Watch a new word variant spread through a social network as agents imitate the majority form used by their neighbours. Toggle prestige weighting to see how high-status speakers can steer an entire community toward a new way of speaking.
Each of 120 agents holds one of two word variants and periodically polls a handful of neighbours on a social network, adopting whichever form is locally dominant with some probability. Small local majorities cascade into global consensus — the same mechanism behind real sound changes and lexical replacement.
Set the initial share of variant B and the conformity strength (how eagerly agents switch), then watch the share-over-time chart. Enable prestige weighting to give high-status agents outsized influence over their neighbours' choices, and observe how much faster consensus forms.
Real language changes almost always follow an S-shaped adoption curve — slow start, rapid middle spread, slow finish — the same shape this simulation produces from purely local, individually simple imitation rules.
This simulation models language change as a diffusion process on a social network: each of 120 speakers holds one of two competing word variants and periodically samples a handful of neighbours, switching to whichever form is locally dominant with some probability. Despite having no central coordination and no agent ever seeing the whole population, purely local imitation reliably tips the entire network toward one variant or the other, producing the S-shaped adoption curves sociolinguists have documented in dozens of real dialect studies. The optional prestige toggle lets high-status agents count for more when neighbours poll their variant, modelling the well-established finding that changes led by socially prominent speakers spread faster and more completely than changes with no prestige gradient.
The model draws on foundational work by William Labov, who showed in the 1960s that sound changes on Martha's Vineyard and in New York City spread through social networks along lines of prestige and identity rather than uniformly across a population, and on Everett Rogers' diffusion-of-innovations framework, which formalised the S-curve shape now considered a signature of successful linguistic (and technological) change. Together they explain why some new words or pronunciations vanish quietly while others sweep an entire speech community within a generation.
It models a population of speakers connected in a social network, each currently using one of two variants of a linguistic feature (a word, pronunciation, or grammatical form). On each interaction, an agent looks at a sample of its network neighbours and, with some probability, switches to whichever variant is more common among them. Repeating this millions of times produces large-scale patterns of language change from purely local decisions.
The "Initial B share" slider sets how many agents start using variant B. "Conformity" controls how readily agents switch to match their neighbours — higher values speed up convergence. "Neighbours polled" controls how much local information each agent samples before deciding. The prestige toggle makes higher-status agents (larger, gold-rimmed nodes) count for more in that poll, and the chart on the right tracks variant B's population share over time.
With majority-rule imitation on a finite, connected network, random drift combined with the "rich get richer" dynamic of majority copying almost always drives the system to one of two absorbing states — everyone using variant A or everyone using variant B — the same outcome predicted by voter-model mathematics. Perfectly balanced 50/50 mixtures are locally unstable, just as they are in real dialect contact zones that eventually resolve toward one dominant form.
Broadly, yes. Sociolinguistic fieldwork since the 1960s — most famously William Labov's studies of Martha's Vineyard and New York City — shows that phonetic and lexical changes spread through social networks via face-to-face imitation, not through top-down instruction. Change typically starts in a socially identifiable subgroup, spreads along network ties, and eventually reaches near-universal adoption, closely matching the network diffusion this simulation reproduces.
Sociolinguists distinguish "overt prestige" (associated with standard, high-status speech) from "covert prestige" (associated with in-group solidarity, e.g. regional or working-class speech). Both function the same way mechanically: certain speakers exert disproportionate influence on which variant others adopt. The prestige toggle here models that asymmetric influence abstractly — enabling it makes high-prestige nodes act as opinion leaders whose usage neighbours are more likely to copy.
Early on, only a few agents use the new variant, so the probability that any given neighbour poll returns a majority for it is low — adoption is slow. As the share grows past roughly 30-40%, polls increasingly return majorities for the new form, and adoption accelerates sharply. Once the old variant becomes rare, remaining holdouts are surrounded by the new form and convert quickly, flattening the curve near 100%. This slow-fast-slow shape is the classic logistic diffusion curve seen in Rogers' innovation-diffusion research.
Yes. This simulation connects each agent to its four nearest neighbours in space, producing tightly clustered local communities bridged by a few longer links. Densely clustered networks tend to produce faster local consensus within clusters but can take longer to reach agreement between clusters, whereas networks with more random long-range ties (closer to a small-world topology) spread changes globally more quickly — a pattern also studied in the Six Degrees of Separation simulation on this site.
In this simplified model, no — with unlimited time, one variant almost always sweeps to full adoption because majority-rule imitation has no mechanism to preserve diversity. Real languages do sustain stable variation for centuries (regional accents, register-dependent word choices), which typically requires additional forces this basic model omits, such as social identity signalling that makes some agents actively prefer minority forms, or geographic isolation that limits mixing.
Diffusion-on-networks models of this kind are used well beyond linguistics: marketers use them to predict adoption of new products, epidemiologists adapt the same majority-rule mechanics to model behaviour change during health campaigns, and computational sociolinguists use agent-based models to test competing theories about why some innovations (new slang, hashtags, emoji usage) go viral while most quietly die out.