Translation, Localization, And Conversational Design In a Multilingual World

Many forward-looking companies are embracing chatbots and working with skilled conversational designers to engage new markets.

But global enterprises—even smaller businesses—risk damaging their reputations by not making local translation (called localization) a priority.

Conversations host Hans van Dam explores this fascinating challenge with Thorben Stemann, a Conversational Linguist with Kocarek, in Germany. Kocarek is an experienced translation service provider for small and medium-sized businesses. The focus of its translation activities is on technology, legal, economics, marketing and communication.

Translation isn’t a new issue for global businesses. They have been considering tone of voice, brand presentation, and cultural nuances for decades now. 

However, in voice tech Thorben points out that at the core of all issues with language technology is the fact that everything is demoed and prototyped in English. 

He explains: “Most of it works great on English. Just adding one more language . . . usually you realize: ‘Oh, that sounds rough’. You always have to add these slashes to add both genders and it doesn't look so great.”

So because different languages have different grammatical structures, designers may need to subtract or add information. 

“For example, in the Slavic languages, when you speak in the past tense the verbs all take a gender. That means you need to know the gender of the person talking. And ideally, you would need to know the gender of the person you’re talking to,” says Thorben.

“As well, will the bot be using formal or informal speech, and will it universally use the same forms to address someone? Those are just technically sometimes also challenging, not just from a design standpoint. But how do I actually make sure that's working in the infrastructure of the bot?”

Thorben says the number one mistake businesses are making in voice tech is placing localization at the end of the conversational design step. A translation, he explains, is a rewrite in another language. The whole process is being done again, so ideally, it should be taken into account right from day one. 

Inclusivity is another important issue in voice tech. Hans notes that copywriters need to understand nuances within a culture, so that everyone understands what is being said. 

Kocarek is working to solve these and many more challenges. Thorben’s role is to be a bridge between designers and linguistics, working with engineers and copywriters to ensure that solutions to linguistic problems are found, and that the finished dialogue sounds natural and makes sense when in use. 

“We want the linguists to be able to adapt to the technological possibilities, and maybe refrain from doing things that are just outright impossible in the technological tool that you're using.” At the same time, Thorben works to sensitize engineers to language so that they recognize when something sounds weird or doesn’t evoke emotion.

Subscribe on multiple platforms! 

Show links:

Thorben Stemann's Socials: 


  • 04:55: The first language of Polish, and now we've localized into German, German for Swiss, French, regular and Swiss, Russian, Czech, Dutch, Flemish.
  • 08:39: If you're dealing with multilingual, conversational AI, you have an exponential growth of challenges.
  • 11:08: If you have a brand with global core values, they manifest themselves differently in different cultures.
  • 18:10: The prototype example for this is the whole data sets in English are trained on the so-called ‘White man English’. And, you know, to mix that up. You need to know what are actually the minorities in the country that you’re localizing to.
  • 19:46: The implications of not training for these things and not understanding what people say . . .it's a conversation that we didn't intend. You actually insulted or hurt someone's identity by not actually considering them when you were training the model.
  • 28:46: Most companies are multilingual enough already to facilitate (inclusivity). Usually, if you're a big conglomerate, you might have a corporate language. Then your different offices obviously operate in their local language every day.