Motivation A common feature of a number of current spoken dialogue systems for information retrieval is that little emphasis is placed on the generation of system contributions to the dialogue. In these systems, utterances have mostly been produced from templates, see for instance (Whittaker and Attwater 1994; Blomberg et al. 1993; Oerder and Aust 1993; Meng et al. 1996) . This is a valid approach in system initiative type systems and in systems where utterances stand in a one-to-one relation

more »

... communicative goals. In mixed initiative systems, however, user and system might both lead the dialogue by providing several pieces of information and pursuing several different goals within one utterance. Hence, in this kind of dialogue we cannot predict what information the user chooses to provide, and hence cannot predict the system's response. We argue that in any system of reasonable size, the number of templates would be too large to determine a priori. Instead, in order to achieve efficient and cooperative dialogue, system utterances must be generated using natural language generation (NLG) techniques. NLG has been used by, for instance, (Pan and McKeown 1996; Sadek et al. 1996) , but they put emphasis on generation of system answers, i.e. on offering and providing information. We are concerned with utterances requesting specific information. (Sadek et al. 1996) also generate requests, but they are mostly of a general nature, and the sample dialogue in their paper suggests that the system's requests for specific information can only realize one communicative goal at a time.