|
 |
Google language translator | ||
Tags: Google | Translation | ||
Share with:
| ||
 Language Learning Forum : Links & Internet Resources |  | |
| Splog Diglot Senior Member Czech Republic anthonylauder.c Joined 5604 days ago 1062 posts - 3263 votes  Speaks: English*, Czech Studies: Mandarin | Message 25 of 29 22 June 2009 at 5:40pm | IP Logged | ||||||||
That is only partly correct. Google uses a statistical approach (as mentioned in posts above), but that approach is based on only one million words of bilingual text (human translated, as you state) for each language. It also relies on an analysis for each language of one billion words of monolingual text. Therefore, 99.9% of the text used in the statistical analysis is not reliant on "human translation as a reference" at all. The human translated text forms only an initial seed (of 0.1% of the text processed). Over time, that initial seed becomes irrelevant, as the the volume of text processed increases, and with it the accuracy of the statistical translation. Edited by Splog on 22 June 2009 at 5:48pm 1 person has voted this message useful | |||||||||
| RBenham Triglot Groupie IndonesiaRegistered users can see my Skype Name Joined 5578 days ago 60 posts - 62 votes Speaks: English*, German, French Studies: Indonesian | Message 26 of 29 22 June 2009 at 8:02pm | IP Logged | ||||||||
This simply doesn't make sense. It is not possible for such a system to "learn" without any feedback. It could try to analyse the syntax of a language on the basis of a monoglot corpus, but if it does that, it does a terrible job. Anyway, as I said, there is a (very low) limit to how good a purely statistical system will ever get. When statistical translation was starting up, they used a starting corpus of legal translations from the EU, and it seemed rather promising. But guess what? Legal texts tend to be very formulaic and the EU human translators took some trouble (not as much as one might hope) to ensure such things as consistency of terminology. So it looked pretty impressive for EU legal translations, but has proved hopeless in the real world. Translation based on parsing is capable of better, but it is still strictly limited. As an extreme case, consider how you would translate the following from Swedish into English, or any other language except German:
You willl probably observe that the language is not really Swedish, but ungrammatical German. But the "sentence" occurred in a Swedish movie I once saw, in which a rather severe father was ridiculing his son's efforts to conjugate the German verb sein and castigating his tutor for not being strict enough. The English subtitle read something like:
This is one way, albeit not a very elegant one, of handling the problem. There is no way a machine translation system, without a huge amount of genuine knowledge and intelligence that computers just can't develop these days, could ever come up with anything other than gibberish or a copy of the original (which might be fine if the target audience could be assumed to know enough German to work out that the "sentence" is ungrammatical, but the level of German knowledge varies a lot from one country to the next). 1 person has voted this message useful | |||||||||
| Cainntear Pentaglot Senior Member Scotland linguafrankly.blogsp Joined 5946 days ago 4399 posts - 7687 votes Speaks: Lowland Scots, English*, French, Spanish, Scottish Gaelic Studies: Catalan, Italian, German, Irish, Welsh | Message 27 of 29 22 June 2009 at 10:09pm | IP Logged | ||||||||
Isn't it? In the realm of learning systems, how different is feedback from input? Both are systems for determining the correct end-product for given input. Theoretically, as the training set grows, the need for feedback should diminish. The amount of feedback you could feasibly give a system of that size is insignificant against the size of the training set it has as input. 1 person has voted this message useful | |||||||||
| Retire@Thailand Newbie Thailand Joined 5644 days ago 3 posts - 3 votes Speaks: English* Studies: Thai | Message 28 of 29 23 June 2009 at 7:13pm | IP Logged | ||||||||
|
I love the Google translator, in the same way as I'd love a mischievous child. Sometimes I can only shake my head and wonder. On the plus side, you can drop in a URL, and, in a few seconds, get a rough-and-ready translation of an entire web page. It won't be perfect, but it will quickly give you the essential meaning. In a language like Thai -- which I study -- there are no spaces between the words. Google does a half-decent job of identifying the words, and that's a huge time saver. Alas, the biggest downside I've found with Google so far is the tendency to one-to-one translation. One word may be several meanings, each mutually exclusive, and depending on context. Google will select just one. That's provided some laughs for me, but for a beginner who is depending on Google's translation, it could be risky. -- Peter Chiangmai, Thailand 1 person has voted this message useful | |||||||||
| ILuvEire Diglot Newbie United States iluveire.wordpress.cRegistered users can see my Skype Name Joined 5570 days ago 26 posts - 26 votes Speaks: English*, Sign Language Studies: Esperanto, Italian, Arabic (Written), Danish, Japanese | Message 29 of 29 24 June 2009 at 12:09am | IP Logged | ||||||||
|
It depends greatly on the language. For Danish, Vietnamese and Turkish I've noticed that it was awful.
It works pretty good as a dictionary in Japanese and German, but there are better online dictionaries out there (jisho.org and dict.tu-chemnitz.de respectively). 1 person has voted this message useful |
If you wish to post a reply to this topic you must first login. If you are not already registered you must first register
|  | 
Printable version
|
| Copyright 2024 FX Micheloud - All rights reserved No part of this website may be copied by any means without my written authorization. |
