Adi Vaani

1. At a Glance

• Adi Vaani is the Ministry of Tribal Affairs' AI platform for translating and learning tribal languages. Its beta version was launched on 1 September 2025 [1][2].
• It was built by a consortium led by IIT Delhi, working with BITS Pilani, IIIT Hyderabad, IIIT Nava Raipur and the Tribal Research Institutes (TRIs) of Jharkhand, Odisha, Madhya Pradesh, Chhattisgarh and Meghalaya [2].
• It applies AI to low-resource languages. These are mostly spoken, have little written or digital data, and current language technology largely does not support them [1].
• UPSC relevance: it links tribal welfare (GS-II), society and cultural heritage (GS-I) and AI or digital inclusion (GS-III). It also ties into NEP 2020's push for mother-tongue education [1].

2. Why in the News

• The Working Group on Tribal Languages held its first meeting on 15 October 2026. It discussed making Adi Vaani the National Tribal Language Repository [1].
• The topic is in focus around Janjatiya Gaurav Diwas (15 November), the birth anniversary of Bhagwan Birsa Munda [1].
• As of 15 October 2026, text translation was available in 9 tribal languages. State TRIs had supplied about 13.9 lakh text sentences and about 300 hours of speech data [1].

3. Background & Evolution

• Rationale: language barriers cut tribal speakers off from digital services. AI translation is meant to open up education, healthcare information, public services and markets [1].
• Milestones:
• 1 Sept 2025: beta launched at an event in New Delhi, held under Janjatiya Gaurav Varsh [2].
• At beta: text-to-text and speech-to-text translation live for Santali, Mundari, Bhili and Gondi; Garo and Kui were nearing completion [2].
• Platforms at launch: an Android app and a live web portal [2].
• 2025: the Ministry included Adi Vaani in its Bharatiya Bhasha Utsav 2025 observance [4].

• 15 Oct 2026: first meeting of the Working Group on Tribal Languages; proposal to make Adi Vaani the National Tribal Language Repository [1].

• Related earlier initiatives:

• Janjatiya Gaurav Diwas, instituted in 2021 [1].
• NEP 2020, which favours the mother tongue as the medium of instruction [1].

• The TRI network, of which 29 TRIs receive financial help from the Ministry [1].

• Global context: the UN General Assembly declared 2022–2032 the International Decade of Indigenous Languages through resolution A/RES/74/135 [5]. UNESCO is the lead UN agency, working with UN DESA [5]. UNESCO has issued a Global Action Plan for the Decade [6].

4. Core Static Facts

Key terms:

• Corpus: the body of text and speech data for a language. Its size decides how well the model performs [1].
• Speech-to-text: converts speech into text, which supports documentation [1].
• Speech-to-speech: links recognition, translation and speech synthesis in one pipeline [1].
• Primers vs dictionaries: primers are basic resources for new learners; dictionaries give vocabulary and meanings [1].

5. Multi-Dimensional Analysis

Social

• It narrows the language-driven digital divide for STs, who make up about 8.6% of the population [1].
• It supports learning in the mother tongue. NEP 2020 recommends the home or local language as the medium of instruction at least till Grade 5, preferably till Grade 8 and beyond [1].
• It can help spread health information in tribal languages. The digest links this to the National Sickle Cell Anaemia Elimination Mission (2023), which aims to eliminate the disease as a public health problem by 2047 [1].

Cultural / Historical

• It shifts Adi Vaani from translation alone towards documentation and archiving. The proposed repository would hold corpora, primers, dictionaries and community stories [1].
• It digitises folklore and oral traditions for languages that are mostly passed down by speech [1][2].

Scientific / Technological

• Low-resource languages lack ready data, so the data has to be created from scratch [1].
• Models are retrained as more data comes in, so accuracy should keep improving [1].
• The target is about 1 lakh sentence pairs and 100 hours of speech per language. Corpus size drives model quality [1].

Ethical / Governance

• A community-based AI approach, with native speakers contributing and experts validating, protects authenticity and gives communities ownership [1].
• Open issues include data ownership, consent and the risk of standardising one dialect over others. This is analysis, not a claim from the sources.

Administrative / Federal

• State-level TRIs act as think tanks and data validators. The Ministry funds 29 of them, which makes this a Centre–State collaboration [1].
• Bottlenecks: the data collected is uneven across languages. Only 9 of the 13 covered languages had text translation as of 15 Oct 2026 [1].

6. Recent Developments (last 12–18 months)

• 1 Sept 2025: beta launch, with Santali, Mundari, Bhili and Gondi live [2].
• 2025: featured at the Ministry's Bharatiya Bhasha Utsav 2025 [4].
• 15 Oct 2026: first meeting of the Working Group on Tribal Languages; Adi Vaani proposed as the National Tribal Language Repository; 9 languages had text translation [1].
• 15 Nov 2026: Janjatiya Gaurav Diwas puts the topic back in focus [1].

7. Prelims Hooks

• Adi Vaani is run by the Ministry of Tribal Affairs, not MeitY [1].
• The beta was launched on 1 September 2025 [2].
• The development consortium is led by IIT Delhi [2].
• The consortium includes IIIT Nava Raipur and the TRIs of five states: Jharkhand, Odisha, MP, Chhattisgarh and Meghalaya [2].
• The first four languages were Santali, Mundari, Bhili and Gondi [2].
• Garo (Meghalaya) and Kui (Odisha) followed the first batch [2].
• AdiKosh is Adi Vaani's community-sourced word bank [1].
• The corpus target is about 1,00,000 sentence pairs and 100 hours of speech per language [1].
• As of 15 Oct 2026, State TRIs had supplied about 13.9 lakh sentences and about 300 hours of speech [1].
• 29 TRIs receive financial help from the Ministry of Tribal Affairs [1].
• Janjatiya Gaurav Diwas falls on 15 November, the birth anniversary of Birsa Munda, and was instituted in 2021 [1].
• The Working Group on Tribal Languages first met on 15 October 2026 [1].
• The UN's International Decade of Indigenous Languages runs 2022–2032 under A/RES/74/135, led by UNESCO [5].
• The Census 2011 ST population was about 10.45 crore (8.6%) [1].

8. Why '9 Languages Live' Hides a Speech Data Gap

• Text data is almost on target, but speech data is far behind
• The target is about 1 lakh sentence pairs and 100 hours of speech for each language [1].
• For 13 languages, that adds up to about 13 lakh sentences and about 1,300 hours of speech.

• By 15 Oct 2026, TRIs had supplied about 13.9 lakh sentences but only about 300 hours of speech [1]. That is less than a quarter of the speech target.

• Speech is what matters most for these languages

• These are mostly oral languages, passed on by speaking, with little written data [1].
• Speech-to-speech translation (spoken words in, spoken words out) is the feature an elder or a farmer who cannot read would actually use. It needs speech recognition, translation and speech synthesis (machine-made speech) to work together [1].

• With little speech data, that pipeline works worst for exactly the people it is meant to help.

• One total figure hides gaps between languages

• 13.9 lakh sentences across all languages looks close to the target. But only 9 of the 13 languages had text translation [1], so some languages must hold far more data than others.

• In 2026 Lok Sabha replies, the Ministry still listed only Gondi, Mundari, Bhili and Santali as supported in the app and portal, with Kui and Garo still under development [8].

• Nobody knows yet how many people use it

• In Parliament, the Ministry gave no user or download figures. It said outreach was still being assessed [8].
• Without usage data, the Ministry can count data collected but cannot show any benefit to tribal users.

9. Why Languages With a Script and Recognition Get Ahead

• Santali started with advantages the others lack
• Santali has its own script, Ol Chiki, created in 1925 by Pandit Raghunath Murmu for Santali sounds [9].
• It entered the Eighth Schedule (the Constitution's list of officially recognised languages) through the 92nd Constitutional Amendment in 2003 [9].

• In December 2025 the President released the Constitution in Santali, written in Ol Chiki [9]. A language with that much printed material is easy to feed into OCR (software that reads printed text) and text models.

• Languages with less written material fall behind

• A text-first model learns from written sentences. A language with few books or documents gives OCR little to read, so its data must be recorded from speakers from scratch [1].

• So the languages most at risk are the slowest to get tools.

• Recognition itself is still unsettled

• A government statement said demands to add 35 more languages to the Eighth Schedule were pending, and that there were no objective criteria for adding them [10].
• Bhashini (MeitY's AI translation platform) is aimed mainly at the scheduled languages (see section 9). Adi Vaani is filling the gap for tribal languages left outside that list.
• The risk: two separate government systems, one for 'recognised' languages and one for tribal languages, with separate data and separate apps.

10. Who Owns a Community's Words Once They Enter a Government Repository

• A repository changes who controls the data
• The proposed National Tribal Language Repository would hold corpora, primers, dictionaries and community stories [1].
• Native speakers give their data through TRIs, and experts check it [1]. But the sources say nothing on who must give consent, who can reuse the data, or whether a company can train its own AI on it.

• Some oral stories are sacred or meant only for the community. Once digitised in a central store, the community cannot easily pull them back.

• The world is moving towards community control

• UNESCO has issued guidelines on indigenous data sovereignty in AI (the right of indigenous peoples to control data about them). They deal with consent and cultural appropriation (outsiders using a culture's knowledge without permission) [12].
• UNESCO, Microsoft and The GovLab started an Indigenous Language Data Commons Incubator, which supports the communities themselves in governing how their language data is collected and used [13].

• The CARE principles for indigenous data sovereignty (Collective benefit, Authority to control, Responsibility, Ethics) appear in submissions to the UN's Global Digital Compact process [14].

• Why this matters in India

• A TRI is a government body. Consent given through a TRI is not the same as consent given by the community.
• Choosing one dialect as 'the' standard Gondi or Bhili for the repository can push out the other dialects. The note raises this risk, and no source shows a rule against it.

11. Does a Translation App Actually Save a Language?

• The strongest objection
• A language survives when children grow up speaking it at home and in school. An app that translates tribal speech into Hindi or English mainly helps the state and the market reach tribal people.

• Easier translation could even speed the shift towards Hindi and English. On this view, the money is better spent on teachers who speak the tribal language.

• What is right about it

• Translation and archiving keep a record of a language. They do not create new speakers.

• The Ministry reports data collected [1] but no figures on use [8]. So there is no evidence yet that the platform is helping people learn or keep using these languages.

• Why the platform still matters

• NEP 2020 asks for teaching in the home language at least till Grade 5 [1]. Teachers cannot do that without primers and dictionaries, and for most tribal languages these do not exist. Adi Vaani is building them [1].

• Tribal speakers also need public services and health information now, for example under the Sickle Cell Anaemia Elimination Mission [1]. A translation tool helps with that while schools catch up.

• Balanced view for Mains

• Adi Vaani is necessary but not enough. Judge it by whether its primers reach classrooms, such as Eklavya Model Residential Schools, not by how many sentences it collects.

12. What the Ministry and Its Partners Should Do Next

• Ministry of Tribal Affairs: publish data and use figures for each language
• One total (13.9 lakh sentences, 300 hours) hides which languages are behind [1].

• Report sentences, speech hours, accuracy and active users for every language, so Parliament can track progress.

• Ministry of Tribal Affairs: set a speech target and a deadline

• Speech data is under a quarter of its target [1]. TRIs should record elders and storytellers first, because oral knowledge is lost when they die.

• Ministry of Tribal Affairs and MeitY: link Adi Vaani with Bhashini

• Bhashini already runs Bhasha Daan, where citizens give text and speech data to build Indian-language AI models [11].

• Shared tools and a shared data standard would stop two ministries from building the same system twice. Tribal-language models could then work inside Bhashini-based government services.

• Working Group on Tribal Languages: write consent and ownership rules before the repository opens

• Use UNESCO's indigenous data sovereignty guidelines [12] and the community data-commons model [13] as templates.

• Let the community decide on access to sacred material. Make every dialect visible in the repository, not just one standard form.

• Ministry: give a clear order for the next languages

• Parliament was told that Katkari, Koya, Kokborok, Betta Kuruba, Thadou Kuki, Tangkhul and Chodri 'may be considered' in later phases [8]. That is not a commitment.
• A published timeline would let the TRIs of those states plan their data collection.

13. Anchors for Answers

• Data: About 300 hours of speech collected against a combined target of about 1,300 hours (100 hours x 13 languages); text is near target at about 13.9 lakh sentences [1]
• Data: ST population about 10.45 crore (8.6%), Census 2011, the group Adi Vaani is meant to reach [1]
• Report/Committee: Working Group on Tribal Languages, first meeting 15 Oct 2026 [1]; UNESCO guidelines on indigenous data sovereignty in AI [12]
• Law/Case: Eighth Schedule; 92nd Constitutional Amendment (2003) added Santali; 35 more languages are waiting to be added [9][10]
• Comparison: UNESCO–Microsoft–GovLab Indigenous Language Data Commons Incubator: the community, not the state, governs its language data [13]; CARE principles [14]
• Scheme: Bhashini's Bhasha Daan, citizen crowdsourcing of language data, a model for joint work [11]; National Sickle Cell Anaemia Elimination Mission, health outreach in tribal languages [1]

14. Mains Relevance

• GS-I: Indian society (tribal identity, linguistic diversity) and the preservation of cultural heritage.
• GS-II: welfare schemes for vulnerable sections (STs); government policies; education (NEP 2020).
• GS-III: science and technology (AI applications, indigenisation of technology); inclusive growth.
• Possible question stems: 1. Low-resource tribal languages face a dual threat of extinction and digital exclusion. Examine how AI-driven platforms like Adi Vaani can address both. (GS-II/III, 15 marks) 2. Discuss the role of Tribal Research Institutes in evidence-based tribal policy, with reference to recent language-technology initiatives. (GS-II, 10 marks) 3. Mother-tongue education under NEP 2020 cannot succeed for tribal children without learning resources in their languages. Critically analyse. (GS-II, 15 marks)

15. Related Topics to Study Next

• Bhashini / National Language Translation Mission: MeitY's AI translation platform for scheduled languages; compare it with Adi Vaani (verify the details on meity.gov.in).
• International Decade of Indigenous Languages (2022–32): the global framework for saving languages [5][6].
• NEP 2020's language provisions: the mother tongue as the medium of instruction [1].
• Tribal Research Institutes: their structure, funding and role [1].
• Janjatiya Gaurav Diwas and Birsa Munda: the historical anchor for tribal-pride initiatives [1].
• Eklavya Model Residential Schools: a likely user base for Adi Vaani's primers and dictionaries [1].
• National Sickle Cell Anaemia Elimination Mission: a health outreach use-case in tribal languages [1].
• Fifth/Sixth Schedules and Article 350A: the constitutional framework for tribal areas and primary education in the mother tongue (not sourced in this note; verify).

16. NCERT Concepts to Revise

• Tribal Research Institute (Polity): TRIs supply and validate the language data and are the Centre–State link in Adi Vaani.
• National Education Policy 2020 (Economy): Its mother-tongue instruction rule is the main education reason for building tribal-language primers and dictionaries.
• Eighth Schedule (Polity): Shows which languages get official recognition; Santali is in it, many tribal languages are not.
• Scheduled languages (Geography): Explains the 22 recognised languages, and why tribal languages outside the list need a separate platform.
• BHASHINI (Polity): MeitY's language-AI platform for scheduled languages, the comparison and possible partner for Adi Vaani.
• Artificial intelligence (Economy): Adi Vaani uses AI for translation and speech, so the basic idea helps explain how it works.
• Digital divide (Geography): Language barriers cut tribal speakers off from digital services; Adi Vaani aims to narrow this gap.
• Inclusive growth (Economy): The platform opens education, health and markets to marginalised tribal groups, a core inclusion argument for Mains.
• UNESCO (Polity): Lead UN agency for the Indigenous Languages Decade and source of indigenous data sovereignty guidelines.
• United Nations General Assembly (Polity): It declared the 2022–2032 International Decade of Indigenous Languages through a resolution.
• Birsa Munda movement (History): Birsa Munda's birth anniversary is Janjatiya Gaurav Diwas, the occasion that puts this topic in the news.
• Community (Polity): Community-based AI depends on native speakers contributing data, and raises questions of community ownership and consent.
• Data (Geography): Corpus size and data collected drive model quality; the data gap is central to the note.

17. Common Errors / Trap Areas

• Wrong ministry: Adi Vaani belongs to the Ministry of Tribal Affairs. Bhashini is MeitY's platform. Do not mix them up [1].
• 13 vs 9: 13 languages are covered, but only 9 had text translation as of 15 Oct 2026 [1].
• Kui's status: one PIB release (Bharatiya Bhasha Utsav 2025) lists Kui in the first phase [4]. The launch release says Kui was still nearing completion [2]. Rely on the launch release.
• Two "15th" dates: the Working Group met on 15 October 2026; Janjatiya Gaurav Diwas is 15 November [1].
• AdiKosh vs TRIs: AdiKosh is the word bank; TRIs are the institutions that contribute and validate the data [1].
• Launch year: some secondary sources give 2024. The official launch was 1 September 2025 [2].