[Date Prev][Date Next][Thread Prev][Thread Next][Chronological Index][Thread Index]

Re: EAGLESEVAL: A question from George Doddington.

> But doesn't that mean that they HAVEN'T made progress on the underlying
> levels?

Suppose that it will take 15 years of hard work to produce a gigantic
improvement in MT quality, but absolutely no improvement until you get
very close to the new plateau.  Sort of a step-function in the [meta]
search space.  The "greedy search" produced by annual
application-level evaluations will prevent anyone from ever finding
this new plateau [speaking at a meta level, with the researchers
executing the search algorithm].  The hypothesis is that interlingual
MT may have such development characteristics.  Classical science,
where goodness is judged by your peers thinking your work is
"interesting", does not have this greedy-search property (as much), in
my opinion.

I think that's what Ed meant, too.