Print

Print


Translation systems that rely on vast corpora can be valuable. But at the possible expense of originality and accuracy. 

You try out your new automatic translation system with something like the Gettysburg Address, asking for Spanish.  Bam! there it is in perfect Spanish. You try Russian -- perfect too. You translate the Russian into Chinese, then that to Swahili, then put that into Tamil and then the Tamil back to English. Perfect, word for word ! Not the gibberish you would have gotten with Babble fish long ago. Why? Well there the Gettysburg Address already sits in that vast corpora (in 50 languages or more.) All ok. It moves as a single chunk, no need to analyze or parse. Transition done long ago by humans. 

But with original text?
Say you put in an English sentence to be translated to Russian and that your computer finds the selfsame sentence was once used in a Hemingway novel. (You know nothing of this.) That novel was translated into 27 languages, including Russian. CLICK it gives you your first Russian sentence, all nice and grammatical. The next sentence you write was once in a newspaper article in France, and translated into English and Russian. The English was exactly as you inserted it, and now you get the Russian. Clickety-clack.  And so on. 

The more original anything is the worse the translation.

-----Original Message-----
From: International Auxiliary Languages [mailto:[log in to unmask]] On Behalf Of Matthew Barnett
Sent: Thursday, August 28, 2014 6:02 PM
To: [log in to unmask]
Subject: Re: The Coolness of Esperantoland

On 2014-08-29 01:35, Paul Bartlett wrote:
> On 2014-08-28 7:08 PM, Leo Moser wrote:
>
>> Leo reports:
>> I put the Esperanto  sentence given below [ Esperantujo estas 
>> mojosega loko, plena je mojosegaj samideanoj.] through the Google 
>> translation system and
>> got:
>> "Esperanto is wicked place, full of mojosegaj minded."
>>
>> WTF !  My impression was that the Google system was rather good at 
>> short sentences.
>>
>> I suggest that others of you to try this, if you will. Maybe my 
>> system is broken.
>>
>> You might try other automatic translation systems if you have been 
>> using any.
>>
>> I'd like to hear back. This is worrisome.
>
> Odd, to say the very least! I got the same result. The lernu! vortaro, 
> both on the website and as a standalone online application, lists 
> 'mojosa' as a slang word with the meanings I gave, and -eg- of course 
> is the augmentative suffix which I translated idiomatically as 
> "really" in this context. I haven't used Google Translate much, but in 
> this instance it seems to fail badly.
>
If you're interested in how Google Translate works:

http://googlesystem.blogspot.co.uk/2010/08/how-google-translate-works.html