Kwa maudhui

Machine Translation vs Esperanto

ya Vestitor, 6 Januari 2013

Ujumbe: 22

Lugha: English

Vestitor (Wasifu wa mtumiaji) 6 Januari 2013 2:25:15 asubuhi

I come down on the side of Vilius as he set out his argument in the Ido thread. In addition machine translation - as it stands now and perhaps in the foreseeable future - is limited to very literal translating and is most certainly not useful for natural conversation. What Esperanto does in common with natural languages is allow a free play of spontaneous linguistic interaction; machine translating is nowhere near that and with the obstacle of people having to render one language into another before sending their message to be 'decoded' it never can be.

For machine translation to be useful like real language interplay it would need to be quicker, without the cumbersome decoding process. The actual basis of the computer (especially how we interact with it) has really not changed all that dramatically in several decades. The sci-fi notion of a computer implanted in the brain so that you can instantly interact with it as part of your thought processes is as far away as it ever was, but that is really the sort of thing required for machine translation to be as useful as Esperanto or any spoken language. Fiddling about with some smartphone app and hoping to be both accurate and spontaneous is, at present, a pipe dream.

Tempodivalse (Wasifu wa mtumiaji) 6 Januari 2013 6:04:16 asubuhi

I suppose it's theoretically possible to create a machine with a very large database of idioms and a heruistic tool for evaluating intended meaning based on surrounding words. Still, this would be a gargantuan task, especially for languages such as English which rely heavily on idioms. And even if translating out of English into a more idiom-neutral language became feasible, the challenges of having a machine render proper idioms back into natural-sounding English are even more insurmountable.

Still, the one thing I've learned about computing is, never say never. As I indicated in the other thread, things change rapidly in this space. The heruistics which professional chess programs use to filter through millions of possible moves to find a single good move are astounding - a machine can find a very difficult checkmate in a second or less, where an experienced human would take an hour. So, given enough speed I don't think a decoding process would necessarily be "cumbersome". Computers have to rely on brute-force speed for calculations and decisions of these kinds because they don't have the human intuition to dismiss obviously bad choices.

sudanglo (Wasifu wa mtumiaji) 6 Januari 2013 11:54:18 asubuhi

For machine translation to be useful like real language interplay it would need to be quicker
Imagine the following Vestitor - Google is working on this already, as I understand it.

You take out your smart phone and speak into it (the phone has been trained on your voice). The phone converts speech to text and invokes translation to the target language (centrally, or in the phone). This is transmitted to the smart phone of the person you are talking to where it is converted to speech and the other party then hears what you have said in his/her language.

This is just like having a normal conversation with the exception that what you say is heard by the other party in his/her language.

Is that convenient enough for you?

acdibble (Wasifu wa mtumiaji) 6 Januari 2013 12:10:43 alasiri

sudanglo:
For machine translation to be useful like real language interplay it would need to be quicker
Imagine the following Vestitor - Google is working on this already, as I understand it.

You take out your smart phone and speak into it (the phone has been trained on your voice). The phone converts speech to text and invokes translation to the target language (centrally, or in the phone). This is transmitted to the smart phone of the person you are talking to where it is converted to speech and the other party then hears what you have said in his/her language.

This is just like having a normal conversation with the exception that what you say is heard by the other party in his/her language.

Is that convenient enough for you?
Just two days ago, I was in Italy, sitting at a table with nothing but Italians, but non parlo italiano. Using Google Translate was hit and miss. Sometimes you have to translate between the lines and hope that was the intended meaning of what they were trying to convey.

sudanglo (Wasifu wa mtumiaji) 6 Januari 2013 12:16:56 alasiri

When discussing the threat that machine translation poses to Esperanto, I made the point (in another thread) that the translation doesn't have to be perfect, just good enough.

Then faced with the alternative of weeks/months/years of study of some language (be it Esperanto or a national language) most people will prefer the good enough convenience of machine translation.

In face to face interaction (see my previous post) any uncertainties can be instantly resolved by rephrasing or asking questions, just as they are currently with another party who has imperfect command of the relevant language.

In an important sense Esperanto is also a good enough solution.

Few Esperantists will have the same level of command of Esperanto that they have of their own language, but most Esperantists will have sufficient command to communicate in a situation where both parties do not have a common national language.

Also in principle, Esperanto is not designed to compete with mother tongues - ie. do everything you can do in the national languages. Esperanto may have an adequate grammar, but comes nowhere close to having the lexicon, or the range of emotional expression, or extent of connotation, of a language like English.

sudanglo (Wasifu wa mtumiaji) 6 Januari 2013 12:20:44 alasiri

Just two days ago, I was in Italy, sitting at a table with nothing but Italians, but non parlo italiano. Using Google Translate was hit and miss
Do not judge the threat to Esperanto by what can be achieved with machine translation at the moment.

Judge it by what will be able to be achieved in a few years time.

jchthys (Wasifu wa mtumiaji) 6 Januari 2013 7:56:38 alasiri

From SpecGram, I came across a cartoon on statistics-based machine translation that I found amusic.

Needless to say, corpus-based machine translation has fared somewhat better.

Maverynthia (Wasifu wa mtumiaji) 6 Januari 2013 9:05:30 alasiri

It might not be as far off as you think. I read an article that Microsoft is close to releasing some awesome new translation tool for Windows 8 that supposedly translated the persons words AND voice on the fly.

Of course I'll believe it when I see it but... we actually are not too far off of a LOT of things as far as technology goes.

Oijos (Wasifu wa mtumiaji) 6 Januari 2013 9:32:54 alasiri

Tempodivalse, you can not spell heuristic correctly. You know what Google Translate do in cases like that? It doesn't even try to translate. It leaves the word in source language. And when you try to understand some Internet forum threats through Google Translate, you are completely lost because they typically contain many misspellings and grammar errors.

So: ju more formal text, des useful output in Google Translate. (Sorry, don't know what "ju ... des ..." is in English. And what about this kind of situations? Human usually can understand, computer translation can't.)

And when talking about human conversations through smartphones, consider this: Google Translate is what you sudanglo think it is, only in very few language pairs! So, for example French-English, English-German, Italian-English, Swedish-English, etc. What about Georgian to Icelandic and vice versa? LOL. Just try. Generally pairs which include English are much more workable.

And I have read that using nowadays-like computer technology, the laws of physics are soon going to end the growth of computational power. Intel may be able to make 5 nm processors, but then what? -10nm? It could be possible to make another kind of computer, but that is uncertain.

And when all people will be able to afford a smartphone? (and of course Internet and the needed infrastructure)
NEVER! There will always be poor people. And humanity is soon to confront peak oil for example.

For me as a Finnish speaker, Google Translate is still very useless. And to some uncritical folks it can be even hazardous, if they translate something important and if they believe blindly that the translations can't include big errors.

P.S. not threats, but threads. But you understood. Computer didn't.

hebda999 (Wasifu wa mtumiaji) 6 Januari 2013 11:23:54 alasiri

Without AI (artificial intelligence) machine translation is not going to be as perfect as human translation. The words and phrases have many different meanings and without understanding it is not possible to translate good. So don't be so sure of those technical gadgets - they are just toys and no serious businessman will ever use them during making a contract...
As far as I know AI has not been realized as yet. What we see are just simple translating systems based on probability of word occurrence which sometimes leads to essential errors.

Kurudi juu