Language Change: How a Word Variant Takes Over

A language is a population of copiers

Every speaker of a language learned it by imitating the people around them, and keeps adjusting their usage, in small ways, throughout their life based on who they talk to. Treat a language community as a network of agents who occasionally hear a new word or pronunciation and decide whether to adopt it, and you have exactly the kind of system agent-based models were built for — the same mathematical toolkit used for epidemics and rumors, applied to how a linguistic variant spreads and eventually replaces its competitor.

Majority-rule copying

The simplest models have each agent adopt whichever variant is more common among their immediate neighbours — a form of conformity bias sometimes called utterance selection. If speakers copy proportionally to frequency, a variant that starts with any small edge in local usage tends to be reinforced and grows; if speakers copy the strict majority regardless of margin, even a slim local majority can snowball into total local uniformity. Neither rule requires any speaker to consciously prefer one word over the other — the outcome emerges purely from repeated local imitation.

Prestige bias

Real speech communities rarely copy strictly by numbers. Sociolinguist William Labov's decades of fieldwork on sound change in American English repeatedly found that innovations often spread outward from a socially prestigious group, even one that is a numerical minority — listeners disproportionately imitate speakers they perceive as high-status. Adding a prestige weight to a small subset of agents in the model lets a minority-held variant overturn a numerical majority, something a pure majority-rule model can never produce on its own.

P(agent adopts variant A) ∝ sum over neighbours j of:
    weight(j) * [neighbour j currently uses variant A]

weight(j) = 1               for an ordinary speaker
weight(j) = w > 1            for a high-prestige speaker

with equal weights, adoption tracks raw local frequency (majority rule)
with w > 1, a prestigious minority can still tip the outcome

The S-curve

Historical linguists, notably Anthony Kroch's influential 1989 study of syntactic change, have repeatedly documented that language changes progress through a population following a characteristic S-shaped curve when adoption rate is plotted against time: a slow start while the innovation is rare, a rapid middle phase once it becomes common enough that most speakers hear it regularly, and a slow final phase as only the most resistant holdouts remain. This shape is not unique to language — it is the generic signature of any diffusion process where the probability of adopting something scales with how many others have already adopted it, the same pattern that shows up in disease spread and technology adoption.

Why network structure slows things down

A speaker community is rarely one uniform, fully mixed network — it is usually a set of tightly connected local clusters (families, friend groups, regional communities) joined by comparatively few connections between them. Sociologist Mark Granovetter's concept of weak ties highlights why this matters: an innovation can saturate one cluster quickly through dense internal contact, but crossing into a different cluster depends entirely on those sparser bridging connections. A change can therefore look essentially complete within one community while barely having started in a neighbouring one, purely because of how the underlying social network is wired.

Frequently asked questions

Try it live

Everything above runs in your browser — open Language Change and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.