Mansi-Language AI Learns to Speak Short Phrases
Researchers at the Yugra Research Institute of Information Technologies have taught a neural network to reliably generate spoken short phrases in the Mansi language. Professor Andrey Melnikov announced the achievement at the International IT Forum with the participation of BRICS and SCO countries in Khanty-Mansiysk.
Previously, the model struggled with short utterances and often made errors when generating speech. According to the researchers, that issue has now been resolved.
At the same time, researchers digitized nearly all written Mansi-language materials held by the Yugra State Library – about 400 books - and created a unified electronic database. This will help preserve the language and integrate it into modern AI applications.
More than 30,000 members of Russia's Indigenous peoples of the North - the Khanty, Mansi, and Nenets - live in Yugra, accounting for about 2% of the region's population. Approximately 4,500 of them maintain a traditional way of life.
