Abstract
Of the approximately seven thousand languages currently spoken, somewhere between forty and fifty percent are projected by various measures to fall out of intergenerational transmission within this century. The headline figures are familiar; the structural questions behind them are less so. Endangerment is not a single phenomenon but a family of distinct trajectories, each with its own causal architecture, its own characteristic timeline, and its own set of feasible interventions. A language with two hundred elderly speakers in a community committed to revitalization is in a fundamentally different situation from a language with two thousand fluent speakers whose children are not learning it, and both differ from a language with twenty thousand speakers in a state that actively suppresses its use. This paper proposes a working typology of contemporary language endangerment, surveys the documentation and revitalization responses appropriate to each type, and considers what can realistically be preserved when full transmission cannot be maintained. The argument is neither that all languages can be saved nor that the situation is hopeless, but that a clearer understanding of the causal structure of endangerment would substantially improve the allocation of the limited resources available.
I. The Inadequacy of Single-Variable Measures
The dominant frameworks for assessing language endangerment — UNESCO’s nine-factor scale, Fishman’s Graded Intergenerational Disruption Scale (GIDS), and Lewis and Simons’s Expanded GIDS (EGIDS) — share a common architecture. They place languages on a single ordinal scale running from full vitality to extinction, with intermediate stages defined primarily by patterns of intergenerational transmission. These frameworks are useful and have done substantial work in making the field tractable, but they obscure differences that matter for response. Two languages at the same EGIDS level may face entirely different threats, respond to entirely different interventions, and require entirely different documentation strategies.
A more useful typology disaggregates endangerment along several dimensions: the size and demographic structure of the speech community, the proximate cause of transmission failure, the political and legal environment, the degree of existing documentation, the typological distinctiveness of the language, and the community’s own orientation toward the language’s future. These dimensions are not independent — a small community in a hostile state with no documentation faces compounded difficulties — but they are separable, and separating them clarifies what is actually at stake in any given case.
II. A Typology by Causal Architecture
The following types are not mutually exclusive — many endangered languages exhibit features of more than one — but they capture the dominant causal structures observed in current cases.
Type 1: Demographic collapse. The speech community is small in absolute terms, often under a few hundred speakers, and the population itself is not reproducing at replacement rates regardless of language choices. The language declines because the community declines. This pattern is characteristic of small Indigenous communities in the Americas, Siberia, and parts of Australia where historical violence, disease, and reproductive disruption reduced populations to thresholds from which recovery is demographically difficult. Examples include several languages of the Pacific Northwest, several Yeniseian languages, and a number of Australian languages where the speaker population is now elderly and the community itself is small.
Type 2: Transmission interruption without demographic collapse. The community is stable or growing, but parents are not transmitting the language to children. This is the most common pattern globally and characterizes the majority of endangered languages by count. The community has shifted to a dominant language for reasons that range from economic opportunity to social stigma to the practical difficulties of raising children bilingually in environments structured around the dominant language. The language remains widely known among older adults but is no longer the language of childhood. Examples are spread across every continent: Welsh in some communities, many Indigenous languages of Latin America, Sámi languages, many languages of the Philippines and Indonesia, and a substantial portion of African languages where shift to a regional or colonial language is in progress.
Type 3: Active suppression. The state or dominant institutions actively discourage or prohibit the language’s use in education, public administration, media, or domestic life. Suppression varies in intensity from formal legal prohibition to informal stigma enforced through schooling and employment practices. The language may have a substantial speaker population that is being driven underground and whose intergenerational transmission is being deliberately disrupted. The Kurdish languages in several states, Uyghur in present-day China, several languages of central Africa during specific political periods, and historically many Indigenous languages under boarding-school regimes fit this pattern.
Type 4: Catastrophic displacement. The community has been physically displaced from its territory through war, environmental disaster, or forced migration, and the dispersal disrupts the social structures that supported the language. Speakers may survive in numbers, but they survive scattered across host populations where the language has no domain of use. Several languages of Syria and Iraq are now in this position, as are some languages of Sudan, Myanmar, and parts of Central America. The Yazidi situation has produced this kind of dispersal for the Kurdish dialects of the affected communities, and similar patterns appear in other conflict zones.
Type 5: Domain contraction. The community remains intact and the language is being transmitted, but the domains in which it is used are shrinking. The language survives in the home and in some traditional practices but has been displaced from work, education, media, and formal institutions. This pattern is intermediate between vitality and decline; it can persist for generations or can collapse rapidly into Type 2 if the remaining domains are lost. Many regional languages of Europe, including some celtic languages in particular configurations, are in this state, as are many languages of South Asia where state and educational use has shifted to a dominant regional or national language.
Type 6: Speaker dispersal in a globalized economy. The community remains in its territory in some form, but a substantial portion of the working-age population has migrated to urban centers or to other countries for economic reasons, leaving an elderly population at home and a younger population abroad in environments where the language has no community to support it. The migration is not catastrophic in the sense of Type 4 — it is voluntary and economically motivated — but its effects on transmission can be similar. Many languages of the Pacific, the highlands of Mexico and Guatemala, and rural China face this configuration.
Type 7: Climate and environmental displacement. The community’s territory is becoming uninhabitable through sea-level rise, desertification, or ecosystem collapse, forcing migration on a scale and timeline that the community cannot manage. Several Pacific island language communities face this directly, as do communities in low-lying areas of South Asia and parts of the Arctic where ice loss is changing the viability of traditional subsistence patterns. This type is small in current count but is projected to grow substantially over the coming decades.
Type 8: Stable diglossia under pressure. The language coexists with a dominant language in a stable functional distribution that has persisted for generations, but external pressures are destabilizing the equilibrium. The language has not been actively suppressed and the community is intact, but new economic, educational, or media pressures are eroding the boundary that previously protected the language’s domains. This pattern characterizes some languages of South Asia, parts of Switzerland and the Iberian peninsula, and parts of West Africa.
III. A Second Axis: Documentation State
Cross-cutting the causal typology is a second axis that determines what can be done in the time remaining: the existing state of documentation. Endangerment without documentation is qualitatively different from endangerment with documentation, because the question of what can be preserved if revitalization fails is not the same in the two cases.
State A: Comprehensive documentation. A reference grammar, a substantial dictionary, an extensive text corpus, and audio-video recordings of varied registers and genres exist. The language could in principle be relearned by community members from the documentation, even if the last fluent speaker died tomorrow. This is the situation of perhaps a few hundred endangered languages globally — those that have been the focus of sustained academic or missionary linguistic work over decades.
State B: Partial documentation. A grammar of some kind exists, a word list or partial dictionary, and some recordings, but coverage is uneven. Specific domains, registers, or genres may be entirely absent from the record. Several thousand endangered languages fall in this state, with the partial documentation often produced decades ago by single researchers and not updated.
State C: Minimal documentation. A short word list, perhaps a sketch grammar, possibly a few recordings. Many endangered languages, especially in New Guinea, the Amazon basin, and parts of Africa, are in this state. The documentation is insufficient to support full revitalization or to fully characterize the language’s structure.
State D: Effectively undocumented. Names of the language, geographic location, perhaps a few words in patrol-officer or anthropologist field notes. Some unknown number of languages are in this state — by definition we have only fragmentary information about how many — and time is short.
The documentation state interacts with the causal type to determine the realistic intervention space. A Type 1 language in State A can lose its last fluent speakers without losing the language itself in any final sense; the documentation supports later relearning if the community decides to pursue it. A Type 1 language in State D will simply disappear: there will be no trace from which to recover it. A Type 3 language in State A may survive politically what it cannot survive demographically. A Type 5 language in State C is in a more precarious position than a Type 5 language in State A, even if their current speaker numbers are identical.
IV. Community Orientation as a Third Axis
A third axis, often underweighted in academic frameworks, is the orientation of the community itself. Communities differ in whether they want their language preserved, in whether they want it transmitted to children, in whether they want outsiders involved in documentation, and in what they understand the language to be for. These orientations are not external variables that documentation programs can ignore; they determine what is possible and what is appropriate.
Some communities have made deliberate decisions to shift to a dominant language and consider the heritage language a part of the past. Some communities are actively committed to revitalization and have organized themselves to pursue it. Some communities are divided, with internal disagreement about whether revitalization is desirable or feasible. Some communities want documentation but not revitalization, treating the language as an inheritance to be preserved but not necessarily lived. Some communities want revitalization but not documentation, distrusting the academic and institutional apparatus that documentation entails.
These differences are real and matter for what interventions are appropriate. The same kind of project that would be welcomed in one community would be intrusive in another. The same kind of revitalization program that would succeed in one community would fail in another because the underlying social commitment is absent. Effective work on endangered languages is not a matter of applying generic methods to whatever language is at hand; it is a matter of matching methods to the specific configuration of community, language, and circumstance.
V. Documentation Strategies by Type
Given the typology, documentation priorities can be matched to circumstance rather than treated uniformly.
For Type 1 languages with elderly speakers, the priority is comprehensive recording while the speakers are still living and able to work productively. Recording sessions should aim at variety of register and genre rather than only the standard elicitation list: narrative, conversation, song, ritual language if the speakers are willing, technical vocabulary from traditional subsistence, kinship and place-name knowledge, and whatever metalinguistic commentary the speakers can offer. The window is short and does not reopen. Working with elderly speakers requires attention to fatigue, health, and the ethics of asking people to perform their language under conditions of awareness that the recording is for posterity.
For Type 2 languages with transmission interruption, the documentation question is different. Speakers exist in numbers; the question is what kind of documentation supports revitalization rather than only preservation. Pedagogical materials, teacher-training resources, and corpora suitable for language-acquisition contexts matter as much as reference grammars. Documentation work that does not feed back into community use is less valuable here than documentation work that produces materials the community can use to teach the language to children.
For Type 3 languages under suppression, documentation may need to be conducted under constraints that ordinary academic work does not face. Records may need to be held in locations outside the suppressing state. Speakers may need protection. The documentation itself may be politically sensitive. The standard methods of academic linguistics, designed for cooperative environments, sometimes need adaptation. In some cases the documentation that exists has been produced by diasporic communities working in safer locations, and supporting that work may matter more than attempting fieldwork in the affected territory.
For Type 4 languages in catastrophic displacement, the documentation effort is a salvage operation. Speakers are scattered, social structures are disrupted, and the practical conditions for sustained work are absent. What can be done is to locate speakers in their dispersal, conduct recording sessions where possible, and accept that the documentation will be uneven and incomplete. Coordinating across diaspora locations is itself a non-trivial effort.
For Type 5 languages in domain contraction, documentation should attend specifically to the domains being lost. The vocabulary of traditional subsistence, the language of ritual or religious life, the technical terminology of crafts, the language of children’s games and pedagogy — these are the domains that disappear first when a language is pushed back into the home. They are also the domains that are typically least represented in standard linguistic documentation, which tends to focus on grammar and basic vocabulary rather than specialized lexicons.
For Type 6 languages experiencing speaker dispersal, documentation work has the unusual feature that the speakers are in two or more locations simultaneously, often in locations where one set of speakers is elderly and rooted while another set is younger and mobile. Effective work captures both populations, recognizing that the variety spoken in the homeland may differ in significant ways from the variety being maintained or partially maintained in diaspora.
For Type 7 languages facing climate displacement, the timeline question is acute in a different way than for Type 1. The community will exist after displacement, but the conditions of its existence may not support the language. Documentation should be planned with the displacement in view, prioritizing materials that will be usable by a dispersed and possibly diminishing community in the decades after relocation.
For Type 8 languages in destabilizing diglossia, documentation work has a longer horizon and can be more comprehensive, but the destabilization can accelerate in ways that compress the timeline. Monitoring the changing functional distribution and adjusting documentation priorities as domains are lost is part of the work.
VI. Revitalization Strategies by Type
Documentation preserves; revitalization restores. The two are connected but distinct, and the strategies that work for revitalization vary substantially by type.
For Type 1 languages, full revitalization in the sense of restoring intergenerational transmission to a viable scale is in many cases not realistic. Demographic constraints alone make it unlikely. What can be pursued is community-scale relearning: a smaller number of community members, often adults rather than children, gaining functional fluency through intensive programs, and using the language in specific community contexts even if it does not return to being the language of daily life for a majority. This is not failure; it is a different and more realistic goal.
The Master-Apprentice model, developed for several California languages, is well suited to this configuration. Pairs consisting of a fluent elder and a committed adult learner work together intensively over months or years, with the learner gaining functional fluency and the language continuing to be transmitted, even if to a smaller cohort than would be needed for full demographic vitality.
For Type 2 languages, where the demographic base exists but transmission has failed, the strategies focus on rebuilding transmission. Immersion preschools and elementary schools — the kohanga reo model from Maori, the punana leo model from Hawaiian, the various immersion school networks now operating in Wales, Brittany, and elsewhere — have shown that it is possible to interrupt the interruption. The conditions for success are demanding: a critical mass of committed parents, sufficient funding, qualified teachers (which is often the binding constraint), curriculum and materials, and the persistence to sustain the program for the multiple generations required to see results.
Adult education programs are necessary alongside child-focused programs, because children growing up in immersion settings need adults around them who can speak the language. The Welsh experience has shown that adult fluency can be built at scale if the institutional commitment is present, but also that the adult learners do not always become primary domestic speakers, which limits the effects on intergenerational transmission unless the school programs continue.
For Type 3 languages, revitalization is shaped primarily by the political environment. Where suppression eases or ends, revitalization can move quickly because the underlying community is often intact. The Basque case, the Catalan case, and several post-Soviet cases illustrate the possibility. Where suppression continues, revitalization is largely conducted in private and in diaspora, with public revitalization waiting for political change.
For Type 4 languages in displacement, revitalization is constrained by the disrupted community structures. Where diaspora communities can establish institutions in their new locations — schools, cultural centers, religious institutions that use the language — partial revitalization can proceed. The conditions are difficult, and the language often becomes a heritage language for the second generation rather than a primary language, but the trajectory is not always toward complete loss.
For Type 5 languages in domain contraction, revitalization often takes the form of expanding the language back into domains it has lost: media, education, public administration. This is the standard trajectory of modern minority-language policy in supportive states, and it can succeed when the political and economic commitment is present. The Welsh, Catalan, and Basque experiences again provide models, though the underlying community vitality in those cases was higher than in many comparable situations elsewhere.
For Type 6 languages in dispersal, revitalization is complicated by the geographic split. Effective programs often involve some combination of homeland-based work and diaspora-based work, with technology making it possible to maintain connections that would have been impossible in earlier eras. Online classes, video communication with elder speakers, and digital materials produced in the homeland and used in diaspora are now standard tools.
For Type 7 languages facing climate displacement, revitalization planning is necessarily forward-looking. Communities planning for relocation can plan for language continuity in the new location, choosing relocation patterns that keep the community concentrated rather than scattered, and establishing the institutions needed to sustain the language in advance of the move. This is a difficult and politically charged undertaking, but it is more tractable than attempting reconstruction after dispersal has occurred.
For Type 8 languages in destabilizing diglossia, revitalization often takes the form of stabilizing the diglossic equilibrium rather than expanding the language’s domains. The goal is to prevent the slide from Type 8 to Type 5 or Type 2, which is a different goal from active expansion and requires different policy tools.
VII. What Can Be Preserved When Revitalization Fails
For some languages, full revitalization will not happen. The community may not want it, the demographic base may not support it, the political environment may not permit it, the resources may not be available. The honest question is what can be preserved when transmission cannot be maintained.
Several distinct things can be preserved through documentation alone. The grammar of the language — its phonology, morphology, syntax, and the principles by which they interact — can be preserved in reference works that capture its structure for future scholarship and for any later attempt at relearning. The lexicon can be preserved in dictionaries that capture not only the inventory of words but their semantic ranges, their collocations, and their cultural associations. The texts of the language — narratives, songs, ritual speech, technical discourse — can be preserved in audio-video recordings and transcriptions that constitute the language’s literary and oral heritage. The cultural knowledge embedded in the language — place names, kinship terms, taxonomies of plants and animals, technical vocabularies of traditional crafts — can be preserved as a cultural and scientific record even when the language as a daily medium is gone.
The Hebrew case demonstrates that languages preserved comprehensively in documentation can in principle be returned to spoken use even after long periods of disuse, though the conditions that supported the Hebrew revival are not generally available. More commonly, documentation supports partial relearning, ceremonial use, scholarly access, and the maintenance of the language as a recognized inheritance even if not as a living daily medium. This is not the same as language vitality, but it is not nothing, and it preserves possibilities that complete loss forecloses.
The communities themselves often distinguish between language use and language inheritance in ways that academic frameworks have been slow to recognize. A community may consider its language preserved when comprehensive documentation exists and when ceremonial and identity functions continue, even if daily transmission has ended. This is a defensible position and one that documentation programs should be prepared to support rather than override.
VIII. The Resource Question
The global resources available for endangered language work are small relative to the scale of the need. Endangered Languages Documentation Programme, the Endangered Languages Project, the Documentation of Endangered Languages program at the National Science Foundation, various foundation initiatives, and the work of several hundred academic linguists worldwide constitute the formal infrastructure. Total funding is probably well under one hundred million dollars per year globally, distributed across thousands of languages.
This is structurally inadequate for comprehensive work on every endangered language. Triage is unavoidable. The question is what principles should govern it.
Several principles can be defended. Languages with elderly fluent speakers and no younger speakers should be prioritized for documentation because the window is closing. Languages that are typologically distinctive — that exhibit features rare or unattested elsewhere — should be prioritized because their loss subtracts unique information from the human record. Languages whose communities have organized for their preservation should be prioritized because the work has a higher probability of bearing fruit. Languages in catastrophic situations — Type 4 displacements, Type 7 climate cases — should be prioritized because the timeline is compressed by external forces. Languages that lack any prior documentation should be prioritized over languages with substantial existing documentation, even where the current speaker numbers favor the better-documented case.
These principles sometimes conflict, and the conflicts have to be adjudicated case by case. The point is that allocation should be deliberate rather than driven by accidents of researcher interest, geographic accessibility, or institutional convenience. A coordinated global program for endangered language work, attentive to typological coverage and to the time-sensitivity of different cases, would extract more value from the available resources than the current uncoordinated arrangement.
IX. The Role of Community Researchers
The shift over the past three decades from documentation conducted exclusively by outside academic researchers to documentation conducted in partnership with community researchers, and increasingly by community researchers themselves, is one of the genuinely positive developments in the field. The reasons are several. Community researchers have access to speakers and registers that outsiders do not, can sustain documentation work over longer periods than externally funded research grants permit, and produce materials that are more directly usable in community revitalization efforts.
The implications for the resource question are significant. Investment in community researcher training — in linguistics, in documentation methods, in archival practice — multiplies the documentation capacity available for any given language and reduces the dependence on the limited supply of outside academic researchers. Programs like the Indigenous Language Institute, the Master of Linguistics for Indigenous Language Documentation programs at several universities, and various regional training initiatives have demonstrated the model. Scaling it would change what is achievable.
The shift also changes the character of the documentation produced. Community researchers tend to prioritize different things from outside researchers — more attention to specialized vocabulary, ritual language, place names, traditional knowledge, and pedagogical materials, less attention to the typological questions that drive academic interest. Both kinds of documentation are valuable, and the mix produced by collaborative work is generally richer than either alone.
X. Technology and Its Limits
Technology has changed what is possible for endangered language work. Audio-video recording is now affordable enough that comprehensive multimedia documentation is feasible where it once would have been prohibitive. Digital archiving makes records accessible across geographic distance. Online platforms allow scattered speakers and learners to maintain contact. Mobile applications can deliver language-learning content to populations that cannot attend in-person classes. Speech recognition, machine translation, and automated transcription are beginning to be usable for low-resource languages, though they remain bottlenecked by training data.
The temptation is to overstate what technology contributes. Recording equipment does not produce documentation; it produces recordings, which then have to be transcribed, glossed, and analyzed by trained people. Mobile applications do not produce speakers; they produce people who have used apps. Online classes do not replace the immersive social environments in which languages are actually transmitted. Technology amplifies effective human work and partially substitutes for some kinds of human work, but it does not substitute for the underlying human commitment, training, and social organization that language preservation requires.
The most promising technological developments are in tools that reduce the overhead of documentation work — forced alignment of recordings to transcriptions, semi-automated glossing, integrated archiving and metadata systems — and in tools that make existing documentation more accessible to communities. Less promising are claims that machine learning will somehow substitute for documentation that has not been produced, or that technological mediation can replace the social conditions of language transmission.
XI. Institutional Architecture
The institutions that currently support endangered language work are a patchwork. Academic linguistics departments in a small number of countries train documentation specialists. Several archives — ELAR in London, the Endangered Languages Archive at SOAS, AILLA at Texas, PARADISEC at Sydney, and others — preserve recordings and transcriptions. SIL International continues to do substantial work in many regions. Government and quasi-governmental bodies in some countries support work on languages within their borders. Indigenous-led organizations, where they exist and are funded, conduct community-scale work.
The patchwork has gaps. Many regions of the world have no archive within reasonable distance and no academic infrastructure for documentation training. Many languages fall between the priorities of all the existing institutions and receive attention only from individual researchers. Coordination across institutions is partial; the same language is sometimes documented redundantly while neighboring languages are entirely neglected.
A coordinated global infrastructure for endangered language work is not on the immediate horizon, and it may not be desirable in any case — centralized coordination of work that depends on local relationships has its own pathologies. What is more achievable is better information sharing, more deliberate triage among institutions that work in overlapping regions, and stronger investment in regional capacity in parts of the world where institutional infrastructure is currently thin.
XII. The Time Frame
The time frame within which the current wave of endangerment will play out is roughly the next three to five generations. Languages that lose intergenerational transmission in this generation will, absent revitalization, be effectively extinct as community languages within sixty to a hundred years, when the last people who learned them in childhood die. Languages that retain transmission but in shrinking communities may persist longer but on declining trajectories. The work that is done or not done in the coming decades will largely determine what survives.
This time frame is short enough that decisions matter and long enough that sustained programs are possible. It is not short enough to justify panic or long enough to justify complacency. The pattern of the past several decades, in which substantial progress has been made on some fronts while overall losses have continued, is likely to continue, with the proportion of progress to loss depending on how the work is organized.
XIII. Conclusion
The endangerment of nearly half the world’s languages over the coming century is a real phenomenon and not a rhetorical exaggeration. It is also not a single phenomenon but a set of distinct trajectories, each with its own causal structure, its own timeline, and its own set of feasible responses. Effective work on the situation depends on understanding which type of endangerment any given language faces, what state of documentation already exists, what the community itself wants, and what resources are available.
Universal solutions do not exist. Universal pessimism is also unwarranted. Some languages will be preserved in full vitality; some will be preserved through revitalization that produces partial vitality; some will be preserved as comprehensively documented inheritances even when daily transmission ends; and some will be lost without record. The proportion that falls into each category depends on choices being made now and over the coming decades, by communities, by institutions, and by the relatively small number of researchers and funders whose work makes the difference.
The honest framing is that the work is worth doing, that it can succeed in particular cases when matched to circumstances, and that the resources currently devoted to it are inadequate to the scale of the need. The question is not whether the situation can be saved in some global sense — it cannot, in the sense that not every language at risk will survive — but what can be saved in particular cases, and how the limited resources available can be allocated to save as much as possible. That is a tractable question, and its answers depend on the kind of careful disaggregation that this paper has attempted to begin.
