By Stuart Kerr, Technology Correspondent
Published: 27 July 2025
Last Updated: 27 July 2025
Contact: liveaiwire@gmail.com | Twitter: @LiveAIWire
Author Bio: About Stuart Kerr
Across the globe, thousands of languages teeter on the edge of extinction. Some have just a few speakers left. Others exist only in fragments—words scrawled in notebooks or echoing faintly in childhood memories. But a growing number of researchers believe artificial intelligence may hold the key to keeping them alive.
Algorithms as Anthropologists
AI is being trained to do what once took teams of linguists decades—preserve and revive lost languages. Projects like Dartmouth’s NüshuRescue demonstrate how large language models can reconstruct forgotten scripts from shockingly little data. In this case, just 35 example sentences were enough to begin teaching a neural network how to write in Nüshu, a centuries-old script used exclusively by women in parts of China.
Academic work like the arXiv paper on NüshuRescue explains how GPT-4 Turbo models were fine-tuned with historical texts to produce accurate, stylistically faithful renditions of the script.
Meanwhile, UNESCO’s consultation on indigenous language preservation explores broader applications: using AI for OCR, machine translation, and even spell-checking in languages long ignored by tech giants.
From Data Fragments to Digital Fluency
Beyond linguistics, AI is digitising entire cultures. In “AI Digitising Cultural Heritage”, LiveAIWire reported on how museums are training neural networks to categorise artefacts, translate ancient inscriptions, and upscale archival images to preserve detail that would otherwise fade away.
Google’s Woolaroo experiment offers another creative application—an AI tool that translates photos into endangered languages like Māori, Tamazight, and Louisiana Creole. It’s part dictionary, part camera, part cultural revivalist.
These projects are more than digital novelties. They serve as proof that AI can help transform endangered languages into living, searchable, speakable forms again.
Cultural Context in Code
But is AI capable of nuance? Can a model trained on word frequency also understand spiritual symbolism, humour, or ancestral pride? A growing body of researchers say it can—if designed correctly.
The arXiv study on context-aware AI translation explores how multi-agent systems can preserve subtle cultural meanings during translation. These models attempt not just to parse words, but to infer how meaning shifts based on who is speaking and why.
Still, challenges remain. As explored in LiveAIWire’s "AI and Conspiracy Culture", algorithms can also distort narratives, especially when fed imbalanced or politicised data. Cultural preservation, then, isn’t just about training smarter models—it’s about curating the right datasets.
Identity, Memory, and the Machine
What happens when AI becomes the last fluent speaker of a language? Or when a digital archive becomes more comprehensive than any living memory? The philosophical implications are staggering.
As discussed in "The AI Identity Crisis", the line between human and machine memory is blurring. What began as an effort to preserve the past could end up reshaping how we define culture itself.
The UNU paper on AI and language learning makes a clear recommendation: community-driven training, open-source datasets, and transparency in model behaviour must remain foundational principles if AI is to amplify rather than overwrite indigenous knowledge.
Code-Switching the Future
AI is not a panacea for cultural loss. But when handled with care—and built with communities, not just coders—it can become a powerful ally in the fight to preserve humanity’s most fragile expressions.
In the museum of the future, AI might not just catalogue our past. It may help us speak it again.
About the Author
Stuart Kerr is the Technology Correspondent for LiveAIWire. He writes about artificial intelligence, ethics, and how technology is reshaping everyday life. Read more.