I have a question regarding writing style. It started out with an AI question where I got this feedback(*):
As an aside, what the AI is generated isn’t really what I’d consider “good” writing in the context of a parser game. It’s over-wrought and over-detailed and everything gets the same amount of focus regardless of whether or not it’s actually important to the scene.
And this:
I think most players would prefer:
You are in a tool shed. A door leads south. An orange life ring is here.
Thus, the general feedback was: “give me shorter more distinct texts” in favor of a more colorful language. Okay, the AI text example may be long, but I still like the mood it sets with “You stand within the cramped confines of a tool shack, where the air hangs heavy with the scent of oil and the slow decay of forgotten things.” vs “You are in a tool shed”. There was actually an intention behind that colorful generated text.
It stems from a desire to write an “impressionistic” game - something that really sets a mood and atmosphere - which I don’t think the short “You are in a tool shed” does.
So, I wonder; if you take out the AI background and focus on a short vs “atmospheric” writing style - is there any kind of consensus or guidelines? Do you prefer short text because it is a game where you want to progress fast instead of reading long windy texts?
Its a genuine question and let me share the background; I spent my summer vacation travelling in Iceland where I was presented with the most amazing scenes, waterfalls, glaciers, desert, boiling mud and much more, and it triggered the idea of embedding it in a text game - a fairly straight forward story but with a very impressionistic style of writing that would try to share the impressions that I felt on the locations.
Thanks
(*) sorry for the double copy, but the editor insists on showing the wrong part of the quoted link and I can’t fix that.
To me, it comes down to what the author wants to convey.
A poetic style can be great for conveying certain things, like the atmosphere you’re describing. A more prosaic style can be great for describing “exits are east and south, there’s a lamp here”. Different games will want to do it differently.
The problem with the AI version, imo, is that it’s taking the information “exits are east and south, there’s a lamp here” and obfuscating it behind poetic language, while that poetic language isn’t really conveying anything, because it’s all AI hallucination.
I also think it’s easier to get away with that in a choice-based game where you have another mechanism (links) for conveying what’s important.
In a parser game if you say
You stand within the cramped confines of a tool shack, where the air hangs heavy with the scent of oil and the slow decay of forgotten things.
then I’m going to immediately try to smell, smell oil, x oil, get oil, search forgotten things, etc.
It is possible to do an impressionistic style in a parser setting but I think it’s a particular skill-set to do it without implying a lot of red herrings…
Exactly! The text in a novel serves the dual purposes of conveying themes and mood to the reader, and also conveying a plot. But the text in a parser game serves the additional purpose of conveying what parts of the world can be meaningfully interacted with, because no game (AI or not) can truly implement everything—if it tries, well, we’ve seen how AI Dungeon tends to go.
If the theme and mood get in the way of the world model, you get something atmospheric but difficult to interact with. Which is a problem for interactive fiction.
Long text in cut-scenes is OK and appreciated, I like to see the author expressing himself.
The more I have to do, the less I prefer to read. I do appreciate a bit of colour in room descriptions, but this has to be the first time I enter in a room only; subsequent look invokations should return a briefer text.
In general impressionistic style can be pulled off while keeping the gaming side of things smooth and functional, just avoid long descriptions that cannot be meaningfully interacted with!
I want to be able to easily see what I need to pay attention to. A great example is from Spider and Web by Andrew Plotkin:
End of Alley
It’s a narrow dead end here, with walls rising oppressively high in three directions. The alley is quite empty, bare even of trash. (Your guidebook warned you: the police are as efficient about litter laws as about everything else they do.) You can retreat to the south.
A plain metal door faces you to the east, near the alley’s end. It’s firmly shut.
Notice how the first paragraph is description. You can EXAMINE WALL, etc. but the important thing is clearly the metal door, offset with a terse paragraph about it.
Sometimes people say my descriptions are too long, but others like them. I try to avoid people getting lost, because I get the concern… but I write because I like writing and make games for people who enjoy reading what I write. That’s my audience.
I am offering an option to capitalize all nouns and directions in my next game to make it easier for players to keep track of things.
I tend to cut loose a bit more with cutscenes. I write narrative games and have to put the narrative someplace.
edit: I’m skipping over the unimplemented nouns because it seems that’s covered. I’d say as a last piece of advice that all text should be genuinely interesting (which is hard!) and reflect authorial vision. I’d prefer bad idiosyncratic writing to the example in the OP.
The answer to the question of whether sentences should be short or long and colorful is “whatever is most interesting”. While I tend to strive for minimalism, there are times when a long, rich sentence can be more provocative for the reader. I believe the reader’s imagination is more productive than the writer’s imagination.
For example:
While evoking an atmosphere is certainly important, I would find the former difficult to read because it doesn’t leave much to the reader’s imagination. The generated text tells you straight away the tool shack is decrepit due to age. It fixes the object/scenery into place and there isn’t much to work off from besides being a longer way to say “this tool shack is old and eerie”.
This is emblematic of how I feel about “purple prose”: they are long sentences that don’t give any freedom to the reader to imagine the space. Now, I’m not attacking long sentences or dense paragraphs – a writer like Lovecraft is able to create strange imagery that still provokes more thought with his florid sentences – but I find what I consider purple prose to be “thought-terminating” for lack of a better word. They explain how you should feel instead of letting you figure out what the space should look like.
This is not to say I find “You are in a tool shed” wonderful prose. This is extremely terse and to the point, which is useful for puzzle parser games. But if you are interested in evoking atmosphere, I feel that adding objects (like the orange life ring) and quirks that the tool shed has will create enough points of interest that stimulate some imagination.
I think interactive fiction as a whole is very similar to tabletop games, but your dungeon/game master is offline for the most part. The game master says a few lines about the setting and then lets the players evoke their own imagination to see what they can do in said setting. Colorful scenes can certainly help in creating this scene, but imagine a game master who keeps describing how wonderful the vistas should be when the player just wants to move on. They should be used carefully and perhaps not all the time.
It’s important to think about what motivates and interests the player when writing descriptions. It’s a lesson I keep relearning all the time when I make games. I’ve recently read The Art of Description by Mark Doty and while I don’t agree with everything he’s written, I found his analysis of descriptions in poetry useful to understand why certain descriptions work and why others don’t. For one, when a narrator describes a scene, it not only paints a picture of the world but also the narrator’s inner world. A pointlessly florid or terse sentence doesn’t help us understand what the narrator is saying, emphasizing, and omitting. We don’t get a sense of the narrator and how they feel about the world around them.
While Wittgenstein certainly meant something else, the first sentence to his Tractatus Logico-Philosophicus reads, “The limits of my language mean the limits of my world.” I find this useful when thinking about description. The way we reach the limits of our own language is very particular to our worldview, and description is always describing a scene and us. When we describe anything, we are not only evoking a specific atmosphere but also ourselves into the picture.
What I find interesting about parser games is that the better games actually understand this well. The Spider and Web extract Hidnook helpfully quoted gives you several examples: “walls rising oppressively high”, “quite empty, bare even of trash”, “a plain metal door … it’s firmly shut.” Not only do you get a sense of how highly policed the space is but you also get some insight on how the player character observes the world: they know to be concise and precise with details as if they’re trained to do that. Getting the narrator’s voice right is important in understanding what a scene description should feel like.
Long story short: I believe there needs to be some intention behind the length of your descriptions and how they help the player understand and imagine the world better. I like to think of descriptions as “prompts” for players to spark their imagination since it’s not only more useful for me but also it stops me from caring too much about how wonderful my prose is. I used to write descriptions not for the reader’s or story’s sake but because I really liked it. There’s no problem with that, of course, but even I recognized back then I was being very silly about it. There wasn’t rhyme or reason behind my descriptions; I just thought I was the next Nabokov. Since I won’t be writing the next Pale Fire any time soon, I am more interested in developing my own voice and thoughts about how the audience should read my work.
When I originally wrote that piece of feedback, my gripe was not necessarily with the descriptive nature of the text but with the fact that the LLM applies that descriptiveness to everything willy-nilly. Description is something that draws the player’s focus and can add wonderful texture to the game experience, but it has to be applied in moderation. Call it the fondant of the IF world.
Let’s walk through the original description you generated:
You stand within the cramped confines of a tool shack, where the air hangs heavy with the scent of oil and the slow decay of forgotten things. Shadows cling to the corners, and the dim light reveals the outlines of ancient tools and shelves thick with dust. The silence is profound, broken only by the faint creak of the shack’s timbers, as if the building itself remembers a time long past.
This is definitely atmospheric, and while a little over-wrought I definitely dig what it’s doing even if I’d personally want to both cut it down and vary the sentence structure more. Let’s continue:
On a battered hook, a life-saving ring dangles, its once-bright orange now mottled and faded with age. The rope attached to it is stiff and frayed, a relic of desperate hope in a place that feels abandoned by time.
The LLM is using the same florid style to describe a single object here. It’s not getting anything new across that wasn’t already conveyed in the room description – we already know this place is old and decrepit! This is another two sentences at the same level of descriptiveness and with the same general structure, and it’s getting old.
To the south, a weathered door stands slightly ajar, leading out of the tool shack and into the unknown.
And by the time we get to our last sentence (which is also similarly structured to the previous!) I’ve started skimming.
A human writer here would vary the sentence structure, length, and level of descriptiveness to both set the mood and draw the player’s attention to important gameplay objects. If everything is overly described nothing stands out to the player in either way. This is a mistake that plenty of newer authors make as well, but LLMs don’t have understanding of what is important to the player’s experience because they don’t “understand” anything the way we do – a human author can be easily trained out of it but I’m not sure an LLM can.
As a demonstration, I’ve assembled and lightly edited various bits of the text posted in your other thread into something that I think nicely balances atmosphere with gameplay concerns:
You stand within the cramped confines of a tool shack, where the air hangs heavy with the scent of oil and the slow decay of forgotten things. The silence is profound, broken only by the faint creak of the shack’s timbers.
A faded orange life ring hangs on the wall. It still has a rope attached, albeit barely.
As another example from the other side, I’ve pasted the first paragraph of the LLM-generated description and some commands I may input:
You stand within the cramped confines of a tool shack, where the air hangs heavy with the scent of oil and the slow decay of forgotten things. Shadows cling to the corners, and the dim light reveals the outlines of ancient tools and shelves thick with dust. The silence is profound, broken only by the faint creak of the shack’s timbers, as if the building itself remembers a time long past.
> x shack
> smell oil
> smell
> x light
> x tools
> x shelves
> x dust
> x timbers
> listen
Notice how lots of irrelevant details are examined. The expectation is that all of these things, if mentioned, are important, or at the very least have some modicum of importance (enough to warrant a description), else they wouldn’t be included. If something like “tools” are explicitly mentioned but not implemented, it seems messy, and, again, makes it hard to figure out what actually is important.
Yeah, this is where the difference between interactive and static fiction really comes into play! You can’t “X LIFE RING” in a book to get more information, so what you see on the page is what you get. But in parser IF the structure allows you to go more in depth for things you choose to implement.
To follow up on Hidnook’s train of thought - what details are being held back in the initial description to be elaborated on later? I.e. what’s the difference between the generated room description for the life ring and the text you’d get by examining it? Will the player get the same flowery description of the life ring every time they enter the room if they haven’t already taken it?
The issue with the example is its lack of utility. I’m defining “utility” broadly. It doesn’t really reveal anything about the story, or the character, or why the room matters. Or effectively identify nouns, as we’ve seen. I think it’s constructive to think about what the text is doing within the game.
Yes, this is how I feel as well. I actually like the AI-generated description of the shack itself. It gets the atmosphere across, and is interesting (to the extent that text generated by an AI can be interesting - it’d be more interesting if it were generated by a person). But the descriptions of the life ring and the door are too much. They could work as a long description that you get when you EXAMINE them, but not in the room description. For example, the door leading into the “unknown” is obvious, because I haven’t gone there yet! And if that word is still used when I have checked out what’s on the other side of the door, then that’s a miss too.
Is it your plan to actually try this? i.e. write either your trip to Iceland, or a game/experience you’ve thought of, and to do it in this style?
The replies in this topic are wall-to-wall excellent. I just know that my advice when someone asks an elemental question to which the answers from the experienced are deep and numerous, my own advice is that they try the thing they’re thinking of doing.
The analogy being, have you gone to the canvas and wiggled a brush around yet? It’s daunting to sort out all the craft and skills needed to paint a particular kind of picture if you haven’t tried to paint any kind of picture yet (I don’t know that you haven’t).
In your case, maybe write and implement a room of your trip, or imagined game. In trying, you’ll start to see which problems and challenges are yours, which won’t be yours because you find you don’t care about them, what you actually want or like or don’t like.
Thanks for all your insightful answers to this. I can see “cutscenes” or one-time introductions to be a constructive way forward if I want some place to pour out my fascination for a place. I also get the point about using evocative language where needed vs just pouring out over it all (for instance object descriptions). And I certainly do get the part about not introducing random things that a player will feel compelled to examine.
Set a mood. Encourage me to imagine details that aren’t spelled out. Give me an idea how the player character feels about this place or what my role in it is supposed to be. Make sure to cover the important things present and maybe one or two fun extras.
Nothing like a trip to a foreign place for inspiration. There was plenty of time and plenty of magic scenery to inspire, so I actually have a notebook stuffed with scenes and some backstories. Even a plot, quests and ending. Always dreamed about writing down all the stuff in my head in a fantasy story, but also very well aware that I would never get through all the grinding of fleshing out the story.
Then AI shows up, and I’m like “wow, if I can structure my notes and, layout the facts, maintain game state as normal - then I can actually make my story come alive by asking the AI to narrate it”. And while I agree on all the points people have shared here, I would never in my life have been able to phrase that room introduction myself. So to some degree I actually enjoyed playing the first scenes myself.
And then of course there’s the factor that I’m a software developer, not a literary writer, which means I always end up focusing on programming and loose sight of the narrative and its inspiration.
I really enjoyed the feedback here. Made me much more aware of how people see and interact with IF. I’ve been around long enough to be fascinated by Zork 1 years ago, but haven’t focused on IF otherwise.
Something worth noting is that most of Infocom’s implementers weren’t writers first. There are AFAIK only a few exceptions – Michael Berlyn (Suspended, Infidel) comes to mind, though no one is reading his stuff for the prose – but many of their games remain classics because their economical prose works really well with the medium and they didn’t have the file size to write more than what is needed anyway.
In other threads, people have noted that what makes Infocom titles classics is not only their originality but by how well playtested and responsive they are. Playing these games feels smooth for parser players who know how to navigate their way. This responsiveness is what I think makes their titles quite magical, even to this day.
I’m much more of a Hemingway iceberg theory guy. A few, well picked, poetic sentences that sets the mood and trigger the reader to fill the description with his own personal experiences. Long overwrought descriptions quickly becomes tedious in a location visited many times. Richly described objects that isn’t implemented is only frustrating.
Hand on heart, I’m not sure if I ever read the whole description of the volcano in Crowther & Woods Adventure.
Yeah, even before Kastel commented, I was thinking I should find one of the YouTube videos where Jon Ingold talks about being a math guy in secondary school and that it was more at university where he got disillusioned with math and turned to writing.
So it’s something you can learn, if you want to. TBH it can be kind of fun, pulling bits of literary theory and writing craft into your programming mindset.
