This article, at first, introduces lexical-functional theory as a theory which: (a) has psychological reality; (b) is computationally exact; and (c) makes relationship between different levels of language. Then, it deals with sentences which are ambiguous for translation machine and, consequently, the machine generates two constituent structures for them: i.e. distinguishing "phrasal verb + noun phrase" from "verb + prepositional phrase" and distinguishing "noun phrase + prepositional phrase" from "noun phrase" and "prepositional phrase". To disambiguate these sentences, the solution taken by a human translator, i.e. using semantic information, is modulated for machine through lexical mapping theory.