Artificial intelligence is creating a new global linguistic hierarchy

Artificial intelligence (AI) has the potential to transform healthcare, education, governance and socioeconomic equity, but its benefits remain concentrated in a small number of languages (Bender, 2019; Blasi et al., 2022; Joshi et al., 2020; Ranathunga and de Silva, 2022; Young, 2015). Language AI - the technologies that underpin widely-used conversational systems such as ChatGPT - could provide major benefits if available in people’s native languages, yet most of the world’s 7,000+ linguistic communities currently lack access and face persistent digital marginalization. Here we present a global longitudinal analysis of social, economic and infrastructural conditions across languages to assess systemic inequalities in language AI. We first analyze the existence of AI resources for 6003 languages. We find that despite efforts of the community to broaden the reach of language technologies (Bapna et al., 2022; Costa-Jussà et al., 2022), the dominance of a handful of languages is exacerbating disparities on an unprecedented scale, with divides widening exponentially rather than narrowing. Further, we contrast the longitudinal diffusion of AI with that of earlier IT technologies, revealing a distinctive hype-driven pattern of spread. To translate our findings into practical insights and guide prioritization efforts, we introduce the Language AI Readiness Index (EQUATE), which maps the state of technological, socio-economic, and infrastructural prerequisites for AI deployment across languages. The index highlights communities where capacity exists but remains underutilized, and provides a framework for accelerating more equitable diffusion of language AI. Our work contributes to setting the baseline for a transition towards more sustainable and equitable language technologies.

💡 Research Summary

The paper provides a comprehensive, data‑driven investigation of how artificial intelligence—specifically large language model (LLM)‑based conversational systems—has reshaped the global linguistic landscape and created a new hierarchy of language access. Using a longitudinal dataset that combines Hugging Face model and dataset metadata with monthly Wayback Machine snapshots from 2020 to 2024, the authors examine AI resource availability for 6,003 languages that have at least one documented lexical item. Their analysis shows that the distribution of language models and datasets follows a power‑law (Zipf) pattern with an exponent close to 1, but English is a massive outlier, possessing orders of magnitude more resources than predicted. While English alone sees up to 50,000 new models per year, the average under‑resourced language gains only about 4.2 models annually, indicating a “rich‑get‑richer” dynamic that intensifies rather than mitigates inequality.

A second line of inquiry correlates the number of models per language with speaker population size. An ordinary least‑squares regression yields a modest positive relationship (β₁ = 0.312, R² = 0.304), yet many high‑population languages in Sub‑Saharan Africa, South Asia, and the Middle East fall far below the regression line, highlighting systemic bias. Conversely, several European minority languages (Finnish, Inari Sámi, etc.) and even extinct languages such as Latin and Ancient Greek appear over‑represented, reflecting historical prestige, copyright‑free corpora, and strong institutional support for language preservation. This pattern demonstrates that the disparity is not a simple Global‑North versus Global‑South divide; rather, it is driven by a complex mix of colonial legacies, academic incentives, and policy priorities.

The authors then compare diffusion dynamics of language AI with classic ICT diffusion (mobile phones, personal computers, electric vehicles). While traditional technologies follow the classic S‑shaped Gompertz curve—slow uptake, rapid growth, and eventual saturation—language models exhibit an early hyper‑growth phase (displacement rate b = 0.927, growth constant c = 1.31, R² = 0.866). After this burst, the growth curve flattens, but this deceleration reflects a consolidation of dominance by high‑resource languages rather than a catch‑up by low‑resource ones. The diffusion is “top‑down”: models are trained simultaneously on hundreds of languages using global web data, so inclusion is dictated by data availability and commercial priorities, not by organic community‑driven adoption. This creates a two‑stage process—initial hype‑driven expansion followed by a lock‑in phase—that entrenches linguistic and socioeconomic gaps.

To move from diagnosis to actionable guidance, the paper introduces the Language AI Readiness Index (EQUATE). EQUATE aggregates 25 indicators spanning technological infrastructure (internet penetration, computing capacity), socioeconomic factors (education levels, GDP per capita), and data ecosystem metrics (availability of annotated corpora, open‑source contributions). Each language receives a score from 0 to 100, allowing stakeholders to identify “high‑potential but under‑utilized” communities (e.g., many Indian languages with strong infrastructure but scarce AI datasets) and “low‑readiness” regions requiring foundational investment (e.g., many Sub‑Saharan languages lacking both connectivity and data). The index is designed for policymakers, NGOs, and private firms to prioritize funding, guide multilingual model development, and monitor progress toward more equitable AI diffusion.

In conclusion, the study demonstrates that AI‑driven language technologies are not neutral tools; they actively reproduce and amplify existing digital inequities, forming a new global linguistic hierarchy. The concentration of resources in a handful of high‑resource languages, the mismatch between speaker populations and model availability, and the atypical diffusion pattern all point to systemic bias. Addressing these challenges requires more than scaling up model counts; it demands targeted, data‑sovereign investments that respect the socioeconomic and infrastructural realities of each language community. By providing both a rigorous empirical baseline and the EQUATE framework, the authors offer a roadmap for steering the AI ecosystem toward a more inclusive, multilingual future.