i18n@... wrote:

>Mark E. Shoulson wrote:
>
>
>
>>True fully automatic high-quality machine translation is known to be
>>impossible (or at least AI-hard), so there's only so high one's hopes
>>can be set. But at least such things can be convenient in saving you
>>trips to the dictionary, at least.
>>
>>~mark
>>
>>
>>
>
>Does "AI-hard" mean there is no known algorithm (hence it is an open
>question)?
>
>

I, at least, mean by it that it is as hard as producing "true"
artificial intelligence. That is, in order to understand language well
enough to translate it believably, you essentially have to be working
with approximately human intelligence and thinking patterns (think Cmdr.
Data).

Note that I'm not asserting that it is that hard, but reporting a
proof/assertion by Yehoshua Bar-Hillel that I've heard of. A Google
search finds me
http://ourworld.compuserve.com/homepages/WJHutchins/Miles-6.htm for you
to look at, which discusses it some (I haven't read it all through).

>I am not so convinced that it will always remain an open issue. I used
>to be, but I have been working on some approaches I haven't found in the
>literature, vast and old as it is....
>
>

I'm not sure that fully human-like AI will always remain an open issue
either.

~mark