The comparative method

node

The procedure that turns a suspicion of relatedness into a demonstration, and the only one historical linguistics has that produces a testable result. It runs in four steps. Assemble candidate cognates — words of similar meaning across the languages being compared, drawn preferentially from vocabulary that is rarely borrowed: body parts, kin terms, low numerals, pronouns, basic verbs. Extract the sound correspondences: not resemblances, but recurring equations of the form "wherever language A has p, language B has f, in this position". Reconstruct the proto-form that would yield every attested outcome by the fewest independent changes, and mark it with an asterisk because it is an inference and not an attestation. Then test the reconstruction against words not used to build it. The force of the method comes entirely from the third and fourth steps: a correspondence set that holds across hundreds of unrelated meanings cannot be produced by chance, because chance does not repeat, and it is not easily produced by borrowing, because borrowers take words rather than rules. The method is also what defines a family — a family is not a group of similar languages but a group for which a shared proto-language has been reconstructed.

A node is not a place. Drawing it on a map would assert something about the world that no stored fact supports.