The generation task is solved by two modules. The microplanner decides about syntactic structures and word choice using constraint-satisfaction techniques. The syntactic generator uses a declarative grammar to produce a syntactically correct utterance. The generator's output is an (annotated) English utterance which is handed over to the speech synthesis component for articulation. The input to VM-GECO is a sequence of semantic representations for utterances in the target language. They have been produced by the transfer component of the VERBMOBIL system by mapping semantic representations for German or Japanese input utterances onto semantic structures for English.