Study 5: the raven and the dove

The lost-civilisation reading. In Fingerprints of the Gods (1995), Underworld (2002) and Magicians of the Gods (2015), Graham Hancock collects flood stories from Mesopotamia, Greece, India, China, the Americas, Africa and Oceania and argues that their number and their agreement cannot be coincidence: they are, on his reading, the memory of a real, worldwide catastrophe, which he places at the end of the Younger Dryas around 11,600 years ago, when meltwater and, he argues, comet fragments raised the seas and drowned the coasts of an advanced civilisation whose survivors carried the story, and the arts of civilisation, to the rest of the world. Plato's date for the sinking of Atlantis, nine thousand years before Solon, is close to that horizon, and the coincidence is part of the case. The stories are said to agree on the essentials: a warning, a chosen survivor, a vessel, animals saved, a landing on a mountain, a bird sent to find land, a sacrifice, a promise.

The ordinary reading. Rivers flood, coasts flood, tsunamis strike, and every people that lives by water has a story about the time the water came. Frazer collected the world's flood stories in 1918 and concluded that most were local memories of local floods, some were etiological (why the rocks are shaped so, why the tribes are separate), and a number in the Americas, Africa and the Pacific had taken on details from Genesis after contact with missionaries, which is why the raven and the dove turn up in places that had never heard of Noah before the eighteenth century. The Genesis story itself has a documented ancestry: it descends, in detail, from the Babylonian flood in the Epic of Gilgamesh, which descends from Atra-Hasis and the Sumerian Ziusudra, a chain of transmission over two thousand years within one region. That chain is what a real transmission looks like in a text: the same bird, the same mountain, the same sacrifice, the same sequence.

What the pilot can distinguish. It can describe the distribution of the selected features in this collection. The original design expected an ancient common source to produce little regional structure; that is not a necessary consequence of common ancestry, since inherited stories can diversify. Concentrations of biblical-looking details can motivate a borrowing hypothesis, but source history is needed to test its direction and timing. The original expectations remain in the preregistration; they are not treated here as decisive tests.

Mark Isaak's Flood Stories from Around the World is the open compendium used here: summaries, each with its source (Frazer, Gaster, Vitaliano, Dundes, the ethnographies), grouped by region and, on a companion page, by language family. It is a secondary source of summaries, and the limits of that are stated at the end. It is also exactly the kind of collection the lost-civilisation books draw on, which makes it the right thing to test.

Claude Sonnet 5 read each summary and answered 26 questions with yes, no or unclear, from the text only: is the flood a punishment; does a god warn the hero; is a vessel built, and of what kind; do the survivors flee to a mountain or a tree; are animals or seeds taken; is a bird sent out, and is it a raven or a dove; does another animal scout, or dive for earth; is a duration stated, and is it forty; is there a sacrifice after, a rainbow or promise; are people re-made from stones or seeds, or by sibling marriage; is anyone bearded or pale; is there a fire or darkness as well; a previous race of giants; does the text itself mention Noah, the Bible or missionaries; is the flood local or the whole world. The coder was told not to name or guess the culture. "Unclear" was used on average times per story out of 26, mostly on the shortest summaries.

Two groups of details were fixed in advance. Generic: any flood story might have them (punishment, a lone survivor or family, a mountain, the whole world, a vessel). Diagnostic: the details that make the Genesis account the Genesis account, and that it shares with its Mesopotamian ancestors (a warning from a god, animals taken aboard, a bird sent out, the raven or dove by name, a sacrifice after, a rainbow or promise, the number forty). The diagnostic score of a story is how many of those seven it has.

Share of stories in each region with each detail. Darker is more common. Diagnostic details are marked.

Every story from outside the ancient literate traditions that carries three or more of the seven diagnostic details, with its region, its diagnostic details, any Christian names or terms in the summary, and the opening of the summary. Read them as leads for a source-history investigation. The keyword flag undercounts. The Masai story at the top of the list, which Frazer took from the German officer Moritz Merker's 1904 account, has a righteous man who "found grace in God's eyes", an ark, animals, a dove and a rainbow, and no keyword; Frazer himself doubted it was untouched by Genesis. The Altaic story has an ark on a mountain and birds sent out, from a people long in contact with Islam, whose scripture has Noah. The Zapotec, Tarascan, Popoluca and Totonac stories were recorded in Mexico centuries after the friars, and two of them name angels, saints and Noéh. These contexts show opportunities for transmission; they do not date the first appearance of each detail or demonstrate that it was borrowed.

Similarity between stories is the share of details they have in common (Jaccard, over 25 yes/no details, the Bible-mention flag excluded). These are the closest stories to the Hebrew account, and to the Babylonian one.

Summaries, not texts. Isaak's entries are paragraph summaries of stories that are often much longer. A detail absent from a summary may be present in the story; the measured prevalences describe what the summaries and coder captured, not guaranteed lower bounds on the full traditions. The right full study codes the sources themselves: Frazer's three hundred pages, Dundes's anthology, the ethnographies.

Dates are publication dates. The year attached to a story is when its source was published, not when the story was first told or first recorded, and for most of the world the two are separated by contact. A proper stratigraphy needs the collection date of each telling, which the sources contain and the summaries do not.

One coder. A single model coded each story once. Study 3 showed how much two coders can disagree on images; on short texts the task is easier, and spot checks agree, but a second coder and an agreement statistic are the first improvement.

Contact is not the same as borrowing. Christian words in a summary, contact with missionaries and a demonstrated history of a particular borrowed detail are different observations. The keyword flag cannot settle direction or timing. The absence of a flag is not evidence of isolation; its presence is not proof that the whole story derives from Genesis.

The corpus is the compilers' corpus. It over-represents what nineteenth- and twentieth-century English-language ethnography recorded, and North America above all. Regional comparisons are between samples of unequal size and quality.

What would improve the inference. Independently check the coding against source passages, distinguish collectors’ notes from narrators’ words, and trace changes within individual traditions. A story can have both ancient ancestry and regional structure. A stronger claim of borrowing needs evidence of particular details entering later versions, not merely a post-contact publication date.