Nepal's indigenous languages get digital boost with AI translation system

Communities and academics in Nepal are leveraging artificial intelligence to digitize and preserve the nation's indigenous languages, a critical effort to pass them on to younger generations. A significant advancement comes from Kathmandu University's Information and Language Processing Research Lab. They recently launched a trilingual translation system connecting Nepali, Tamang, and English. This system, powered by a Google Academic Research Award, was trained using over 100,000 parallel sentences and employed AI, specifically Meta AI's NLLB-200 model, on Nepali and Tamang datasets. Professor Bal Krishna Bal, the lead researcher, highlighted the challenge of English's dominance, which encroaches on Nepali usage. He believes the new system will enable Tamang speakers to access global knowledge and share their culture more widely. "English has become so dominant that we find it increasingly difficult to write and speak even in Nepali. The language faces ever more encroachment," Bal stated. He also noted a renewed emphasis on mother tongue preservation among Nepalis living abroad, who wish for their children to learn cultural traditions. Amrit Yonjan, coordinator of the Tamang translation team, expressed renewed hope for indigenous language preservation due to this technology. The team is working to integrate Tamang into Google Translate by mid-March 2027, a process that involves translating a designated English corpus into Tamang. Currently, only Nepali, Maithili, and Nepal Bhasha (Newari) from Nepal are available on Google Translate. The success with Tamang has spurred interest from other communities, including Magar, Sherpa, and Tharu, who are exploring similar AI projects. The Magar community has committed to providing over 25,000 sentences and words to aid AI model training.

“

English has become so dominant that we find it increasingly difficult to write and speak even in Nepali. The language faces ever more encroachment.