What’s grandpa saying? First Hokkien speech translator could boost communication between generations

The company hopes to bridge communication gaps between generations. PHOTO: PIXABAY

SINGAPORE - Hokkien-speaking grandparents may soon be able to converse with their English-speaking grandchildren using a speech translation system developed by artificial intelligence (AI) laboratory Meta AI. 

The AI-powered system, the first of its kind to translate between Hokkien and English, is currently in the research phase. Meta AI belongs to Meta, which runs Facebook.

Earlier this month, the lab released a video demonstration by Meta founder and chief executive Mark Zuckerberg and Mr Chen Peng-jen, who led the development of the software.

In the video, Mr Chen, Meta AI’s software engineer, spoke in Hokkien, and Mr Zuckerberg in English. The system translated their conversation seamlessly.

Mr Chen, 37, grew up in Taiwan speaking Mandarin. His father Chen Sheng-Jiang, a 70-year-old retired factory lead technician, spoke Taiwanese Hokkien. 

“I have always wished my father could communicate with everyone in Taiwanese Hokkien, which is the language he’s most comfortable speaking,” said Mr Chen, who is now based in New York.  “He understands Mandarin well, but speaks more slowly when communicating about complex topics.” 

Hokkien, which has a history of more than 1,700 years, is spoken by around 50 million people in parts of China, Taiwan, and among the Chinese diaspora in South-east Asian countries like Singapore, Indonesia, Malaysia and the Philippines.

The Hokkien dialect group, at around 45 per cent of the Chinese community in Singapore, is the largest Chinese dialect group here. The dialect used to be the lingua franca of the Chinese communities in Singapore before the Government pushed for the widespread use of Mandarin to replace dialects with its Speak Mandarin Campaign from 1979.  

While using computers to translate languages is not new, previous efforts have focused on written languages. However, of the more than 7,000 languages currently in use, over 40 per cent are primarily oral, like Hokkien. They do not have a standard or widely known writing system.

There is little speech data on Hokkien, and the dialect has not been paired with other languages like Spanish or English. There are also few human English-to-Hokkien translators, making it difficult to collect and annotate data to train the AI system.

To get around these problems, Meta researchers used text written in Chinese, which is similar to Hokkien. The team also worked closely with Hokkien speakers to ensure that the translations were correct.

Meta researchers believe AI could help solve many communication challenges for speakers of oral languages. It could, for instance, make it easier for people to navigate the Internet and communicate in different languages virtually or in real life. 

Hokkien is just the first step to translating other oral languages for Meta AI. The researchers will make their model, code and training data freely available to allow others in the community or industry to build on their work.

The effort to translate Hokkien comes under what Meta calls a Universal Speech Translator. The company hopes to translate many more languages in real-time to help millions of people break down language barriers wherever they are located – even in the metaverse. It also hopes to bridge communication gaps between generations and prevent some languages from dying out.

“The ability to communicate with anyone in any language – that’s a superpower people have dreamed of forever, and AI is going to deliver that within our lifetimes,” said Mr Zuckerberg in the online presentation.  

Join ST's WhatsApp Channel and get the latest news and must-reads.