Predicting Intonational Boundaries Automatically from Text: The ATIS Domain

Rela t ing the intonat ional characteristics of an u t ter ance to other features inferable f rom its text is impor t an t bo th for speech recognition and for speech synthesis. This work investigates techniques for predic t ing the locat ion of intonat ional phrase boundaries in na tu ra l speech, th rough analyzing a ut terances from the D A R P A Air Travel In format ion Service database. For s ta t is t ical model ing, we employ Classification and Regression Tree ( C A R T ) techniques. We achieve success rates o f jus t