Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Evaluating Radio News Intonation - Autosegmental Versus Superpositional Modelling

Maria Wolters (1), Hansjörg Mixdorff (2)

(1) Universität Bonn, Germany
(2) TU Dresden, Germany

This study examines prosodic correlates of the givenness of discourse entities in German radio news speech. The material comes from the Stuttgart Radio News Corpus. Both GToBI intonation labels and a Fujisaki-style parametrization of the intonation contour were examined. We find strong word-class specific accentuation defaults; the influence of entity status is rather small and varies with word class. However, there are strong influences of newness on phrasing. The results of autosegmental and superpositional approaches complement each other nicely.


Full Paper

Bibliographic reference.  Wolters, Maria / Mixdorff, Hansjörg (2000): "Evaluating radio news intonation - autosegmental versus superpositional modelling", In ICSLP-2000, vol.1, 581-584.