Separating connection from coincidence

Two stories can look alike for three different reasons: they descend from a common ancestor, one borrowed from the other, or they were invented independently and merely rhyme. Myth Tracker's similarity engine is built to tell these apart — and, above all, to avoid the classic trap of calling every resemblance a connection.

The Seven Sisters problem

The motivating case: Greek mythology has Orion chasing seven sisters who become the Pleiades; First Nations Australian traditions — isolated from Eurasia for roughly 50,000 years before 1788 — tell of seven sisters, pursued by a male figure identified with Orion, who rise into the sky as the same cluster. Both call them seven although most eyes see six stars, and both explain the discrepancy with a lost sister. If these stories are genuinely related, their common source predates the out-of-Africa dispersals: it would be close to 100,000 years old.

But the careful analysis (Berezkin's motif statistics) shows why the naive conclusion is unsafe:

• The Pleiades and Orion are among the most salient objects in the night sky — every culture tells stories about them (340+ Pleiades motifs vs fewer than 100 Seven Sisters motifs in Berezkin's database), so the base rate for coincidental overlap is enormous.
• Earth's rotation makes Orion visibly follow the Pleiades across the sky — the "chase" is written in the sky for anyone to read off independently.
• "Man pursues woman/women" is one of the most common motifs on Earth, especially natural for hunter cultures.
• Core elements of the Australian telling — the ice-bodied sisters, the frost, the initiation ritual triggered by the cluster's rising — are entirely absent from the Greek version. Under common descent you expect core elements to survive somewhere; their total loss in one branch argues for independent origin.
• What does survive scrutiny is the narrow, arbitrary motif of "seven stars, six visible, one lost" — plausibly a memory of the star Pleione when it stood visibly apart from Atlas, 70–100 thousand years ago.

The engine below is that reasoning, made mechanical and applied to every pair of stories in the database.

Step 1 — Decompose stories into motifs

Following Yuri Berezkin's analytical catalogue (and the older Thompson Motif Index / ATU tale-type tradition), each telling is described as a bundle of discrete motifs — "birds sent to find land", "the lost Pleiad", "serpent withholds the waters". Motifs are the unit of comparison because they survive transmission far better than whole plots: stories are re-cut at every retelling, but their components persist.

Each motif in a telling is tagged with a role: core (the story is unrecognisable without it), supporting, or peripheral (local colour).

Step 2 — Weight shared motifs by global rarity

Every motif carries an estimate of how many of the world's traditions attest it (out of ~1000, Berezkin-style). Sharing a near-universal motif is weak evidence; sharing a rare, arbitrary one is strong evidence — the same logic as inverse document frequency in text search, or shared derived characters in cladistics.

"Hunter pursues women" (~400/1000 traditions) gets weight ≈ 0.26; "birds released to scout for land" (~25/1000) gets ≈ 0.60. A pair of stories sharing three rare motifs will vastly outscore a pair sharing three universal ones.

Step 3 — Overlap, weighted by role

The raw similarity is a weighted overlap: shared motifs (counted at the average of their two roles — core-in-one but peripheral-in-the-other is intermediate evidence) divided by shared plus unshared material. Unshared motifs count against the score but are damped: in cladistic terms they are autapomorphies — a story accumulating its own elaborations over millennia is weaker counter-evidence of kinship than a missing shared core.

Step 4 — The core-retention penalty (the coincidence filter)

This is the filter that weeds out stories that seem similar but miss too many essential elements. For each story in the pair, we ask: what fraction of its core motifs is entirely absent from the other? Under real descent or borrowing, core elements tend to survive in some recognisable form; wholesale absence is the signature of coincidence.

This is exactly what demotes Greek–Australian Seven Sisters: healthy motif overlap, but each side's core includes elements (ritual/ice-water on one side, the shame-of-Merope aetiology pattern on the other) that the other side lacks entirely.

Step 5 — Verdict tiers, not fake probabilities

Mythological data is too soft for the score to be read as a probability, so it maps to tiers:

What the score deliberately ignores

Geography and chronology are shown on the maps but kept out of the similarity number. That is intentional: transmission plausibility (could these peoples have shared an ancestor or a contact route recently enough?) is a separate question from textual resemblance, and collapsing the two lets each contaminate the other. A high score across an "impossible" gap — like the seven-count of the Pleiades across the Australian isolation barrier — is precisely the interesting kind of result, because it forces a choice: extreme antiquity, unrecorded contact, or a hole in the motif model.

Method lineage

• Yuri Berezkin — analytical catalogue of ~3000 motifs across ~1000 world traditions; areal statistics linking motif distributions to prehistoric migration routes.
• Julien d'Huy — phylogenetic (cladistic/Bayesian) reconstruction of myth families such as the Cosmic Hunt, treating motifs as heritable characters.
• Aarne–Thompson–Uther / Thompson Motif Index — the classical decomposition of folk narrative into types and motifs.
• Michael Witzel — the Laurasian/Gondwanan macro-family hypothesis for world mythology.

← Back to the Myth Index