Neuronale maschinelle und KI-generierte Übersetzungen: Versuch einer Kategorisierung von Stärken und Schwächen in der Sprachkombination Spanisch-Deutsch

Abstract Neural machine translation has fundamentally changed the practice of translation for more than a decade. The translation industry is now undergoing another fundamental transformation due to advances in generative artificial intelligence (GenAI) in the form of large language models (LLMs), such as ChatGPT. For professional translators, however, automatically generated translations are primarily a productivity tool. This process is known as MTPE (machine translation post-editing) and involves combining machine translations with human expertise. This article identifies and categorizes the strengths and weaknesses of automatically generated translations using a series of examples. To this end, the neural machine translation systems DeepL and Google Translate were compared with ChatGPT’s GPT-4o large language model, both with the source text and with each other. The results show that the examined automatic translations offer numerous advantages and are changing the translation profession extensively. However, they are not yet capable of ensuring an accurate translation, meaning that checking or post-editing by a human translator remains necessary.