I heard from a friend his religion, the Baha'i Faith, is advocating for the need for an international language. Basically all the nations would agree on a second language that all the schools would make mandatory or something. People's native language wouldn't go away, just everyone's second language would be the same. I mean think of all the barriers that are put up because people can't communicate.

Of course implementing this looks impossible right now...

If it's starting to happen though, it looks like english is becoming that international language.

(he also told me the Baha'is in Iran are terribly persecuted and recently some Baha'i leaders have been imprisoned for no reason, just like the 1980s when thousands were executed just for not recanting their faith... sorta sucks)

5/24/2008 11:31:30 PM

Isn't English pretty much the de facto international language of today. Used to be french until the last couple centuries then Imperial England reached its height followed by the US gaining dominance post-WW2

5/24/2008 11:47:35 PM

Quote :

"Attn: users with >1000 posts"

5/24/2008 11:53:01 PM

Didn't they try that with Esperanto and fail?

But yeah, I always thought English was the de facto international language, at least for now. And also, studying a second language in school in no way means people will actually be competent in it. That's something I see every day where I teach now.

5/24/2008 11:59:09 PM

When I was in India I found that there were so many dialects of Hindi that a person from northern India could hardly communicate with someone from southern India. To bridge the gap they all just learned really really shitty english. It's pretty funny to watch.

5/25/2008 3:22:31 AM

International language is a silly concept. First off, language and culture are intertwined and can hardly be divorced from each other. To create a usable "international language" it would have to somehow encapsulate a generic culture as well so as to convey idioms, figures of speech, etc.

More so -- on a more basic level -- the whole concept is just ridiculous for humans. The most spoken language is English. The second most spoken (albeit not as "widely") is Mandarin. Try to find the common ground in basic communication between Mandarin and English. There's not much. English as a language is not tonal; Mandarin is exquisitely tonal down to the syllable level. What is the "international language" that approximates between a non-tonal and a tonal language exactly? How about the written script that approximates between romanization and phonographs?

As to English becoming some kind of "global standard" -- well, yes, but I think mostly as a "glue" language. People will always prefer to speak in their native tongues -- and by speak, I mean, to communicate subtle cultural assumptions that don't translate well. Not to mention that native fluency in a language tends to draw a "trust" barrier around a culture in many situations.

This language business isn't easy and it's not just about syntax and grammar.

5/25/2008 6:37:23 AM

It's kind of a silly idea. For starters, any constructed language (WHAT'S UP, ESPERANTO?!?) is more or less bound to fail when compared to languages that have more naturally evolved. Now, the universal language doesn't have to be constructed, you could indeed just make it English or French or whatever, but that won't really accomplish anything. Say you can teach English to literally every person in the world. What is accomplished that makes it worth the effort? And, finally, I think there is GREAT value in different peoples speaking different languages, primarily because of the way that language can shape culture, thoughts, and even the development of the brain.

http://en.wikipedia.org/wiki/Sapir%E2%80%93Whorf_hypothesis

5/25/2008 10:12:47 AM

Quote :

"Say you can teach English to literally every person in the world. What is accomplished that makes it worth the effort?"

Easy communication between all people. I'd say that's significant. Currently, language barriers limit the exchange of ideas and create various costs. Teaching everyone a single language would require a high initial investment but have longterm benefits. That said, I suspect technology will solve this problem, both through split-second translators and vastly superior ways to teach language.

5/25/2008 3:10:11 PM

^

Technology will not solve this problem anytime soon. Natural language processing is a ridiculously hard problem (theoretically, not just in terms of "throwing man hours at it").

But more so, I think your notion of "cost reduction" is purely theoretical. The existence of a "universal language" is inherently based on some cultural dominance in the global economy. Which is what we have now, between the U.S. and the E.U. But fast forward twenty or thirty years, in which timespan China will likely have a bigger GDP by a significant margin than either U.S. or E.U. Will it be possible then to expect all parties at the economic bargaining table to accept English as "the glue language?"

I think the international language discussion is primarily a U.S.-centric worldview; it usually consists of our college graduates in coffee shops drinking a latte saying "yea, man, I'd love a world language! As long as it's MY language!"

5/25/2008 3:46:07 PM

"Hey Goldenviper, nice weather today, huh?"
"Someday technology will make weather obsolete."
(pause)
"Are you always on?"

---

The basic system of lingua franca has sufficed pretty well in the past. The average person has little need to speak anything other than their national/regional language. People in international business or relations have historically picked a common language -- French for a while, now generally English -- and in that capacity, it makes sense. For the world at large? A lot of cost for not a lot of benefit.

5/25/2008 3:48:03 PM

Uh, I don't think y'all realize how good current translators are or how quickly processing power is advancing.

Quote :

"But more so, I think your notion of "cost reduction" is purely theoretical."

So translators don't get paid in real dollars?

Quote :

"The existence of a "universal language" is inherently based on some cultural dominance in the global economy."

To get started, perhaps. Once it happened, the process might sustain itself.

5/25/2008 3:53:48 PM

vi vs. emacs

5/25/2008 3:56:08 PM

^^

Quote :

"Uh, I don't think y'all realize how good current translators are or how quickly processing power is advancing."

Gee, GoldenViper, livin' and workin' out here in Silicon Valley, I'd never have had ANY clue how fast processing power is advancing in the computer industry!

Of course linear (or polynomial, for that matter) increases in processing power do NOT in any way, shape, or form come close to allowing us to solve natural language processing. In computing there are classes of problems that are "too hard" to solve algorithmically, just as in physics you can't apply Newton's laws to all classes of problems. The solutions that approximate NLP are getting better, sure, but they are not going to be good enough for real speech/"universal translator" applications anytime soon.

Quote :

" So translators don't get paid in real dollars?"

They do, but you missed my point. Your assumption was that the "cost" of teaching everyone the "world language" is only borne once. But what about if and when the agreed-upon world language changes due to very real, non-theoretical changes in the makeup of the world economy / dominant cultures / etc.

I'd rather think bearing the cost twice, or N times, kind of torpedoes your "cost reduction" viewpoint.

Quote :

"Once it happened, the process might sustain itself."

Of course, you haven't given any evidence to back up this viewpoint; whereas I can easily point to the fact that the economic balance -- and therefore the cultural balance -- in the world is changing very rapidly. Rapidly, say, in favor of a country whose government tells its citizens what words they must use for "husband" and "wife," what every character in the script looks like, etc. etc.

[Edited on May 25, 2008 at 4:16 PM. Reason : foo]

[Edited on May 25, 2008 at 4:16 PM. Reason : foo]

5/25/2008 4:14:59 PM

Quote :

"The solutions that approximate NLP are getting better, sure, but they are not going to be good enough for real speech/"universal translator" applications anytime soon."

Depends what you mean by soon. With a decade or two, absolutely. Folks already use free computer translators to access books in languages they don't understand. The prototypes are far, far better.

Quote :

"But what about if and when the agreed-upon world language changes due to very real, non-theoretical changes in the makeup of the world economy / dominant cultures / etc."

If handled as a traditional lingua franca, sure. We're talking about making everyone proficient in one language. Don't assume they'd necessarily stop using this language once it became established. If the ties between countries were strong enough, they'd continue using the tongue all speak. Yes, by extension this would enshrine one culture over others. It's still the most efficient longterm solution.

5/25/2008 4:34:10 PM

^

It won't happen in a decade or two, just as we won't have perpetual motion machines, either. Seriously, you just don't know what you're talking about here. People use translators as an _aid_ in understanding other languages but they're laughably bad; as to prototypes being "far, far better," that's like saying they're 100% better while still being a million times too bad for widespread usage (and an order of magnitude worse than that in terms of making language differences "obsolete").

There are areas of NLP that aren't even well understood theoretically (Context being a huge one), much less in terms of practical implementation. The general computability problem is considered to be one of the hardest in Computer Science, directly equivalent to solving all AI-related problems completely.

Maybe if there's some non-linear advance in technology that brings about computers that can solve NP-Complete problems quickly or someone proves P = NP or whatever, we might get there. I'm also holding my breath waiting for that infinite free energy, teleportation, etc. etc.

Quote :

"We're talking about making everyone proficient in one language."

With the minor problem that "everyone" isn't born yet. And then there's your silly idea that in a multi-polar world (economically), there would be any consensus at all on language (or anything else for that matter).

I just really think you underestimate the role of language in human culture. There are theories that people _think differently_ based on what language you speak fluently. The idea that language is just a tool, and the differences represent a mere "compatibility problem" is, well -- again, it's facile coffee shop reasoning, not at all grounded in reality.

5/25/2008 6:05:38 PM

Quote :

"It won't happen in a decade or two, just as we won't have perpetual motion machines, either. Seriously, you just don't know what you're talking about here."

Ha, what? Those two things aren't remotely comparable. Perpetual motion machines violate the laws of physics. Good computer translators do not. The fact that attempt to equate them suggests you're the one lacking knowledge. I do think this is something informed people could disagree on, however. I apologize for being dismissive earlier. No one knows the future. We only guess. For context, note that I expect AIs to match humans by 2030 or so. I'm a devoted follower of Ray Kurzweil. I drink the Kool-Aid proudly.

Quote :

"And then there's your silly idea that in a multi-polar world (economically), there would be any consensus at all on language (or anything else for that matter)."

When did I say this? I agree it'd hard to implement. No question about that. As I said, I think technology will solve the problem for us.

Quote :

"The idea that language is just a tool, and the differences represent a mere "compatibility problem" is, well -- again, it's facile coffee shop reasoning, not at all grounded in reality."

Hardly. Don't put too much stock in the Sapir-Whorf hypothesis. Language has some effect on thought, of course, but not as much as you're suggesting. It is basically a tool, and the compatibility problem affects countless people every day. I want to get past that.

5/25/2008 8:15:46 PM

it would be pretty cool to have 1 language we all had in common. things could get done a lot faster, even in my field


but it would be pretty hard to spread one in particular and give the reason why it should be that specific one.

5/25/2008 8:27:44 PM

Quote :

"it would be pretty cool to have 1 language we all had in common. things could get done a lot faster, even in my field"

Exactly. By romanticizing linguistic differences you forget how problematic being unable to communicate is.

5/25/2008 8:33:42 PM

All we need to do is invent the Babel Fish and we're cool.

5/26/2008 12:36:58 AM

why cant the us just take over the world and make everyone speak american

[Edited on May 26, 2008 at 4:01 AM. Reason : by hypnotizing them so we live in a utopia]

5/26/2008 4:01:02 AM

^^ I wish I had one of those right now

5/26/2008 6:43:42 AM

+

=

5/26/2008 8:46:11 AM

Quote :

"Those two things aren't remotely comparable. Perpetual motion machines violate the laws of physics. Good computer translators do not. The fact that attempt to equate them suggests you're the one lacking knowledge."

LOL! Classic Soap Box. I'm being lectured about my own field by someone who's completely ignorant about it.

OK, I am a retired windbag for this reason. Moving on ...

5/26/2008 1:49:51 PM

That's not an argument, Smoker4. Regardless of what field you work in, strong AI doesn't violate the laws of physics. It's not equivalent to a perpetual motion machine. The main similarity is that both concepts appear in science fiction. This does not make them equally reasonable. If you disagree, please explain the mechanism that makes strong AI impossible. We're talking about duplicating something that already exists. Why should only sacks of flesh be capable of intelligence?

This is a classic mistake that shows superficial understanding of technological possibilities. You have to examine each potential advance based on current science, not fictional representation or social significance. The fact that strong AI will change everything in no way makes it less likely. The laws of physics give considerable room to work with. Scientifically plausible technologies can be as fantastic as unreasonable ones.

[Edited on May 26, 2008 at 2:04 PM. Reason : classic mistake]

5/26/2008 1:58:12 PM

there should be an international language and it should be english.

fuck da rest

5/26/2008 2:39:42 PM

Based on current trends, English won't become the international language. Its prominence is declining. It'll remain important for the conceivable future, but faces competition from Mandarin, Spanish, Arabic, and Hindi.

5/26/2008 3:27:07 PM

^ u are an idiot if you think Mandarin will become the international language of commerce, airlines, etc.

5/26/2008 5:05:06 PM

Quote :

"If you disagree, please explain the mechanism that makes strong AI impossible."

I already did. The problem here is you don't understand the first thing about Computer Science. Implementing strong AI requires either a solution to P = NP, or the implementation of technologies that can reasonably execute algorithms that are "super-polynomial" in nature.

The former is the hardest theoretical problem in Computer Science. It amounts to saying that nearly-intractable problems can be somehow reduced to tractable ones without a loss in fidelity.

The latter -- well, substitute your favorite Science Fiction concept of the future for what paradigm of computing supersedes good ol' Von Neumann. Quantum? Nano? Hybrid-Electric? Take your pick. The point here is that neither you nor Ray Kurzweil can predict when or how it will come, how effective it will be, and whether it'll be commercially applicable.

So, as to my "superficial understanding" -- with all due respect, fuck you. I'm the only one here talking about the actual problems involved and the real science behind it. Just because you read some Kurzweil doesn't mean you understand technology.

5/26/2008 5:36:19 PM

Quote :

"u are an idiot if you think Mandarin will become the international language of commerce, airlines, etc."

Yea, it's completely impossible -- it could happen, maybe, if the script were replaced by something romanized (pinyin, maybe). But most people aren't willing to spend ten years studying characters just so they can find the lavatory ...

5/26/2008 5:37:53 PM

^^

Also as a handy guide for what constitutes a reasonable argument here.

If Smoker4 (that's me) says something like:

"Strong solutions to AI require proving P = NP"

A reasonable counter-argument might be:

"Well, not necessarily, some NP-Complete problems can be solved in a way that is 'good enough' for futuristic applications. For example, in the case of Natural Language, it can be likened to such-and-such problem ..."

An unreasonable counter-argument sounds like this:

"You're stupid, you don't read enough Kurzweil, I read Kurzweil but I can't interpret his ideas well enough to discuss them here," etc. etc.

Get it? Argument 1 is about the actual point I made and the theory involved. Argument 2 is about dick-waving.

[Edited on May 26, 2008 at 5:50 PM. Reason : foo]

5/26/2008 5:49:22 PM

Quote :

"I already did."

No, you did not. Difficult isn't remotely close to impossible. Creating strong AI seems almost insurmountable at this time. A perpetual motion machine blatantly violates one of the core laws of physics. I'm baffled that you can't distinguish between the two.

Quote :

"Just because you read some Kurzweil doesn't mean you understand technology."

Just because you program doesn't mean you understand futurism.

Quote :

"The point here is that neither you nor Ray Kurzweil can predict when or how it will come, how effective it will be, and whether it'll be commercially applicable."

Agreed. We can, however, make educated guesses. That's futurism. The uncertainly doesn't magically make strong AI or computer translation equivalent to perpetual motion machines. You're dead wrong there.

Note also that a functional real-time translator wouldn't require fully fledged strong AI. Far from it. Such translators would merely have to good enough for their intended purposes. There are no perfect translations anyway, only ones more or less suited to the circumstances. Making one for basic commercial transactions would be relatively easy. You might need a different program for academic discussion.

I'll post more when I get the chance.

[Edited on May 26, 2008 at 6:28 PM. Reason : failboat]

5/26/2008 6:13:41 PM

language is nice from the cultural aspect, but from other aspects, a variety of them does get in the way.

We're Americans, so everyone else can learn our language right? That means more than you might think. While many Europeans know 3+ languages, the time learning those would have been directed towards other activities here.

It's actually a pretty key economic advantage.

5/26/2008 8:19:51 PM

Quote :

"We're Americans, so everyone else can learn our language right? That means more than you might think. While many Europeans know 3+ languages, the time learning those would have been directed towards other activities here."

Not necessarily; many are taught in school. While our kids our wasting excessive time as teachers "teach the test" (aritimitic/reading comp), European students also get foreign language studies.

Even if we are talking solely recreational time then i guess we would have less TV watching/video games/ sitting at the bar time if we were more proactive about learning another language.

[Edited on May 26, 2008 at 8:26 PM. Reason : l]

5/26/2008 8:23:53 PM

If you went to an American teacher and told them, "okay, add these 2 languages to the curriculum" what do you think they would say?

Whatever that time is filled with, it's indisputable that it's already allocated in some way, the rest is our lifestyle. Other countries have tests they have to teach for. In Japan, practically someone's entire life is determined by college entrance exams, and I hear South Korea isn't any better. Germany I hear is a bit different, but also have evaluation problems with their universities complicated by that fact that just about everyone goes to university - a serious problem for them is that one can't tell which are the good ones. Argue whichever way you want about the value of this, but those exams not only force teaching of things, but give society itself an indispensable evaluation method. I also despise them, and wish we could at least trash the EOGs, NCLB, etc., but this is a different discussion. My claim is that we have those resources and are using them for something else that creates a benefit to society. Saying we would be better off without the product that that time and effort creates is simply indefensible.

Not to mention that you can't go so far as to decree what activities would be given up to institutionalize a European level foreign language instruction in public schools. If you tried to implement this, kids/families would probably refuse to reduce TV watching time, it would come out of history, math or something like that. Culture never changes because someone tells it to.

5/26/2008 9:15:09 PM

Quote :

"It won't happen in a decade or two, just as we won't have perpetual motion machines, either. Seriously, you just don't know what you're talking about here. People use translators as an _aid_ in understanding other languages but they're laughably bad; as to prototypes being "far, far better," that's like saying they're 100% better while still being a million times too bad for widespread usage (and an order of magnitude worse than that in terms of making language differences "obsolete")."

NLP will get substantially better in the next 2 decades, reaching very close or equivalent to a human's ability.

Quote :

"There are areas of NLP that aren't even well understood theoretically (Context being a huge one), much less in terms of practical implementation. The general computability problem is considered to be one of the hardest in Computer Science, directly equivalent to solving all AI-related problems completely. Maybe if there's some non-linear advance in technology that brings about computers that can solve NP-Complete problems quickly or someone proves P = NP or whatever, we might get there. I'm also holding my breath waiting for that infinite free energy, teleportation, etc. etc. "

There's a lot of good research in quantum computing that would aid in the NP problem set. Quantum computers are pretty much what the definition of what a non-deterministic turing machine is. Solving NP-complete is a different issue than P=NP though.

In any case, you seem to be doubting that in the next 20 years we could have hardware that can process NP problems natively, where I would bet at least at the research-level someone will be using quantum computing to do this.

In any case, with companies like Intel experimenting with 80 core CPUs, the IBMs Cell, Sun's Niagra, and even Intel's Larrabbee GPU, computing is headed to the massively parallel realm, which would make solving NP problems significantly faster than what we can do now. On top of that, new types of solid state disks are going to go a long ways with mitigating the HD bottleneck. Saying we won't have good NLP in 20 years is like people saying the Wright brothers would never fly a plane.

And personally, I think treating the context side of NLP as a strict P=NP problem is narrowminded, considering that humans can't even understand language with the performance expected with such an algorithm. Something heuristic could be created that can even surpass what a human can do.

5/26/2008 9:49:30 PM

^

It's just not enough for the purpose of a reasonable debate to just wave your hands and say "Oh look! Lots of cool shit is happening right now! We'll DEFINITELY have near-human NLP in twenty years!"

Obviously solving an NP-Complete problem per se is different than the P = NP, hence why I gave _two_ alternatives, one is the proven theorem, and the other is "magical advances in technology" aka "Lots of cool shit is happening!" So let's look at the "cool shit" angle from the perspective of NLP.

How does massively parallel computing solve NLP? Is NLP suddenly some kind of "embarrassingly parallel problem?" It seems to be that processing subsequent morphemes in context is as sequential as it gets. I can see how a "computing cloud" type setup can help with the overall job; but that's the kind of "thousand monkeys typing" approach that doesn't exactly equal major technological progress. I do fully expect that companies like Google/IBM/Intel/etc will angle towards the "mother brain" approach and make measurable progress on the problem.

Given that, let's also talk turkey about the terms of this discussion -- I am _NOT_ saying that NLP will not improve in twenty years. Somehow this has been twisted around from the original premise, was that NLP would be good enough soon to make human translators obsolete, etc.

My assertion is this: NLP will be much better in twenty years because of better approximations and more computing power, plus an accumulation of raw statistical input, but it will not even be close to "human-quality." And the notion that it CAN be, represents a total misunderstanding of the complexity of the problem, above and beyond the NP-Complete situation with basic _processing_ of morphemes.

Whatever problems humans have in understanding and interpreting language -- which are obviously far less than ye olde VN machines -- are made up for by an incredible amount of world knowledge and fine-grained contextual interaction. This particular aspect amounts to solving AI in general -- a wonderful academic brick wall that plenty of great minds have haplessly thrown themselves against.

It's OK; we can just reverse engineer the human brain, build machines that model NP-Complete problems in polynomial time, and develop cultural and contextual sensitivity in computers. In twenty years!

5/26/2008 11:40:42 PM

Quote :

"Creating strong AI seems almost insurmountable at this time. A perpetual motion machine blatantly violates one of the core laws of physics. I'm baffled that you can't distinguish between the two."

Well, I think you're splitting hairs. It doesn't "seem" almost insurmountable. It actually _is_ insurmountable. If we want to have a debate over whether something is technically impossible (by analogy) or impossible in all practical terms, then -- well, not willing to go there, my time is worth more.

Quote :

" Just because you program doesn't mean you understand futurism."

And? I don't get the first rat's ass about what Ray Kurzweil thinks. I don't care about futurism. It has absolutely nothing to do with this debate.

The issue here is your intellectual honesty, not my understanding of futurism. Which is to say you won't engage the debate on its own terms -- WHAT, exactly, is required to make NLP possible. That is the ONLY thing that matters in a discussion about the future of NLP technology, would you not agree?

I am trying to bring this discussion to the level of -- what is NLP, what are the problems with it, how do you make NLP work in reality, etc. etc. As soon as we get an inch in, you start bringing out this futurism crap that deals mostly in generalities about human progress. I mean, if we're going to discuss on this level, we may as well bring in the Mayan Calendar which supposedly says the world will end in 2012 and therefore we won't have NLP in twenty years, QED. How's that for futurism?

Sheesh.

5/26/2008 11:54:00 PM

time will figure it all out

5/27/2008 12:00:06 AM

Quote :

"How does massively parallel computing solve NLP? Is NLP suddenly some kind of "embarrassingly parallel problem?""

The context side of it is, I think.

Quote :

"It seems to be that processing subsequent morphemes in context is as sequential as it gets. I can see how a "computing cloud" type setup can help with the overall job; but that's the kind of "thousand monkeys typing" approach that doesn't exactly equal major technological progress. I do fully expect that companies like Google/IBM/Intel/etc will angle towards the "mother brain" approach and make measurable progress on the problem. "

Within very specific contexts (like translating colors or something lame), we can already do translations that work very well. But for 99.99% of translations, a specific context is insufficient.

The thing about context is that humans derive it literally for years and years of forming contextual correlations in the form of life experiences. You'd be an idiot to think you can hard program a computer to develop "life experiences" enough to determine context. Even feeding large amounts of data from an encyclopedia for example only relates to very, very limited contexts. You'd could use textual data from an encyclopedia cross-referenced with something like twitter (but MUCH, MUCH more mundane) indexed using a bayesian algorithm to keywords also derived from that bayesian analysis, across languages. Then, do dumb conversions of the sentence(s) you have to convert, determine the likely lexicon using a bayesian analysis, then using this contextual lexicon, re-do your translation (weighted with markov chains similar to OCR algorithms). The indexing and cross-referencing are highly parallel-izable. The database would not fit on a typical home computer (thus requiring a cloud-computing setup). And the really difficult (REALLY difficult) part is picking the right sources to create the database. The "easy" thing to do would be to create essentially robots or a distributed computing project that creates a database of references emergently, rather than try and build the database yourself (I personally don't think it's possible to hand-pick a viable database to use for this purpose).

Obviously i'm light on details because I haven't tried this myself (yet), but I don't see any major roadblocks to making something as good as a human, and I don't know of anyone trying anything near this approach (a non-dictionary based one).

Quote :

" My assertion is this: NLP will be much better in twenty years because of better approximations and more computing power, plus an accumulation of raw statistical input, but it will not even be close to "human-quality." "

To recap the above, I think it will get better solely and only because we'll find better algorithms. I suspect it's possible even on current hardware to make something that runs very well.

[Edited on May 27, 2008 at 12:20 AM. Reason : ]

5/27/2008 12:17:49 AM

^

Thanks for your insightful reply.

One of my basic problems with context as an issue -- aside from the processing details you discussed -- is that as a theoretical concept, we don't really know what context is. That to me is the big intractable problem that really makes NLP hard on an other-worldly level. It might be said to be world knowledge -- but it's really not, sometimes it's simply a mental representation in someone's mind of what world knowledge is. e.g., one's own assumptions about the world versus facts.

This aspect actually interests me a lot because, as English speakers, we're rather addicted to the idea that words have strictly objective meanings underlying them. We have this vague, "common sense" notion of the "literal sense" of words that is actually a mental formulation and not a well-defined (theoretically) concept. I think if you look at other languages -- Mandarin comes to mind -- these ideas about "literal" or objective meanings lose out to what is well-understood or communicated telegraphically.

So -- without hand-waving too much here -- I kind of believe that, as we approach true "solutions" to NLP, the problem converges into the basic Turing test. i.e. if computers really can 'understand' natural language, then basically they must also be able to pass a Turing test since many of the same basic skills involved in deriving meanings from words are involved. I'm sure this is a fairly obvious idea but it's worth mentioning.

Anyway, I like the idea of readily available translation services (in a cloud somewhere) that make best-guess translations available easily to humans. And I really like the idea of humans feeding their input back into the system to improve the algorithm, as you suggest. That's cool stuff. I can see how an ever-expanding index of human communication -- obvious privacy concerns aside -- can potentially lead to very good translation. Unfortunately it's a bit of a boil-the-ocean scheme but it's an innovative idea ...

[Edited on May 27, 2008 at 2:49 AM. Reason : foo]

[Edited on May 27, 2008 at 2:50 AM. Reason : foo]

5/27/2008 2:48:28 AM

Quote :

"Yea, it's completely impossible -- it could happen, maybe, if the script were replaced by something romanized (pinyin, maybe). But most people aren't willing to spend ten years studying characters just so they can find the lavatory ..."

Exactly. That shit takes way too much time for the average person to spend their time on. I could only see Mandarin as a viable option if it was written in pinyin for international usage.

5/27/2008 8:46:47 AM

^ Could we maybe see more asian languages move to a Korean-esque writing system?

Mandarin really is too complicated for the world to learn, and is barely simple enough for the masses of one country to learn. This type of system probably made a lot more sense 1000 years ago.

5/27/2008 10:21:18 AM

^ I highly doubt it at least with Mandarin. It's been tried again and again throughout Chinese history but never had the support to sustain the effort. Pinyin is probably the latest attempt, it was originally developed to replace Chinese characters by the CCP back in the 1950s, but while it became the romanization scheme for the language, it never displaced the characters themselves. From what I gather about the Japanese language, the written language would be nearly incomprehensible without the use of Chinese characters (pure Hiragana with no Kanji) given the extremely large number of homonyms. Korean suffers this to a lesser extent; while the reduction of hanja (Chinese characters) has improved literacy, it forces a lot more reliance on context to understand more complex writings. That's why hanja is still critical in the legal field, where precise meanings are required.

The irony of course is that if there is any possible "universal" character set or language for Confucian circles, it would be Chinese characters given how heavily embedded they are in the heritage and vocabulary of four of the largest languages (Chinese, Japanese, Korean, and Vietnamese). With the exception of Vietnam, characters are still taught throughout the region and are clearly linked to each language. There are a lot of things that would need to be standardized, character meanings and vocabulary have mutated a bit in each culture, but its common enough that you can get basic points across. I suppose the closest Western equivalent would be Latin back at the height of the Church in Europe.

5/27/2008 10:44:23 AM

Well now, Vietnam may have a slight competitive edge with their system. I don't know the language, but the fact that they use things that look like characters to us probably helps usability.

I think Korean is helped a lot by the fact that the phonetic alphabet is slightly larger than Japanese. It's very similar to Japanese, and if not for that phonetic alphabet difference it should have a similar depth of homonyms. Their sounds also have very very subtle differences, like p, b, t versus pp, bb, tt. I definitely have trouble with the differences, but I think it's a useful tool for distinguishing words that would otherwise be about the same thing.

Japanese really is incomprehensible without the characters, even though it's perfectly possible to write it as such. But Hangul (Korean) implemented spaces. It's practically never valid to use spaces in Chinese derivatives or Japanese. It just looks wrong and even messes up the uniform character size - something I find frustrating about Korean.

5/27/2008 11:26:04 AM

The spaces in Hangul were only recently introduced from the West to improve readability especially as Chinese characters are phased out of newspapers and popular text. The spaces aren't necessary at all; if you look at older Korean documents or in traditional calligraphy, you won't find any spaces. Korean is grammatically very similar to Japanese with an identical word order and using particles to show the relationship between nouns and verbs, and the spaces are usually used after each particle.

5/27/2008 12:26:56 PM

The truly international language is mathematics. Peoples of every continent understand its grammar and content independent of their culture.

5/27/2008 12:51:22 PM

Quote :

"Isn't English pretty much the de facto international language of today. Used to be french until the last couple centuries then Imperial England reached its height followed by the US gaining dominance post-WW2"

good thing we don't consider Africa a real place...because they generally speak french as a second language.

Also some parts of Southeast Asia.

5/27/2008 1:26:01 PM

I think we should avoid what you call dick-waving, Smoker4. I've apologized for being dismissive earlier. I'll try to be more respectful as the debate continues. I'll begin with a question: Why do you write about linear increases in processing power? Moore's Law and company are explicitly exponential. That's a basic principle of Kurzweil-style future studies. Indeed, evidence shows that even the rate of doubling is increasing. This rate of growth is absolutely key. Linear progress would take forever to get us anywhere, but that's not what we have in information technology.

As for natural language processing, Google and others have already made considerable progress by using relatively straightforward statistical algorithms. This suggests that the problem isn't as difficult as you and other skeptics claim. Why would this advancement stop? If nothing else, we'll be able to duplicate a human language ability via brain modeling. This would the Kurzweil approach to AI. Our knowledge of the brain expands daily. Within a few decades, we'll be able to copy it.

For a discussion of NLP and the coming conversational interface, see John Smart's excellent work. I'll quote Smart for emphasis:

Quote :

"If 2020 is our expected transition point, then CI's are a bit farther off than some of our most optimistic technology futurists would today have us believe. Yet they are also sooner than the naysayers, who tell us NLP is riddled with near-insoluable problems, and who don't understand how far we've advanced already with simple statistically based systems. I have heard that Google, for example, has won U.S. NIST's benchmark language translation competition, over IBM's and other's ontological and mixed systems, with a relatively simple statistical NLP approach for at least two years in a row (2005 and 2006)."

http://www.accelerationwatch.com/lui.html

I reiterate that computer translators won't have to be flawless to be functional. Once we have conversational interfaces, basic real-time translation should become possible. Perhaps just enough to get by in foreign country. As with all information technology, these translation programs will improve exponentially.

Note that real-time translation is only one of the many ways by which science can shatter the language barrier. As our understand of language grows, we'll create better ways to teach it. Semi-intelligent digital tutors could make the process easier and more appealing. Even assuming the worst, portable dictionaries activated by voice or thought will be a boon to anyone learning a new tongue.

Another possibility would be duplicating the extraordinary language ability of certain individuals. Daniel Tammet, for example, learned Icelandic in a week. We could perhaps alter our brains to gain similar powers. (Ideally without the negative effects of autism, of course.) Technology opens up nearly endless potential. For every predicted advance that fails, a breakthrough appears. The Singularity is coming. It's only a matter of when. I want it sooner rather than later.

5/27/2008 2:13:04 PM

talking on the phone with Indians for customer support is bad enough. If anyone makes me deal with a robot by voice recognition, I'm never working with them again.

I don't care how spectacular it seems in your lab. All programs that mimic a human saying something are still crap to the rest of us.

5/27/2008 2:27:41 PM

a colocar "em up"

5/27/2008 2:33:04 PM