A passion project by Derin Savasan

The Shapes
of Stories

Every screenplay is a line that rises and falls. This is what 1,627 Hollywood films look like when you plot their feelings from first page to last.

Kurt Vonnegut at the chalkboard, 2004 — The protagonist moves between good fortune (G) and ill fortune (I) from the beginning (B) to the end (E) of a story. The different arcs represent different stories.

Kurt Vonnegut spent the second half of his life lecturing on an idea from the first.

Every story ever written, he said, has a shape. We follow a character moving between good fortune and ill fortune, from the beginning to the end of a story.

He named a few of these shapes and drew them on chalkboards in lecture halls across the US. The 2004 version is one of the few that survived on tape.

Vonnegut’s idea was originally for books, not films.
Computers, he said, could probably find the rest.

He didn’t live to do it. But analysis methods got cheap, screenplay archives went online, and his question turned into something a laptop could answer in an afternoon.

So I picked up where he left off and asked his question for Hollywood.
How many shapes are in Hollywood? And what do these shapes look like?

warm up

Here’s what one looks like as a line.

A single screenplay, page one to the end, plotted as the emotional pulse of its writing. A sentiment model reads it in twenty chunks and scores each one. The line rises with joy and falls with sorrow.

Highs and lows.

Peaks are …’s emotional highs. Valleys are its emotional lows. Keep scrolling to follow the pulse.

Watch where it turns.

Each dot is a turning point where the story visibly changes its mind. A great screenplay doesn’t need many of them, but it does need the right ones. Hover over the dots to watch the scenes.

zoom out

What happens if we do this to 1,627 film screenplays?

Every screenplay on Script Slug from 1980 to today. Same process, same grid. Vonnegut said a computer could find the common shapes. So I gave it a go with machine learning.

the answer

Six shapes. That’s it.

Six shapes came back for 1,627 screenplays. Reagan et al. found the same six for books back in 2016. They line up with what Vonnegut drew on the chalkboard, give or take. He named them. Click one to see what inspired their name.

one by one

Now, let’s look at one shape at a time.

The bold line is the average across hundreds of screenplays. The faint lines are the twenty films that hug the average closest. Same shape, different stories.

1 of 6

Oedipus

fall → rise → fall

across time

Has Hollywood’s grip on these six shapes changed?

Every film leans most on one of the six shapes. Sort all 1,627 of them by that shape, decade by decade. Has the mix Hollywood reaches for changed?

the steady state

The grip holds.

Every dot is a film, sorted by the shape it leans on most. Some decades have more films than others, but the proportions across shapes barely move. Hover over an archetype or a decade to compare across the grid.

and yet

If the shapes haven’t changed, why do today’s films feel different?

Data suggests it could be for a few reasons.

first reason

People talk a lot more now.

Horror and thrillers used to leave room for things to happen without anyone talking. Not anymore. By the 2010s, nearly every genre runs on dialogue, catching up to Romance, which was already nearly all talk. The scripts that used to sit at the quiet end like Friday the 13th and Body Heat don’t get written anymore. Hover for the specifics.

second reason

Things are more dramatic and less forgiving.

The genre mix has bent toward drama. Comedy has halved. Adventure has nearly vanished. But the bigger story sits inside the genres themselves. In Drama, Rags to Riches has collapsed. In Horror, Icarus has surged. In other words, we’re watching heavier movies now. When things go well, it doesn’t last. When they go badly, our heroes don’t recover.

your turn

Pick one. Watch its pulse.

Pick one. Or rummage through all 1,627. Dots mark the turning points.

Select a film

Hover a dot for the moment behind it

coda

Kurt Vonnegut was right. Stories have shapes.

And these shapes are equipment for living, kind of like practice runs for the future. The shapes haven’t changed.
But our heroes inside them aren’t recovering the way they used to. So what kind of future are we practicing for?

Read the methodology →

Methodology

How the lines were drawn.

Corpus

1,627 screenplays scraped from Script Slug using a modified version of the Movie Script Database pipeline (Saha, 2021). Final drafts only, 1980 to 2025. Titles and release years cross-checked against TMDb and IMDb.

Parsing

Each script is a PDF. Some are clean. Some are scans of scans. I pulled text out two ways: pypdf first, ocrmypdf for the ones that fought back. From there a rule-based parser from Ramakrishna et al. (2017) separates dialogue from action and character names from slug lines, so the script can be read by a machine the way it’s read by a person. I patched two edge cases the original parser misses.

Sentiment

A RoBERTa-based sentiment model reads each script in twenty overlapping windows scaled to its length. (I tried various window sizes before settling on this one.)Each window gets a signed score: positive class probability minus negative. String the scores together and you get the arc. Each script is z-scored within itself, so a quiet drama and a loud thriller sit on the same axis. The approach extends Reagan et al. (2016), who did this for 1,327 books in Project Gutenberg.

Clustering

First I tried sorting the arcs into hard buckets. Ward’s hierarchical clustering, HDBSCAN, DP-GMM. The arcs refused. UMAP showed they sit on a smooth continuum, not in clean groups. So instead of forcing them into boxes, I used Archetypal Analysis. It finds the six most extreme shapes in the corpus and lets every film sit on a mixture of them. Most films are blends. A few are pure. The six archetypes map onto the shapes Reagan et al. found in literary fiction: Oedipus, Icarus, Man in a Hole, Tragedy, Rags to Riches, Cinderella.

Turning points

Local highs and lows on a Savitzky-Golay smoothed version of the arc, filtered by prominence (threshold 0.5, tested across 0.3 to 0.7). Small wobbles get filtered out. Only the moments where the story visibly changes its mind survive. The notes on each turn are mine, written by hand, anchored to the page they happen on.

Caveats

Sentiment is not plot. A war film can read positive on the page if the dialogue is hopeful. These shapes are the emotional rhythm of the writing. Not the events.

The Shapesof Stories