Identifying segments for routing emergency response dialogues

In crisis management situations information is exchanged in different ways. In general, information is exchanged through spoken dialogues or text messaging conversations. Part of this exchanged dialogue information is often relevant to other actors involved in managing the crisis. Due to the dynamic character of the situation, dialogue partners may not be aware of who else needs the exchanged information. We present a coarse-grained segmentation method for automatically recognizing coherent dialogue segments which are then used for routing. We investigate the effectiveness of our features for recognizing boundaries of segments on transcribed emergency response dialogues and we compare classification by relevance of the identified information segments to the ideal topic segments.