Theses Doctoral

Found Things: Variations in information density in long-form narrative

Bohannon, Catherine Ridder

This dissertation makes the case that treating digitized corpora of literary works as cognitive artifacts can provide particular insight into how the reading mind apprehends events within an imagined world and, thereby, provide potentially useful functional models for event perception, emotional memory, and determining what’s “real.” Most essentially, it will make the case that the deepest feature of narrative cognition may involve an “information distribution” assessment, wherein the variation of information density over time cues the mind to attend to denser events with increased attention, potentially saving more of their content for long-term memory. This mimics what cognitive research has frequently established for real-world processing of emotionally stimulating events, wherein emotional memory tends to be better retained over time, with more detail, fewer conflations, and more resistance to fading, while neutral events tend to be relegated to gist or forgotten.

Put together, this produces an ordering of autobiographical memory that resembles a glimmering string of pearls: densely detailed memories strung together over time, separated by thinner, looser memories and gist, with a particular cluster of these “pearls” towards the middle for the memory bump of the mid-teens to mid-twenties. While many have argued for larger schemas or socially influenced self-regard as the major driver for memory emphasis in one’s Life Story, if autobiographical memory is anything like a novel, it may prove a bit simpler: most of the bigger pearls mark where one’s sensory array “dilated” in moments of arousal, and their lustrous, persistent “shine” may be a matter of how likely it was that one returned to those memories over time.

Chapter 1 examines what we do and don’t know about the reading mind, settling on a narrower definition of immersive narrative reading as an exceptional cognitive state which moves in and out of what cognitive psychologists call “flow” and a more passive, vivid “daydream.” This is an inherently unstable activity that requires a great deal of assistance from the text, thereby providing useful targets of analysis for researchers interested in perception, emotion, and memory, with a particular eye towards embodied cognition. It then discusses key gaps in the scientific literature and literary scholarship around event perception and narrative cognition, some of which this project aims to partially fill through quantitative analysis of literary texts. This chapter will also discuss the promise and perils of treating literary corpora like the novels in Project Gutenberg as cognitive artifacts: the known limitations of using “canon” texts as a representative sample of literature in general, the rarity of reading, and what it means to “backsolve” cognition through its artifacts.

Chapter 2 describes a series of experiments conducted on a corpus of a few thousand novels and nonfiction narratives contained in Project Gutenberg and the Nickels and Dimes Project. Leaning on the “string of pearls” metaphor for autobiographical memory organization, this chapter will promote a model of long-form narrative’s fundamental mnemonics as something that mimics that organizational pattern: information density that varies over time, predicting not only the pace of in-narrative time passing, but which “moments” or features of the narrative will be important for the reader to remember over multiple reading events, while others will be forgotten or relegated to gist. This pattern closely mimics models of autobiographical memory in cognitive psychology, not only of so-called “flashbulb memory” or surprising, high-affect events, but also of Life Story in general: vast periods of fleeting detail, with dense memory clusters around events that were encoded in moments of arousal, with curious memory affects just before and after those events, possibility illustrating what Jefferey Zacks presents as a “gating” model of event perception.

Drawing on the scientific literature on event segmentation, arousal and memory, and time perception, and likewise drawing on literary scholarship on time and stylistics in the novel, this chapter will explore the implications and limitations of using POS tagging to try and tease out quantifiable units of “information” from large corpora of novels utilizing one-way repeated measures MANOVA. Applications for these findings in literary scholarship will be discussed throughout—for instance, while scenes involving sex or violence are predictably information-dense in most texts in the corpus that were hand-scored for accuracy (and subsequently used as training texts for the algorithm), in-book variation from the norm and from nearby passages is more predictive than a raw density score alone. For example, when Stephen Dedalus has sex in Portrait of the Artist as a Young Man, imagistic detail goes down compared to nearby scenes and compared to the more detail-dense passages in the text, which seems to be typical of Joyce: while he does vary density according to temporality and that maps roughly to “significant” scenes, the most emotional scenes tend to be written more sparely (spare for that author, that is—Joyce is not Hemingway). That may be an authorial quirk, or it may be that he relies upon a second strategy to stimulate a reader’s emotional response: semantic content that’s normally cued to a strong negative or positive valence.

Chapter 3 will attend to the ways some authors resist narrative’s “ease of use” in order to prompt their readers to interrogate what’s Real. This chapter zooms in on a specific period of American and British literature, and a genre within that brief time: the rise of Creative Nonfiction and/or New Journalism, with a close read or “case study” of George Orwell’s Homage to Catalonia and James Agee’s Let Us Now Praise Famous Men. This chapter proposes that the authors set out to create narratives that would reflect the “real” lives of their subjects, with an objective of making those lives feel real to their readership. But were they successful? Drawing on cognitive psychology research in psychosis, metacognition, and temporal sense, this chapter aims to elucidate how literary narratives like these may “aim to fail” at certain features of deep narrative form (as discussed in the prior chapters) in order to “startle” their readers into a less passive state, in order to better mimic the qualia of witnessing something in the real world, and thereby produce a sense that the subjects within the text are Real. These embedded structural failures are often more subtle than anything Brechtian, but nevertheless can be found both quantitatively and in close reading, which may indicate that when a long-form narrative text purposefully aims to make a reader uncomfortably aware of Reality--especially when motivated by known, deep ethical concerns--it may “work” in ways that have less to do with the subject or content of the text and more to do with form.


This item is currently under embargo. It will be available starting 2027-05-11.

More About This Work

Academic Units
English and Comparative Literature
Thesis Advisors
Dames, Nicholas J.
Tenen, Dennis Y.
Ph.D., Columbia University
Published Here
May 25, 2022