How regional media group TNL harnesses generative AI for its translation engine

The News Lens Group (TNL) is a regional digital media group founded in 2013, covering Taiwan, Hong Kong and Japan. At FIPP Congress, TNL’s chief integration officer Richard Lee went into the brand’s data-driven culture and explained how its ambitious plans to diversify across languages, revenue streams and regional reach are coming to fruition – with the help of generative AI.

“Internationalisation by default”

With brands including The News Lens (in-depth feature stories and infographics), Inside (covering internet, software startups, AI) and Business Yee, TNL has a broad reach. Its other brands include Cool3C (product information and gadget reviews), Sports Vision (expert analysis, athlete stories) and iCook (a leading recipe platform with more than 3m members).

But the growth is still ongoing. Lee began by outlining TNL’s business strategy as one of intentional reach across the region: “The idea is to grow from a single media brand into a media group, via a process of ‘internationalisation by default’,” he explained. In part, this means that news articles are shared and translated across different languages and regions.

On this basis, top editors cover each region/vertical: “We empower them to work on delivering the best ‘local news’,” explained Lee. “The idea is to try to cover many different kinds of culture and topics across the Asia region.”

TNL covers topics that are important to people in particular regions. “For example, in Taiwan, we just had elections. Political news, about democracy, this is popular here. In Taiwan we are also the first Chinese-speaking country to pass the same-sex marriage law.”

They also harness the power of zero-, first- and third-party data, using comprehensive user profiles and analysing reader behaviours.

Surpassing “peak local”: The roadmap to a regional media group

Around 2018, TNL began to sense that they had reached the peak of the local news market. “We found TNL had significantly fewer readers in Taiwan than the top local traditional outlets, probably because they have TV and radio as well as digital news. It was therefore a challenge as a purely digital news media outlet to compete with them.”

This led TNL to try to expand its readership to other places, launching TNL Hong Kong and International editions in 2014 and 2015 respectively, followed by TNL Japan in 2022. In 2020, they also made a series of acquisitions and expanded into data analytics and adtech.

“But despite this, of course it is not all that easy,” said Lee. “Content localisation is difficult, even when you know the Taiwanese market better than most, as we do. But it takes time to create quality local content – our editors have to figure out the best topics to cover. Translations entail delays, sometimes of a couple of days. And that means that the news is not fresh enough. Every time there’s a new angle or vertical, this adds to editorial costs.”

Publishing in different languages: Generative AI to the rescue

Playing around with ChatGPT when it launched early this year, however, the team immediately found it to be a help to both editors and translators. “Adding new languages to new sites has become vastly more efficient with new LLM mode-based AI workflows,” explained Lee.

Generative AI-assisted workflows are simply faster than traditional methods, explained Lee. AI can translate a breaking news story about semiconductors, for instance, in less than three minutes. It then takes editors less than an hour to go over the AI translation and fix any minor issues.

Notably, AI models work best in the English language at the present moment. To get around this, TNL translates local-language content into English, then uses this version as a basis for translating into other languages. TNL’s translation engine is currently in use on its media sites in English, Japanese and Chinese.

Since adopting this AI-based translation workflow, the benefits for TNL have been significant. “Around 15 per cent of content is AI-assisted, with page views and ad revenue from AI content around 10 per cent,” said Lee.

“The most significant thing is that it is much more cost-efficient, and time-efficient,” said Lee. The costs of creating AI-assisted content represent just 20 per cent of the original methods.

“It gives us the chance to do things we just wouldn’t do before. For example previously, we would only have recipe content in Chinese. Now we have it in at least three languages.”

Lee hopes that others can learn from TNL’s experiences. “I hope this gives you confidence about how enabling AI in your workflow can help you to generate new page views and new ad revenue – and also enable your editorial team to try new languages and verticals.”


Your first step to joining FIPP's global community of media leaders

Sign up to FIPP World x