You can maybe try to think about languages differences and model these with a binary tree, like "is feature X found ? " if yes, proceed in one direction, if not, proceed in another direction.
By constructing this search tree efficiently you could end with a rather fast code.
