Amii researchers use AI to preserve African languages and crack old ciphers
Two researchers at the Alberta Machine Intelligence Institute (Amii) are applying AI to opposite ends of language: Ife Adebara is building speech data for underrepresented African languages, while Greg Kondrak uses natural language processing to decipher coded historical texts.
- Ife Adebara co-leads African Next Voices, backed by the Gates Foundation, Meta and African institutions.
- The project collects speech data in Kenya, Nigeria and South Africa across dozens of languages, including isiZulu, Hausa, Yoruba, Kikuyu and Igbo.
- Greg Kondrak uses NLP to decipher coded manuscripts; in 2018 he and Bradley Hauer identified Hebrew as the most probable language of the Voynich Manuscript.
- His team found that the 1897 Dorabella Cipher aligns with musical notation, revealing a playable melody.
Bringing African languages into AI
According to Amii, most AI tools are trained on English, Chinese and European languages, which leaves millions of speakers of African languages severely underrepresented. The institute says this digital divide risks cutting communities off from AI advances in education, healthcare and infrastructure.
Ife Adebara is an Amii Fellow, a Canada CIFAR AI Chair and an assistant professor at the University of Alberta. She co-leads the African Next Voices project, an international collaboration backed by the Gates Foundation, Meta and African institutions. Amii, citing The Conversation, says the initiative is collecting the largest African language voice dataset for AI to date.
Adebara frames the problem this way: "When a language isn’t in the data, its speakers aren’t in the product, and AI cannot be safe, useful or fair for them." (Ife Adebara, quoted by Amii)
How the voice dataset is built
The project focuses on automatic speech recognition (ASR). The team collects ethically sourced speech data in Kenya, Nigeria and South Africa. The samples span dozens of languages, including isiZulu, Hausa, Yoruba, Kikuyu and Igbo, and cover topics from healthcare and agriculture to daily conversations.
According to Amii, training models on spoken dialects lays the foundation for localized voice assistants, medical tools and educational chatbots, so that people can use technology in the languages they live in. Adebara says that if the project succeeds, built-in AI in African languages "will be setting new standards for inclusive, responsible AI worldwide."
Cracking coded manuscripts with NLP
Greg Kondrak, an Amii Fellow and professor of computing science at the University of Alberta, works at the historical end of the spectrum. Amii says roughly 1% of surviving historical manuscripts are written in code, hiding material such as political secrets, love letters and ancient remedies. Deciphering them by hand can take a lifetime. In coverage by La Presse, Kondrak uses natural language processing to crack such ciphers in minutes.
In 2018, Kondrak and researcher Bradley Hauer used NLP to analyze the 15th-century Voynich Manuscript and identified Hebrew as its most probable underlying language. His team later applied algorithmic models to Edward Elgar’s 1897 Dorabella Cipher and found that it aligned with musical notation rather than text, revealing a hidden, playable melody.
What it means
Amii describes the AI models as intelligent "copilots" for historians, able to reveal patterns across millions of characters. Together, the two projects show AI being used both to widen access to current technology and to read records that were previously out of reach.
For speakers of African languages, the stated aim is access to AI-based services in their own languages. Amii, which says it supports a network of more than 500 active researchers, presents the work as part of its broader effort to advance AI science.
