▲ 47 points
back
32 comments
The more I think about the difficulties of translation, the more I think that "Darmok" of Star Trek: TNG season 5, is the rule rather than the exception. Not only is it difficult to replicate the literal meaning of individual words and the structure of phrases, but it seems inextricably hard to replicate the sentiment and cultural context of words (nevermind idioms).
The Olympics recently reminded me of this, when news orgs reported that Brazillians had a colloquial word -- jeitinho -- that could be roughly translated as "MacGyver-ism" [0]. But even that rough translation needs a translation for all native English speakers who were born in America after the 90s.
Side note: The phrase "star-crossed lovers" is something that comes to mind that would be hard to translate. I threw it into Google Translate [1] and got "desventurados amantes", which I guess means "wretched lovers" (when I threw "desventurados" back into Spanish to English). Maybe a little gets lost in the translation, but at least Google Translate didn't return whatever the literal translation for "star" and "crossed" are.
[0] http://www.bloomberg.com/news/articles/2016-08-05/rio-olympi...
[1] https://translate.google.com/#auto/es/star-crossed%20lovers
This site [1] translates desventurado as "unfortunate, ill-fated", which is the same thing that the "star-crossed" metaphor is trying to convey [2] -- outside forces beyond their control are condemning their relationship. The "star" in context was recognized as controlling fate.
I think it translation is lacking the poetic beauty of the original -- in being more literal -- but gets the meaning across. I couldn't find it quickly, but it looks like the professional Spanish translations refer to them lovers "with stars opposed" [3] ("enamorados con estrellas opuestas").
[1] http://www.wordreference.com/es/en/translation.asp?spen=desv...
Human language is like a scaffold of metaphors, on upon the other. George Lakoff, of course, wrote the book on this.
> Not only is it difficult to replicate the literal meaning of individual words and the structure of phrases, but it seems inextricably hard to replicate the sentiment and cultural context of words (nevermind idioms).
And its even harder than either to replicate (or even reliably get the best available approximation of) the interaction of literal meaning with sentiment and cultural context that drives a lot of artistic use of language (not just something like poetry, but even the language of a lot of fiction or oratory.)
The article says "Another unvarying element of the Google Translate versions is the word "boat" for barco, a perfectly correct rendering of a simple Spanish noun. But in English, we have the expression, "When my ship comes in," ... Translating barco as "ship" brings an echo of that deep history of longing into our first, fleeting image of the narrator, Don Diego de Zama, who lives in the hope of a ship that will come to save him."
This sounds rather grandiloquent to me.
But more significantly, my father was the captain of a ship. One of his pet peeves was people mistakenly interchanging the words "boat" and "ship". The two terms are not equivalent (boat being a small vessel, and ship being a large vessel that might even be expected to carry a boat of its own, viz, a lifeboat). If he heard that human translator taking liberties with his nautical vocabulary, he'd have a cow.
So do we want Google to translate more accurately, or do we want it to strive for a literary flair?
I'll be underwhelmed until translation software provides configuration for core assumptions
One could even make the argument that a literal translation of 'star-crossed' would be equally correct; if the corresponding construct of adverse astrology exists in the language being translated into, the literal translation preserves the cosmological scope, while wretched, even though it demonstrates insight into an aspect of what 'star-crossed' means, strips away that crucial facet of the meaning. These lovers aren't wretched because of illness or poverty. They're wretched because of destiny.
+1 for Darmok. If you can handle his particular affinity for the sound of his own pen scratch, Hofstader's Le Ton Beau de Marot is another excellent discussion of how -hard- translation is.
Spoiler: No.
A better title would have been, "How Google fails at translations". The piece is actually interesting to read if you're into either language translations or how Google does stuff that seems magical, but the clickbait headline isn't doing it any favors.
I'd say the answer was "yes" to "Can Google Help Translate," but decisively "no" to "Can Google Translate a Classic Novel?" The interesting thing to me is that Google can supply reliable translations of stock phrases, which helps the translator identify them. More unusual constructions stand out because the translation is nonsensical in context.
Not according to the author of the article:
"No, Google Translate was in no way useful to my translation of the 1956 Argentine novel Zama: let's get that out of the way first thing."
But that's just poor English on the part of the author. The question she was really answering was "Did Google Help Translate a Classic Novel?". To which the answer, in her case, was no.
My guess is that that may be because she is fluent in Spanish and English and is a professional translator.
For the rest of us mortals who do occasionally attempt to read classic works in their original language, the answer to the headline is often "Yes, though not particularly well, but better than me".
Hmm,
So, the article notes that the Google translate of a challenging but not incomprehensible Spanish phrase hasn't improved in the four years the translator tested. I personally haven't noticed Facebook translate improve in any fashion over a similar period.
Yet I had heard of translation as one of the triumphs of recurrent neural networks. It may well be that such translations are doing better on some metric in some competitions yet that such formal improvements aren't palpable to humans yet. I have read previously things about translation now no longer needing language experts - which seems to say that perhaps automatic translation hasn't improved but merely gotten cheaper.
While this raises the "is this deep learning stuff hype?", I think AlphaGo shows that some deep learning stuff is pretty amazing. One might argue also that ConvNets have had the unambiguous triumphs which RNNs' triumph is a bit more ambiguous.
I'd be curious how folks who actually create this this technology would view progress of this sort.
Google Translate is not currently based on neural networks. The best NMT approaches are better than Google Translate, but they're currently too computationally intensive from what I've heard.
I agree Google Translate hasn't improved in a while. It's probably because so much new research is focused on neural nets, not on improving the classic statistical translation approaches.
Facebook, though, I think they DO use NMT. Reasons:
1. They employ many famous neural net people, but not any experts on statistical machine translation that I'm aware of.
2. They came into the machine translation game fairly late. It would be odd to start off implementing an approach perceived to be on the way out.
3. Translation between closely related languages (e.g. Norwegian and Swedish) is far, far better on Facebook than in Google Translate. This is one thing I expect to be a strength of neural machine translation.
They can probably afford to burn more computational cycles, since it's not an on-demand service in the same way.
I haven't found Facebook translate much to write home about either but I've mostly used Spanish to English and Tagalog to English.
I just went back and checked again and the translation quality ranged from ungrammatical-but-broadly-comprehensible to entirely incoherent.
Of course, neither Spanish nor Tagalog is very close to English.
Still, this gives me the impression the RNNs represent at best an incremental improvement in automatic translation.
I'd say that the fact that AlphaGo can play games against itself and there's a clear determiner what makes something successful(ie: a winning strategy), allows it to improve on it's own. I can't think of a way in which you could do this with language without having a human determine whether the output was readable/understandable or not.
This is pretty much it. The rules for Go are known. A perfect game of Go is an NP problem. We don't know how to frame the problem of Natural language interaction like that. We don't if it's even possible to do so.
One scary thing I've noticed is that Google Translate has a new failure mode. While it used to return obviously incomprehensible gibberish when things went off the rails, it now sometimes returns something plausible but subtly incorrect.
For example, it silently dropped the negation in a French -> English translation. Unfortunately, I can't remember the exact phrase but it was so semantically implausible--something like "regardless, we will be liable for any damages" that I went back and looked at the source text. It also sometimes replaces pronouns with oddly specific nouns: "He does X --> John does X" I assume these are both reflections of oddities in the training data.
Reminds me of the photocopier that started changing numbers in documents it copied.
http://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_...
Wow. That must have been absolutely maddening to discover.
It's not a new failure mode, it has always done that.
I've wondered the same thing. I haven't noted any large improvements in Google Translate despite the RNNs demonstrating big gains for at least 3 years now. However, in all the public comments by Googlers I've read, I'm not sure they have rolled out RNN to Google Translate publically yet - this is the same company that created its own ASICs to save electricity on big deployments because Nvidia isn't moving fast enough.
Following Betteridge's Law of Headlines (https://en.wikipedia.org/wiki/Betteridge%27s_law_of_headline...), the answer has to be "no".
To extend this point further, editors aren't simply unaware of Betteridge's Law. A lot of the time, headlines like this are written specifically to evoke the "No" response, and to entice the reader into asking the question, "Why would anyone even make such an argument?"
> Salí de la ciudad, ribera abajo, al encuentro solitario del barco que aguardaba, sin saber cuándo vendría.
> I left the city and made my way downriver alone, to meet the ship I awaited without knowing when it would come.
The human translation seems a little wrong to me.
I'll argue for the following:
I left the city, downstream, to the solitary meeting of the ship I awaited, not knowing when it would arrive.
"encuentro" seems to be a noun here, therefore "meeting". "solitario" is an adjective and describes the "meeting", not the narrator's manner of leaving.
I agree your translation is a lot closer to the original text, the original for example talks about a 'solitary meeting' with the ship, it does not say like in the translators' that the character got there alone, they could have gotten there with others and left them to wait for the boat alone for all the reader knows.
As much as the sentence in the article is quite nice, it seems more of a reinterpretation of the original compared to an exact translation like yours.
I find Google Translate good at translating sentence fragments but generally not full sentences. Also, any Google Translate Japanese to English translation is near jibberish.
J>E is the worst, Spanish is also not great. Google Translate is especially bad in places where the speakers will assume their interlocutors will infer significant amounts of information from context—because Google Translate has not actual knowledge of the context.
Japanese->English can't be "the worst", because I've seen Korean websites telling this one weird trick for better quality: "Translate Korean to Japanese first and then translate that to English, and the result is better than direct translation!"
¯\_(ツ)_/¯
When I tried, it was more of a hit and miss, though.
This is a perfect lesson in understanding the difference between computer and human intelligence. Google "knows" more languages than any human, and how to translate most words between languages, but when it comes to 1) parsing the meaning of the sentence and 2) bringing that into another language, it's generally pretty bad. It's no match for a person who speaks both languages adequately.
EDIT: It's the type of thing that makes me not fear the threat of AI. Seems like we are still pretty far from HAL. Missed his 2001 launch window.
Yes.... ish? I've done the following for French -> English on an essay: 1) Google translate all the single sentences or sentence fragments. Use an online dictionary when the answer looks odd. 2) Re-mangle the machine translated blocks into decent English. It mostly worked. Doing this process on a novel would hurt. ;-)
Yes, but manual translation of a novel would also hurt ;-)
What a beautiful sentence - and challenging.