Measuring the Size of a Story

After considering Gail Carmichael's post about working on her Ph.D. (story coherence in computer games) and trying to keep up with the literature, I wondered if there was any academic literature about the related problem of measuring the size of a story. This also ties in with my Story Points Ethical Calculus, where the meaning of life is measured by how it contributes to the size of the story of the universe.

Commercial writing is measured by a word count. Though is that useful for describing how complex the story is? Canadian CBC radio story teller Stuart McLean is notable for extending a short snippet of story (such as "The Fig Tree" about burying a fig tree to preserve it over winter) into a long tale, with tremendous amounts of background detail about everything and everybody involved and their histories (the owner worked in a furniture factory 50 years ago doing...), and ending up being more about getting old and the hope of springtime. So, is word count relevant? To some extent, but the number of words used by different writers can vary wildly. For example, a six word story: "For sale: baby shoes, never worn" also conveys a lot. BookFox has a nice article looking at short story sizes of different authors - it varies quite a bit, even for the same author. BookFox doesn't say much about complexity, other than longer stories can have more complex plots and character development.

The movie business has ways of estimating the cost to produce a script. One summary from uses script pages per day as one of their metrics, with multipliers for actors moving around in complex patterns or for special effects, then extra costs for days of actor's salaries (based on the number of pages the actor is in and various multiplier factors) and so on. It starts with essentially a word count but then moves to the cost of implementing those words.

Some other measurements about stories include a Narrative Engageability Scale, and more commercially, many ways of measuring how engaging an advertisement is to a target audience. But those count number of messages seen or the impact on an audience, not the size or complexity of the messages.

You could also measure a story by figuring out the plot structure in some sort of abstract form. Maybe extract a graph of plot points and character interactions from a story. It isn't mathematically elegant, but TV Tropes sort of does that by collecting the base patterns used in stories. You could use tropes to measure a story, by finding all matching tropes and counting them up. Possibly the structure of the story could be summarised by finding the structure of the tropes (sequential ordering, a graph of interconnections). Then the resulting data, along with unique descriptions from the story (such as names and descriptions of things that are exceptional) can be used as a lossy compression of the story. From that graph of tropes and specific data, you could reconstruct the original story in an approximate way.

Is there something like Scale-invariant feature transforms but for finding tropes in stories rather than objects in pictures? Instead of doing geometrical transformations of objects, you would mangle words in some way (a thesaurus springs to mind) to go between the actual story element and the closest trope.

Following a line of thought about feature extraction turns up a paper titled Towards Automatically Extracting Story Graphs from Natural Language Stories. It even uses something like tropes to classify characters and objects as female character, anthropomorphic animal character, object, location, and so on. Though I wonder if it would be useful to use the full richness of the collected TV Tropes to classify things. Anyway, that's just a small part of the overall system, which spits out a story graph at the end (though they still have work to do on including time and unifying references to a person better - that guy, him, Joe could all be the same person).

The Association for the Advancement of Artificial Intelligence has several other story graph papers - a motherlode of information. I can see why Gail has trouble finding time to read it all! Story Graphs are used to encode characters, props, locations, and interactions between them. Perhaps we could use the size of a story graph as a way of measuring the story's size and complexity.

There's also an enhanced version called Story Intention Graph (SIG) which adds what the characters are thinking (their model of the world) and their motivation. Detecting Story Analogies from Annotations of Time, Action and Agency describes SIG and uses it to find common parts of stories. When I see that, I think data compression!

Interesting ideas. But for #StoryPointsEC the story size evaluations are a bit too lossy and Human oriented (particularly tropes and story graphs) to be able to describe the universe. I still think something like lossy video compression (but for all senses) would be more useful. Though that implies that the audience sets what is interesting. So viewed through an audience of entirely TV watchers, compressing by Tropes would make sense.

Copyright © 2020 by Alexander G. M. Smith. Feel free to quote portions of this essay if you want to continue discussion on the topic.

$Id: 20200418MeasuringStorySize.html,v 1.5 2020/04/18 23:24:27 agmsmith Exp $