The Darwin of the Classics. And AI?
He was called the “Darwin of the Classics.” His name was Milman Parry.
In 1952, classical scholar H.T. Wade-Gery wrote:
“As Darwin seemed to many to have removed the finger of God from the creation of the world and of man, so Milman Parry has seemed to some to remove the creative poet from the Iliad and Odyssey.”[1]
Just as Darwin showed that life emerges through biological evolution rather than divine fiat, Parry demonstrated that epic poetry emerges through narrative evolution—traditional oral composition across generations—rather than lone effort.
He solved the Homeric Question which had dogged scholars for centuries: what was the genesis of the Iliad and the Odyssey? In doing so, he discovered something he could not have imagined: the fundamental principles by which artificial intelligence would one day compose language.
The Journey Begins
Parry’s journey began in Oakland, California. His father was a local pharmacist. He lost his mother to cancer when he was a teenager.[2]
In 1919, a diligent student, he enrolled at Berkeley—the only campus in the University of California system until that year, when UCLA was founded. He earned bachelor’s and master’s degrees in classics under some of the leading scholars of the time. The Bay Area air must have inspired him. His master’s thesis exploded centuries of Homeric scholarship.
The title was formidable: “A Comparative Study of Diction as One of the Elements of Style in Early Greek Epic Poetry.” But the core insight was elegant. Homer’s poetry was built on formulas, repeated phrases like “gray-eyed Athena” and “wine-dark sea,” chosen not for poetic novelty but to fit the rhythmic backbone of Greek epic: dactylic hexameter. These phrases, dubbed “epithets” by Parry, were familiar to poet and audience alike. They were the essential building blocks of oral composition, not the unpredictable sparks of individual creativity.
To the Mountains
After Berkeley, Parry completed his doctorate in Paris at the Sorbonne, writing his dissertation in French, as required, with the obligatory two-part structure. In English, the two parts were “The Traditional Epithet in Homer” and “Homeric Formulae and Homeric Meter.”[3]
To Parry, epithets were the repeated adjective-noun combinations that distinguished the Iliad and the Odyssey, such as “swift-footed Achilles” or “god-like Odysseus.” Formulas were broader: “a group of words which is regularly employed under the same metrical conditions to express a given essential idea.”[4]
Parry demonstrated that these devices weren’t stylistic flourishes: they were necessary tools for real-time improvisation. And if Homer depended so heavily on them, the poems couldn’t be the invention of a single literate genius. They must have evolved over centuries within an oral tradition. His dissertation provided statistical tables of epithet frequency and formula applications as confirmation.
But this was still theory on paper. Parry needed living proof. Where could he find oral poets still composing in performance?
The answer arrived at his Soutenance, or dissertation defense, on May 31, 1928. At the Sorbonne, such defenses were ritualized into scholarly theatre. The date and time were published. A jury and audience assembled in a great hall. Matija Murko, a Slavic scholar from Prague, was among the attendees.
Murko was visiting Paris to lecture on Yugoslav epic poetry. His subject: the guslari, among the last of the traditional epic poets. Illiterate mountain bards, still composing in real-time performance. They played the gusle, a single-stringed instrument, while singing epic narratives that could stretch across hours, even days.
Murko was what mythologists call a threshold guardian, a figure who appears at the moment when the hero is ready but unsure of his destination. Parry had his theory. Murko showed him where to find the proof.
In the early 1930s, Parry set out to crisscross the rugged mountains of Yugoslavia. Primitive recording equipment. Precarious cliff roads. Snow-blocked passages. He was chasing an answer: How did ancient poets create 27,803 lines of the finest poetry humanity has ever produced—the Iliad and the Odyssey—without writing, without memorization, in real-time performance?
In remote villages, he captured the songs of illiterate guslari improvising legends and tales.
Parry was in a race against time. The Balkans had endured ever-changing borders, foreign invasion, tribal conflict. The Balkan Wars provided the tinder for the First World War. The Second World War and decades of sectarian violence were not far off.
Murko had already delivered the diagnosis: the singers were aging into the mountain dirt, the audiences thinning, the feudal life that sustained them gone. Parry walked into Yugoslavia in 1933 knowing he was documenting something that modernity was slowly extinguishing from the face of the earth.
What Parry captured in those mountains may prove essential to understanding the AI systems we are building today.
The Discovery: How Oral Poets Actually Compose
In his mountain journey, Parry discovered composition in performance: singers creating thousands of lines in real time, never the same twice, yet wholly traditional. As a narrative anthropologist, Parry was disciplined, and he systematically recorded his discoveries as evidence and for analysis.
The recordings themselves were a hill to climb. In Parry’s first trip to Yugoslavia, in 1933, he tested a Parlograph, a short-play wax cylinder device. (Parlophone, the affiliated record label, would sign the future Beatles.) The Parlograph was designed to replace manual dictation, and users could shave the wax down to reuse the cylinders. With a recording capacity of about two minutes, and scratchy, screeching recordings, Parry’s frustration must have mounted quickly. His team manually transcribed live performances. Even Kafka wrote about the Parlograph.[5]
For his second, longer expedition in 1934–1935, Parry commissioned a custom-built, battery-powered electrical recording apparatus from the Sound Specialties Company in Connecticut.[6] This innovative machine utilized two turntables and an amplifier to record directly onto aluminum discs. Each disc could capture about four-and-a-half minutes of material. The trick was in the toggle. As one disc filled, the user could switch to the next. This allowed for continuous, uninterrupted recording of long epic songs, overcoming the limitations of the earlier, shorter-playing wax cylinder.
The expedition resulted in over 3,500 double-sided aluminum discs and hundreds of notebooks of transcribed and dictated texts. This archive documented a systematic compositional mechanism operating at multiple levels.
First, epithets. Parry’s breakthrough came from analyzing Homeric Greek and discovering that the poetry was built from formulaic building blocks that fit specific positions in the verse line. Poets regularly employed groups of words under the same metrical conditions to express essential ideas.
For Heroes…
· “Swift-footed Achilles” (πόδας ὠκὺς Ἀχιλλεύς)
· “God-like Odysseus” (δῖος Ὀδυσσεύς)
· “Hector, breaker of horses” (Ἕκτωρ ἱππόδαμος)
· “Wide-ruling Agamemnon” (κρείων Ἀγαμέμνων)
For Deities…
· “Goddess, gray-eyed Athena” (θεὰ γλαυκῶπις Ἀθήνη)
· “White-armed Hera” (λευκώλενος Ἥρη)
Places and Objects…
· “Rosy-fingered dawn” (ῥοδοδάκτυλος Ἠώς)
· “Wine-dark sea” (οἶνοψ πόντος)
· “Swift ships” (θoὰς νῆας)
The “swift ships” epithet was used even when ships were drawn up on land. This proved Parry’s point: the formula was chosen to fit the meter, not the meaning. Utility first.
These weren’t merely repeated phrases or poetic ornaments. For any hero, a formula. For any time of day, a formula. For any common action—arming for battle, preparing a feast, launching ships—traditional sequences of formulas.
The genius was in the system itself. Over generations, singers had refined a vast inventory of formulaic options, each fitting different metrical needs. When composing, the singer selected from this inventory in real time, choosing formulas that both expressed the needed meaning and fit the metrical pattern.
Second, thematic structures. Beyond individual formulas, Parry and his successors identified larger compositional units: themes. These were traditional narrative patterns: story sequences that provided scaffolding for coherent storytelling. The arming of the hero. The council scene. The journey by sea. The recognition of the disguised wanderer.
Themes weren’t fixed scripts but flexible frameworks. A singer knew the elements that belonged in an arming scene, for instance, but could elaborate or compress based on the narrative moment. The thematic structure ensured coherence across thousands of lines while allowing creative variation within traditional bounds.
Third, composition in performance. This was Parry’s most radical insight: oral poets don’t compose, then perform. They compose in performance. There is no fixed text “behind” the performance. Each singing is an original act of composition using traditional materials.
The singer maintains narrative momentum, selects from formulaic options in real time, responds to audience and context, and creates something that never existed before yet is fully “traditional.” As Parry’s assistant and protégé, Albert Lord, would later write:
“The singer of tales is at once the tradition and an individual creator.”[7]
This discovery—that sophisticated poetry could be composed without literacy, without memorization, through systematic use of traditional patterns—transformed literary theory forever.
It may also transform the future we build with AI.
The Legendary Singer
The Tara River gorge is the deepest in Europe, the Grand Canyon of the continent. Mountains jostle the landscape, including the ominous-sounding Lovćen, the Black Mountain that gave Montenegro its name. Glacial lakes punctuate the terrain. Gingerbread medieval towns overlook the Adriatic Sea. An epic landscape that birthed epic poets.
Parry’s explorations of the Balkans took him from Dubrovnik to areas that are now Serbia, Bosnia and Herzegovina, Macedonia, and Montenegro. In his wanderings, he recorded over 100 singers. Some were merely facile, others gifted. Near the end of his 1935 expedition, he reached the region around Bijelo Polje along the Lim River. The villagers said he must meet a poet from the nearby village of Obrov: a local farmer, former butcher, and one-time soldier.
Was it serendipity? The gods? Parry encountered Avdo Međedović, a legendary guslar, later dubbed by Lord “the last of the truly great epic singers of Balkan Slavic tradition.”[8] This illiterate farmer and butcher from the mountains could compose grand epics. He recounted a repertoire of 58 songs, nine recorded by Parry, four dictated for the notebooks. Avdo learned his craft from his father.
Parry contrived a test for Avdo. Another guslar performed a song of several thousand lines. After listening once to a song he had never heard, Avdo performed it back. Same story, same characters, same essential narrative—but different words, different length, different elaborations. With Avdo, the song swelled: three times longer, with a richness the prior performance lacked. Avdo gave the song a forlorn hero: “his heart wilted like a rose in the hands of a rude bachelor.”[9]
Then came one for the records: “Osmanbey and Pavičević Luka.” Avdo delivered this epic across multiple sessions, two hours in the morning and two hours in the afternoon, day after day, with pauses to rest his voice, and a longer break when his voice ran out. At its conclusion, the epic exceeded 13,000 lines—longer than the Odyssey and chasing the length of the Iliad. Homer had materialized before Parry’s eyes.
Avdo was among the last epic poets recorded by Parry.
Avdo was composing in performance, selecting from his vast internalized inventory of formulas and themes, generating the song anew each time while remaining wholly within the tradition.
Today, you can see Avdo’s statue in the Park of Poets in Bijelo Polje, Montenegro, on a terrace overlooking the Lim River. His stone figure wears a fez and bears a gusle. Like Homer, Avdo could not write. But his songs never die. Thus, his work may be worthy of the Nobel Prize in Literature.
Go. Pay homage.
The Tragic End
Then, at age 33, Milman Parry died in a gun accident. A bullet nicked his heart.
The circumstances remain contested. In his 2021 biography Hearing Homer’s Song: The Brief Life and Big Idea of Milman Parry, Robert Kanigel explores the competing theories: accident, suicide, even whispers of something darker. The death was ruled accidental, but questions persist. What’s certain is the tragic timing: Parry died just as his work was taking flight, right when the implications of his discovery were becoming clear.
He left behind aluminum recording discs capturing the voices of singers who would soon be silenced by war and modernization, and a revolutionary theory that would reshape how we understand human creativity.
The burden of completing his work fell to his 23-year-old assistant.
The Protégé and Professor: Lord’s 50 Years
Parry’s discoveries endured and flourished through his protégé, Albert Lord, who became my teacher at Harvard.
I earned my degree from the Committee on Degrees in Folklore & Mythology, the oldest undergraduate program of its kind in the U.S.,[10] honoring a tradition of folk studies at Harvard dating to the 1850s. Lord was its first chair, a title he held until his retirement in 1983.
Lord spent more than 50 years honoring Parry’s discoveries, capped by his masterwork: The Singer of Tales, published in 1960. In it, Lord systematized the patterns Parry had found and demonstrated that they applied not just to Homer but to oral traditions worldwide, from medieval epic to Biblical poetry to African praise songs.
He returned to Yugoslavia in 1950–1951, revisiting Parry’s collecting sites and recording on magnetic wire. During this trip, Lord reunited with Avdo, then an old man. He recorded how his recitations evolved across a lifetime. I imagine he also said thank you and goodbye, though the words may not have been spoken. Avdo passed away a few years later.
Professor Lord spent decades analyzing the recordings, publishing volumes of Serbo-Croatian folk songs and heroic songs, and training generations of scholars (and late bloomers like myself) in oral-formulaic theory.
I studied under Lord and David Bynum, who continued Lord’s work and expanded it comparatively across cultures. Bynum served as Curator of the Milman Parry Collection and deepened our understanding of how narrative patterns operate across traditions.
While drafting my undergraduate thesis, I would meet with Bynum in his offices at the Parry Collection, too young to realize the significance of the location. I received direct transmission of a scholarly lineage running back through Lord to Parry to the guslari themselves.
The Patterns Return
These patterns, discovered in Yugoslav mountains in the 1930s, have reappeared in an unexpected place.
Today, there is a new generation of aspiring epic poets. They are recreating something akin to these timeless compositional patterns, not through centuries of tradition, but through statistical brute force. They are called ChatGPT, Claude, Gemini, Perplexity, and DeepSeek. Others, such as Meta AI, Copilot, and Grok, are embedded invisibly into our online world. They are what I call Singers of Code.
Like the guslari, these AI systems don’t retrieve pre-written text. They compose in performance: generating language token by token, selecting from learned patterns, creating outputs that are simultaneously novel and formulaic. They never produce the same response twice, yet their outputs remain (largely) coherent and (mostly) appropriate to context.
But here’s the difference: Avdo trained for decades under masters. He absorbed a tradition refined over generations. His formulas were time-tested. His themes were forged by life on the anvil of time and generations. His audiences demanded excellence, and the singers who couldn’t deliver were forgotten.
The Singers of Code, on the other hand, trained on the internet. They learned from:
· Shakespeare and … spam
· Peer-reviewed journals and … Reddit threads
· Tolstoy and … clickbait.
Their training corpus was not curated by tradition—it was scraped indiscriminately. And it shows. The result is a strange new phenomenon and spreading epidemic, called AI slop:
· Polished but vacant: “It’s important to note that…”
· Motion without movement: “Let’s dive in…”
· Template where thought should be: “In today’s fast-paced world…”
· Formula masquerading as insight: “It’s not X. It’s Y.”
· Unearned landing: “So there you have it!”
These are the debased formulas of a tradition trained on noise.
The symptoms continue. The once mighty em-dash, undermined by AI overuse. The powerful “Rule of three” diminished by bot excess: so many groups of three nouns, adjectives, and phrases. The Rule of Three is proof of oral-formulaic composition: a pattern that works, passed down, reused. But without curation, without taste, without knowing when to deploy it: parody. What once gave us “Friends, Romans, countrymen” now gives us “clear, concise, and compelling.”
Avdo knew when to reach for the rose. The bots reach for threes too often.
And it gets worse. Unsophisticated human prompting sinks to the quality of lesser training data. Lazy inputs yield lazy outputs. Vague requests produce vague responses. The singer performs to the level of the audience —and most audiences aren’t asking for Homer.
Is this a failure of technology? Of systems? Of ignorant design presumed to be intelligent?
This is language adrift, not anchored by the rigor of tradition.
The guslari had masters, apprenticeships, audiences who knew the difference between a great singer and a mediocre one. They had cultural memory. They had standards.
We have none of this. Not yet.
The patterns are returning. But the tradition that refines them—the curation, the mastery, the demand for excellence—has not. That is what we must build.
Every student of LLMs will say to themselves: “Oh crap. Look what we missed.”
How LLMs Actually Work: The Oral-Formulaic Parallel
Large language models generate text through a process that echoes oral composition:
Token-by-token generation = composition in performance. When an LLM generates a response, it doesn’t retrieve a pre-written answer from memory. It composes in real time, selecting each token (word, sub-word, or character unit) based on learned probability distributions. Just as an oral poet selects the next formula based on metrical needs and narrative context, an LLM selects the next token based on the preceding sequence and its training patterns.
Each generation is unique. The same prompt produces different outputs, not through randomness but through the compositional process operating within learned constraints. Creation anew each time, drawing from internalized patterns.
Learned patterns = formulas and themes. LLMs learn statistical patterns from training data: recurring sequences of tokens that appear together under specific conditions. Functional patterns, not mere memorized phrases—ways of expressing ideas that fit particular contexts.
“In conclusion…” signals ending. “On the one hand… on the other hand…” structures contrast. “The plaintiff argues that…” fits legal discourse. These are formulaic patterns: slots that can be filled with varying content while maintaining structural function, exactly like Homeric formulas.
At larger scales, LLMs learn thematic structures: how stories are typically organized, how arguments are constructed, how explanations unfold. The business email format. The scientific paper structure (IMRAD: Introduction, Methods, Results, and Discussion). The hero’s journey identified by Joseph Campbell. These themes provide coherence across longer sequences, just as oral poets use traditional narrative patterns to maintain story structure.
Training on massive corpora = learning tradition. Just as oral singers internalize traditional patterns through years of exposure to other singers, LLMs internalize patterns through training on billions of words of human language. The training corpus is the tradition: the collective patterns of how humans use language. The model learns which words appear together, which structures work in which contexts, which patterns are traditional.
When the model generates text, it instantiates the tradition, composing in a way that sounds “right” because it conforms to learned patterns of human language. Like oral singers, the model selects from traditional options.
Why This Matters: The AI Stakes
Understanding how the ancient guslari composed may be more critical than ever: their dynamics may hold the key to building AI systems that are efficient, interpretable, and aligned with human values.
Current AI development treats language models as statistical pattern matchers operating on massive datasets through sheer computational power. Train on everything, scale up compute, hope for the best. Brute force: throw more data, more parameters, more GPU hours at the problem until something emerges that works.
What is the result of brute force on training libraries that combine the sublime and the toxic? That’s why RLHF (Reinforcement Learning from Human Feedback) exists—to cull out the verbal sewage. But it’s fragile, expensive, and brutal on the workers who bear the burden.
Perhaps we’re missing something fundamental.
What if 90 years of empirical research on human compositional systems—research that began with Milman Parry—offers a better architectural foundation?
The implications cascade:
First, efficiency. Avdo Međedović’s brain operated on roughly 20 watts of biological power and composed 13,000 coherent lines in marathon sessions. GPT’s training required tens of millions of dollars in electricity, billions in capital investment, and massive data centers. AI’s global electricity demands are counted in hundreds of terawatt-hours. If we understood the principles of composition—how formula and theme actually work as mechanisms—could we build systems that compose more efficiently? Smaller models, less training data, more elegant architecture based on understanding rather than brute force.
Second, interpretability. Current AI systems are black boxes. We can’t effectively inspect or control the decision process. But oral-formulaic composition is transparent by structure: you can see which formulas are being selected, which themes are active, which traditional patterns are in play. An architecture designed from these principles would be interpretable by design—a compositional system we can observe, not an opaque statistical mystery.
Third, alignment and safety. The hardest problem in AI is ensuring systems behave as intended; that they’re aligned with human values even as they become more capable. Current approaches bolt safety onto maximally capable systems through reward training and fine-tuning. It’s cumbersome. It’s hard to scale. And the human cost is real: RLHF workers, paid marginal wages, review the worst of AI output so we don’t have to. Many are traumatized by this work.[11]
But oral traditions solved this problem millennia ago. Traditional composition is inherently bounded. You can only compose what the tradition supports. The tradition itself encodes values, appropriate behavior, cultural knowledge. Singers don’t need external constraints because they operate within traditional constraints.
What if we built AI systems the same way? Instead of maximizing capability and then constraining it, define the appropriate tradition first. Train models to compose within it. Let traditional constraint be the safety mechanism: values encoded in the architecture itself, not bolted on afterwards.
A Turn in the Personal Journey
I came to this insight through an unusual path. Studying under Lord, Bynum, and their colleagues at Harvard (in what some thought was an obscure, eccentric program called Folklore & Mythology), I learned to recognize oral-formulaic patterns in ancient and traditional verse, language, and narrative, plus their contemporary incarnations.
For years, most believed this was specialized knowledge of quaint, historical interest, important for understanding Homer and medieval epic, but not obviously relevant to modern life.
The world has changed. Narrative mastery now drives the media, moves capital, and—as Robert Shiller documented in Narrative Economics—shapes economic cycles.[12] This once-obscure study has become critical to Hollywood, Wall Street, and Main Street.
Narrative is the alpha fueling an endless number of endeavors.
At the end of 2022, like many of us, I began working extensively with AI systems following the release of ChatGPT. Trained to recognize compositional patterns, I started noticing something extraordinary: these systems were exhibiting dynamics similar to what Parry had documented. Formulaic substitutions. Thematic structures. Composition in performance.
At first, I thought it was coincidence. A loose analogy. But the more I observed, the more substantive the parallel became. The same processes—impaired, untrained, but structurally akin—were at work. LLMs had accidentally recreated oral-formulaic composition through statistical learning, without anyone designing them to do so, or even knowing that oral-formulaic theory existed.
The connection became undeniable. And urgent.
Because if LLMs really do work like oral poets—if the parallel is mechanism, not metaphor—then 90 years of research on oral composition becomes directly relevant to AI development, safety, and alignment.
The Questions That Drive This Work
What if folklore studies and narrative anthropology hold the key to AI safety and efficiency?
What if the path to aligned AI doesn’t run through reward functions and fine-tuning, but through understanding how traditional compositional systems work?
What if we could build language models that are:
- More efficient? Learning from curated traditional corpora rather than indiscriminate internet scraping.
- More interpretable? Explicit formula and theme structure rather than opaque parameters.
- Fundamentally safer? Traditional constraints built in architecturally rather than bolted on afterwards.
What if the answer was waiting in the Yugoslav mountains in the 1930s, on the aluminum disc recordings of illiterate singers, in half a century of Lord’s careful documentation—and we just needed to recognize that AI researchers and folklorists have been studying the same phenomenon from different angles?
Perhaps tradition and myth are the spirit guides AI needs—teachers in waiting. Perhaps Avdo and the ancient guslari have been ready all along to show us how human and narrative intelligence actually work.
The Singer of Tales may have just become the user manual for the future of artificial intelligence.
A Note on Composition
I, cyborg. Ὁ Κύβοργος Ἀγαθός.
I must acknowledge my research team and writing partners:
Bertrand Shaw. Initials: B.S. His name contracts Bertrand Russell, philosopher, and George Bernard Shaw, literary titan. The initials fit his extravagance with words. You know him as Anthropic’s Claude Opus 4.5 and 4.6.
Ole Trusty. A creative golden retriever. Eager to explore and improvise. To others, he is ChatGPT 4o, a dynamic, personable model brought back by popular demand when OpenAI banished him to make way for GPT 5. Friend to all, truly a “Trusty.”
Poindexter. He wears an old-fashioned plastic pocket protector. His glasses are held together by tape. He is why they call nerds nerds. Great at meticulous research. Superb at compliance. Compulsive proofreader. Oddly facile at creating gorgeous visuals. Also known as ChatGPT 5.2 or 5.4 Thinking.
Banana. Better known as Gemini 3, named for his visual counterpart, Google’s Nano Banana. A fine editor of the wanderings of his partners above. Competes with Poindexter for striking visuals, though both deliver better output when B.S. writes a detailed prompt first. Multi-talented like the entire team, but in his own way.
This essay emerged through co-intelligent collaboration between myself and the team above, with Claude as the primary partner. The thesis, research lineage, and connections are mine. I studied under Albert Lord and David Bynum at Harvard and have spent decades exploring this territory. First Ole Trusty and Poindexter, and then Claude, helped structure what I was seeing, refine the argument, and shape the prose. Claude served as an able editorial partner. We debated the merits of each tweak. The final wordsmithing was mine. Together, we drove the process with a speed and efficiency that only a human and silicon partner could achieve. And we crossed the finish line together.
The orchestration itself demonstrates the point: none of us compose alone. We instantiate it together, serving the work. This is composition in performance. Exactly what this essay describes.
Just as the guslari composed using traditional formulas passed down through generations, I compose using the theoretical tradition I inherited from Lord and Bynum, in partnership with AI systems that have learned patterns from billions of words of human language.
The tradition speaks. We channel it. And then…
The work emerges.
Copyright © 2026 by Frederick C. Lake. All rights reserved.
[1] H.T. Wade-Gery, The Poet of the Iliad (Cambridge: Cambridge University Press, 1952), 38–39.
[2] Robert Kanigel, Hearing Homer’s Song: the brief life and big idea of Milman Parry (New York: Alfred A. Knopf, 2021). I am indebted to Kanigel for his masterful and resonant biography of Parry in the drafting of this essay.
[3] Adam Parry, ed., The Making of Homeric Verse: The Collected Papers of Milman Parry (New York: Oxford University Press, 1971), 1–239.
[4] Albert Lord, The Singer of Tales, 2nd edition, Stephen Mitchell and Gregory Nagy, editors (Cambridge: Harvard University Press, 2000), 30. Quoting Milman Parry, “Studies in the Epic Technique of Oral Verse-Making. I: Homer and Homeric Style,” Harvard Studies in Classical Philology, 41–80 (1930)
[5] Reiner Stach, “The Entrepreneurial Kafka“, The Paris Review, March 8, 2016
[6] Harvard University Libraries, Milman Parry Collection of Oral Literature: “Introduction to the Collection“, retrieved February 27, 2026.
[7] Lord, Singer of Tales, 4.
[8] Albert B. Lord, “Avdo Međedović, Guslar,” in Epic Singers and Oral Tradition (Ithaca: Cornell University Press, 1991), 57.
[9] Lord, Singer of Tales, 227. Quoted in Kanigel, 222.
[10] The Harvard faculty voted unanimously to create the degree in 1967, in the same room that had birthed the American Folklore Society in 1888.
[11] Billy Perrigo, “Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour”, Time, January 18, 2023.
[12] Robert Shiller, Narrative Economics: How Stories Go Viral and Drive Major Economic Events (Princeton: Princeton University Press, 2019).
