My ranking:
1. ChatGPT4 - flawless translation. I was blown away
2. DeepL - very close, but one mistake
3. Google Translate - good translation, some mistakes
4. Microsoft Translate - bad translation, many mistakes
I can understand the panic.
I guess we have to get used to software redefining the meaning of words. It was kind of funny when that happened regarding Google Maps / neighborhood names, but with LLMs it's a different ballgame.
For anyone who doesn't speak German, pathetisch means with pathos, impassioned.
A native English speaker probably would only use "pathetic" to mean "emotional" if the emotions were specifically negative. They also would use pathetic to describe someone experiencing non-emotional suffering such as injury or poverty.
Therefore, a native English speaker probably would not use "pathetic" to mean "emotional" in everyday writing. However, I could definitely see someone using it to mean emotional when they were being more poetic. For example, I could see someone calling an essay on the emotional toll of counseling "The Pathetic Class" in order to imply that social workers are a class that society has tasked with confronting negative emotions.
It's the same in Romanian, and I guess many other languages.
Many of the common words of European languages are derived from Greek and Latin, and where the meaning has diverged in English, now (because of its ubiquity) these false friends are being realigned to mean what they do in English.
And as with anything else, with the time it will get improved, too. LLM is not the answer to all linguistic problems.
But the interesting thing IMHO is the nature of the mistakes ChatGPT makes... often they're quite elementary mistakes (e.g., the occasional subject-adjective word order) while it gets the big picture right. Whereas DeepL is sometimes the reverse. ChatGPT also has the advantage of being able to tailor its output to a particular context, e.g., it can tailor legal translations to terms used in Canadian law rather than French law. However, for longer texts, I've noticed that ChatGPT will sometimes omit small parts of the source text from the translation, which is unfortunate.
I have a colleague who says that the Mandarin translations done by ChatGPT are an order of magnitude better than DeepL though, which is interesting.
https://github.com/ogkalu2/Human-parity-on-machine-translati...