Within the SPICOS project, which is part of the joint research-project 'Speech Understanding Systems', a prototype of a dialogue-system was successfully implemented, which allows questioning a database in continuous speech. The answers are output in synthesized speech. For this SPICOS prototype we have developed a powerful and yet flexible approach for syntactic and semantic analysis. The analysis has sucessfully been adapted to different acoustic outputs, which vary with respect to the quality and the quantity of the hypothesized words of the utterance.The stages of the linguistic analysis and problems resulting from the different outputs of the acoustic modules will be pointed out in the paper.The system deals with a vocabulary of about 1000 words of German and covers the type of sentences typical for data-base queries, such as imperatives.wh-questions and sentence questions.