Deepdive into DeepL

Ok, let‘s be honest: we all have used DeepL once or twice in our everyday (business) lives. Interpreters and translators do, too (we have to do our research, don’t we?).

DeepL is not new, and there were times when we were rather surprised at how natural and fluent the output seemed at first glance. Lately, however, we’ve noticed that AI translations are getting worse rather than better. There is an increased number of inconsistencies, it struggles with maintaining accurate grammar and sentence structures, and often the message is different compared to the original. This phenomenon is known as Degenerative AI. As human-generated data, that has been checked for correctness and consistency, is becoming scarce, AI is now training on AI-generated data. And when a translation tool such as DeepL is trained on a lot of its own output, there us a high risk of misinterpretation, data breach, quality loss, lack of consistency, hallucinations, bias and discrimination.

iSo don’t just rely entirely on AI without consulting translation and interpreting professionals to make sure you don’t lose important information or spread misinformation.