Slashdot Mirror
Kurzweil: Human-Level Machine Translation By 2029
An anonymous reader writes "In a video interview with the Huffington Post, noted futurist Ray Kurzweil predicts that machines will reach human levels of translation quality by the year 2029. However, he was quick to highlight that even major technological advances in translation do not replace the need for language learning. 'Even the best translators can't fully translate literature,' he pointed out. 'Some things just can't be expressed in another language.'"
...but the meat is rotten.
Hello computer
You know, I'm a big sucker for futurism as anybody, but Ray Kurzweil makes a lot of predictions about future tech every couple of years, most of which never pan out anywhere near what he predicted. And each time Kurzweil makes a prediction, many of which are just way too optimistic or just play goofy in retrospect, the tech-minded people like slashdot lap it up.
Can't tech futurists find a better spokesman than Ray Kurzweil?
My postings are informational and does not constitute legal advice. Act on it at your risk.
'Some things just can't be expressed in another language.'
Then it's not a human idea, and probably wasn't expressed in the original language.
I can pull stuff out of my ass too, and I'd like to get paid for it.
He's full of bullsh*t. Watson proved that machine learning is possible and all that is needed is enough processing power / memory.
If you get to a place where you can associate every word sequence in a language to a correspondent word sequence in another, then we'll have almost perfect translators.
We'll be doing full translations a lot sooner. I predict this, so GIVE ME A SLASHDOT ARTICLE PLEASE! :)
Then how about this: Some things can't be expressed in another language without having culture notes as long as the original work itself.
Why is anybody listening to his opinion exactly?
Have you seen BLIO? I seem to remember him predicting that it would NOT be a piece of sh*t. But he was wrong.
Wow! Someone who's predictions are always wrong (and are obviously wrong to anyone who knows anything about what he's predicting) is predicting something cool! Yay! Let's put it on Slashdot!
, HAL (If you can't see the above, multilingual posting doesn't work; ah, the irony!)
http://en.wikipedia.org/wiki/Sisu
FCKGW 09F9 42
This guy has been wrong on his previous predictions as everyone already has been emphasizing, but what the fuck is the deal with such a specific year for his prediction? Why not round up to indicate it's a rough measure? 2029, really??
There is a fine line between being a cultivated citizen and being someone else's crop. - A. J. Patrick Liszkie
The guy is off his rocker !! It's the subtle kind of lunacy !!
"wtf?!? stuf! lol cul8r"
"What did my son say?"
"Sir, he inquires if things are quite as they seem. He wishes to seek tranquility, though is in good humour and will be pleased to visit again with you anon."
A feeling of having made the same mistake before: Deja Foobar
"Sisu" sounds a lot like the kind of "perseverance" that U.S. elementary school children hear about every February when teachers drop everything and teach Black history.
Can't tech futurists find a better spokesman than Ray Kurzweil?
Sure, I've always wanted to be a futurist.
In the future, things will be similar, but different in interesting ways. $5, pay up.
Ok, on a serious note, evolutionary, not revolutionary changes. The average Slashdot style website is a BBS without the modems. We don't get flying cars because that's not where cars need to be, we get faster cars, more fuel efficient cars, more luxury cars and cheaper cars.
Also, in the future, Slashdot.org car analogies will be replaced with .Slashdot flying car analogies.
..and I can express the universe with it.
Just after I've finished computing my own goedel number...
We don't understand how the brain works, which happens to be the most complex object known. Not only that, but we don't even fully grasp how a simple neuron actually works. Besides we can hardly define what intelligence actually is.
So IA is not a computing power problem, but a much deeper algorithmic/architectural issue. We just don't have a fucking model. Comparing a computer with a brain is like comparing a car with a horse and telling "look how superior a car is", which happens to be utter bullshit.
And that guy is telling us that in less than 20 years, we'll have a machine that can do one of the most complex operations of the brain? Just bullshit... As was his "documentary", The Transcendent Man. Unstructured, without a serious opposing view, without even trying to define or explain what "intelligence" is, but showing Kurzweil in a zero G plane trying to look at the camera. And he commands speaker fees of like $10000 to tell that stuff.
Nothing so see here, move along.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Kurzweil never predicted crowdsourcing. This didn't make it to Slashdot yet, but apparently the creator of reCaptcha is launching a service of human-aided mass translation.
It might just turn out that language problems are easier to solve by throwing social networks at them rather than hardware. Even if we eventually get hardware that would be able to do it, it would then be used for other problems that computers are already better at than humans.
somebody says something is X years in the future. The translation of "10 years in the future" is "I don't know. If they say "20 years in the future" that means "I really don't know" and if they say "50 years in the future" that means "Go ask my dog, he's more likely to give a correct answer than me."
Did you know 80 to 90% of the moderators on slashdot wouldn't recognize a troll even if one dragged them under a bridge.
...we need reliable voice recognition and automatic transcription. Google is also working on it, but
Smartphones need a better input device than a tiny QWERTY keyboard
No sig for the moment.
Sorry, hit the button without proofreading...
Previous post should read: "Google is also working on it, but it's still far from reliable"
No sig for the moment.
He has predicted a lot of things but seemed far too optimistic about the time lines. This one seems more reasonable. The real prediction is whether any humans will still be reading enough to care.
of mapping one culture to another. machine knows nothing about that. you can load all the databases in the world, but what machine lacks: immersion in the culture, being part of it, have the attitude about it, system of values, so to speak.
The stuff Kurzweil says, well, it's meaningless He's a great inventor, yes, no one argues about that. But the *big picture* he's trying to promote - well, it's mostly about promoting himself.
He past incorrectness was vague, but now he is perfect in his errors. I call that an improvement.
I dug through this thread looking for the surely inevitable reply to ask you for actual evidence to back your claim (that Kurzweil's predictions are often wrong), so I could mod it up.
I can't find one, so I sacrifice my option to mod this thread to call you out. Can you back up your claim?
I certainly don't think Kurzweil has been perfect in his prediction, but I think he does quite a good job. Here is my evidence: http://singularityhub.com/2010/01/19/kurzweil-defends-predictions-for-2009-says-he-is-102-for-108/
The predictions criticized in that article are definitely not entirely accurate, but they're also pretty damn good for having been made in 1998. We are close to where Kurzweil says we should be.
Please defend with counterexamples :-)
Yeah Google translate isn't THAT good at this stage. Here's what it did to your post. My favourite mistranslation: was in Russian. "I did a lot of work with translators and even they get it wrong, so I think Kurzweil is actually a way" can be taken to mean the opposite of "so I think Kurzweil is actually off in a way." which is very misleading. I'd use it cautiously for casual conversation even now, knowing there will be misunderstanding. But I wouldn't dare dream of reliability for anything critical in a decade. It would only need to tranlsate "Well we don't want war" to "We want war" once to kill millions.
Here is your post, translated English -> French -> English
Yeah, but for years, machine translation was blocked on Alta Vista babelfish gloomy ... which was essentially a one to one translation of the dictionary (often without the help of the definition of right) to hilariously bad translations.
A couple of years, Google translate gave a big bump in the whole concept of using United Nations documents (which are usually 5 + languagels) as a faithful translation. It has many failures, but translations of Babel Babble past often unreadable for something that often range from decent translation of something that you can see if you put some thought into it.
I did a lot of work with translators and even they make mistakes, so I think Kurzweil is really a way out. IMO, by the end of this decade, machine translation will often be good enough (really, Google Translate should start looking for context clues and I do not think it will be 19 years away) but n 'there will ever be perfect because the language is not perfect. Watch humans communicate sometimes, it's not a strict protocol, misunderstandings can occur any time between people. But when a machine is wrong, people will point to it as evil, instead of the very nature of language.
Here is your post, translated English -> Chinese -> English
Yes, but over the years, machine translation is boring to stay in Alta Vista's Babelfish platform ... which is basically a one to one translation dictionary (usually not the right to use the definition of) the cheerful bad translation.
A couple years ago, Google translation of a big jolt to the whole concept of using United Nations documents (which is usually 5 + languagels) as a reliable translation. It has a lot of hiccups, but the translation is often unreadable babble from Babel things, often translated from a decent range of things you can calculate, if you want to go some.
I've done a lot of work, even their translation wrong, so I think Kuziweier actually way off. Haishizuzhi, by the end of this decade, machine translation is often not good enough (really, Google translator need to start looking for more context clues, I can not believe it will be 19 years later), but there will never be perfect because language itself is not perfect. A look at human communication, it is not a strict protocol, can be misunderstanding between the people what happened. However, when a machine to get it wrong, people will point to it's bad, rather than the nature of language itself.
Here is your post, translated English -> Russian -> English
Yes, but for many years, machine translation is stuck in a dreary Babelfish Alta Vista in ... which was basically the dictionary translation 12:59 (often without proper identification) to have fun bad translations.
A couple of years ago, Google translate gave a big blow to the whole concept of United Nations documents (which are usually 5 + languagels) as a reliable translation. It has a lot of hiccups, but the translations often traveled from unreadable babble babel to what is often varied from decent translation of the fact that you can find out if you put some thought into it.
I did a lot of work with translators and even they get it wrong, so I think Kurzweil is actually a way. IMO, by the end of this decade, machine translation will often be
These posts express my own personal views, not those of my employer
I can see machine translations improving significantly but until we develop proper machine AI I don't think they're going to be near perfect.
I've done translations from Japanese -> English occasionally for work and I can tell you that sentences often encode different things. As an example I once did a translation of a letter regarding animal imports. Japanese has no distinction between singular and plural and gender isn't encoded at a grammatical level to the degree it is in English. This created problems because nowhere in the original letter was the gender or number of animals involved mentioned. In order to translate the letter into correct, natural English I ended up having to ring around to find the number and gender of the animals involved. As another example, I have a friend who did a translation for a court where the original Japanese had been scattered with borrowed English terms written in Roman characters. It created real problems for her because they definitely didn't fit into the English sentences at all but they were still there and needed to be translated accurately. These were after all documents that had the potential to decide a court-case!
Language is hard and translations definitely have a degree of creativity and artistic skill involved, even technical translations, as in significantly different languages you often find yourself having to rewrite sentences structures that simply don't exist in the target language for the translation. The summary highlights it to a degree but it's not just literature where you come across things that can7t be directly translated.
and most likely deceased. Not to be mean, but the guy keeps adjusting his time frames sooner rather than later. Unless there is some massive push for I-Robot the idea we all used robots for our physical chores can never happen when people need to earn a living, and no we aren't all going to be dealing with Future Crime either.
He claims 86%.
Got any alternate figures, or some sci-fi authors we can compare to?
Why would anyone engrave "Elbereth"?
"Transhumanist with no domain-specific experience makes unambitious prediction about state of technology in distant future".
Good machine translation is five years away, and has been for the last 40 years.
He's totally reliable.... in that he is systematically wrong. One could make a bundle betting against every single one of his "predictions", if you could find any bookie dumb enough to take the bet.
If you want to see some good examples of what "human-level translation" might mean, look around on the engrish.com site, which shows you just how good humans can be at this task. Note that almost everything there is "official" in some sense, intended to inform the public of something; very few of the examples are intended to be funny.
There's also a good "Engrish" classification at failblog.org, if you want lots more examples from a different source.
The Language Log blog has lots of discussions of examples such as these, generally trying to answer the question "How did the translator go so wrong in this case?" You can learn a lot about the "gotchas" of human languages by reading these discussions.
So I'm not too impressed by yet another prediction of "human-level" machine translation. That's a pretty low hurdle to cross, if it means the level of accuracy that current human translators routinely consider good enough to put on signs, menus, etc. Yes, professional translators will generally do a lot better. But that's not what the summary or TFA predicts; they just predict an unspecified "human-level" capability, which would be satisfied by the examples in the above "Engrish" sites.
Those who do study history are doomed to stand helplessly by while everyone else repeats it.
Kurzweil believes in Strong AI which, amongst its absurdities, appears to claim that the reals are countable. In more detail, the continuity of consciousness that we experience is inexplicable in a universe in which Strong AI could be true. Unfortunately the proof is a little complicated and I haven't got all the fine detail figured out yet.
John_Chalisque
Just dumb human translators down.
Isn't that what our government doing?
News flash: there are already computers smarter than Ray Kurzweil. That singularity was kind of a big whoop: we should have set a higher bar, like achieving AI that's smarter than real scientists.
I don't know about his futurism, but damn, the man can build a synthesizer. Music is the universal language. :)
When was the last time something worth reading was written or said in a language other than English? We're not just talking decades here, but generations, possibly centuries.
Can we start marking Kurzweil articles as dupes?
Granted, this is a little less ridiculous than some of his past claims — machine translation has improved a lot in the O(decade) since Babelfish — but translation algorithms are still context-blind for the foreseeable future because no one's yet found a computationally feasible shortcut for the "every Bayesian probability is dependent on every other Bayesian probability" case that natural language seems to teeter in the direction of. Moore's law isn't going to fix it, either, because it's not a polynomial-time problem that can be fixed by just throwing faster clocks or more cores at the problem. We've gotten as far as we have by using dumber, polynomial-time algorithms and just throwing supertankers of training data at the problem, but in the end it's no more contextual than Dissociated Press.
Incidentally, it's clear that natural language translation is not actually as difficult as computing the maximum likelihood of a fully cross-connected Bayes network (i.e. superpolynomial-time), or else the human brain itself would be stumped. But we don't know enough about which shortcuts are useful for convincing human brains versus which shortcuts result in "the vodka is good but the meat is rotten" translations. That means we're stuck theorizing from our armchairs, throwing algorithmic crap at the wall and seeing what sticks, or maybe poking at brains with pointy sticks and fMRIs. On this matter, date predictions are worthless. The breakthrough could come tomorrow or hundreds of years from now, and Kurzweil is no better equipped to predict the date than my cat is.
Range Voting: preference intensity matters
I hate prophets. Religion. Technology. Whatever.
Where is my human level OCR, Mr. Kurzweil? Still waiting for that one.
Due to not paying the journalists who provide(d) the stories?
Now they need slashdot to get readers? Maybe the editors of slashdot should reduce eating popcorn and drinking of soda and instead start to work a little more...
Also predictions for 2029 are nice... so far away that probably nobody remembers then what nonsense mr kurzweil talked decades before......
Kurzweil? Same guy that "predicts" achieving immortality (accidentally, I swear) just before his own death? Wishful thinking is strong with this one. Do not take him seriously.
What modern Obelix would say today? Of course, "Those crazy Americans!".
The linguistics skills of the NSA's A.I. are on par with a human, including its translation capability. Ray Kurzweil doesn't know his ass from his elbow.
At my job (plug: http://gramtrans.com/ ), we're doing human level machine translation now, if by human you're talking about high school level of quality.
But in difference to Google or Microsoft or all the previous attempts, we're NOT doing Statistical Machine Translation; we make Rule-Based MT.
Google's way is to scan huge amounts of parallel texts for good translation candidates, which works to some degree, but it has an upper limit that no amount of parallel text will help them surpass.
Our way involves writing a huge amount of rules so that our software actually understands the input text, and then rules for how things should be translated. This takes a lot longer to develop as you might imagine, but the results are great and there is no limit to how good it gets...if something isn't quite right, add a rule for it.
However, because it takes so long to develop and that we're only 2 people working on it, we only have a few really good translation chains.
Today's most successful approaches use statistical (or hybrid) machine translation. Moses is a great open source example that presents a fine baseline when training with good parallel language data and language models. However, it uses Bayesian statistics that uses phrase co-occurrences and words in context to define a probabilistic generative model for translation. While it provides impressive results, it is still a far cry from fluent (and adequate) output. Additionally, if he wants to claim that speech translation will also be up to par, he is also requiring researchers to perfect speech recognition. Granted, he's giving us researchers 18 years to develop a solution, I still think he's underestimating the challenges statistical systems need to overcome.
If you're interested in learning more about machine translation, I invite you to take a look at the EuroMatrixPlus project.
He's been saying this for years. 2029 = computers reach human intelligence, 2040 = Singularity, where we either merge with them, or we get left behind. It's all in the 2009 documentary Transcendent Man. So... why is this news?
I8-D
is a science fiction author that cannot think of a plot.
It seems to me there's a major issue that Kurzweil is glossing over. When humans speak, regardless of what language is used, there is actually very little information passing through media, whether that be words moving through the air via longitudinal waves or moving over circuits as voltage differentials, or whatever. The actual information passing is very small.
What is NOT small though, is the utterly vast shared experience that humans have. The words themselves are just little signposts or markers for these concepts. Some of them are practical, some of them are abstract. I'm not saying you can't ever teach the essence of these concepts to computers, but it seems to me that it's difficult to know even where to start. For me this is a non-trivial problem and any 'deeper' translation that isn't just shifting words and syntax around will have to grapple with the meaning and intent behind language.
What seems more likely to me is that we could well create some kind of complex AI but it would be essentially 'alien' to us and our everyday experience/dispositions.
I do not want your cheap brainburning drugs. They are useless for work. And I am a working man today.
I mean, Ray Kurzweil believes in "alkalinized water" and dismisses just adding sodium bicarbonate, because the HNO3- molecule won't work as well as the HO- molecule... which entirely disregards that HNO3- interacts with H2O to make H2NO3 and HO-. http://glowing-health.com/alkaline-water/ray-kurzweil-alkaine-water.html
Sodium bicarbonate (aka baking soda) has a molecular formula of NaHCO3. There is no nitrogen in it, and it seems you merely need to s/N/C in your quote. Somewhat ironically given your particular typo, HNO3 is nitric acid, which is a very strong acid (pKa =~ -1.4, which implies that it will completely dissociate in water) and is most certainly not alkaline. Furthermore, note that HNO3 would have no charge as written, and NO3- (nitrate) would have a - charge for the polyatomic ion. However, your typo and the subject molecules are interesting from a biochemistry perspective.
Bicarbonate (HCO3-) is a weak acid/base and is amphiprotic and amphoteric. Many biological organisms, humans included, use bicarbonate as a buffer molecule to maintain blood pH within the very narrow pH band required for the organism's survival.
The rule of thumb for buffer solutions is that they are most effective within pH +/- 1 of the pKa of the relevant moiety. Of course, in biological systems, the buffer isn't "static" and the organism expends energy to maintain homeostasis. Here's an example of the blood pH bicarbonate buffer using the Hendersen-Hasselbach equation. You can see the blood pH (7.4) is outside the +/- 1 pH of the pKa of bicarbonate (~6.1), and therefore the dissociation ratio is 20:1. Obviously, this buffer solution wouldn't work well if the organism didn't constantly rebalance it.
Your body uses the lungs to maintain the buffer, and this is why your blood can become more alkaline if you hyperventilate (respiratory alkalosis).
Haha, how's that for an off-topic tangent?