Interfacing with Computer Narratives: Literary Possibilities for Interactive Fiction
Interactive fictions were initially entirely textual, and attempts to extend and improve the early interfaces were made within this context. All-text interactive fictions reaped great benefits from textual interfaces, although this method of interaction is not without its limitations.
Because the Zork interface, called the Interlogic parser, is archetypal of English-based interaction with interactive fictions, it is important to understand it in detail. The manual to Zork I describes characteristics of this interface:
When you play a ZORK game, you talk to ZORK in plain English, typing in all your requests on your keyboard when you see the prompt (>). (ZORK allows you to abbreviate some words, though, because you use them often.) When you have finished typing a line, press the RETURN key and ZORK will digest your request.
ZORK usually acts as if your sentence begins with "I want to...", although you should not type those words explicitly. ZORK then displays a response that tells you whether what you want to do is possible in the current situation, and if it is, whether anything interesting happens as a result. All the words you type are distinguished by their first six letters and all subsequent letters are ignored.... ZORK "understands" many different types of sentences. A sentence must contain a verb and usually an object. Some examples:
TAKE THE GOLD. DROP GOLD. PICK UP GOLD. PUT DOWN THE GOLD.
GO NORTH. NORTH. WALK NORTH. N. WALK AROUND THE HOUSE.
PUSH BUTTON. PUSH THE RED BUTTON. OPEN WOODEN DOOR....
ZORK tries to be clever about what you really mean when you don't give enough information. ... ZORK will sometimes decide that there was only one possible object you could have meant. ...
ZORK uses many more words than it "understands." ZORK's vocabulary consists of several hundred words which are all you are likely to use when conversing with ZORK. However, ZORK's responses may tend toward "purple prose" at times, and ZORK will display English descriptions that it couldn't possible analyze. ... Completely mystifying sentences will cause ZORK to complain in one way or another. (3)
Although Zork's interface, as described, clearly has limitations, it does allow the reader to participate in the interactive fiction as a literal co-author, adding lines of text that describe the action of the central character. A transcript of the beginning of one Zork interaction follows (the lines beginning with the '>' prompt have been typed by an interactor; other text is produced by Zork):
ZORK I: The Great Underground Empire Copyright (c) 1981, 1982, 1983 Infocom, Inc. All rights reserved. ZORK is a registered trademark of Infocom, Inc. Revision 88 / Serial number 840726 West of House You are standing in an open field west of a white house, with a boarded front door. There is a small mailbox here. >open the mailbox. Opening the small mailbox reveals a leaflet. >take the leaflet out of the mailbox. Taken. >read the leaflet. "WELCOME TO ZORK! ZORK is a game of adventure, danger, and low cunning. In it you will explore some of the most amazing territory ever seen by mortals. No computer should be without one!" >go north. North of House You are facing the north side of a white house. There is no door here, and all the windows are boarded up. To the north a narrow path winds through the trees. >go north. Forest Path This is a path winding through a dimly lit forest. The path heads north-south here. One particularly large tree with some low branches stands at the edge of the path. >go north. Clearing You are in a clearing, with a forest surrounding you on all sides. A path leads south. On the ground is a pile of leaves. >count the leaves. There are 69,105 leaves here.
Using text as a means of interaction has technical benefits, and, even within the constraints of Zork, literary ones as well. Typing commands to a text-only interactive fiction is uniform and consistent, whether the interactor is issuing commands to manipulate objects, look at the landscape or converse with other characters. The rich semantics of text allow for great specificity in what exactly is to be done and how a certain action is to be performed. Finally, the action of typing text - producing language - is engaging.
The uniformity that a textual interface provides is important, although not unique. Any decent interface that does not rely on a different medium of communication and representation will share these features. A textual interface is uniform because throughout the course of interaction the interactor uses the same method of interacting; it is consistent because the interactor is responding to information in a certain medium (text) via the same medium. Consistency is important because the interactor can refer to an object or person that the interactive fiction refers to by using the same term. If the interactor saw a picture that looked like a bladed weapon and had to respond by typing, it might be only after the initial frustration of commands like "take dagger" and "take knife" that the interactor would realize that computer's name for the object was "sword." This frustration presents itself in the original King's Quest, which describes places and objects pictorially (sometimes objects are formed from just a few pixels) but accepts only typed commands for most interactions. In an all-text interactive fiction with a textual interface, the way to refer to an object (if the interface understands references to the object at all) is always evident from its original description.
Another advantage of text is that it is rich semantically, allowing the interactor not only to easily name the object of his or her action but also to specify what action is to be taken. Rather than just "using" objects interactors can eat them, burn them, hand them to other characters or take them apart. Interactors can use textual interfaces to speak with other characters, describing exactly what they wish to say rather than choosing their utterances from a list. In interactive fictions with a conversational paradigm, such as Maur and Racter, the ability of a textual interface to specify exactly which words are spoken to the other character is particularly appropriate, as the following transcript of a brief conversation with Maur illustrates:
Well, well, a visitor. It has been a long time since anyone has come down this way. You say: Who was the last? My last visitor was one of those Northerners. That must have been... certainly forty of what you call years. He actually tried to attack me... naturally I incinerated him on the spot. You say: You breathe fire? Yes... I am so strong that my flame will now melt stone. (The dragon seems rather proud of this.) You say: May I see you demonstrate this destruction of stone? Watch this. (The dragon chuckles malevolently.) (You hear a great inrush of air, and take a few steps backwards. Maur's eyes stay fixed on you, and then, before you can turn to run, he opens his mouth and exhales. A great wind buffets you, hot enough that you can feel your hair begin to crinkle and your skin begins to sting. The air smells hard and metallic. Then there is a great blast of heat and flame, flame white as lightning and hotter than any forge, enveloping you. Then an explosion of pain hits you, and you know no more...)
The interactor has the satisfaction of authoring exactly the desired statements, not just key words or a selection from a list of stock responses.
If the interactive fiction is designed to engage the interactor in rather than distance him or her from its world, the textual interface and its semantic richness can contribute to this goal by requiring the formation and typing of sentences and the conscious search for language to express one's desired action. The deep involvement of the interactor with the process of interaction is arguably a disadvantage in the context of a computer application such as a word processor or spreadsheet (when the focus should be on the desired outcome), but in a literary context it can be beneficial.
Textual interfaces have limitations and disadvantages as well, and the limited vocabulary of Infocom's interactive fictions exemplifies these. When the interactor's Zork character is in the Clearing area (where the 69,105 leaves are), the command >pick up the leaves and scatter them into the air results in the unsatisfying I don't know the word "scatter". Perhaps this attempt at frolicking is not consistent, however, with adventuring and puzzle-solving, and the parser's ignorance of a word such as "scatter" can be excused. In fact, the inability to comprehend every possible action that can be communicated in English is due to constraints that are beyond the control of interactive fiction designers. One of the limitations of computing which has been discovered through the efforts of researchers in artificial intelligence is that programs can succeed at understanding English only in well-specified limited domains; the ambiguities of language have hindered the development of systems for general understanding or translation. Even in an adventuring context, therefore, it might be too much to expect the parser to recognize our attempts to >drop leaves (i.e. put the leaves back down on the ground) and also to >drink a drop of water. (Note 9) Certainly the slang usage >drop the troll (i.e. kill him) or simply >drop (i.e. "hit the deck," fall to the floor) would likely be beyond the abilities of a good parser. The Interlogic parser recognizes only the first usage of "drop."
Less excusable is the parser's inability to recognize a reference to one of the interactive fiction's own phrases. When in A Mind Forever Voyaging the interactor answers Jill's question Hi, hon. Any news from your agent today? with >Tell Jill "No news." the response is [That sentence isn't one I recognize.] (Note 10) Similarly, should the interactor >examine the low branches when the character is in Zork's Forest Path area, the interactive fiction would respond that it does not know the word "low" - despite the appearance of the phrase "low branches" in the description of this area. Here, the context suggests to us that we might wish to refer to the low branches or the high branches specifically. It seems, unfortunately, that even the simple phrase "low branches" and the word "news" in A Mind Forever Voyaging are instances of the "purple prose" which the Zork manual warned us about, and are unknown to the parser, despite being used to describe the landscape of the story. "The inference that one would make is that the game 'agent' is severely brain damaged, since the agent that produces language and the agent that comprehends it are assumed to be one and the same. ... the agent represented more than it 'knew.'" (Laurel 59) Contextual clues like the text of the location's description or the words uttered by a character provide the best types of implicit constraints, which might suggest a parser's vocabulary or subtly prevent the interactor from typing input which cannot be understood or which has no effect. The alternative is an explicit constraint like a message indicating the parser's lack of understanding; but this detracts from the usual flow of interaction. (Laurel 104-105)
Not only is it unclear whether or not we can reference certain objects, but when we do wish to take action, it is not always obvious how to phrase the command. The semantic precision with which text allows interactors to define their commands also constrains them to using only those commands that are recognized by the parser. Beneficial engagement in the text can, at times, degenerate into an unpleasant struggle for acceptable phrasing. In the following example, after the interactor issues the first commands to Zork, it is evident that the rug should be cleared from atop the trap door in order to proceed. But the interactor has some difficulty accomplishing this.
>look under the rug Underneath the rug is a closed trap door. As you drop the corner of the rug, the trap door is once again concealed from view. >pick up the rug and put it aside The rug is extremely heavy and cannot be carried. I don't know the word "aside". >drag the rug to one side I don't know the word "drag". >move the rug to one side I don't know the word "side". >move the rug With a great effort, the rug is moved to one side of the room, revealing the dusty cover of a closed trap door.
Certainly, the task of moving the rug was accomplished only with great effort. In the first case, the parser saw two sequential commands rather than the single action of moving, and thus responded to the first one, "pick up the rug," as if the interactor had wished to begin lugging the rug about; the second part of the sentence triggered a vocabulary problem, as did the next two commands.
A final difficulty is that either consistency or uniformity must be abandoned if the interactive fiction is to utilize any medium other than text. If consistency is sacrificed, as in King's Quest, interactors may see something to which they wish to refer but be prevented from such a reference because the name or term for the object is not evident. If uniformity is sacrificed, interactors have to shift gears between different media and new problems arise at the seams of the interactor's various interfaces: each different input method has different ways of communicating the interactor's message (e.g. a double-click on a mouse does something different than does pressing a joystick button twice rapidly; typing requires that RETURN be pressed after each phrase of input whereas voice-recognition systems require words to be uttered slowly, with a pause between each).
Finally, existing text interfaces are not what Carnegie Mellon researchers call "highly interactive", that is, responding to input at all times. (Kelso et. al. 1) Interfaces now allow the interactor to influence the course of the narrative only after blocks of text appear, not at every moment. Infocom's one interactive fiction in which time runs independent of the interactor's input, Border Zone, is no better than any other work in this regard. At one point in Border Zone, the character whom the interactor controls is on a train. Agents enter his compartment, search it, and leave; the interactor cannot perform any actions during this search, and in fact has no option to interact with the agents at all, since they have left by the 'time' the command prompt reappears. Although events occur in Mindwheel independent of the interactor's input, scrolling on to the screen whether or not the interactor has recently typed a command, the interface to this work is also not highly interactive, since series of events occur that the interactor cannot interrupt. There is no technical barrier to a highly interactive text interface, other than the slow speed of typing relative to other means of interaction, like pointing and clicking a mouse. For interactive fictions that emulate conversations, a highly interactive text interface would be particularly appropriate, allowing the interactor to interrupt long statements as they scrolled onto the screen and allowing both fiction and interactor to append statements to their earlier comments if a reaction was not soon forthcoming. The person-to-person communication UNIX utility talk, which allows two parties at different computers or terminals to type asynchronously in two sections of the screen, is this type of highly interactive conversational interface, although it connects people rather than joining a fiction with an interactor.
The press toward such multimedia has already made all-text interactive fictions with textual interfaces, the dominant form of the 1980s, an anomaly in the computing landscape. But there is a place for text in the brave new surround-sound, graphical world. Interactive fiction designers in academic and shareware circles as well as a few commercial sources like Eastgate Systems are continuing to develop and distribute all-text interactive fictions and tools to aid others in developing these works. While there has been a flurry of publicity on behalf of multimedia, the benefits of text among other electronic media have not been unsung of late.
Interactive multimedia leaves very little to the imagination. Like a Hollywood film, multimedia narrative includes such specific representations that less and less is left to the mind's eye. By contrast, the written word sparks images and evokes metaphors that get much of their meaning from the reader's imaginations and experiences.
The above was not written by a novelist, nor by a literary critic or any other aficionado of traditional text. It is a statement by Nicholas Negroponte, professor of media technology at MIT and founding director of the Media Lab, who says he "does not like to read." (8).