How Deutsche Welle is reducing the workload of content adaptation with plain X

Global broadcasters like Deutsche Welle and the BBC face a dilemma - they want to offer their content in as many languages as possible, but manual content adaptation eats up time and budget. At the same time, they can't relinquish quality control. The solution doesn't lie in fully automated AI, but in intelligent support with "humans in the loop." Deutsche Welle, together with Priberam, a Portuguese specialist in Natural Language Processing (NLP), has developed a platform for exactly this purpose: plain X.

I spoke with Mirko Lorenz about what the platform delivers and which users benefit most from it. He's Innovation Manager in Deutsche Welle's Research and Cooperation Team (ReCo) in Bonn - a kind of satellite department of DW that isn't financed from the federal budget like the main organization, but primarily through funding from EU and German research projects.

Mirko Lorenz is also the founder of Datawrapper. The data journalism tool, launched in 2012, is now used by around 600 major newsrooms worldwide, has one million registered users, and produces 10,000 charts daily. Full disclosure: Mirko Lorenz and I have known each other since our early journalism days. More than 30 years ago, we were both freelance contributors to the nationally distributed university magazine Unicum.

But now, on to plain X:

plain X combines four functions in a collaborative work environment:

What makes it special: The platform was developed from research for real life business solutions. "This is a rare example of a real product coming out of scientific research," explains Lorenz. Deutsche Welle's Research and Cooperation Team has been working in "Human Language Technologies" for 10 years - experience that fed into the development of plain X.

Deutsche Welle holds a stake in the project and has been using plain X itself as a standard tool for nearly two years, with full integration into the production environment and connection to cloud services. plain X is operated and developed by the company Priberam, based in Lisbon. The two partners met through joint research projects. The Portuguese team brings decades of language expertise - the company has been running the largest online dictionary for Portuguese since 1996.

The crucial difference from many AI tools: plain X is designed for journalistic quality assurance. While content creators are being forgiven for using quick automated YouTube video subtitling, that would be unacceptable for DW. "We can't suddenly accept a bunch of typos. That would mean giving away our competence and credibility," says Lorenz.

The platform allows you to comment on transcripts, check the spelling of names, or invite native speakers for review and to create glossaries. These glossaries ensure that names and technical terms are written consistently. Former British Prime Minister Rishi Sunak’s name, for example, is written with a Z by most transcription systems. "With glossaries, we can prevent spending a week changing Zunak to Sunak during the transition to Keir Starmer," says Lorenz.

plain X is "engine-agnostic"—the platform currently connects to 15 transcription services and eight translation engines, including Azure, Google, Whisper, DeepL, and ElevenLabs Scribe. The advantage: You can choose the best available engine for each task.

The quality of language engines varies greatly by language and especially language pairs. DeepL is used for about 70 percent of all cases, but for "low-resource languages"—languages with less training material—specialized engines can be better. ElevenLabs Scribe, for example, works well for Kiswahili, important for the German public broadcaster Westdeutscher Rundfunk (WDR) that has a studio in Nairobi.

The approach deliberately supports small, specialized providers like OpenTrad (Galician, Catalan) or Lesan AI (Amharic). "We expect that through linguistic focus, a small engine can become better in the long run than a large generic one. At Google, Galician just runs along, they're not really interested in it. OpenTrad, on the other hand, does nothing but Galician," explains Lorenz.

plain X also enables the use of synthetic voices. But this function is currently only used for very specific use cases. "The currently available voices are already very good. But our sense of hearing is even more finely developed than our sense of sight. An artificial voice that sounds really good in isolation still doesn't work for a longer documentary," says Lorenz. The current generation of speech technology works well for short news videos. But longer documentaries still lack crucial subtleties—nuances, speech pauses, changing emotions that we immediately pick up on in real speech.

In addition to Deutsche Welle, users include public broadcaster WDR, Lusa (Portugal's news agency), CNN Portugal, and ChatEurope (AFP, dpa, El País), all using the platform in regular operations.

Lusa deploys plain X with 80 editors daily - but not for video adaptation, rather for transcripts. "Editors need the original quotes from politicians that ran in the morning shows, and transfer them into a text stream," explains Lorenz. 

The Doha Institute in Qatar uses plain X pro bono. Journalism students work with it for transcription, translation, and subtitling. "This helps us better understand the quality of transcription and translation when dealing with Modern Standard Arabic and regional dialects," says Lorenz.

plain X promises up to 85 percent time savings for transcription, up to 80 percent for translation, and up to 95 percent for subtitling and voice-over. These efficiency gains are based on templating options and the application of subtitling rules.

Critical for professional use: Integration into existing workflows. At Deutsche Welle, plain X is connected to internal systems. This allows texts, audio, and especially videos to be processed in plain X, with results sent to Adobe Premiere for final editing, for example.

Smaller newsrooms can also benefit. The pricing model is volume-based with no monthly fees, including a seven-day free trial.

Lorenz sees particular potential for networks: "Imagine if we could enable networks of investigative journalists to easily translate their content into 30 languages with this tool—that would be no problem. They might not need it frequently, but when it matters, they could offer a quickly scaled 30-language translation done overnight."

To enable such usage models, plain X uses volume-based billing—the number of editors isn't a factor in pricing. “Otherwise, we'd make temporary teamwork almost impossible, but precisely this form of collaboration is important," adds Mirko Lorenz.

Public broadcasters could also better reach their communities. Lorenz cites a German example: "Why don't we have a Ukrainian program where our content is translated into Ukrainian? The refugees would be grateful, and a ten-minute daily update would be enough."

For Lorenz, it's about more than efficiency: "The internet offers an enormous amount of information, which is precisely why context matters. What's true is currently bitterly disputed. Climate change denial is just one example. Conveying knowledge here, constantly providing facts - that’s important. And it's easier when you speak the respective language."

For years, Deutsche Welle has pursued a strategy of reaching users through each country's popular platforms—Facebook, Instagram, YouTube, or TikTok. As a result, DW now reaches 337 million users per week and reports as an international broadcaster in 32 languages worldwide. "When you meet people from South America, Africa, or Asia, you often encounter Deutsche Welle viewers. People recognize that DW strives to offer balanced reporting. This is in demand in times with sometimes very extreme political styles and viewpoints," emphasizes Lorenz.

plain X is a tool to advance this strategy—not through blind automation, but through intelligent support of journalistic work with clear quality control.

For individual journalists:

For small newsrooms with minority languages:

For mid-sized newsrooms:

For global news organizations:

For all newsrooms: