Case Study: Translating Better with AI

For a long time, specialist translations have been produced according to a fixed procedure: translators read through a text and then translate it into another language before passing the draft translation to proofreaders. The latter then review the text, correcting the spelling and grammar as necessary. For an average-sized text of 20,000 words, this process takes an average of 15 days. “Translations produced by traditional translation agencies are highly inefficient,” says Christopher Kränzler. “Not only do they take a long time, but they also tend to be expensive.”

Can machine learning help?

Kränzler found the traditional method too slow and too expensive. He saw potential for improvement here – and founded the technology company “lengoo” as a result. “lengoo” uses a subspecies of artificial intelligence, namely machine learning, to translate a text into another language within a few milliseconds. “We train our programme by feeding it sample texts – on the exact same subject and from the same client – of possible translations in a certain language pair. The more samples, the better,” he explains. From these sample texts, which consist of at least 10,000 sentences, the AI then recognises what type of text is involved in the new translation, as well as its subject area and its prevailing tone of voice.

Neural networks are supposed to recognise everything

This type of artificial intelligence is a self-learning system. It works with artificial neural networks. These networks translate words into vectors – that is, numbers – and detect the likelihood of certain words occurring next to one another. In the translation process, the system is therefore able to reflect the clients’ tone of voice very accurately and to use the latter’s usual vocabulary. All texts translated by the system are then proofread by human translators, because, even at “lengoo”, human beings are as irreplaceable as ever. Nevertheless, entrepreneur Kränzler is convinced that AI will be taking over even more tasks in the future: “A century from now, the connection between human and machine will be even more intensive.”

High hopes for the future

In fact, his software for specialist translations is just one of many examples of how AI could be used, particularly in the field of language: many languages are threatened with extinction because there are simply no longer enough people able to speak them fluently. Coming to the rescue, however, are chatbots that, in the meantime, have been developed and programmed precisely for these languages. They can converse with people and even recognise and correct mistakes in the spoken language. In this way, these languages can be preserved.

Researchers from Columbia University in New York have even found a way to use AI to deduce language from the auditory centre in the brain, while their colleagues from San Francisco are then able to form entire sentences from it. In this way, it’s possible to translate people’s silent thoughts into audible speech. The method has not yet been perfected, but in the future it could help patients who are no longer able to speak for themselves.

AI offers an enormous range of opportunities in this respect – but it also faces challenges: does AI have the potential to make jobs obsolete? Could AI alter our languages too much? One example in this regard is what happened at Facebook: in 2017, the AI used at the company invented its own secret language that was no longer comprehensible to humans – even to the developers themselves.

But, despite these concerns, it’s highly likely that AI will continue to be used in the future – and probably even more extensively than today.