By Carlos de Paula, who has translated over 100,000 documents since 1982.
If you want to see for yourself how terrible AI translation can be, do this little experiment. You do not even need to speak the target translation language:
Run a moderately complex, 500-word English text in a AI translation engine. Choose as target a romance language, such as Spanish, Portuguese, French and Italian.
Run the resulting translation through the engine, translating it back to English.
Run the resulting back-translation through the engine, translate it again to the chosen target language.
Do this back and forth a few times, always using the same text, same languages. By the fifth, or sixth run, the original text may look very different, depending on the ambiguity, punctuation and quality of the original English text.
If you do the same experiment using a very complex text (say, a Court judgment), the results will be even more astounding.
If you use different AI engines (Google, MS, DeepL, ChatGPT) the results will be even scarier. If you translate into different languages, the text may not be recognizable after the fifth, sixth run.
You see, every time the engine translates it may be interpreting things differently, misunderstanding the ambiguities, overlooking style, making wrong assumptions and hallucinating. If AI translation were the mathematical operation purveyors claim it is, there would be little change, translating back-and-forth would create smooth results.That is the way it used to be 15 years ago, it remains so.
(877) 626-0642
WhatsApp (786) 543-1687
https://legaltranslationsystems.com