Touch of the author lost and severely degraded in machine translation

First of all, THANK YOU :heart_eyes: for creating this wonderful website. Etimonline has always been a to go place whenever I see a new word, or whenever I just want to read.
That is why I am not really satisfied with the way how the translated versions are created. Please consider hiring human, professional, mother tongue translators, and using fonts that are common for each language. Machine translation may work in some languages, but definitely not in my language, which is Japanese.

2 Likes

I apologize for that. Text translated by a machine is always wrong and insulting to a human mind. I wish it could be done as you ask, but there is a great deal of text on etymonline, in several languages, and no money to pay the translators. And the text is constantly being edited and updated.

Thank you very much for your feedback.

Indeed, the machine translation we’re currently using needs improvement and should only be taken as a reference. For the sake of accuracy, please refer to the original text in English. Translating the content of the dictionary is particularly challenging due to the use of abbreviated expressions for concise writing. A lot of subtle information is lost in translation.

We have thought about involving community volunteers and users in providing feedback and improving translations, but we still don’t have a feasible plan yet.

Hi, Doug and Chongwei
Thank you for your reply. :smiley:
It may be more insulting to the great mind and work of the author than to the readers. I hope the translated versions will be as good as the original version someday.

1 Like

Absolutely agree with your point of view, and I understand your concerns.

However, a large part of our user base comes from non-English speaking countries, and it’s very difficult for most of them to read the original English text directly. Therefore, as a compromise, we’ve employed machine translation to help them read and understand. Of course, there’s a lot that needs to be improved, and we will continue to invest resources to enhance the quality of the translations. :nerd_face:

For instance, I see this snip from Japanese etymonline quoted on a Japanese website explaining what a “hostage” is. Translating the Japanese back into English with Google, it reads:

It is a proto-Indo-European root meaning “stranger, guest, host,” and it means “one who has a duty of mutual hospitality,” referring to the “relations of mutual exchange” that were so important to ancient Indo-European societies. It represents. However, this word is divisive, as the Gentiles are both guests and potential enemies.

Granted, it is a double translation. But the original text on the site is:

Proto-Indo-European root meaning “stranger, guest, host,” properly “someone with whom one has reciprocal duties of hospitality,” representing “a mutual exchange relationship highly important to ancient Indo-European society” [Watkins]. But as strangers are potential enemies as well as guests, the word has a forked path.

Starts out well enough then flies off the rails. Never send a machine to do a human job. I’d be embarrassed to have written that, but computers make me write that in Japan.

Granted, machine translation is a minefield. But not all hope is lost.

Quite obviously in order to provide a reasonably acceptable translation the translator must understand the subject and master pretty well both the source and the target languages, which is more than I would dare say even for the majority of human translators. But apparently this seemingly impossible goal is approaching at an impressive speed thanks to modern LLM’s like ChatGPT.
Lately I fed ChatGPT4 with a piece originally written by me in Italian, and it provided two quite decent - occasionally excellent - translations in English and German: the bloody thing translated the meaning of the sentences, not just the words with an eye to grammar and syntax. Some of the translated sentences had actually a pretty different syntactic structure, but the meaning was definitely there - all of it, unscathed, occasionally even a bit enhanced.
Then, as I confronted it with the one single mistake it made - translating an idiom literally rather than ad sensum - it humbly apologized, said that translating idioms is an art, and promised me never to do it again. Flabbergasting.

Even a little disquieting, if you will: that means that in some way the “thing” understood what I meant to say and took the liberty of changing the semantic structure to better render the sense.
I’m wondering what the future has in stock for us in the next few years…

1 Like

We have found similar surprises when using GPT as a translation tool, when it has far exceeded our expectations. But more often than not, it’s the frustration: GPT has a hard time delivering consistent results, and sometimes it makes stupid mistakes, and we don’t know when it’s making them.

1 Like

Absolutely true, and that’s why I said that the goal is approaching: we aren’t there yet, and the devil knows when the LLM’s will become at least as reliable as a good human translator (sloppy and mediocre ones are never in short supply).
But Hope is the last goddess, and hoping is still tax-free :wink:

1 Like