The Theory of Narrative Selection

Sarah Perry is a contributing editor of Ribbonfarm.

Why do some stories become popular, retold for hundreds of years, while others are forgotten? Why do you see the particular stories that you see in your social media feed and on the news? How can we tell whether stories are true? And why are false stories so maddeningly popular?

Here I will look at stories as if they were biological organisms. Stories can’t reproduce themselves; they rely on humans for their survival and reproduction. In that sense, stories are symbiotic (or perhaps parasitic) in their relationship with humans. New stories are constantly being invented, using existing and novel devices and elements. They take as their subject matter factual happenings, imaginings, or both. They are transmitted and retold at different rates; most stories peter out and die, while a few sweep across the world in hours. They exhibit all the hallmarks for natural selection to act: variation, differential survival and reproduction, and heritability. Reproduction is complicated. Stories may transmit copies of themselves (reprintings, oral retellings), or they may transmit their traits to new generations of stories.

The following are a few of the mechanisms of narrative selection for all kinds of stories, from detective fiction to scientific papers.

Technological Progress

In “The Slaughterhouse of Literature,” Franco Moretti (2000) begins by wondering why Sherlock Holmes has been popular for a century, while Arthur Conan Doyle’s many contemporaries are all forgotten. Moretti examines the features of mystery stories published during the last decade of the nineteenth century, and makes a surprising discovery: most stories didn’t use clues at all. Some writers mentioned clues (“I wish we had a clue!”) but failed to actually introduce any. Of those that did use clues, many were not made visible to the reader, but rather were mentioned for the first time during the detective’s explanation of the solution. Of those that used visible clues, in most cases they had no relation to the solution. And of those who produced visible, relevant clues, including even Doyle, the clues were often not reasonably decodable by the reader. Each of these stages – clues, visibility, relevance, decodability – represents a technological innovation.

Doyle innovated the integration of clues into a coherent story, and as a result, his work is remembered while that of his contemporaries is forgotten. But it would not be until a generation later, with Agatha Christie, that the genre would take as a given the four-part structure.

I am reminded of a different story of cultural evolution: that of the development of the pintle-and-gudgeon rudder in the Mediterranean (Mott 1991). Side-mounted rudders (kind of like oars) were simple, flexible technology, and dominated the Mediterranean from the first century to the fourteenth. Rudders mounted to the stern with iron fittings (pintle-and-gudgeon) were innovated by northern shipbuilders, but not adopted in the Mediterranean (despite ample contact) for centuries. In order to offer any real advantage over the side-mounted rudder system, pintle-and-gudgeon rudders had to wait for vertical sternposts (and novel sails and rigging) to evolve. Only when all parts of the package were present did the new, high-tech solution become dominant.

Each trait, or feature, then, may not be adaptive on its own; only when all the pieces fit together does a technological advance go viral. As for rudders, so for clues. Moretti (2000) says,

Detective fiction, writes Todorov, is made of two separate stories (crime and investigation, past and present, fabula and sjuzhet), and these two stories “have no point in common.” Well, not quite: clues are precisely that point in common. An incredibly central position, where the past is suddenly in touch with the present; a hinge that joins the two halves together, turning the story into something more than the sum of its parts: a structure. And the tightening up starts a morphological virtuous circle that somehow improves every part of the story: if you are looking for clues, each sentence becomes “significant,” each character “interesting”; descriptions lose their inertia; all words become sharper, stranger.

p. 218, citation omitted.

Elements of Fit

In the academic study of narrative selection, a popular approach has been to identify individual elements that might explain the success of particular stories. One body of research concerns minimally counterintuitive entities and elements: narrative elements that are counterintuitive enough to be interesting and memorable, but not so counterintuitive as to lose coherence (for instance, a mouse that talks, but not a mouse that talks, is 500 feet high, and can control the weather). There was even a suggestion that the ideal story has between one and three counterintuitive elements (see, e.g., Stubbersfield et al., 2013).

Another element suspected to have adaptive benefits for stories is the elicitation of emotion. Heath et al. (2001) found that stories that evoked disgust were more likely to be shared, and the more disgusting, the better. Food contamination legends remain one of the most popular categories of legend – even more so if the horror is magnified by the incorporation of human body parts. Food contamination is extremely relevant to the survival interests of humans. But only a minority of stories evoke disgust.

Brewer et al. (1980) found that evoking the emotions of curiosity, suspense, and surprise were crucial to viewers enjoying stories, and perceiving them as stories. Evoking these emotions, however, is a function of the entire structure of the story, not an isolated element of a story. Stories must evoke suspense and then resolve it, or evoke curiosity and then surprise.

A story’s quality, memorability, and story-ness are more than the sum of its parts. Upal (2011) found that minimally counterintuitive elements only predicted the memorability of a story if those elements “contribute to the global cohesion of the overall story.” As with the pintle-and-gudgeon rudder, the fit between all the elements of a story, and not those separate elements themselves, underlies success and memorability. Upal says:

Comprehension researchers have identified global cohesion among the elements of a narrative as a key factor in making it memorable. Cohesion of a piece of text is defined as connections among the elements of the text. Cohesion is not just a function of the text itself but also of the background knowledge that the reader possesses. The connections that make a text more or less cohesive include co-references, as well as causal and logical connections among its elements. A text is better remembered if its constituents can be made coherent by the reader. Furthermore, the more effort a reader spends in making a text coherent, the more memorable the text….counterintuitive ideas are better remembered because they attract a reader’s attention by violating the reader’s expectations about what is to come next in the text. When a reader’s expectations are violated, he/she will attempt to resolve the situation by reasoning to justify the inclusion of expectation-violating information by invoking the relevant mental knowledge that the reader possesses. If this post-diction effort is successful, the expectation-violating concepts become richly linked to the reader’s existing mental representations, which were retrieved to explain the inconsistency to derive a coherent theme, and to the story theme itself.

p. 32, citations omitted.

The Gospel of Mark is the earliest of the Synoptic Gospels; it ends at the empty tomb of Jesus, in a state of suspense, with no resolution. The later gospels of Matthew and Luke add a resolution: Jesus returns to his disciples and explains things, and promises to return. The addition of a resolution may be a mutation that increases the fitness of the story in a rare way: by adding an element that ameliorates the fit between the story elements. The resolution is almost always incorporated into retellings of the Christ story, although of course Mark is preserved in the canon.

Surprise, Suspense, and Curiosity

Hoeken et al. (2000) suggest that three emotions are core to the success of a story: surprise, suspense, and curiosity. For Moretti (2000), curiosity is backwards-looking: the murdered corpse evokes curiosity as to what happened, and then this is resolved. Suspense is forward-looking: events that suggest danger occur, and then the suspense is resolved. Interestingly, the emotion of surprise can be elicited even if readers know how a story ends.

Surprise is crucial to Murray Davis’ (1971) taxonomy of philosophical interestingness. Davis lists twelve ways in which an argument can be considered interesting, each one a pair of opposites. Each of these is a different way of evoking surprise. For instance:

What seems to be a good phenomenon is really a bad phenomenon.
What seems to be a bad phenomenon is really a good phenomenon.

What seems to be a stable and unchanging phenomenon is really an unstable and changing phenomenon.
What seems to be a changing and unstable phenomenon is really a stable and unchanging phenomenon.

What seem to be assorted heterogeneous phenomena are in reality composed of a single element.
What seems to be a single phenomenon is in reality composed of assorted heterogeneous elements.

The underlying structure of interestingness is surprise: an assumption (what seems to be this) and a challenge (is actually that). A particularly popular type of story is the mythical myth: a story about what people once believed. Here, the surprise is backwards: we all know this, but once people believed that.

Stories about past myths might be particularly likely to be transmitted, even if false, because they provide surprise and novelty and at the same time credibility, a feeling of strengthening one’s memetic immune system even as one admits a false belief. Consider the iconic, simple story of the tulip bubble, a purported crazy belief of our gullible predecessors, and the more complex story that suggests that tulip buyers were not so irrational, and that it is perhaps wrong to say that there even was a bubble. The tulip bubble story is surprising, fits together well, and is easy to remember; Cochrane’s analysis is complex and requires us to incorporate concepts like futures contracts and the peculiarities of Dutch society. The tulip bubble may never have taken place, but the story of the tulip bubble seems to possess greater memetic fitness than the boring truth.

Ingroup and Outgroup

The emotions of surprise, curiosity, and suspense are important predictors of the success not only of fictional stories, but also of stories told as true. Television news to this day uses the suspense tactic of “teasing” a story before reporting on it (“Find out which product in your house could kill you!”). “Active shooters” and car chases in progress evoke suspense on the news as well as on Twitter. And there are thousands of more subtle ways of evoking curiosity than “one weird trick.”

I suspect that there is another important feature of narrative success: establishing the goodness of the ingroup and the badness of the outgroup. The same story may be told in different ways for different audiences, to emphasize this boundary. The category of legends known as “subversion myths” refers to myths that some outgroup is plotting against us, is responsible for our troubles, and possibly eats our babies in Satanic rituals. Today, the ingroups and outgroups evoked are primarily political. Friggeri et al. (2014) found that by far the most common category of rumor shared on Facebook was the political rumor. Stories are constantly being selected to have the “right” villain, the “right” victim, and the “right” structure of events to confirm and inflame the outgroup-hatred of any given ingroup.

False vs. True

Why are false stories so popular? Are false stories more popular than true stories?

Friggeri et al. (2014), studying rumor cascades on Facebook, found that while only 45% of the rumors covered on Snopes.com (the famous urban legend debunking site) are false, 62% of the rumors shared on Facebook were false. 26% of the stories on Snopes are labeled true, while only 9% of the rumors shared on facebook were true. False stories seem to enjoy an advantage at the outset.

Interestingly, however, the few true stories that were shared had longer “rumor cascades” of subsequent shares than false stories (an average of 163 shares per upload for true stories, vs. 108 for false stories). Unfortunately, being “Snoped” (a user commenting on the rumor with a link to Snopes) does not seem to hinder virality much; over half of the shares of false and mixed items occurred after the item was Snoped.

If the keys to success of a story are to evoke the emotions of curiosity, surprise, and suspense, as well as disgust or ingroup-outgroup hatred, and for its parts to fit together into a coherent structure to produce these emotions, then the truth is limited. A true story is heavily constrained: it must have a source in events in reality, and its elements may not change.

False stories, on the other hand, are unconstrained: they may change the order of elements, add and subtract elements, and invent entirely new elements. They may have some or no basis in fact. They may attach indications of credibility to themselves (using named entities, references to particular times and places, etc.) and claim trustworthy sources. The space of false stories is larger and more varied than the space of true stories. The best stories, then, are likely to be false – contrary to the maxim that truth is stranger than fiction.

In the criminological study of deception, one result is that impromptu lies (as from a suspect being questioned) lack details and do not form a coherent story. Ordinary truths have detail and form a coherent story, but usually do not produce much emotion or surprise. Carefully-crafted (or -evolved) lies, however, can provide a supernormal stimulus of truth: their parts are so well-selected and fit together so well that they produce emotion better than the truth.

If all the parts of a story work together perfectly, with nothing extraneous, and if the story produces strong emotion, we should be extra careful to check its veracity. The maxim “too good to be true” applies.

Horizontal and Vertical Transmission

In biological science, vertical transmission of a parasite (or symbiote) is transmission from one generation to the next, as from parent to child. Horizontal transmission is transmission between organisms of similar ages. When parasites evolve under vertical transmission, they tend to be gentler on their hosts: less contagious, and less virulent, even beneficial (Stewart et al., 2005). But when parasites evolve under horizontal transmission, they can be more virulent and contagious at the expense of their hosts, whose fitness is now not so relevant to the parasite’s survival.

Some stories and practices are transmitted vertically; these probably have the most fitness benefits, and the least cost. Horizontally transmitted stories and practices can afford to be more harmful to us.

Olivier (2015) suggests that children’s culture is horizontally transmitted: from child to child, rather than from parent to child. In my previous essay on children’s dares, I noted that dares exact a major fitness cost. Risky practices culturally transmitted between children may survive in spite of this cost specifically because they are horizontally transmitted.

Most stories in the age of mass media and the internet are horizontally transmitted. Few stories hurt us. But to the extent that stories can harm us, we would expect “viral” horizontally-transmitted stories to be more harmful than stories passed between generations.

Science and Narrative Selection

I have cited many scientific papers here. Most of them are probably false.

Smaldino et al. (2016), in “The Natural Selection of Bad Science,” analyzed 44 review papers describing published papers in the behavioral sciences over the past sixty years. They found that the statistical power of the studies to detect small effects averaged only about .24 – that is, if the researchers picked correct hypotheses 100% of the time, they should fail to get a positive result three out of four times. The problem of underpowered studies has long been noted in psychology, but Smaldino et al. found that the power of studies did not change over time: science got no better. (Of course, positive results remain much more common than 25%.)

The authors hypothesize that since positive results are so important for career success, science labs that produce positive results – even false ones – will be more successful than labs that are more rigorous and produce fewer positive results. In turn, bad (but successful) methods will be reproduced by the next “generation” of labs.

Science is not immune to narrative selection. Positive results are surprising and make a good story. Counterintuitive results are ideal. Replications are rare, and failed replications (and even revelation of actual fraud) often have little effect on a theory’s memetic success.

An Acquired Memetic Immune System?

How can we protect ourselves from false stories? How can we tell the difference between true and false stories?

Friggeri et al. (2014) examined the role of Snopes.com, a website dedicated to cataloging and establishing the truth or falsity of rumors and legends, in memetic immunity. It seems the effectiveness was limited. Shall we teach “critical thinking” and awareness of sources in order to immunize ourselves against falling for false stories?

An acquired immune system might not do much good in the presence of constantly-evolving antagonists. Hedrick (2004) suggested that invertebrates, which don’t have an acquired immune system but only an innate immune system, experience no more burden from disease than vertebrates. The acquired immune system of vertebrates has functioned like antibiotics, selecting new strains specifically adapted to undermine it – often using it against itself.

We do have something like an acquired memetic immune system. We are suspicious of certain kinds of stories and sources based on past experience; certain features trigger skepticism. But stories are constantly being adapted to get around our defenses. Those “mythical myths” mentioned above specifically hijack our memetic immune system, sneaking in false stories under the guise of feeling less gullible than people in the past.

The best we can do is to hold our beliefs lightly, realizing that it is difficult to tell truth from fiction. Preventing harm from beliefs is more important than preventing false beliefs as such. We should enjoy and study stories, and science, without expecting them to be true a majority of the time. Epistemic caution is important. But so is fun.

---

Works Cited

Brewer, William F., and Edward H. Lichtenstein (1980). “Event schemas, story schemas, and story grammars.” Center for the Study of Reading Technical Report; no. 197.

Davis, Murray S. (1971). “That’s interesting: Towards a phenomenology of sociology and a sociology of phenomenology.” Philosophy of the social sciences 1:4, 309.

Friggeri, Adrien, Lada A. Adamic, Dean Eckles, and Justin Cheng (2014). “Rumor Cascades.” ICWSM.

Heath, Chip, Chris Bell, and Emily Sternberg (2001). Emotional selection in memes: The case of urban legends. Journal of Personality and Social Psychology 81:6, 1028-1041.

Hedrick, Stephen M. (2004). “The acquired immune system: a vantage from beneath.” Immunity 21:5, 607-615. h/t to Adam Sandberg for finding this paper and suggesting that costly ritual is a memetic immune system.

Hoeken, Hans, and Mario van Vliet (2000). “Suspense, curiosity, and surprise: How discourse structure influences the affective and cognitive processing of a story.” Poetics 27:4, 277-286

Moretti, Franco (2000). The Slaughterhouse of Literature. Modern Language Quarterly 61:1, 207-227.

Morin, Olivier (2015). How Traditions Live and Die. Oxford University Press.

Mott, Lawrence V. (1991). The Development of the Rudder, A.D. 100-1600: A Technological Tale (Master’s Thesis).

Smaldino, Paul E., and Richard McElreath (2016). The Natural Selection of Bad Science. (Under review.) arXiv.

Stewart, Andrew D., John M. Logsdon, and Steven E. Kelley (2005). “An empirical study of the evolution of virulence under both horizontal and vertical transmission.” Evolution 59:4, 730-739.

Stubbersfield, Joseph, and, Jamshid Tehrani (2013). Expect the Unexpected? Testing for Minimally Counterintuitive (MCI) Bias in the Transmission of Contemporary Legends: A Computational Phylogenetic Approach. Social Science Computer Review 31: 90–102.

Upal, M. Afzal (2011). “Memory, Mystery and Coherence: Does the Presence of 23 Counterintuitive Concepts Predict Cultural Success of a Narrative?” Journal of Cognition and Culture 11:1, 23-48.