Can Prosody Aid the Automatic Classification of Dialog Acts in Conversational Speech?

Investigated whether current approaches to automatically classifying dialog acts (DAs) in natural conversation could be improved by adding prosodic information. Using conversations from the switchboard corpus, researchers annotated DAs, automatically extracting prosodic features for each DA. Results indicated that DAs are redundantly marked in natural conversation. A variety of automatically extractable prosodic features could aid dialog processing in speech applications. (SM)