h a l f b a k e r y"Not baked goods, Professor; baked bads!" -- The Tick
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
Much like the SETI distributed computing project that used home PCs to crunch numbers for a good cause, I propose using the global human resource of Internet users to teach a computer how to translate.
Bilingual users would sign-up at a website, or download an applet that would send them a paragraph
every so often. It would also send what it thought was a good translation of that paragraph, then ask the user to correct it. Once corrected, the user would send it back, and the computer would compare this answer to the thousands of others it had received for the same chunk of text.
As more and more chunks of language received the human touch, the computer would learn all the rules (even though it couldn’t understand why) of translating any particular language into any other.
Over a period of months, a database of languages would be constructed. This could then be linked with voice recognition technology and used to translate speech in real time, for example between a Japanese and an American businessman. It could also be linked to services such as instant messenger and email. How about instantaneous subtitles for any movie, in the language of your choice?
I’m sure that many users around the world would be happy to participate in such a project, in the spirit of bringing humanity closer together, and breaking down the cultural barriers that so often lead to conflict.
A non-profit organization could be set up to administer the computers and write the software. This translation service would then be provided free of charge for anyone wishing to use it.
"Ebonify" standard English.
http://www.joel.net...NICS/translator.asp
[snarfyguy, Oct 17 2004, last modified Oct 21 2004]
Lost in translation
http://www.tashian.com/multibabel/
It talk pretty one day : A day enough speech [bristolz, Oct 17 2004, last modified Oct 21 2004]
Tha Shizzolator
http://asksnoop.com/
Teach a computer program shizzolate perfectly, fo' free, know what I'm sayin'? [thumbwax, Oct 17 2004, last modified Oct 21 2004]
(?) Machine translation
http://www.diplomac...slation/machine.htm
Introduction to pros and cons. [squeak, Oct 17 2004, last modified Oct 21 2004]
Why computers can't translate like humans.
http://www.ttt.org/theory/barker.html
A theory. [squeak, Oct 17 2004, last modified Oct 21 2004]
Statistical language translation techniques
http://www.google.c...anguage+translation
This is an area of great research now. [krelnik, Oct 17 2004]
Update on using people power to translate on a massive scale
http://www.ted.com/..._collaboration.html
Just another update [TIB, Dec 10 2011]
Google Quick Draw
https://quickdraw.withgoogle.com/#
It is pretty hopeless... [neutrinos_shadow, May 14 2017]
Please log in.
If you're not logged in,
you can see what this page
looks like, but you will
not be able to add anything.
Destination URL.
E.g., https://www.coffee.com/
Description (displayed with the short name and URL.)
|
| |
Given the amount of illiteracy I see on the web, the program probably won't be any better than Babelfish, will it? |
|
| |
Why not just feed a computer lots of texts with good translations? It wouldn't be free, and internet users wouldn't get to help, but maybe it would work. |
|
| |
Hey [snarfyguy], you just gave me another idea: a program that translates crappy North American English (like mine) that kids use on the web these days, into real English, fit for the queen of England. |
|
| |
It’d be great for job applications: |
|
| |
“Dude, wazzup? FWIW, I’m like, looking to score some blingbling, and need $. I’ve RTFM, and am your man – LOL ;) |
|
| |
// a program that translates crappy North American English (like mine) that kids use on the web these days, into real English, fit for the queen of England.// |
|
| |
There's already a program that does the reverse (sort of), that ebonification thing (link). |
|
| |
The original idea should work. |
|
| |
Say the computer sent out 1000 examples it wanted corrected. They’d all have to be done in the same incorrect way to screw the system up. |
|
| |
The computer would look for the most similar responses, and choose the ‘average’ as the correct translation. |
|
| |
The computer could also specify a response as it would be spoken, or as it would be written. |
|
| |
This assumes languages are a sort of concrete and absolute set of words, which they arent, they are very mutable. This translator would be permanently out of date. And I also suspect minority dialects of a language would be "shouted down" by their more numerous counterparts. |
|
| |
Read up first. There is an unbelievable amount of information concerning machine translation on the web. I've added a link which is a pretty good introduction to it's strengths and weaknesses. If [tobyp] turns up in this string, listen to the man. He knows what he's talking about. |
|
| |
Bad idea. Just two bilingual people never agree on a translation. I mean NEVER. |
|
| |
Hmm. Your link seems to offer more reasons why M.T. is flawed than why it might work [squeak]. I don't want to expand the annotations to talk about M.T. as a general concept, but I personally have done quite a lot of work on automated textual analysis and in my experience trying to fit a rule-based system round any language (with all their historical ambiguities and contradictions) can only be at best a partial solution.
With reference to [TIB]'s idea, I think its a little too simplistic an approach. |
|
| |
I'm with [Kreuner]. The English language is so abstract at times, it takes a human brain to understand it. I remember my favorite example from Modern Linguistics (a class I took in college), the sentence:
"Time flies like an arrow."
How does a computer parse this? Unless it knows the context, it could interpret this several different ways. *Time flies* could be flies (like fruit flies) of the time variety (is time the noun or is flies the noun?), *like* being approving of, or similarly, etc... |
|
| |
Computers, darnit, although useful, just aren't smart when it comes to our ridiculously complex language. What we need is a new language that computers completely understand. That gives me an idea; I should learn to speak C++ orally. |
|
| |
I think it's fair to say that a distributed computing approach stands to do a better job than something like Babelfish. I think there's some merit to the idea on that basis. |
|
| |
Those are all points that can be figured out later. They don't detract from the merit of the idea, in my opinion. Regardless of the point of the idea, it seems logical to me that a distributed computing method stands a good chance of coming closer to perfect translation than a stand-alone system does. |
|
| |
Kreuner, you just like to hear yourself argue. |
|
| |
[custardlove] and [kreuner]. I am aware that machine translation works pretty well in some situations (meteo in Canada being the usual example given) and that it also has many, well documented problems. I put the link on to show [TIB] some of the ups and downs as it doesn't sound like he's done much (if any) reading up. The idea, in fact, sounds almost identical to translation memory systems which are thoroughly baked (though I'm not saying they work or not, just that they exist).Google gives 444,000 hits. So [M-F-D]. |
|
| |
See second link for more pertinent text concerning computers and their problems with translation. |
|
| |
I think the vast majority of human languages contain shortcuts because humans are lazy. These shortcuts bring in ambiguity of meaning that won't be understandable to, or translatable by, any one/thing that doesn't understand laziness. Once you get your computer to understand laziness, it won't translate anything for you anyway. |
|
| |
Since it’s so difficult to accomplish any type of machine translation that would ever approach the level of a proficient human translator, I guess we should just give up trying.
I was really looking forward to speaking to the in-laws in their own language ;)
Check out cyc.com. That looks similar to my idea. |
|
| |
Where can hamsters get free pearls online? |
|
| |
There was an interesting article in the New York Times recently about using statistical methods to train computers to translate. Basically you feed a learning program a bunch of matching source texts in the same pair of languages, and it learns how to translate. Its a computerized version of the Rosetta Stone. They claim it actually works. (I'd link to the article, but the Times has already punted it to their archive so it can't be read online for free). |
|
| |
Anyway, you could use something like this idea to supply the matching texts for input to the statistical program. So I believe that this idea could be useful, if these statistical translation methods actually do work. |
|
| |
See link for many resources on the web about statistical translation. It is an area of much research, and some researchers have already built quite good translators using this technique. |
|
| |
*Server farm, maintenance, programming, bandwidth,
advertising, quality assurance, management, and
documentation not included. |
|
| |
Yeah [Voice], that's true, but remember there's many
possible revenue streams to power the gizmo, and
the teaching part is at least free. |
|
| |
..but if you think about it hard enough, the only thing
truly free in life is gravity. |
|
| |
That was a great TED talk I thought! |
|
| |
Let a million monkeys pound on a million typewriters and you
will get the internet. |
|
| |
Unfortunately, when you get the masses to help with something technical like this, you end up with garbage.
Recently, Google did it's "Quick Draw" experiment, to get a computer (AI, whatever...) to recognise things from quick sketches. Because most people are useless, the system is mostly unable to recognise much of anything. |
|
| |