What if methods designed to predict species extinction could help explain disappearing languages—but only after being carefully redesigned? This review shows where the analogy is powerful, and where it breaks down.

Language diversity is under threat, with between a third to a half of all languages considered endangered, and predicted rates of loss equivalent to one language per month for the rest of the century. Rather than reviewing the extensive body of linguistic research on endangered languages, this review focuses specifically on the interdisciplinary transfer of methods developed in conservation biology, macroecology and macroevolution to the study of language endangerment and loss.

While the causes of language endangerment and loss are different to those for species, studying patterns of diversity of species and languages involves similar analytical challenges, associated with testing hypotheses and identifying causal relationships. Solutions developed in biology can be adapted to illuminate patterns in language endangerment, but other tools from conservation biology may be much less use in understanding or predicting language endangerment.

This review concerns the application of methods developed in conservation biology, macroecology and macroevolution to the patterns and causes of language endangerment. This is not a general review of language endangerment because there are many excellent reviews written by experts in linguistics.

Instead, it is focused specifically on the way that methods originally developed in biology have been adapted and applied to endangered languages. Like biodiversity, language diversification is a continuous process of generation of new languages, sometimes accompanied by the loss or replacement of other languages.

Languages, like species, can be lost if there are no new generations using the language to communicate; Etruscan ceased to be used around two thousand years ago and left no descendant languages. Language loss, like species loss, can occur when the speaker population is catastrophically reduced, for example, through violence or disease.

However, in contrast to biodiversity, language loss also occurs through language shift, when speakers cease to speak their heritage language, or their children no longer learn it, as they adopt a different language to communicate. Decline through language shift makes language loss much more complicated than species extinction.

Language shift may be forced upon a population, occur through lack of support for a minority language, or accompany economic, social or geographic movement. Language shift complicates the relationship between population size and number of speakers, and this makes some techniques from conservation biology difficult to apply to language endangerment.

Speaker population size has been used as an indicator of language endangerment, including IUCN criteria based on small speaker populations, restricted range and speaker population declines. However, there are several limitations with using range and population size as a primary indicator for language endangerment.

Quantitative data on declines in speaker number are available for relatively few languages, so Amano and colleagues could include only nine percent of the world’s languages in their analysis. More generally, population size is not always a good indicator of language vitality.

Translation of IUCN-style endangerment metrics to languages is made problematic by differences in the way both population size and range size are recorded. Language range maps are not like species range maps. The IUCN Red List provides a centralised database and a global standard for assessing and communicating species endangerment, but currently there is no equivalent internationally agreed standard for language endangerment.

The standard indicators of species endangerment underlying IUCN ratings—population size and range decline—are problematic for evaluating language endangerment. Instead, language endangerment scales tend to focus on transmission from one generation to the next and domains of use.

Several global databases of endangered languages exist, each using a different endangerment scale and having different coverage. While the different scales and databases broadly agree, there are some differences in rankings for individual languages, including White Lachi and Bahing.

Table one compares four language-endangerment systems: EGIDS, AES, UNESCO, and LEI, including their databases, accessibility, language coverage, and estimated endangered percentages. It reports seven thousand one hundred fifty-one languages and forty-three percent endangered for EGIDS, eight thousand five hundred sixty-five and sixty-three percent for AES, two thousand five hundred and fifty percent for UNESCO, and three thousand four hundred fifty-nine and forty-three percent for LEI.

The aligned categories also show how assessments incorporate speaker numbers, intergenerational transmission, domains of use, recognition, and stability, while noting that communities may prefer terms such as “Sleeping” to labels like “extinct” or “dormant.” Identifying correlates of endangerment, whether internal factors or external threats, requires separating out patterns due to shared environment, patterns of relatedness and covariation between traits.

For example, it has been suggested that polysynthetic languages are particularly vulnerable to decline. Words in polysynthetic languages are formed from many different morphemes, allowing complex ideas to be coded into single words. Because polysynthetic languages can be challenging for adults to learn, it has been proposed that they are prone to simplification and loss in contact situations.

The challenge in testing this hypothesis is that there are many covarying factors to be untangled. Fortunately, there are methods designed to untangle such causal interconnections caused by relatedness, proximity and covariation. Related languages tend to be more similar to each other in many features, even if those features are not structurally related.

This problem is known in biology as phylogenetic nonindependence, but is often referred to in cultural evolution studies as Galton’s problem. Relationships between languages can be used to select datapoints that approximate statistical independence, or to inform a matrix of expected covariance due to shared ancestry.

Figure two shows how L1 speaker population size relates to two language-endangerment scales: the AES from Glottolog and the EGIDS from Ethnologue. Speaker population is plotted on a logarithmic axis, while the panels show endangerment categories and their correspondence across the two systems, with some languages marked by three-letter ISO codes.

This matters because the figure makes clear that language endangerment is assessed through intergenerational transmission and use, rather than population size alone. To avoid being led astray by indirect associations between variables, it is important to evaluate the explanatory power of variables above and beyond their covariation with other variables.

Languages show a latitudinal diversity gradient: the tropics have more languages, which tend to have smaller populations and more restricted areas. When the effect of proximity, relatedness and covariation are taken into account, there is no evidence that lower GDP is associated with higher language endangerment.

Only road density and average years of schooling have a significant, worldwide association with language endangerment, above and beyond their covariation with other variables. Figure 3 maps the relative numbers of spoken first-language languages classified as Sustainable, Endangered, or Sleeping across Natural Earth regions, using the EGIDS scale and a database of 6,511 languages.

The stacked bars make regional differences visible, including substantial endangered representation in North America and very high sleeping representation in the Caribbean. This matters because regional history and geography can create associations—such as between polysynthesis and endangerment—that might otherwise be mistaken for causal effects.

Forms of future projection commonly employed in conservation biology appear to be of limited use in language endangerment and loss. Population viability analysis projects species’ trajectories by modelling birth and death rates, in order to evaluate species endangerment and inform management strategies.

However, this approach will not work for languages because vitality of languages is based on the transmission of language rather than human reproduction, so language speaker trends may be unconnected to population demography. The number of L1 speakers of the Australian language Bardi declined for over a century even as the Bardi population was growing, and the Bardi language is now critically endangered.

Language diversity and biodiversity show some striking similarities, including, unfortunately, their current catastrophic rates of loss. There is room for biologists and linguists to join forces to share useful tools and insights in understanding and protecting endangered species and languages.

However, tools from conservation biology should not be adopted uncritically into studies of language diversity without examining whether the method captures patterns and processes relevant to language endangerment and loss. Broad-scale analyses of language endangerment patterns do not provide a comprehensive picture of the current vitality and future prospects of any specific language, but they contribute to understanding patterns of language diversity and loss.

The central lesson is selective transfer: biological methods can reveal broad patterns and predictors of language endangerment, but language vitality depends on transmission and social context, so tools must be tested rather than adopted uncritically.