Tal Cohen has a personal opinion about Douglas R. Hofstadter’s book “Le Ton beau de Marot: In Praise of the Music of Language”. Hofstadter, of course, has an opinion about the process of translating an expression of an idea from one language into another language.
I have a simpler (?) problem to consider. An utterance in one language (say, language A) can be translated into one of several utterances in another language (say, language B). The original utterance can have one of several possible meanings, depending upon the context. To each meaning correspond one or more translations in language B.
Now, in a computerized translation system, we would like to be able to specify, for each utterance in the original text, enough of the context in which it lies, for the software to be able to fit a good translation into language B.
The question is how to specify context. We need to know also how much context to specify. If we specify more than this amount of context, we are really translating into a different dialect of language B.
I once wrote a story which made “Fruit flies like a banana, time flies like an arrow” translate the opposite of what you'd've thought. It required about two pages of SF story to get the reader to the point where this is the right context. I believe “determining enough context” and “formalizing context” to be AI-Complete 🙂
(http://livejournal.com/users/moshez)