Section 02

Language Preservation through AI & Digital Archives

A language dies when parents stop speaking it to children — everything else is a symptom. This strategy attacks the causes: prestige, usefulness, and presence in the digital world where the next generation actually lives.

The diagnosis

Why are indigenous languages declining?

48.4%Assamese mother-tongue share in Assam, Census 2011 — from ~58% in 1991. Absolute speakers grew (~15.3 million), but the share keeps sliding.
~250Indian languages lost in the last five decades, by the People's Linguistic Survey of India's estimate — the Northeast is the highest-risk zone.
<50kSpeakers remaining for several Assam languages — Tai Phake, Tai Aiton, Tai Khamti number in the low thousands; UNESCO lists Deori and Mising as vulnerable-to-endangered.
~0.1%Rough share of web and AI training content available in Assamese versus its speaker population — the digital-presence gap that this strategy exists to close.

Language shift in the Northeast is not one problem but a chain of pressures, each individually rational for the family making the choice:

  • The prestige economy. English- and Hindi-medium schooling is read as the path to jobs; parents choose it out of love, not neglect. When the languages of income and the language of home diverge, home loses within two generations.
  • Urban migration. Moving to Guwahati, Bengaluru, or Delhi for work breaks the village-level density a small language needs for daily use — and distress migration (see Economy) accelerates it.
  • The digital vacuum. A teenager's world is a phone. If keyboards, memes, games, voice assistants, and search barely function in your language, the phone quietly teaches you that your language is for the past.
  • Oral tradition without archives. Much of the region's literature — Mising oi-nitom lyric poetry, Karbi sabin alun epics, Bodo oral histories, tribal ecological knowledge — exists only in living memory. Every elder who passes without being recorded is a library fire.
  • Script and standardisation friction. Several tribal languages still contest scripts or lack standardised orthography, which blocks textbooks, publishing, and Unicode-native software.

The lesson from every successful revival (see Case Studies): you do not save a language by pleading. You save it by making it useful, prestigious, and present — in schools, on screens, and in pay-cheques.

The strategy

Four programmes, in priority order

🎙️

1. The Northeast Digital Archive

A ten-year, community-run recording mission: oral epics, songs, rituals, ecological knowledge, and everyday speech from every language community — captured on video and audio with metadata, stored openly, mirrored internationally (following the model of SOAS's Endangered Languages Archive). Pay village youth as trained field recordists: preservation becomes employment.

🤖

2. Language AI as public infrastructure

Build on, not from scratch: India's Bhashini mission and AI4Bharat (IIT Madras) already ship open models — IndicTrans2 translation and IndicBERT — that include Assamese. The Assam move is a state Language Technology Mission that contributes corpora (from the archive above), fine-tunes speech-to-text, text-to-speech, and translation for Assamese and Bodo, Mising, Karbi and others, and publishes everything as open APIs any local startup can use. Iceland's national Language Technology Programme (2019–2023) proved a small language can buy its digital future for the price of a flyover.

📱

3. Daily-life presence

Keyboards, fonts, spell-checkers, and voice input for every scheduled and tribal language; Assamese-first interfaces on state digital services; Wikipedia and Wiktionary edit-a-thons; expansion of community dictionaries like Xobdo; children's content (a modern successor to the Xofura magazine tradition) and gaming/reels ecosystems that make the language cool where the audience actually is.

🏫

4. Intergenerational transmission

The Māori Kōhanga Reo insight: revival happens in early childhood or not at all. Mother-tongue-medium pre-primary and primary (backed by NEP 2020 §4.11), language-nest crèches run by elders, periodic Assamese-language quizzes and competitions, and adult night-school classes so parents can keep up with children.

The ownership angle

Every corpus, model, and archive built under this programme is owned by Assam-based institutions under open licences — never exclusively licensed away. Language data is the one raw material where Assam holds a natural global monopoly; treat it like the strategic asset it is.

Measurable targets

What success looks like, in numbers

IndicatorBaseline (2026)2030 target2035 target
Hours of archived audio/video across NE languagesFragmented, mostly private25,000 hrs, openly catalogued100,000 hrs; every living language covered
Open speech-recognition word-error rate (Assamese)Usable but weakNear-Hindi parityProduction-grade for Bodo, Mising, Karbi too
Children in mother-tongue-medium early educationMinority, declining+30% enrolmentMother-tongue pre-primary the default option statewide
Assamese Wikipedia articlesTens of thousands3× baselineLargest NE-language knowledge base; tribal-language editions active
Working digital keyboards/fontsAssamese only, patchyAll 8th-Schedule + major tribal languagesVoice input for every language with >100k speakers

Sources & further reading

  1. Census of India 2011, C-16 Language tables; Registrar General of India mother-tongue data 1991–2011.
  2. G.N. Devy (ed.), People's Linguistic Survey of India (2013–) — language loss estimates and NE volumes.
  3. UNESCO, Atlas of the World's Languages in Danger — entries for Deori, Tai Phake, Tai Aiton, Tai Khamti, Mising, and other NE languages.
  4. National Education Policy 2020, §4.11 — medium of instruction in home language/mother tongue until at least Grade 5.
  5. Bhashini / National Language Translation Mission (MeitY, 2022); AI4Bharat (IIT Madras) — IndicTrans2, IndicBERT open models.
  6. Icelandic Government, Language Technology Programme for Icelandic 2019–2023.
  7. Endangered Languages Archive (ELAR), SOAS University of London — community archiving methodology.
  8. Kōhanga Reo National Trust (NZ) — language-nest model documentation.