The Storytelling Terms "AI" Keeps Getting Wrong

Exchange of Information vs Denial of Misunderstanding

I was reading a post about how to collaborate with LLMs when it comes to matters of craft, a post that advanced the notion that LLMs know what they’re talking about when you mention specific terms of art, such as protagonist, inciting incident, or key event.

In the last ten years of doing this, I’ve not found a set of accepted definitions for these terms of art. What I have found is a lot of disagreement and people using terms however they want. Artsy people tend to do that. Whod’ve thunk it, right? “We don’t need no stinking rules.” Mention rules and they bristle.

Take the definition of a protagonist, for example. A few years back I asked a writing group about recs for resources on writing dual protagonists. All of the answers I got had to do with writing dual (or multiple) viewpoints. The conflation of viewpoint character with protagonist is rampant. Ridiculously so.

Same with “inciting incident.”

Let’s start with the word “inciting.” I certainly did, so I went over to dictionary.com and typed it in. I kid you not, here’s the answer:

https://www.dictionary.com/browse/inciting...

Dude, really?

Pick your sources well, and that goes for people and for LLMs:

For our purposes we are obviously talking about an event that stirs up or moves someone to action because we’re talking about Story.

Problem is, when I was doing critiques (which I did for years) and then when I was teaching at various literary conventions, it became quite clear that some people thought that the opening of a story was the inciting incident. Makes, sense, doesn’t it? I mean, truly, I don’t blame them for thinking that. It’s what gets things rolling.

Others thought the inciting incident/event was the end of Act 1 (as we work it based on the 4-Act structure from My Story Can Beat Up Your Story (MSCBUYS) and K.M. Weiland’s 3-Act structure). Only a minuscule number of writers understood it correctly—the point at which the character (or the audience on behalf of the character) get emotionally involved, which occurs about halfway into Act 1, or at about one-eight of the story.

So, based on what my convention sample and on what my critique-group cohort “knew,” the “inciting incident” could be at 0%, 12.5%, or 25% of the story, depending on whose definition you used. And when we don’t need no stinking rules we scoff at definitions too—they’re not rules so much as guidelines, amirite?

Here’s Google’s definition of inciting incident. Good as far as it goes, but still a source of misunderstanding because it doesn’t deny misunderstanding. This is where my technical/scientific writing training kicks in, because effective communication is not just about exchange of information, but about denying misunderstanding.

The Google answer above is an exchange of information, but not a denial of misunderstanding.

The MasterClass people define it thusly:

The inciting incident of a story is the event that sets the main character or characters on the journey that will occupy them throughout the narrative. Typically, this incident will upset the balance within the main character’s world.—(source)

Not a bad definition, but not very clear. Neither is their example:

It’s unclear what “An example of this is Luke Skywalker’s recruitment in the original Star Wars film from 1977.” actually means. From studying MSCBUYS we know it’s the scene where Luke discovers Leia’s message, but that’s hardly “recruitment” unless you intend for recruitment to mean getting emotionally involved, which Luke does, because she’s a pretty girl, not because he’s a hero doing heroic stuff (which is actually good because Joe Ego (aka Gary-Stu) putting his hands on his hips and declaring he’s here to to save the day can just die, die, die; seriously, I hate these guys—all of them). But Luke doesn’t know she’s in danger so it’s not like it’s his recruitment into the rebel cause.

Luke made no deliberate choice to do anything when he found the message. You might argue that his first step was to go after R2-D2 who’d run off, which is a later scene. There is also no mention of placement within the narrative, not for pacing purposes or anything else, which is why it’s so difficult to “talk to” AI/LLMs about whether or not something is an inciting incident. If it’s anything like people, some of whom think it’s the opening scene versus the start of the story, some of whom think it’s answering the call to adventure versus the call to adventure itself, some of whom know it as the midpoint of Act 1 (aka PP05 MSCBUYS), we are using English and the words “inciting” and “incident” but we all mean different things.

Fictionary got it right, however, so giving credit where it’s due:

Source: https://fictionary.co/journal/inciting-incident/#:~:text=What%20Is%20an%20Inciting%20Incident,what%27s%20best%20for%20your%20story.

But again we run into what does “a dramatic way” actually mean. With how many stories start with action, often without context, dropping us into a white room with talking heads, running around like chickens with their heads cut off, how do we define “dramatic way?” It’s all drama, 24/7, without worldbuilding, context, and worst of all, without a reason to give a damn. Ooh, they’re in danger? So what? Everyone’s in danger all the time.

“Kicks off the story” is also unclear. That’s the opening scene, isn’t it? Nope!

If you do look at Fictionary’s map, be aware, their “Plot Point 1” is not our PP01, but our Turning Point 1. And their Plot Point 2 is our Turning Point 3.

So when we’re arguing with someone, whether it’s on the a forum/thread, or an LLM, and we’re talking about something to do with Plot Point 2, do we mean the middle of the story at 50%, or do we mean the 75%-mark?

This is one of the reasons I like to use percentages, even though numbers turn some people off. It’s why I tolerate people making fun of me using percentages and rolling their eyes at me about it.

Terminology is one of my major sources of frustration when working with an LLM (and with people too). Even when I created my own custom GPT with my definitions, with a clear path of what goes where and why, any conversation with the GPT ultimately breaks down as it goes. The “AI” doesn’t know where it is in the story, even when given reference materials that clearly state what schema we are using.

I run into issues when discussing other terms of art as well, such as head hopping. Like the writer population at large, it doesn’t understand what head hopping is. It conflates loss of perspective, loss of viewpoint, etc. with head hopping.

Here’s ChatGPT 4.0 on head-hopping:

When a writer floats viewpoint, it’s not necessarily a head hop. A head hop requires a shift from omniscient author-narration to character-narration, usually via a shift to first-person present tense. It’s basically direct internal monologue.

In the example given by Chat GPT, the first line is Sarah’s perspective. The second line is John’s, but there is no head hop, not technically. It’s sloppy floating of viewpoint, but not a head hop. The lack of transition from Sarah (first line) to John (second line) is just a sloppy float/shift of viewpoint. Clear?

Then in the third line, there is another perspective switch, and we have an actual head hop with “Why couldn’t he just admit he was wrong” and in the fourth line with “This again.” Typically we see these as first-person present-tense italicized, in order to meet the textbook definition of head-hopping by the omniscient author-narrator. Here the AI didn’t bother formatting them. Regardless of formatting, both those sentences are head hops because they are character-narration using character-voice when they should be author-narration.

Adding some distance would fix the head hops:

Sarah wondered why he couldn’t just admit he was wrong.

and

John felt his temper rising and pondered why this was happening yet again.

In both these cases, the omniscient author-narrator maintains distance by observing mindset (allowed in omniscient narration) without violating viewpoint by using character-narrator voice for narration.

Now, the fix (the single viewpoint example), is correct, so I do have to give ChatGPT that. Let’s all give it a pat on the head. I wonder if we kept asking the same question if it would change its answer. Do run your own test and let me know. I suspect that the answers ChatGPT gives me have been somewhat fine-tuned at this point.

I switched to o3 with the same question and it gave me this:

At least this answer had the added caveat for “internal viewpoint” making it better. So I asked for an example.

Finally, a correct answer, but one that also betrays an AI-ism. That of putting “dialogue” (in this case the direct internal) after physical beats again and again. This is Bad Writing 101 and I often see it in ingenues. I’ve seen as many as 20 consecutive lines set up like this, with all the dialogue coming after a physical beat. Sigh.

No, I’m not saying, “never put dialogue after a physical beat” but I am saying that when I see a lot of it this is how I know you’re new or you’re using AI to generate or rewrite your prose. This is AI-slop, a term I learned yesterday, ironically, from someone who teaches writers to use LLMs. Good to know that even people who use AI see the slop it outputs and aren’t slavish devotees to generative-AI (my precious can do no wrong you haters).

To get back on-topic, only an omniscient narrator can head hop and he does so when he uses the character’s own voice via direct internal monologue. When a non-omniscient author-narrator does it, it may be a loss of perspective/viewpoint, a viewpoint violation, authorial intrusion, or just sloppy writing, but it’s not technically a head hop.

This is what makes this term-of-art nuanced. I’m told the concept of head hopping as something to be avoided didn’t exist until the last fifty years or so, so be warned that some older works are full of it or may seem like they’re full of it.

I’ve been listening to some early 20th-century material, and while some transitions are sloppy and result in unclear viewpoint shifts, the first-person present-tense shifts aren’t that prevalent.

Here is Grok’s answer:

Here too, it conflates channel-surfing (the shifting of heads without going to direct internal monologue) with head hopping. The “Head hopping” example (the first bullet point) is actually a good example of omniscient narration without head hopping, but with the sin of channel-surfing.

It would have been head hopping if an omniscient narrator would have written:

I’m nervous about the meeting, Alice thought, palms sweating.

Bob wondered why she looked to tense. Is she hiding something?

I wish I could leave, Alice thought.

Bob smirked, amused by her discomfort.

Here the head hopping occurs because the author-narrator uses character-voice for unspoken dialogue. Anyone else getting Dune flashbacks?

The “No head hopping” bullet point is essentially correct, using a distant single PoV, that of Alice as a filter, either for an omniscient author-narrator (note the use of the filters) or a third-distant author-narrator (ditto).

Yes, it is better to stick to one perspective, no matter if it’s done in omniscient or third-limited, because few writers do the transitions that float viewpoint well, thus calling attention to the shift itself. Sometimes it goes beyond calling attention to something badly written and crosses into loss of viewpoint or reader confusion.

Another reason to avoid the channel-surfing is that even when done correctly, the switching of heads is jarring to some readers who then go “look, a head hop” even when it’s technically not one.

Ultimately, what I (and almost every writers I’ve talked to about this) want is the ability to define our own paradigm, our own schema, our own terms of art, for the LLM/AI to use so when we’re having a conversation with it, we’re on the same page.

I really don’t care for its consensus-based hot-take on matters of craft, because I have yet to see it apply terms of art like “inciting incident,” “key event,” “turning point,” “head-hop,” or “pinch-point” consistently.

If so many writers didn’t subscribe to the “We don’t need no stinking rules” mindset, we wouldn’t have these problems in the first place. But the issue is that with that mindset, The Consensus (TM) isn’t necessarily what you want. When one schema calls the 75%-mark “Turning Point 3” and another calls it “Plot Point 2” or one group calls the one-eight mark the “Inciting Event” and another the “Key Event” (which sometimes happens before the story starts) it becomes confusing for everyone and leads not just to arguments, but to errors, for people and for LLMs.

Those who’ve made a study of different schemas (like myself) and who’ve witnessed multiple food-fights about headhopping, know to start with definitions. It’s why I make sure to start with these when I give presentations because I don’t see how we can have a meaningful conversation about craft when we have different definitions.

Thank you paid subscribers. To access articles more than four weeks old, please consider upgrading to a paid subscription.