Submitted by Such a development would cause a soul-shattering upheaval in my mental life. Although I fully understand the fascination of trying to get machines to translate well, I am not in the least eager to see human translators replaced by inanimate machines. Indeed, the idea frightens and revolts me. To my mind, translation is an incredibly subtle art that draws constantly on one’s many years of experience in life, and on one’s creative imagination. If, some “fine” day, human translators were to become relics of the past, my respect for the human mind would be profoundly shaken, and the shock would leave me reeling with terrible confusion and immense, permanent sadness.
As a translator myself, I can indeed confirm Google Translate is complete and utter garbage, but the idea that I would “mourn” the end of translators seems outlandish to me. The unstoppable march of technology has eliminated countless jobs over the course of human existence, and if translators are next, I don’t see any reason to mourn the end of my occupation. Of course, it’d suck for me personally, but that’s about it.
That being said, I’m not afraid of running out of work any time soon. Google Translate’s results are pretty terrible, and they only seem to be getting worse for me, instead of getting better. There’s no doubt in my mind that machine translation will eventually get good enough, but I think it’ll take at least another 20 years, if not more, to get there.
Yes it’s horrible. Makes me wonder if there’s any translators or scholars involved in google (getting paid for it).
I think google treats it like a crappy “work on it for free for us via snippet†without digging deep into context and syntax rules.
Or perhaps it is already achieving its intended purpose. It allows someone with zero understanding of a certain language to translate a very basic gist of a paragraph, getting at least a basic understanding of the passage.
Any words that seem completely incorrect can be exchanged for alternatives until something resembling sense can be made.
Of course it’s not as good as a real translator. But it’s free, it’s instant, and infinitely better than simply staring at an unfamiliar language and knowing nothing of what it means.
This!
I don’t use Google Translate to read poetry.
I don’t use Google Translate to read my favorite SF in it’s original lanuage.
I use it to read science/technical papers to see if there is interesting research. And for that it works.
For other stuff I understand even the translators themselves sometimes can not agree on how sentences should be translated. Heck, as an anime fan I find it hard to explain to some English speaking friends what I mean when I reference something that is clear in Japanese but does not translate 1-to-1 in Western culture.
Edited 2018-02-06 15:48 UTC
Actually, I would think Google Translate is a lot better at translating poetry than actual novels, especially SF ones
The basic gist is OK for a translation of a 3 word phrase, but if you’re trying to translate a page or a sign (a feature that they market as one of the major features), or taking a picture of something complex and with depth such as an essay or a colloquial use of language, it fails horribly.
My uses are more scholar, to which Google Translate is not suitable and definitively misses the contextual analysis of languages, a very important issue when translating things longer than one sentence or with subtle variances in syntax and regional use of words.
Not in my experience. I moved the Netherlands, and do not speak anything like sufficient Dutch. I’ve used Google translate to translate letters from both the local council & the Dutch tax office, and it did an adequate job of allowing me to understand what the letter was about, at least enough for me to home in on the important parts and translate those more accurately.
Google Translate is also perfectly adequate for translating most web pages (via. Chrome’s “Translate to…” functionality).
Translating languages is basically just a proof-of-concept for something much greater – that is teaching machines to understand our communication.
This feat is a cornerstone for integrating robots into our society not only as mere translators but our servants in daily life. Business-wise the applications would have no limit.
Actually I’m seeing that real translators that translate to my native language or from it to English are also becoming worse, this is more prevalent as the companies tend to not hire even a half specialized translators but lazy general ones that when they find an unknown or ambiguous word don’t search for its meaning and base it’s translation in a supposition.
ML, AI and “neural networks” just try to learn and decide from experience. In other words learning from statistical analysis.
Do anyone really think that statistical analysis can translate emotions? That can decide if something is beautiful or ugly? Computers are deterministic machines, have no notion about emotion and will never have.
Edited 2018-02-06 00:45 UTC
These automated translators are complete failure and they have so many poor quality translations or complete opposite meaning – until recently GT translated “배불러요” (I’m full) in Korean as “I’m hungry”, or “내꺼야” (mine as for belonging) into Polish “kopalnia” – which has meaning of underground mine. Until we develop fully sentient AI which can perform on at least of the human level, I think human translators are pretty safe.
Edited 2018-02-06 03:29 UTC
That sounds suspiciously like it translated the Korean into English and then into Polish, and since “mine” is a homonym in English for both possession and underground tunnels, it perhaps randomly chose the latter meaning for the final translation.
I have noticed that both Google and Bing translators work better the more context/sentence structure you provide for them. Perhaps we need a mode where it asks you for context instead of just randomly offering a word-to-word translation when all you provide is one word.
And speaking of Bing, it still does what you indicated Google used to: “내꺼야” becomes “Kopalni”.
That is how Google translate seems to work for many languages: Translate to English first and thence to whatever ‘minor’ language. Many one word translations from, say, Spanish to Belarusian result in an English word because not enough Belarusian speakers have corrected it.
Yeap, English as pivot language is the way to go.
… implying that human minds are NOT deterministic machines?
How can you be so sure?
A bold claim.
I’m not sure, but I am also not sure human mind IS deterministic and can be emulated by computers. This is something we, humans, love to debate (as computer would never do :-)) and I am not seeing this stopping any time soon.
But I would bet (and feel) it’s not deterministic
Ultimately, they will likely be able to simulate a human brain… we already simulate neural systems of simpler beings.
Oh, and human learning is also based on the past, past experience, on ~statistic (what works and what doesn’t)
You won’t go far if you resort to insults to convince people. And doubt is important to ingenuity.
What insults?… And what I wrote in the previous post is precisely about doubt… (of widespread beliefs in our exceptionality)
Edited 2018-02-07 20:42 UTC
Humans are Turing complete and thus at least our termination is indeterminable.
That definition makes computers not deterministic.
Theoretical infinitely larger computers anyway. But yes, I believe computers can be programmed to be non-deterministic.
The difference is that humans are naturally non-deterministic and can be trained to work deterministic. While current computers are the other way.
Edited 2018-02-06 23:52 UTC
Uh, no. This is provably false. You can make a machine which is non-deterministic. For example, you can use a radioactive sample to create a true hardware random number generator and incorporate it into the same enclosure as a computer. If you then program the computer to use the RNG, it’s output will be non-deterministic. But the source of non-determinism is the quantum uncertainty of the radioactive sample, not the programming of the computer. You can not program a computer to be non-deterministic. Period. Full stop. Any claim to the contrary is either provably false or it is using one of the terms in a non standard way, such as referring to a computer plus hardware RNG device simply as a computer.
Whether or not human beings are non-deterministic is very much an open question.
I think the author of the article is missing the point of google translate as used by most people, which is to read the menu at restaurants and ask for directions.
I don’t know how much research will be going into a broader contextual transation ability but I’m not sure google translate, a free tool, is where this would surface to disrupt the translation service industry worth ~$38bn
Jehovah’s Witneses. The most translated site in the world(600 languagesjw.org), and even the most translated video. (450 languages)
Kinda funny they’re trying to reach so many people, when only 144,000 of them can actually go to heaven.
Source ? Pardon, verse ?
I was intrigued and google autocompleted “jehova 144” to this https://www.jw.org/en/bible-teachings/questions/go-to-heaven/
I am not going to argue about any of this, just pointing out that your answer was longer than my Google search
At least Google translated your search into a correct answer.
Well, “is the earth flat” doesn’t have provided me a… flat out answer.
First of all: Badum tsss
Second: I copy/pasted the text from https://www.jw.org/en/bible-teachings/questions/go-to-heaven/ into https://translate.google.com/#auto/nl. I wouldn’t call the result “complete and utter garbish” (Thom and I are Dutch, but please test with your own language
You can clearly see some nice things, which I suspect are user suggested. For example “Skip to content” becomes “Meteen naar de inhoud” (Immediately to the content)
You can also see some errors, especially when there is no context
* Home -> Huis (House))
* Bible teachings -> Bijbel lesgeven (bible tutoring)
But the main article, which has full sentences and context is actually translated fairly nicely.
Would a professional EN-NL translator do a better job? Surely
Would a random person that understands English and Dutch do a better job? Not quite sure
Would a random person do a better job translating instantly to-and-from 100 languages? Don’t be ridiculous!
Conclusion: YEAH for professional translators, but YEAH for Google translate as well
Sidenote: I deal with translated help-pages on a daily basis and they have gotten quite good in the last two years. The pages are now (only?) produced in English and immediately published in all other languages. Professional translators seem to be used as well, but not so much to translate individual pages but to improve the automatic translation in general, basically working to make themselves unneeded for this work that nobody likes to do.
Usually similar languages are translated pretty well.
Like Dutch, German and English (same/similar ancestors).
But try something like Japanese and English, good luck !
These are languages with different sentence structure, etc. that’s really hard.
I try Japanese fairly regularly because my Japanese is at extreme beginnerslevel and my wife is Japanese. It surely works less well, but it still makes things understandable.
Don’t rely on it, don’t translate poetry or literature, but treat it like most IT: A useful tool among many tools
Definitely still better than nothing. 🙂
Most of the time English/Japanese not such good results, sometimes you get a surprise better result than expected.
At my work the translators will dump documents into GT, then fix up the wording. They claim it is faster doing it this way instead of doing a straight translation, otherwise they can’t meet the demands they have daily.
If I use GT for just a phrase and send it to them to check, they always reply back saying it needs the certain changes and why, so clearly they catch errors.
Belgium is a multilingual country, so as a Belgian website developer, I have to make a lot of multilingual sites.
GT saves me a lot of typing. Improving a GT translated text is much faster the writing a translation from scratch.
I’ve found it easier to translate well if I do it from scratch. I often use Google Translate (and other tools) when I have trouble coming up with the right word, though.
Google Translate is far from garbage. It serves its’ purpose just fine much of the time and will only continue to get better. For the people who insist that it’s complete crap, I’d like to see you do better. There’s no shortage of people who struggle with their own native language & grammar, much less do a great job of translating it.
The Google Translate AI has also been successfully helping decode ancient languages that have stumped real people who specialize in that field. I’m pretty sure none of them would describe GT as `horrible`, `complete and utter garbage`, and `complete failure`.
Are you high? It’s perfectly serviceable for simpler phrases and sentences.
If you want garbage there’s Facebook’s translations.
People can give machines a run for their money when it comes to unintentionally hilarious translations.
Long before there was such a thing as Google translate, human translators gave use the joys of “All your base are belong to us”. And more recently, they’ve introduced us to the pleasures of delicacies like “Stir fried wikipedia” ( http://ourfounder.typepad.com/leblog/2007/10/jimmy-wales-gro.html ).
In a school in western Norway some years ago, a bunch of students began a letter-writing exercise in Spanish with a cheerful “Brezal!”, after having looked up “hei” in a Norwegian-Spanish dictionary and selected the wrong Spanish word (“hei” in Norwegian means “hello”, but can also mean “heath”).
Would you hire a translator who didn’t know how to say hello in the target language?
AI seems to be already better than average human. That’s probably enough to bring marked improvement to the world… I’d guess Thom would be fine with a self-driving car that’s better than average driver (that would already bring down road accidents/fatalities)
Machines won’t start translating as human beings do unless they start actually understanding the text which they don’t and I’m afraid we’ll need general AI for that which is currently nothing more than a pipe dream.
Now I don’t remember the quote, but paraphrasing it basically was: technology that just seems around the corner will take a few more years than you would expect and technology that seems far away is actually closer than you think.
Predicting is actually hard. 🙂
Edit here is an article that talks about the future:
https://gizmodo.com/the-ai-revolution-how-far-away-are-our-robot-ove…
Edited 2018-02-06 12:40 UTC
This is the most advanced Artificial Narrow Intelligence we have now as far as I know:
https://deepmind.com/blog/alphago-zero-learning-scratch/
One AI researcher said:
__
For over 50 years there have been about 10 human abilities we haven’t even figured out how we would do that with AI. We’ve now started/some what solved the first one.
___
And in my view, we’ve only made proper practical progress on the first because we now have the hardware to do it.
And the hardware is on a Moore’s Law curve… so we might be going faster from here… and after that a lot faster.
Edited 2018-02-06 13:01 UTC
As a technical translator, there’s no chance for GT to replace humans anytime soon for translating even slightly complex material adequately for business purposes. You get a rough idea of the content, that’s all. I believe it’s become better now than it used to be, but still, GT is no competitor to me in my business.
There is one additional reason why it won’t happen soon: because the source itself is often not perfect. An important task for a translator is to convey the author’s intent and not necessarily what they literally said.
Edited 2018-02-06 12:54 UTC
Are you a programmer? That is what we have to do all the time.
I’m not a programmer, but among other things, I localize SW for a big vendor. They do it, too. The results are, again, better than they used to be, but still shit by the common standards of translation industry. I guess they know it too, because they pay me for machine-translated strings the same as for untranslated ones. (Of course, by default, they wanted to pay less for MT strings but I refused, and they didn’t even attempt to haggle.)
In narrow contexts, you can get reasonably good results from MT by using tightly-controlled language and being extremely careful with terminology, but that’s not always possible. Where I live, I’ve heard about such setups being used in the energy industry.
Moreover, as an additional complication, now people who write the English source barely speak English themselves. This confuses the hell out of MT engines and meatware translators alike.
I can’t imagine computers will ever wholly replace human translators. Some languages are written in a way that two different words can be written with the same letters, and the reader has to know the context to know which word is being used (e.g. Arabic, which omits short vowels from the standard written form). At the very least, a proof-reader will be required to correct mistakes and make sure idioms are correctly translated, and they will need to know both languages.
That’s not a problem for computers. Watson showed that contextual understanding is possible, given the right training.
The problem is that language is a moving target, influenced by culture on all time scales at varying proportions, and still spoken more often than written. By the time things get written down, the culture will have already moved on.
So for computers to make progress, they very much have to be equipped with voice recognition, and exposed to language and culture all the time, and it must start to communicate with itself in that language, like we do when we talk to ourselves or thinking of what to say.
Google Translate can be passable for reading articles on the Web, such as technical articles in Italian. It’s not perfect, but good enough to follow across the odd glitch.
A different skill altogether is translation of literary works, where a good translation may have to contend, for example, with an author’s deliberate use of double meanings. This is where the work of a skilled translator really shines, and I cannot see machine translation approaching the apex of this art for quite some time yet.
I’ve been using GT for Polish-to-English and Russian-to-English translations of documents I’ve come across in my genealogy research. Nowhere near what I’d expect from a professional, but definitely good enough for me to make some sense of the document.
GT is quite trash, but some are better. Case in point: http://www.deepl.com
They’re a department of linguee and use their vast linguistic database. It’s far (very far) from perfect, but from what I can tell (i.e. when I can translate myself) it is much better than GT.
I’m now waiting for it to integrate other languages I can’t speak and I’m interested in.
In Deepl’s translation of La Marseillaise:
Aux armes, citoyens,
Formez vos bataillons,
Marchons, marchons!
Qu’un sang impur
Abreuve nos sillons!
becomes…
Guns, citizens,
Train your battalions,
Let’s walk, let’s walk!
That impure blood
Show us our furrows!
“Let’s Walk, let’s walk!” is priceless. Not exactly martially motivating. And the less said about furrows the better.
Wikipedia has that stanza as:
To arms, citizens,
Form your battalions,
Let’s march, let’s march!
Let an impure blood
Soak our fields!
In the United States, its near impossible to get some people to understand plain standard English written at a 7th grade level, even amongst my graduate degree holding peers. If two humans speaking the same language can’t understand each other with the frequent explanation of a phrase, I think you are being too hard on the computers. Humans suck at comprehension even more than they do.
Is differentiating “its” from “it’s” at 7th grade level?
I’m completely bilingual in French & English and recently had to translate a technical (medical) paper that I had written into French.
I thought I’d give Google Translate a chance, just to laugh at the howlers, but it wasn’t bad at all.
Spelling was spot on, except when there were ambiguities, and it got all the accents right. The French was rather clumsy, which needed smoothing out, but the grammar was mostly correct. The punctuation was awful, but then French does have all sorts of arcane rules.
I was pleasantly surprised at how well it did, though I probably spent more time editing than I would have done writing from scratch, but not much.
What I _did_ find interesting was that the paper came out some 30% longer – made realise how succinct English can be – e.g., no word for “siblings” in French, it has to be « frères et soeurs » (don’t forget the spaces before the guillemets).
MaC
English invariably has appropriate words for each and every possible situation. We simply borrow or make up any term necessary eg schadenfreude or weekend.
Google is switching from rules based translation to deep learning. The new method is already almost as good as human translators for some langauges.
https://www.theverge.com/2016/9/27/13078138/google-translate-ai-mach…
I predict that native or near-native real time translation will be available withing five years for most major language pairs eg English-Spanish.