Chris Bartley, a computer science student at Sheffield University, is deploying AI text-to-speech and speech recognition systems to preserve Manx Gaelic, a critically endangered language with approximately 2,200 speakers remaining on the Isle of Man.
The project focuses on converting spoken Manx into searchable text and enabling text-to-speech output for screen readers—infrastructure that transforms language preservation from unpaid volunteer work into scalable digital tooling that supports learning, accessibility, and cultural heritage without extracting from the community.
UNESCO Declared Manx Extinct, Then Reversed Classification
Manx is a Goidelic language closely related to Irish and Scottish Gaelic. UNESCO declared it extinct in 2009 despite hundreds of active speakers on the Isle of Man, drawing sharp criticism from historians and linguists who pointed out that people were still having productive conversations in the language daily.
UNESCO reversed its classification to "critically endangered" after backlash, but the episode highlighted how languages without digital presence risk institutional erasure. The 2021 census counted roughly 2,200 people with knowledge of Manx—up from 1,823 in 2011—indicating modest revival driven by language schools, radio broadcasts, and cultural preservation efforts.
However, the scarcity of fluent speakers means every hour spent manually transcribing archival recordings represents scarce human capital diverted from higher-value work like teaching, creating lessons, or coaching pronunciation. This is where AI shifts the burden from people to infrastructure.
Practical Implementation: Machine Draft, Human Final
Bartley's speech recognition model targets the highest-friction tasks: rough transcription that fluent reviewers clean up rather than attempting to replace human fluency entirely. The workflow operates as "machine draft, human final," shortening the time native speakers spend on low-value typing and pushing effort toward teaching and content creation.
The system supports faster transcription of archival audio, provides pronunciation feedback for language learners, and critically improves accessibility for visually impaired users through text-to-speech functionality. Without speech synthesis, web pages, learning materials, and digital messages remain inaccessible to screen reader users unless manually narrated—a bottleneck that effectively excludes disabled people from participating in the language community.
Modern AI handles these tasks by learning from limited datasets, a crucial capability for low-resource languages. Manx lacks the massive training corpora available for dominant languages like English or Mandarin, so the model must extract patterns from smaller archives of recordings, many captured on older equipment with variable audio quality.
Why Accuracy Matters More Than Speed
Inaccurate AI translation or transcription actively harms language preservation by spreading errors that learners internalize as correct usage. One researcher demonstrated this risk by inputting "Men and women hold on to the land" in Te Reo Māori into a popular translation tool, which returned "white man, white woman, keep the land"—a catastrophic mistranslation that inverts meaning and introduces colonial framing absent from the original.
For Manx, maintaining accuracy means community control over training data, transparent model development, and human verification of outputs before deployment. Bartley's student-led approach prioritizes these constraints over flashy demos or scale, focusing on shipped tooling that people can test, critique, and improve rather than extracting data for external commercial purposes.
Cultural Heritage Gains Digital Support
Language preservation extends beyond saving words—it protects identity, indigenous knowledge, spiritual beliefs, and social structures that vanish when languages die. Of the world's approximately 7,000 languages, UNESCO estimates 43% are endangered and could disappear by the end of the century.
AI cannot replace fluent speakers or the cultural context they provide, but it can reduce the operational friction that makes preservation work exhausting and unsustainable. By automating transcription rough drafts and enabling text-to-speech accessibility, AI shifts scarce human attention from mechanical tasks toward high-impact teaching and content creation.
The Manx preservation effort demonstrates how AI serves communities best when built as infrastructure for people already doing the work rather than as replacement technology that claims to automate away human expertise. As Bartley's project gains usage, the next challenge becomes ensuring the tools remain community-controlled rather than captured by external platforms with different incentives.
