China’s DeepZang AI aims to transform Tibetan language tech, promote ethnic unity

Creators of China’s first Tibetan-language AI model have pledged to speed up upgrades aimed at promoting ethnic integration.

Approaches to advancing the large language model, DeepZang, were discussed at a seminar on Monday that brought together representatives from research institutes, universities and enterprises, according to state-run portal Chinanews.com.

It said the event held in Hohhot, in China’s northern Inner Mongolia autonomous region, addressed issues such as improving Tibetan artificial intelligence (AI) technologies and expanding the model’s application into multilingual simultaneous interpretation.

Speakers included Tenzin Norbu, the model’s founder, who vowed to continuously refine Tibetan AI technologies and advance their practical use. He also emphasised the use of technological innovation to “foster ethnic interaction and integration” and support digital development in ethnic minority regions.

DeepZang’s launch in March came roughly three years after the unveiling of Monlam AI, a Tibetan-language LLM developed by Monlam Tibetan IT Research Centre, a non-profit organisation based in Dharamsala – the Indian hillside town that has hosted the 14th Dalai Lama and the Tibetan administration-in-exile since 1960.

Tibet remains a top priority in ethnic and religious affairs for Beijing, which has consistently bolstered domestic Tibetan studies to regain control of international discourse on the issue. DeepZang is part of this broader effort to reshape global narratives around Tibet with its own authoritative data and technology.

These efforts are also reflected in Beijing’s push since 2023 to use the romanised Pinyin name “Xizang” instead of “Tibet” in English articles, projecting its sovereign narrative globally.

Xi Jinping attends Tibet anniversary events

China’s policies in ethnic minority regions including Tibet have been a major focus of Western governments and rights groups, which repeatedly accuse China of human rights violations. Beijing steadfastly denies any wrongdoing.

In March, China passed its Law on Promoting Ethnic Unity and Progress, which is designed to counter Western ideological influence and lay the legal foundation for assimilating minority groups. The law took effect in July.

DeepZang has been trained on nearly 70 million Tibetan texts and over 30,500 hours of audio covering the language’s three main dialects. It powers AI chat, real-time translation and speech-to-text features across Tibetan, Chinese and English.

Tibetan AI technologies have long faced low-resource challenges due to unique linguistic structures, dialect variations and scarce digital corpora, making precision and versatility difficult to achieve.

Monday’s Chinanews.com report said that DeepZang had continuously broken bottlenecks, “bridging key technical voids in the field and bolstering China’s technology ecosystem for ethnic language AI”.

As cited in the report, Tenzin Norbu noted that AI development in other Chinese ethnic minority languages also faced low-resource constraints, making cross-regional and cross-linguistic collaboration essential.

The seminar also outlined future plans to integrate DeepZang into sectors including public services, education, healthcare and local governance, while “empowering the digital transmission of fine traditional Tibetan culture and ensuring that all ethnic groups equitably share the benefits of technological progress”.