Tuesday, May 24, 2005

Google Translator: The Universal Language

Google Translator: The Universal Language:

"today, English is much more universal. 30 countries have it as an official language, and in many other countries it is taught in school and understood fairly well. The internet can be suspected to further increase the adoption of English.

Still, many people can’t speak English. The collected, shared knowledge that makes up the web is therefore only partly accessible to them. The reverse, of course, is true as well. When you surf the web, you will sometimes come across languages and characters you don’t understand – like Chinese, Arabic, Korean, French, German, Italian, Spanish, or Japanese. Would you be able to fluently read these languages, those sites wouldn’t be a dead end for you. You would discover a wealth of knowledge, and more importantly, opinions. If you’re an US citizen, how many Arabic, German or French sources do you read to get a good understanding of how the world sees the US? How many blogs do you read in foreign languages? Probably not many, unless you’re fluent in those languages.

At the recent web cast of the Google Factory Tour, researcher Franz Och presented the current state of the Google Machine Translation Systems. He compared translations of the current Google translator, and the status quo of the Google Research Lab’s activities. The results were highly impressive. A sentence in Arabic which is now being translated to a nonsensical “Alpine white new presence tape registered for coffee confirms Laden” is now in the Research Labs being translated to “The White House Confirmed the Existence of a New Bin Laden Tape.”

How do they do that? It’s certainly complex to program such a system, but the underlying principle is easy – so easy in fact that the researchers working on this enabled the system to translate from Chinese to English without any researcher being able to speak Chinese. To the translation system, any language is treated the same, and there is no manually created rule-set of grammar, metaphors and such. Instead, the system is learning from existing human translations. Google relies on a large corpus of texts which are available in multiple languages.

This is the Rosetta Stone approach of translation. Let’s take a simple example: if a book is titled “Thus Spoke Zarathustra” in English, and the German title is “Also sprach Zarathustra”, the system can begin to understand that “thus spoke” can be translated with “also sprach”. (This approach would even work for metaphors – surely, Google researchers will take the longest available phrase which has high statistical matches across different works.) All it needs is someone to feed the system the two books and to teach it the two are translations from language A to language B, and the translator can create what Franz Och called a “language model.” I suspect it’s crucial that the body of text is immensely large, or else the system in its task of translating would stumble upon too many unlearned phrases. Google used the United Nations Documents to train their machine, and all in fed 200 billion words. This is brute force AI, if you want – it works on statistical learning theory only and has not much real “understanding” of anything but patterns.

One can suspect Google will release their new translation system soon (possibly, this or next year). The question is: what will they do with it – where will they integrate it – and what side-effects would it have? If via Google we get our universal language, would that resolve many global problems by fostering cross-cultural understanding, like Zamenhof was hoping for? Here is a speculative list of translation applications Google might implement; the key is auto-translation.
The Google Translation Service

This one is the most obvious: Google will still allow you to translate any document from their search results by the click of a link. What might be less obvious is that they might enable you to search foreign languages in your native language. All translating would be done behind-the-scenes, so that when you search for “thus spoke”, you might as well get results which only contain “also sprach.”
The Google Browser

If Google ever releases their own browser, they could seamlessly integrate translations of foreign languages; the user would just have to define what languages she reads fluently. It would be the Google Auto-translator (and surely it would be attacked using similar arguments than those brought forth against Google’s auto-linking.) And if it’s not a Google Browser, it would be a Google Toolbar feature.

Now imagine this: you specified you speak English only. What does the Google Browser do when it encounters a Japanese page? It will show you an English version of it. You wouldn’t even notice it’s Japanese, except for text contained within graphics or Flash, and a little icon Google might show that indicates Auto-translation has been triggered. After a while, you might even forget about the Auto-translation. To you, the web would just be all-English. Your surfing behavior could drastically change because you’re now reading many Japanese sources, as well as the ones in all other languages."

No comments: