calculated optimal null-prior value

I wonder is it possible to calculate balanced prior probability of NULL alignment to fit optimal score between recall for precision? I bet [GIZA](https://github.com/moses-smt/giza-pp) or [MGIZA](https://github.com/moses-smt/mgiza) aligners calculates it, any chance eflomal could do something like that?