VERBMOBIL makes use of different translation methods which are ordered along the line of one so-called translation track each. The deep track employs HPSG-grammars to obtain syntax and semantics which are in turn encoded in VIT's (VERBMOBIL interface terms) [Bos et al. 1998]. From these VIT's, together with pragmatic information, direx expressions are extracted and transferred to the dialogue manager for completion, storage and evaluation.