Contact, borrowing and the limits of resemblance

node

The single largest confound in genealogical classification, and the reason similarity is evidence of nothing on its own. Languages in prolonged contact exchange words freely — English has taken thousands from French and Latin without ceasing to be Germanic — and prolonged contact also transfers structure: sounds, word order, grammatical categories. Where several unrelated languages share a region for long enough, they can converge into a linguistic area, or sprachbund, in which typological similarity is high and genealogical relatedness is nil; the Balkans and South Asia are the classic cases. The comparative method's defences against this are indirect but effective. Loanwords tend to enter after the relevant sound changes have finished, so they fail to show the expected correspondences. Basic vocabulary and bound morphology — irregular verb paradigms, pronoun sets, inflectional endings — are borrowed far less readily than nouns for goods and institutions, so agreement in irregularity is worth more than agreement in vocabulary. But the defence is statistical rather than absolute, and where contact is intense and ancient, it can be impossible to tell an inheritance from a very old loan.

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