Mountain language diversity

Why can a single mountain valley system hold more languages than a whole plain ten times its size?

The island of New Guinea holds something like a thousand languages — Papua New Guinea alone counts around 840, roughly an eighth of the world's total, on about a third of one per cent of its land. Daghestan, a corner of the Caucasus, holds several dozen from three unrelated families. Meanwhile the North China Plain and the Eurasian steppe, vastly larger and more populous, get by on a handful.

The standard answer is that mountains isolate people, and isolated people drift apart. But the Siberian taiga and the Canadian north are as isolating as any mountain range and are linguistically thin. So isolation on its own evidently does not do it.

Separate two things the standard answer runs together. Isolation can only preserve differences; it cannot create them. Languages diverge continuously wherever speakers stop talking to each other often enough, and that happens everywhere. The interesting variable is not what keeps groups apart but what makes them come together — because contact is what erases difference, and its absence is what lets difference accumulate.

So ask: what forces people to maintain wide-ranging relationships with people far away? Chiefly, risk. If your food supply fails in some years and you cannot store your way through, you must have partners elsewhere — kin, exchange networks, obligations you can call on. Those relationships are expensive and they require a shared language across a large area. Where the growing season is long and reliable, a community can feed itself from a small territory in almost every year, and needs no such insurance.

That is the argument Daniel Nettle made, and it predicts the global map better than ruggedness does: language diversity tracks growing-season length and rainfall reliability more closely than it tracks terrain. It explains Vanuatu, roughly a hundred languages among a few hundred thousand people — flat islands, no mountains to speak of, but a year-round growing season. And it explains the taiga, where a household may need a hunting range of hundreds of square kilometres and cannot survive without distant ties.

Where do mountains come in, then? In two ways, neither of them "isolation" in the simple sense.

The first is vertical stacking. Walk a few kilometres up a mountainside and you cross several ecological bands: irrigated terraces, orchards, forest, summer pasture, high grazing. A valley community can hold a full portfolio of resources inside a very small horizontal territory, moving stock up and down rather than trading across a plain. Mountains do not merely divide; they let a self-sufficient community be small. That is the same mechanism as a long growing season, arriving by a different route.

The second is political. A plain is cheap to cross with an army, cheap to administer, cheap to tax, and cheap to farm at scale. One expanding polity or farming population rolls across it — Han, Indo-European, Bantu, Turkic, Arabic — and what was there before is absorbed. Plains are not places where diversity failed to arise; they are places where it was erased, repeatedly. Mountains resist because occupation and administration are expensive per head, so the same expansion that flattens a plain stalls in the valleys above it.

This gives the useful distinction between a spread zone, where one expansion periodically overwrites everything, and a residual zone, where languages accumulate because nothing overwrites them. The Caucasus is the standard residual zone; the steppe beside it is the standard spread zone. Some go further and argue that highland peoples actively cultivated fluid, fragmented identities as a way of staying ungovernable — an influential reading of upland Southeast Asia, and a contested one, since it credits deliberate strategy to a pattern that low tax yields might explain on their own.

Three qualifications, in rough order of how much they should worry you.

Counting is contested. Whether two speech forms are one language or two is a judgement, and the reference counts everyone cites are compiled unevenly. A region studied intensively by linguists who split finely will out-count one surveyed coarsely. Some of the mountain-versus-plain gap is real; some is who did the surveying.

The statistics are correlational and spatially awkward. Neighbouring places resemble each other on every variable at once, so growing season, ruggedness, isolation and state history are all tangled together. Nettle's ecological-risk account is well argued and widely cited, not universally accepted.

And terrain is not a single thing. A range is a barrier across and a corridor along; valleys concentrate movement rather than blocking it. What matters is not ruggedness in the abstract but whether a self-sufficient community's territory can be small, and whether a state ever found it worth swallowing.

• Why a lingua franca can spread across a diverse region without displacing the local languages beneath it, and when it stops doing so.
• How the same ecological-risk logic predicts the size of exchange networks and kinship obligations, not just of language communities.

Every term the collection defines is gathered in the glossary.