Context construction as subtask of dialogue processing - the VERBMOBIL case

Abstract

This paper presents the dialogue component of the speech translation system VERBMOBIL. In particular, it describes the Dialogue Memory which has been developed to represent contextual information acquired during dialogue processing. Information is stored both chronologically, i.e. in the order of appearance (in the Sequence Memory) and conceptually (in the Thematic Structure). We show how the Dialogue Memory is used to draw contextual inferences, some of which serve as basis for the detection of indirect speech acts.