Finite-state models for lexical reordering in spoken language translation