Taiwan Ministry of Digital Affairs launches Sovereign AI corpus collection

Taiwan's Ministry of Digital Affairs has launched a campaign to collect private linguistic data for the ‘Taiwan Sovereign AI Training Corpus.’ The initiative aims to ensure AI models better understand Taiwan's specific language, culture, history, and social context, addressing the current dominance of Simplified Chinese in international large language models.

Minister Lin Yi-jing initiated the program by donating her own literary works, including ‘The Happy Ghost Island’ and ‘Bandits and Innovators.’ Several prominent authors, publishers, and e-book platforms—including Readmoo, Ink Publishing, and ShowSpeed—have signed authorization agreements to participate. The collection focuses on authorized publications, book introductions, trial readings, and public domain works.

The Ministry emphasized that the project operates on three core principles: voluntary participation, explicit authorization, and the ability to opt out. This mechanism is designed to protect the rights and interests of copyright holders while building a trustworthy AI with local characteristics. As of August, the corpus has reached approximately 2.2 billion tokens, primarily utilizing government data.