The end of interpreters?

A "Tower of Babel" device that gives the illusion of being bilingual is being developed by US scientists.

Users simply have to silently mouth a word in their own language for it to be translated and read out in another.

Read the end of this article on the BBC news website
Excellent analysis of the sytem on the ever brilliant Language Log

By |2006-10-26T07:38:26+00:00October 26th, 2006|Interpreting|9 Comments

About the Author:

I am Céline Graciet, a freelance English to French translator. Since 2003 I’ve been writing on all sorts of areas linked to translation and the life of a translator.


  1. neij October 26, 2006 at 9:48 am

    Hi Céline, I saw this article and immediately thought of you… headed over here and there was your article about it!
    I’m always amused by articles like this in the press written by non-linguists. The article mentions a “vocabulary” of 100 to 200 words. Given that this is about one-fifth of the working active vocabulary of a THREE-year old, I’m amazed that the “system” is regarded as being anywhere near newsworthy. If the best efforts of publicly-available machine translation services such as Babble Fish are anything to go by, Céline and her colleagues have nothing to worry about in the near future!

  2. céline October 26, 2006 at 10:01 am

    100 to 200 words is only a start. Once they’ve perfected the system, added millions of words, attached various meanings to various contexts, who knows what the possibilities will be?
    And let’s forget about pesky interpreters and their self-centered worries: wouldn’t it be amazing if such a tool was to be developed? No more language barrier, trouble-free communication with every human being! Finally, we can be as clever as Jean-Luc Picard and Captain Janeway!

  3. Josep Tarrés October 26, 2006 at 10:37 am

    I think it would be too expensive (if possible) to maintain a set of such TM.
    The idea of processing just words, with a few context indicators, is completely unworkable, since human communication mixes always two or more contexts, and human speakers often make mistakes which should also be expected by the machine.
    Since machines cannot ‘think’, the only possible way to allow such an interpretation would be the use of huge translation memories, but again, it is unworkable because of size and time to maintain them.

  4. Tony October 26, 2006 at 1:54 pm

    Irresponsible of New Scientist to publish a piece about this, and irresponsible of the BBC to pick it up.
    You don’t need to be a linguist to see that this not even worth the effort of rubbishing.

  5. Jemima October 26, 2006 at 3:58 pm

    The thing I like is that despite it having a vocab of only 100-200 words, it still gets it wrong about 20% of the time. And that’s without the subtlety of context to worry about. The genuinely clever thing, though, is the real time translation element. I wonder if the resulting speed would mean that confusion could be clarified quickly enough to make it workable, once the vocab is larger.

  6. céline October 26, 2006 at 4:02 pm

    The real-time aspect is definitely very exciting, but again, how can the program cope with regional accents, which mean that even a simple word like “film” just isn’t pronounced using exactly the same muscles whether you’re from Glasgow or Tunbridge Wells?

  7. Jemima October 26, 2006 at 4:10 pm

    I guess regional accents could be dealt with by extra sampling combined, perhaps, with some level of setup to align the speaker with the most likely range of sounds known to the machine. I guess. I guess, too, that you’d have to use headphones, or you’d get very confused hearing your speech in a different language – a little like when there’s echo on a phone line.

  8. Mark Liberman October 27, 2006 at 10:08 pm

    You might be interested in a discussion of the technology behind the story:

  9. céline October 28, 2006 at 8:39 am

    Fantastic. That’s exactly the kind of analysis I was after and that the BBC article completely failed to provide. I really should have thought of checking the excellent Language Log.

Comments are closed.