At the most reductionist level, IF is a program that takes input in the form of text, and returns output, also in the form of text. While there might be z-abuses that do not do even this, I think most people would accept the argument that such works do not fall in the category of IF.

IF also always has a set beginning. The program starts somewhere; there is some initial text. (One could randomize and have the initial text be different from occasion to occasion; that doesn't matter. There must be *some* commencement.)

An IF program also constrains the form of input. I could, if I wished, create a program that took whatever the player cared to type in and perform numerical calculations on the ascii values of the characters and through some mechanisms of its own return a value. This, I think most people would agree, would also not be IF in the form that we understand it. IF generally pretends to understand the player, or attempt to do so, for a value of "understand" equal to "recognize and assign to some internally defined set of categories."

Beyond that, it is difficult to say what IF is. Is there always a story, or even, as you put it, a story-generating engine? No; Threading The Labyrinth had no story. Is there always a modelled physical environment with which to interact? No; one could argue about Space Under the Window, but I would say that it presents a textual surface beneath which the state of the world has only a vague representation, and that it at least pushes the bounds of world-representation at all. TTL doesn't even try to present a world.

Almost universally, however, the body of what we consider IF *tends* to represent, in some form, an environment or imagined world whose physical space we can explore. This kind of modelling is at the heart of all the standard IF-writing languages: there must be first and foremost a way to represent the place where the player is; the movement between that place and other places that the player might visit; the objects encountered; and certain basic physical interactions between those objects. The vast majority of the truly standardized interactions have to do with placement and the constraints on placement: all of the go verbs, which place the player; opening and closing objects; moving them in and out of the player's possession; placing objects inside or on top of other objects. And the IF-work represents these things to the player consistently and allows the player to affect the situation in certain generally predictable ways. The text produced, then, is generally either a description of the current state of the world model (as summoned up inventory, look, etc) or a description of a change to that state ("Taken.", "Dropped.", "You have put the enormous emerald in the Ming vase.").

In that sense it is also ipso facto a story-generating engine, if you are willing to accept as story any sequence of narrated events. I think most people would want to define "story" a bit more narrowly, or, failing that, would drag in the word "plot": a narrative arc of some kind, a sense of progression through stages.

If there is progression, then there is probably also a constrained ending or set of endings. A purely exploratory work like some art show entries, or like, say, Schroedinger's Cat, enforce no ending on the player of any kind, and the only way to terminate the experience is by quitting the program. Most IF, however, presents the player with at least one ending-- the "winning" ending, which represents the ultimate state of progress in the game. It needn't be a *happy* ending -- note most of Adam Cadre's oeuvre -- but it does signal that there is no further progress to be made with the work, or at least no further progress along the path chosen by the player.

Depending on its genre and pedigree, the work of IF may present a number of endings instead of just one, and there are two typical modalities for this: all "losing" endings except for the one "winning" ending, or an assortment of equally "winning" endings which vary in tone but which nonetheless convey in their wording that the IF work is somehow "finished": that the player has done his job working a batch of text out of the game. There are a very few examples of games with many "losing" *and* many "winning" endings. The win-vs-loss paradigm is one encouraged by the structure of the library of Inform, at least, since it allows the author to flag the termination of gameplay in several ways. Flag a losing ending, and the player is invited to undo or restart the game (an unsubtle reminder that Work Remains To Be Done); Flag a winning one, and he may be offered the opportunity to read a sort of Afterword from the author (typically "Amusing").

The signalling of a complete vs. an incomplete experience is central to the nature of IF as most people play it. As a player, one often feels something amounting almost to a responsibility to play through the "entirety" of the game, to see the whole thing. This doesn't equate to reading all of the text that the IF-work is capable of producing, though many players want to see at least all the good bits, and this is why the Amusing option is provided in the Inform library: to allow authors to alert their players to interesting and entertaining "side material" that they may have neglected to notice.

Where the IF-machine generates a definite "winning" ending for the player, it begins to approximate a story-generator more closely than the purely world-modelling device mentioned above. Most people would, I think, accept the statement that a story needs to have a beginning *and an end*; if the IF-work produces no ending, or, frustratingly, only a bunch of misleadingly "losing" endings (I have encountered no such game, though I have sometimes in despair supposed myself to be playing one), then it is less story-like and more model-like.

So for the vast majority of IF-works it is fair to say that they are both world-models and story-making machines.

[This begs the question of what happens when there are *multiple* winning endings. What is the "complete" experience of the IF-work then? Is it one playthrough? Three? Ten? Does one have to reach all the endings? If not, how does one know when one is finished? What is the IF-work then producing? It's a story machine, yes, but it is making such different stories; where is the sum of the parts? A game like Galatea allows the player, through repeated playings, to create the sense of the deeper narrative under the surface one produced on a single playing: certain things are constant. (Galatea is only a so-so example because some of the endings are mutually contradictory; reconciling them is left as an exercise to the post-modern reader. Perhaps better than "deeper narrative" is "deeper structure", and that structure, in the case of Galatea, is not a single story about the sequence of events that occurred, but a concept of personality and character, and the forces that created said personality and character.)]

In any case, if we accept the story-making-machine idea, we nonetheless find that in most cases the story the machine is able to produce varies considerably at a microscopic level (player a chooses to take inventory; player b chooses to kill the troll; the texts do not resemble each other remotely.) But there is an upper limit to this variability because, first of all, there is only one "winning" ending and because, second, progress towards that ending is marked by certain benchmark actions. These may be the solution of what we call puzzles, or they may be simply crossing a certain amount of map territory, or waiting a certain number of turns. Benchmarks are usually also unique, semi-unpredictable, and marked by a change in the level of generated prose.

By unique, I mean that the player can only experience them once in the course of a given playthrough.

By semi-unpredictable, I mean that while the player may have had a general goal of achieving the action of the benchmark, the results of doing so were at least partially unknown to him in advance, at least the first time he played. This sets a strong contrast with the results of average actions on the microscopic level: the player can type look three times in a row and know exactly what he will see each time. Moreover, the fact that the answer does not change is not a flaw in the design of the game.

The change in level of generated prose simply means that usually the author of the game has crafted the descriptions of benchmark events by hand, rather than allowing them to be machine-generated. They are outcomes that are not prescripted in the underlying library of the game: the parser produces responses like "Taken." and "Dropped." and even "You turn on the brass lantern.", but not "Suddenly, the bolted door shivers, and with an eerie groan tears itself from its hinges..." That action might even be the result of a partially prescripted system-- eg, say, a game with a set of rules that allowed the player to make inanimate objects animate. But the prose indicating an important juncture in the text tends to need to be handcrafted: because it is unique, because it is not predictable, and because it needs to be striking enough to make the player aware that a benchmark has been achieved. [Imagine if you will a game in which you would not be allowed to progress until you had performed a series of completely banal and completely random tasks: picking up a blue disk from one room, say, and putting it on top of another. If there were no special text that occurred when you performed one of these actions, you would never be able to distinguished the instance of "Taken." that was important.]

There is additional variability if the player is allowed to perform the benchmark actions in a sequence of his choosing. This kind of variation is what Graham Nelson has tended to describe as "breadth", as opposed to "linearity": a game with a large number of options available at a time and many benchmarks to be achieved in any order is broad; one in which benchmarks are presented to the player one-at-a-time is linear.

In most cases, regardless, the story-making-machine waits for benchmarks presenting either a new stage of the game (in the form of new possibilities for the player to explore, a new section of the world, another character, a scene change...) or the end. "Losing" endings are, among other things, a way to keep the player from straying too far from the intended path. Defy the internal logic of the IF-work and you will find yourself floating home with no oxygen.

Some games progress by internally automated benchmarks: Rameses is a game at such an extreme of the spectrum that each *move* achieves a benchmark, no turn is exactly like any other, and failure to make progress is impossible. After a certain number of turns has elapsed, the story will deposit you at the "winning" ending. (That it is an extremely depressing winning ending has no relevance to this discussion.) The more a game automates benchmark-achievement, forcing the player to make progress, the more we are inclined to call it "railroaded." Spellbreaker is very linear, in the sense that one has to perform the benchmarks in order; it is not railroaded, however, because it is possible to sit there failing to make progress for hundreds of turns. Though linear games needn't be railroaded, railroaded games are nearly always linear -- how can they help it, considering that the player isn't required to add much? (I would submit, as a possible exception, Best of Three, where it is possible for the player to have some effect on the course of the game, but where it is also very easy to get through the whole thing to a "winning" ending without exercising much initiative. This is, in my opinion, another flaw.)

The existence of benchmarks gives us more of an excuse to talk about the "plot" of a game: if we are required to generate *certain specific pieces of text* in order to make progress through the game and arrive at the "winning" ending, then several things result. First, a uniformity of sorts is imposed on the experience of the players, since all those who complete the game successfully do so through related processes. Second, an especial importance is given to the benchmarks themselves; they are marked out in the text as the part of that text that is relevant to the story as story-ness.

Here I will wax general and point out that humans are generally inclined to take a reductionist approach to gobbets of narrative text; we strain out the "Taken."s and the "So *then* Bob and I went into this little antique store..."s and the "Little Red Riding Hood's cape was fringed with beads and gold trim...", and latch onto what we consider the most central aspects of story, which we perceive like the skeletal structure under flesh. There have been students of Homeric poetry who, in the interests of finding out about oral tradition, have studied at great length and taken dictation of the performances of oral poets in Serbia: among their discoveries was the fact that a poet might claim to perform the "same" poem or give the "same" performance on two occasions even though the recorded documentation shows numerous differences between the two works. The point is that for the poet the event did not change *in its essentials.* Writing-- text-- gives a fixity to narrative that privileges every word equally: The letters spelling "The" or "for instance" are as lasting and immutable as the letters spelling "Lancelot seduced Guinevere." But our appreciation of the underlying plot structure exists, nonetheless.

[Still not a word about puzzles, not really. Hang in there and I will come to that.]

Some story-making-machines, ambitiously, attempt to extend the variability by allowing for, as mentioned, different "winning" endings, depending on a character's crucial choice near the end. Kathleen Fischer's Masquerade follows this description, as does Ian Finley's Kaged. This allows some flexibility to the player, which constitutes, at least in these two works, a sort of permission to put what spin on the story one will: the *events* are all in place, established by the player's benchmark actions, but the outcome and thus the meaning of those events rests in the player's own hands, and is a real choice.

Yet more ambitious is the kind of variability that allows the player to select one benchmark rather than another. Adam Cadre's I-0 is a famous example: the "winning" ending is to the best of my knowledge always the same, but the player is allowed to experience quite different events according to her choice. And here the story-generator is putting what we might legitimately call "plot" in the hands of the player to control. How *much* control the player has, or how meaningful it is to the player's experience, depends to a large degree on how clear the decisions are that lead to one benchmark rather than another. If the player understands in advance what will happen, that's one thing; if he understands only in retrospect, that's arguably a lesser kind of participation *in the making of the plot*. If he never understands at all, then the story-making machine seems capricious and flawed, and this problem to some degree plagued my own Best of Three.

So far I have talked as though IF were a composite of two kinds of interaction: the wholly prefabricated kind, the take and drop and inventory actions that are similar on the whole from game to game, and the actions-that-lead-to-benchmarks, which are unique, plot-determining, and author-scripted. Obviously, this is not the case; there are also those elements of the game -- locations, objects, the interactions of objects -- that are added by the author and which can be experienced repeatedly through a composite of pre-crafted and automatically generated text. Take the average room description. There is the name of the room, often printed by the library in a boldface or other marked font. Then there follows a description of the room as a whole, which is also written by the author in full in advance. Then there is the description of the objects *in* the room, and this description may be generated automatically ("You see a platinum bar here.") or customized to some degree by the author ("A gleaming platinum bar rests in a dusty corner of the room."). Generally description that covers the interrelations of numerous movable objects ("You see a table. On the table are a platinum bar, a leather pouch (containing 2 gems), and a book.") is highly automated, because it is both complex and changeable. [Footnote: on background events in Best of Three.]

Readers of static fiction may find it startling that the incongruity of diction isn't more of an impediment to the IF player. A lavish and lyrical room description may have tagged on, as if the most thoughtless appendix, a list of objects in the most rudimentary format. If the text is taken simply as literature, the effect is a bit bathetic. But here again there is a question of how to read: one understands that certain text is more important than certain other text, and, generally speaking, the automated text is automated to allow for the intrusion of the player's own will.

It is in this automation that the other pleasure of IF occurs. One experiences a story stitched together by a machine, with plot points of its choosing; but one can also affect the story, especially if one understands the rules of operation. Here we return to the issue of world modelling. In the world of the author's creation, there are objects and places unfamiliar to the player and whose interaction and response to stimuli ("UNLOCK METAL BOX WITH GREY KEY", "TURN ON FLOOBLE-SYNTHESIZER") are unknown at the beginning of the process. The player generally wants to find out what these things are and how they work. How does one room relate geographically to the rest of the rooms around it? Easily learned, but made more complex by mazes. How does a complex object behave? (Look at Jigsaw's Enigma machine for the ultimate in Complex Object Fun.)

More satisfying yet if the player can find out what set of general rules governs the interaction of whole groups of things in the modelled universe. Some general rules are nearly universal to IF, rules about putting things in other things, not putting things in themselves, about passing only through open doors and not through shut ones, about light and visibility. But these are a basic physical model, over which can lay another and often quite intriguing system of interactive possibilities. These have included, for instance, the spell system of the Enchanter series, where specialized verbs introduced additional possibilities for the player; the behavior of puzzle-box games like Colours or Schroedinger's Cat; the machines in Metamorphoses; the gadgets in Spider and Web. The finale of Augmented Fourth is another wonderful example of learned information coming together.

It's some combination of the desire to explore and the desire to hear a story that drives players through IF games. Not all players possess those motivations in equal degrees: for some an easily-completed but narratively powerful IF-work such as Photopia will be more enjoyable; others will appreciate a narratively rambly but exceedingly explorable world like that of Curses.

When exploratory achievements (determining how to manipulate something, whether it be a crotchety NPC or a lock and a key) become in themselves goals and benchmarks of the game, they are puzzles. I say "goals and benchmarks" not in a spirit of redundancy, but because there's a general requirement *that the player know in advance* that he is working towards a certain thing in order for it to be a legitimate puzzle (at least according to my definition of the term). Hence "goal." The puzzle might be as vague as "figure out what is going on here and then take appropriate action" (as in Pytho's Mask) or as specific as "solve this fifteen-puzzle in order to release an object into your possession." A puzzle-benchmark can also be a plot-benchmark, and these are, I think, the most satisfying kind: those moments where the player interprets the nature of the world, understands it, and manipulates it according to his understanding in a way that also causes a new piece of the plot to fall into place. At such times, it feels as though the story plot has been *earned* by the player and is consonant with his own intentions.

[In his review of Dennis Jerz's Fine-Tuned, Adam Cadre talks about its "participatory humor:" moments where the player is caused, by the natural behavior of the game, to set up and be the instigating factor in humorous events. In my emphasis on world exploration and plot, I don't mean to suggest that there aren't also other things to be enjoyed in IF, such as humor, tone, the quality of the writing. But rare as it is for an author to allow the player to participate fully in plot events (that is, to want something to happen and thus to form and execute a plan to bring it about, rather than receiving the plot advance as the "reward" of a somewhat unrelated puzzle or through effectively inadvertent action), it is even rarer that the player is able to participate significantly in shaping humor, tone, thematics, and so on.

Some of Nick Montfort's work allows the player to make choices that do affect the diction of the game: Ad Verbum adopts abbreviations in describing directions if the player also uses abbreviated directions. Winchester's Nightmare, in an attempt to make the player a true co-author of the prose that constitutes the game's fabric, refuses to accept abbreviations at all. My experience of the latter effect was mostly that it is annoying, and impeded the other interactions I might have with the game world which were, in my opinion, more significant than the privilege of creating some fully legible English prose. It might have been significantly different if I had in fact had the full expressive range of natural language at my disposal; but disguising IF pidgin as "real English" seemed a lesser accomplishment. The experiment was unquestionably interesting, however.

More enjoyable was The Space Under the Window, which offered little real opportunity to affect the gameworld or plot intentionally, but which allowed the player to shift emphasis, tonality, focus: the block of text at the top of the screen was removed and returned in a different format, or extended in length, by each choice. The effect was close to what I imagine interactive poetry might be like: imagine a sonnet in which you could demand the elaboration of a conceit, or reject it wholly--]

On this understanding of IF, what we call "puzzles" are in fact the natural effect of allowing the player to interact with the gameworld/plot in ways whose effects he can reasonably anticipate. If the player cannot affect the gameworld/plot significantly, then there may be no real puzzles; if the effects of his acts are so inscrutable as to be effectively random, then there may also be no puzzles. Neither of these options seems to appeal to people much.

Note that there are two stages: the reasonable anticipation must precede the successful manipulation. Most puzzles consist of one part exploring until the rules of the gadget/NPC/puzzle/interaction/universe are understood, and one part applying those rules. The problem with classic lock-and-key puzzles is that the rules are trivial and don't take any thought, and that finding the key is often tied to some quite arbitrary unrelated actions. The problem with traditional mazes is that the rules of puzzle solution are understood, but that the application is lengthy and tedious -- a nasty double-whammy. And there are puzzles that simply require exhaustive thoroughness (LOOK UNDER BED, SEARCH PILE OF STUFF) when there is no reason to expect that one will find anything, but one feels that it is one's duty to look under, search, move, rattle, bite, sniff, etc., each and every object one meets with, and pick up every portable item for the inevitable moment when it will be required of one. So there is, actually, a set of rules-of-the-universe for this kind of game: the rules say that the model world was invented teleologically by the author; that there is nothing in it that does not ultimately contribute to the accomplishment of some benchmark or other; and that certain actions can be (thoroughly and repeatedly) applied to achieve progress. The progress itself is pursued in a procedural rather than a goal-oriented way: the player does things not because he expects a given plot shape to occur or because he cares about the end of the narrative but because progress, in general, is good, and it is Enough if his Score Goes Up. In the oldest forms of IF, it was enough to present a locked door, and the player would wish to open it and strive fearlessly and ceaselessly until he did so, even if the door were locked with bolts of titanium and set in a medieval fortress inhabited by alien ninjas. The goal was there, but the sense of relevance and the relationship to the overarching narrative structure was not.

I'm not entirely knocking this. I loved Curses; I've played it at least twice, probably three times, in full, and gone back to bits of it, so full of wit and charm and lithesome grace were the descriptions. But my favorite parts of it were the ones where the puzzle, if arbitrary, at least allowed me to anticipate exactly what would occur and know why I desired that occurrence. Not the locked doors, then, with who-knows-what behind them, but the clear glass pane through which I could see a Rube Goldberg device that would do something I could guess at from afar.

People who write or play games with the plot-aspect and the story-generating machine foremost in their minds are liable to get a little impatient with randomly inserted puzzles. Some of the early writing I read about IF kept saying, "and here is how you insert puzzles..."; and my thought was, "why would I want to?" From my point of view the goal isn't to litter my player's way with thorns; heaven help us, I have enough trouble enabling the player to move the plot forward. If there are no plot benchmarks, no events significant in shaping a narrative, then inserting as many puzzles as possible does make sense. Likewise, perhaps, if there are plot events, but the player has no way to anticipate or control them; I would say that this is the case in Curses, which does at least begin to have a story, but it is not your story. The player's experience is about coming up with clever ways to open doors and travel between worlds.

So I think that what I have sometimes referred to as my desire to write "puzzleless" IF is simply a desire to integrate intentional interaction into the game world in such a way that it does not seem artificial or contrived, *even* in the context of a complex and character-driven plot. People have pointed to Galatea as a puzzle game because one can, on repeated playings, build up enough of a sense of the NPC's personality to begin to affect her behavior intentionally, rather than stumbling on results by guess.

The desire to coopt the player into participating meaningfully in the development of a plot goes hand-in-glove with the desire not to force him into any tedious, plotless behavior; and this, I think, is where we get games with temporal discontinuities, where, when one benchmark is achieved, the player character is transported to a later time and the player is given a chance to re-enter the action when it gets interesting again. Otherwise, the player would be forced to play through scenarios in which there was no kind of interaction that wasn't there purely as connective tissue: long sequences of driving a car to and from work, making oneself dinner, watching television, going to bed, before one could get back to the fascinating matter of the office intrigue, the secretary who is a secret agent, and so on. A game that attempts to generate a coherent and plausible plot, rather than simply a string of puzzles to push through, may often be structured in such a way that these narrative temporal gaps will naturally occur. This is what I meant when I said that the gaps allow an alternative means of pacing to the pure insertion of puzzles.

Temporal cutting has a cost: the player is ripped away from direct identification with the player character, because the PC experiences things that the player never sees.

It also gives rise to the possibility of coy narrative tricks. Both All Roads and Photopia challenge the player to fill in the gaps and make sense of the connections between apparently disconnected scenes; they allow major plot points to occur off-screen, unseen and only reported, or, worse, only interpreted. This shifts the interactivity of the experience largely into the realm of the pure intellectual, in the player's head: do you understand, or do you not? Does the plot take shape, or doesn't it? Manipulating the game world can only generate clues, in the form of the text, and that is where the interaction focuses. The deeper plot is unchangeable.

Where that leaves us is with a good deal of unexplored territory and few simple conclusions. I continue to think that IF that allows some interaction with the diction of the text is an interesting idea, or with the tone in which things are presented. But it's difficult to see how that would work, beyond very simplistic commands ("Type GOTHIC to see this story done in a style florid morbidity.")

It's even more difficult to see how it would engage the player the way world-manipulation does. As human animals we wander around all the time manipulating the physical world in small actions with some larger goal in mind: drawing shapes on a piece of paper, which we refer to as "writing a check", which we expect will result, eventually, through the operations of the extremely complicated universe in which we live, in someone at the power company consenting to continue to send us electricity, which will manifest itself in a bright light when one's finger is applied to a switch-- and all this our minds understand and describe in many levels of abstraction, speaking now of "writing a check" and now of "being responsible" and now of the strange progress by which we have become adult and find ourselves needing to do responsible things like pay utility bills. The brain seeks a narrative. It retrofits significance. It comes up with benchmarks.

IF, at least in potentio, is like a form of life where the activity-to-narrative conversion is cleaner and less problematic. Which events in my life are part of my "life story": that's a very subjective judgement. Some would say none, and that people's lives aren't stories, and that attempting to impose that kind of organization on reality is foolhardy. But we like stories, and IF offers, at times, the immediacy and directness of life-like actions mingled with the pleasing elegance of a well-told tale.