On September 20, at a Qinghai Normal University achievement showcase, China's first Tibetan multilingual omni-modal AI input method—Zhida AI Input Method—and a companion Tibetan font set were officially released, developed by the State Key Laboratory of Tibetan Intelligent Technology.
The input method strictly follows the national standard Tibetan keyboard layout and supports Tibetan Latin transliteration to lower the learning barrier. Voice input covers the written and spoken forms of the three major dialects—Ü-Tsang, Amdo and Kham—plus mixed Chinese, Tibetan and English speech. OCR input can photograph or screenshot mixed Han-Tibetan-English text for literature digitization, teaching materials and office scenarios.
AI optimization is the highlight: smart correction catches spelling, grammar and collocation errors optimized per language; smart polishing converts spoken to written Tibetan in formal, friendly or academic tones; smart translation supports bidirectional, context-aware translation. The lexicon covers 23 disciplines, and the team released four font styles—bold, regular, artistic and brush.
The underlying Zhida model ships in 9 versions from 0.8B to 100B parameters, with three main online models supporting 56-language translation and image/video understanding for education, research and government.
Source: China Daily / IT Home / Phoenix
