That is an interesting project. I would like to take that to the next level, where translation is the easy part. It's converting old handwritten letters, journals, and texts with now archaic cursive, into text. We have fascinating material from before WWII, in various languages - but trying to decipher it accurately, is a challenge. As the older generations died off, it is hard to find anyone to read that text.
I’ve got my great-grandfather’s autobiography and one of the striking features is the language. It’s extremely matter of fact with very little prose to connect his life together, just a lot of details like “moved to X in year Y” and “went to school for X between Y and Z years”. It’s got a lot of facts, but little coherent narrative.
If someone is important enough to warrant a biographer they’ll put more work into the narrative but old autobiographies seem to be very mechanical records.
Isn't a narrative just a way of presenting facts? Given facts, you can write a narrative. The more facts you have, the more detailed the narrative can be.
Not exactly. Peter Turchi came up with the analogy of mapmaking for narrative which really goes a long way to making sense of this since in both it’s about being able to find a sensible way to connect the details in a usable way.
Great read ! My father has published 12 books in his native language, I am in process of converting them in English using chatgpt. This gives me idea !!
Most probably the author knows this but there are some photos and information here [1]. His grandfather is a member of "The Righteous Among the Nations" [2].
Translation has always been interesting to me as I speak Mandarin, Swedish, and English.
I think a lot of subtlety gets lost in translations, even from person to person. Is the more "academically correct" translation better, or is the one that captures the slang and the vibe in modern terms and memes in a way we can feel the speaker more correct?
I think most people who choose the former look at translations as something that is absolute and objective, but I think translations, like most communication, needs to be subjective to both the speaker and the reader.
And just as humans can get each of these "wrong," LLMs can and will, too
There’s a great essay by Lydia Davis which appeared in the Paris Review in 2014 (I really should have a better citation, but oh well), which talks about the challenges of translation. To a certain extent, even if you’re writing something original, you’re still writing a translation in terms of trying to find a way to express the idea in your mind within the limitations of the words available to you.
If I want to learn about a culture and language, an objective translation is what I want. E.g. I want to learn the idioms of the time and place, not have them changed by the whim of the translator to something similar but perhaps changing the intent
If I want to persuade or illuminate the topic, then by all means a more approachable translation with modern idiom is more useful.
An example I know of is, an early Dolby cassette recording from a Japanese source had the marketing copy "Smells of the lamp of high technology!" Which is meaningless, even humorous to a Westerner. Not good marketing copy.
But I enjoyed learning something of Japanese culture. That things could 'smell of' something as a way of imagining them.
I look at translations as works by themselves. I don't think it's possible to recreate the feeling of experiencing a work in its original language without actually learning that language. One truly cannot experience Shakespeare unless one reads him in the original Klingon.
That doesn't mean that translations are useless, but you're kidding yourself if you think you understand the phrase that translated to "smells of" in the same way that a Japanese person would. It's evocative to you, but native speakers may not even think of it as a metaphor. As an example, do you think of a human face when someone refers to the "face" of a clock? Do you imagine an organist when someone says they are "pulling out all the stops"?
Only real Japanese speakers can experience works of art in Japanese, just as Germans who don't speak English can't actually experience Hasselhoff the way Americans do, try as they might. People can create works of art inspired by the originals, and call them translations, but they are new works.
I would watch the movie adaptation of a book I enjoyed, or vice versa, but I wouldn't read a book in translation that I'd already read in the original, and I don't think many people would. (I might read the original after the translation).
So in some sense a translation is less of a unique work than a film adaptation.
Classical scholars are an exception here, who might read Homer in half a dozen translations. But that's for two reasons: firstly, classical scholars are likely translators themselves, or at least students of the art of translation; secondly, there are only so many great Ancient Greek texts to read in the original.
> E.g. I want to learn the idioms of the time and place, not have them changed by the whim of the translator to something similar but perhaps changing the intent
But how can you learn the meaning of an idiom if it's translated literally? For example, even in context, "and now we have the salad" can be pretty baffling.
I am also a Mandarin speaker, and a very interesting thing happens when I try to read classical Chinese by dictionary or my modern understanding of the language: It's vague, lacking nuance, and basically unreadable. It requires very talented translators and skilled academics to make readable modern translations.
For classical Chinese poetry, the best English translations have been other poets, even though most have been unable to read any Chinese! Check out "Lament of the Frontier Guard" by Tang dynasty poet Li Bo (李白), as written by Ezra Pound more than 100 years ago:
By the North Gate, the wind blows full of sand,
Lonely from the beginning of time until now!
Trees fall, the grass goes yellow with autumn.
I climb the towers and towers
to watch out the barbarous land:
I am sure professional translators would find issues on technical and stylistic points, but could any be as evocative with their own translations?
I think we have to be prepared to give LLMs some leeway with language ... or empower humans (both editors and readers) with the tools to "tune" the translations as appropriate to their tastes, understanding, and subjective needs. This is basically what Korny did, trying different tools and consulting other humans as needed with certain colloquialisms.
Translating poetry is, in many ways, an impossibility, which is why the best translations are by poets and not linguists as they’re essentially writing a new poem with the skeleton of the original.
I thought about something similar today. My kid (2.5 years) watches some stuff on Netflix which is translated to Swedish from English.
It's very obvious that the voice over and subtitles are done by different teams/companies, because they often mismatch.
What I find more interesting is how they differ, and it's related to the "slang vs academic" but it's about culture and context.
Stuff like word puns that doesn't work when translated directly, or named being "translated" in one but used directly in the other.
Honestly I think it's extremely lazy and cheap on Netflix' part, for adults it's one thing, but for kids it's pretty confusing. (My son is not reading the subtitles yet though)
I was watching a German show on Amazon Prime with both German audio and subtitles. I was confused as to why on practically every line, the wording had been changed despite the meaning being similar.
I eventually figured out that the German subtitles had been machine-translated from the English subtitles!
To add to this, things can also get "lost in translation" even if you're reading something in your native language.
If you're reading something written in the 19th or early 20th century then even something you think you understand might be going over your head, because it's referencing some cultural context obvious to anyone at the time.
It’s also way harder for Chinese too. Different regions have different dialects despite using the same writing system. Different dialects usually mean different culture which means they use different words in different ways like slangs.
reply