A simple branching model that reproduces language family and language population distributions

Human history leaves fingerprints in human languages. Little is known over language evolution and its study is of great importance. Here, we construct a simple stochastic model and compare its results to statistical data of real languages. The model bases on the recent findings that language changes occur independently on the population size. We find agreement with the data additionally assuming that languages may be distinguished by having at least one among a finite, small number of different features. This finite set is used also in order to define the distance between two languages, similarly to linguistics tradition since Swadesh.

💡 Research Summary

The paper presents a minimalist stochastic model that captures two well‑known statistical regularities in world languages: the power‑law distribution of language‑family sizes and the log‑normal distribution of speaker populations for individual languages. Building on recent empirical work suggesting that the rate of linguistic change is essentially independent of the number of speakers, the authors model each language as a fixed‑length binary feature vector (typically 20–30 bits). The distance between two languages is defined as the Hamming distance between their vectors, echoing the Swadesh‑type lexical distance used in historical linguistics.

The dynamics consist of two probabilistic events applied at each discrete time step. With probability p_branch a language “branches”, creating a daughter language that inherits the parent’s feature vector except for a single randomly chosen bit that is flipped (the mutation). Simultaneously, each language’s speaker population evolves according to a log‑normal growth process, but crucially the branching and mutation probabilities are independent of population size. The simulation starts from a single proto‑language with a small population and runs for thousands of steps, generating a tree‑like phylogeny of languages.

Results show that the simulated family‑size distribution follows a power law f(k) ∝ k^‑α with α≈1.5, matching the empirical distribution derived from Ethnologue and other linguistic databases. The distribution of speaker numbers across languages is log‑normal, reproducing both the bulk and the heavy tail of real‑world data, including the few languages with tens of millions of speakers. Sensitivity analyses reveal that the model’s success hinges on a moderate number of features (≈20) and a low mutation rate (≈1–5 %). Too few features produce overly discrete distance spectra, while too many lead to excessive sparsity and unrealistic fragmentation of families.

The authors discuss the implications of these findings. By stripping away geographic, social, and political complexities, the model demonstrates that the macroscopic statistical patterns of language evolution can emerge from simple, population‑size‑independent branching and mutation processes. The finite feature set provides a concrete operationalization of linguistic distance, offering a bridge between abstract statistical modeling and traditional comparative methods. Limitations are acknowledged: the model ignores spatial diffusion, network effects, and the multi‑valued nature of many linguistic traits. Nevertheless, its ability to reproduce real data with such parsimonious assumptions suggests that language change may be governed primarily by stochastic, scale‑free mechanisms rather than by detailed demographic or cultural factors.

Future work is proposed in two directions. First, integrating geographic spread and social network structures could explain regional variations in language diversity and the emergence of language isolates. Second, extending the feature representation to allow multi‑state or continuous variables would increase realism and enable direct comparison with phonological or syntactic typology databases. Overall, the paper offers a compelling proof‑of‑concept that a simple branching model, grounded in empirically motivated assumptions, can account for the observed statistical regularities of language families and speaker populations, thereby providing a valuable baseline for more elaborate theories of language evolution.