Machine Translation Errors: Their Main Types and Solutions to Address Them

As machine translation (MT) develops at a crescendo pace, many believe human translation will soon sink into oblivion.

It can possibly be assumed to be true if you are a tourist trying to communicate in a foreign hotel. However, should your endeavour be slightly more complex, you will surely meet what is called machine translation errors. Here are some of them.

MT cannot reproduce all linguistic subtleties and capture cultural sensitivities and the style and tonality of the message. It does not translate terms consistently: if you have entered into an agreement and translate it with MT, be ready to find the local equivalents of “agreement”, “contract” or “treaty” in your translation.

Well, have you probably heard anything about hallucinations? AI systems hallucinate too – by inserting words not found in the original into their translation. And then, there is a so-called “additional toxicity”, when MP uses toxic (vulgar, abusive or violent) language, which the original lacks.

To top it all off, MT systems have learned with masterly skill to “pack” all these errors into perfectly structured sentences where they might be difficult to find.

However, there is always a way out of the foxhole. Any text generated by AI should be proofread by a professional translator.