What is Digg?66 Comments
- Norweed, on 10/12/2007, -1/+28 Actually Natural Language processing is TOTALLY different from a speech recognition program. It's like the difference between a calculator and a program that does proofs.
Solving the Natural Language problems is one of the major topics in AI and is certainly one of the most important steps in making a machine that understands natural language and it's structure. Though I seriously doubt that that this machine can spit out machine understandable code from natural language input, if it can that's an amazingly large leap forward in AI. - berwiki, on 10/12/2007, -10/+19if a machine could pass the Turing Test, that would be headline news, Languages have structure and rules to follow, its no surprise a computer program is able to replicate these rules.
http://en.wikipedia.org/wiki/Turing_test
This is just another Dragon Naturally Speaking program. - schestowitz, on 10/12/2007, -6/+15In Soviet Russia, text parses Chomsky.
[mod me down now, but I couldn't resist] - vann, on 10/12/2007, -2/+9I also have a background in Computational Linguistics and to say you've "solved the problem" of syntax parsing is absurd. I'll believe it when I see their syntax trees.
- enicholas, on 10/12/2007, -2/+8 @Yorn: No computer has ever passed the Turing test, period.
As for the Chinese Room argument, I'm afraid that you've missed the fundamental point of the argument. The basic argument is that you can (in theory) give a person a book with rules to follow -- a program, if you will. That person can then, by following these (fantastically complex) rules take an input sequence of Chinese characters and produce an output sequence of Chinese characters. Because the rules are so good (they necessarily emulate the intelligence and learning capacity of a human being), the person can carry out a conversation in Chinese as if he spoke it. But, of course, he doesn't -- he's just following rules.
The argument is that this is all a computer ever does. Even if you had a program which was smart enough to understand and speak Chinese, the argument states, the computer doesn't actually "understand" it -- it's just following rules and cannot be anything more than a dumb machine.
Of course, this argument is complete, utter, unmitigated *****, because all it says is that the computer's *processor* (the "person" in the thought experiment) wouldn't understand Chinese, which I accept. It fails to consider that the system composed of processor (person) + program (rules) *does* in fact understand Chinese. - rkuchiki, on 10/12/2007, -0/+6Demo or it didn't happen..?
- junk, on 10/12/2007, -1/+6For your NanoPenis I presume?
- snuffulupagus, on 10/12/2007, -7/+12Buried as inaccurate, I'm a linguist with a background in CS and I know this is highly unlikely. Also, it's not as if it's parsing language and talking to you, but parsing language for a search engine which doesn't take much work.
- barthosch, on 10/12/2007, -1/+5I have a Counter-Strike background too.
- UltraNurd, on 10/12/2007, -0/+4Don't forget to reverse the polarity!
- JonForTheWin, on 10/12/2007, -0/+3Aww if this makes it to getting used for writing code it'll take all the fun out of it.
- PirateFSM, on 10/12/2007, -5/+8...But will it Blend?
Damn Viral marketing... - bmartin, on 10/12/2007, -3/+6Agreed, Norweed. I did my MS in Computation Linguistics, focusing on question answering systems. One of the biggest hindrances to computer productivity is that computers do exactly what we tell them to, rather than what we mean. This will create amazing advances in robots and search engines and the like.
- thePoopSmith, on 10/12/2007, -0/+3Vaporware almost as old as the computer itself
- benwellington, on 10/12/2007, -1/+4I am getting my PhD in this stuff... this article is meaningless. You can't "solve" the problem. You can improve results by adding new ideas. But it is not the kind of think you solve. It's like saying that you solved the weather prediction problem. You would have to be 100% perfect to say it is solved. So, yeah, this is lame.
- MadOtaku, on 10/12/2007, -2/+5@Gottschalk
Most humans can't instantly calculate the value of hundreds of cells in a spreadsheet either. Just sayin'. - VSKBadCRC, on 10/12/2007, -1/+4Oh this is an easy one, just ask the computer to install a recursive algorithm, problem solved. That always made the holograms on Star Trek work.
- kakapu4u, on 10/12/2007, -0/+3Way to use that 'nano' buzzword. Here, have some funding.
70's version: Syntaxonics
80's version: MicroSyntax
90's version: eSyntax
Coming soon: QuantumSyntax - UltraNurd, on 10/12/2007, -0/+3Heh... I've seen them present their work. It's really, really good, but it is not solving the natural language problem.
- strictnein, on 10/12/2007, -2/+5Another CS major who did a little bit of study of languages (like any CS major really).
I know what we're all thinking right now: What would Chomsky think? - zbeast, on 10/12/2007, -0/+2I would just be happy if this "system" could make for better voice recognition.
I also want it to not take tons of expensive hardware to get this to work.
The software should be simple enough to be used in a phone answering machine.
So it can filter off unwanted callers, advertisers and bill collectors.
Please stay on the line I have a very important call for. *click*. - Archangeleon, on 10/12/2007, -0/+2I too have a background in computational linguistics and if they actually solved this problem, then I'm shocked that I didn't hear anything about this.
The part that really gets me is the way they make whatever algorithm they came up with sound like Doc Brown in Back to the Future talking about the flux capacitor. "It's the thing that makes NLP possible!!!" - foobario, on 10/12/2007, -1/+3Oh come on... I know *people* who can't parse natural human language.
- MikeTheKitty, on 10/12/2007, -0/+2I worked in NLP for several years. Both statistical and straight linguistics. IMHO, they have probably come as close as any of the other active research groups out there who can parse various corpora to 95+% precision.
The only real news here is that in a year we might be able to type an actual question into Google, and have more than simple keyword searches.
That being said, I for one welcome our Nano Agent overlords. [Ed: comment added in case they're WATCHING US RIGHT NOW.] - Feodoric, on 10/12/2007, -0/+2I think one of the important words in the summary is the word "normative". As in "normative human language". This means that they claim their software can parse human languages with a certain formal structure. So there is probably a whole class of utterances this software can't deal with.
But they don't claim (at least from what I read) to have software that can parse ALL human language. Just human language with specific structure.
So I agree with you, to a point. The question is, what is the normative structure that the algorithm parses, and how well does it work on that structure? 100 percent? 95 percent? - creamyhorror, on 10/12/2007, -0/+1So where's the proof, the prototype? If it's all proprietary, shouldn't there at least be reviews or recorded demonstrations of the capabilities of the software? This isn't quite newsworthy, folks.
- Yorn, on 10/12/2007, -0/+1@enicholas:
Oh really? What about the tests done where a human sits on one end and enters either 1 or 0 and on the other end is either a computer or human who can only respond with a 1 or a 0?
The complexity of a Turing test is not simply defined as a chatbot, however that is the capacity that has been tested recently and is arguably more effective. After all, if you were talking to a Chinese chatbot and a Chinese human, would you as an English-speaker only be able to differentiate between who was human and who was computer?
Older much more simpler tests actually required the judge to determine which of two responding to a signalling system were human and which were computer. In such a test, computers can easily be mistaken for humans. - JK124, on 10/12/2007, -1/+2Not enough information.
I would have to agree with Nells that a computer must be able to feel or be given the definition of feelings, which is a very hard if not impossible task. There isn't any information or insights being provided on this new technology, therefore I would have to question at what stage of their development their technology is. What kind of testing have they done? There are no papers available yet to confirm this either.. - Feodoric, on 10/12/2007, -0/+1Enicholas-
What happens when the processor (the person) has worked with the rules for so long, that he no longer needs to refer to the book to decide what response to make? The output is exactly the same as before.
In this case, the processor has the same "understanding" of chinese as the system of processor+rules had. But this seems different from the "understanding" of chinese that an actual speaker of chinese would have. - toxonix, on 10/12/2007, -0/+1They probably have something they mistakenly think they can patent, and are looking for money. Most areas of research in AI, including NLP, do not present concrete solveable problems. You might have a better approach at something that gives you more reliable or realistic AI, but even a Turing test is subjective. Does anyone have a logical or mathematical proof of self awareness?
- Urusai, on 10/12/2007, -0/+1A problem with parsing natural language is that there are intractable external elements coming from culture. Humans create new words based on things that a non-intelligent (as in an actual AI) computer would not be able to handle. Natural languages are ambiguous, both syntactically and semantically; the meaning of a sentence changes based not merely on the language context, but the physical context as well.
There is no "solution" to the problem, it's a matter of having an AI that can "see" the context, which means access to massive amounts of data and better input devices than we have today. That's the obstacle I see to AI right now, is that people want it to work on toy domains with bounded state spaces. Real intelligence is unbounded, and needs a lot of input. We don't even have computers to where they can analyze a photograph too well even now, although progress is being made. - nepawoods, on 10/12/2007, -0/+1@vann
"I also have a background in Computational Linguistics and to say you've "solved the problem" of syntax parsing is absurd."
So you believe the grey matter in our heads does something which is inherently undescribable, and follows no physical laws? Because if it can conceivably be described, and follows laws, it can in principle be emulated through computation. - nepawoods, on 10/12/2007, -0/+1parsing language and character recognition are completely different things
- vertinox, on 10/12/2007, -0/+1@"One researcher has argued that someone could sit in a boxed room and take in chinese symbols as input, put them in a stack, and spit out chinese symbols and eventually reach a level where they are so good at associating symbols that they could pass for being human"
The problem with the "Chinese Room" theory is that by definition it just means that person knows Chinese at that point if he can associate them with others. He may not know the actual definition of words, but he knows the relation of one word to another so he might as well know Chinese.
Example... I may not know what PCI, VESA, or AGP actual definition by memory, but if you showed me a mother board I could tell you that the brown slot is an AGP and the white is a PCI and that usually the graphics card will go in the AGP (barring PCI express cards... but you get the point).
The reason why the Chinese room explanation of AI fails is that is how humans minds actually work with word relations rather than memorizing millions of definitions we just relate to context in the sentences and our minds fill in the gaps. - syspro, on 10/12/2007, -0/+1It may not be as far-fetched as you think,
I remember hearing of a method to filter meaningful requests from complete sentences, that the speaker called the 'falafel stall' technique,
The idea is that you can teach someone who does not understand even one word of the local language to sell falafel. simply by teaching them to listen for the words they need to know about, and ask 'canned' questions regarding ambiguities,
And finally to repeat the order they understood for confirmation.
I have heard of SMS query applications for bus time-tables that return excellent results, that were supposedly based on this technique.
Maybe they have found a method to easily build the vocabulary, logic and inference rules.
Maybe the 'falafel stall' idea is from the same Israeli company,
Maybe they are just looking for venture capital,
Who knows.. - philo23, on 10/12/2007, -0/+1well there goes the idea of random didgets and letters to stop robots from logging into websites if this goes online. (its called something like capatures, right?)
- vertinox, on 10/12/2007, -0/+1@" i too have computational linguistics"
Gee... It seems like half of Digg has a degree in Computational Linguistics.
Or whatever the related subject major of the article maybe. ;) - adragons, on 10/12/2007, -0/+1I have no education in this field what so ever and even I highly doubt it. Does it understand slang? Does it take into account that natural languages keep changing? If not, then it has not by any of my standards, parsed the human language. It sounds like they just created a new algorithm that works well enough to understand basic sentences.
- nepawoods, on 10/12/2007, -0/+1I beg to differ.
Do you believe the human brain is doing something other than computation? And if so, is it doing something that can't be emulated by computation? Because so far, I don't think anyone has demonstrated any physical process (and the brain parsing language is a physical process) that can't be emulated by computation (true randomness aside, but that's irrelevant).
It's a difficult problem, but there is already a machine that solves it: the brain. - corsar, on 10/12/2007, -0/+0Imagine this...computers now understand the natural human language - the first to use this new feature will be the military...and some day, someone in the nuclear missiles control room will say, after seeing the news : "God, how I would like to kill all ......." (replace .... with an enemy country).
While I believe in AI (I have a degree in cybernetics), this won't happen soon. - evilTak, on 10/12/2007, -1/+1So, is it highly unlikely, or does it not take much work? ;-)
J/K - As a fellow CS graduate, I also believe it highly unlikely. - JrGhoull, on 10/12/2007, -0/+0i think there are a few key pieces of software which will come out soon and allow for some very big computer changes. along with computers being able to really understand what you are saying, theres also software and hardware being made to allow a computer to understand anyone right away, without any pre setting up, as well as (though this last part is a few decades away at least) decent AIs that are being developed to allow things such as search engines to be able to find specifically what u want, (as well as its many ohter various applications)
- EthylAdded, on 10/12/2007, -0/+0Aye. It's very clear that whether or not Linguistic Agents has good technology and/or good engineers, they do have some very good marketing people.
- thomashauk, on 10/12/2007, -0/+0So I have to type in twice as much to find basic stuff. eg this is how things would go.
Me: Foo
Google: Huh?
Me: Sigh
Me: Tell me about foo
Google: Yeah cause I like questions that take me an age
Me:...
Google: BE MORE SPECIFIC!
Me: I dunno I just wanted to know more about it... And there's no need to shout.
Google: Sorry
Google: Just I get millions of these every day and it make me all RAUGH...
...
And so on...
'Tis the future guys! - freff, on 10/12/2007, -1/+1But, how could that possibly be? It says right in the article that "Linguistic Agents does not appear to be a hoax".
When has the internet ever led us astray before?
Looking at Linguistic Agents' site, my first impression is that I'm not impressed. Just saying. - grow, on 10/12/2007, -0/+0I think they installed Ruby and didn't realize it.
- TetchyTony, on 10/12/2007, -0/+0What would they deduce, for instance, from the turn of phrase so often seen here: "...I could care less"?
Is it a US neologism? A hip misperception? A rhetorical inversion? For myself, 'I couldN'T care less'.
Or in those famous words: "Frankly my dear, I could give a damn." - Figs, on 10/12/2007, -0/+0Ok, here's the solution to the Chinese room paradox (the way I solved it in Junior year HS):
It doesn't matter whether the man knows Chinese or not, because he is merely a part of the entire system of the room. You need to take the rule book and the I/O method into account. The "box" you describe knows Chinese. The man is just a piece in the box.
A person has neurons which follow some set of rules to produce the output. Does a neuron know Chinese? No. But can a man, using his neurons, produce Chinese? Yes.
The problem is with the way people look at the analogy in my opinion is that they're breaking the black-box rule. You shouldn't know how the box works; just that it does. If I presented you with a box that could produce output in Chinese, then the box can produce Chinese! This same concept can be used to say why something can be considered intelligent, despite being based upon a set of rules.
Look up "brain in a jar" for an interesting analogy in reverse. - Nells, on 10/12/2007, -2/+1Actually, I think it is possible that a program understands language, if it implements the brain's algorithm (which isn't necessarily complex) and is given the same kind of input the human brain is given. I guess the first step would be to put those artificial brains in a kind of simulation of the world, so one wouldn't have to design devices to reproduce the behavior of the nose, eyes, mouth, skin, and ears. And all it could endanger would be that simulation...
- Aninhumer, on 10/12/2007, -3/+2While we're at it isn't the brain just a bunch of, highly complex, chemical reactions?
-
Show 51 - 66 of 66 discussions
Check out the new & improved 
© Digg Inc. 2009
— Content created and posted by Digg users is