I posted this a few years ago, but I think a lot of it is still relevant:
I used to study linguistics. It is very interesting, but also makes you feel very humble. Human language has far more subtleties than most people credit it with. It is true that if you're a Chomskian, you will tend to see languages as more similar than different (the opposite of most non-academic views). However, even if you do believe in Universal Grammar and all that the idea entails, it has to be said that there are some fundamental difficulties in machine translation. As I'm not in the field, I don't know if they've been solved yet, but I imagine they haven't reconciled the:
1) Differences in language syntactic structure. How do you reconcile a VSO language with an OSV language and still maintain real-time processing? More specifically, if, in, say, language 1 one would form a sentence like "John buys milk" (Subject-Verb-Object, like English) but in lanuguge two you would say "buys milk John", how do you begin to immediately translate, word for word, when the words are not in the same order? Answer is, you don't. The longer and more clausal the sentence gets, the more or a problem this becomes. This assumes the translator is going to have to decide where to pause so it can rearrange the sentence, parse and translate it. This is fine, except that:
2) Natural speech doesn't necessarily follow the same rules as written language. So the speaker many not speak in nice, neat, parseable chunks. So the translation machine has to start making some decisions. For the benefit of the doubt, let's say that we're going to pause nicely after each complete sentence to let the translator do its work. You still have the problem of:
3) Context. A.k.a. the "frame" problem (to some degree, though not exactly). Computers have no context w/ regards to language (they have no actual experiential knowledge of meaning), and thus have no concept of relevance (if you believe in Relevance theory pragmatics). They have no basis upon which to "guess" at word meaning or pull meaning out of inferential utterances -- no basis which to understand sarcasm, humor, hyberbole, or anything your lit professor taught about -- and here's the kicker folks, all of that plays a role in figuring out meaning, which is usually the tiebreaker in any case of:
4) Ambiguity. Wonder why Babelfish only works half the time? Because idiomatic expressions exist. Because words are ambiguous -- one word can have multiple meanings and multiple words can mean the same thing. One word can have different meanings to different people. (BTW, if you want to explode your head, just *begin* to study semantics).
This will probably be another "nobody will ever need more than 16k of RAM" quote, but I think we'll have a hell of a time getting machine translation up to human standards until the machine is thinking for itself. Not that i'm arguing it can't be done, it's just not as straightforward as L&H, or IBM, or the Office of Naval Research would have you believe.
Hmm... Shouldn't this be tried on Caltrain (http://www.caltrain.org/)? I mean, you want a tech-savvy audience -- how about putting it on a commuter train between San Francisco and Silicon Valley... I think ridership would increase and people would be willing to pay extra for the service. The trick would be getting a company to help offset the cost of the installation.
Yes. What I meant is whether or not another effect (say, building a massive root system around the mine) could be triggered by the color change. I don't suppose why not.
Sounds crazy, but if they can do this, what is stopping them from having the plant develop a massive root system *around* the mine? That way, either the roots would eventually trigger the mine, or more likely (and safer), one could actually use the plant to help mine extraction. A third, far fetched possibility, is for the plant (or a sibling plant) to use a chemical process to attack the mine if it had encountered one. That said, I don't know how you would breed "if" logic into a plant:)
People don't like interacting with robots in their current state because, more often than not, they are cumbersome. However, there are exceptions to the rule, such as the oft-mentioned ATM and pay at the pump services.
Now, these robots are getting more and more sophisticated, and at some point people won't mind, and even prefer dealing with them. They will become much more "human-like" to deal with and much more capable of handling situtations that don't fit their simple paradigm; in other words, the expert systems will increase their range of expertise. As you said, it may not happen soon, but I'm fairly certain it will happen.
I agree with the Hibernate suggestion. However, I'm not so sure about the collection type comment. Yes, in a sense the interface is a standard Collections interface (Set, List, etc) but the actual underlying implementation is a Hibernate data type. IIRC, this was done to allow lazy-loading of collections.
Ok, perhaps I shoot a bit higher res than your average casual shooter, and shoot more often. I use a 6mp Canon D60 at large/fine (approx 2.5mb per image), and I use as my primary CF card a 1gb microdrive. That translates into about 400 shots at ISO 100, fewer at higher ISOs. I've easily taken over 150 shots in a day (digital encourages you shoot more pix). So, hmm, three days worth of pix for a two week trip? Imagine shooting RAW (which a lot of people do, I would too if not for the storage and processing issues) at 5mb+ per image....
Oddly enough, this afternoon, Hitachi announced a 4GB microdrive, which is great news, especially given that consumer digicams are getting into the 4/5 mp range and file sizes are increasing rapidly. Keep in mind too that Canon just released an 11mp body, and Kodak is just about to release a 14mp one, and this is probably not going to stop any time soon (though you could make an argument that 14mp is already past what a 35mm lens can resolve, so why bother).
Yes, i'm aware of the digital wallet, but it has its flaws, esp. wrt quality control. As mentioned in a previous post, the Archos unit is tempting and would probably work fine in combo with the microdrive, but ultimately (as a Mac user and longtime Apple hardware admirer) an iPod with digital-wallet capabilities is really what I want, and i'm pretty sure i'm not the only one out there with this desire.
Unfortunately, a lot of people seem to be missing the point (or at least, a major selling point) of a "video iPod". It's not about the video -- though in the future it could be -- it's about the stills.
The digicam market is growing very quickly as are digicam file sizes. Even large CF cards are not sufficiently large to hold enough images for a 2 week vacation -- for an avid photographer, no where near sufficient. Apple could kill two birds with one stone (kickass MP3 player, kickass digicam storage, etc) by releasing something llke this.
Ironically, I've emailed them at least twice asking them to consider updating the iPod design to be able to store CF files directly from the card. If this rumor is true, photogs everywhere will be cheering.
Nobody ever said premium Mac systems were cheap. Quad proc Daystars were quite expensive in the day.
I said "competitive", but I meant in performance terms. Price competitiveness is a whole different issue, though as another poster hinted, price/performance might not be as bad as one would think, esp. if IBM isn't plagued by the same fab issues Motorola has come across with the G4 (and history has shown that they probably won't be).
My guess is that we're gonna see "Four brains are better than one" right around the time the 970 arrives, and that this time it will make the Mac competetive (unlike the Daystar days).
This kind of thing might work, however, if a given company's IT department (say, the large nameless Networking company I work for) decided to install a client on all the company machines that would do this and ran it systematically. It could loan out the CPU cycles of an entire building at a time to institutions. The biggest issue is, in our case, that we're pushing laptops, and any non-personal computer (i.e. server) is doing real work as it is.
Interesting talk in the railroad.net forums
on
Jet Turbine Locomotives
·
· Score: 4, Informative
I plugged "JetTrain bombardier" into google and got postings on the railroad.net forums. Seems quite a few folks there are pretty skeptical. A good read, in all:
I'm a software guy, so admittedly I only partly understand the issues involved in any hardware port. However, since, as I understand it, the Athlon was mimicing x86 instructions and breaking them down into a more Risc-like set, couldn't it do something similar with the PPC instruction set?
I'm sure 64 bit brings its own set of problems, and my money's on the IBM Power4 mutant, but in theory at least, couldn't something from AMD be *tweaked* to emulate PPC (or would that kill any performance advantage and make for a really warm Crusoe:) ?
For Java news, I turn to http://www.cafeaulait.org/ and for XML news, I go to http://www.cafeconleche.org/. Eliotte Rusty Harold does a really good job on these sites. They tend to be no BS, and keep you updated on what's goin' on in both the open source and payware worlds (with definite props to open source). Highly recommended.
A friend working at the "Death Star" (big black Sony bldg in San Jose) showed me a Sony product called a PetaSite that they used for storing broadcast-quality video during editing. Apparently the product is marketed as the PetaFile in Japan..
In case anyone wants to make the case against electric motor performance, they can look at the AC Propulsion T Zero (http://www.acpropulsion.com/). 0-60 in four seconds.
Actually most people consider the Tour de France to be the largest sporting event in the world. Depends on how you classify it. If it means viewers, you're spot on, with the Olympics likely second.
Depends on how you see it. IIRC, in generative phonology you can have CV pairs (viable in *any* language) with a silent V, and thus C(V). More extreme interpretations posited a (C)V pattern. Looking at prosody in this fashion accounted for many of the phonological effects that were difficult if not impossible to explain using traditional phoneme-based rules.
Another issue is how to translate a haiku rule such as 5-7-5 mora to a language that, as you put it, has no concept of it? That doesn't necessarily translate to a pattern 5-7-5 syllables. I'm not saying that haiku is correct, neither is the issue straightforward.
Slashdot Mirror
Try IntelliJ IDEA.
I posted this a few years ago, but I think a lot of it is still relevant:
1 3 for original
I used to study linguistics. It is very interesting, but also makes you feel very humble. Human language has far more subtleties than most people credit it with. It is true that if you're a Chomskian, you will tend to see languages as more similar than different (the opposite of most non-academic views). However, even if you do believe in Universal Grammar and all that the idea entails, it has to be said that there are some fundamental difficulties in machine translation. As I'm not in the field, I don't know if they've been solved yet, but I imagine they haven't reconciled the:
1) Differences in language syntactic structure. How do you reconcile a VSO language with an OSV language and still maintain real-time processing? More specifically, if, in, say, language 1 one would form a sentence like "John buys milk" (Subject-Verb-Object, like English) but in lanuguge two you would say "buys milk John", how do you begin to immediately translate, word for word, when the words are not in the same order? Answer is, you don't. The longer and more clausal the sentence gets, the more or a problem this becomes. This assumes the translator is going to have to decide where to pause so it can rearrange the sentence, parse and translate it. This is fine, except that:
2) Natural speech doesn't necessarily follow the same rules as written language. So the speaker many not speak in nice, neat, parseable chunks. So the translation machine has to start making some decisions. For the benefit of the doubt, let's say that we're going to pause nicely after each complete sentence to let the translator do its work. You still have the problem of:
3) Context. A.k.a. the "frame" problem (to some degree, though not exactly). Computers have no context w/ regards to language (they have no actual experiential knowledge of meaning), and thus have no concept of relevance (if you believe in Relevance theory pragmatics). They have no basis upon which to "guess" at word meaning or pull meaning out of inferential utterances -- no basis which to understand sarcasm, humor, hyberbole, or anything your lit professor taught about -- and here's the kicker folks, all of that plays a role in figuring out meaning, which is usually the tiebreaker in any case of:
4) Ambiguity. Wonder why Babelfish only works half the time? Because idiomatic expressions exist. Because words are ambiguous -- one word can have multiple meanings and multiple words can mean the same thing. One word can have different meanings to different people. (BTW, if you want to explode your head, just *begin* to study semantics).
This will probably be another "nobody will ever need more than 16k of RAM" quote, but I think we'll have a hell of a time getting machine translation up to human standards until the machine is thinking for itself. Not that i'm arguing it can't be done, it's just not as straightforward as L&H, or IBM, or the Office of Naval Research would have you believe.
see http://slashdot.org/article.pl?sid=01/01/19/22222
Hmm... Shouldn't this be tried on Caltrain (http://www.caltrain.org/)? I mean, you want a tech-savvy audience -- how about putting it on a commuter train between San Francisco and Silicon Valley... I think ridership would increase and people would be willing to pay extra for the service. The trick would be getting a company to help offset the cost of the installation.
First of all, it wasn't the first Google result, it was the second :)
l 1/flywheel1.htm
Second of all, somehow an "l" got added to the end which shouldn't have been there.
The real link should be
http://www.upei.ca/~physics/p261/projects/flywhee
Thirdly, lighten up.
How about just using a large scale underground flywheel to store the energy?
l 1/flywheel1.html
http://www.upei.ca/~physics/p261/projects/flywhee
I guess that's why they make a clear variety.. (see article)
Yes. What I meant is whether or not another effect (say, building a massive root system around the mine) could be triggered by the color change. I don't suppose why not.
Sounds crazy, but if they can do this, what is stopping them from having the plant develop a massive root system *around* the mine? That way, either the roots would eventually trigger the mine, or more likely (and safer), one could actually use the plant to help mine extraction. A third, far fetched possibility, is for the plant (or a sibling plant) to use a chemical process to attack the mine if it had encountered one. That said, I don't know how you would breed "if" logic into a plant :)
People don't like interacting with robots in their current state because, more often than not, they are cumbersome. However, there are exceptions to the rule, such as the oft-mentioned ATM and pay at the pump services.
Now, these robots are getting more and more sophisticated, and at some point people won't mind, and even prefer dealing with them. They will become much more "human-like" to deal with and much more capable of handling situtations that don't fit their simple paradigm; in other words, the expert systems will increase their range of expertise. As you said, it may not happen soon, but I'm fairly certain it will happen.
I agree with the Hibernate suggestion. However, I'm not so sure about the collection type comment. Yes, in a sense the interface is a standard Collections interface (Set, List, etc) but the actual underlying implementation is a Hibernate data type. IIRC, this was done to allow lazy-loading of collections.
... or use adapters lik the Archos unit.
Ok, perhaps I shoot a bit higher res than your average casual shooter, and shoot more often. I use a 6mp Canon D60 at large/fine (approx 2.5mb per image), and I use as my primary CF card a 1gb microdrive. That translates into about 400 shots at ISO 100, fewer at higher ISOs. I've easily taken over 150 shots in a day (digital encourages you shoot more pix). So, hmm, three days worth of pix for a two week trip? Imagine shooting RAW (which a lot of people do, I would too if not for the storage and processing issues) at 5mb+ per image....
Oddly enough, this afternoon, Hitachi announced a 4GB microdrive, which is great news, especially given that consumer digicams are getting into the 4/5 mp range and file sizes are increasing rapidly. Keep in mind too that Canon just released an 11mp body, and Kodak is just about to release a 14mp one, and this is probably not going to stop any time soon (though you could make an argument that 14mp is already past what a 35mm lens can resolve, so why bother).
Yes, i'm aware of the digital wallet, but it has its flaws, esp. wrt quality control. As mentioned in a previous post, the Archos unit is tempting and would probably work fine in combo with the microdrive, but ultimately (as a Mac user and longtime Apple hardware admirer) an iPod with digital-wallet capabilities is really what I want, and i'm pretty sure i'm not the only one out there with this desire.
Unfortunately, a lot of people seem to be missing the point (or at least, a major selling point) of a "video iPod". It's not about the video -- though in the future it could be -- it's about the stills.
The digicam market is growing very quickly as are digicam file sizes. Even large CF cards are not sufficiently large to hold enough images for a 2 week vacation -- for an avid photographer, no where near sufficient. Apple could kill two birds with one stone (kickass MP3 player, kickass digicam storage, etc) by releasing something llke this.
Ironically, I've emailed them at least twice asking them to consider updating the iPod design to be able to store CF files directly from the card. If this rumor is true, photogs everywhere will be cheering.
Nobody ever said premium Mac systems were cheap. Quad proc Daystars were quite expensive in the day.
I said "competitive", but I meant in performance terms. Price competitiveness is a whole different issue, though as another poster hinted, price/performance might not be as bad as one would think, esp. if IBM isn't plagued by the same fab issues Motorola has come across with the G4 (and history has shown that they probably won't be).
My guess is that we're gonna see "Four brains are better than one" right around the time the 970 arrives, and that this time it will make the Mac competetive (unlike the Daystar days).
This kind of thing might work, however, if a given company's IT department (say, the large nameless Networking company I work for) decided to install a client on all the company machines that would do this and ran it systematically. It could loan out the CPU cycles of an entire building at a time to institutions. The biggest issue is, in our case, that we're pushing laptops, and any non-personal computer (i.e. server) is doing real work as it is.
I plugged "JetTrain bombardier" into google and got postings on the railroad.net forums. Seems quite a few folks there are pretty skeptical. A good read, in all:
I D=5208
http://www.railroad.net/forums/messages.asp?Topic
I'm a software guy, so admittedly I only partly understand the issues involved in any hardware port. However, since, as I understand it, the Athlon was mimicing x86 instructions and breaking them down into a more Risc-like set, couldn't it do something similar with the PPC instruction set?
:) ?
I'm sure 64 bit brings its own set of problems, and my money's on the IBM Power4 mutant, but in theory at least, couldn't something from AMD be *tweaked* to emulate PPC (or would that kill any performance advantage and make for a really warm Crusoe
For Java news, I turn to http://www.cafeaulait.org/ and for XML news, I go to http://www.cafeconleche.org/. Eliotte Rusty Harold does a really good job on these sites. They tend to be no BS, and keep you updated on what's goin' on in both the open source and payware worlds (with definite props to open source). Highly recommended.
just gotta gloat a little:2 637
http://slashdot.org/comments.pl?sid=26184&cid=284
A friend working at the "Death Star" (big black Sony bldg in San Jose) showed me a Sony product called a PetaSite that they used for storing broadcast-quality video during editing. Apparently the product is marketed as the PetaFile in Japan..
In case anyone wants to make the case against electric motor performance, they can look at the AC Propulsion T Zero (http://www.acpropulsion.com/). 0-60 in four seconds.
Actually most people consider the Tour de France to be the largest sporting event in the world. Depends on how you classify it. If it means viewers, you're spot on, with the Olympics likely second.
I think the poster was referring to the haiku that started with:
...
"Money tree"
Change your threshold and you'll see it.
Depends on how you see it. IIRC, in generative phonology you can have CV pairs (viable in *any* language) with a silent V, and thus C(V). More extreme interpretations posited a (C)V pattern. Looking at prosody in this fashion accounted for many of the phonological effects that were difficult if not impossible to explain using traditional phoneme-based rules.
Another issue is how to translate a haiku rule such as 5-7-5 mora to a language that, as you put it, has no concept of it? That doesn't necessarily translate to a pattern 5-7-5 syllables. I'm not saying that haiku is correct, neither is the issue straightforward.