Slashdot Mirror
DARPA Starts Ultimate Language Translation Project
An anonymous reader writes "Defense Advanced Research Projects Agency (DARPA) has launched the ultimate speech translation engine project that would be capable of real-time interpretation of television and radio programs as well as printed or online textual information in order to be summarized, abstracted, and presented to human analysts emphasizing points of particular interest." If combined with the tower of babel project we discussed earlier, it could only lead to awesomeness.
to understand DARPAese. (Try reading some of their PPT slides.)
By 'lead to awesomeness' do you mean 'lead to you not having to attempt to edit summaries and fail at both grammar and spelling'?
My work here is dung.
The right to offend is far more important than the right not to be offended. (Rowan Atkinson)
If you consider that now the government will be able to spy on you in your native language to be awesome, then I suppose giving the Feds this sort of technology can only lead to awesomeness.
Surveillance of civilian populations under the guise of "monitoring terrorists" is not something that I'd consider awesome. Irksome, yes. But not awesome.
Just feed this new system a few reruns of Japanese television game shows. After that, we will be safe from automated snooping for at least another decade. As a plus, all artificial intelligence projects at the DARPA will be set back by another decade as well.
[
What's wrong with using humans? This is exactly what humans are good at. While there most certainly are fields where machines can replace humans, this is _not_ one of them.
d c.php/
http://lyricslist.com/lyrics/artist_albums/16/ac-
Dragon Speak is going out of business of guess
awesomeness, to be sure, if you consider the ultimate outcome. Remember, this is DARPA, so they're looking at potential military applications. I read it as: "translate (military) communications in real-time, ... then destroy one or both parties."
All I really want is a free online translator for web pages (ala Babelfish and Google) that aren't terrible at it. Seriously, the quality of Babelfish translations has stayed constant since it came on the scene in the late 90s, even though machine translation in general has made some rather significant advances. I don't really use them enough to justify plopping down $500 on the professional packages, but the current systems are just terrible.
I read the internet for the articles.
w0nT b3 phUn 1ph th3y d0nT d3w l337sp3ak
owned
Es una trampa!
its lame, its old and yet i cannot help it
i can see a translated japanese movie coming...
"all your base are belong to us, make your time"
If you look like your passport photo, you're too ill to travel. - Will Kommen
As long as it can handle swear words and a good variety of sex acts I'm interested in I say thumbs up!
This project, along with CMU's Tower of Babel, certainly get props in the coolness category, but the practicality is still lacking. I believe DARPA is barking up the wrong tree for now, or at least biting off more than they can chew.
Speech Recognition is the hardest problem to tackle on the path to recognition, and MUST be addressed before there is a viable real-time (or even delayed) translation engine. Currently, even the best speech recognition software can achieve at best ~80% accuracy when faced with a large vocabulary with no limits on speakers/dialects, and this level of accuracy is typically not achieved in real-time. While this 80% level is actually pretty good when transcribing to text (since the reader can typically decipher what the computer meant), it's downright awful if trying to translate the resulting text to another language.
For example, if I say "I like ice cream" into voice recognition software and 'hears' "I like, I scream", the reader might understand what this means, particularly if they say it in context and aloud. However, let's say we translate each sentence into Spanish ("Tengo gusto del helado" and "Tengo gusto, yo grito" respectively, according to Babel Fish), and the speaker would be completely lost as the out of context phrases don't sound anything alike. In a natural language translation, even under relatively accurate recognition scenarios, would be frought with misunderstandings.
Once speech recognition is tackled, it's just a matter of translation then voice synthesis. Fortunately these problems aren't nearly as difficult, and current solutions would suffice (with the only pitfall being poor grammer in the destination language, and a robotic sounding voice).
Crack - Free with every butt and set of boobs
I read that as "Defense Against Research Projects Agency"!
Ultimate Language Translation researchers should probably compete along a stretch of the Mojave desert so as not to injure or offend nearby native speakers
Obviously such a thing will not work well without an advertising filter (imagine an analyzer sifting through washing powder ads).
So they will have to develop one.
This will be integrated into VCRs to stop/start recording when advertising starts/stops.
Great!
Slashdot headline ten years from now... "Creators forgot to implement 1337 speak into translator matrix..."
Or the WRATH OF AN ANGRY GOD! heh
"Hey... didn't I make them all speak different languages to teach those uppity humans a lesson? Now they what? The end routed me on that one? Oh I don't think so!"
I'm a fiscal conservative, it's a pity we don't have a political party anymore
The space station is being built again. India is planning manned missions into space. A shift in power in the US Government. Now we're creating a Universal Translator! How exciting these times we live in.
when people do that.
But seriously, I've basically made my living off of DARPA grants and I fully support the criticism leveled at them above. It is truly a classic government buearacracy, very wasteful, not entirely straight about what they are doing, and you have to have personal connections to get money from them.
...Romulan, Klingon, and Vulcan?
GetOuttaMySpace - The Anti-Social Network
The metal gear project. I mean, honestly, the DARPA chief isn't going to be jailed in the secret arctic base for the ultimat language translator.
In Soviet Russia, dots slash you!
Seems likely to be very useful for specifically what they suggest it is for (flagging potentially interesting material for further review by human analysts, a kind of time-saving filtering device for the limited pool of translators available.)
But beyond that, I wouldn't give too much faith in any kind of mechanical translation as particularly reliable on its own except on narrow kinds of material. It conceivably might work for strictly literal usages, or for fairly stable idiomatic uses, but unless you have frequent collection and incorporation of usage data from every culture and subculture that may be a source of translated material, its going to fail, sometimes subtly and sometimes spectacularly, for a lot of idiom. Similarly, even within the same language, different groups using it will have different idiomatic uses that sometimes will produce different or opposing meanings for similar usages, which will require accurate identification of the source at more than just the language level to get correct results from. There's a lot of evolving cultural context that informs the use of language...
init 11 - for when you need that edge.
To properly translate all the nuances of some languages actually requires a lot of skill, and sometimes translating can be ask much interpreting as anything. Granted, this is something a human could handle better than a machine, but the problem is that humans also have a bias. Yes, there have been cases wherein human translation has caused problems because of bias or even due to being outright wrong.
I reminds me of the old joke:
Guard: Now tell me where you hid the money, or you will suffer
Translator: Tell him where the money is, or you will suffer
Prisoner: I'll never speak
Translator: He says he won't tell you
Guard: *putting gun to prisoner's head* Tell him I will blow his brains out if he doesn't tell me immediately
Translator: He will shoot you in the head unless you tell him now
Prisoner: I buried a million dollars under the floorboards in the old woodshed
Translator: *pauses* He says you don't have the guts to shoot him...
While being still quite far from an adaptive Star Trek-style universal translator, it is conceivable that one day with the help of portable devices (like Palm/cell phones/iPod) that we could indeed have an on-demand personal translator that would work anywhere. I think this is the beginning of such a capability.
The US department of Defense is openly claiming to be able to solve one of the world's hardest AI problems, and you don't believe it? Big surprise.
If the US military had anything close to real A.I., you wouldn't hear about it. It would be a classified information.
The NSA would love to have anything close to a system capable of understanding language as well as a native speaker can; as would the CIA, or any other clandestine organization. Any system smart enough to understand and generate English probably also came with a breakthrough in CS theory that will give them better tanks, planes, and communications systems. And those would be classified, too.
In short, this is just an excuse to spend money, and to hide the funding for any secret research projects that they really are working on.
What is it about "ultimate, do-everything" project sponsored by the government that sets off every alarm bell signaling imminent failure?
Back in 2000 or 2001, I saw IBM television ads about telephones that translate in real-time. In the ad an English-speaking woman was speaking to a Turkish person, with the telephone doing the translation. This was promised in the "near future" by IBM. Anyone know what became of it?
By the way, mod parent up. "Offtopic" is not a good replacement for "I don't get the joke".
that may be true for most languages, but just about every government defense/security agency i've heard of is desperate for speakers of arabic, persian, malay, javanese, pashtu, etc. And they much prefer native speakers, for obvious reasons -- very few people ever achieve native-like proficiency in a language they learn after 12 or so (the critical period for language acquisition is pretty brief, sadly).
that said, a skilled non-native translator will no doubt beat the crap out of a computer. this DARPA project sounds pretty far-fetched, even for them. i doubt they'll end up with much more than the rudimentary string of words altavista creates. (seriously, if you want a good laugh, hook babelfish up to a chinese or japanese newspaper -- see if you can guess what the headlines are actually about).
All your base.
Slashdot Burying Stories About Slashdot Media Owned
>the government already has the language skills it needs even without a whizbang translation machine.
Sadly, they don't. The FBI has something like two guys who speak Arabic, and there are numerous instances in the news recently where some fed is bewailing the lack of language skills in his department. On a diplomatic note, how many US Ambassadors actually speak the language of their host country? It might be useful if they had some way to understand the locals.Some mornings it's hardly worth chewing through the restraints to get out of bed.
Headline: "Lack of translator support kill "l337" and "IM" speak. World rejoices as collective intelligence level rises."
fahqtut [fully automated high quality translation of unrestricted text] will never happen, boys and girls. Language is too diverse and expressions are often untranslatable. I work in this industry and all the technology I've seen sofar is laughably inadequate. Yes, you can have accurate translations for a restricted range of text, straightforward, present tense stuff sure, it'll work mostly. As soon as you start with the real language that people use your system hits the bricks in a decidedly unflattering manner.
When complex sentences with multiple clauses are used [and this is stuff you're translating, right?] it just cannot keep up. On top of that there's noise in the background, interruptions, people don't enunciate well, use crummy language, use the wrong word altogether, contradict themselves in the same sentence, start sentences they don't complete. This DARPA uber translator thingamajig is going to just come along and happily munch through all that gibberish and deliver crisp and pristine language? Ain't happenin'.
They've been playing with this technology for decades now. Language is our most flexible, most diverse tool. The only way they're ever going to make that work with any measure of reliability is when they figure out how brains work and find a way to build one.
Hey, somebody is going to be working with some exciting technology. I say: let them play.
In order to recognize speech, one needs a context-sensitive parser. In order to make a context-sensitive parser which is fast enough to interpret the text, the computer should have the pattern-matching capacity of a grown up human. The human brain contains 500-1000 trillion synapses! even if one makes the assumption that one synapse equals one bit, in order to understand the context, one would need a computer with a tremendous amount of memory which could be searched in parallel.
Of course if you narrow the problem down to specific terms, then it is doable. But then it would not be 'ultimate' any more.
They just speak a little sslloowweerr and LOUDER! The natives usually catch on.
Deleted
I attend one of the many Universities where DOD research is currently being conducted. Portions of our graduate student body and faculty are working on the powered armor concept in conjunction with UC-Berkley (they're doing the frame and kinematics, we're doing the control theory/system and power supply). We're actually making quite a bit of progress in the field of alternative batteries (the current iteration is a peroxide-fueled hydraulic hybrid-type system widget) and mechanical interface control theory application. So, while God knows we won't see cap' troopers in 'suits any time soon, we are at least progressing towards that end while developing widely applicable technologies along the way (this is, if I may remind you, the way many technologies we love dearly were spun off from the space program et. al).
You'd think the FBI would be a prime customer for something like this, but apparently keeping a huge backlog of documents to translate and a staff that's too small to handle it is more important to the mechanics of their bureaucracy.
The point being, if this tech works, great, but will it be used?
Translation *is* the hardest part in the Gale project. So much harder than in the current evaluations the translations are so bad that the impact of the ASR errors on the final result is not significantly detectable it seems. We hope that the MT teams are going to make some massive progress fast (they may, they get a *lot* of new data from the project) so that working on the ASR actually means something, but more importantly so that the project goes on.
OG.
Take that, Wikipedia! It's not just some plot device!
Errr...I mean, soon, it won't be a plot device anymore!
Crap, I mean, eventually it might not be just a plot device...
I mean...oh, fuck it. This is DARPA after all.
... ASR on people talking to each other naturally Just Doesn't Work[tm]. As in 70-80% error rate or worse.
Gale is about TV/Radio news, not random people conversations.
OG.
Seriously though, I just don't believe it. I've worked on a number of DARPA robot projects, and have heard a lot of their babble. They claim to be funding all these fantastic ideas, but none of them ever work except in a limited capacity.
..." and translated it as "I absolutely would mind to tell you about ..." which is the exact opposite. Many languages, such as Russian, Spanish and Portuguese (and no doubt others) use double negatives to express negation. "I don't know nobody" is quite correct in Russian, Spanish and Portuguese although it is quite grammatically incorrect in English if your intention was to say "I don't know anybody". Programs that translate into English from languages that use double negatives often fail to correctly translate the negation. Maybe there are some that get it right, but I've never seen any. Text translation programs are very poor at distinguishing between words that have uses as different parts of speech. Here's an example:
This is a big pipe dream that is extremely unlikely to work any time soon. How do I know that? Right now, I think it would be reasonable to conclude that computer technology today is good enough to do accurate text translation. Can it? Well, it depends on how picky you are. There are always mistakes, sometimes glaring ones, in text to text translation programs. I can speak Russian and for convenience (to get quick rough translations) at one time I owned what is probably the best Russian-English text translation program. It's much more accurate than Babelfish. It still left a lot to be desired. It would be about 80-90% accurate, but no more. I remember one time when it took a statement in Russian that said "I absolutely would not mind to tell you about
She sings like an angel.
In this sentence, "like" is an adverb, but it can also be a verb ("She likes to go shopping."). A text translation program might fail to correctly understand that "like" is an adverb here and say something like:
She sings and angel is pleasing to her.
I could give a lot more examples, but these are enough. If we can't even do a better job right now at text translation, how on earth is DARPA going to get speech translation right? This is the kind of project that gets funded by idiots who have never studied foreign languages and believe that the Star Trek idea of a Universal Translator is only a few years away.
Yeah. Computers monitoring all "public" forms of communication - e.g. telephone, e-mail, radio, &c. Key phrase[s] pop up, computer flags it, sends a message to the Thought Police, who come, and bust down your door.
While it may not be practical to have a living human listening in on e.g. every telephone conversation going on at the same time, with government money, one could throw a lot of computing power at this issue. This is DARPA, after all.
I, for one, do not welcome our new, Digital Thought Police, Overlords.
We could probably collect as many cellphone and internet messages as we want, but there arent enough people to sift through them.
Several terrorists in Colorado Supermax prisons sent over a hundred unread Arabic letters overseas because they have just one part time guy reading them down there. Quite a scandal there.
Just say the title out loud to get some idea of why speech recognition is hard, nevermind translation. Translation has long been regarded as "AI-complete" because to do it well you have to understand what is being said, which involves solving all the other difficult AI problems. The current translation systems are lousy because they don't understand what is being said and most of them don't even attempt to.
So my guess is that this program will be a boondoggle for researchers with little practical result.
Alright...anyone who is not aware that all these AI "translators" have been a complete failure should grab a AI history book. It might be incrementally better to what's out there; however, I'm almost certain it'll fail to accomplish it set out to do...at least from any practical perspective.
Now if we could just teach English to the Marines.
... speach researchers at Carnegie Mellon University has recently demoed a prototype of a device ...
I'm sure the multi-lingual people out there are laughing at the very concept of "Real-time translation". Unless you're doing something trivial (Italian to Spanish?), this just isn't possible.
Some languages place verbs at the very end of the sentence. Assuming that the computer could understand each of the words, the entire sentence still has to be re-composed in English. For long sentences, the speaker has already moved on.
Other languages, like French, use some crazy sentence structures that effectively do the same thing. A French sentence can be like a string of comma-separated pronouns with the object of the pronoun in the very last bit.
Both of these cases will induce delays and still don't account for some problems of context. In my own personal experience, I've found spots where whole paragraphs really need to be translated if we want to keep the original meaning.
Seriously, I think "Real-time" could be roughly translated :) as within a few seconds.
I look forward to them distributing all the translated content on torrent, as a part of the freedom of info act or something, so that we can get English subbed foreign TV programs. I think that the anime shares will be most popular
be very suck. If you anything secret to say you better be hurry.
Previously: "Linux... Toward the Sunrise..." Now: "Linux... Toward the-- No, now, part of Every Sunrise"
Seriously, try it. Input the sentence, "My dog has fleas." Go from English to Japanese, copy and paste the Japanese into the entry box, and translate back to English. "There is a chisel in my dog."
Just one of many reasons that I'm not that worried about my career as a Japanese - English translator. :P
"What in the name of Fats Waller is that?"
"A four-foot prune."
I am billdar, and I approve this message.
As I just noted in another post, the current publicly available state of machine translation gives me little to fear as a professional translator. You note:
I'd like to point out that "poor grammar" can often have disastrous consequences for the meaning. Take my previous example, "My dog has fleas." Babelfish's Japanese output is "Watashi no inu ni nomi ga aru." This backtranslates to "There is a chisel in my dog." The bolded word here is the kicker -- aru means "is / has / to be" depending on context, but is reserved mostly for inanimate subjects. Nomi is the subject here, which could be either "flea(s)" or "chisel(s)". When using aru, it becomes a chisel, an inanimate. Using correct grammar, Babelfish should have used iru , "is / has / to be" for *animate* subjects.
Let's take another example from Spanish. "I'm lost" should usually be translated as "Estoy perdido", using estoy to describe a transient state. With bad grammar, a translator (machine or otherwise) might use soy for permanent states, and produce "Soy perdido", which means something more like "I've lost my virginity." Woot!
Granted, context can make a lot of this much clearer, but it's still awfully muddy when bad grammar is thrown into the mix. Your comment suggests that poor grammar is a minor problem, but I'd beg to differ -- bad grammar can radically alter the meaning on the one hand, and it requires a full understanding of the context and intent of the source text to produce an accurate target, which is no mean feat in programming terms.
Which, happily enough given the complexity of the problem, leaves me pretty secure job-wise. :)
Cheers,
"What in the name of Fats Waller is that?"
"A four-foot prune."
I think we need to sit down and think about analog computers again, using waves to add and subtract and create filters. Of course, this can be simulated in digital space with fourier transforms and stuff.
Cool! Amazing Toys.
In 10 years the opening Wikipedia article sentence will be:
The DARPA Ultimate Language Translation Project by effectively removing all barriers to communication between different races and cultures, has caused more and bloodier wars than anything else in the history of creation.
We could stop canning translators because they are gay (no liberty should be exempted from removal in our eternal war against evil, except the liberty from having gays look at your crotch, I guess).
OR
We could start learning some foreign languages. Everyone who graduates high school should learn at least two. Fluently. And no, not Spanish. A language NOT spoken by your neighbors. A *foreign* language. Arabic would be damned helpful.
I'd agree, as there are two fields here that are extremely difficult -- voice recognition and machine translation -- which makes this all seem like so much pie in the sky. Anyone who's ever used voice recognition knows how spotty it can be, and anyone who's ever played with Babelfish (like this guy) knows how much humour can result. Now imagine these two lovely examples put to use on the battlefield, or at intel HQ, and some very unhappy possibilities arise.
I'm all a fan of research for learning's sake, but the article here makes it sound like we're going to get something Real Soon Now (TM), which I seriously doubt. All of which seems to lend credence to your statement:
So yeah, "wow that would be handy", but don't anyone hold their breath. :p
"What in the name of Fats Waller is that?"
"A four-foot prune."
This is an interesting concept, and I'd be really interested to work on it, given that I'm a business analyst with an undergrad degree in Linguistics. However, I have to question how accurate this system would be, and if they really understand the complexity of the problem they're trying to solve. Cited is a current project (elsewhere):
The device is dubbed the "Tower of Babel". Currently the device can handle a small vocabulary of 100-200 words at about 80% accuracy, and accuracy drops off significantly beyond that vocabulary.
With a vocabulary this small, 80% accuracy is VERY BAD. Imagine if 2/10 of the words I use to communicate are UNINTELLIGIBLE. Perhaps you might *&^% I am trying to %&@@# you. I would live in great fear of a government that trusted such a system for spying. It might be a decent first cut at translation that could then be fixed by a translator, but could never, ever be very useful as intended (that is: "capable of real-time interpretation of television and radio programs").
The problem is more solvable in "formal" speech (i.e. a news broadcast). The subjects and words are usually clear and idiom/jargon free. But imagine getting a computer to understand "Wayne's World" or any common sitcom, or a conversation between frieds. How do you translate "Doood, that party was goin' OFF last night"?
There are other linguistic and cultural hurdles to overcome. Polish, for example, does not have separate words for "hand" and "arm". So, a literal translation (and remember, computers are VERY literal) from Polish would proably be "I broke my arm", which if stated in English could have been "I broke my hand". How about emotions? Some languages have many words for "love", whereas English (generally) only has one. When I say I love my brother, this is not meant as sexual statement, but I have to choose the correct word for "love" in the target language to keep it "brotherly".
Which brings us to context. I may (in a twisted universe) actually mean that as a sexual statement. How do you ever expect a computer to pick up on the context (which often is not clear to a human) and immediately give a correct translation?
To develop a literal, word-to-word translator is not a stunning achievement. Automating a language-to-language dictionary cuold be a first semester comp sci project. I suspect that this is basically what this project will produce. But if you want something better, it pays to delve into the world of Linguistics to understand how humans use language, starting from the phonemic level all the way up to semantic parsing and psycolinguistics. Take what you learn there, and apply it to a system like this and your results will be better. I'm not going to promise perfection, but certainly better than 80%.
--Jonathan
Then, of course there are cultural differences too. A Xhosa girl would like to be told "You remind me of that fat cow over there", whereas your average American chick might not.
Engineering is the art of compromise.
Requires conceptual processing which no one has solved yet.
Fergeddaboutit.
Until conceptual processing is able to be performed, ANY form of human language translation will be inadequate. It might be usable in some respects, but not adequate for most real purposes.
Richard Steven Hack - This sig is TOO GODDAMN SHORT TO DO ANYTHING USEFUL WITH! MORONS!
I thought the Universal Translators tech came after the hyperdrive one, or is this the one that only translates into an incomprehensible dead language? Joking aside I don't envy them the challenge, humans who speak the same language often have a hard enough time understanding each other.
Although many may think this new technology is a bad idea, think about the communication unit in Star Trek. I know this is the real world and all, but advances like this can lead to a better understanding of each other. A unifying device like this can make views and beliefs from other cultures more understandable and somehow through this, we'll be able to make this world better in some small way.
If man must go to the moon then yes, he will go there....
Hooray for Herb!
I don't know if they'll pull this off, but if they do (and I can get access to it), I'll be able to pull out all the old albums, like some of Bob Dylan's, and actually enjoy the lyrics. "Ohhh! So that's what the song is about!"