|
 |
Learning completely offline/analog | ||
Share with:
| ||
 Language Learning Forum : Learning Techniques, Methods & Strategies |  | |
| Juаn Senior Member Colombia Joined 5141 days ago 727 posts - 1830 votes  Speaks: Spanish* | Message 17 of 34 23 October 2014 at 5:42pm | IP Logged | |||||||||||
|
The computer is a machine, and as long as you treat it as such it can be a useful tool. It is invaluable to me as dictionary, encyclopedia, and source of bibliographical information.
Those higher activities that require a fuller functioning of our intellectual and spiritual capacities beyond the simple retrieval of information, such as reading good books or listening to music, are better conducted sheltered from its influence. About those dreams of a machine reproducing human language or doing translations beyond the rudimentary stage, they are just silly and a result of a misunderstanding about what humans are - that is, not a sequence of logical steps. 1 person has voted this message useful | ||||||||||||
| robarb Nonaglot Senior Member United States languagenpluson Joined 4855 days ago 361 posts - 921 votes Speaks: Portuguese, English*, German, Italian, Spanish, Dutch, Swedish, Esperanto, French Studies: Mandarin, Danish, Russian, Norwegian, Cantonese, Japanese, Korean, Polish, Greek, Latin, Nepali, Modern Hebrew | Message 18 of 34 23 October 2014 at 9:02pm | IP Logged | |||||||||||
I didn't claim that machine translation would remove the need to learn foreign languages, only that it would remove some of the possible motivations. That would include things like wanting foreign-language webpages to be comprehensible. Once machine translation is consistently acceptable, you wouldn't need to learn the language for that. Or for occasional written communication, where there's no motivation to understand the culture of the people you'll be emailing with a few times each year. People who want that kind of basic practical communication will not need to learn the language. Then there are things like literary experience, gaining enough mastery to understand the subtleties, composing things originally in the target language, gaining rapport with the natives, etc., etc. People will continue to learn languages for those things. Machine translation is like a calculator: it tells you what 0.15*34.76 equals, which is fine for calculating tips, but if you really care about math you'll still want to know what it means and how to do it yourself.
Actually, text machine translation is not that far off. Yes, it's very, very far off from being humanlike, and it would likely be impossible to make it humanlike without a complete redesign of how the machine works and how it approaches language. State-of-the-art techniques, for example, don't involve representing the meaning at all. That could be done in principle with a machine, but not the kind of machine we know how to build in 2014. The error is in thinking the things we know how to build could be combined and extended to produce humanlike intelligence- but it's manifestly true that there is some configuration of matter that will perform linguistic tasks. If we ever get to the point where we can build such a thing, I doubt we will be describing its operation as a "series of logical steps." That said, machine translations are already pretty comprehensible in the languages with the most efforts, as long as you don't use a lot of nonstandard words and confusing constructions, and the languages share some similarities in their conceptual structure. It comes out unidiomatic and with a few errors, but is understandable:
It's a little garbled, but you get the take-home message: The tax on cigars will soon be the same as that on cigarettes. I don't know what "human consumption" is supposed to mean, but as a French learner I also didn't understand the original "droit de consommation." Babelfish-style speech translation is already in prototyping: http://www.theverge.com/2014/5/27/5756166/microsofts-skype-t ranslator-will-translate-voice-calls-on- the- fly It doesn't work very well yet, but why shouldn't it be in 50 years where text translation is now? You just have to improve speech recognition and then adapt your translator to the peculiarities of spoken language. I doubt it'll deal well with slang and emotions, but in 50 years I bet it'll be good enough to convey a message- better than, say, trying to use a language that you speak below C1 level. Let me say again: these techniques do not understand the language, they do not deal well with subtleties, they cannot rephrase things to make them culturally appropriate. But they do produce comprehensible output with few errors under favorable conditions, and they will get better at it. Edited by robarb on 23 October 2014 at 9:10pm 6 persons have voted this message useful | ||||||||||||
| Cavesa Triglot Senior Member Czech Republic Joined 4805 days ago 3277 posts - 6779 votes Speaks: Czech*, FrenchC2, EnglishC1 Studies: Spanish, German, Italian | Message 19 of 34 24 October 2014 at 12:02am | IP Logged | |||||||||||
|
While I believe machine translation French-English may be already quite usable, the Czech-English is nowhere near that and I doubt it is the only language with horrible results. The machine translated message is usually not understandable at all and it is really painful to read. I am not requiring a machine translation to be humanlike, I just think it should make sense and convey at least the main meaning of the message. So, I don't think in fifty years, a czech would say: "let's put the email for our finnish client to the machine translator and send it". They will act just as today: "they surely speak English, don't they? Or should we pay someone to translate?".
C1 level is not the treshold of conveying a message well and surely not something a machine could aspire to get close to. I can agree the "babbel fish" could theoretically replace A2 skills, perhaps B1, surely the touristy inconvenient paper conversation use. But somewhere in the middle of B1 you already begin to include much more to your speech then just straightforward survival sentences and memorised phrases so my bet would go on the B1/B2 speaker winning over a machine any day of the week next year or in a century. Really, I think we are just as likely to have telepatic translation technology first, some of the results of research on brain are quite promising/terrifying. So the unhappy tourist wouldn't use an app to speak into a device and let it translate. Instead, they might wear a helmet that would translate those straightforward things for them, just from the thought. :-D 1 person has voted this message useful | ||||||||||||
| robarb Nonaglot Senior Member United States languagenpluson Joined 4855 days ago 361 posts - 921 votes Speaks: Portuguese, English*, German, Italian, Spanish, Dutch, Swedish, Esperanto, French Studies: Mandarin, Danish, Russian, Norwegian, Cantonese, Japanese, Korean, Polish, Greek, Latin, Nepali, Modern Hebrew | Message 20 of 34 24 October 2014 at 4:09am | IP Logged | |||||||||||
That would be cool, but I'm kind of astonished you think this will be practical before machine translation- mediated conversation. Portable brain recorders such as EEG caps don't have anywhere near the resolution to read off someone's thoughts, not even in principle with perfect knowledge. Today's thought decoding is rather crude and requires either invasive recordings--opening the head--or MRI machines, which are huge, loud, not portable, expensive, and not safe to use around metals. Currently, we can use these to decode one concept at a time, but we don't know how to do it combinatorially. Plus, even if you could decode the message, how would you communicate it to the recipient? You'd probably have to reconstruct it into language! It would also be pretty jarring to speak your messages to people who share a language with you, but think them to people who don't. What if there are some of each? I'm not saying this will never exist, but I'm pretty certain it's farther than translation.
That's fair. Less common language pairs (Maltese-Swahili?) do even worse. The techniques are the same though, they just need more data to work with. I imagine once the techniques improve enough to support more practical applications, people will adapt them to more languages. Czech translation will take longer than French, just like phone voice assistants appeared several years earlier in English than in Czech. Regardless, improved machine translation will happen. I don't know how long it will take for it to surpass the typical non-immigrant L2 learner's comprehension, I only stated it's imaginable that it will be under 50 years. It's also imaginable that it will be more than 100 years. It definitely won't be 1000+ years unless something bad happens to humanity. Regardless, there is little to fear: it won't spell the end of language learning, and it won't even spell the end of professional translation, not unless and until we build machines with humanlike general intelligence, which is not in striking distance of our technology. Edited by robarb on 24 October 2014 at 4:15am 1 person has voted this message useful | ||||||||||||
 Iversen Super Polyglot Moderator Denmark berejst.dk Joined 6499 days ago 9078 posts - 16473 votes Speaks: Danish*, French, English, German, Italian, Spanish, Portuguese, Dutch, Swedish, Esperanto, Romanian, Catalan Studies: Afrikaans, Greek, Norwegian, Russian, Serbian, Icelandic, Latin, Irish, Lowland Scots, Indonesian, Polish, Croatian Personal Language Map | Message 21 of 34 24 October 2014 at 7:19am | IP Logged | |||||||||||
|
I do think that machine translation would take a giant leap forward if the main player in this area (Google) wasn't so stubborn in using statistical methods and only statistical methods. Some combinations are right now terrible because there isn't even a sufficient (or adequate) corpus of parallel texts to analyze - Latin-anything springs to mind - but the majority of the really terrible mistakes are due to the lack of a grammar check after the initial construction of a translation. The frequent loss of negations during translations is a result of this, and some quirks concerning wordorder could also be avoided - like the frequent transformations of "AB and CD" into "ABC and D". Besides there should for each combination be a list of proper names (not least geographical names) and a few other categories with canonical translations, and proper names etc. outside this list shold NOT be changed.
I'm sick and tired of getting the wrong place names and coinages and institution names in my translations - but even with these problems Google translate learns new languages faster than I do, and I can normally spot the worst errors (at least in the direction fra target to base language). Edited by Iversen on 24 October 2014 at 7:50am 1 person has voted this message useful | ||||||||||||
| Cavesa Triglot Senior Member Czech Republic Joined 4805 days ago 3277 posts - 6779 votes Speaks: Czech*, FrenchC2, EnglishC1 Studies: Spanish, German, Italian | Message 22 of 34 24 October 2014 at 2:40pm | IP Logged | |||||||||||
|
EEG has quite nothing to do with it, as far as I've read about the matter. The main part of the research lies in the MRI (which has been improved to work much faster, it was a huge news a few years ago), and monitoring various subconscious responses to external stimuli and internal thoughts and so on (much less annoying and more portable devices monitoring things like all the eye movement). There were even such results as finding the spot on MRI that "shines" where you remember doing activity one and another where you remember doing activity two (which could be used for patients with extremely limited communication options, in a manner like "if you agree, remember playing tennis, if not, remember reading a book"). All those results are unfortunately quite likely to soon interpret emotions, objects of thought and consideration and so on. The first to interpret the results will be marketing experts :-( And I wouldn't be surprised by an antiuthopic future in which a tiny machine finds out you are becoming angry while reading the newspaper and administers an "appropriate" dose of mood modulator drug subcutaneously (all parts that could be abused this way are already being successfully researched for nicer purposes). So, I don't think an interpretation of the message into words somehow would be so much more difficult when the base will already be there. But of course, I consider this to be just as difficult and hardly realistic as machine translation of speech.
Of course it takes longer to develop the Czech mutation, I wasn't complaining. As far as I am concerned, the phone voice assisstants and such inventions making people dumber and dumber (and more dependent on phones even in situations like driving) could have stayed English only or better on paper only. :-) I agree with Iversen the trouble is not only the corpus. Obviously, the google translator ignores the existence of grammar or rather takes it as a set of rules accidentally applied (perhaps following a statistics that each word is most often translated in a certain grammar form). The main trouble is that English is a language where one word represents quite often both a noun and a verb (or adjective) but those are different words in another language. And the grammar a machine has trouble translating is the key to understanding. A word to word translation from English to Czech (just an exemple) is decypherable only by a speaker of both English and Czech with lots of imagination on top of that. Really, I'd say it's easier to learn a language than to become a cryptographer. But you are correct even existence of these imperfect tools is already becoming an excuse for lazy people not to learn languages. A few more excuses applied to education and brain utilization and the humankind will have to be renamed idiotkind in a few generations. Actually, the scientists are already considering 2050 as the year Homo Sapiens Sapiens will have evolved into a new species. Perhaps Homo Non Sapiens? :-D 2 persons have voted this message useful | ||||||||||||
| Ari Heptaglot Senior Member Norway Joined 6378 days ago 2314 posts - 5695 votes Speaks: Swedish*, English, French, Spanish, Portuguese, Mandarin, Cantonese Studies: Czech, Latin, German | Message 23 of 34 24 October 2014 at 3:50pm | IP Logged | |||||||||||
|
I have little doubt that machine translation will continue to improve, albeit slowly, and speech recognition as well. Translation of speech is already something that's being done and it will likely improve over time. People once thought a computer would never be able to play chess, or drive a car, or recognize faces. I strongly suspect I'll see computers writing literature indistinguishable from the literature pruduced by humans in my lifetime (I'm 31). A human is just a computer, after all.
Regarding people, it's been said in all ages that humanity is deteriorating and getting dumber, and people have always been wrong about it before, so I see no reason to believe it now. You don't have to go back far to find a time when most people couldn't even read, let alone learn a foreign language. 3 persons have voted this message useful | ||||||||||||
 emk Diglot Moderator United States Joined 5328 days ago 2615 posts - 8806 votes Speaks: English*, FrenchB2 Studies: Spanish, Ancient Egyptian Personal Language Map | Message 24 of 34 24 October 2014 at 7:02pm | IP Logged | |||||||||||
|
Machine translation currently suffers from a number of problems, including:
1) Existing machine translation systems have no "model" of the world, or of human nature. This makes it hard for them to sanity-check translation results. This is an area of extremely active research, and we're currently seeing huge year-to-year gains in problems like image recognition. This will eventually spill over to machine translation, though it may take a while. But for now, machine translation works at a purely linguistic level with no world knowledge. 2) Machine translation requires a huge corpus of parallel text to get first-rate results. Basically, we know how to make machines find patterns in a couple billion words, but we don't know how to extract as much as possible from a couple million. Again, this is a very active area of research. 2 persons have voted this message useful |
|  | 
Printable version
|
| Copyright 2024 FX Micheloud - All rights reserved No part of this website may be copied by any means without my written authorization. |
