China Releases First Tibetan Multimodal AI Input Method Across All Terminals
Qinghai Normal University's State Key Laboratory of Tibetan Intelligence released the Zhida AI Input Method, China's first Tibetan multi-language full-modal AI input tool for mobile and desktop devices. Powered by the underlying Zhida LLM, it supports text input, dialect voice recognition, and cross-language optical character recognition (OCR). This release represents a significant advancement in minority language processing by bringing multimodal AI capabilities to Tibetan speakers, educators, and researchers. It helps bridge digital accessibility gaps for low-resource languages by enabling seamless cross-language communication and document digitization across Chinese, Tibetan, and English. The input method supports Tibetan Latin transliteration alongside national standard keyboard layouts, voice recognition across three major Tibetan dialects (Ü-Tsang, Amdo, and Kham), and OCR for mixed-language documents. The underlying Zhida LLM encompasses 9 version configurations ranging from 800 million to 100 billion parameters, supporting translation across 56 languages.
## BACKGROUND
Tibetan natural language processing faces unique technical challenges due to complex script geometry, distinct regional spoken dialects, and historical reliance on Latin transliteration systems like Wylie for computer input. Multimodal AI models allow software to process text, speech, and images simultaneously within a unified framework, helping overcome digital resource scarcity in low-resource minority languages.