Pixelstorm-Award 2011:
__ main
   __ Einreichungen
   __ Call for entries
   __ Jury

__ Main
__ About
__ Best of Pixelstorm-Award
__ Community (Who is Who)
__ Guestbook

__ 2009
__ 2008
__ 2007
__ 2005
__ 2004
__ 2003

__ Kontakt
__ Credits
__ Presse

__ dock18
__ Pixelstorm
__ Claudia
__ Equipo
__ FAQ42
__ Hauser, Schwarz
__ Ted Davis
__ station21




by Sandra Phillips, Computer Linguist, Vancouver 2009

  1. What's translation to a Linguist? Let me start with words since we love to dissect them.  They are not only something written but alive. In the sense that they are the verbal expression of our thoughts, dreams, ideas, beliefs, myths, and stories and they create our cultural landscapes. We only use or invent words that we need. So my grandparents didn't need to know what "Skype" is or that you could "skype". The name Skype by the way comes from Sky-Peer-to-Peer, which morphed into Skyper, then Skype and by now has even made it into the Cambridge dictionary. Speaking more than one language gives me access to different cultural landscapes. So translation for me is the means of transport between them, not only at the word level but at a cultural one.
  2. Now I'm not just a linguist, I'm a Computer Linguist. Computer Linguist? That sounds a bit geeky! We'll we might be, but mostly we're nice peple. We have looked at all the books, articles, webpages, and official documents which have to be translated and kind of felt sorry for the translators. How could we help out? We decided to speed the whole process up and build the bullet trains of translation -Machine Translation Systems.
  3. Let's go back in time, to where it all started: In 1954, the US Air Force was in desperate need to translate some Russian snippets of text during the Cold War. So they paid two linguists, an American and a Russian one, and a programmer of IBM, to come up with an automated solution. The three programmed a computer with six grammar rules and a small lexicon of 250 words. Surprisingly, this system could translate over 60 Russian sentences into English. We all know that this is not a lot; I could probably order a coffee in Italian. Anyhow, Overnight, Machine Translation became a little superstar and money was pumped into building even better Machine Translation Systems. These early systems were all based on rules and dictionaries, kind of how we were taught Latin and then were expected to make some sense of Ovid's Metamorphoses. They really can't handle phrases or idioms: So "Spass am Umgang mit Menschen" turns into "fun at handling humans".
  4. After the initial euphoria, most researchers lost interest in Machine Translation. Only in the late 1990s interest , when bigger processors could  handle more input faster, interest sparked off again. Google became intersted in Machine Translation. The approach was very different though: Google is good at collecting a huge amount of data and applying statistical learning techniques. Google Translate was fed billions of words of monolingual text. It also parsed aligned texts, meaning for each sentence in one language there is translation in another. Once the system has processed all the texts, it comes up with some rules of how a language works. These rules have nothing to do with what we know as grammar rule: a rule would for instance be "if I know tongue in cheek mostly appears together, it will be translated as "mit einem Augenzwinkern". So, Google Translate gets the "Spass am Umgang mit Menschen" correctly and translates it as "Enjoy working with people". But Google Translate is very weak at grammar and can't deal with structurally complicated sentences (see Fabian Roth's entry).
  5. Google's success breathes new life into Machine Translation research. It's a hot topic among Computer Linguists today. Some systems try to merge the two approaches to get the best of both worlds. They are fed rules, dictionaries plus huge amounts of texts over which they apply statistical means. Since June 2009, Babelfish is such a hybrid Machine Translation System. Meanwhile Google is implementing their own chat programme, called Google Wave. One of the features will be a little robot called Rosy Etta. It should be able to detect the input language just from what you're typing and should translate in real time while you type. But you will have to wait a little until Google Wave is released.
  6. So is Machine Translation really worth it? Or are they jsut providing entertainment with all the  funny menus and signs that we can find while travelling? I say they are worth it: to speed up the process of manual translation. It also gives people with limited language abilities a glimpse into a different culture. Of course they are a far cry from dealing deal with the subtelties of language such as irony, politeness or cultural differences. For Babelfish, Google Translate or Rosy Etta words are dead matter, strings of bits and bites they deal with. Only a human translator lives the word. Or to say it in Emily Dickinson's words:

A WORD is dead
When it is said,
Some say.

I say it just
Begins to live
That day.