The world's linguistic diversity is facing a significant threat, with nearly half of the approximately 7,000 documented languages considered endangered. This alarming rate of language loss, estimated to be one language every one to three months, surpasses the extinction rates observed in many animal species. Unlike biodiversity loss, which has seen statistically rigorous analysis, predictions of language loss have lacked such rigor. This study aims to address this gap by providing a global analysis to model current and future language endangerment patterns and compare the predictive power of various potential drivers of language loss. The analysis incorporates a broad range of factors, including demographic factors, linguistic resources, socioeconomic settings, language ecology, connectivity, land use, environment, climate, and biodiversity, overcoming statistical challenges such as phylogenetic non-independence, spatial autocorrelation, and covariation among variables. Furthermore, the model incorporates demographic and environmental variables that can be projected into the future, allowing for predictions of future language endangerment patterns in time and space. While language change is a natural process, colonization and globalization have dramatically accelerated language loss. This study seeks to identify widespread general factors contributing to endangerment to help predict which languages may face increasing threats in the future, complementing more focused qualitative studies on the topic.

Previous research has highlighted the significant threat to global linguistic diversity, emphasizing the high rate of language endangerment and loss. Studies have pointed to various factors contributing to this, including colonization, globalization, and socioeconomic shifts. However, previous analyses have often focused on limited sets of predictors and failed to account for important statistical challenges like phylogenetic non-independence and spatial autocorrelation. Some studies used speaker population size and geographic distribution as proxies for endangerment, overlooking the fact that small languages can be stable while large languages can be vulnerable if intergenerational transmission is weak. This study expands upon this prior work by utilizing a more comprehensive dataset and advanced statistical methods to more accurately model the complex interplay of factors leading to language endangerment.

The study analyzed data on 6,511 spoken languages, representing over 90% of the world's spoken languages. Fifty-one predictor variables were used, encompassing various aspects of language maintenance, including language transmission, language shift, and language policy. These variables covered demographic factors, linguistic resources, socioeconomic settings, language ecology, connectivity, land use, environment, climate, and biodiversity. The researchers addressed major statistical challenges by simultaneously accounting for phylogenetic non-independence, spatial autocorrelation, and covariation among variables. This was achieved using a novel statistical approach combining phylogenetic, distance, and contact matrices within an autoregressive ordinal probit model. The model was constructed using a stepwise selection procedure applied to a training dataset (two-thirds of the languages) and tested on the remaining one-third. The best model was chosen based on its predictive power, and model parameters were re-estimated using all 6,511 languages. The EGIDS (Expanded Graded Intergenerational Disruption Scale) score from Glottolog was used as the dependent variable, representing the language endangerment level. The analysis incorporated an interaction term between region and each independent predictor to account for regional variations. Finally, the study used demographic information, as well as projections for climate and land-use change, to predict future patterns of language endangerment.

The best-fit model, accounting for 34% of the variation in language endangerment, identified several key predictors. The number of first-language (L1) speakers emerged as the strongest predictor, highlighting the importance of speaker population size. However, the study surprisingly found that direct contact with neighboring languages is not a major driver of endangerment. Instead, increased road density, possibly facilitating population movement and language shift, was associated with increased endangerment. Higher average years of schooling also showed a strong association with greater endangerment, suggesting that formal education systems can negatively impact language diversity if bilingual education is not adequately supported. Additionally, a higher proportion of endangered languages in a region also increased the risk for individual languages, indicating the influence of broader regional factors. The model predicted a significant increase in language loss in the coming decades: at least a tripling of language loss in the next 40 years and a nearly fivefold increase in 'Sleeping' languages (languages with no living L1 speakers) by the end of the century, potentially resulting in the loss of over 1,500 languages. Geographical hotspots for future language loss include the west coast of North America, Central America, the Amazon rainforest, West Africa, the north coast of New Guinea, and northern Australia. The study also found regional variations in the influence of predictors, with factors like land use and climate showing stronger associations in specific regions.

The findings challenge the common perception that language contact is the primary driver of language loss. The study demonstrates that the relationship between language endangerment and contact is complex, suggesting that other socioeconomic and political factors often co-occur with contact but are not synonymous with it. The strong association between road density and endangerment highlights the indirect impact of infrastructure development on language vitality, suggesting that increased connectivity can lead to the erosion of heritage languages. The negative association between higher years of schooling and language maintenance underscores the importance of policies promoting bilingual education to safeguard linguistic diversity. The regional variations in the importance of different predictors highlight the need for context-specific interventions to support language vitality. The alarming predictions of future language loss underscore the urgency of proactive measures to prevent the irreversible loss of linguistic heritage.

This study provides a statistically robust global analysis of language endangerment, identifying key predictors and offering alarming predictions for the future. The results challenge common assumptions about the drivers of language loss and highlight the critical role of road density and educational policies. Urgent investment in language documentation, bilingual education programs, and community-based initiatives is crucial to mitigate the impending loss of linguistic diversity. Future research should focus on more fine-grained analyses of regional variations and the long-term impacts of specific interventions on language vitality.

The study acknowledges limitations due to data availability and the inherent complexities of language endangerment. The reliance on national averages for certain socioeconomic variables might mask variations within countries. Furthermore, historical factors and local-scale dynamics not fully captured in the global data could influence language vitality. While the model includes variables representing change over time, historical processes are not directly captured. The study’s predictive power is also limited by the assumption of current trends continuing into the future, which may not hold true. The model focuses on L1 speakers, overlooking the potential for language maintenance through L2 use or revitalization efforts.