Thursday, February 23, 2006

Altavista Babelfish Lost in Translation.

Now that I had fun with Google Translator, I decided to try the Altavista Babelfish translator and see if you can obtain any better results.

I slightly modified my Java program (The Translator Buster) to hit the Babelfish site instead of the Google site. I ran the first test with the simple phrase “I am a writer”. I had it converted into Italian and than back into English; this methodology is described in great details in my previous post about Google Translator.

Sit down! Ready? Here is the news! The result is EXACTLY the same obtained with Google!!! Errors and everything! The phrase gets translated first into “Sono un produttore” and then into “They are a producer”, exactly like it happens using Google Translator.

Is that a coincidence? Well, if both translations were correct, then I would conclude that both translators work perfectly. The fact that both translators completely screwed up the translation in the EXACT same way seems very odd. Two translators, completly lost in translation, ending up in the same place. What are the odds?

Let’s compare more complicated results.

The following is a quote from Johann Wolfgang von Goethe:
All truly wise thoughts have been thought already thousands of times; but to make them truly ours, we must think them over again honestly, till they take root in our personal experience.

I report here the result of translating this back and forth from English to Italian 10 times with both Babelfish and Google translator. Here we go:

Babelfish results:
The thoughts all that in order to align must examine it have been thoughts of the migliaia already of the periods; but for so that it renders in order to align ours, than that that what those that that what must honestly think them anchorage of the surplus, the finchè must they that the puttinges that it uproots in our personal experience.
Google results:
The thoughts all that in order to align must examine it have been thoughts of the migliaia already of the periods; but for so that it renders in order to align ours, than that that what those that that what must honestly think them anchorage of the surplus, the finchè must they that the puttinges that it uproots in our personal experience.

You got it!! The results are the exact same! Coincidence? I don’t think so!

Running enough tests I observed cases where I got slightly different results. Here is an example. The original quote is from Alvin Toffler:

The illiterate of the 21st century will not be those who cannot read and write, but those who cannot learn, unlearn, and relearn.

And the results obtained with the two translators are:

Babelfish results:
Those the illetterati ones of ventunesimo the century will not be those who cannot read and write, but those who cannot learn, unlearn and relearn.
Google results:
Illetterato one of ventunesimo the century will not be those who cannot read and write, but those who cannot learn, the unlearn and not relearn.


My fist wild guess after running these tests is that both Google and Babelfish run the same translation back-end engine. I coudn't find any mention of this fact anywhere, but it seems pretty obvious to me.

Given that Altavista Babelfish has alot more supported languages, it is possible that Google bought the translator from Altavista.

BTW, I also tried with some other languages (like French and German) with very similar results.

1 comment:

Tony Ruscoe said...

Google and Babelfish both use Systran for their MT engine, as do most websites offering free machine translations.

Another alternative is FreeTranslation.com,which uses SDL International's Enterprise Translation Server.

Also, it's worth noting that translating sentences to and from English doesn't prove anything about the accuracy of a machine translation system.