Language extinction and diversification: The two sides of the same coin

Language extinction is conventionally treated as a conservation crisis to be arrested, and language diversification as a historical explanandum to be reconstructed. The two phenomena are studied separately, by different communities, with different methods.

This separation is theoretically incoherent. Extinction and diversification are reciprocal outcomes of a single evolutionary dynamic operating on four nested scales of linguistic organisation. Whether a given dynamic is labelled "extinction" or "diversification" depends on the observer's reference frame, not on the mechanism.

I integrate lexical diffusion (Wang), utterance selection (Croft), feature-pool ecology (Mufwene), iterated learning (Kirby), agent-based simulation (Gong et al.), and the Extended Evolutionary Synthesis (Laland et al.). A worked numerical example, using a Pólya-urn model with the Price-equation decomposition, demonstrates that identical micro-dynamics produce divergence or replacement depending on a single pair of ecological parameters (inter-group contact × prestige alignment).

Five falsifiable predictions follow, each with a concrete operationalisation, from symmetric sign-reversal of endangerment predictors in phylogenetic diversification data, to measurable Large Language Model (LLM) induced diversity collapse in post-2022 corpora.

The framework reframes revitalisation policy (target niche conditions, not feature prescription), Artificial Intelligence (AI) design (diversity as a reward signal, not an afterthought), and the conceptual status of endangered languages (ecologically reconstructible, not ontologically tragic).

Language extinction and language diversification are not opposing phenomena; they are the same evolutionary process running in two directions. A single multi-level selection framework that spans sub-lexical, feature-pool, population, and machine-mediated scales predicts both divergence and death, and a tractable Pólya-urn simulation shows how identical micro-dynamics flip between the two under one ecological parameter.