Examine the potential and limitations of AI-based tools such as Adi Vaani in preserving India's endangered tribal languages.

Census 2011 recorded 2,843 mother tongues, and 1,474 of them are still unclassified. The Ministry of Education's SPPEL scheme lists 117 endangered languages, each with fewer than 10,000 speakers [3]. Adi Vaani is the Ministry of Tribal Affairs' AI translator, built by an IIT Delhi-led consortium [2]. It offers a fast way to record these languages, but a tool alone cannot keep a language alive.

Potential

• Scale and features: It translates text in nine tribal languages and also offers speech, OCR and video subtitling [1].
• National repository: Users can search 300+ primers written by Tribal Research Institutes (TRIs). The plan is to grow it into a national tribal language repository [1].
• Welfare delivery: Using tribal languages improves last-mile outreach of government schemes [1].
• Education base: Its text collections and primers are raw material for mother-tongue-based multilingual education [1].
• Cooperative federalism: State TRIs contributed about 13.9 lakh sentences in 10 languages [1].

Limitations

• Data bias: AI needs large datasets, but only about 300 hours of speech have been collected so far [1]. The first languages to go live were large ones: Santali, Bhili, Mundari and Gondi [4]. The smallest and most endangered languages risk being left behind.
• Dialect flattening: IIT Delhi asked for standard spelling and writing rules [1]. This could raise one dialect above the others.
• Low uptake: The app has 48,000+ downloads but only about 6,500 monthly users [1].
• Archive, not revival: SPPEL has documented languages since 2013 [3], yet a recorded language can still lose its speakers. The Seven Pillars framework puts mother-tongue proficiency first and technology only fifth [1].
• Duplication: SPPEL's Sanchika portal [3] and Adi Vaani [1] are building similar archives side by side.
• Consent: Plans for community-led documentation of oral traditions [1] raise the question of who controls cultural material used to train AI.

Way forward: Adi Vaani is a necessary first step, but only a first step. CIIL has proposed phase-wise prioritisation [1]; languages should be ranked by risk, not by how much data they have. SPPEL's archive should be merged into the common repository, communities should give consent before their material is used, and the corpora should be put to use in classrooms and offices. This would help fulfil the promise of Article 29: that communities can protect their own language and culture.