The War of the Ghosts
F.C. Bartlett, 1920. A Cambridge psychologist passed a Kathlamet ghost story down chains of English retellers and measured the drift: the unfamiliar cut, the strange domesticated — and the ghosts, the ones in the title, gone from every chain almost at once. Legend fading to myth, caught on paper.
In 1920 Frederic Bartlett was an experimental psychologist at Cambridge with a parlour game and a measurement problem. Everyone believed stories change in the retelling; nobody had measured how. So he built an instrument out of people. In “Some Experiments on the Reproduction of Folk-Stories” — the journal Folklore, vol. 31, no. 1, pp. 30–47, and the whole paper is free to read — he describes two methods: repeated reproduction, one person retelling a story at widening intervals, and serial reproduction, a chain — the parlour game telephone, run as apparatus — each teller reproducing only the version the teller before them left behind (pp. 31–32). Subjects read the story twice at a normal pace, wrote it from memory fifteen minutes later, and never saw the original again (pp. 32, 35). Twelve years later he expanded the experiments into the book everyone cites, Remembering (1932), where his idea of the schema got its name. The 1920 paper is the primary record, and the chains are printed in it verbatim.
A control that was already a retelling
The story he fed the machine was “The War of the Ghosts,” and its own paper trail is the first lesson. It is a Chinookan narrative in the Kathlamet dialect, told to the anthropologist Franz Boas by an aging Kathlamet speaker named Charles Cultee, and published in 1901 in Kathlamet Texts, Bulletin 26 of the Bureau of American Ethnology (the Chinook Indian Nation tells the Cultee–Boas story itself). Bartlett noted in a footnote that he had “slightly adapted” it; he does not say how. So the fixed original — the baseline every distortion was measured against — was already an old man’s remembered telling, carried across a language into print, then trimmed by the experimenter’s hand. The control was a retelling, three transmissions deep before the first subject read it twice.
The ghosts leave first
The story is a ghost story; the ghosts are in the title. A young man is drawn into a war party the text lets you slowly realize is made of the dead. And in every serial chain Bartlett ran — every one — the ghosts dropped out almost immediately (p. 35). His Cambridge subjects had nothing in their own stock of ideas that would take the weight of a ghost war-party, and material a reteller cannot rationalise is not argued with. It is simply not there anymore. The story kept its river, its battle, its dying; it lost the thing it was about.
Bartlett sorted the wreckage with a vocabulary worth preserving, because it is about to matter that it is his. Material fell out of the chains in three classes of omission — the irrelevant, the unfamiliar, the unpleasant (pp. 34–37). What stayed was worked over by three engines. Familiarisation: the unfamiliar swapped for the nearest familiar thing — “canoe” became “boat,” “paddling” became “rowing,” and in a Central African story run through the same machinery, “pea-nut” became “acorn” (pp. 36, 43–45). Rationalisation: retellers splicing in “therefore” and “because” the original never had, until stories grew morals nobody wrote. Dominance: one vivid detail bending the whole narrative around itself. He even logged the opposite effect, the persistence of the trivial — an odd, meaningless detail riding a whole chain untouched (pp. 32–33, 43).
Rationalisation even repairs the damage the omissions cause. The young man’s wound belonged to the ghost-logic; with the ghosts cut, the wound dangled, unexplained — so the chains explained it, inventing an arrow that fell away (“sore wounded”), returned (“pierced through the heart”), and finally settled as plain “mortally wounded,” cause unneeded (pp. 38–40). Nobody decided this. The story healed its own continuity around the amputation, one honest misremembering at a time.
Something black
The best evidence in the paper is the death scene, laid out across pp. 40–42 like a specimen series. The original ends the young man’s life on an image with no explanation attached:
When the sun rose, he fell down. Something black came out of his mouth. His face became contorted. The people jumped up and cried. He was dead. — the death scene as Bartlett’s subjects first met it · F.C. Bartlett, “Some Experiments on the Reproduction of Folk-Stories,” Folklore 31(1), 1920, pp. 40–42
“Something black.” Not a soul, not a spirit — a thing, opaque and unassigned. The retellings could not leave it alone. It becomes “a black thing,” then “a great black thing” that “flew” from his mouth; then the frame arrives — “his soul fled black from his mouth”; then “his soul passed out”; then “his spirit fled” — a fully domesticated Western death, the strangeness converted without residue into the most familiar dying image his subjects owned. One of them told Bartlett exactly how it happened in his head: “I was thinking of the Greek myth, and visualised a picture in which the soul is flying from a dying man’s mouth.” A subject catching his own culture rewriting a foreign image, in the first person, on the record, 1920.
Not his words
Here the study turns into a demonstration of itself. If you met this experiment in a psychology course, you likely met it as “assimilation, leveling, and sharpening.” Those words are not in the paper. They are later textbook shorthand — a relabeling from the lineage of Allport and Postman’s 1947 rumor research, which built on Bartlett — an error now propagating through nearly every study guide that mentions him. His terms are the ones above. Also not in the paper: a sample size. The tidy headcount repeated across revision sites appears nowhere in the 1920 text, which never states one; some subjects read the story, some heard it, and intervals varied — loose by design. The most famous memory study in the field is remembered with a vocabulary it never used and a precision it never had. A paper about drift, drifted.
What the drift actually says
The replication file has the same shape. For decades the repeated-reproduction result carried a shaky record — worth savoring, given the subject — until Bergman and Roediger rebuilt the protocol in 1999 with the same story, testing recall at fifteen minutes, one week, and six months. Content decayed and rationalisation grew with the interval: Bartlett’s qualitative pattern held. (Cited here from the paper’s indexed abstract, not its full text — which is why it appears without quotation marks.)
Then in 2022 a reappraisal in the journal Memory, building on Wagoner’s work on Bartlett’s schema concept, pressed the sharper point: the pop-science moral — Bartlett proved memory is unreliable — misreads him. His theory predicts accurate recall when material fits the tellers’ culture, and he reported strikingly accurate recall himself. Run the chain on a story the tellers already own — the reappraisal used the “Vanishing Hitchhiker,” a familiar urban legend — and it survives retelling comparatively intact, more so when tellers expect a strict audience to check them. Memory is not a bad recorder. It is a strong editor with a house style, and Bartlett named the house style in 1920: “a common tendency to change all presented material into such a form that it may be accepted without uneasiness, and without question” (p. 37).
Which returns this page to the week it lives in. The Eye of the World opens by asserting a decay chain as cosmology — memory becomes legend, legend fades to myth, and even myth is long forgotten — and then runs the engine where you can watch. Bartlett had run that chain as a procedure seventy years earlier, and his results say the fade has a direction. The unfamiliar is cut; the strange is domesticated; the black thing becomes a soul, the canoe becomes a boat — and the ghosts leave first, because nothing in the new tellers’ world will hold them. When Robert Jordan’s gleeman tells a fireside that Lenn flew to the moon in the belly of an eagle made of fire, that is the Cambridge data, run forward an Age. The Wheel turns at the speed of a nine-person chain.
We point; we don’t reproduce. The 1920 paper was read here in full, page by page, and every page number and quotation above comes from it — including the death-scene chain and the subject’s Greek-myth report. The Bergman–Roediger findings are cited from the indexed abstract only, so no words are put in their mouths. The sample size that circulates in study guides appears nowhere in Bartlett’s text, so it appears nowhere here. And “assimilation, leveling, sharpening” is quoted above only to say plainly that Bartlett never wrote it. Boas’s 1901 original was not re-read for this page; nothing from it is quoted beyond what Bartlett’s own footnote carries.