With the rapid rise of artificial intelligence (AI) as a major driver of technological progress, democratizing access to digital opportunities through greater representation and inclusion of diverse languages has never been more important.  

Linguistic diversity remains a defining feature of our global cultural landscape. An estimated 7,000 languages are spoken worldwide, each a signifier of heritage and identity. Today, linguists estimate that most of the world’s unique languages face severe endangerment or extinction by the end of this century, threatening a loss of entire knowledge systems and jeopardizing the preservation of much of humanity’s rich cultural heritage. The digital realm is facing similar linguistic imbalances, with most online activity taking place in high-resource languages, such as English.  

Low-resource languages—such as Twi (mostly spoken in Ghana), Sinhala (mostly spoken in Sri Lanka) and Quechua (an Andean-Equatorial language)—appear on less than 0.1 percent of websites. This is creating a digital divide that limits equitable access to digital innovation. Many communities lack fluency in the second or third languages needed to access essential online services such as e-learning or telemedicine. 

AI models underperform for speakers of low-resource languages 

Large language models (LLMs) rely heavily on data scraped from the Internet, where the world’s highest-resource languages are overrepresented. As a result, AI models underperform for the 1.2 billion native speakers of low-resource languages. Even when AI systems can process queries in low-resource languages, their outputs tend to be slower, less culturally relevant, and insufficiently covered by model safeguards. Voice capabilities are also limited—presenting a significant accessibility barrier, particularly for communities with strong oral traditions or low literacy.  

The consequences of this disparity are far-reaching—and disproportionately impact communities in the Global South. Closing the language divide in AI is critical to catalyse locally relevant innovation, empower marginalized communities, and ensure no one is left behind no matter the language they speak. 

Addressing the divides with diverse projects on the ground 

Despite these challenges, promising initiatives are emerging across public, private, non-profit, and international actors to address the lack of low-resource language representation in AI.  

On the African continent, organizations such as Lelapa AI, Masakhane, GhanaNLP, or Professor Vukosi Marivate’s Data Science for Social Impact Research Group, are making significant progress in incentivizing community efforts to deploy datasets for African low-resource languages and integrating them into AI systems.   

Global organizations like Lacuna Fund, GIZ FairForward, International Development Research Centre (IDRC), Mozilla Common Voice, Google Research, and OpenAI are supporting these efforts and contributing resources to making low-resource language data and African Language AI products more widely available.