Can AI help us save endangered languages? - What in the World podcast, BBC World Service

Key Concepts

• Linguicide: The erasure and loss of languages due to external threats, colonialism, or socioeconomic pressures.
• Language Hierarchy: The societal perception that certain languages are more "prestigious" or economically valuable than others, leading to the marginalization of minority languages.
• Endangered Language: A language at risk of falling out of use, often identified by a lack of intergenerational transmission (parents not teaching children) and absence in public/educational spheres.
• AI Hallucination: The phenomenon where AI models generate inaccurate or fabricated information, particularly problematic when misidentifying or mistranslating endangered languages.
• Data Sovereignty: The ethical concern regarding who owns the data used to train AI models and whether the speaker communities have provided informed consent.

---

1. The Crisis of Language Loss

Linguists estimate that approximately half of the world’s languages could disappear by the end of this century. This process, termed "linguicide," is occurring at an unprecedented rate.

• Drivers of Loss: The primary variables identified by researchers are the expansion of infrastructure (building roads) and the rise of socioeconomic status. As communities integrate into the global economy, a "language hierarchy" often emerges, where dominant languages are prioritized for education and employment, relegating local languages to a status of lower importance.
• Impact: Language loss is not merely the loss of words; it involves the erasure of stories, songs, history, and unique worldviews. Research from British Columbia suggests a correlation between the loss of indigenous languages and higher youth suicide rates, indicating that language maintenance may be a vital component of community resilience and cultural identity.

2. Defining Endangered Languages

Contrary to popular belief, the number of speakers is not the sole indicator of endangerment.

• Sustainability: Some languages, such as the Karuk language in Northern California, were historically sustainable with only 1,500 speakers.
• Danger Signs: Linguists prioritize qualitative markers over raw numbers:
  • Are parents passing the language to their children?
  • Is the language present in public life?
  • Is the language integrated into the education system?

3. AI as a Double-Edged Sword

AI presents both a potential lifeline and a significant threat to linguistic diversity.

• Preservation Efforts: Projects are underway to build translation tools for under-resourced languages. For example, developers in Ghana are creating smartphone tools for local languages, and researchers are using AI to preserve Nüshu, an ancient language created by women in China, by training models on small datasets.
• Risks and Failures:
  • Inaccuracy: AI models often "hallucinate" or misidentify languages. For instance, a model misidentified the Emilian dialect as a "constructed language" (conlang) and incorrectly labeled the Manx word for Manx (Gaelg) as "English."
  • Linguistic Misinformation: AI often reinforces colonial biases by labeling indigenous languages as "dialects" rather than distinct languages, a practice historically used to devalue them.
  • Ethical Concerns: There are critical issues regarding data security, consent, and the risk that AI will simply replicate the inequalities of the "analogue world" rather than bridge them.

4. Framework for Future Protection

Sophia Smith Galer emphasizes that grassroots activism is necessary but insufficient on its own. A sustainable model for language preservation requires:

1. Institutional Support: Recognition of endangered languages in national constitutions and formal policy.
2. Ethical AI Development: Tech companies must invest in proper, expert-vetted tools rather than relying on flawed, automated systems.
3. Community-Led Initiatives: AI tools must be built based on the genuine needs and consent of the speaker communities, rather than being imposed from the outside.

5. Synthesis

The preservation of language is as critical to human society as biodiversity is to the natural world. While AI offers innovative technical pathways to document and teach endangered languages, it currently suffers from significant biases and accuracy issues. To prevent further linguicide, the global community must move beyond volunteer-led efforts and ensure that institutional power and ethical technology development work in tandem to protect the world's linguistic heritage.