I get it: translation is hard. Heck, I’m a reasonably bright native English speaker and often have difficulty translating my own thoughts into understandable English.
This is a message that was posted in a woodcarving group:
“Hello, I am writing a message to help my father. And I see myself. Only on the American or Canadian woodcarving site and no response. It’s just for the books. And politeness. It’s when it’s repetitive that it’s not funny. But you how many millions to be connected. I find that very embarrassing. Administrators must take their jobs seriously. I have already reported them, I pass the imfermire contest as if I was going to sew up a person at any time, have a nice day everyone.”
The author’s native language is, I think, Italian. Or perhaps French. I suspect not an English speaker, although it’s possible that his grasp of English is better than my grasp of his native language. I cannot tell if the message is the result of automatic translation, or if the author did the translation himself with the help of a Italian-to-English dictionary. Either way, I cannot make any sense of it.
Which is weird. I’ve seen bad translations before. But usually I can get the gist of a message that’s been automatically translated: a “hook” that gives me a broad idea, and from there puzzle out a few details. For example, the word “imfermire” in the above text looked promising. It looks like a misspelling of the Italian word “infermiere,” meaning “nurse.” The best I can guess is that the author is having trouble getting some woodcarving books for his father. Not sure where the nurse comes in.
The author’s responses to comments provide no useful information. Which isn’t too much of a surprise. I imagine he has to translate the question, then write and translate a response. The combined errors inherent in that process aren’t conducive to understanding. Automatic translation software is especially bad at round-tripping because errors accumulate very quickly.
Can the technology that powers the new crop of generative AIs be put to good use in the automatic translation space? I imagine feeding an Italian-to-English translation to a tool that can leverage its knowledge of translation errors and spit out a short and meaningful summary. Is such a tool within our grasp?