Taiwan: A History Lesson - from TWSC to Taiwan AI Service Corporation

Language has been the foundation of identity, history, and resistance for centuries. From the whispered codes of ancient tribes to the defiant survival of dialects erased by colonization, languages hold more than words – they have power. Today, as artificial intelligence shapes the digital world, a quiet battle is unfolding: Who controls language, and who gets to preserve it?

Taiwan, a tech powerhouse known for its semiconductor exports, is proving that old languages have a future. While global AI giants like Google prioritize widely spoken tongues – those they can commercialize and monitor – Taiwan has chosen a different path.

A leading Taiwan giant is listed company ASUS, which partnered with NVIDIA, Intel, and AMD. Together with its subsidiary TWSC, now renamed Taiwan AI Service Corporation, they deliver comprehensive AI infrastructure solutions and lead hardware computer manufacturing globally.

It may be a small nation, but Taiwan has recognized the importance of preserving, revitalizing, and equalizing the languages of its ethnic minorities. Using VOX technology, programmes have been developed based on the text-to-speech (TTS) System designed for Taiwanese Hakka. It supports six distinct dialects and addresses the scarcity of Hakka language resources. Why?

The Hakka people are an intriguing ethnic subgroup from Northern China with a distinct culture, cuisine, and history. From the early times of the Jin Dynasty (265CE), many travelled and migrated to other parts of the world. Today, over 80 million worldwide, the Hakka have a fascinating story of migration, hardship, and preservation of ancient traditions. Today, a few million in the diaspora still speak Hakka, but in Taiwan, the government has ensured that older adults feel understood and cared for in public health services.

Nothing exemplifies this more than the loss of languages around the world. Many regional languages are slowly dying out as new generations of native speakers have no choice but to learn languages that will help them in their future careers.

The slow erasure of indigenous languages is not an accident. Around the world, AI models are trained primarily on dominant languages – English, Mandarin, and Spanish—while thousands of others are left without digital representation. Google’s AI models, for instance, prioritize languages it can monetize, meaning many communities are left without AI resources in their own tongue. In Nigeria, for example, where over 400 languages are spoken, only five are recognized as official. The pattern repeats across the world.

But Taiwan has recognized what others have ignored: AI is an opportunity to fight back.

The Hakka language of Taiwan, though officially recognized as one of the country’s national languages and the third most commonly spoken language on the island, is spoken by only a handful of people.

“There is a danger of Hakka dying out in one or two generations,” says Dr. Yuan-Fu Liao, Director and Professor at the Institute of Artificial Intelligence (AI) Innovation, Industry, Academia Innovation School of the National Yang Ming Chiao Tung University (NYCU). “Only older people speak it anymore, and almost none of the younger generations can.”

Many Indigenous languages face the same problem: economic realities push younger generations toward dominant languages that promise global job opportunities. However, Taiwan’s response is a blueprint for what AI can do beyond profit-driven language models. It reminds us that AI can serve culture, not just corporations.

Wu and the Taiwan AI Service Corporation team contacted Dr. Liao at NYCU to do this. “Many of these people are elderly and live in remote or faraway places,” recounts Dr. Liao. “Since they are all elderly people, we couldn’t expect them to come to us – we had to go to them.”

The Taiwan AI Service Corporation-NYCU team met with 300 people and recorded many hours of Hakka. TWSC then took the recordings and spent around four years building the LLM that is capable of not only translating inputs into Hakka but also correcting mispronunciations (among many other things).

“ChatGPT has made LLMs a big deal,” says Wu. “This is why the Taiwan government decided to invest in this program. Roughly, it costs around USD 3-4 million to build the dataset app, while the hardware infrastructure costs around USD 160 million.”

Wu is immensely proud to mention that most of the core work is complete, and the government is ready to open the API to the public for further development. In addition, two pilot applications are in the testing phase right now – one for use in the healthcare sector so older Hakka speakers can benefit from improved quality of service, and the other is a language teaching phone application that will target younger people.

Both the government and I are happy with how things have progressed. The truth is that there is a dearth of resources for keeping languages like Hakka relevant today. Larger entities like Google and Microsoft can’t or are not interested in investing time and effort into their products for a niche language like this, and even here at home, most similar efforts are directed towards Taiwanese since that is the most commonly spoken language.”

Lessons from Hakka Leadership

The Hakka people have always been influential despite being historically marginalized. Deng Xiaoping, the leader behind China’s economic rise, was of Hakka descent. Lee Kuan Yew also oversaw Singapore’s transformation into a global financial hub. Their contributions underscore a broader truth: old languages are not relics of the past; they are the foundation of innovation, resilience, and leadership.

By investing in AI-powered language preservation, Taiwan proves that technology does not have to erase history – it can revive it. The government partnered with Taiwanese tech giant ASUS and its subsidiary, Taiwan AI Service Corporation, a company now positioning itself as a serious AI contender. They have indeed demonstrated the AI for purpose model, investing years gathering voice data from elderly speakers, building an AI model to record the dialect of Hakka, and correcting possible mispronunciations. The project cost roughly $3-4 million to develop the dataset app, and $160 million for the infrastructure – an investment in history and the future.

Now, the application is ready for public use. A healthcare app is being tested to improve services for elderly Hakka speakers, while a mobile learning app aims to engage younger generations working with those living in rural areas. Buoyed by its success, Taiwan AI Service Corporation is now preparing for an IPO, signaling that AI for good exists—it can serve communities.

The fight to preserve old and Indigenous languages is less about cultural nostalgia and more about autonomy in an AI-driven world. If the power to decide which languages thrive and which disappear remains concentrated in a few tech giants, then the digital future will be one of exclusion, not inclusion.

Taiwan’s approach is an urgent lesson for governments, communities, and innovators: The democratization of AI for Indigenous languages has a right to exist, not as museum artifacts, but as living, evolving mediums of expression. The technology exists to make this happen.