As Wired reports , the International Association of Professional Translators and Interpreters (IAPTI) held a competition in South Korea in February, in which four people competed against AI-assisted translation programs by Google , Naver and Systran . They had 50 minutes to translate randomly selected literary and non-literary texts into Korean and vice versa. The humans clearly outperformed the machines in the end. Without knowing whose results they were looking at, two professional translators awarded an average 25 out of 30 points to the texts translated by people, while artificial intelligence texts were given 10 to 15 points.
This was due partly to the context, which the machines did not understand, meaning that they would not have passed the Winograd Schema Challenge , and due partly to lack of expressiveness: 90% of AI texts were quite awkwardly constructed and “grammatically awkward”. However, it is probably only a matter of time before that changes. More and more companies rely on NLP systems (Natural Language Processing), which are even capable of teaching themselves new languages.