It's relatively easy to translate structured English to pseudo code or pseudo
code to source code. The challenge is that no human being is good at writing
sufficiently precise English or pseudo code such that the final source code is
correct even syntactically, much less semantically. In the end, I'm sure you
can write a program that can translate successfully, but I suspect the generated
source code will be too imprecise to be correct.
As to translating between natural languages, I can't comment on the state of the
art. I understand that those systems that have had the greatest success have
been commercial tools on which developers spent a LOT of time hand customizing
the translation process. AFAIK, no theoretical models for machine translation
have accomplished a useful level of performance without major hand tuning for
idioms and semantic subtleties.
Randy Crawford http://www.yqcomputer.com/
~rand rand AT rice DOT edu