be to have the interactive fiction generate a textual description of the scene in the "viewport," whenever the interactor types a certain command (e.g. "DESCRIBE"). The computer description could be generated not from the actual image, (Note 14) but from the internal representation of the objects and structures pictured. Then, the interactor could refer unambiguously to objects and direct actions to the fictional world as well as converse with characters through the text interface. The result would be a continuous interface in first-person graphical perspective and the ability to bridge the consistency gap between a graphically seen world that must be referred to textually. The Origin interactive fiction Bioforge, although it does not have a text interface and displays the interactor's character in a third-person view, is an example of software that uses this type of technology in a narrative. Bioforge follows a cinematic, rather than written, paradigm, and does not allow for full co-authorship, but this product demonstrates that the technology exists for graphical interactive that renders scenes as the interactor moves a character through an environment.
It is worth noting that this type of interactive fiction could also be a (in this case perhaps literal) "hall-of-mirrors" work, since rules for designing environments in response to the interactor's actions can be contrived. Sim City, Civilization and a wide variety of related strategy games are examples of interactive virtual environments that grow and change according to rules and input.
These interfaces are examples of ways to bring the interactor more fully into the process of creation that is essential to a new form of interactive fiction. While there is literary potential in existing interactive fiction, this new direction of open-ended narrative with continuous interaction has potential that is, as yet, entirely untapped.
Several issues that lie beyond the scope of this essay must be investigated if interactive fiction is to realize its potential and extend into the realm of open-ended, generative fiction. There is still much work to be done on deeper features of the narrative, on that which the interface connects to the interactor, as Joseph Bates of the Carnegie Mellon University OZ Project states:
[While] ... human-interface technique ... [is] ... important, we see exclusive attention to [it] as something like studying celluloid instead of cinema, paper instead of literature, or cathode ray tubes instead of television. To reach our dream of "interactive cinema," we must also look at the underlying content of the world we want to model. (4)
Although Bates may neglect the elevated importance of interface in interactive media, as opposed to the traditional narrative forms he enumerates, he is correct in noting that interface should not be the limit of investigations into making this new form literary.
Further work must be done on ways to generate plots. If the plot of a narrative is to branch arbitrarily, it will be necessary for the computer to generate new concluding segments for the plot that is being traversed. Although the success of story-generating software has been only partial, two factors suggest that there is hope for automatic plot generation in interactive fiction. First, there is a long tradition of work that suggests that traditional narrative, both oral and written, has well-defined high-level structures (Polti, Propp). Second, interactive narratives do not have to be generated tabula rosa and without any intervention from start to finish - in fact, they require human input along the way. This input could be used not only to determine points of branching but also to helping the software overcome some of the obstacles it faces in story generation. But using human input in this way will require further work on defining how exactly the input is to be used in the generative process.
Character generation also will require great additional effort. The work of artificial intelligence researchers so far has focused on categorizing character traits and generating actions based on these traits (Bates), and on developing actions that are consistent with those the character has performed before (Lebowitz). Since characters are defined in existing systems by structures that store their traits and prior actions, the groundwork for generating characters automatically has been laid. Now, systems must be developed that generate these defining structures rather than simply using the structures to develop actions.
Setting has been given the most attention by interactive fiction designers. Since characters are more difficult to simulate than environments, and branching plots are new and difficult ground, the usual way to provide interesting narrative features has been to enhance the setting of a work. If environments are to be generated in the same way character and plot could be, however, rather than just being laid down beforehand by a designer, we must explore ways of setting up rules for generating environments so as to create settings that are interesting and contribute to a narrative.
Finally, the interfaces sketched here must be described in detail and developed. Questions about each interface remain. Regarding the asynchronous text interface, is a system with more than one computer voice feasible or desirable? Could the system be used for narratives other than dialogues, and, if so, how? Regarding the interruptable text interface, how long between "interruptable points" is appropriate? Would the speed of the scrolling text need to vary at different points, or should it always be the same, and should the user or the computer determine the speed? Regarding the perspective control and text input interface, how will the slow speed of typing allow control in a continuously moving world? Since this interface envisions the interactor as controlling a character, how will the interactor's full role as co-author be realized, and how exactly will a generative system work with this interface? All of these interfaces will require attention to the problems associated with existing text interfaces, such as limited vocabulary and the inherent ambiguity of language. Additional questions, likely more difficult ones, will undoubtedly arise during the development process.
To find a new direction for interactive fiction, existing positive features like consistency, uniformity, simplicity, and the use of implicit rather than explicit constraints should be maintained. There are two critical qualities, however, that must be added to existing interfaces in order to fully exploit the narrative possibilities of the medium. First, high interactivity, also called continuousness, should be added. The interactor should be able to influence the narrative at any point, so the experience is as engaging as possible. Second, interfaces that convey semantics appropriately should be crafted. The specificity required for puzzle-solving is not necessary in an open-ended, generative system, but the inarticulate gestures of the mouse are inappropriate if the interactor is to be a co-author of sorts, rather than just an adventurer or reader.
The processes that are created by means of the three interfaces described herein add these two critical qualities. They allow interactive fiction designers to use the tactics of revealing unknown text or media, and in fact, of revealing newly generated text which will may have never existed before and may never be generated again. These interfaces also facilitate the infinite chaining of narratives. Although much remains to be done in order to develop interactive fictions along these lines, there are encouraging signals that indicate that the necessary progress is possible. When combined with open-ended, generative systems, continuous interfaces that can convey the needed semantics, like the three described here, hold the possibilities for a new literature in a new medium. End